xref: /linux/kernel/trace/trace_uprobe.c (revision 50f2944009a25bb39a09f2f7bab64a73ce928bef)
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/bpf-cgroup.h>
11 #include <linux/security.h>
12 #include <linux/ctype.h>
13 #include <linux/module.h>
14 #include <linux/uaccess.h>
15 #include <linux/uprobes.h>
16 #include <linux/namei.h>
17 #include <linux/string.h>
18 #include <linux/rculist.h>
19 #include <linux/filter.h>
20 
21 #include "trace_dynevent.h"
22 #include "trace_probe.h"
23 #include "trace_probe_tmpl.h"
24 
25 #define UPROBE_EVENT_SYSTEM	"uprobes"
26 
27 struct uprobe_trace_entry_head {
28 	struct trace_entry	ent;
29 	unsigned long		vaddr[];
30 };
31 
32 #define SIZEOF_TRACE_ENTRY(is_return)			\
33 	(sizeof(struct uprobe_trace_entry_head) +	\
34 	 sizeof(unsigned long) * (is_return ? 2 : 1))
35 
36 #define DATAOF_TRACE_ENTRY(entry, is_return)		\
37 	((void*)(entry) + SIZEOF_TRACE_ENTRY(is_return))
38 
39 static int trace_uprobe_create(const char *raw_command);
40 static int trace_uprobe_show(struct seq_file *m, struct dyn_event *ev);
41 static int trace_uprobe_release(struct dyn_event *ev);
42 static bool trace_uprobe_is_busy(struct dyn_event *ev);
43 static bool trace_uprobe_match(const char *system, const char *event,
44 			int argc, const char **argv, struct dyn_event *ev);
45 
46 static struct dyn_event_operations trace_uprobe_ops = {
47 	.create = trace_uprobe_create,
48 	.show = trace_uprobe_show,
49 	.is_busy = trace_uprobe_is_busy,
50 	.free = trace_uprobe_release,
51 	.match = trace_uprobe_match,
52 };
53 
54 /*
55  * uprobe event core functions
56  */
57 struct trace_uprobe {
58 	struct dyn_event		devent;
59 	struct uprobe_consumer		consumer;
60 	struct path			path;
61 	struct inode			*inode;
62 	char				*filename;
63 	unsigned long			offset;
64 	unsigned long			ref_ctr_offset;
65 	unsigned long			nhit;
66 	struct trace_probe		tp;
67 };
68 
69 static bool is_trace_uprobe(struct dyn_event *ev)
70 {
71 	return ev->ops == &trace_uprobe_ops;
72 }
73 
74 static struct trace_uprobe *to_trace_uprobe(struct dyn_event *ev)
75 {
76 	return container_of(ev, struct trace_uprobe, devent);
77 }
78 
79 /**
80  * for_each_trace_uprobe - iterate over the trace_uprobe list
81  * @pos:	the struct trace_uprobe * for each entry
82  * @dpos:	the struct dyn_event * to use as a loop cursor
83  */
84 #define for_each_trace_uprobe(pos, dpos)	\
85 	for_each_dyn_event(dpos)		\
86 		if (is_trace_uprobe(dpos) && (pos = to_trace_uprobe(dpos)))
87 
88 static int register_uprobe_event(struct trace_uprobe *tu);
89 static int unregister_uprobe_event(struct trace_uprobe *tu);
90 
91 struct uprobe_dispatch_data {
92 	struct trace_uprobe	*tu;
93 	unsigned long		bp_addr;
94 };
95 
96 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs);
97 static int uretprobe_dispatcher(struct uprobe_consumer *con,
98 				unsigned long func, struct pt_regs *regs);
99 
100 #ifdef CONFIG_STACK_GROWSUP
101 static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
102 {
103 	return addr - (n * sizeof(long));
104 }
105 #else
106 static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
107 {
108 	return addr + (n * sizeof(long));
109 }
110 #endif
111 
112 static unsigned long get_user_stack_nth(struct pt_regs *regs, unsigned int n)
113 {
114 	unsigned long ret;
115 	unsigned long addr = user_stack_pointer(regs);
116 
117 	addr = adjust_stack_addr(addr, n);
118 
119 	if (copy_from_user(&ret, (void __force __user *) addr, sizeof(ret)))
120 		return 0;
121 
122 	return ret;
123 }
124 
125 /*
126  * Uprobes-specific fetch functions
127  */
128 static nokprobe_inline int
129 probe_mem_read(void *dest, void *src, size_t size)
130 {
131 	void __user *vaddr = (void __force __user *)src;
132 
133 	return copy_from_user(dest, vaddr, size) ? -EFAULT : 0;
134 }
135 
136 static nokprobe_inline int
137 probe_mem_read_user(void *dest, void *src, size_t size)
138 {
139 	return probe_mem_read(dest, src, size);
140 }
141 
142 /*
143  * Fetch a null-terminated string. Caller MUST set *(u32 *)dest with max
144  * length and relative data location.
145  */
146 static nokprobe_inline int
147 fetch_store_string(unsigned long addr, void *dest, void *base)
148 {
149 	long ret;
150 	u32 loc = *(u32 *)dest;
151 	int maxlen  = get_loc_len(loc);
152 	u8 *dst = get_loc_data(dest, base);
153 	void __user *src = (void __force __user *) addr;
154 
155 	if (unlikely(!maxlen))
156 		return -ENOMEM;
157 
158 	if (addr == FETCH_TOKEN_COMM)
159 		ret = strlcpy(dst, current->comm, maxlen);
160 	else
161 		ret = strncpy_from_user(dst, src, maxlen);
162 	if (ret >= 0) {
163 		if (ret == maxlen)
164 			dst[ret - 1] = '\0';
165 		else
166 			/*
167 			 * Include the terminating null byte. In this case it
168 			 * was copied by strncpy_from_user but not accounted
169 			 * for in ret.
170 			 */
171 			ret++;
172 		*(u32 *)dest = make_data_loc(ret, (void *)dst - base);
173 	}
174 
175 	return ret;
176 }
177 
178 static nokprobe_inline int
179 fetch_store_string_user(unsigned long addr, void *dest, void *base)
180 {
181 	return fetch_store_string(addr, dest, base);
182 }
183 
184 /* Return the length of string -- including null terminal byte */
185 static nokprobe_inline int
186 fetch_store_strlen(unsigned long addr)
187 {
188 	int len;
189 	void __user *vaddr = (void __force __user *) addr;
190 
191 	if (addr == FETCH_TOKEN_COMM)
192 		len = strlen(current->comm) + 1;
193 	else
194 		len = strnlen_user(vaddr, MAX_STRING_SIZE);
195 
196 	return (len > MAX_STRING_SIZE) ? 0 : len;
197 }
198 
199 static nokprobe_inline int
200 fetch_store_strlen_user(unsigned long addr)
201 {
202 	return fetch_store_strlen(addr);
203 }
204 
205 static unsigned long translate_user_vaddr(unsigned long file_offset)
206 {
207 	unsigned long base_addr;
208 	struct uprobe_dispatch_data *udd;
209 
210 	udd = (void *) current->utask->vaddr;
211 
212 	base_addr = udd->bp_addr - udd->tu->offset;
213 	return base_addr + file_offset;
214 }
215 
216 /* Note that we don't verify it, since the code does not come from user space */
217 static int
218 process_fetch_insn(struct fetch_insn *code, void *rec, void *dest,
219 		   void *base)
220 {
221 	struct pt_regs *regs = rec;
222 	unsigned long val;
223 
224 	/* 1st stage: get value from context */
225 	switch (code->op) {
226 	case FETCH_OP_REG:
227 		val = regs_get_register(regs, code->param);
228 		break;
229 	case FETCH_OP_STACK:
230 		val = get_user_stack_nth(regs, code->param);
231 		break;
232 	case FETCH_OP_STACKP:
233 		val = user_stack_pointer(regs);
234 		break;
235 	case FETCH_OP_RETVAL:
236 		val = regs_return_value(regs);
237 		break;
238 	case FETCH_OP_IMM:
239 		val = code->immediate;
240 		break;
241 	case FETCH_OP_COMM:
242 		val = FETCH_TOKEN_COMM;
243 		break;
244 	case FETCH_OP_DATA:
245 		val = (unsigned long)code->data;
246 		break;
247 	case FETCH_OP_FOFFS:
248 		val = translate_user_vaddr(code->immediate);
249 		break;
250 	default:
251 		return -EILSEQ;
252 	}
253 	code++;
254 
255 	return process_fetch_insn_bottom(code, val, dest, base);
256 }
257 NOKPROBE_SYMBOL(process_fetch_insn)
258 
259 static inline void init_trace_uprobe_filter(struct trace_uprobe_filter *filter)
260 {
261 	rwlock_init(&filter->rwlock);
262 	filter->nr_systemwide = 0;
263 	INIT_LIST_HEAD(&filter->perf_events);
264 }
265 
266 static inline bool uprobe_filter_is_empty(struct trace_uprobe_filter *filter)
267 {
268 	return !filter->nr_systemwide && list_empty(&filter->perf_events);
269 }
270 
271 static inline bool is_ret_probe(struct trace_uprobe *tu)
272 {
273 	return tu->consumer.ret_handler != NULL;
274 }
275 
276 static bool trace_uprobe_is_busy(struct dyn_event *ev)
277 {
278 	struct trace_uprobe *tu = to_trace_uprobe(ev);
279 
280 	return trace_probe_is_enabled(&tu->tp);
281 }
282 
283 static bool trace_uprobe_match_command_head(struct trace_uprobe *tu,
284 					    int argc, const char **argv)
285 {
286 	char buf[MAX_ARGSTR_LEN + 1];
287 	int len;
288 
289 	if (!argc)
290 		return true;
291 
292 	len = strlen(tu->filename);
293 	if (strncmp(tu->filename, argv[0], len) || argv[0][len] != ':')
294 		return false;
295 
296 	if (tu->ref_ctr_offset == 0)
297 		snprintf(buf, sizeof(buf), "0x%0*lx",
298 				(int)(sizeof(void *) * 2), tu->offset);
299 	else
300 		snprintf(buf, sizeof(buf), "0x%0*lx(0x%lx)",
301 				(int)(sizeof(void *) * 2), tu->offset,
302 				tu->ref_ctr_offset);
303 	if (strcmp(buf, &argv[0][len + 1]))
304 		return false;
305 
306 	argc--; argv++;
307 
308 	return trace_probe_match_command_args(&tu->tp, argc, argv);
309 }
310 
311 static bool trace_uprobe_match(const char *system, const char *event,
312 			int argc, const char **argv, struct dyn_event *ev)
313 {
314 	struct trace_uprobe *tu = to_trace_uprobe(ev);
315 
316 	return strcmp(trace_probe_name(&tu->tp), event) == 0 &&
317 	   (!system || strcmp(trace_probe_group_name(&tu->tp), system) == 0) &&
318 	   trace_uprobe_match_command_head(tu, argc, argv);
319 }
320 
321 static nokprobe_inline struct trace_uprobe *
322 trace_uprobe_primary_from_call(struct trace_event_call *call)
323 {
324 	struct trace_probe *tp;
325 
326 	tp = trace_probe_primary_from_call(call);
327 	if (WARN_ON_ONCE(!tp))
328 		return NULL;
329 
330 	return container_of(tp, struct trace_uprobe, tp);
331 }
332 
333 /*
334  * Allocate new trace_uprobe and initialize it (including uprobes).
335  */
336 static struct trace_uprobe *
337 alloc_trace_uprobe(const char *group, const char *event, int nargs, bool is_ret)
338 {
339 	struct trace_uprobe *tu;
340 	int ret;
341 
342 	tu = kzalloc(struct_size(tu, tp.args, nargs), GFP_KERNEL);
343 	if (!tu)
344 		return ERR_PTR(-ENOMEM);
345 
346 	ret = trace_probe_init(&tu->tp, event, group, true);
347 	if (ret < 0)
348 		goto error;
349 
350 	dyn_event_init(&tu->devent, &trace_uprobe_ops);
351 	tu->consumer.handler = uprobe_dispatcher;
352 	if (is_ret)
353 		tu->consumer.ret_handler = uretprobe_dispatcher;
354 	init_trace_uprobe_filter(tu->tp.event->filter);
355 	return tu;
356 
357 error:
358 	kfree(tu);
359 
360 	return ERR_PTR(ret);
361 }
362 
363 static void free_trace_uprobe(struct trace_uprobe *tu)
364 {
365 	if (!tu)
366 		return;
367 
368 	path_put(&tu->path);
369 	trace_probe_cleanup(&tu->tp);
370 	kfree(tu->filename);
371 	kfree(tu);
372 }
373 
374 static struct trace_uprobe *find_probe_event(const char *event, const char *group)
375 {
376 	struct dyn_event *pos;
377 	struct trace_uprobe *tu;
378 
379 	for_each_trace_uprobe(tu, pos)
380 		if (strcmp(trace_probe_name(&tu->tp), event) == 0 &&
381 		    strcmp(trace_probe_group_name(&tu->tp), group) == 0)
382 			return tu;
383 
384 	return NULL;
385 }
386 
387 /* Unregister a trace_uprobe and probe_event */
388 static int unregister_trace_uprobe(struct trace_uprobe *tu)
389 {
390 	int ret;
391 
392 	if (trace_probe_has_sibling(&tu->tp))
393 		goto unreg;
394 
395 	/* If there's a reference to the dynamic event */
396 	if (trace_event_dyn_busy(trace_probe_event_call(&tu->tp)))
397 		return -EBUSY;
398 
399 	ret = unregister_uprobe_event(tu);
400 	if (ret)
401 		return ret;
402 
403 unreg:
404 	dyn_event_remove(&tu->devent);
405 	trace_probe_unlink(&tu->tp);
406 	free_trace_uprobe(tu);
407 	return 0;
408 }
409 
410 static bool trace_uprobe_has_same_uprobe(struct trace_uprobe *orig,
411 					 struct trace_uprobe *comp)
412 {
413 	struct trace_probe_event *tpe = orig->tp.event;
414 	struct inode *comp_inode = d_real_inode(comp->path.dentry);
415 	int i;
416 
417 	list_for_each_entry(orig, &tpe->probes, tp.list) {
418 		if (comp_inode != d_real_inode(orig->path.dentry) ||
419 		    comp->offset != orig->offset)
420 			continue;
421 
422 		/*
423 		 * trace_probe_compare_arg_type() ensured that nr_args and
424 		 * each argument name and type are same. Let's compare comm.
425 		 */
426 		for (i = 0; i < orig->tp.nr_args; i++) {
427 			if (strcmp(orig->tp.args[i].comm,
428 				   comp->tp.args[i].comm))
429 				break;
430 		}
431 
432 		if (i == orig->tp.nr_args)
433 			return true;
434 	}
435 
436 	return false;
437 }
438 
439 static int append_trace_uprobe(struct trace_uprobe *tu, struct trace_uprobe *to)
440 {
441 	int ret;
442 
443 	ret = trace_probe_compare_arg_type(&tu->tp, &to->tp);
444 	if (ret) {
445 		/* Note that argument starts index = 2 */
446 		trace_probe_log_set_index(ret + 1);
447 		trace_probe_log_err(0, DIFF_ARG_TYPE);
448 		return -EEXIST;
449 	}
450 	if (trace_uprobe_has_same_uprobe(to, tu)) {
451 		trace_probe_log_set_index(0);
452 		trace_probe_log_err(0, SAME_PROBE);
453 		return -EEXIST;
454 	}
455 
456 	/* Append to existing event */
457 	ret = trace_probe_append(&tu->tp, &to->tp);
458 	if (!ret)
459 		dyn_event_add(&tu->devent, trace_probe_event_call(&tu->tp));
460 
461 	return ret;
462 }
463 
464 /*
465  * Uprobe with multiple reference counter is not allowed. i.e.
466  * If inode and offset matches, reference counter offset *must*
467  * match as well. Though, there is one exception: If user is
468  * replacing old trace_uprobe with new one(same group/event),
469  * then we allow same uprobe with new reference counter as far
470  * as the new one does not conflict with any other existing
471  * ones.
472  */
473 static int validate_ref_ctr_offset(struct trace_uprobe *new)
474 {
475 	struct dyn_event *pos;
476 	struct trace_uprobe *tmp;
477 	struct inode *new_inode = d_real_inode(new->path.dentry);
478 
479 	for_each_trace_uprobe(tmp, pos) {
480 		if (new_inode == d_real_inode(tmp->path.dentry) &&
481 		    new->offset == tmp->offset &&
482 		    new->ref_ctr_offset != tmp->ref_ctr_offset) {
483 			pr_warn("Reference counter offset mismatch.");
484 			return -EINVAL;
485 		}
486 	}
487 	return 0;
488 }
489 
490 /* Register a trace_uprobe and probe_event */
491 static int register_trace_uprobe(struct trace_uprobe *tu)
492 {
493 	struct trace_uprobe *old_tu;
494 	int ret;
495 
496 	mutex_lock(&event_mutex);
497 
498 	ret = validate_ref_ctr_offset(tu);
499 	if (ret)
500 		goto end;
501 
502 	/* register as an event */
503 	old_tu = find_probe_event(trace_probe_name(&tu->tp),
504 				  trace_probe_group_name(&tu->tp));
505 	if (old_tu) {
506 		if (is_ret_probe(tu) != is_ret_probe(old_tu)) {
507 			trace_probe_log_set_index(0);
508 			trace_probe_log_err(0, DIFF_PROBE_TYPE);
509 			ret = -EEXIST;
510 		} else {
511 			ret = append_trace_uprobe(tu, old_tu);
512 		}
513 		goto end;
514 	}
515 
516 	ret = register_uprobe_event(tu);
517 	if (ret) {
518 		if (ret == -EEXIST) {
519 			trace_probe_log_set_index(0);
520 			trace_probe_log_err(0, EVENT_EXIST);
521 		} else
522 			pr_warn("Failed to register probe event(%d)\n", ret);
523 		goto end;
524 	}
525 
526 	dyn_event_add(&tu->devent, trace_probe_event_call(&tu->tp));
527 
528 end:
529 	mutex_unlock(&event_mutex);
530 
531 	return ret;
532 }
533 
534 /*
535  * Argument syntax:
536  *  - Add uprobe: p|r[:[GRP/]EVENT] PATH:OFFSET[%return][(REF)] [FETCHARGS]
537  */
538 static int __trace_uprobe_create(int argc, const char **argv)
539 {
540 	struct trace_uprobe *tu;
541 	const char *event = NULL, *group = UPROBE_EVENT_SYSTEM;
542 	char *arg, *filename, *rctr, *rctr_end, *tmp;
543 	char buf[MAX_EVENT_NAME_LEN];
544 	enum probe_print_type ptype;
545 	struct path path;
546 	unsigned long offset, ref_ctr_offset;
547 	bool is_return = false;
548 	int i, ret;
549 
550 	ret = 0;
551 	ref_ctr_offset = 0;
552 
553 	switch (argv[0][0]) {
554 	case 'r':
555 		is_return = true;
556 		break;
557 	case 'p':
558 		break;
559 	default:
560 		return -ECANCELED;
561 	}
562 
563 	if (argc < 2)
564 		return -ECANCELED;
565 
566 	if (argv[0][1] == ':')
567 		event = &argv[0][2];
568 
569 	if (!strchr(argv[1], '/'))
570 		return -ECANCELED;
571 
572 	filename = kstrdup(argv[1], GFP_KERNEL);
573 	if (!filename)
574 		return -ENOMEM;
575 
576 	/* Find the last occurrence, in case the path contains ':' too. */
577 	arg = strrchr(filename, ':');
578 	if (!arg || !isdigit(arg[1])) {
579 		kfree(filename);
580 		return -ECANCELED;
581 	}
582 
583 	trace_probe_log_init("trace_uprobe", argc, argv);
584 	trace_probe_log_set_index(1);	/* filename is the 2nd argument */
585 
586 	*arg++ = '\0';
587 	ret = kern_path(filename, LOOKUP_FOLLOW, &path);
588 	if (ret) {
589 		trace_probe_log_err(0, FILE_NOT_FOUND);
590 		kfree(filename);
591 		trace_probe_log_clear();
592 		return ret;
593 	}
594 	if (!d_is_reg(path.dentry)) {
595 		trace_probe_log_err(0, NO_REGULAR_FILE);
596 		ret = -EINVAL;
597 		goto fail_address_parse;
598 	}
599 
600 	/* Parse reference counter offset if specified. */
601 	rctr = strchr(arg, '(');
602 	if (rctr) {
603 		rctr_end = strchr(rctr, ')');
604 		if (!rctr_end) {
605 			ret = -EINVAL;
606 			rctr_end = rctr + strlen(rctr);
607 			trace_probe_log_err(rctr_end - filename,
608 					    REFCNT_OPEN_BRACE);
609 			goto fail_address_parse;
610 		} else if (rctr_end[1] != '\0') {
611 			ret = -EINVAL;
612 			trace_probe_log_err(rctr_end + 1 - filename,
613 					    BAD_REFCNT_SUFFIX);
614 			goto fail_address_parse;
615 		}
616 
617 		*rctr++ = '\0';
618 		*rctr_end = '\0';
619 		ret = kstrtoul(rctr, 0, &ref_ctr_offset);
620 		if (ret) {
621 			trace_probe_log_err(rctr - filename, BAD_REFCNT);
622 			goto fail_address_parse;
623 		}
624 	}
625 
626 	/* Check if there is %return suffix */
627 	tmp = strchr(arg, '%');
628 	if (tmp) {
629 		if (!strcmp(tmp, "%return")) {
630 			*tmp = '\0';
631 			is_return = true;
632 		} else {
633 			trace_probe_log_err(tmp - filename, BAD_ADDR_SUFFIX);
634 			ret = -EINVAL;
635 			goto fail_address_parse;
636 		}
637 	}
638 
639 	/* Parse uprobe offset. */
640 	ret = kstrtoul(arg, 0, &offset);
641 	if (ret) {
642 		trace_probe_log_err(arg - filename, BAD_UPROBE_OFFS);
643 		goto fail_address_parse;
644 	}
645 
646 	/* setup a probe */
647 	trace_probe_log_set_index(0);
648 	if (event) {
649 		ret = traceprobe_parse_event_name(&event, &group, buf,
650 						  event - argv[0]);
651 		if (ret)
652 			goto fail_address_parse;
653 	} else {
654 		char *tail;
655 		char *ptr;
656 
657 		tail = kstrdup(kbasename(filename), GFP_KERNEL);
658 		if (!tail) {
659 			ret = -ENOMEM;
660 			goto fail_address_parse;
661 		}
662 
663 		ptr = strpbrk(tail, ".-_");
664 		if (ptr)
665 			*ptr = '\0';
666 
667 		snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_0x%lx", 'p', tail, offset);
668 		event = buf;
669 		kfree(tail);
670 	}
671 
672 	argc -= 2;
673 	argv += 2;
674 
675 	tu = alloc_trace_uprobe(group, event, argc, is_return);
676 	if (IS_ERR(tu)) {
677 		ret = PTR_ERR(tu);
678 		/* This must return -ENOMEM otherwise there is a bug */
679 		WARN_ON_ONCE(ret != -ENOMEM);
680 		goto fail_address_parse;
681 	}
682 	tu->offset = offset;
683 	tu->ref_ctr_offset = ref_ctr_offset;
684 	tu->path = path;
685 	tu->filename = filename;
686 
687 	/* parse arguments */
688 	for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
689 		trace_probe_log_set_index(i + 2);
690 		ret = traceprobe_parse_probe_arg(&tu->tp, i, argv[i],
691 					is_return ? TPARG_FL_RETURN : 0);
692 		if (ret)
693 			goto error;
694 	}
695 
696 	ptype = is_ret_probe(tu) ? PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL;
697 	ret = traceprobe_set_print_fmt(&tu->tp, ptype);
698 	if (ret < 0)
699 		goto error;
700 
701 	ret = register_trace_uprobe(tu);
702 	if (!ret)
703 		goto out;
704 
705 error:
706 	free_trace_uprobe(tu);
707 out:
708 	trace_probe_log_clear();
709 	return ret;
710 
711 fail_address_parse:
712 	trace_probe_log_clear();
713 	path_put(&path);
714 	kfree(filename);
715 
716 	return ret;
717 }
718 
719 int trace_uprobe_create(const char *raw_command)
720 {
721 	return trace_probe_create(raw_command, __trace_uprobe_create);
722 }
723 
724 static int create_or_delete_trace_uprobe(const char *raw_command)
725 {
726 	int ret;
727 
728 	if (raw_command[0] == '-')
729 		return dyn_event_release(raw_command, &trace_uprobe_ops);
730 
731 	ret = trace_uprobe_create(raw_command);
732 	return ret == -ECANCELED ? -EINVAL : ret;
733 }
734 
735 static int trace_uprobe_release(struct dyn_event *ev)
736 {
737 	struct trace_uprobe *tu = to_trace_uprobe(ev);
738 
739 	return unregister_trace_uprobe(tu);
740 }
741 
742 /* Probes listing interfaces */
743 static int trace_uprobe_show(struct seq_file *m, struct dyn_event *ev)
744 {
745 	struct trace_uprobe *tu = to_trace_uprobe(ev);
746 	char c = is_ret_probe(tu) ? 'r' : 'p';
747 	int i;
748 
749 	seq_printf(m, "%c:%s/%s %s:0x%0*lx", c, trace_probe_group_name(&tu->tp),
750 			trace_probe_name(&tu->tp), tu->filename,
751 			(int)(sizeof(void *) * 2), tu->offset);
752 
753 	if (tu->ref_ctr_offset)
754 		seq_printf(m, "(0x%lx)", tu->ref_ctr_offset);
755 
756 	for (i = 0; i < tu->tp.nr_args; i++)
757 		seq_printf(m, " %s=%s", tu->tp.args[i].name, tu->tp.args[i].comm);
758 
759 	seq_putc(m, '\n');
760 	return 0;
761 }
762 
763 static int probes_seq_show(struct seq_file *m, void *v)
764 {
765 	struct dyn_event *ev = v;
766 
767 	if (!is_trace_uprobe(ev))
768 		return 0;
769 
770 	return trace_uprobe_show(m, ev);
771 }
772 
773 static const struct seq_operations probes_seq_op = {
774 	.start  = dyn_event_seq_start,
775 	.next   = dyn_event_seq_next,
776 	.stop   = dyn_event_seq_stop,
777 	.show   = probes_seq_show
778 };
779 
780 static int probes_open(struct inode *inode, struct file *file)
781 {
782 	int ret;
783 
784 	ret = security_locked_down(LOCKDOWN_TRACEFS);
785 	if (ret)
786 		return ret;
787 
788 	if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
789 		ret = dyn_events_release_all(&trace_uprobe_ops);
790 		if (ret)
791 			return ret;
792 	}
793 
794 	return seq_open(file, &probes_seq_op);
795 }
796 
797 static ssize_t probes_write(struct file *file, const char __user *buffer,
798 			    size_t count, loff_t *ppos)
799 {
800 	return trace_parse_run_command(file, buffer, count, ppos,
801 					create_or_delete_trace_uprobe);
802 }
803 
804 static const struct file_operations uprobe_events_ops = {
805 	.owner		= THIS_MODULE,
806 	.open		= probes_open,
807 	.read		= seq_read,
808 	.llseek		= seq_lseek,
809 	.release	= seq_release,
810 	.write		= probes_write,
811 };
812 
813 /* Probes profiling interfaces */
814 static int probes_profile_seq_show(struct seq_file *m, void *v)
815 {
816 	struct dyn_event *ev = v;
817 	struct trace_uprobe *tu;
818 
819 	if (!is_trace_uprobe(ev))
820 		return 0;
821 
822 	tu = to_trace_uprobe(ev);
823 	seq_printf(m, "  %s %-44s %15lu\n", tu->filename,
824 			trace_probe_name(&tu->tp), tu->nhit);
825 	return 0;
826 }
827 
828 static const struct seq_operations profile_seq_op = {
829 	.start  = dyn_event_seq_start,
830 	.next   = dyn_event_seq_next,
831 	.stop   = dyn_event_seq_stop,
832 	.show	= probes_profile_seq_show
833 };
834 
835 static int profile_open(struct inode *inode, struct file *file)
836 {
837 	int ret;
838 
839 	ret = security_locked_down(LOCKDOWN_TRACEFS);
840 	if (ret)
841 		return ret;
842 
843 	return seq_open(file, &profile_seq_op);
844 }
845 
846 static const struct file_operations uprobe_profile_ops = {
847 	.owner		= THIS_MODULE,
848 	.open		= profile_open,
849 	.read		= seq_read,
850 	.llseek		= seq_lseek,
851 	.release	= seq_release,
852 };
853 
854 struct uprobe_cpu_buffer {
855 	struct mutex mutex;
856 	void *buf;
857 };
858 static struct uprobe_cpu_buffer __percpu *uprobe_cpu_buffer;
859 static int uprobe_buffer_refcnt;
860 
861 static int uprobe_buffer_init(void)
862 {
863 	int cpu, err_cpu;
864 
865 	uprobe_cpu_buffer = alloc_percpu(struct uprobe_cpu_buffer);
866 	if (uprobe_cpu_buffer == NULL)
867 		return -ENOMEM;
868 
869 	for_each_possible_cpu(cpu) {
870 		struct page *p = alloc_pages_node(cpu_to_node(cpu),
871 						  GFP_KERNEL, 0);
872 		if (p == NULL) {
873 			err_cpu = cpu;
874 			goto err;
875 		}
876 		per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf = page_address(p);
877 		mutex_init(&per_cpu_ptr(uprobe_cpu_buffer, cpu)->mutex);
878 	}
879 
880 	return 0;
881 
882 err:
883 	for_each_possible_cpu(cpu) {
884 		if (cpu == err_cpu)
885 			break;
886 		free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf);
887 	}
888 
889 	free_percpu(uprobe_cpu_buffer);
890 	return -ENOMEM;
891 }
892 
893 static int uprobe_buffer_enable(void)
894 {
895 	int ret = 0;
896 
897 	BUG_ON(!mutex_is_locked(&event_mutex));
898 
899 	if (uprobe_buffer_refcnt++ == 0) {
900 		ret = uprobe_buffer_init();
901 		if (ret < 0)
902 			uprobe_buffer_refcnt--;
903 	}
904 
905 	return ret;
906 }
907 
908 static void uprobe_buffer_disable(void)
909 {
910 	int cpu;
911 
912 	BUG_ON(!mutex_is_locked(&event_mutex));
913 
914 	if (--uprobe_buffer_refcnt == 0) {
915 		for_each_possible_cpu(cpu)
916 			free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer,
917 							     cpu)->buf);
918 
919 		free_percpu(uprobe_cpu_buffer);
920 		uprobe_cpu_buffer = NULL;
921 	}
922 }
923 
924 static struct uprobe_cpu_buffer *uprobe_buffer_get(void)
925 {
926 	struct uprobe_cpu_buffer *ucb;
927 	int cpu;
928 
929 	cpu = raw_smp_processor_id();
930 	ucb = per_cpu_ptr(uprobe_cpu_buffer, cpu);
931 
932 	/*
933 	 * Use per-cpu buffers for fastest access, but we might migrate
934 	 * so the mutex makes sure we have sole access to it.
935 	 */
936 	mutex_lock(&ucb->mutex);
937 
938 	return ucb;
939 }
940 
941 static void uprobe_buffer_put(struct uprobe_cpu_buffer *ucb)
942 {
943 	mutex_unlock(&ucb->mutex);
944 }
945 
946 static void __uprobe_trace_func(struct trace_uprobe *tu,
947 				unsigned long func, struct pt_regs *regs,
948 				struct uprobe_cpu_buffer *ucb, int dsize,
949 				struct trace_event_file *trace_file)
950 {
951 	struct uprobe_trace_entry_head *entry;
952 	struct trace_event_buffer fbuffer;
953 	void *data;
954 	int size, esize;
955 	struct trace_event_call *call = trace_probe_event_call(&tu->tp);
956 
957 	WARN_ON(call != trace_file->event_call);
958 
959 	if (WARN_ON_ONCE(tu->tp.size + dsize > PAGE_SIZE))
960 		return;
961 
962 	if (trace_trigger_soft_disabled(trace_file))
963 		return;
964 
965 	esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
966 	size = esize + tu->tp.size + dsize;
967 	entry = trace_event_buffer_reserve(&fbuffer, trace_file, size);
968 	if (!entry)
969 		return;
970 
971 	if (is_ret_probe(tu)) {
972 		entry->vaddr[0] = func;
973 		entry->vaddr[1] = instruction_pointer(regs);
974 		data = DATAOF_TRACE_ENTRY(entry, true);
975 	} else {
976 		entry->vaddr[0] = instruction_pointer(regs);
977 		data = DATAOF_TRACE_ENTRY(entry, false);
978 	}
979 
980 	memcpy(data, ucb->buf, tu->tp.size + dsize);
981 
982 	trace_event_buffer_commit(&fbuffer);
983 }
984 
985 /* uprobe handler */
986 static int uprobe_trace_func(struct trace_uprobe *tu, struct pt_regs *regs,
987 			     struct uprobe_cpu_buffer *ucb, int dsize)
988 {
989 	struct event_file_link *link;
990 
991 	if (is_ret_probe(tu))
992 		return 0;
993 
994 	rcu_read_lock();
995 	trace_probe_for_each_link_rcu(link, &tu->tp)
996 		__uprobe_trace_func(tu, 0, regs, ucb, dsize, link->file);
997 	rcu_read_unlock();
998 
999 	return 0;
1000 }
1001 
1002 static void uretprobe_trace_func(struct trace_uprobe *tu, unsigned long func,
1003 				 struct pt_regs *regs,
1004 				 struct uprobe_cpu_buffer *ucb, int dsize)
1005 {
1006 	struct event_file_link *link;
1007 
1008 	rcu_read_lock();
1009 	trace_probe_for_each_link_rcu(link, &tu->tp)
1010 		__uprobe_trace_func(tu, func, regs, ucb, dsize, link->file);
1011 	rcu_read_unlock();
1012 }
1013 
1014 /* Event entry printers */
1015 static enum print_line_t
1016 print_uprobe_event(struct trace_iterator *iter, int flags, struct trace_event *event)
1017 {
1018 	struct uprobe_trace_entry_head *entry;
1019 	struct trace_seq *s = &iter->seq;
1020 	struct trace_uprobe *tu;
1021 	u8 *data;
1022 
1023 	entry = (struct uprobe_trace_entry_head *)iter->ent;
1024 	tu = trace_uprobe_primary_from_call(
1025 		container_of(event, struct trace_event_call, event));
1026 	if (unlikely(!tu))
1027 		goto out;
1028 
1029 	if (is_ret_probe(tu)) {
1030 		trace_seq_printf(s, "%s: (0x%lx <- 0x%lx)",
1031 				 trace_probe_name(&tu->tp),
1032 				 entry->vaddr[1], entry->vaddr[0]);
1033 		data = DATAOF_TRACE_ENTRY(entry, true);
1034 	} else {
1035 		trace_seq_printf(s, "%s: (0x%lx)",
1036 				 trace_probe_name(&tu->tp),
1037 				 entry->vaddr[0]);
1038 		data = DATAOF_TRACE_ENTRY(entry, false);
1039 	}
1040 
1041 	if (print_probe_args(s, tu->tp.args, tu->tp.nr_args, data, entry) < 0)
1042 		goto out;
1043 
1044 	trace_seq_putc(s, '\n');
1045 
1046  out:
1047 	return trace_handle_return(s);
1048 }
1049 
1050 typedef bool (*filter_func_t)(struct uprobe_consumer *self,
1051 				enum uprobe_filter_ctx ctx,
1052 				struct mm_struct *mm);
1053 
1054 static int trace_uprobe_enable(struct trace_uprobe *tu, filter_func_t filter)
1055 {
1056 	int ret;
1057 
1058 	tu->consumer.filter = filter;
1059 	tu->inode = d_real_inode(tu->path.dentry);
1060 
1061 	if (tu->ref_ctr_offset)
1062 		ret = uprobe_register_refctr(tu->inode, tu->offset,
1063 				tu->ref_ctr_offset, &tu->consumer);
1064 	else
1065 		ret = uprobe_register(tu->inode, tu->offset, &tu->consumer);
1066 
1067 	if (ret)
1068 		tu->inode = NULL;
1069 
1070 	return ret;
1071 }
1072 
1073 static void __probe_event_disable(struct trace_probe *tp)
1074 {
1075 	struct trace_uprobe *tu;
1076 
1077 	tu = container_of(tp, struct trace_uprobe, tp);
1078 	WARN_ON(!uprobe_filter_is_empty(tu->tp.event->filter));
1079 
1080 	list_for_each_entry(tu, trace_probe_probe_list(tp), tp.list) {
1081 		if (!tu->inode)
1082 			continue;
1083 
1084 		uprobe_unregister(tu->inode, tu->offset, &tu->consumer);
1085 		tu->inode = NULL;
1086 	}
1087 }
1088 
1089 static int probe_event_enable(struct trace_event_call *call,
1090 			struct trace_event_file *file, filter_func_t filter)
1091 {
1092 	struct trace_probe *tp;
1093 	struct trace_uprobe *tu;
1094 	bool enabled;
1095 	int ret;
1096 
1097 	tp = trace_probe_primary_from_call(call);
1098 	if (WARN_ON_ONCE(!tp))
1099 		return -ENODEV;
1100 	enabled = trace_probe_is_enabled(tp);
1101 
1102 	/* This may also change "enabled" state */
1103 	if (file) {
1104 		if (trace_probe_test_flag(tp, TP_FLAG_PROFILE))
1105 			return -EINTR;
1106 
1107 		ret = trace_probe_add_file(tp, file);
1108 		if (ret < 0)
1109 			return ret;
1110 	} else {
1111 		if (trace_probe_test_flag(tp, TP_FLAG_TRACE))
1112 			return -EINTR;
1113 
1114 		trace_probe_set_flag(tp, TP_FLAG_PROFILE);
1115 	}
1116 
1117 	tu = container_of(tp, struct trace_uprobe, tp);
1118 	WARN_ON(!uprobe_filter_is_empty(tu->tp.event->filter));
1119 
1120 	if (enabled)
1121 		return 0;
1122 
1123 	ret = uprobe_buffer_enable();
1124 	if (ret)
1125 		goto err_flags;
1126 
1127 	list_for_each_entry(tu, trace_probe_probe_list(tp), tp.list) {
1128 		ret = trace_uprobe_enable(tu, filter);
1129 		if (ret) {
1130 			__probe_event_disable(tp);
1131 			goto err_buffer;
1132 		}
1133 	}
1134 
1135 	return 0;
1136 
1137  err_buffer:
1138 	uprobe_buffer_disable();
1139 
1140  err_flags:
1141 	if (file)
1142 		trace_probe_remove_file(tp, file);
1143 	else
1144 		trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
1145 
1146 	return ret;
1147 }
1148 
1149 static void probe_event_disable(struct trace_event_call *call,
1150 				struct trace_event_file *file)
1151 {
1152 	struct trace_probe *tp;
1153 
1154 	tp = trace_probe_primary_from_call(call);
1155 	if (WARN_ON_ONCE(!tp))
1156 		return;
1157 
1158 	if (!trace_probe_is_enabled(tp))
1159 		return;
1160 
1161 	if (file) {
1162 		if (trace_probe_remove_file(tp, file) < 0)
1163 			return;
1164 
1165 		if (trace_probe_is_enabled(tp))
1166 			return;
1167 	} else
1168 		trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
1169 
1170 	__probe_event_disable(tp);
1171 	uprobe_buffer_disable();
1172 }
1173 
1174 static int uprobe_event_define_fields(struct trace_event_call *event_call)
1175 {
1176 	int ret, size;
1177 	struct uprobe_trace_entry_head field;
1178 	struct trace_uprobe *tu;
1179 
1180 	tu = trace_uprobe_primary_from_call(event_call);
1181 	if (unlikely(!tu))
1182 		return -ENODEV;
1183 
1184 	if (is_ret_probe(tu)) {
1185 		DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_FUNC, 0);
1186 		DEFINE_FIELD(unsigned long, vaddr[1], FIELD_STRING_RETIP, 0);
1187 		size = SIZEOF_TRACE_ENTRY(true);
1188 	} else {
1189 		DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_IP, 0);
1190 		size = SIZEOF_TRACE_ENTRY(false);
1191 	}
1192 
1193 	return traceprobe_define_arg_fields(event_call, size, &tu->tp);
1194 }
1195 
1196 #ifdef CONFIG_PERF_EVENTS
1197 static bool
1198 __uprobe_perf_filter(struct trace_uprobe_filter *filter, struct mm_struct *mm)
1199 {
1200 	struct perf_event *event;
1201 
1202 	if (filter->nr_systemwide)
1203 		return true;
1204 
1205 	list_for_each_entry(event, &filter->perf_events, hw.tp_list) {
1206 		if (event->hw.target->mm == mm)
1207 			return true;
1208 	}
1209 
1210 	return false;
1211 }
1212 
1213 static inline bool
1214 trace_uprobe_filter_event(struct trace_uprobe_filter *filter,
1215 			  struct perf_event *event)
1216 {
1217 	return __uprobe_perf_filter(filter, event->hw.target->mm);
1218 }
1219 
1220 static bool trace_uprobe_filter_remove(struct trace_uprobe_filter *filter,
1221 				       struct perf_event *event)
1222 {
1223 	bool done;
1224 
1225 	write_lock(&filter->rwlock);
1226 	if (event->hw.target) {
1227 		list_del(&event->hw.tp_list);
1228 		done = filter->nr_systemwide ||
1229 			(event->hw.target->flags & PF_EXITING) ||
1230 			trace_uprobe_filter_event(filter, event);
1231 	} else {
1232 		filter->nr_systemwide--;
1233 		done = filter->nr_systemwide;
1234 	}
1235 	write_unlock(&filter->rwlock);
1236 
1237 	return done;
1238 }
1239 
1240 /* This returns true if the filter always covers target mm */
1241 static bool trace_uprobe_filter_add(struct trace_uprobe_filter *filter,
1242 				    struct perf_event *event)
1243 {
1244 	bool done;
1245 
1246 	write_lock(&filter->rwlock);
1247 	if (event->hw.target) {
1248 		/*
1249 		 * event->parent != NULL means copy_process(), we can avoid
1250 		 * uprobe_apply(). current->mm must be probed and we can rely
1251 		 * on dup_mmap() which preserves the already installed bp's.
1252 		 *
1253 		 * attr.enable_on_exec means that exec/mmap will install the
1254 		 * breakpoints we need.
1255 		 */
1256 		done = filter->nr_systemwide ||
1257 			event->parent || event->attr.enable_on_exec ||
1258 			trace_uprobe_filter_event(filter, event);
1259 		list_add(&event->hw.tp_list, &filter->perf_events);
1260 	} else {
1261 		done = filter->nr_systemwide;
1262 		filter->nr_systemwide++;
1263 	}
1264 	write_unlock(&filter->rwlock);
1265 
1266 	return done;
1267 }
1268 
1269 static int uprobe_perf_close(struct trace_event_call *call,
1270 			     struct perf_event *event)
1271 {
1272 	struct trace_probe *tp;
1273 	struct trace_uprobe *tu;
1274 	int ret = 0;
1275 
1276 	tp = trace_probe_primary_from_call(call);
1277 	if (WARN_ON_ONCE(!tp))
1278 		return -ENODEV;
1279 
1280 	tu = container_of(tp, struct trace_uprobe, tp);
1281 	if (trace_uprobe_filter_remove(tu->tp.event->filter, event))
1282 		return 0;
1283 
1284 	list_for_each_entry(tu, trace_probe_probe_list(tp), tp.list) {
1285 		ret = uprobe_apply(tu->inode, tu->offset, &tu->consumer, false);
1286 		if (ret)
1287 			break;
1288 	}
1289 
1290 	return ret;
1291 }
1292 
1293 static int uprobe_perf_open(struct trace_event_call *call,
1294 			    struct perf_event *event)
1295 {
1296 	struct trace_probe *tp;
1297 	struct trace_uprobe *tu;
1298 	int err = 0;
1299 
1300 	tp = trace_probe_primary_from_call(call);
1301 	if (WARN_ON_ONCE(!tp))
1302 		return -ENODEV;
1303 
1304 	tu = container_of(tp, struct trace_uprobe, tp);
1305 	if (trace_uprobe_filter_add(tu->tp.event->filter, event))
1306 		return 0;
1307 
1308 	list_for_each_entry(tu, trace_probe_probe_list(tp), tp.list) {
1309 		err = uprobe_apply(tu->inode, tu->offset, &tu->consumer, true);
1310 		if (err) {
1311 			uprobe_perf_close(call, event);
1312 			break;
1313 		}
1314 	}
1315 
1316 	return err;
1317 }
1318 
1319 static bool uprobe_perf_filter(struct uprobe_consumer *uc,
1320 				enum uprobe_filter_ctx ctx, struct mm_struct *mm)
1321 {
1322 	struct trace_uprobe_filter *filter;
1323 	struct trace_uprobe *tu;
1324 	int ret;
1325 
1326 	tu = container_of(uc, struct trace_uprobe, consumer);
1327 	filter = tu->tp.event->filter;
1328 
1329 	read_lock(&filter->rwlock);
1330 	ret = __uprobe_perf_filter(filter, mm);
1331 	read_unlock(&filter->rwlock);
1332 
1333 	return ret;
1334 }
1335 
1336 static void __uprobe_perf_func(struct trace_uprobe *tu,
1337 			       unsigned long func, struct pt_regs *regs,
1338 			       struct uprobe_cpu_buffer *ucb, int dsize)
1339 {
1340 	struct trace_event_call *call = trace_probe_event_call(&tu->tp);
1341 	struct uprobe_trace_entry_head *entry;
1342 	struct hlist_head *head;
1343 	void *data;
1344 	int size, esize;
1345 	int rctx;
1346 
1347 	if (bpf_prog_array_valid(call)) {
1348 		u32 ret;
1349 
1350 		ret = bpf_prog_run_array_sleepable(call->prog_array, regs, bpf_prog_run);
1351 		if (!ret)
1352 			return;
1353 	}
1354 
1355 	esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1356 
1357 	size = esize + tu->tp.size + dsize;
1358 	size = ALIGN(size + sizeof(u32), sizeof(u64)) - sizeof(u32);
1359 	if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE, "profile buffer not large enough"))
1360 		return;
1361 
1362 	preempt_disable();
1363 	head = this_cpu_ptr(call->perf_events);
1364 	if (hlist_empty(head))
1365 		goto out;
1366 
1367 	entry = perf_trace_buf_alloc(size, NULL, &rctx);
1368 	if (!entry)
1369 		goto out;
1370 
1371 	if (is_ret_probe(tu)) {
1372 		entry->vaddr[0] = func;
1373 		entry->vaddr[1] = instruction_pointer(regs);
1374 		data = DATAOF_TRACE_ENTRY(entry, true);
1375 	} else {
1376 		entry->vaddr[0] = instruction_pointer(regs);
1377 		data = DATAOF_TRACE_ENTRY(entry, false);
1378 	}
1379 
1380 	memcpy(data, ucb->buf, tu->tp.size + dsize);
1381 
1382 	if (size - esize > tu->tp.size + dsize) {
1383 		int len = tu->tp.size + dsize;
1384 
1385 		memset(data + len, 0, size - esize - len);
1386 	}
1387 
1388 	perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1389 			      head, NULL);
1390  out:
1391 	preempt_enable();
1392 }
1393 
1394 /* uprobe profile handler */
1395 static int uprobe_perf_func(struct trace_uprobe *tu, struct pt_regs *regs,
1396 			    struct uprobe_cpu_buffer *ucb, int dsize)
1397 {
1398 	if (!uprobe_perf_filter(&tu->consumer, 0, current->mm))
1399 		return UPROBE_HANDLER_REMOVE;
1400 
1401 	if (!is_ret_probe(tu))
1402 		__uprobe_perf_func(tu, 0, regs, ucb, dsize);
1403 	return 0;
1404 }
1405 
1406 static void uretprobe_perf_func(struct trace_uprobe *tu, unsigned long func,
1407 				struct pt_regs *regs,
1408 				struct uprobe_cpu_buffer *ucb, int dsize)
1409 {
1410 	__uprobe_perf_func(tu, func, regs, ucb, dsize);
1411 }
1412 
1413 int bpf_get_uprobe_info(const struct perf_event *event, u32 *fd_type,
1414 			const char **filename, u64 *probe_offset,
1415 			bool perf_type_tracepoint)
1416 {
1417 	const char *pevent = trace_event_name(event->tp_event);
1418 	const char *group = event->tp_event->class->system;
1419 	struct trace_uprobe *tu;
1420 
1421 	if (perf_type_tracepoint)
1422 		tu = find_probe_event(pevent, group);
1423 	else
1424 		tu = trace_uprobe_primary_from_call(event->tp_event);
1425 	if (!tu)
1426 		return -EINVAL;
1427 
1428 	*fd_type = is_ret_probe(tu) ? BPF_FD_TYPE_URETPROBE
1429 				    : BPF_FD_TYPE_UPROBE;
1430 	*filename = tu->filename;
1431 	*probe_offset = tu->offset;
1432 	return 0;
1433 }
1434 #endif	/* CONFIG_PERF_EVENTS */
1435 
1436 static int
1437 trace_uprobe_register(struct trace_event_call *event, enum trace_reg type,
1438 		      void *data)
1439 {
1440 	struct trace_event_file *file = data;
1441 
1442 	switch (type) {
1443 	case TRACE_REG_REGISTER:
1444 		return probe_event_enable(event, file, NULL);
1445 
1446 	case TRACE_REG_UNREGISTER:
1447 		probe_event_disable(event, file);
1448 		return 0;
1449 
1450 #ifdef CONFIG_PERF_EVENTS
1451 	case TRACE_REG_PERF_REGISTER:
1452 		return probe_event_enable(event, NULL, uprobe_perf_filter);
1453 
1454 	case TRACE_REG_PERF_UNREGISTER:
1455 		probe_event_disable(event, NULL);
1456 		return 0;
1457 
1458 	case TRACE_REG_PERF_OPEN:
1459 		return uprobe_perf_open(event, data);
1460 
1461 	case TRACE_REG_PERF_CLOSE:
1462 		return uprobe_perf_close(event, data);
1463 
1464 #endif
1465 	default:
1466 		return 0;
1467 	}
1468 }
1469 
1470 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs)
1471 {
1472 	struct trace_uprobe *tu;
1473 	struct uprobe_dispatch_data udd;
1474 	struct uprobe_cpu_buffer *ucb;
1475 	int dsize, esize;
1476 	int ret = 0;
1477 
1478 
1479 	tu = container_of(con, struct trace_uprobe, consumer);
1480 	tu->nhit++;
1481 
1482 	udd.tu = tu;
1483 	udd.bp_addr = instruction_pointer(regs);
1484 
1485 	current->utask->vaddr = (unsigned long) &udd;
1486 
1487 	if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1488 		return 0;
1489 
1490 	dsize = __get_data_size(&tu->tp, regs);
1491 	esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1492 
1493 	ucb = uprobe_buffer_get();
1494 	store_trace_args(ucb->buf, &tu->tp, regs, esize, dsize);
1495 
1496 	if (trace_probe_test_flag(&tu->tp, TP_FLAG_TRACE))
1497 		ret |= uprobe_trace_func(tu, regs, ucb, dsize);
1498 
1499 #ifdef CONFIG_PERF_EVENTS
1500 	if (trace_probe_test_flag(&tu->tp, TP_FLAG_PROFILE))
1501 		ret |= uprobe_perf_func(tu, regs, ucb, dsize);
1502 #endif
1503 	uprobe_buffer_put(ucb);
1504 	return ret;
1505 }
1506 
1507 static int uretprobe_dispatcher(struct uprobe_consumer *con,
1508 				unsigned long func, struct pt_regs *regs)
1509 {
1510 	struct trace_uprobe *tu;
1511 	struct uprobe_dispatch_data udd;
1512 	struct uprobe_cpu_buffer *ucb;
1513 	int dsize, esize;
1514 
1515 	tu = container_of(con, struct trace_uprobe, consumer);
1516 
1517 	udd.tu = tu;
1518 	udd.bp_addr = func;
1519 
1520 	current->utask->vaddr = (unsigned long) &udd;
1521 
1522 	if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1523 		return 0;
1524 
1525 	dsize = __get_data_size(&tu->tp, regs);
1526 	esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1527 
1528 	ucb = uprobe_buffer_get();
1529 	store_trace_args(ucb->buf, &tu->tp, regs, esize, dsize);
1530 
1531 	if (trace_probe_test_flag(&tu->tp, TP_FLAG_TRACE))
1532 		uretprobe_trace_func(tu, func, regs, ucb, dsize);
1533 
1534 #ifdef CONFIG_PERF_EVENTS
1535 	if (trace_probe_test_flag(&tu->tp, TP_FLAG_PROFILE))
1536 		uretprobe_perf_func(tu, func, regs, ucb, dsize);
1537 #endif
1538 	uprobe_buffer_put(ucb);
1539 	return 0;
1540 }
1541 
1542 static struct trace_event_functions uprobe_funcs = {
1543 	.trace		= print_uprobe_event
1544 };
1545 
1546 static struct trace_event_fields uprobe_fields_array[] = {
1547 	{ .type = TRACE_FUNCTION_TYPE,
1548 	  .define_fields = uprobe_event_define_fields },
1549 	{}
1550 };
1551 
1552 static inline void init_trace_event_call(struct trace_uprobe *tu)
1553 {
1554 	struct trace_event_call *call = trace_probe_event_call(&tu->tp);
1555 	call->event.funcs = &uprobe_funcs;
1556 	call->class->fields_array = uprobe_fields_array;
1557 
1558 	call->flags = TRACE_EVENT_FL_UPROBE | TRACE_EVENT_FL_CAP_ANY;
1559 	call->class->reg = trace_uprobe_register;
1560 }
1561 
1562 static int register_uprobe_event(struct trace_uprobe *tu)
1563 {
1564 	init_trace_event_call(tu);
1565 
1566 	return trace_probe_register_event_call(&tu->tp);
1567 }
1568 
1569 static int unregister_uprobe_event(struct trace_uprobe *tu)
1570 {
1571 	return trace_probe_unregister_event_call(&tu->tp);
1572 }
1573 
1574 #ifdef CONFIG_PERF_EVENTS
1575 struct trace_event_call *
1576 create_local_trace_uprobe(char *name, unsigned long offs,
1577 			  unsigned long ref_ctr_offset, bool is_return)
1578 {
1579 	enum probe_print_type ptype;
1580 	struct trace_uprobe *tu;
1581 	struct path path;
1582 	int ret;
1583 
1584 	ret = kern_path(name, LOOKUP_FOLLOW, &path);
1585 	if (ret)
1586 		return ERR_PTR(ret);
1587 
1588 	if (!d_is_reg(path.dentry)) {
1589 		path_put(&path);
1590 		return ERR_PTR(-EINVAL);
1591 	}
1592 
1593 	/*
1594 	 * local trace_kprobes are not added to dyn_event, so they are never
1595 	 * searched in find_trace_kprobe(). Therefore, there is no concern of
1596 	 * duplicated name "DUMMY_EVENT" here.
1597 	 */
1598 	tu = alloc_trace_uprobe(UPROBE_EVENT_SYSTEM, "DUMMY_EVENT", 0,
1599 				is_return);
1600 
1601 	if (IS_ERR(tu)) {
1602 		pr_info("Failed to allocate trace_uprobe.(%d)\n",
1603 			(int)PTR_ERR(tu));
1604 		path_put(&path);
1605 		return ERR_CAST(tu);
1606 	}
1607 
1608 	tu->offset = offs;
1609 	tu->path = path;
1610 	tu->ref_ctr_offset = ref_ctr_offset;
1611 	tu->filename = kstrdup(name, GFP_KERNEL);
1612 	if (!tu->filename) {
1613 		ret = -ENOMEM;
1614 		goto error;
1615 	}
1616 
1617 	init_trace_event_call(tu);
1618 
1619 	ptype = is_ret_probe(tu) ? PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL;
1620 	if (traceprobe_set_print_fmt(&tu->tp, ptype) < 0) {
1621 		ret = -ENOMEM;
1622 		goto error;
1623 	}
1624 
1625 	return trace_probe_event_call(&tu->tp);
1626 error:
1627 	free_trace_uprobe(tu);
1628 	return ERR_PTR(ret);
1629 }
1630 
1631 void destroy_local_trace_uprobe(struct trace_event_call *event_call)
1632 {
1633 	struct trace_uprobe *tu;
1634 
1635 	tu = trace_uprobe_primary_from_call(event_call);
1636 
1637 	free_trace_uprobe(tu);
1638 }
1639 #endif /* CONFIG_PERF_EVENTS */
1640 
1641 /* Make a trace interface for controlling probe points */
1642 static __init int init_uprobe_trace(void)
1643 {
1644 	int ret;
1645 
1646 	ret = dyn_event_register(&trace_uprobe_ops);
1647 	if (ret)
1648 		return ret;
1649 
1650 	ret = tracing_init_dentry();
1651 	if (ret)
1652 		return 0;
1653 
1654 	trace_create_file("uprobe_events", TRACE_MODE_WRITE, NULL,
1655 				    NULL, &uprobe_events_ops);
1656 	/* Profile interface */
1657 	trace_create_file("uprobe_profile", TRACE_MODE_READ, NULL,
1658 				    NULL, &uprobe_profile_ops);
1659 	return 0;
1660 }
1661 
1662 fs_initcall(init_uprobe_trace);
1663