xref: /linux/kernel/trace/trace_uprobe.c (revision 7d8d6ad659c02ed5d2387777194c22e8e81dbb2b)
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/cleanup.h>
12 #include <linux/ctype.h>
13 #include <linux/filter.h>
14 #include <linux/module.h>
15 #include <linux/namei.h>
16 #include <linux/percpu.h>
17 #include <linux/rculist.h>
18 #include <linux/security.h>
19 #include <linux/string.h>
20 #include <linux/uaccess.h>
21 #include <linux/uprobes.h>
22 
23 #include "trace.h"
24 #include "trace_dynevent.h"
25 #include "trace_probe.h"
26 #include "trace_probe_tmpl.h"
27 
28 #define UPROBE_EVENT_SYSTEM	"uprobes"
29 
30 struct uprobe_trace_entry_head {
31 	struct trace_entry	ent;
32 	unsigned long		vaddr[];
33 };
34 
35 #define SIZEOF_TRACE_ENTRY(is_return)			\
36 	(sizeof(struct uprobe_trace_entry_head) +	\
37 	 sizeof(unsigned long) * (is_return ? 2 : 1))
38 
39 #define DATAOF_TRACE_ENTRY(entry, is_return)		\
40 	((void*)(entry) + SIZEOF_TRACE_ENTRY(is_return))
41 
42 static int trace_uprobe_create(const char *raw_command);
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 			int argc, const char **argv, 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 uprobe_consumer		consumer;
63 	struct path			path;
64 	char				*filename;
65 	struct uprobe			*uprobe;
66 	unsigned long			offset;
67 	unsigned long			ref_ctr_offset;
68 	unsigned long __percpu		*nhits;
69 	struct trace_probe		tp;
70 };
71 
72 static bool is_trace_uprobe(struct dyn_event *ev)
73 {
74 	return ev->ops == &trace_uprobe_ops;
75 }
76 
77 static struct trace_uprobe *to_trace_uprobe(struct dyn_event *ev)
78 {
79 	return container_of(ev, struct trace_uprobe, devent);
80 }
81 
82 /**
83  * for_each_trace_uprobe - iterate over the trace_uprobe list
84  * @pos:	the struct trace_uprobe * for each entry
85  * @dpos:	the struct dyn_event * to use as a loop cursor
86  */
87 #define for_each_trace_uprobe(pos, dpos)	\
88 	for_each_dyn_event(dpos)		\
89 		if (is_trace_uprobe(dpos) && (pos = to_trace_uprobe(dpos)))
90 
91 static int register_uprobe_event(struct trace_uprobe *tu);
92 static int unregister_uprobe_event(struct trace_uprobe *tu);
93 
94 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs,
95 			     __u64 *data);
96 static int uretprobe_dispatcher(struct uprobe_consumer *con,
97 				unsigned long func, struct pt_regs *regs,
98 				__u64 *data);
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 = strscpy(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 	} else
174 		*(u32 *)dest = make_data_loc(0, (void *)dst - base);
175 
176 	return ret;
177 }
178 
179 static nokprobe_inline int
180 fetch_store_string_user(unsigned long addr, void *dest, void *base)
181 {
182 	return fetch_store_string(addr, dest, base);
183 }
184 
185 /* Return the length of string -- including null terminal byte */
186 static nokprobe_inline int
187 fetch_store_strlen(unsigned long addr)
188 {
189 	int len;
190 	void __user *vaddr = (void __force __user *) addr;
191 
192 	if (addr == FETCH_TOKEN_COMM)
193 		len = strlen(current->comm) + 1;
194 	else
195 		len = strnlen_user(vaddr, MAX_STRING_SIZE);
196 
197 	return (len > MAX_STRING_SIZE) ? 0 : len;
198 }
199 
200 static nokprobe_inline int
201 fetch_store_strlen_user(unsigned long addr)
202 {
203 	return fetch_store_strlen(addr);
204 }
205 
206 static unsigned long translate_user_vaddr(unsigned long file_offset)
207 {
208 	unsigned long base_addr;
209 	struct uprobe_dispatch_data *udd;
210 
211 	udd = (void *) current->utask->vaddr;
212 
213 	base_addr = udd->bp_addr - udd->tu->offset;
214 	return base_addr + file_offset;
215 }
216 
217 /* Note that we don't verify it, since the code does not come from user space */
218 static int
219 process_fetch_insn(struct fetch_insn *code, void *rec, void *edata,
220 		   void *dest, void *base)
221 {
222 	struct pt_regs *regs = rec;
223 	unsigned long val;
224 	int ret;
225 
226 	/* 1st stage: get value from context */
227 	switch (code->op) {
228 	case FETCH_OP_REG:
229 		val = regs_get_register(regs, code->param);
230 		break;
231 	case FETCH_OP_STACK:
232 		val = get_user_stack_nth(regs, code->param);
233 		break;
234 	case FETCH_OP_STACKP:
235 		val = user_stack_pointer(regs);
236 		break;
237 	case FETCH_OP_RETVAL:
238 		val = regs_return_value(regs);
239 		break;
240 	case FETCH_OP_COMM:
241 		val = FETCH_TOKEN_COMM;
242 		break;
243 	case FETCH_OP_FOFFS:
244 		val = translate_user_vaddr(code->immediate);
245 		break;
246 	default:
247 		ret = process_common_fetch_insn(code, &val);
248 		if (ret < 0)
249 			return ret;
250 	}
251 	code++;
252 
253 	return process_fetch_insn_bottom(code, val, dest, base);
254 }
255 NOKPROBE_SYMBOL(process_fetch_insn)
256 
257 static inline void init_trace_uprobe_filter(struct trace_uprobe_filter *filter)
258 {
259 	rwlock_init(&filter->rwlock);
260 	filter->nr_systemwide = 0;
261 	INIT_LIST_HEAD(&filter->perf_events);
262 }
263 
264 static inline bool uprobe_filter_is_empty(struct trace_uprobe_filter *filter)
265 {
266 	return !filter->nr_systemwide && list_empty(&filter->perf_events);
267 }
268 
269 static inline bool is_ret_probe(struct trace_uprobe *tu)
270 {
271 	return tu->consumer.ret_handler != NULL;
272 }
273 
274 static bool trace_uprobe_is_busy(struct dyn_event *ev)
275 {
276 	struct trace_uprobe *tu = to_trace_uprobe(ev);
277 
278 	return trace_probe_is_enabled(&tu->tp);
279 }
280 
281 static bool trace_uprobe_match_command_head(struct trace_uprobe *tu,
282 					    int argc, const char **argv)
283 {
284 	char buf[MAX_ARGSTR_LEN + 1];
285 	int len;
286 
287 	if (!argc)
288 		return true;
289 
290 	len = strlen(tu->filename);
291 	if (strncmp(tu->filename, argv[0], len) || argv[0][len] != ':')
292 		return false;
293 
294 	if (tu->ref_ctr_offset == 0)
295 		snprintf(buf, sizeof(buf), "0x%0*lx",
296 				(int)(sizeof(void *) * 2), tu->offset);
297 	else
298 		snprintf(buf, sizeof(buf), "0x%0*lx(0x%lx)",
299 				(int)(sizeof(void *) * 2), tu->offset,
300 				tu->ref_ctr_offset);
301 	if (strcmp(buf, &argv[0][len + 1]))
302 		return false;
303 
304 	argc--; argv++;
305 
306 	return trace_probe_match_command_args(&tu->tp, argc, argv);
307 }
308 
309 static bool trace_uprobe_match(const char *system, const char *event,
310 			int argc, const char **argv, struct dyn_event *ev)
311 {
312 	struct trace_uprobe *tu = to_trace_uprobe(ev);
313 
314 	return (event[0] == '\0' ||
315 		strcmp(trace_probe_name(&tu->tp), event) == 0) &&
316 	   (!system || strcmp(trace_probe_group_name(&tu->tp), system) == 0) &&
317 	   trace_uprobe_match_command_head(tu, argc, argv);
318 }
319 
320 static nokprobe_inline struct trace_uprobe *
321 trace_uprobe_primary_from_call(struct trace_event_call *call)
322 {
323 	struct trace_probe *tp;
324 
325 	tp = trace_probe_primary_from_call(call);
326 	if (WARN_ON_ONCE(!tp))
327 		return NULL;
328 
329 	return container_of(tp, struct trace_uprobe, tp);
330 }
331 
332 /*
333  * Allocate new trace_uprobe and initialize it (including uprobes).
334  */
335 static struct trace_uprobe *
336 alloc_trace_uprobe(const char *group, const char *event, int nargs, bool is_ret)
337 {
338 	struct trace_uprobe *tu;
339 	int ret;
340 
341 	tu = kzalloc_flex(*tu, tp.args, nargs);
342 	if (!tu)
343 		return ERR_PTR(-ENOMEM);
344 
345 	tu->nhits = alloc_percpu(unsigned long);
346 	if (!tu->nhits) {
347 		ret = -ENOMEM;
348 		goto error;
349 	}
350 
351 	ret = trace_probe_init(&tu->tp, event, group, true, nargs);
352 	if (ret < 0)
353 		goto error;
354 
355 	dyn_event_init(&tu->devent, &trace_uprobe_ops);
356 	tu->consumer.handler = uprobe_dispatcher;
357 	if (is_ret)
358 		tu->consumer.ret_handler = uretprobe_dispatcher;
359 	init_trace_uprobe_filter(tu->tp.event->filter);
360 	return tu;
361 
362 error:
363 	free_percpu(tu->nhits);
364 	kfree(tu);
365 
366 	return ERR_PTR(ret);
367 }
368 
369 static void free_trace_uprobe(struct trace_uprobe *tu)
370 {
371 	if (!tu)
372 		return;
373 
374 	path_put(&tu->path);
375 	trace_probe_cleanup(&tu->tp);
376 	kfree(tu->filename);
377 	free_percpu(tu->nhits);
378 	kfree(tu);
379 }
380 
381 static struct trace_uprobe *find_probe_event(const char *event, const char *group)
382 {
383 	struct dyn_event *pos;
384 	struct trace_uprobe *tu;
385 
386 	for_each_trace_uprobe(tu, pos)
387 		if (strcmp(trace_probe_name(&tu->tp), event) == 0 &&
388 		    strcmp(trace_probe_group_name(&tu->tp), group) == 0)
389 			return tu;
390 
391 	return NULL;
392 }
393 
394 /* Unregister a trace_uprobe and probe_event */
395 static int unregister_trace_uprobe(struct trace_uprobe *tu)
396 {
397 	int ret;
398 
399 	if (trace_probe_has_sibling(&tu->tp))
400 		goto unreg;
401 
402 	/* If there's a reference to the dynamic event */
403 	if (trace_event_dyn_busy(trace_probe_event_call(&tu->tp)))
404 		return -EBUSY;
405 
406 	ret = unregister_uprobe_event(tu);
407 	if (ret)
408 		return ret;
409 
410 unreg:
411 	dyn_event_remove(&tu->devent);
412 	trace_probe_unlink(&tu->tp);
413 	free_trace_uprobe(tu);
414 	return 0;
415 }
416 
417 static bool trace_uprobe_has_same_uprobe(struct trace_uprobe *orig,
418 					 struct trace_uprobe *comp)
419 {
420 	struct trace_probe_event *tpe = orig->tp.event;
421 	struct inode *comp_inode = d_real_inode(comp->path.dentry);
422 	int i;
423 
424 	list_for_each_entry(orig, &tpe->probes, tp.list) {
425 		if (comp_inode != d_real_inode(orig->path.dentry) ||
426 		    comp->offset != orig->offset)
427 			continue;
428 
429 		/*
430 		 * trace_probe_compare_arg_type() ensured that nr_args and
431 		 * each argument name and type are same. Let's compare comm.
432 		 */
433 		for (i = 0; i < orig->tp.nr_args; i++) {
434 			if (strcmp(orig->tp.args[i].comm,
435 				   comp->tp.args[i].comm))
436 				break;
437 		}
438 
439 		if (i == orig->tp.nr_args)
440 			return true;
441 	}
442 
443 	return false;
444 }
445 
446 static int append_trace_uprobe(struct trace_uprobe *tu, struct trace_uprobe *to)
447 {
448 	int ret;
449 
450 	ret = trace_probe_compare_arg_type(&tu->tp, &to->tp);
451 	if (ret) {
452 		/* Note that argument starts index = 2 */
453 		trace_probe_log_set_index(ret + 1);
454 		trace_probe_log_err(0, DIFF_ARG_TYPE);
455 		return -EEXIST;
456 	}
457 	if (trace_uprobe_has_same_uprobe(to, tu)) {
458 		trace_probe_log_set_index(0);
459 		trace_probe_log_err(0, SAME_PROBE);
460 		return -EEXIST;
461 	}
462 
463 	/* Append to existing event */
464 	ret = trace_probe_append(&tu->tp, &to->tp);
465 	if (!ret)
466 		dyn_event_add(&tu->devent, trace_probe_event_call(&tu->tp));
467 
468 	return ret;
469 }
470 
471 /*
472  * Uprobe with multiple reference counter is not allowed. i.e.
473  * If inode and offset matches, reference counter offset *must*
474  * match as well. Though, there is one exception: If user is
475  * replacing old trace_uprobe with new one(same group/event),
476  * then we allow same uprobe with new reference counter as far
477  * as the new one does not conflict with any other existing
478  * ones.
479  */
480 static int validate_ref_ctr_offset(struct trace_uprobe *new)
481 {
482 	struct dyn_event *pos;
483 	struct trace_uprobe *tmp;
484 	struct inode *new_inode = d_real_inode(new->path.dentry);
485 
486 	for_each_trace_uprobe(tmp, pos) {
487 		if (new_inode == d_real_inode(tmp->path.dentry) &&
488 		    new->offset == tmp->offset &&
489 		    new->ref_ctr_offset != tmp->ref_ctr_offset) {
490 			pr_warn("Reference counter offset mismatch.");
491 			return -EINVAL;
492 		}
493 	}
494 	return 0;
495 }
496 
497 /* Register a trace_uprobe and probe_event */
498 static int register_trace_uprobe(struct trace_uprobe *tu)
499 {
500 	struct trace_uprobe *old_tu;
501 	int ret;
502 
503 	guard(mutex)(&event_mutex);
504 
505 	ret = validate_ref_ctr_offset(tu);
506 	if (ret)
507 		return ret;
508 
509 	/* register as an event */
510 	old_tu = find_probe_event(trace_probe_name(&tu->tp),
511 				  trace_probe_group_name(&tu->tp));
512 	if (old_tu) {
513 		if (is_ret_probe(tu) != is_ret_probe(old_tu)) {
514 			trace_probe_log_set_index(0);
515 			trace_probe_log_err(0, DIFF_PROBE_TYPE);
516 			return -EEXIST;
517 		}
518 		return append_trace_uprobe(tu, old_tu);
519 	}
520 
521 	ret = register_uprobe_event(tu);
522 	if (ret) {
523 		if (ret == -EEXIST) {
524 			trace_probe_log_set_index(0);
525 			trace_probe_log_err(0, EVENT_EXIST);
526 		} else
527 			pr_warn("Failed to register probe event(%d)\n", ret);
528 		return ret;
529 	}
530 
531 	dyn_event_add(&tu->devent, trace_probe_event_call(&tu->tp));
532 
533 	return ret;
534 }
535 
536 DEFINE_FREE(free_trace_uprobe, struct trace_uprobe *, if (_T) free_trace_uprobe(_T))
537 
538 /*
539  * Argument syntax:
540  *  - Add uprobe: p|r[:[GRP/][EVENT]] PATH:OFFSET[%return][(REF)] [FETCHARGS]
541  */
542 static int __trace_uprobe_create(int argc, const char **argv)
543 {
544 	struct traceprobe_parse_context *ctx __free(traceprobe_parse_context) = NULL;
545 	struct trace_uprobe *tu __free(free_trace_uprobe) = NULL;
546 	const char *trlog __free(trace_probe_log_clear) = NULL;
547 	const char *event = NULL, *group = UPROBE_EVENT_SYSTEM;
548 	struct path path __free(path_put) = {};
549 	unsigned long offset, ref_ctr_offset;
550 	char *filename __free(kfree) = NULL;
551 	char *arg, *rctr, *rctr_end, *tmp;
552 	char *gbuf __free(kfree) = NULL;
553 	char *buf __free(kfree) = NULL;
554 	enum probe_print_type ptype;
555 	bool is_return = false;
556 	int i, ret;
557 
558 	ref_ctr_offset = 0;
559 
560 	switch (argv[0][0]) {
561 	case 'r':
562 		is_return = true;
563 		break;
564 	case 'p':
565 		break;
566 	default:
567 		return -ECANCELED;
568 	}
569 
570 	if (argc < 2)
571 		return -ECANCELED;
572 
573 	trlog = trace_probe_log_init("trace_uprobe", argc, argv);
574 
575 	if (argc - 2 > MAX_TRACE_ARGS) {
576 		trace_probe_log_set_index(2);
577 		trace_probe_log_err(0, TOO_MANY_ARGS);
578 		return -E2BIG;
579 	}
580 
581 	if (argv[0][1] == ':')
582 		event = &argv[0][2];
583 
584 	if (!strchr(argv[1], '/'))
585 		return -ECANCELED;
586 
587 	filename = kstrdup(argv[1], GFP_KERNEL);
588 	if (!filename)
589 		return -ENOMEM;
590 
591 	/* Find the last occurrence, in case the path contains ':' too. */
592 	arg = strrchr(filename, ':');
593 	if (!arg || !isdigit(arg[1]))
594 		return -ECANCELED;
595 
596 	trace_probe_log_set_index(1);	/* filename is the 2nd argument */
597 
598 	*arg++ = '\0';
599 	ret = kern_path(filename, LOOKUP_FOLLOW, &path);
600 	if (ret) {
601 		trace_probe_log_err(0, FILE_NOT_FOUND);
602 		return ret;
603 	}
604 	if (!d_is_reg(path.dentry)) {
605 		trace_probe_log_err(0, NO_REGULAR_FILE);
606 		return -EINVAL;
607 	}
608 
609 	/* Parse reference counter offset if specified. */
610 	rctr = strchr(arg, '(');
611 	if (rctr) {
612 		rctr_end = strchr(rctr, ')');
613 		if (!rctr_end) {
614 			rctr_end = rctr + strlen(rctr);
615 			trace_probe_log_err(rctr_end - filename,
616 					    REFCNT_OPEN_BRACE);
617 			return -EINVAL;
618 		} else if (rctr_end[1] != '\0') {
619 			trace_probe_log_err(rctr_end + 1 - filename,
620 					    BAD_REFCNT_SUFFIX);
621 			return -EINVAL;
622 		}
623 
624 		*rctr++ = '\0';
625 		*rctr_end = '\0';
626 		ret = kstrtoul(rctr, 0, &ref_ctr_offset);
627 		if (ret) {
628 			trace_probe_log_err(rctr - filename, BAD_REFCNT);
629 			return ret;
630 		}
631 	}
632 
633 	/* Check if there is %return suffix */
634 	tmp = strchr(arg, '%');
635 	if (tmp) {
636 		if (!strcmp(tmp, "%return")) {
637 			*tmp = '\0';
638 			is_return = true;
639 		} else {
640 			trace_probe_log_err(tmp - filename, BAD_ADDR_SUFFIX);
641 			return -EINVAL;
642 		}
643 	}
644 
645 	/* Parse uprobe offset. */
646 	ret = kstrtoul(arg, 0, &offset);
647 	if (ret) {
648 		trace_probe_log_err(arg - filename, BAD_UPROBE_OFFS);
649 		return ret;
650 	}
651 
652 	/* setup a probe */
653 	trace_probe_log_set_index(0);
654 	if (event) {
655 		gbuf = kmalloc(MAX_EVENT_NAME_LEN, GFP_KERNEL);
656 		if (!gbuf)
657 			return -ENOMEM;
658 
659 		ret = traceprobe_parse_event_name(&event, &group, gbuf,
660 						  event - argv[0]);
661 		if (ret)
662 			return ret;
663 	}
664 
665 	if (!event) {
666 		char *tail;
667 		char *ptr;
668 
669 		tail = kstrdup(kbasename(filename), GFP_KERNEL);
670 		if (!tail)
671 			return -ENOMEM;
672 
673 		ptr = strpbrk(tail, ".-_");
674 		if (ptr)
675 			*ptr = '\0';
676 
677 		buf = kmalloc(MAX_EVENT_NAME_LEN, GFP_KERNEL);
678 		if (!buf)
679 			return -ENOMEM;
680 		snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_0x%lx", 'p', tail, offset);
681 		event = buf;
682 		kfree(tail);
683 	}
684 
685 	argc -= 2;
686 	argv += 2;
687 
688 	tu = alloc_trace_uprobe(group, event, argc, is_return);
689 	if (IS_ERR(tu)) {
690 		ret = PTR_ERR(tu);
691 		/* This must return -ENOMEM otherwise there is a bug */
692 		WARN_ON_ONCE(ret != -ENOMEM);
693 		return ret;
694 	}
695 	tu->offset = offset;
696 	tu->ref_ctr_offset = ref_ctr_offset;
697 	tu->path = path;
698 	/* Clear @path so that it will not freed by path_put() */
699 	memset(&path, 0, sizeof(path));
700 	tu->filename = no_free_ptr(filename);
701 
702 	ctx = kzalloc_obj(*ctx);
703 	if (!ctx)
704 		return -ENOMEM;
705 	ctx->flags = (is_return ? TPARG_FL_RETURN : 0) | TPARG_FL_USER;
706 
707 	/* parse arguments */
708 	for (i = 0; i < argc; i++) {
709 		trace_probe_log_set_index(i + 2);
710 		ret = traceprobe_parse_probe_arg(&tu->tp, i, argv[i], ctx);
711 		if (ret)
712 			return ret;
713 	}
714 
715 	ptype = is_ret_probe(tu) ? PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL;
716 	ret = traceprobe_set_print_fmt(&tu->tp, ptype);
717 	if (ret < 0)
718 		return ret;
719 
720 	ret = register_trace_uprobe(tu);
721 	if (!ret)
722 		tu = NULL;
723 
724 	return ret;
725 }
726 
727 int trace_uprobe_create(const char *raw_command)
728 {
729 	return trace_probe_create(raw_command, __trace_uprobe_create);
730 }
731 
732 static int create_or_delete_trace_uprobe(const char *raw_command)
733 {
734 	int ret;
735 
736 	if (raw_command[0] == '-')
737 		return dyn_event_release(raw_command, &trace_uprobe_ops);
738 
739 	ret = dyn_event_create(raw_command, &trace_uprobe_ops);
740 	return ret == -ECANCELED ? -EINVAL : ret;
741 }
742 
743 static int trace_uprobe_release(struct dyn_event *ev)
744 {
745 	struct trace_uprobe *tu = to_trace_uprobe(ev);
746 
747 	return unregister_trace_uprobe(tu);
748 }
749 
750 /* Probes listing interfaces */
751 static int trace_uprobe_show(struct seq_file *m, struct dyn_event *ev)
752 {
753 	struct trace_uprobe *tu = to_trace_uprobe(ev);
754 	char c = is_ret_probe(tu) ? 'r' : 'p';
755 	int i;
756 
757 	seq_printf(m, "%c:%s/%s %s:0x%0*lx", c, trace_probe_group_name(&tu->tp),
758 			trace_probe_name(&tu->tp), tu->filename,
759 			(int)(sizeof(void *) * 2), tu->offset);
760 
761 	if (tu->ref_ctr_offset)
762 		seq_printf(m, "(0x%lx)", tu->ref_ctr_offset);
763 
764 	for (i = 0; i < tu->tp.nr_args; i++)
765 		seq_printf(m, " %s=%s", tu->tp.args[i].name, tu->tp.args[i].comm);
766 
767 	seq_putc(m, '\n');
768 
769 	trace_probe_dump_args(m, &tu->tp);
770 
771 	return 0;
772 }
773 
774 static int probes_seq_show(struct seq_file *m, void *v)
775 {
776 	struct dyn_event *ev = v;
777 
778 	if (!is_trace_uprobe(ev))
779 		return 0;
780 
781 	return trace_uprobe_show(m, ev);
782 }
783 
784 static const struct seq_operations probes_seq_op = {
785 	.start  = dyn_event_seq_start,
786 	.next   = dyn_event_seq_next,
787 	.stop   = dyn_event_seq_stop,
788 	.show   = probes_seq_show
789 };
790 
791 static int probes_open(struct inode *inode, struct file *file)
792 {
793 	int ret;
794 
795 	ret = security_locked_down(LOCKDOWN_TRACEFS);
796 	if (ret)
797 		return ret;
798 
799 	if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
800 		ret = dyn_events_release_all(&trace_uprobe_ops);
801 		if (ret)
802 			return ret;
803 	}
804 
805 	return seq_open(file, &probes_seq_op);
806 }
807 
808 static ssize_t probes_write(struct file *file, const char __user *buffer,
809 			    size_t count, loff_t *ppos)
810 {
811 	return trace_parse_run_command(file, buffer, count, ppos,
812 					create_or_delete_trace_uprobe);
813 }
814 
815 static const struct file_operations uprobe_events_ops = {
816 	.owner		= THIS_MODULE,
817 	.open		= probes_open,
818 	.read		= seq_read,
819 	.llseek		= seq_lseek,
820 	.release	= seq_release,
821 	.write		= probes_write,
822 };
823 
824 /* Probes profiling interfaces */
825 static int probes_profile_seq_show(struct seq_file *m, void *v)
826 {
827 	struct dyn_event *ev = v;
828 	struct trace_uprobe *tu;
829 	unsigned long nhits;
830 	int cpu;
831 
832 	if (!is_trace_uprobe(ev))
833 		return 0;
834 
835 	tu = to_trace_uprobe(ev);
836 
837 	nhits = 0;
838 	for_each_possible_cpu(cpu) {
839 		nhits += per_cpu(*tu->nhits, cpu);
840 	}
841 
842 	seq_printf(m, "  %s %-44s %15lu\n", tu->filename,
843 		   trace_probe_name(&tu->tp), nhits);
844 	return 0;
845 }
846 
847 static const struct seq_operations profile_seq_op = {
848 	.start  = dyn_event_seq_start,
849 	.next   = dyn_event_seq_next,
850 	.stop   = dyn_event_seq_stop,
851 	.show	= probes_profile_seq_show
852 };
853 
854 static int profile_open(struct inode *inode, struct file *file)
855 {
856 	int ret;
857 
858 	ret = security_locked_down(LOCKDOWN_TRACEFS);
859 	if (ret)
860 		return ret;
861 
862 	return seq_open(file, &profile_seq_op);
863 }
864 
865 static const struct file_operations uprobe_profile_ops = {
866 	.owner		= THIS_MODULE,
867 	.open		= profile_open,
868 	.read		= seq_read,
869 	.llseek		= seq_lseek,
870 	.release	= seq_release,
871 };
872 
873 struct uprobe_cpu_buffer {
874 	struct mutex mutex;
875 	void *buf;
876 	int dsize;
877 };
878 static struct uprobe_cpu_buffer __percpu *uprobe_cpu_buffer;
879 static int uprobe_buffer_refcnt;
880 #define MAX_UCB_BUFFER_SIZE PAGE_SIZE
881 
882 static int uprobe_buffer_init(void)
883 {
884 	int cpu, err_cpu;
885 
886 	uprobe_cpu_buffer = alloc_percpu(struct uprobe_cpu_buffer);
887 	if (uprobe_cpu_buffer == NULL)
888 		return -ENOMEM;
889 
890 	for_each_possible_cpu(cpu) {
891 		struct page *p = alloc_pages_node(cpu_to_node(cpu),
892 						  GFP_KERNEL, 0);
893 		if (p == NULL) {
894 			err_cpu = cpu;
895 			goto err;
896 		}
897 		per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf = page_address(p);
898 		mutex_init(&per_cpu_ptr(uprobe_cpu_buffer, cpu)->mutex);
899 	}
900 
901 	return 0;
902 
903 err:
904 	for_each_possible_cpu(cpu) {
905 		if (cpu == err_cpu)
906 			break;
907 		free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf);
908 	}
909 
910 	free_percpu(uprobe_cpu_buffer);
911 	return -ENOMEM;
912 }
913 
914 static int uprobe_buffer_enable(void)
915 {
916 	int ret = 0;
917 
918 	lockdep_assert_held(&event_mutex);
919 
920 	if (uprobe_buffer_refcnt++ == 0) {
921 		ret = uprobe_buffer_init();
922 		if (ret < 0)
923 			uprobe_buffer_refcnt--;
924 	}
925 
926 	return ret;
927 }
928 
929 static void uprobe_buffer_disable(void)
930 {
931 	int cpu;
932 
933 	lockdep_assert_held(&event_mutex);
934 
935 	if (--uprobe_buffer_refcnt == 0) {
936 		for_each_possible_cpu(cpu)
937 			free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer,
938 							     cpu)->buf);
939 
940 		free_percpu(uprobe_cpu_buffer);
941 		uprobe_cpu_buffer = NULL;
942 	}
943 }
944 
945 static struct uprobe_cpu_buffer *uprobe_buffer_get(void)
946 {
947 	struct uprobe_cpu_buffer *ucb;
948 	int cpu;
949 
950 	cpu = raw_smp_processor_id();
951 	ucb = per_cpu_ptr(uprobe_cpu_buffer, cpu);
952 
953 	/*
954 	 * Use per-cpu buffers for fastest access, but we might migrate
955 	 * so the mutex makes sure we have sole access to it.
956 	 */
957 	mutex_lock(&ucb->mutex);
958 
959 	return ucb;
960 }
961 
962 static void uprobe_buffer_put(struct uprobe_cpu_buffer *ucb)
963 {
964 	if (!ucb)
965 		return;
966 	mutex_unlock(&ucb->mutex);
967 }
968 
969 static struct uprobe_cpu_buffer *prepare_uprobe_buffer(struct trace_uprobe *tu,
970 						       struct pt_regs *regs,
971 						       struct uprobe_cpu_buffer **ucbp)
972 {
973 	struct uprobe_cpu_buffer *ucb;
974 	int dsize, esize;
975 
976 	if (*ucbp)
977 		return *ucbp;
978 
979 	esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
980 	dsize = __get_data_size(&tu->tp, regs, NULL);
981 
982 	ucb = uprobe_buffer_get();
983 	ucb->dsize = tu->tp.size + dsize;
984 
985 	BUILD_BUG_ON(MAX_UCB_BUFFER_SIZE < MAX_PROBE_EVENT_SIZE);
986 	if (WARN_ON_ONCE(ucb->dsize > MAX_UCB_BUFFER_SIZE)) {
987 		ucb->dsize = MAX_UCB_BUFFER_SIZE;
988 		dsize = MAX_UCB_BUFFER_SIZE - tu->tp.size;
989 	}
990 
991 	store_trace_args(ucb->buf, &tu->tp, regs, NULL, esize, dsize);
992 
993 	*ucbp = ucb;
994 	return ucb;
995 }
996 
997 static void __uprobe_trace_func(struct trace_uprobe *tu,
998 				unsigned long func, struct pt_regs *regs,
999 				struct uprobe_cpu_buffer *ucb,
1000 				struct trace_event_file *trace_file)
1001 {
1002 	struct uprobe_trace_entry_head *entry;
1003 	struct trace_event_buffer fbuffer;
1004 	void *data;
1005 	int size, esize;
1006 	struct trace_event_call *call = trace_probe_event_call(&tu->tp);
1007 
1008 	WARN_ON(call != trace_file->event_call);
1009 
1010 	if (trace_trigger_soft_disabled(trace_file))
1011 		return;
1012 
1013 	esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1014 	size = esize + ucb->dsize;
1015 	entry = trace_event_buffer_reserve(&fbuffer, trace_file, size);
1016 	if (!entry)
1017 		return;
1018 
1019 	if (is_ret_probe(tu)) {
1020 		entry->vaddr[0] = func;
1021 		entry->vaddr[1] = instruction_pointer(regs);
1022 		data = DATAOF_TRACE_ENTRY(entry, true);
1023 	} else {
1024 		entry->vaddr[0] = instruction_pointer(regs);
1025 		data = DATAOF_TRACE_ENTRY(entry, false);
1026 	}
1027 
1028 	memcpy(data, ucb->buf, ucb->dsize);
1029 
1030 	trace_event_buffer_commit(&fbuffer);
1031 }
1032 
1033 /* uprobe handler */
1034 static int uprobe_trace_func(struct trace_uprobe *tu, struct pt_regs *regs,
1035 			     struct uprobe_cpu_buffer **ucbp)
1036 {
1037 	struct event_file_link *link;
1038 	struct uprobe_cpu_buffer *ucb;
1039 
1040 	if (is_ret_probe(tu))
1041 		return 0;
1042 
1043 	ucb = prepare_uprobe_buffer(tu, regs, ucbp);
1044 
1045 	rcu_read_lock();
1046 	trace_probe_for_each_link_rcu(link, &tu->tp)
1047 		__uprobe_trace_func(tu, 0, regs, ucb, link->file);
1048 	rcu_read_unlock();
1049 
1050 	return 0;
1051 }
1052 
1053 static void uretprobe_trace_func(struct trace_uprobe *tu, unsigned long func,
1054 				 struct pt_regs *regs,
1055 				 struct uprobe_cpu_buffer **ucbp)
1056 {
1057 	struct event_file_link *link;
1058 	struct uprobe_cpu_buffer *ucb;
1059 
1060 	ucb = prepare_uprobe_buffer(tu, regs, ucbp);
1061 
1062 	rcu_read_lock();
1063 	trace_probe_for_each_link_rcu(link, &tu->tp)
1064 		__uprobe_trace_func(tu, func, regs, ucb, link->file);
1065 	rcu_read_unlock();
1066 }
1067 
1068 /* Event entry printers */
1069 static enum print_line_t
1070 print_uprobe_event(struct trace_iterator *iter, int flags, struct trace_event *event)
1071 {
1072 	struct uprobe_trace_entry_head *entry;
1073 	struct trace_seq *s = &iter->seq;
1074 	struct trace_uprobe *tu;
1075 	u8 *data;
1076 
1077 	entry = (struct uprobe_trace_entry_head *)iter->ent;
1078 	tu = trace_uprobe_primary_from_call(
1079 		container_of(event, struct trace_event_call, event));
1080 	if (unlikely(!tu))
1081 		goto out;
1082 
1083 	if (is_ret_probe(tu)) {
1084 		trace_seq_printf(s, "%s: (0x%lx <- 0x%lx)",
1085 				 trace_probe_name(&tu->tp),
1086 				 entry->vaddr[1], entry->vaddr[0]);
1087 		data = DATAOF_TRACE_ENTRY(entry, true);
1088 	} else {
1089 		trace_seq_printf(s, "%s: (0x%lx)",
1090 				 trace_probe_name(&tu->tp),
1091 				 entry->vaddr[0]);
1092 		data = DATAOF_TRACE_ENTRY(entry, false);
1093 	}
1094 
1095 	if (trace_probe_print_args(s, tu->tp.args, tu->tp.nr_args, data, entry) < 0)
1096 		goto out;
1097 
1098 	trace_seq_putc(s, '\n');
1099 
1100  out:
1101 	return trace_handle_return(s);
1102 }
1103 
1104 typedef bool (*filter_func_t)(struct uprobe_consumer *self, struct mm_struct *mm);
1105 
1106 static int trace_uprobe_enable(struct trace_uprobe *tu, filter_func_t filter)
1107 {
1108 	struct inode *inode = d_real_inode(tu->path.dentry);
1109 	struct uprobe *uprobe;
1110 
1111 	tu->consumer.filter = filter;
1112 	uprobe = uprobe_register(inode, tu->offset, tu->ref_ctr_offset, &tu->consumer);
1113 	if (IS_ERR(uprobe))
1114 		return PTR_ERR(uprobe);
1115 
1116 	tu->uprobe = uprobe;
1117 	return 0;
1118 }
1119 
1120 static void __probe_event_disable(struct trace_probe *tp)
1121 {
1122 	struct trace_uprobe *tu;
1123 	bool sync = false;
1124 
1125 	tu = container_of(tp, struct trace_uprobe, tp);
1126 	WARN_ON(!uprobe_filter_is_empty(tu->tp.event->filter));
1127 
1128 	list_for_each_entry(tu, trace_probe_probe_list(tp), tp.list) {
1129 		if (!tu->uprobe)
1130 			continue;
1131 
1132 		uprobe_unregister_nosync(tu->uprobe, &tu->consumer);
1133 		sync = true;
1134 		tu->uprobe = NULL;
1135 	}
1136 	if (sync)
1137 		uprobe_unregister_sync();
1138 }
1139 
1140 static int probe_event_enable(struct trace_event_call *call,
1141 			struct trace_event_file *file, filter_func_t filter)
1142 {
1143 	struct trace_probe *tp;
1144 	struct trace_uprobe *tu;
1145 	bool enabled;
1146 	int ret;
1147 
1148 	tp = trace_probe_primary_from_call(call);
1149 	if (WARN_ON_ONCE(!tp))
1150 		return -ENODEV;
1151 	enabled = trace_probe_is_enabled(tp);
1152 
1153 	/* This may also change "enabled" state */
1154 	if (file) {
1155 		if (trace_probe_test_flag(tp, TP_FLAG_PROFILE))
1156 			return -EINTR;
1157 
1158 		ret = trace_probe_add_file(tp, file);
1159 		if (ret < 0)
1160 			return ret;
1161 	} else {
1162 		if (trace_probe_test_flag(tp, TP_FLAG_TRACE))
1163 			return -EINTR;
1164 
1165 		trace_probe_set_flag(tp, TP_FLAG_PROFILE);
1166 	}
1167 
1168 	tu = container_of(tp, struct trace_uprobe, tp);
1169 	WARN_ON(!uprobe_filter_is_empty(tu->tp.event->filter));
1170 
1171 	if (enabled)
1172 		return 0;
1173 
1174 	ret = uprobe_buffer_enable();
1175 	if (ret)
1176 		goto err_flags;
1177 
1178 	list_for_each_entry(tu, trace_probe_probe_list(tp), tp.list) {
1179 		ret = trace_uprobe_enable(tu, filter);
1180 		if (ret) {
1181 			__probe_event_disable(tp);
1182 			goto err_buffer;
1183 		}
1184 	}
1185 
1186 	return 0;
1187 
1188  err_buffer:
1189 	uprobe_buffer_disable();
1190 
1191  err_flags:
1192 	if (file)
1193 		trace_probe_remove_file(tp, file);
1194 	else
1195 		trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
1196 
1197 	return ret;
1198 }
1199 
1200 static void probe_event_disable(struct trace_event_call *call,
1201 				struct trace_event_file *file)
1202 {
1203 	struct trace_probe *tp;
1204 
1205 	tp = trace_probe_primary_from_call(call);
1206 	if (WARN_ON_ONCE(!tp))
1207 		return;
1208 
1209 	if (!trace_probe_is_enabled(tp))
1210 		return;
1211 
1212 	if (file) {
1213 		if (trace_probe_remove_file(tp, file) < 0)
1214 			return;
1215 
1216 		if (trace_probe_is_enabled(tp))
1217 			return;
1218 	} else
1219 		trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
1220 
1221 	__probe_event_disable(tp);
1222 	uprobe_buffer_disable();
1223 }
1224 
1225 static int uprobe_event_define_fields(struct trace_event_call *event_call)
1226 {
1227 	int ret, size;
1228 	struct uprobe_trace_entry_head field;
1229 	struct trace_uprobe *tu;
1230 
1231 	tu = trace_uprobe_primary_from_call(event_call);
1232 	if (unlikely(!tu))
1233 		return -ENODEV;
1234 
1235 	if (is_ret_probe(tu)) {
1236 		DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_FUNC, 0);
1237 		DEFINE_FIELD(unsigned long, vaddr[1], FIELD_STRING_RETIP, 0);
1238 		size = SIZEOF_TRACE_ENTRY(true);
1239 	} else {
1240 		DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_IP, 0);
1241 		size = SIZEOF_TRACE_ENTRY(false);
1242 	}
1243 
1244 	return traceprobe_define_arg_fields(event_call, size, &tu->tp);
1245 }
1246 
1247 #ifdef CONFIG_PERF_EVENTS
1248 static bool
1249 __uprobe_perf_filter(struct trace_uprobe_filter *filter, struct mm_struct *mm)
1250 {
1251 	struct perf_event *event;
1252 
1253 	list_for_each_entry(event, &filter->perf_events, hw.tp_list) {
1254 		if (event->hw.target->mm == mm)
1255 			return true;
1256 	}
1257 
1258 	return false;
1259 }
1260 
1261 static inline bool
1262 trace_uprobe_filter_event(struct trace_uprobe_filter *filter,
1263 			  struct perf_event *event)
1264 {
1265 	return __uprobe_perf_filter(filter, event->hw.target->mm);
1266 }
1267 
1268 static bool trace_uprobe_filter_remove(struct trace_uprobe_filter *filter,
1269 				       struct perf_event *event)
1270 {
1271 	bool done;
1272 
1273 	write_lock(&filter->rwlock);
1274 	if (event->hw.target) {
1275 		list_del(&event->hw.tp_list);
1276 		done = filter->nr_systemwide ||
1277 			(event->hw.target->flags & PF_EXITING) ||
1278 			trace_uprobe_filter_event(filter, event);
1279 	} else {
1280 		filter->nr_systemwide--;
1281 		done = filter->nr_systemwide;
1282 	}
1283 	write_unlock(&filter->rwlock);
1284 
1285 	return done;
1286 }
1287 
1288 /* This returns true if the filter always covers target mm */
1289 static bool trace_uprobe_filter_add(struct trace_uprobe_filter *filter,
1290 				    struct perf_event *event)
1291 {
1292 	bool done;
1293 
1294 	write_lock(&filter->rwlock);
1295 	if (event->hw.target) {
1296 		/*
1297 		 * event->parent != NULL means copy_process(), we can avoid
1298 		 * uprobe_apply(). current->mm must be probed and we can rely
1299 		 * on dup_mmap() which preserves the already installed bp's.
1300 		 *
1301 		 * attr.enable_on_exec means that exec/mmap will install the
1302 		 * breakpoints we need.
1303 		 */
1304 		done = filter->nr_systemwide ||
1305 			event->parent || event->attr.enable_on_exec ||
1306 			trace_uprobe_filter_event(filter, event);
1307 		list_add(&event->hw.tp_list, &filter->perf_events);
1308 	} else {
1309 		done = filter->nr_systemwide;
1310 		filter->nr_systemwide++;
1311 	}
1312 	write_unlock(&filter->rwlock);
1313 
1314 	return done;
1315 }
1316 
1317 static int uprobe_perf_close(struct trace_event_call *call,
1318 			     struct perf_event *event)
1319 {
1320 	struct trace_probe *tp;
1321 	struct trace_uprobe *tu;
1322 	int ret = 0;
1323 
1324 	tp = trace_probe_primary_from_call(call);
1325 	if (WARN_ON_ONCE(!tp))
1326 		return -ENODEV;
1327 
1328 	tu = container_of(tp, struct trace_uprobe, tp);
1329 	if (trace_uprobe_filter_remove(tu->tp.event->filter, event))
1330 		return 0;
1331 
1332 	list_for_each_entry(tu, trace_probe_probe_list(tp), tp.list) {
1333 		ret = uprobe_apply(tu->uprobe, &tu->consumer, false);
1334 		if (ret)
1335 			break;
1336 	}
1337 
1338 	return ret;
1339 }
1340 
1341 static int uprobe_perf_open(struct trace_event_call *call,
1342 			    struct perf_event *event)
1343 {
1344 	struct trace_probe *tp;
1345 	struct trace_uprobe *tu;
1346 	int err = 0;
1347 
1348 	tp = trace_probe_primary_from_call(call);
1349 	if (WARN_ON_ONCE(!tp))
1350 		return -ENODEV;
1351 
1352 	tu = container_of(tp, struct trace_uprobe, tp);
1353 	if (trace_uprobe_filter_add(tu->tp.event->filter, event))
1354 		return 0;
1355 
1356 	list_for_each_entry(tu, trace_probe_probe_list(tp), tp.list) {
1357 		err = uprobe_apply(tu->uprobe, &tu->consumer, true);
1358 		if (err) {
1359 			uprobe_perf_close(call, event);
1360 			break;
1361 		}
1362 	}
1363 
1364 	return err;
1365 }
1366 
1367 static bool uprobe_perf_filter(struct uprobe_consumer *uc, struct mm_struct *mm)
1368 {
1369 	struct trace_uprobe_filter *filter;
1370 	struct trace_uprobe *tu;
1371 	int ret;
1372 
1373 	tu = container_of(uc, struct trace_uprobe, consumer);
1374 	filter = tu->tp.event->filter;
1375 
1376 	/*
1377 	 * speculative short-circuiting check to avoid unnecessarily taking
1378 	 * filter->rwlock below, if the uprobe has system-wide consumer
1379 	 */
1380 	if (READ_ONCE(filter->nr_systemwide))
1381 		return true;
1382 
1383 	read_lock(&filter->rwlock);
1384 	ret = __uprobe_perf_filter(filter, mm);
1385 	read_unlock(&filter->rwlock);
1386 
1387 	return ret;
1388 }
1389 
1390 static void __uprobe_perf_func(struct trace_uprobe *tu,
1391 			       unsigned long func, struct pt_regs *regs,
1392 			       struct uprobe_cpu_buffer **ucbp)
1393 {
1394 	struct trace_event_call *call = trace_probe_event_call(&tu->tp);
1395 	struct uprobe_trace_entry_head *entry;
1396 	struct uprobe_cpu_buffer *ucb;
1397 	struct hlist_head *head;
1398 	void *data;
1399 	int size, esize;
1400 	int rctx;
1401 
1402 #ifdef CONFIG_BPF_EVENTS
1403 	if (bpf_prog_array_valid(call)) {
1404 		const struct bpf_prog_array *array;
1405 		u32 ret;
1406 
1407 		rcu_read_lock_trace();
1408 		array = rcu_dereference_check(call->prog_array, rcu_read_lock_trace_held());
1409 		ret = bpf_prog_run_array_uprobe(array, regs, bpf_prog_run);
1410 		rcu_read_unlock_trace();
1411 		if (!ret)
1412 			return;
1413 	}
1414 #endif /* CONFIG_BPF_EVENTS */
1415 
1416 	esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1417 
1418 	ucb = prepare_uprobe_buffer(tu, regs, ucbp);
1419 	size = esize + ucb->dsize;
1420 	size = ALIGN(size + sizeof(u32), sizeof(u64)) - sizeof(u32);
1421 	if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE, "profile buffer not large enough"))
1422 		return;
1423 
1424 	preempt_disable();
1425 	head = this_cpu_ptr(call->perf_events);
1426 	if (hlist_empty(head))
1427 		goto out;
1428 
1429 	entry = perf_trace_buf_alloc(size, NULL, &rctx);
1430 	if (!entry)
1431 		goto out;
1432 
1433 	if (is_ret_probe(tu)) {
1434 		entry->vaddr[0] = func;
1435 		entry->vaddr[1] = instruction_pointer(regs);
1436 		data = DATAOF_TRACE_ENTRY(entry, true);
1437 	} else {
1438 		entry->vaddr[0] = instruction_pointer(regs);
1439 		data = DATAOF_TRACE_ENTRY(entry, false);
1440 	}
1441 
1442 	memcpy(data, ucb->buf, ucb->dsize);
1443 
1444 	if (size - esize > ucb->dsize)
1445 		memset(data + ucb->dsize, 0, size - esize - ucb->dsize);
1446 
1447 	perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1448 			      head, NULL);
1449  out:
1450 	preempt_enable();
1451 }
1452 
1453 /* uprobe profile handler */
1454 static int uprobe_perf_func(struct trace_uprobe *tu, struct pt_regs *regs,
1455 			    struct uprobe_cpu_buffer **ucbp)
1456 {
1457 	if (!uprobe_perf_filter(&tu->consumer, current->mm))
1458 		return UPROBE_HANDLER_REMOVE;
1459 
1460 	if (!is_ret_probe(tu))
1461 		__uprobe_perf_func(tu, 0, regs, ucbp);
1462 	return 0;
1463 }
1464 
1465 static void uretprobe_perf_func(struct trace_uprobe *tu, unsigned long func,
1466 				struct pt_regs *regs,
1467 				struct uprobe_cpu_buffer **ucbp)
1468 {
1469 	__uprobe_perf_func(tu, func, regs, ucbp);
1470 }
1471 
1472 int bpf_get_uprobe_info(const struct perf_event *event, u32 *fd_type,
1473 			const char **filename, u64 *probe_offset,
1474 			u64 *probe_addr, bool perf_type_tracepoint)
1475 {
1476 	const char *pevent = trace_event_name(event->tp_event);
1477 	const char *group = event->tp_event->class->system;
1478 	struct trace_uprobe *tu;
1479 
1480 	if (perf_type_tracepoint)
1481 		tu = find_probe_event(pevent, group);
1482 	else
1483 		tu = trace_uprobe_primary_from_call(event->tp_event);
1484 	if (!tu)
1485 		return -EINVAL;
1486 
1487 	*fd_type = is_ret_probe(tu) ? BPF_FD_TYPE_URETPROBE
1488 				    : BPF_FD_TYPE_UPROBE;
1489 	*filename = tu->filename;
1490 	*probe_offset = tu->offset;
1491 	*probe_addr = tu->ref_ctr_offset;
1492 	return 0;
1493 }
1494 #endif	/* CONFIG_PERF_EVENTS */
1495 
1496 static int
1497 trace_uprobe_register(struct trace_event_call *event, enum trace_reg type,
1498 		      void *data)
1499 {
1500 	struct trace_event_file *file = data;
1501 
1502 	switch (type) {
1503 	case TRACE_REG_REGISTER:
1504 		return probe_event_enable(event, file, NULL);
1505 
1506 	case TRACE_REG_UNREGISTER:
1507 		probe_event_disable(event, file);
1508 		return 0;
1509 
1510 #ifdef CONFIG_PERF_EVENTS
1511 	case TRACE_REG_PERF_REGISTER:
1512 		return probe_event_enable(event, NULL, uprobe_perf_filter);
1513 
1514 	case TRACE_REG_PERF_UNREGISTER:
1515 		probe_event_disable(event, NULL);
1516 		return 0;
1517 
1518 	case TRACE_REG_PERF_OPEN:
1519 		return uprobe_perf_open(event, data);
1520 
1521 	case TRACE_REG_PERF_CLOSE:
1522 		return uprobe_perf_close(event, data);
1523 
1524 #endif
1525 	default:
1526 		return 0;
1527 	}
1528 }
1529 
1530 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs,
1531 			     __u64 *data)
1532 {
1533 	struct trace_uprobe *tu;
1534 	struct uprobe_dispatch_data udd;
1535 	struct uprobe_cpu_buffer *ucb = NULL;
1536 	unsigned int flags;
1537 	int ret = 0;
1538 
1539 	tu = container_of(con, struct trace_uprobe, consumer);
1540 
1541 	this_cpu_inc(*tu->nhits);
1542 
1543 	udd.tu = tu;
1544 	udd.bp_addr = instruction_pointer(regs);
1545 
1546 	current->utask->vaddr = (unsigned long) &udd;
1547 
1548 	if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1549 		return 0;
1550 
1551 	flags = trace_probe_load_flag(&tu->tp);
1552 	if (flags & TP_FLAG_TRACE)
1553 		ret |= uprobe_trace_func(tu, regs, &ucb);
1554 
1555 #ifdef CONFIG_PERF_EVENTS
1556 	if (flags & TP_FLAG_PROFILE)
1557 		ret |= uprobe_perf_func(tu, regs, &ucb);
1558 #endif
1559 	uprobe_buffer_put(ucb);
1560 	return ret;
1561 }
1562 
1563 static int uretprobe_dispatcher(struct uprobe_consumer *con,
1564 				unsigned long func, struct pt_regs *regs,
1565 				__u64 *data)
1566 {
1567 	struct trace_uprobe *tu;
1568 	struct uprobe_dispatch_data udd;
1569 	struct uprobe_cpu_buffer *ucb = NULL;
1570 	unsigned int flags;
1571 
1572 	tu = container_of(con, struct trace_uprobe, consumer);
1573 
1574 	udd.tu = tu;
1575 	udd.bp_addr = func;
1576 
1577 	current->utask->vaddr = (unsigned long) &udd;
1578 
1579 	if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1580 		return 0;
1581 
1582 	flags = trace_probe_load_flag(&tu->tp);
1583 	if (flags & TP_FLAG_TRACE)
1584 		uretprobe_trace_func(tu, func, regs, &ucb);
1585 
1586 #ifdef CONFIG_PERF_EVENTS
1587 	if (flags & TP_FLAG_PROFILE)
1588 		uretprobe_perf_func(tu, func, regs, &ucb);
1589 #endif
1590 	uprobe_buffer_put(ucb);
1591 	return 0;
1592 }
1593 
1594 static struct trace_event_functions uprobe_funcs = {
1595 	.trace		= print_uprobe_event
1596 };
1597 
1598 static struct trace_event_fields uprobe_fields_array[] = {
1599 	{ .type = TRACE_FUNCTION_TYPE,
1600 	  .define_fields = uprobe_event_define_fields },
1601 	{}
1602 };
1603 
1604 static inline void init_trace_event_call(struct trace_uprobe *tu)
1605 {
1606 	struct trace_event_call *call = trace_probe_event_call(&tu->tp);
1607 	call->event.funcs = &uprobe_funcs;
1608 	call->class->fields_array = uprobe_fields_array;
1609 
1610 	call->flags = TRACE_EVENT_FL_UPROBE | TRACE_EVENT_FL_CAP_ANY;
1611 	call->class->reg = trace_uprobe_register;
1612 }
1613 
1614 static int register_uprobe_event(struct trace_uprobe *tu)
1615 {
1616 	init_trace_event_call(tu);
1617 
1618 	return trace_probe_register_event_call(&tu->tp);
1619 }
1620 
1621 static int unregister_uprobe_event(struct trace_uprobe *tu)
1622 {
1623 	return trace_probe_unregister_event_call(&tu->tp);
1624 }
1625 
1626 #ifdef CONFIG_PERF_EVENTS
1627 struct trace_event_call *
1628 create_local_trace_uprobe(char *name, unsigned long offs,
1629 			  unsigned long ref_ctr_offset, bool is_return)
1630 {
1631 	enum probe_print_type ptype;
1632 	struct trace_uprobe *tu;
1633 	struct path path;
1634 	int ret;
1635 
1636 	ret = kern_path(name, LOOKUP_FOLLOW, &path);
1637 	if (ret)
1638 		return ERR_PTR(ret);
1639 
1640 	if (!d_is_reg(path.dentry)) {
1641 		path_put(&path);
1642 		return ERR_PTR(-EINVAL);
1643 	}
1644 
1645 	/*
1646 	 * local trace_kprobes are not added to dyn_event, so they are never
1647 	 * searched in find_trace_kprobe(). Therefore, there is no concern of
1648 	 * duplicated name "DUMMY_EVENT" here.
1649 	 */
1650 	tu = alloc_trace_uprobe(UPROBE_EVENT_SYSTEM, "DUMMY_EVENT", 0,
1651 				is_return);
1652 
1653 	if (IS_ERR(tu)) {
1654 		pr_info("Failed to allocate trace_uprobe.(%d)\n",
1655 			(int)PTR_ERR(tu));
1656 		path_put(&path);
1657 		return ERR_CAST(tu);
1658 	}
1659 
1660 	tu->offset = offs;
1661 	tu->path = path;
1662 	tu->ref_ctr_offset = ref_ctr_offset;
1663 	tu->filename = kstrdup(name, GFP_KERNEL);
1664 	if (!tu->filename) {
1665 		ret = -ENOMEM;
1666 		goto error;
1667 	}
1668 
1669 	init_trace_event_call(tu);
1670 
1671 	ptype = is_ret_probe(tu) ? PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL;
1672 	if (traceprobe_set_print_fmt(&tu->tp, ptype) < 0) {
1673 		ret = -ENOMEM;
1674 		goto error;
1675 	}
1676 
1677 	return trace_probe_event_call(&tu->tp);
1678 error:
1679 	free_trace_uprobe(tu);
1680 	return ERR_PTR(ret);
1681 }
1682 
1683 void destroy_local_trace_uprobe(struct trace_event_call *event_call)
1684 {
1685 	struct trace_uprobe *tu;
1686 
1687 	tu = trace_uprobe_primary_from_call(event_call);
1688 
1689 	free_trace_uprobe(tu);
1690 }
1691 #endif /* CONFIG_PERF_EVENTS */
1692 
1693 /* Make a trace interface for controlling probe points */
1694 static __init int init_uprobe_trace(void)
1695 {
1696 	int ret;
1697 
1698 	ret = dyn_event_register(&trace_uprobe_ops);
1699 	if (ret)
1700 		return ret;
1701 
1702 	ret = tracing_init_dentry();
1703 	if (ret)
1704 		return 0;
1705 
1706 	trace_create_file("uprobe_events", TRACE_MODE_WRITE, NULL,
1707 				    NULL, &uprobe_events_ops);
1708 	/* Profile interface */
1709 	trace_create_file("uprobe_profile", TRACE_MODE_READ, NULL,
1710 				    NULL, &uprobe_profile_ops);
1711 	return 0;
1712 }
1713 
1714 fs_initcall(init_uprobe_trace);
1715