xref: /linux/kernel/trace/trace_uprobe.c (revision 44f57d78)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * uprobes-based tracing events
4  *
5  * Copyright (C) IBM Corporation, 2010-2012
6  * Author:	Srikar Dronamraju <srikar@linux.vnet.ibm.com>
7  */
8 #define pr_fmt(fmt)	"trace_uprobe: " fmt
9 
10 #include <linux/ctype.h>
11 #include <linux/module.h>
12 #include <linux/uaccess.h>
13 #include <linux/uprobes.h>
14 #include <linux/namei.h>
15 #include <linux/string.h>
16 #include <linux/rculist.h>
17 
18 #include "trace_dynevent.h"
19 #include "trace_probe.h"
20 #include "trace_probe_tmpl.h"
21 
22 #define UPROBE_EVENT_SYSTEM	"uprobes"
23 
24 struct uprobe_trace_entry_head {
25 	struct trace_entry	ent;
26 	unsigned long		vaddr[];
27 };
28 
29 #define SIZEOF_TRACE_ENTRY(is_return)			\
30 	(sizeof(struct uprobe_trace_entry_head) +	\
31 	 sizeof(unsigned long) * (is_return ? 2 : 1))
32 
33 #define DATAOF_TRACE_ENTRY(entry, is_return)		\
34 	((void*)(entry) + SIZEOF_TRACE_ENTRY(is_return))
35 
36 struct trace_uprobe_filter {
37 	rwlock_t		rwlock;
38 	int			nr_systemwide;
39 	struct list_head	perf_events;
40 };
41 
42 static int trace_uprobe_create(int argc, const char **argv);
43 static int trace_uprobe_show(struct seq_file *m, struct dyn_event *ev);
44 static int trace_uprobe_release(struct dyn_event *ev);
45 static bool trace_uprobe_is_busy(struct dyn_event *ev);
46 static bool trace_uprobe_match(const char *system, const char *event,
47 			       struct dyn_event *ev);
48 
49 static struct dyn_event_operations trace_uprobe_ops = {
50 	.create = trace_uprobe_create,
51 	.show = trace_uprobe_show,
52 	.is_busy = trace_uprobe_is_busy,
53 	.free = trace_uprobe_release,
54 	.match = trace_uprobe_match,
55 };
56 
57 /*
58  * uprobe event core functions
59  */
60 struct trace_uprobe {
61 	struct dyn_event		devent;
62 	struct trace_uprobe_filter	filter;
63 	struct uprobe_consumer		consumer;
64 	struct path			path;
65 	struct inode			*inode;
66 	char				*filename;
67 	unsigned long			offset;
68 	unsigned long			ref_ctr_offset;
69 	unsigned long			nhit;
70 	struct trace_probe		tp;
71 };
72 
73 static bool is_trace_uprobe(struct dyn_event *ev)
74 {
75 	return ev->ops == &trace_uprobe_ops;
76 }
77 
78 static struct trace_uprobe *to_trace_uprobe(struct dyn_event *ev)
79 {
80 	return container_of(ev, struct trace_uprobe, devent);
81 }
82 
83 /**
84  * for_each_trace_uprobe - iterate over the trace_uprobe list
85  * @pos:	the struct trace_uprobe * for each entry
86  * @dpos:	the struct dyn_event * to use as a loop cursor
87  */
88 #define for_each_trace_uprobe(pos, dpos)	\
89 	for_each_dyn_event(dpos)		\
90 		if (is_trace_uprobe(dpos) && (pos = to_trace_uprobe(dpos)))
91 
92 #define SIZEOF_TRACE_UPROBE(n)				\
93 	(offsetof(struct trace_uprobe, tp.args) +	\
94 	(sizeof(struct probe_arg) * (n)))
95 
96 static int register_uprobe_event(struct trace_uprobe *tu);
97 static int unregister_uprobe_event(struct trace_uprobe *tu);
98 
99 struct uprobe_dispatch_data {
100 	struct trace_uprobe	*tu;
101 	unsigned long		bp_addr;
102 };
103 
104 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs);
105 static int uretprobe_dispatcher(struct uprobe_consumer *con,
106 				unsigned long func, struct pt_regs *regs);
107 
108 #ifdef CONFIG_STACK_GROWSUP
109 static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
110 {
111 	return addr - (n * sizeof(long));
112 }
113 #else
114 static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
115 {
116 	return addr + (n * sizeof(long));
117 }
118 #endif
119 
120 static unsigned long get_user_stack_nth(struct pt_regs *regs, unsigned int n)
121 {
122 	unsigned long ret;
123 	unsigned long addr = user_stack_pointer(regs);
124 
125 	addr = adjust_stack_addr(addr, n);
126 
127 	if (copy_from_user(&ret, (void __force __user *) addr, sizeof(ret)))
128 		return 0;
129 
130 	return ret;
131 }
132 
133 /*
134  * Uprobes-specific fetch functions
135  */
136 static nokprobe_inline int
137 probe_mem_read(void *dest, void *src, size_t size)
138 {
139 	void __user *vaddr = (void __force __user *)src;
140 
141 	return copy_from_user(dest, vaddr, size) ? -EFAULT : 0;
142 }
143 /*
144  * Fetch a null-terminated string. Caller MUST set *(u32 *)dest with max
145  * length and relative data location.
146  */
147 static nokprobe_inline int
148 fetch_store_string(unsigned long addr, void *dest, void *base)
149 {
150 	long ret;
151 	u32 loc = *(u32 *)dest;
152 	int maxlen  = get_loc_len(loc);
153 	u8 *dst = get_loc_data(dest, base);
154 	void __user *src = (void __force __user *) addr;
155 
156 	if (unlikely(!maxlen))
157 		return -ENOMEM;
158 
159 	if (addr == FETCH_TOKEN_COMM)
160 		ret = strlcpy(dst, current->comm, maxlen);
161 	else
162 		ret = strncpy_from_user(dst, src, maxlen);
163 	if (ret >= 0) {
164 		if (ret == maxlen)
165 			dst[ret - 1] = '\0';
166 		else
167 			/*
168 			 * Include the terminating null byte. In this case it
169 			 * was copied by strncpy_from_user but not accounted
170 			 * for in ret.
171 			 */
172 			ret++;
173 		*(u32 *)dest = make_data_loc(ret, (void *)dst - base);
174 	}
175 
176 	return ret;
177 }
178 
179 /* Return the length of string -- including null terminal byte */
180 static nokprobe_inline int
181 fetch_store_strlen(unsigned long addr)
182 {
183 	int len;
184 	void __user *vaddr = (void __force __user *) addr;
185 
186 	if (addr == FETCH_TOKEN_COMM)
187 		len = strlen(current->comm) + 1;
188 	else
189 		len = strnlen_user(vaddr, MAX_STRING_SIZE);
190 
191 	return (len > MAX_STRING_SIZE) ? 0 : len;
192 }
193 
194 static unsigned long translate_user_vaddr(unsigned long file_offset)
195 {
196 	unsigned long base_addr;
197 	struct uprobe_dispatch_data *udd;
198 
199 	udd = (void *) current->utask->vaddr;
200 
201 	base_addr = udd->bp_addr - udd->tu->offset;
202 	return base_addr + file_offset;
203 }
204 
205 /* Note that we don't verify it, since the code does not come from user space */
206 static int
207 process_fetch_insn(struct fetch_insn *code, struct pt_regs *regs, void *dest,
208 		   void *base)
209 {
210 	unsigned long val;
211 
212 	/* 1st stage: get value from context */
213 	switch (code->op) {
214 	case FETCH_OP_REG:
215 		val = regs_get_register(regs, code->param);
216 		break;
217 	case FETCH_OP_STACK:
218 		val = get_user_stack_nth(regs, code->param);
219 		break;
220 	case FETCH_OP_STACKP:
221 		val = user_stack_pointer(regs);
222 		break;
223 	case FETCH_OP_RETVAL:
224 		val = regs_return_value(regs);
225 		break;
226 	case FETCH_OP_IMM:
227 		val = code->immediate;
228 		break;
229 	case FETCH_OP_COMM:
230 		val = FETCH_TOKEN_COMM;
231 		break;
232 	case FETCH_OP_FOFFS:
233 		val = translate_user_vaddr(code->immediate);
234 		break;
235 	default:
236 		return -EILSEQ;
237 	}
238 	code++;
239 
240 	return process_fetch_insn_bottom(code, val, dest, base);
241 }
242 NOKPROBE_SYMBOL(process_fetch_insn)
243 
244 static inline void init_trace_uprobe_filter(struct trace_uprobe_filter *filter)
245 {
246 	rwlock_init(&filter->rwlock);
247 	filter->nr_systemwide = 0;
248 	INIT_LIST_HEAD(&filter->perf_events);
249 }
250 
251 static inline bool uprobe_filter_is_empty(struct trace_uprobe_filter *filter)
252 {
253 	return !filter->nr_systemwide && list_empty(&filter->perf_events);
254 }
255 
256 static inline bool is_ret_probe(struct trace_uprobe *tu)
257 {
258 	return tu->consumer.ret_handler != NULL;
259 }
260 
261 static bool trace_uprobe_is_busy(struct dyn_event *ev)
262 {
263 	struct trace_uprobe *tu = to_trace_uprobe(ev);
264 
265 	return trace_probe_is_enabled(&tu->tp);
266 }
267 
268 static bool trace_uprobe_match(const char *system, const char *event,
269 			       struct dyn_event *ev)
270 {
271 	struct trace_uprobe *tu = to_trace_uprobe(ev);
272 
273 	return strcmp(trace_event_name(&tu->tp.call), event) == 0 &&
274 		(!system || strcmp(tu->tp.call.class->system, system) == 0);
275 }
276 
277 /*
278  * Allocate new trace_uprobe and initialize it (including uprobes).
279  */
280 static struct trace_uprobe *
281 alloc_trace_uprobe(const char *group, const char *event, int nargs, bool is_ret)
282 {
283 	struct trace_uprobe *tu;
284 
285 	if (!event || !group)
286 		return ERR_PTR(-EINVAL);
287 
288 	tu = kzalloc(SIZEOF_TRACE_UPROBE(nargs), GFP_KERNEL);
289 	if (!tu)
290 		return ERR_PTR(-ENOMEM);
291 
292 	tu->tp.call.class = &tu->tp.class;
293 	tu->tp.call.name = kstrdup(event, GFP_KERNEL);
294 	if (!tu->tp.call.name)
295 		goto error;
296 
297 	tu->tp.class.system = kstrdup(group, GFP_KERNEL);
298 	if (!tu->tp.class.system)
299 		goto error;
300 
301 	dyn_event_init(&tu->devent, &trace_uprobe_ops);
302 	INIT_LIST_HEAD(&tu->tp.files);
303 	tu->consumer.handler = uprobe_dispatcher;
304 	if (is_ret)
305 		tu->consumer.ret_handler = uretprobe_dispatcher;
306 	init_trace_uprobe_filter(&tu->filter);
307 	return tu;
308 
309 error:
310 	kfree(tu->tp.call.name);
311 	kfree(tu);
312 
313 	return ERR_PTR(-ENOMEM);
314 }
315 
316 static void free_trace_uprobe(struct trace_uprobe *tu)
317 {
318 	int i;
319 
320 	if (!tu)
321 		return;
322 
323 	for (i = 0; i < tu->tp.nr_args; i++)
324 		traceprobe_free_probe_arg(&tu->tp.args[i]);
325 
326 	path_put(&tu->path);
327 	kfree(tu->tp.call.class->system);
328 	kfree(tu->tp.call.name);
329 	kfree(tu->filename);
330 	kfree(tu);
331 }
332 
333 static struct trace_uprobe *find_probe_event(const char *event, const char *group)
334 {
335 	struct dyn_event *pos;
336 	struct trace_uprobe *tu;
337 
338 	for_each_trace_uprobe(tu, pos)
339 		if (strcmp(trace_event_name(&tu->tp.call), event) == 0 &&
340 		    strcmp(tu->tp.call.class->system, group) == 0)
341 			return tu;
342 
343 	return NULL;
344 }
345 
346 /* Unregister a trace_uprobe and probe_event */
347 static int unregister_trace_uprobe(struct trace_uprobe *tu)
348 {
349 	int ret;
350 
351 	ret = unregister_uprobe_event(tu);
352 	if (ret)
353 		return ret;
354 
355 	dyn_event_remove(&tu->devent);
356 	free_trace_uprobe(tu);
357 	return 0;
358 }
359 
360 /*
361  * Uprobe with multiple reference counter is not allowed. i.e.
362  * If inode and offset matches, reference counter offset *must*
363  * match as well. Though, there is one exception: If user is
364  * replacing old trace_uprobe with new one(same group/event),
365  * then we allow same uprobe with new reference counter as far
366  * as the new one does not conflict with any other existing
367  * ones.
368  */
369 static struct trace_uprobe *find_old_trace_uprobe(struct trace_uprobe *new)
370 {
371 	struct dyn_event *pos;
372 	struct trace_uprobe *tmp, *old = NULL;
373 	struct inode *new_inode = d_real_inode(new->path.dentry);
374 
375 	old = find_probe_event(trace_event_name(&new->tp.call),
376 				new->tp.call.class->system);
377 
378 	for_each_trace_uprobe(tmp, pos) {
379 		if ((old ? old != tmp : true) &&
380 		    new_inode == d_real_inode(tmp->path.dentry) &&
381 		    new->offset == tmp->offset &&
382 		    new->ref_ctr_offset != tmp->ref_ctr_offset) {
383 			pr_warn("Reference counter offset mismatch.");
384 			return ERR_PTR(-EINVAL);
385 		}
386 	}
387 	return old;
388 }
389 
390 /* Register a trace_uprobe and probe_event */
391 static int register_trace_uprobe(struct trace_uprobe *tu)
392 {
393 	struct trace_uprobe *old_tu;
394 	int ret;
395 
396 	mutex_lock(&event_mutex);
397 
398 	/* register as an event */
399 	old_tu = find_old_trace_uprobe(tu);
400 	if (IS_ERR(old_tu)) {
401 		ret = PTR_ERR(old_tu);
402 		goto end;
403 	}
404 
405 	if (old_tu) {
406 		/* delete old event */
407 		ret = unregister_trace_uprobe(old_tu);
408 		if (ret)
409 			goto end;
410 	}
411 
412 	ret = register_uprobe_event(tu);
413 	if (ret) {
414 		pr_warn("Failed to register probe event(%d)\n", ret);
415 		goto end;
416 	}
417 
418 	dyn_event_add(&tu->devent);
419 
420 end:
421 	mutex_unlock(&event_mutex);
422 
423 	return ret;
424 }
425 
426 /*
427  * Argument syntax:
428  *  - Add uprobe: p|r[:[GRP/]EVENT] PATH:OFFSET [FETCHARGS]
429  */
430 static int trace_uprobe_create(int argc, const char **argv)
431 {
432 	struct trace_uprobe *tu;
433 	const char *event = NULL, *group = UPROBE_EVENT_SYSTEM;
434 	char *arg, *filename, *rctr, *rctr_end, *tmp;
435 	char buf[MAX_EVENT_NAME_LEN];
436 	struct path path;
437 	unsigned long offset, ref_ctr_offset;
438 	bool is_return = false;
439 	int i, ret;
440 
441 	ret = 0;
442 	ref_ctr_offset = 0;
443 
444 	switch (argv[0][0]) {
445 	case 'r':
446 		is_return = true;
447 		break;
448 	case 'p':
449 		break;
450 	default:
451 		return -ECANCELED;
452 	}
453 
454 	if (argc < 2)
455 		return -ECANCELED;
456 
457 	if (argv[0][1] == ':')
458 		event = &argv[0][2];
459 
460 	if (!strchr(argv[1], '/'))
461 		return -ECANCELED;
462 
463 	filename = kstrdup(argv[1], GFP_KERNEL);
464 	if (!filename)
465 		return -ENOMEM;
466 
467 	/* Find the last occurrence, in case the path contains ':' too. */
468 	arg = strrchr(filename, ':');
469 	if (!arg || !isdigit(arg[1])) {
470 		kfree(filename);
471 		return -ECANCELED;
472 	}
473 
474 	trace_probe_log_init("trace_uprobe", argc, argv);
475 	trace_probe_log_set_index(1);	/* filename is the 2nd argument */
476 
477 	*arg++ = '\0';
478 	ret = kern_path(filename, LOOKUP_FOLLOW, &path);
479 	if (ret) {
480 		trace_probe_log_err(0, FILE_NOT_FOUND);
481 		kfree(filename);
482 		trace_probe_log_clear();
483 		return ret;
484 	}
485 	if (!d_is_reg(path.dentry)) {
486 		trace_probe_log_err(0, NO_REGULAR_FILE);
487 		ret = -EINVAL;
488 		goto fail_address_parse;
489 	}
490 
491 	/* Parse reference counter offset if specified. */
492 	rctr = strchr(arg, '(');
493 	if (rctr) {
494 		rctr_end = strchr(rctr, ')');
495 		if (!rctr_end) {
496 			ret = -EINVAL;
497 			rctr_end = rctr + strlen(rctr);
498 			trace_probe_log_err(rctr_end - filename,
499 					    REFCNT_OPEN_BRACE);
500 			goto fail_address_parse;
501 		} else if (rctr_end[1] != '\0') {
502 			ret = -EINVAL;
503 			trace_probe_log_err(rctr_end + 1 - filename,
504 					    BAD_REFCNT_SUFFIX);
505 			goto fail_address_parse;
506 		}
507 
508 		*rctr++ = '\0';
509 		*rctr_end = '\0';
510 		ret = kstrtoul(rctr, 0, &ref_ctr_offset);
511 		if (ret) {
512 			trace_probe_log_err(rctr - filename, BAD_REFCNT);
513 			goto fail_address_parse;
514 		}
515 	}
516 
517 	/* Parse uprobe offset. */
518 	ret = kstrtoul(arg, 0, &offset);
519 	if (ret) {
520 		trace_probe_log_err(arg - filename, BAD_UPROBE_OFFS);
521 		goto fail_address_parse;
522 	}
523 
524 	/* setup a probe */
525 	trace_probe_log_set_index(0);
526 	if (event) {
527 		ret = traceprobe_parse_event_name(&event, &group, buf,
528 						  event - argv[0]);
529 		if (ret)
530 			goto fail_address_parse;
531 	} else {
532 		char *tail;
533 		char *ptr;
534 
535 		tail = kstrdup(kbasename(filename), GFP_KERNEL);
536 		if (!tail) {
537 			ret = -ENOMEM;
538 			goto fail_address_parse;
539 		}
540 
541 		ptr = strpbrk(tail, ".-_");
542 		if (ptr)
543 			*ptr = '\0';
544 
545 		snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_0x%lx", 'p', tail, offset);
546 		event = buf;
547 		kfree(tail);
548 	}
549 
550 	argc -= 2;
551 	argv += 2;
552 
553 	tu = alloc_trace_uprobe(group, event, argc, is_return);
554 	if (IS_ERR(tu)) {
555 		ret = PTR_ERR(tu);
556 		/* This must return -ENOMEM otherwise there is a bug */
557 		WARN_ON_ONCE(ret != -ENOMEM);
558 		goto fail_address_parse;
559 	}
560 	tu->offset = offset;
561 	tu->ref_ctr_offset = ref_ctr_offset;
562 	tu->path = path;
563 	tu->filename = filename;
564 
565 	/* parse arguments */
566 	for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
567 		tmp = kstrdup(argv[i], GFP_KERNEL);
568 		if (!tmp) {
569 			ret = -ENOMEM;
570 			goto error;
571 		}
572 
573 		trace_probe_log_set_index(i + 2);
574 		ret = traceprobe_parse_probe_arg(&tu->tp, i, tmp,
575 					is_return ? TPARG_FL_RETURN : 0);
576 		kfree(tmp);
577 		if (ret)
578 			goto error;
579 	}
580 
581 	ret = register_trace_uprobe(tu);
582 	if (!ret)
583 		goto out;
584 
585 error:
586 	free_trace_uprobe(tu);
587 out:
588 	trace_probe_log_clear();
589 	return ret;
590 
591 fail_address_parse:
592 	trace_probe_log_clear();
593 	path_put(&path);
594 	kfree(filename);
595 
596 	return ret;
597 }
598 
599 static int create_or_delete_trace_uprobe(int argc, char **argv)
600 {
601 	int ret;
602 
603 	if (argv[0][0] == '-')
604 		return dyn_event_release(argc, argv, &trace_uprobe_ops);
605 
606 	ret = trace_uprobe_create(argc, (const char **)argv);
607 	return ret == -ECANCELED ? -EINVAL : ret;
608 }
609 
610 static int trace_uprobe_release(struct dyn_event *ev)
611 {
612 	struct trace_uprobe *tu = to_trace_uprobe(ev);
613 
614 	return unregister_trace_uprobe(tu);
615 }
616 
617 /* Probes listing interfaces */
618 static int trace_uprobe_show(struct seq_file *m, struct dyn_event *ev)
619 {
620 	struct trace_uprobe *tu = to_trace_uprobe(ev);
621 	char c = is_ret_probe(tu) ? 'r' : 'p';
622 	int i;
623 
624 	seq_printf(m, "%c:%s/%s %s:0x%0*lx", c, tu->tp.call.class->system,
625 			trace_event_name(&tu->tp.call), tu->filename,
626 			(int)(sizeof(void *) * 2), tu->offset);
627 
628 	if (tu->ref_ctr_offset)
629 		seq_printf(m, "(0x%lx)", tu->ref_ctr_offset);
630 
631 	for (i = 0; i < tu->tp.nr_args; i++)
632 		seq_printf(m, " %s=%s", tu->tp.args[i].name, tu->tp.args[i].comm);
633 
634 	seq_putc(m, '\n');
635 	return 0;
636 }
637 
638 static int probes_seq_show(struct seq_file *m, void *v)
639 {
640 	struct dyn_event *ev = v;
641 
642 	if (!is_trace_uprobe(ev))
643 		return 0;
644 
645 	return trace_uprobe_show(m, ev);
646 }
647 
648 static const struct seq_operations probes_seq_op = {
649 	.start  = dyn_event_seq_start,
650 	.next   = dyn_event_seq_next,
651 	.stop   = dyn_event_seq_stop,
652 	.show   = probes_seq_show
653 };
654 
655 static int probes_open(struct inode *inode, struct file *file)
656 {
657 	int ret;
658 
659 	if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
660 		ret = dyn_events_release_all(&trace_uprobe_ops);
661 		if (ret)
662 			return ret;
663 	}
664 
665 	return seq_open(file, &probes_seq_op);
666 }
667 
668 static ssize_t probes_write(struct file *file, const char __user *buffer,
669 			    size_t count, loff_t *ppos)
670 {
671 	return trace_parse_run_command(file, buffer, count, ppos,
672 					create_or_delete_trace_uprobe);
673 }
674 
675 static const struct file_operations uprobe_events_ops = {
676 	.owner		= THIS_MODULE,
677 	.open		= probes_open,
678 	.read		= seq_read,
679 	.llseek		= seq_lseek,
680 	.release	= seq_release,
681 	.write		= probes_write,
682 };
683 
684 /* Probes profiling interfaces */
685 static int probes_profile_seq_show(struct seq_file *m, void *v)
686 {
687 	struct dyn_event *ev = v;
688 	struct trace_uprobe *tu;
689 
690 	if (!is_trace_uprobe(ev))
691 		return 0;
692 
693 	tu = to_trace_uprobe(ev);
694 	seq_printf(m, "  %s %-44s %15lu\n", tu->filename,
695 			trace_event_name(&tu->tp.call), tu->nhit);
696 	return 0;
697 }
698 
699 static const struct seq_operations profile_seq_op = {
700 	.start  = dyn_event_seq_start,
701 	.next   = dyn_event_seq_next,
702 	.stop   = dyn_event_seq_stop,
703 	.show	= probes_profile_seq_show
704 };
705 
706 static int profile_open(struct inode *inode, struct file *file)
707 {
708 	return seq_open(file, &profile_seq_op);
709 }
710 
711 static const struct file_operations uprobe_profile_ops = {
712 	.owner		= THIS_MODULE,
713 	.open		= profile_open,
714 	.read		= seq_read,
715 	.llseek		= seq_lseek,
716 	.release	= seq_release,
717 };
718 
719 struct uprobe_cpu_buffer {
720 	struct mutex mutex;
721 	void *buf;
722 };
723 static struct uprobe_cpu_buffer __percpu *uprobe_cpu_buffer;
724 static int uprobe_buffer_refcnt;
725 
726 static int uprobe_buffer_init(void)
727 {
728 	int cpu, err_cpu;
729 
730 	uprobe_cpu_buffer = alloc_percpu(struct uprobe_cpu_buffer);
731 	if (uprobe_cpu_buffer == NULL)
732 		return -ENOMEM;
733 
734 	for_each_possible_cpu(cpu) {
735 		struct page *p = alloc_pages_node(cpu_to_node(cpu),
736 						  GFP_KERNEL, 0);
737 		if (p == NULL) {
738 			err_cpu = cpu;
739 			goto err;
740 		}
741 		per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf = page_address(p);
742 		mutex_init(&per_cpu_ptr(uprobe_cpu_buffer, cpu)->mutex);
743 	}
744 
745 	return 0;
746 
747 err:
748 	for_each_possible_cpu(cpu) {
749 		if (cpu == err_cpu)
750 			break;
751 		free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf);
752 	}
753 
754 	free_percpu(uprobe_cpu_buffer);
755 	return -ENOMEM;
756 }
757 
758 static int uprobe_buffer_enable(void)
759 {
760 	int ret = 0;
761 
762 	BUG_ON(!mutex_is_locked(&event_mutex));
763 
764 	if (uprobe_buffer_refcnt++ == 0) {
765 		ret = uprobe_buffer_init();
766 		if (ret < 0)
767 			uprobe_buffer_refcnt--;
768 	}
769 
770 	return ret;
771 }
772 
773 static void uprobe_buffer_disable(void)
774 {
775 	int cpu;
776 
777 	BUG_ON(!mutex_is_locked(&event_mutex));
778 
779 	if (--uprobe_buffer_refcnt == 0) {
780 		for_each_possible_cpu(cpu)
781 			free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer,
782 							     cpu)->buf);
783 
784 		free_percpu(uprobe_cpu_buffer);
785 		uprobe_cpu_buffer = NULL;
786 	}
787 }
788 
789 static struct uprobe_cpu_buffer *uprobe_buffer_get(void)
790 {
791 	struct uprobe_cpu_buffer *ucb;
792 	int cpu;
793 
794 	cpu = raw_smp_processor_id();
795 	ucb = per_cpu_ptr(uprobe_cpu_buffer, cpu);
796 
797 	/*
798 	 * Use per-cpu buffers for fastest access, but we might migrate
799 	 * so the mutex makes sure we have sole access to it.
800 	 */
801 	mutex_lock(&ucb->mutex);
802 
803 	return ucb;
804 }
805 
806 static void uprobe_buffer_put(struct uprobe_cpu_buffer *ucb)
807 {
808 	mutex_unlock(&ucb->mutex);
809 }
810 
811 static void __uprobe_trace_func(struct trace_uprobe *tu,
812 				unsigned long func, struct pt_regs *regs,
813 				struct uprobe_cpu_buffer *ucb, int dsize,
814 				struct trace_event_file *trace_file)
815 {
816 	struct uprobe_trace_entry_head *entry;
817 	struct ring_buffer_event *event;
818 	struct ring_buffer *buffer;
819 	void *data;
820 	int size, esize;
821 	struct trace_event_call *call = &tu->tp.call;
822 
823 	WARN_ON(call != trace_file->event_call);
824 
825 	if (WARN_ON_ONCE(tu->tp.size + dsize > PAGE_SIZE))
826 		return;
827 
828 	if (trace_trigger_soft_disabled(trace_file))
829 		return;
830 
831 	esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
832 	size = esize + tu->tp.size + dsize;
833 	event = trace_event_buffer_lock_reserve(&buffer, trace_file,
834 						call->event.type, size, 0, 0);
835 	if (!event)
836 		return;
837 
838 	entry = ring_buffer_event_data(event);
839 	if (is_ret_probe(tu)) {
840 		entry->vaddr[0] = func;
841 		entry->vaddr[1] = instruction_pointer(regs);
842 		data = DATAOF_TRACE_ENTRY(entry, true);
843 	} else {
844 		entry->vaddr[0] = instruction_pointer(regs);
845 		data = DATAOF_TRACE_ENTRY(entry, false);
846 	}
847 
848 	memcpy(data, ucb->buf, tu->tp.size + dsize);
849 
850 	event_trigger_unlock_commit(trace_file, buffer, event, entry, 0, 0);
851 }
852 
853 /* uprobe handler */
854 static int uprobe_trace_func(struct trace_uprobe *tu, struct pt_regs *regs,
855 			     struct uprobe_cpu_buffer *ucb, int dsize)
856 {
857 	struct event_file_link *link;
858 
859 	if (is_ret_probe(tu))
860 		return 0;
861 
862 	rcu_read_lock();
863 	list_for_each_entry_rcu(link, &tu->tp.files, list)
864 		__uprobe_trace_func(tu, 0, regs, ucb, dsize, link->file);
865 	rcu_read_unlock();
866 
867 	return 0;
868 }
869 
870 static void uretprobe_trace_func(struct trace_uprobe *tu, unsigned long func,
871 				 struct pt_regs *regs,
872 				 struct uprobe_cpu_buffer *ucb, int dsize)
873 {
874 	struct event_file_link *link;
875 
876 	rcu_read_lock();
877 	list_for_each_entry_rcu(link, &tu->tp.files, list)
878 		__uprobe_trace_func(tu, func, regs, ucb, dsize, link->file);
879 	rcu_read_unlock();
880 }
881 
882 /* Event entry printers */
883 static enum print_line_t
884 print_uprobe_event(struct trace_iterator *iter, int flags, struct trace_event *event)
885 {
886 	struct uprobe_trace_entry_head *entry;
887 	struct trace_seq *s = &iter->seq;
888 	struct trace_uprobe *tu;
889 	u8 *data;
890 
891 	entry = (struct uprobe_trace_entry_head *)iter->ent;
892 	tu = container_of(event, struct trace_uprobe, tp.call.event);
893 
894 	if (is_ret_probe(tu)) {
895 		trace_seq_printf(s, "%s: (0x%lx <- 0x%lx)",
896 				 trace_event_name(&tu->tp.call),
897 				 entry->vaddr[1], entry->vaddr[0]);
898 		data = DATAOF_TRACE_ENTRY(entry, true);
899 	} else {
900 		trace_seq_printf(s, "%s: (0x%lx)",
901 				 trace_event_name(&tu->tp.call),
902 				 entry->vaddr[0]);
903 		data = DATAOF_TRACE_ENTRY(entry, false);
904 	}
905 
906 	if (print_probe_args(s, tu->tp.args, tu->tp.nr_args, data, entry) < 0)
907 		goto out;
908 
909 	trace_seq_putc(s, '\n');
910 
911  out:
912 	return trace_handle_return(s);
913 }
914 
915 typedef bool (*filter_func_t)(struct uprobe_consumer *self,
916 				enum uprobe_filter_ctx ctx,
917 				struct mm_struct *mm);
918 
919 static int
920 probe_event_enable(struct trace_uprobe *tu, struct trace_event_file *file,
921 		   filter_func_t filter)
922 {
923 	bool enabled = trace_probe_is_enabled(&tu->tp);
924 	struct event_file_link *link = NULL;
925 	int ret;
926 
927 	if (file) {
928 		if (tu->tp.flags & TP_FLAG_PROFILE)
929 			return -EINTR;
930 
931 		link = kmalloc(sizeof(*link), GFP_KERNEL);
932 		if (!link)
933 			return -ENOMEM;
934 
935 		link->file = file;
936 		list_add_tail_rcu(&link->list, &tu->tp.files);
937 
938 		tu->tp.flags |= TP_FLAG_TRACE;
939 	} else {
940 		if (tu->tp.flags & TP_FLAG_TRACE)
941 			return -EINTR;
942 
943 		tu->tp.flags |= TP_FLAG_PROFILE;
944 	}
945 
946 	WARN_ON(!uprobe_filter_is_empty(&tu->filter));
947 
948 	if (enabled)
949 		return 0;
950 
951 	ret = uprobe_buffer_enable();
952 	if (ret)
953 		goto err_flags;
954 
955 	tu->consumer.filter = filter;
956 	tu->inode = d_real_inode(tu->path.dentry);
957 	if (tu->ref_ctr_offset) {
958 		ret = uprobe_register_refctr(tu->inode, tu->offset,
959 				tu->ref_ctr_offset, &tu->consumer);
960 	} else {
961 		ret = uprobe_register(tu->inode, tu->offset, &tu->consumer);
962 	}
963 
964 	if (ret)
965 		goto err_buffer;
966 
967 	return 0;
968 
969  err_buffer:
970 	uprobe_buffer_disable();
971 
972  err_flags:
973 	if (file) {
974 		list_del(&link->list);
975 		kfree(link);
976 		tu->tp.flags &= ~TP_FLAG_TRACE;
977 	} else {
978 		tu->tp.flags &= ~TP_FLAG_PROFILE;
979 	}
980 	return ret;
981 }
982 
983 static void
984 probe_event_disable(struct trace_uprobe *tu, struct trace_event_file *file)
985 {
986 	if (!trace_probe_is_enabled(&tu->tp))
987 		return;
988 
989 	if (file) {
990 		struct event_file_link *link;
991 
992 		link = find_event_file_link(&tu->tp, file);
993 		if (!link)
994 			return;
995 
996 		list_del_rcu(&link->list);
997 		/* synchronize with u{,ret}probe_trace_func */
998 		synchronize_rcu();
999 		kfree(link);
1000 
1001 		if (!list_empty(&tu->tp.files))
1002 			return;
1003 	}
1004 
1005 	WARN_ON(!uprobe_filter_is_empty(&tu->filter));
1006 
1007 	uprobe_unregister(tu->inode, tu->offset, &tu->consumer);
1008 	tu->inode = NULL;
1009 	tu->tp.flags &= file ? ~TP_FLAG_TRACE : ~TP_FLAG_PROFILE;
1010 
1011 	uprobe_buffer_disable();
1012 }
1013 
1014 static int uprobe_event_define_fields(struct trace_event_call *event_call)
1015 {
1016 	int ret, size;
1017 	struct uprobe_trace_entry_head field;
1018 	struct trace_uprobe *tu = event_call->data;
1019 
1020 	if (is_ret_probe(tu)) {
1021 		DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_FUNC, 0);
1022 		DEFINE_FIELD(unsigned long, vaddr[1], FIELD_STRING_RETIP, 0);
1023 		size = SIZEOF_TRACE_ENTRY(true);
1024 	} else {
1025 		DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_IP, 0);
1026 		size = SIZEOF_TRACE_ENTRY(false);
1027 	}
1028 
1029 	return traceprobe_define_arg_fields(event_call, size, &tu->tp);
1030 }
1031 
1032 #ifdef CONFIG_PERF_EVENTS
1033 static bool
1034 __uprobe_perf_filter(struct trace_uprobe_filter *filter, struct mm_struct *mm)
1035 {
1036 	struct perf_event *event;
1037 
1038 	if (filter->nr_systemwide)
1039 		return true;
1040 
1041 	list_for_each_entry(event, &filter->perf_events, hw.tp_list) {
1042 		if (event->hw.target->mm == mm)
1043 			return true;
1044 	}
1045 
1046 	return false;
1047 }
1048 
1049 static inline bool
1050 uprobe_filter_event(struct trace_uprobe *tu, struct perf_event *event)
1051 {
1052 	return __uprobe_perf_filter(&tu->filter, event->hw.target->mm);
1053 }
1054 
1055 static int uprobe_perf_close(struct trace_uprobe *tu, struct perf_event *event)
1056 {
1057 	bool done;
1058 
1059 	write_lock(&tu->filter.rwlock);
1060 	if (event->hw.target) {
1061 		list_del(&event->hw.tp_list);
1062 		done = tu->filter.nr_systemwide ||
1063 			(event->hw.target->flags & PF_EXITING) ||
1064 			uprobe_filter_event(tu, event);
1065 	} else {
1066 		tu->filter.nr_systemwide--;
1067 		done = tu->filter.nr_systemwide;
1068 	}
1069 	write_unlock(&tu->filter.rwlock);
1070 
1071 	if (!done)
1072 		return uprobe_apply(tu->inode, tu->offset, &tu->consumer, false);
1073 
1074 	return 0;
1075 }
1076 
1077 static int uprobe_perf_open(struct trace_uprobe *tu, struct perf_event *event)
1078 {
1079 	bool done;
1080 	int err;
1081 
1082 	write_lock(&tu->filter.rwlock);
1083 	if (event->hw.target) {
1084 		/*
1085 		 * event->parent != NULL means copy_process(), we can avoid
1086 		 * uprobe_apply(). current->mm must be probed and we can rely
1087 		 * on dup_mmap() which preserves the already installed bp's.
1088 		 *
1089 		 * attr.enable_on_exec means that exec/mmap will install the
1090 		 * breakpoints we need.
1091 		 */
1092 		done = tu->filter.nr_systemwide ||
1093 			event->parent || event->attr.enable_on_exec ||
1094 			uprobe_filter_event(tu, event);
1095 		list_add(&event->hw.tp_list, &tu->filter.perf_events);
1096 	} else {
1097 		done = tu->filter.nr_systemwide;
1098 		tu->filter.nr_systemwide++;
1099 	}
1100 	write_unlock(&tu->filter.rwlock);
1101 
1102 	err = 0;
1103 	if (!done) {
1104 		err = uprobe_apply(tu->inode, tu->offset, &tu->consumer, true);
1105 		if (err)
1106 			uprobe_perf_close(tu, event);
1107 	}
1108 	return err;
1109 }
1110 
1111 static bool uprobe_perf_filter(struct uprobe_consumer *uc,
1112 				enum uprobe_filter_ctx ctx, struct mm_struct *mm)
1113 {
1114 	struct trace_uprobe *tu;
1115 	int ret;
1116 
1117 	tu = container_of(uc, struct trace_uprobe, consumer);
1118 	read_lock(&tu->filter.rwlock);
1119 	ret = __uprobe_perf_filter(&tu->filter, mm);
1120 	read_unlock(&tu->filter.rwlock);
1121 
1122 	return ret;
1123 }
1124 
1125 static void __uprobe_perf_func(struct trace_uprobe *tu,
1126 			       unsigned long func, struct pt_regs *regs,
1127 			       struct uprobe_cpu_buffer *ucb, int dsize)
1128 {
1129 	struct trace_event_call *call = &tu->tp.call;
1130 	struct uprobe_trace_entry_head *entry;
1131 	struct hlist_head *head;
1132 	void *data;
1133 	int size, esize;
1134 	int rctx;
1135 
1136 	if (bpf_prog_array_valid(call) && !trace_call_bpf(call, regs))
1137 		return;
1138 
1139 	esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1140 
1141 	size = esize + tu->tp.size + dsize;
1142 	size = ALIGN(size + sizeof(u32), sizeof(u64)) - sizeof(u32);
1143 	if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE, "profile buffer not large enough"))
1144 		return;
1145 
1146 	preempt_disable();
1147 	head = this_cpu_ptr(call->perf_events);
1148 	if (hlist_empty(head))
1149 		goto out;
1150 
1151 	entry = perf_trace_buf_alloc(size, NULL, &rctx);
1152 	if (!entry)
1153 		goto out;
1154 
1155 	if (is_ret_probe(tu)) {
1156 		entry->vaddr[0] = func;
1157 		entry->vaddr[1] = instruction_pointer(regs);
1158 		data = DATAOF_TRACE_ENTRY(entry, true);
1159 	} else {
1160 		entry->vaddr[0] = instruction_pointer(regs);
1161 		data = DATAOF_TRACE_ENTRY(entry, false);
1162 	}
1163 
1164 	memcpy(data, ucb->buf, tu->tp.size + dsize);
1165 
1166 	if (size - esize > tu->tp.size + dsize) {
1167 		int len = tu->tp.size + dsize;
1168 
1169 		memset(data + len, 0, size - esize - len);
1170 	}
1171 
1172 	perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1173 			      head, NULL);
1174  out:
1175 	preempt_enable();
1176 }
1177 
1178 /* uprobe profile handler */
1179 static int uprobe_perf_func(struct trace_uprobe *tu, struct pt_regs *regs,
1180 			    struct uprobe_cpu_buffer *ucb, int dsize)
1181 {
1182 	if (!uprobe_perf_filter(&tu->consumer, 0, current->mm))
1183 		return UPROBE_HANDLER_REMOVE;
1184 
1185 	if (!is_ret_probe(tu))
1186 		__uprobe_perf_func(tu, 0, regs, ucb, dsize);
1187 	return 0;
1188 }
1189 
1190 static void uretprobe_perf_func(struct trace_uprobe *tu, unsigned long func,
1191 				struct pt_regs *regs,
1192 				struct uprobe_cpu_buffer *ucb, int dsize)
1193 {
1194 	__uprobe_perf_func(tu, func, regs, ucb, dsize);
1195 }
1196 
1197 int bpf_get_uprobe_info(const struct perf_event *event, u32 *fd_type,
1198 			const char **filename, u64 *probe_offset,
1199 			bool perf_type_tracepoint)
1200 {
1201 	const char *pevent = trace_event_name(event->tp_event);
1202 	const char *group = event->tp_event->class->system;
1203 	struct trace_uprobe *tu;
1204 
1205 	if (perf_type_tracepoint)
1206 		tu = find_probe_event(pevent, group);
1207 	else
1208 		tu = event->tp_event->data;
1209 	if (!tu)
1210 		return -EINVAL;
1211 
1212 	*fd_type = is_ret_probe(tu) ? BPF_FD_TYPE_URETPROBE
1213 				    : BPF_FD_TYPE_UPROBE;
1214 	*filename = tu->filename;
1215 	*probe_offset = tu->offset;
1216 	return 0;
1217 }
1218 #endif	/* CONFIG_PERF_EVENTS */
1219 
1220 static int
1221 trace_uprobe_register(struct trace_event_call *event, enum trace_reg type,
1222 		      void *data)
1223 {
1224 	struct trace_uprobe *tu = event->data;
1225 	struct trace_event_file *file = data;
1226 
1227 	switch (type) {
1228 	case TRACE_REG_REGISTER:
1229 		return probe_event_enable(tu, file, NULL);
1230 
1231 	case TRACE_REG_UNREGISTER:
1232 		probe_event_disable(tu, file);
1233 		return 0;
1234 
1235 #ifdef CONFIG_PERF_EVENTS
1236 	case TRACE_REG_PERF_REGISTER:
1237 		return probe_event_enable(tu, NULL, uprobe_perf_filter);
1238 
1239 	case TRACE_REG_PERF_UNREGISTER:
1240 		probe_event_disable(tu, NULL);
1241 		return 0;
1242 
1243 	case TRACE_REG_PERF_OPEN:
1244 		return uprobe_perf_open(tu, data);
1245 
1246 	case TRACE_REG_PERF_CLOSE:
1247 		return uprobe_perf_close(tu, data);
1248 
1249 #endif
1250 	default:
1251 		return 0;
1252 	}
1253 	return 0;
1254 }
1255 
1256 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs)
1257 {
1258 	struct trace_uprobe *tu;
1259 	struct uprobe_dispatch_data udd;
1260 	struct uprobe_cpu_buffer *ucb;
1261 	int dsize, esize;
1262 	int ret = 0;
1263 
1264 
1265 	tu = container_of(con, struct trace_uprobe, consumer);
1266 	tu->nhit++;
1267 
1268 	udd.tu = tu;
1269 	udd.bp_addr = instruction_pointer(regs);
1270 
1271 	current->utask->vaddr = (unsigned long) &udd;
1272 
1273 	if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1274 		return 0;
1275 
1276 	dsize = __get_data_size(&tu->tp, regs);
1277 	esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1278 
1279 	ucb = uprobe_buffer_get();
1280 	store_trace_args(ucb->buf, &tu->tp, regs, esize, dsize);
1281 
1282 	if (tu->tp.flags & TP_FLAG_TRACE)
1283 		ret |= uprobe_trace_func(tu, regs, ucb, dsize);
1284 
1285 #ifdef CONFIG_PERF_EVENTS
1286 	if (tu->tp.flags & TP_FLAG_PROFILE)
1287 		ret |= uprobe_perf_func(tu, regs, ucb, dsize);
1288 #endif
1289 	uprobe_buffer_put(ucb);
1290 	return ret;
1291 }
1292 
1293 static int uretprobe_dispatcher(struct uprobe_consumer *con,
1294 				unsigned long func, struct pt_regs *regs)
1295 {
1296 	struct trace_uprobe *tu;
1297 	struct uprobe_dispatch_data udd;
1298 	struct uprobe_cpu_buffer *ucb;
1299 	int dsize, esize;
1300 
1301 	tu = container_of(con, struct trace_uprobe, consumer);
1302 
1303 	udd.tu = tu;
1304 	udd.bp_addr = func;
1305 
1306 	current->utask->vaddr = (unsigned long) &udd;
1307 
1308 	if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1309 		return 0;
1310 
1311 	dsize = __get_data_size(&tu->tp, regs);
1312 	esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1313 
1314 	ucb = uprobe_buffer_get();
1315 	store_trace_args(ucb->buf, &tu->tp, regs, esize, dsize);
1316 
1317 	if (tu->tp.flags & TP_FLAG_TRACE)
1318 		uretprobe_trace_func(tu, func, regs, ucb, dsize);
1319 
1320 #ifdef CONFIG_PERF_EVENTS
1321 	if (tu->tp.flags & TP_FLAG_PROFILE)
1322 		uretprobe_perf_func(tu, func, regs, ucb, dsize);
1323 #endif
1324 	uprobe_buffer_put(ucb);
1325 	return 0;
1326 }
1327 
1328 static struct trace_event_functions uprobe_funcs = {
1329 	.trace		= print_uprobe_event
1330 };
1331 
1332 static inline void init_trace_event_call(struct trace_uprobe *tu,
1333 					 struct trace_event_call *call)
1334 {
1335 	INIT_LIST_HEAD(&call->class->fields);
1336 	call->event.funcs = &uprobe_funcs;
1337 	call->class->define_fields = uprobe_event_define_fields;
1338 
1339 	call->flags = TRACE_EVENT_FL_UPROBE;
1340 	call->class->reg = trace_uprobe_register;
1341 	call->data = tu;
1342 }
1343 
1344 static int register_uprobe_event(struct trace_uprobe *tu)
1345 {
1346 	struct trace_event_call *call = &tu->tp.call;
1347 	int ret = 0;
1348 
1349 	init_trace_event_call(tu, call);
1350 
1351 	if (traceprobe_set_print_fmt(&tu->tp, is_ret_probe(tu)) < 0)
1352 		return -ENOMEM;
1353 
1354 	ret = register_trace_event(&call->event);
1355 	if (!ret) {
1356 		kfree(call->print_fmt);
1357 		return -ENODEV;
1358 	}
1359 
1360 	ret = trace_add_event_call(call);
1361 
1362 	if (ret) {
1363 		pr_info("Failed to register uprobe event: %s\n",
1364 			trace_event_name(call));
1365 		kfree(call->print_fmt);
1366 		unregister_trace_event(&call->event);
1367 	}
1368 
1369 	return ret;
1370 }
1371 
1372 static int unregister_uprobe_event(struct trace_uprobe *tu)
1373 {
1374 	int ret;
1375 
1376 	/* tu->event is unregistered in trace_remove_event_call() */
1377 	ret = trace_remove_event_call(&tu->tp.call);
1378 	if (ret)
1379 		return ret;
1380 	kfree(tu->tp.call.print_fmt);
1381 	tu->tp.call.print_fmt = NULL;
1382 	return 0;
1383 }
1384 
1385 #ifdef CONFIG_PERF_EVENTS
1386 struct trace_event_call *
1387 create_local_trace_uprobe(char *name, unsigned long offs,
1388 			  unsigned long ref_ctr_offset, bool is_return)
1389 {
1390 	struct trace_uprobe *tu;
1391 	struct path path;
1392 	int ret;
1393 
1394 	ret = kern_path(name, LOOKUP_FOLLOW, &path);
1395 	if (ret)
1396 		return ERR_PTR(ret);
1397 
1398 	if (!d_is_reg(path.dentry)) {
1399 		path_put(&path);
1400 		return ERR_PTR(-EINVAL);
1401 	}
1402 
1403 	/*
1404 	 * local trace_kprobes are not added to dyn_event, so they are never
1405 	 * searched in find_trace_kprobe(). Therefore, there is no concern of
1406 	 * duplicated name "DUMMY_EVENT" here.
1407 	 */
1408 	tu = alloc_trace_uprobe(UPROBE_EVENT_SYSTEM, "DUMMY_EVENT", 0,
1409 				is_return);
1410 
1411 	if (IS_ERR(tu)) {
1412 		pr_info("Failed to allocate trace_uprobe.(%d)\n",
1413 			(int)PTR_ERR(tu));
1414 		path_put(&path);
1415 		return ERR_CAST(tu);
1416 	}
1417 
1418 	tu->offset = offs;
1419 	tu->path = path;
1420 	tu->ref_ctr_offset = ref_ctr_offset;
1421 	tu->filename = kstrdup(name, GFP_KERNEL);
1422 	init_trace_event_call(tu, &tu->tp.call);
1423 
1424 	if (traceprobe_set_print_fmt(&tu->tp, is_ret_probe(tu)) < 0) {
1425 		ret = -ENOMEM;
1426 		goto error;
1427 	}
1428 
1429 	return &tu->tp.call;
1430 error:
1431 	free_trace_uprobe(tu);
1432 	return ERR_PTR(ret);
1433 }
1434 
1435 void destroy_local_trace_uprobe(struct trace_event_call *event_call)
1436 {
1437 	struct trace_uprobe *tu;
1438 
1439 	tu = container_of(event_call, struct trace_uprobe, tp.call);
1440 
1441 	kfree(tu->tp.call.print_fmt);
1442 	tu->tp.call.print_fmt = NULL;
1443 
1444 	free_trace_uprobe(tu);
1445 }
1446 #endif /* CONFIG_PERF_EVENTS */
1447 
1448 /* Make a trace interface for controling probe points */
1449 static __init int init_uprobe_trace(void)
1450 {
1451 	struct dentry *d_tracer;
1452 	int ret;
1453 
1454 	ret = dyn_event_register(&trace_uprobe_ops);
1455 	if (ret)
1456 		return ret;
1457 
1458 	d_tracer = tracing_init_dentry();
1459 	if (IS_ERR(d_tracer))
1460 		return 0;
1461 
1462 	trace_create_file("uprobe_events", 0644, d_tracer,
1463 				    NULL, &uprobe_events_ops);
1464 	/* Profile interface */
1465 	trace_create_file("uprobe_profile", 0444, d_tracer,
1466 				    NULL, &uprobe_profile_ops);
1467 	return 0;
1468 }
1469 
1470 fs_initcall(init_uprobe_trace);
1471