xref: /linux/kernel/trace/trace_uprobe.c (revision 5159938e10d876fd23d3a5474689a9f6e18a446e)
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 #include <linux/percpu.h>
21 
22 #include "trace_dynevent.h"
23 #include "trace_probe.h"
24 #include "trace_probe_tmpl.h"
25 
26 #define UPROBE_EVENT_SYSTEM	"uprobes"
27 
28 struct uprobe_trace_entry_head {
29 	struct trace_entry	ent;
30 	unsigned long		vaddr[];
31 };
32 
33 #define SIZEOF_TRACE_ENTRY(is_return)			\
34 	(sizeof(struct uprobe_trace_entry_head) +	\
35 	 sizeof(unsigned long) * (is_return ? 2 : 1))
36 
37 #define DATAOF_TRACE_ENTRY(entry, is_return)		\
38 	((void*)(entry) + SIZEOF_TRACE_ENTRY(is_return))
39 
40 static int trace_uprobe_create(const char *raw_command);
41 static int trace_uprobe_show(struct seq_file *m, struct dyn_event *ev);
42 static int trace_uprobe_release(struct dyn_event *ev);
43 static bool trace_uprobe_is_busy(struct dyn_event *ev);
44 static bool trace_uprobe_match(const char *system, const char *event,
45 			int argc, const char **argv, struct dyn_event *ev);
46 
47 static struct dyn_event_operations trace_uprobe_ops = {
48 	.create = trace_uprobe_create,
49 	.show = trace_uprobe_show,
50 	.is_busy = trace_uprobe_is_busy,
51 	.free = trace_uprobe_release,
52 	.match = trace_uprobe_match,
53 };
54 
55 /*
56  * uprobe event core functions
57  */
58 struct trace_uprobe {
59 	struct dyn_event		devent;
60 	struct uprobe_consumer		consumer;
61 	struct path			path;
62 	char				*filename;
63 	struct uprobe			*uprobe;
64 	unsigned long			offset;
65 	unsigned long			ref_ctr_offset;
66 	unsigned long __percpu		*nhits;
67 	struct trace_probe		tp;
68 };
69 
70 static bool is_trace_uprobe(struct dyn_event *ev)
71 {
72 	return ev->ops == &trace_uprobe_ops;
73 }
74 
75 static struct trace_uprobe *to_trace_uprobe(struct dyn_event *ev)
76 {
77 	return container_of(ev, struct trace_uprobe, devent);
78 }
79 
80 /**
81  * for_each_trace_uprobe - iterate over the trace_uprobe list
82  * @pos:	the struct trace_uprobe * for each entry
83  * @dpos:	the struct dyn_event * to use as a loop cursor
84  */
85 #define for_each_trace_uprobe(pos, dpos)	\
86 	for_each_dyn_event(dpos)		\
87 		if (is_trace_uprobe(dpos) && (pos = to_trace_uprobe(dpos)))
88 
89 static int register_uprobe_event(struct trace_uprobe *tu);
90 static int unregister_uprobe_event(struct trace_uprobe *tu);
91 
92 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs);
93 static int uretprobe_dispatcher(struct uprobe_consumer *con,
94 				unsigned long func, struct pt_regs *regs);
95 
96 #ifdef CONFIG_STACK_GROWSUP
97 static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
98 {
99 	return addr - (n * sizeof(long));
100 }
101 #else
102 static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
103 {
104 	return addr + (n * sizeof(long));
105 }
106 #endif
107 
108 static unsigned long get_user_stack_nth(struct pt_regs *regs, unsigned int n)
109 {
110 	unsigned long ret;
111 	unsigned long addr = user_stack_pointer(regs);
112 
113 	addr = adjust_stack_addr(addr, n);
114 
115 	if (copy_from_user(&ret, (void __force __user *) addr, sizeof(ret)))
116 		return 0;
117 
118 	return ret;
119 }
120 
121 /*
122  * Uprobes-specific fetch functions
123  */
124 static nokprobe_inline int
125 probe_mem_read(void *dest, void *src, size_t size)
126 {
127 	void __user *vaddr = (void __force __user *)src;
128 
129 	return copy_from_user(dest, vaddr, size) ? -EFAULT : 0;
130 }
131 
132 static nokprobe_inline int
133 probe_mem_read_user(void *dest, void *src, size_t size)
134 {
135 	return probe_mem_read(dest, src, size);
136 }
137 
138 /*
139  * Fetch a null-terminated string. Caller MUST set *(u32 *)dest with max
140  * length and relative data location.
141  */
142 static nokprobe_inline int
143 fetch_store_string(unsigned long addr, void *dest, void *base)
144 {
145 	long ret;
146 	u32 loc = *(u32 *)dest;
147 	int maxlen  = get_loc_len(loc);
148 	u8 *dst = get_loc_data(dest, base);
149 	void __user *src = (void __force __user *) addr;
150 
151 	if (unlikely(!maxlen))
152 		return -ENOMEM;
153 
154 	if (addr == FETCH_TOKEN_COMM)
155 		ret = strscpy(dst, current->comm, maxlen);
156 	else
157 		ret = strncpy_from_user(dst, src, maxlen);
158 	if (ret >= 0) {
159 		if (ret == maxlen)
160 			dst[ret - 1] = '\0';
161 		else
162 			/*
163 			 * Include the terminating null byte. In this case it
164 			 * was copied by strncpy_from_user but not accounted
165 			 * for in ret.
166 			 */
167 			ret++;
168 		*(u32 *)dest = make_data_loc(ret, (void *)dst - base);
169 	} else
170 		*(u32 *)dest = make_data_loc(0, (void *)dst - base);
171 
172 	return ret;
173 }
174 
175 static nokprobe_inline int
176 fetch_store_string_user(unsigned long addr, void *dest, void *base)
177 {
178 	return fetch_store_string(addr, dest, base);
179 }
180 
181 /* Return the length of string -- including null terminal byte */
182 static nokprobe_inline int
183 fetch_store_strlen(unsigned long addr)
184 {
185 	int len;
186 	void __user *vaddr = (void __force __user *) addr;
187 
188 	if (addr == FETCH_TOKEN_COMM)
189 		len = strlen(current->comm) + 1;
190 	else
191 		len = strnlen_user(vaddr, MAX_STRING_SIZE);
192 
193 	return (len > MAX_STRING_SIZE) ? 0 : len;
194 }
195 
196 static nokprobe_inline int
197 fetch_store_strlen_user(unsigned long addr)
198 {
199 	return fetch_store_strlen(addr);
200 }
201 
202 static unsigned long translate_user_vaddr(unsigned long file_offset)
203 {
204 	unsigned long base_addr;
205 	struct uprobe_dispatch_data *udd;
206 
207 	udd = (void *) current->utask->vaddr;
208 
209 	base_addr = udd->bp_addr - udd->tu->offset;
210 	return base_addr + file_offset;
211 }
212 
213 /* Note that we don't verify it, since the code does not come from user space */
214 static int
215 process_fetch_insn(struct fetch_insn *code, void *rec, void *edata,
216 		   void *dest, void *base)
217 {
218 	struct pt_regs *regs = rec;
219 	unsigned long val;
220 	int ret;
221 
222 	/* 1st stage: get value from context */
223 	switch (code->op) {
224 	case FETCH_OP_REG:
225 		val = regs_get_register(regs, code->param);
226 		break;
227 	case FETCH_OP_STACK:
228 		val = get_user_stack_nth(regs, code->param);
229 		break;
230 	case FETCH_OP_STACKP:
231 		val = user_stack_pointer(regs);
232 		break;
233 	case FETCH_OP_RETVAL:
234 		val = regs_return_value(regs);
235 		break;
236 	case FETCH_OP_COMM:
237 		val = FETCH_TOKEN_COMM;
238 		break;
239 	case FETCH_OP_FOFFS:
240 		val = translate_user_vaddr(code->immediate);
241 		break;
242 	default:
243 		ret = process_common_fetch_insn(code, &val);
244 		if (ret < 0)
245 			return ret;
246 	}
247 	code++;
248 
249 	return process_fetch_insn_bottom(code, val, dest, base);
250 }
251 NOKPROBE_SYMBOL(process_fetch_insn)
252 
253 static inline void init_trace_uprobe_filter(struct trace_uprobe_filter *filter)
254 {
255 	rwlock_init(&filter->rwlock);
256 	filter->nr_systemwide = 0;
257 	INIT_LIST_HEAD(&filter->perf_events);
258 }
259 
260 static inline bool uprobe_filter_is_empty(struct trace_uprobe_filter *filter)
261 {
262 	return !filter->nr_systemwide && list_empty(&filter->perf_events);
263 }
264 
265 static inline bool is_ret_probe(struct trace_uprobe *tu)
266 {
267 	return tu->consumer.ret_handler != NULL;
268 }
269 
270 static bool trace_uprobe_is_busy(struct dyn_event *ev)
271 {
272 	struct trace_uprobe *tu = to_trace_uprobe(ev);
273 
274 	return trace_probe_is_enabled(&tu->tp);
275 }
276 
277 static bool trace_uprobe_match_command_head(struct trace_uprobe *tu,
278 					    int argc, const char **argv)
279 {
280 	char buf[MAX_ARGSTR_LEN + 1];
281 	int len;
282 
283 	if (!argc)
284 		return true;
285 
286 	len = strlen(tu->filename);
287 	if (strncmp(tu->filename, argv[0], len) || argv[0][len] != ':')
288 		return false;
289 
290 	if (tu->ref_ctr_offset == 0)
291 		snprintf(buf, sizeof(buf), "0x%0*lx",
292 				(int)(sizeof(void *) * 2), tu->offset);
293 	else
294 		snprintf(buf, sizeof(buf), "0x%0*lx(0x%lx)",
295 				(int)(sizeof(void *) * 2), tu->offset,
296 				tu->ref_ctr_offset);
297 	if (strcmp(buf, &argv[0][len + 1]))
298 		return false;
299 
300 	argc--; argv++;
301 
302 	return trace_probe_match_command_args(&tu->tp, argc, argv);
303 }
304 
305 static bool trace_uprobe_match(const char *system, const char *event,
306 			int argc, const char **argv, struct dyn_event *ev)
307 {
308 	struct trace_uprobe *tu = to_trace_uprobe(ev);
309 
310 	return (event[0] == '\0' ||
311 		strcmp(trace_probe_name(&tu->tp), event) == 0) &&
312 	   (!system || strcmp(trace_probe_group_name(&tu->tp), system) == 0) &&
313 	   trace_uprobe_match_command_head(tu, argc, argv);
314 }
315 
316 static nokprobe_inline struct trace_uprobe *
317 trace_uprobe_primary_from_call(struct trace_event_call *call)
318 {
319 	struct trace_probe *tp;
320 
321 	tp = trace_probe_primary_from_call(call);
322 	if (WARN_ON_ONCE(!tp))
323 		return NULL;
324 
325 	return container_of(tp, struct trace_uprobe, tp);
326 }
327 
328 /*
329  * Allocate new trace_uprobe and initialize it (including uprobes).
330  */
331 static struct trace_uprobe *
332 alloc_trace_uprobe(const char *group, const char *event, int nargs, bool is_ret)
333 {
334 	struct trace_uprobe *tu;
335 	int ret;
336 
337 	tu = kzalloc(struct_size(tu, tp.args, nargs), GFP_KERNEL);
338 	if (!tu)
339 		return ERR_PTR(-ENOMEM);
340 
341 	tu->nhits = alloc_percpu(unsigned long);
342 	if (!tu->nhits) {
343 		ret = -ENOMEM;
344 		goto error;
345 	}
346 
347 	ret = trace_probe_init(&tu->tp, event, group, true, nargs);
348 	if (ret < 0)
349 		goto error;
350 
351 	dyn_event_init(&tu->devent, &trace_uprobe_ops);
352 	tu->consumer.handler = uprobe_dispatcher;
353 	if (is_ret)
354 		tu->consumer.ret_handler = uretprobe_dispatcher;
355 	init_trace_uprobe_filter(tu->tp.event->filter);
356 	return tu;
357 
358 error:
359 	free_percpu(tu->nhits);
360 	kfree(tu);
361 
362 	return ERR_PTR(ret);
363 }
364 
365 static void free_trace_uprobe(struct trace_uprobe *tu)
366 {
367 	if (!tu)
368 		return;
369 
370 	path_put(&tu->path);
371 	trace_probe_cleanup(&tu->tp);
372 	kfree(tu->filename);
373 	free_percpu(tu->nhits);
374 	kfree(tu);
375 }
376 
377 static struct trace_uprobe *find_probe_event(const char *event, const char *group)
378 {
379 	struct dyn_event *pos;
380 	struct trace_uprobe *tu;
381 
382 	for_each_trace_uprobe(tu, pos)
383 		if (strcmp(trace_probe_name(&tu->tp), event) == 0 &&
384 		    strcmp(trace_probe_group_name(&tu->tp), group) == 0)
385 			return tu;
386 
387 	return NULL;
388 }
389 
390 /* Unregister a trace_uprobe and probe_event */
391 static int unregister_trace_uprobe(struct trace_uprobe *tu)
392 {
393 	int ret;
394 
395 	if (trace_probe_has_sibling(&tu->tp))
396 		goto unreg;
397 
398 	/* If there's a reference to the dynamic event */
399 	if (trace_event_dyn_busy(trace_probe_event_call(&tu->tp)))
400 		return -EBUSY;
401 
402 	ret = unregister_uprobe_event(tu);
403 	if (ret)
404 		return ret;
405 
406 unreg:
407 	dyn_event_remove(&tu->devent);
408 	trace_probe_unlink(&tu->tp);
409 	free_trace_uprobe(tu);
410 	return 0;
411 }
412 
413 static bool trace_uprobe_has_same_uprobe(struct trace_uprobe *orig,
414 					 struct trace_uprobe *comp)
415 {
416 	struct trace_probe_event *tpe = orig->tp.event;
417 	struct inode *comp_inode = d_real_inode(comp->path.dentry);
418 	int i;
419 
420 	list_for_each_entry(orig, &tpe->probes, tp.list) {
421 		if (comp_inode != d_real_inode(orig->path.dentry) ||
422 		    comp->offset != orig->offset)
423 			continue;
424 
425 		/*
426 		 * trace_probe_compare_arg_type() ensured that nr_args and
427 		 * each argument name and type are same. Let's compare comm.
428 		 */
429 		for (i = 0; i < orig->tp.nr_args; i++) {
430 			if (strcmp(orig->tp.args[i].comm,
431 				   comp->tp.args[i].comm))
432 				break;
433 		}
434 
435 		if (i == orig->tp.nr_args)
436 			return true;
437 	}
438 
439 	return false;
440 }
441 
442 static int append_trace_uprobe(struct trace_uprobe *tu, struct trace_uprobe *to)
443 {
444 	int ret;
445 
446 	ret = trace_probe_compare_arg_type(&tu->tp, &to->tp);
447 	if (ret) {
448 		/* Note that argument starts index = 2 */
449 		trace_probe_log_set_index(ret + 1);
450 		trace_probe_log_err(0, DIFF_ARG_TYPE);
451 		return -EEXIST;
452 	}
453 	if (trace_uprobe_has_same_uprobe(to, tu)) {
454 		trace_probe_log_set_index(0);
455 		trace_probe_log_err(0, SAME_PROBE);
456 		return -EEXIST;
457 	}
458 
459 	/* Append to existing event */
460 	ret = trace_probe_append(&tu->tp, &to->tp);
461 	if (!ret)
462 		dyn_event_add(&tu->devent, trace_probe_event_call(&tu->tp));
463 
464 	return ret;
465 }
466 
467 /*
468  * Uprobe with multiple reference counter is not allowed. i.e.
469  * If inode and offset matches, reference counter offset *must*
470  * match as well. Though, there is one exception: If user is
471  * replacing old trace_uprobe with new one(same group/event),
472  * then we allow same uprobe with new reference counter as far
473  * as the new one does not conflict with any other existing
474  * ones.
475  */
476 static int validate_ref_ctr_offset(struct trace_uprobe *new)
477 {
478 	struct dyn_event *pos;
479 	struct trace_uprobe *tmp;
480 	struct inode *new_inode = d_real_inode(new->path.dentry);
481 
482 	for_each_trace_uprobe(tmp, pos) {
483 		if (new_inode == d_real_inode(tmp->path.dentry) &&
484 		    new->offset == tmp->offset &&
485 		    new->ref_ctr_offset != tmp->ref_ctr_offset) {
486 			pr_warn("Reference counter offset mismatch.");
487 			return -EINVAL;
488 		}
489 	}
490 	return 0;
491 }
492 
493 /* Register a trace_uprobe and probe_event */
494 static int register_trace_uprobe(struct trace_uprobe *tu)
495 {
496 	struct trace_uprobe *old_tu;
497 	int ret;
498 
499 	mutex_lock(&event_mutex);
500 
501 	ret = validate_ref_ctr_offset(tu);
502 	if (ret)
503 		goto end;
504 
505 	/* register as an event */
506 	old_tu = find_probe_event(trace_probe_name(&tu->tp),
507 				  trace_probe_group_name(&tu->tp));
508 	if (old_tu) {
509 		if (is_ret_probe(tu) != is_ret_probe(old_tu)) {
510 			trace_probe_log_set_index(0);
511 			trace_probe_log_err(0, DIFF_PROBE_TYPE);
512 			ret = -EEXIST;
513 		} else {
514 			ret = append_trace_uprobe(tu, old_tu);
515 		}
516 		goto end;
517 	}
518 
519 	ret = register_uprobe_event(tu);
520 	if (ret) {
521 		if (ret == -EEXIST) {
522 			trace_probe_log_set_index(0);
523 			trace_probe_log_err(0, EVENT_EXIST);
524 		} else
525 			pr_warn("Failed to register probe event(%d)\n", ret);
526 		goto end;
527 	}
528 
529 	dyn_event_add(&tu->devent, trace_probe_event_call(&tu->tp));
530 
531 end:
532 	mutex_unlock(&event_mutex);
533 
534 	return ret;
535 }
536 
537 /*
538  * Argument syntax:
539  *  - Add uprobe: p|r[:[GRP/][EVENT]] PATH:OFFSET[%return][(REF)] [FETCHARGS]
540  */
541 static int __trace_uprobe_create(int argc, const char **argv)
542 {
543 	struct trace_uprobe *tu;
544 	const char *event = NULL, *group = UPROBE_EVENT_SYSTEM;
545 	char *arg, *filename, *rctr, *rctr_end, *tmp;
546 	char buf[MAX_EVENT_NAME_LEN];
547 	char gbuf[MAX_EVENT_NAME_LEN];
548 	enum probe_print_type ptype;
549 	struct path path;
550 	unsigned long offset, ref_ctr_offset;
551 	bool is_return = false;
552 	int i, ret;
553 
554 	ref_ctr_offset = 0;
555 
556 	switch (argv[0][0]) {
557 	case 'r':
558 		is_return = true;
559 		break;
560 	case 'p':
561 		break;
562 	default:
563 		return -ECANCELED;
564 	}
565 
566 	if (argc < 2)
567 		return -ECANCELED;
568 
569 	if (argv[0][1] == ':')
570 		event = &argv[0][2];
571 
572 	if (!strchr(argv[1], '/'))
573 		return -ECANCELED;
574 
575 	filename = kstrdup(argv[1], GFP_KERNEL);
576 	if (!filename)
577 		return -ENOMEM;
578 
579 	/* Find the last occurrence, in case the path contains ':' too. */
580 	arg = strrchr(filename, ':');
581 	if (!arg || !isdigit(arg[1])) {
582 		kfree(filename);
583 		return -ECANCELED;
584 	}
585 
586 	trace_probe_log_init("trace_uprobe", argc, argv);
587 	trace_probe_log_set_index(1);	/* filename is the 2nd argument */
588 
589 	*arg++ = '\0';
590 	ret = kern_path(filename, LOOKUP_FOLLOW, &path);
591 	if (ret) {
592 		trace_probe_log_err(0, FILE_NOT_FOUND);
593 		kfree(filename);
594 		trace_probe_log_clear();
595 		return ret;
596 	}
597 	if (!d_is_reg(path.dentry)) {
598 		trace_probe_log_err(0, NO_REGULAR_FILE);
599 		ret = -EINVAL;
600 		goto fail_address_parse;
601 	}
602 
603 	/* Parse reference counter offset if specified. */
604 	rctr = strchr(arg, '(');
605 	if (rctr) {
606 		rctr_end = strchr(rctr, ')');
607 		if (!rctr_end) {
608 			ret = -EINVAL;
609 			rctr_end = rctr + strlen(rctr);
610 			trace_probe_log_err(rctr_end - filename,
611 					    REFCNT_OPEN_BRACE);
612 			goto fail_address_parse;
613 		} else if (rctr_end[1] != '\0') {
614 			ret = -EINVAL;
615 			trace_probe_log_err(rctr_end + 1 - filename,
616 					    BAD_REFCNT_SUFFIX);
617 			goto fail_address_parse;
618 		}
619 
620 		*rctr++ = '\0';
621 		*rctr_end = '\0';
622 		ret = kstrtoul(rctr, 0, &ref_ctr_offset);
623 		if (ret) {
624 			trace_probe_log_err(rctr - filename, BAD_REFCNT);
625 			goto fail_address_parse;
626 		}
627 	}
628 
629 	/* Check if there is %return suffix */
630 	tmp = strchr(arg, '%');
631 	if (tmp) {
632 		if (!strcmp(tmp, "%return")) {
633 			*tmp = '\0';
634 			is_return = true;
635 		} else {
636 			trace_probe_log_err(tmp - filename, BAD_ADDR_SUFFIX);
637 			ret = -EINVAL;
638 			goto fail_address_parse;
639 		}
640 	}
641 
642 	/* Parse uprobe offset. */
643 	ret = kstrtoul(arg, 0, &offset);
644 	if (ret) {
645 		trace_probe_log_err(arg - filename, BAD_UPROBE_OFFS);
646 		goto fail_address_parse;
647 	}
648 
649 	/* setup a probe */
650 	trace_probe_log_set_index(0);
651 	if (event) {
652 		ret = traceprobe_parse_event_name(&event, &group, gbuf,
653 						  event - argv[0]);
654 		if (ret)
655 			goto fail_address_parse;
656 	}
657 
658 	if (!event) {
659 		char *tail;
660 		char *ptr;
661 
662 		tail = kstrdup(kbasename(filename), GFP_KERNEL);
663 		if (!tail) {
664 			ret = -ENOMEM;
665 			goto fail_address_parse;
666 		}
667 
668 		ptr = strpbrk(tail, ".-_");
669 		if (ptr)
670 			*ptr = '\0';
671 
672 		snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_0x%lx", 'p', tail, offset);
673 		event = buf;
674 		kfree(tail);
675 	}
676 
677 	argc -= 2;
678 	argv += 2;
679 
680 	tu = alloc_trace_uprobe(group, event, argc, is_return);
681 	if (IS_ERR(tu)) {
682 		ret = PTR_ERR(tu);
683 		/* This must return -ENOMEM otherwise there is a bug */
684 		WARN_ON_ONCE(ret != -ENOMEM);
685 		goto fail_address_parse;
686 	}
687 	tu->offset = offset;
688 	tu->ref_ctr_offset = ref_ctr_offset;
689 	tu->path = path;
690 	tu->filename = filename;
691 
692 	/* parse arguments */
693 	for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
694 		struct traceprobe_parse_context ctx = {
695 			.flags = (is_return ? TPARG_FL_RETURN : 0) | TPARG_FL_USER,
696 		};
697 
698 		trace_probe_log_set_index(i + 2);
699 		ret = traceprobe_parse_probe_arg(&tu->tp, i, argv[i], &ctx);
700 		traceprobe_finish_parse(&ctx);
701 		if (ret)
702 			goto error;
703 	}
704 
705 	ptype = is_ret_probe(tu) ? PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL;
706 	ret = traceprobe_set_print_fmt(&tu->tp, ptype);
707 	if (ret < 0)
708 		goto error;
709 
710 	ret = register_trace_uprobe(tu);
711 	if (!ret)
712 		goto out;
713 
714 error:
715 	free_trace_uprobe(tu);
716 out:
717 	trace_probe_log_clear();
718 	return ret;
719 
720 fail_address_parse:
721 	trace_probe_log_clear();
722 	path_put(&path);
723 	kfree(filename);
724 
725 	return ret;
726 }
727 
728 int trace_uprobe_create(const char *raw_command)
729 {
730 	return trace_probe_create(raw_command, __trace_uprobe_create);
731 }
732 
733 static int create_or_delete_trace_uprobe(const char *raw_command)
734 {
735 	int ret;
736 
737 	if (raw_command[0] == '-')
738 		return dyn_event_release(raw_command, &trace_uprobe_ops);
739 
740 	ret = trace_uprobe_create(raw_command);
741 	return ret == -ECANCELED ? -EINVAL : ret;
742 }
743 
744 static int trace_uprobe_release(struct dyn_event *ev)
745 {
746 	struct trace_uprobe *tu = to_trace_uprobe(ev);
747 
748 	return unregister_trace_uprobe(tu);
749 }
750 
751 /* Probes listing interfaces */
752 static int trace_uprobe_show(struct seq_file *m, struct dyn_event *ev)
753 {
754 	struct trace_uprobe *tu = to_trace_uprobe(ev);
755 	char c = is_ret_probe(tu) ? 'r' : 'p';
756 	int i;
757 
758 	seq_printf(m, "%c:%s/%s %s:0x%0*lx", c, trace_probe_group_name(&tu->tp),
759 			trace_probe_name(&tu->tp), tu->filename,
760 			(int)(sizeof(void *) * 2), tu->offset);
761 
762 	if (tu->ref_ctr_offset)
763 		seq_printf(m, "(0x%lx)", tu->ref_ctr_offset);
764 
765 	for (i = 0; i < tu->tp.nr_args; i++)
766 		seq_printf(m, " %s=%s", tu->tp.args[i].name, tu->tp.args[i].comm);
767 
768 	seq_putc(m, '\n');
769 	return 0;
770 }
771 
772 static int probes_seq_show(struct seq_file *m, void *v)
773 {
774 	struct dyn_event *ev = v;
775 
776 	if (!is_trace_uprobe(ev))
777 		return 0;
778 
779 	return trace_uprobe_show(m, ev);
780 }
781 
782 static const struct seq_operations probes_seq_op = {
783 	.start  = dyn_event_seq_start,
784 	.next   = dyn_event_seq_next,
785 	.stop   = dyn_event_seq_stop,
786 	.show   = probes_seq_show
787 };
788 
789 static int probes_open(struct inode *inode, struct file *file)
790 {
791 	int ret;
792 
793 	ret = security_locked_down(LOCKDOWN_TRACEFS);
794 	if (ret)
795 		return ret;
796 
797 	if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
798 		ret = dyn_events_release_all(&trace_uprobe_ops);
799 		if (ret)
800 			return ret;
801 	}
802 
803 	return seq_open(file, &probes_seq_op);
804 }
805 
806 static ssize_t probes_write(struct file *file, const char __user *buffer,
807 			    size_t count, loff_t *ppos)
808 {
809 	return trace_parse_run_command(file, buffer, count, ppos,
810 					create_or_delete_trace_uprobe);
811 }
812 
813 static const struct file_operations uprobe_events_ops = {
814 	.owner		= THIS_MODULE,
815 	.open		= probes_open,
816 	.read		= seq_read,
817 	.llseek		= seq_lseek,
818 	.release	= seq_release,
819 	.write		= probes_write,
820 };
821 
822 /* Probes profiling interfaces */
823 static int probes_profile_seq_show(struct seq_file *m, void *v)
824 {
825 	struct dyn_event *ev = v;
826 	struct trace_uprobe *tu;
827 	unsigned long nhits;
828 	int cpu;
829 
830 	if (!is_trace_uprobe(ev))
831 		return 0;
832 
833 	tu = to_trace_uprobe(ev);
834 
835 	nhits = 0;
836 	for_each_possible_cpu(cpu) {
837 		nhits += per_cpu(*tu->nhits, cpu);
838 	}
839 
840 	seq_printf(m, "  %s %-44s %15lu\n", tu->filename,
841 		   trace_probe_name(&tu->tp), nhits);
842 	return 0;
843 }
844 
845 static const struct seq_operations profile_seq_op = {
846 	.start  = dyn_event_seq_start,
847 	.next   = dyn_event_seq_next,
848 	.stop   = dyn_event_seq_stop,
849 	.show	= probes_profile_seq_show
850 };
851 
852 static int profile_open(struct inode *inode, struct file *file)
853 {
854 	int ret;
855 
856 	ret = security_locked_down(LOCKDOWN_TRACEFS);
857 	if (ret)
858 		return ret;
859 
860 	return seq_open(file, &profile_seq_op);
861 }
862 
863 static const struct file_operations uprobe_profile_ops = {
864 	.owner		= THIS_MODULE,
865 	.open		= profile_open,
866 	.read		= seq_read,
867 	.llseek		= seq_lseek,
868 	.release	= seq_release,
869 };
870 
871 struct uprobe_cpu_buffer {
872 	struct mutex mutex;
873 	void *buf;
874 	int dsize;
875 };
876 static struct uprobe_cpu_buffer __percpu *uprobe_cpu_buffer;
877 static int uprobe_buffer_refcnt;
878 
879 static int uprobe_buffer_init(void)
880 {
881 	int cpu, err_cpu;
882 
883 	uprobe_cpu_buffer = alloc_percpu(struct uprobe_cpu_buffer);
884 	if (uprobe_cpu_buffer == NULL)
885 		return -ENOMEM;
886 
887 	for_each_possible_cpu(cpu) {
888 		struct page *p = alloc_pages_node(cpu_to_node(cpu),
889 						  GFP_KERNEL, 0);
890 		if (p == NULL) {
891 			err_cpu = cpu;
892 			goto err;
893 		}
894 		per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf = page_address(p);
895 		mutex_init(&per_cpu_ptr(uprobe_cpu_buffer, cpu)->mutex);
896 	}
897 
898 	return 0;
899 
900 err:
901 	for_each_possible_cpu(cpu) {
902 		if (cpu == err_cpu)
903 			break;
904 		free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf);
905 	}
906 
907 	free_percpu(uprobe_cpu_buffer);
908 	return -ENOMEM;
909 }
910 
911 static int uprobe_buffer_enable(void)
912 {
913 	int ret = 0;
914 
915 	BUG_ON(!mutex_is_locked(&event_mutex));
916 
917 	if (uprobe_buffer_refcnt++ == 0) {
918 		ret = uprobe_buffer_init();
919 		if (ret < 0)
920 			uprobe_buffer_refcnt--;
921 	}
922 
923 	return ret;
924 }
925 
926 static void uprobe_buffer_disable(void)
927 {
928 	int cpu;
929 
930 	BUG_ON(!mutex_is_locked(&event_mutex));
931 
932 	if (--uprobe_buffer_refcnt == 0) {
933 		for_each_possible_cpu(cpu)
934 			free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer,
935 							     cpu)->buf);
936 
937 		free_percpu(uprobe_cpu_buffer);
938 		uprobe_cpu_buffer = NULL;
939 	}
940 }
941 
942 static struct uprobe_cpu_buffer *uprobe_buffer_get(void)
943 {
944 	struct uprobe_cpu_buffer *ucb;
945 	int cpu;
946 
947 	cpu = raw_smp_processor_id();
948 	ucb = per_cpu_ptr(uprobe_cpu_buffer, cpu);
949 
950 	/*
951 	 * Use per-cpu buffers for fastest access, but we might migrate
952 	 * so the mutex makes sure we have sole access to it.
953 	 */
954 	mutex_lock(&ucb->mutex);
955 
956 	return ucb;
957 }
958 
959 static void uprobe_buffer_put(struct uprobe_cpu_buffer *ucb)
960 {
961 	if (!ucb)
962 		return;
963 	mutex_unlock(&ucb->mutex);
964 }
965 
966 static struct uprobe_cpu_buffer *prepare_uprobe_buffer(struct trace_uprobe *tu,
967 						       struct pt_regs *regs,
968 						       struct uprobe_cpu_buffer **ucbp)
969 {
970 	struct uprobe_cpu_buffer *ucb;
971 	int dsize, esize;
972 
973 	if (*ucbp)
974 		return *ucbp;
975 
976 	esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
977 	dsize = __get_data_size(&tu->tp, regs, NULL);
978 
979 	ucb = uprobe_buffer_get();
980 	ucb->dsize = tu->tp.size + dsize;
981 
982 	store_trace_args(ucb->buf, &tu->tp, regs, NULL, esize, dsize);
983 
984 	*ucbp = ucb;
985 	return ucb;
986 }
987 
988 static void __uprobe_trace_func(struct trace_uprobe *tu,
989 				unsigned long func, struct pt_regs *regs,
990 				struct uprobe_cpu_buffer *ucb,
991 				struct trace_event_file *trace_file)
992 {
993 	struct uprobe_trace_entry_head *entry;
994 	struct trace_event_buffer fbuffer;
995 	void *data;
996 	int size, esize;
997 	struct trace_event_call *call = trace_probe_event_call(&tu->tp);
998 
999 	WARN_ON(call != trace_file->event_call);
1000 
1001 	if (WARN_ON_ONCE(ucb->dsize > PAGE_SIZE))
1002 		return;
1003 
1004 	if (trace_trigger_soft_disabled(trace_file))
1005 		return;
1006 
1007 	esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1008 	size = esize + ucb->dsize;
1009 	entry = trace_event_buffer_reserve(&fbuffer, trace_file, size);
1010 	if (!entry)
1011 		return;
1012 
1013 	if (is_ret_probe(tu)) {
1014 		entry->vaddr[0] = func;
1015 		entry->vaddr[1] = instruction_pointer(regs);
1016 		data = DATAOF_TRACE_ENTRY(entry, true);
1017 	} else {
1018 		entry->vaddr[0] = instruction_pointer(regs);
1019 		data = DATAOF_TRACE_ENTRY(entry, false);
1020 	}
1021 
1022 	memcpy(data, ucb->buf, ucb->dsize);
1023 
1024 	trace_event_buffer_commit(&fbuffer);
1025 }
1026 
1027 /* uprobe handler */
1028 static int uprobe_trace_func(struct trace_uprobe *tu, struct pt_regs *regs,
1029 			     struct uprobe_cpu_buffer **ucbp)
1030 {
1031 	struct event_file_link *link;
1032 	struct uprobe_cpu_buffer *ucb;
1033 
1034 	if (is_ret_probe(tu))
1035 		return 0;
1036 
1037 	ucb = prepare_uprobe_buffer(tu, regs, ucbp);
1038 
1039 	rcu_read_lock();
1040 	trace_probe_for_each_link_rcu(link, &tu->tp)
1041 		__uprobe_trace_func(tu, 0, regs, ucb, link->file);
1042 	rcu_read_unlock();
1043 
1044 	return 0;
1045 }
1046 
1047 static void uretprobe_trace_func(struct trace_uprobe *tu, unsigned long func,
1048 				 struct pt_regs *regs,
1049 				 struct uprobe_cpu_buffer **ucbp)
1050 {
1051 	struct event_file_link *link;
1052 	struct uprobe_cpu_buffer *ucb;
1053 
1054 	ucb = prepare_uprobe_buffer(tu, regs, ucbp);
1055 
1056 	rcu_read_lock();
1057 	trace_probe_for_each_link_rcu(link, &tu->tp)
1058 		__uprobe_trace_func(tu, func, regs, ucb, link->file);
1059 	rcu_read_unlock();
1060 }
1061 
1062 /* Event entry printers */
1063 static enum print_line_t
1064 print_uprobe_event(struct trace_iterator *iter, int flags, struct trace_event *event)
1065 {
1066 	struct uprobe_trace_entry_head *entry;
1067 	struct trace_seq *s = &iter->seq;
1068 	struct trace_uprobe *tu;
1069 	u8 *data;
1070 
1071 	entry = (struct uprobe_trace_entry_head *)iter->ent;
1072 	tu = trace_uprobe_primary_from_call(
1073 		container_of(event, struct trace_event_call, event));
1074 	if (unlikely(!tu))
1075 		goto out;
1076 
1077 	if (is_ret_probe(tu)) {
1078 		trace_seq_printf(s, "%s: (0x%lx <- 0x%lx)",
1079 				 trace_probe_name(&tu->tp),
1080 				 entry->vaddr[1], entry->vaddr[0]);
1081 		data = DATAOF_TRACE_ENTRY(entry, true);
1082 	} else {
1083 		trace_seq_printf(s, "%s: (0x%lx)",
1084 				 trace_probe_name(&tu->tp),
1085 				 entry->vaddr[0]);
1086 		data = DATAOF_TRACE_ENTRY(entry, false);
1087 	}
1088 
1089 	if (trace_probe_print_args(s, tu->tp.args, tu->tp.nr_args, data, entry) < 0)
1090 		goto out;
1091 
1092 	trace_seq_putc(s, '\n');
1093 
1094  out:
1095 	return trace_handle_return(s);
1096 }
1097 
1098 typedef bool (*filter_func_t)(struct uprobe_consumer *self, struct mm_struct *mm);
1099 
1100 static int trace_uprobe_enable(struct trace_uprobe *tu, filter_func_t filter)
1101 {
1102 	struct inode *inode = d_real_inode(tu->path.dentry);
1103 	struct uprobe *uprobe;
1104 
1105 	tu->consumer.filter = filter;
1106 	uprobe = uprobe_register(inode, tu->offset, tu->ref_ctr_offset, &tu->consumer);
1107 	if (IS_ERR(uprobe))
1108 		return PTR_ERR(uprobe);
1109 
1110 	tu->uprobe = uprobe;
1111 	return 0;
1112 }
1113 
1114 static void __probe_event_disable(struct trace_probe *tp)
1115 {
1116 	struct trace_uprobe *tu;
1117 	bool sync = false;
1118 
1119 	tu = container_of(tp, struct trace_uprobe, tp);
1120 	WARN_ON(!uprobe_filter_is_empty(tu->tp.event->filter));
1121 
1122 	list_for_each_entry(tu, trace_probe_probe_list(tp), tp.list) {
1123 		if (!tu->uprobe)
1124 			continue;
1125 
1126 		uprobe_unregister_nosync(tu->uprobe, &tu->consumer);
1127 		sync = true;
1128 		tu->uprobe = NULL;
1129 	}
1130 	if (sync)
1131 		uprobe_unregister_sync();
1132 }
1133 
1134 static int probe_event_enable(struct trace_event_call *call,
1135 			struct trace_event_file *file, filter_func_t filter)
1136 {
1137 	struct trace_probe *tp;
1138 	struct trace_uprobe *tu;
1139 	bool enabled;
1140 	int ret;
1141 
1142 	tp = trace_probe_primary_from_call(call);
1143 	if (WARN_ON_ONCE(!tp))
1144 		return -ENODEV;
1145 	enabled = trace_probe_is_enabled(tp);
1146 
1147 	/* This may also change "enabled" state */
1148 	if (file) {
1149 		if (trace_probe_test_flag(tp, TP_FLAG_PROFILE))
1150 			return -EINTR;
1151 
1152 		ret = trace_probe_add_file(tp, file);
1153 		if (ret < 0)
1154 			return ret;
1155 	} else {
1156 		if (trace_probe_test_flag(tp, TP_FLAG_TRACE))
1157 			return -EINTR;
1158 
1159 		trace_probe_set_flag(tp, TP_FLAG_PROFILE);
1160 	}
1161 
1162 	tu = container_of(tp, struct trace_uprobe, tp);
1163 	WARN_ON(!uprobe_filter_is_empty(tu->tp.event->filter));
1164 
1165 	if (enabled)
1166 		return 0;
1167 
1168 	ret = uprobe_buffer_enable();
1169 	if (ret)
1170 		goto err_flags;
1171 
1172 	list_for_each_entry(tu, trace_probe_probe_list(tp), tp.list) {
1173 		ret = trace_uprobe_enable(tu, filter);
1174 		if (ret) {
1175 			__probe_event_disable(tp);
1176 			goto err_buffer;
1177 		}
1178 	}
1179 
1180 	return 0;
1181 
1182  err_buffer:
1183 	uprobe_buffer_disable();
1184 
1185  err_flags:
1186 	if (file)
1187 		trace_probe_remove_file(tp, file);
1188 	else
1189 		trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
1190 
1191 	return ret;
1192 }
1193 
1194 static void probe_event_disable(struct trace_event_call *call,
1195 				struct trace_event_file *file)
1196 {
1197 	struct trace_probe *tp;
1198 
1199 	tp = trace_probe_primary_from_call(call);
1200 	if (WARN_ON_ONCE(!tp))
1201 		return;
1202 
1203 	if (!trace_probe_is_enabled(tp))
1204 		return;
1205 
1206 	if (file) {
1207 		if (trace_probe_remove_file(tp, file) < 0)
1208 			return;
1209 
1210 		if (trace_probe_is_enabled(tp))
1211 			return;
1212 	} else
1213 		trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
1214 
1215 	__probe_event_disable(tp);
1216 	uprobe_buffer_disable();
1217 }
1218 
1219 static int uprobe_event_define_fields(struct trace_event_call *event_call)
1220 {
1221 	int ret, size;
1222 	struct uprobe_trace_entry_head field;
1223 	struct trace_uprobe *tu;
1224 
1225 	tu = trace_uprobe_primary_from_call(event_call);
1226 	if (unlikely(!tu))
1227 		return -ENODEV;
1228 
1229 	if (is_ret_probe(tu)) {
1230 		DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_FUNC, 0);
1231 		DEFINE_FIELD(unsigned long, vaddr[1], FIELD_STRING_RETIP, 0);
1232 		size = SIZEOF_TRACE_ENTRY(true);
1233 	} else {
1234 		DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_IP, 0);
1235 		size = SIZEOF_TRACE_ENTRY(false);
1236 	}
1237 
1238 	return traceprobe_define_arg_fields(event_call, size, &tu->tp);
1239 }
1240 
1241 #ifdef CONFIG_PERF_EVENTS
1242 static bool
1243 __uprobe_perf_filter(struct trace_uprobe_filter *filter, struct mm_struct *mm)
1244 {
1245 	struct perf_event *event;
1246 
1247 	list_for_each_entry(event, &filter->perf_events, hw.tp_list) {
1248 		if (event->hw.target->mm == mm)
1249 			return true;
1250 	}
1251 
1252 	return false;
1253 }
1254 
1255 static inline bool
1256 trace_uprobe_filter_event(struct trace_uprobe_filter *filter,
1257 			  struct perf_event *event)
1258 {
1259 	return __uprobe_perf_filter(filter, event->hw.target->mm);
1260 }
1261 
1262 static bool trace_uprobe_filter_remove(struct trace_uprobe_filter *filter,
1263 				       struct perf_event *event)
1264 {
1265 	bool done;
1266 
1267 	write_lock(&filter->rwlock);
1268 	if (event->hw.target) {
1269 		list_del(&event->hw.tp_list);
1270 		done = filter->nr_systemwide ||
1271 			(event->hw.target->flags & PF_EXITING) ||
1272 			trace_uprobe_filter_event(filter, event);
1273 	} else {
1274 		filter->nr_systemwide--;
1275 		done = filter->nr_systemwide;
1276 	}
1277 	write_unlock(&filter->rwlock);
1278 
1279 	return done;
1280 }
1281 
1282 /* This returns true if the filter always covers target mm */
1283 static bool trace_uprobe_filter_add(struct trace_uprobe_filter *filter,
1284 				    struct perf_event *event)
1285 {
1286 	bool done;
1287 
1288 	write_lock(&filter->rwlock);
1289 	if (event->hw.target) {
1290 		/*
1291 		 * event->parent != NULL means copy_process(), we can avoid
1292 		 * uprobe_apply(). current->mm must be probed and we can rely
1293 		 * on dup_mmap() which preserves the already installed bp's.
1294 		 *
1295 		 * attr.enable_on_exec means that exec/mmap will install the
1296 		 * breakpoints we need.
1297 		 */
1298 		done = filter->nr_systemwide ||
1299 			event->parent || event->attr.enable_on_exec ||
1300 			trace_uprobe_filter_event(filter, event);
1301 		list_add(&event->hw.tp_list, &filter->perf_events);
1302 	} else {
1303 		done = filter->nr_systemwide;
1304 		filter->nr_systemwide++;
1305 	}
1306 	write_unlock(&filter->rwlock);
1307 
1308 	return done;
1309 }
1310 
1311 static int uprobe_perf_close(struct trace_event_call *call,
1312 			     struct perf_event *event)
1313 {
1314 	struct trace_probe *tp;
1315 	struct trace_uprobe *tu;
1316 	int ret = 0;
1317 
1318 	tp = trace_probe_primary_from_call(call);
1319 	if (WARN_ON_ONCE(!tp))
1320 		return -ENODEV;
1321 
1322 	tu = container_of(tp, struct trace_uprobe, tp);
1323 	if (trace_uprobe_filter_remove(tu->tp.event->filter, event))
1324 		return 0;
1325 
1326 	list_for_each_entry(tu, trace_probe_probe_list(tp), tp.list) {
1327 		ret = uprobe_apply(tu->uprobe, &tu->consumer, false);
1328 		if (ret)
1329 			break;
1330 	}
1331 
1332 	return ret;
1333 }
1334 
1335 static int uprobe_perf_open(struct trace_event_call *call,
1336 			    struct perf_event *event)
1337 {
1338 	struct trace_probe *tp;
1339 	struct trace_uprobe *tu;
1340 	int err = 0;
1341 
1342 	tp = trace_probe_primary_from_call(call);
1343 	if (WARN_ON_ONCE(!tp))
1344 		return -ENODEV;
1345 
1346 	tu = container_of(tp, struct trace_uprobe, tp);
1347 	if (trace_uprobe_filter_add(tu->tp.event->filter, event))
1348 		return 0;
1349 
1350 	list_for_each_entry(tu, trace_probe_probe_list(tp), tp.list) {
1351 		err = uprobe_apply(tu->uprobe, &tu->consumer, true);
1352 		if (err) {
1353 			uprobe_perf_close(call, event);
1354 			break;
1355 		}
1356 	}
1357 
1358 	return err;
1359 }
1360 
1361 static bool uprobe_perf_filter(struct uprobe_consumer *uc, struct mm_struct *mm)
1362 {
1363 	struct trace_uprobe_filter *filter;
1364 	struct trace_uprobe *tu;
1365 	int ret;
1366 
1367 	tu = container_of(uc, struct trace_uprobe, consumer);
1368 	filter = tu->tp.event->filter;
1369 
1370 	/*
1371 	 * speculative short-circuiting check to avoid unnecessarily taking
1372 	 * filter->rwlock below, if the uprobe has system-wide consumer
1373 	 */
1374 	if (READ_ONCE(filter->nr_systemwide))
1375 		return true;
1376 
1377 	read_lock(&filter->rwlock);
1378 	ret = __uprobe_perf_filter(filter, mm);
1379 	read_unlock(&filter->rwlock);
1380 
1381 	return ret;
1382 }
1383 
1384 static void __uprobe_perf_func(struct trace_uprobe *tu,
1385 			       unsigned long func, struct pt_regs *regs,
1386 			       struct uprobe_cpu_buffer **ucbp)
1387 {
1388 	struct trace_event_call *call = trace_probe_event_call(&tu->tp);
1389 	struct uprobe_trace_entry_head *entry;
1390 	struct uprobe_cpu_buffer *ucb;
1391 	struct hlist_head *head;
1392 	void *data;
1393 	int size, esize;
1394 	int rctx;
1395 
1396 #ifdef CONFIG_BPF_EVENTS
1397 	if (bpf_prog_array_valid(call)) {
1398 		u32 ret;
1399 
1400 		ret = bpf_prog_run_array_uprobe(call->prog_array, regs, bpf_prog_run);
1401 		if (!ret)
1402 			return;
1403 	}
1404 #endif /* CONFIG_BPF_EVENTS */
1405 
1406 	esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1407 
1408 	ucb = prepare_uprobe_buffer(tu, regs, ucbp);
1409 	size = esize + ucb->dsize;
1410 	size = ALIGN(size + sizeof(u32), sizeof(u64)) - sizeof(u32);
1411 	if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE, "profile buffer not large enough"))
1412 		return;
1413 
1414 	preempt_disable();
1415 	head = this_cpu_ptr(call->perf_events);
1416 	if (hlist_empty(head))
1417 		goto out;
1418 
1419 	entry = perf_trace_buf_alloc(size, NULL, &rctx);
1420 	if (!entry)
1421 		goto out;
1422 
1423 	if (is_ret_probe(tu)) {
1424 		entry->vaddr[0] = func;
1425 		entry->vaddr[1] = instruction_pointer(regs);
1426 		data = DATAOF_TRACE_ENTRY(entry, true);
1427 	} else {
1428 		entry->vaddr[0] = instruction_pointer(regs);
1429 		data = DATAOF_TRACE_ENTRY(entry, false);
1430 	}
1431 
1432 	memcpy(data, ucb->buf, ucb->dsize);
1433 
1434 	if (size - esize > ucb->dsize)
1435 		memset(data + ucb->dsize, 0, size - esize - ucb->dsize);
1436 
1437 	perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1438 			      head, NULL);
1439  out:
1440 	preempt_enable();
1441 }
1442 
1443 /* uprobe profile handler */
1444 static int uprobe_perf_func(struct trace_uprobe *tu, struct pt_regs *regs,
1445 			    struct uprobe_cpu_buffer **ucbp)
1446 {
1447 	if (!uprobe_perf_filter(&tu->consumer, current->mm))
1448 		return UPROBE_HANDLER_REMOVE;
1449 
1450 	if (!is_ret_probe(tu))
1451 		__uprobe_perf_func(tu, 0, regs, ucbp);
1452 	return 0;
1453 }
1454 
1455 static void uretprobe_perf_func(struct trace_uprobe *tu, unsigned long func,
1456 				struct pt_regs *regs,
1457 				struct uprobe_cpu_buffer **ucbp)
1458 {
1459 	__uprobe_perf_func(tu, func, regs, ucbp);
1460 }
1461 
1462 int bpf_get_uprobe_info(const struct perf_event *event, u32 *fd_type,
1463 			const char **filename, u64 *probe_offset,
1464 			u64 *probe_addr, bool perf_type_tracepoint)
1465 {
1466 	const char *pevent = trace_event_name(event->tp_event);
1467 	const char *group = event->tp_event->class->system;
1468 	struct trace_uprobe *tu;
1469 
1470 	if (perf_type_tracepoint)
1471 		tu = find_probe_event(pevent, group);
1472 	else
1473 		tu = trace_uprobe_primary_from_call(event->tp_event);
1474 	if (!tu)
1475 		return -EINVAL;
1476 
1477 	*fd_type = is_ret_probe(tu) ? BPF_FD_TYPE_URETPROBE
1478 				    : BPF_FD_TYPE_UPROBE;
1479 	*filename = tu->filename;
1480 	*probe_offset = tu->offset;
1481 	*probe_addr = 0;
1482 	return 0;
1483 }
1484 #endif	/* CONFIG_PERF_EVENTS */
1485 
1486 static int
1487 trace_uprobe_register(struct trace_event_call *event, enum trace_reg type,
1488 		      void *data)
1489 {
1490 	struct trace_event_file *file = data;
1491 
1492 	switch (type) {
1493 	case TRACE_REG_REGISTER:
1494 		return probe_event_enable(event, file, NULL);
1495 
1496 	case TRACE_REG_UNREGISTER:
1497 		probe_event_disable(event, file);
1498 		return 0;
1499 
1500 #ifdef CONFIG_PERF_EVENTS
1501 	case TRACE_REG_PERF_REGISTER:
1502 		return probe_event_enable(event, NULL, uprobe_perf_filter);
1503 
1504 	case TRACE_REG_PERF_UNREGISTER:
1505 		probe_event_disable(event, NULL);
1506 		return 0;
1507 
1508 	case TRACE_REG_PERF_OPEN:
1509 		return uprobe_perf_open(event, data);
1510 
1511 	case TRACE_REG_PERF_CLOSE:
1512 		return uprobe_perf_close(event, data);
1513 
1514 #endif
1515 	default:
1516 		return 0;
1517 	}
1518 }
1519 
1520 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs)
1521 {
1522 	struct trace_uprobe *tu;
1523 	struct uprobe_dispatch_data udd;
1524 	struct uprobe_cpu_buffer *ucb = NULL;
1525 	int ret = 0;
1526 
1527 	tu = container_of(con, struct trace_uprobe, consumer);
1528 
1529 	this_cpu_inc(*tu->nhits);
1530 
1531 	udd.tu = tu;
1532 	udd.bp_addr = instruction_pointer(regs);
1533 
1534 	current->utask->vaddr = (unsigned long) &udd;
1535 
1536 	if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1537 		return 0;
1538 
1539 	if (trace_probe_test_flag(&tu->tp, TP_FLAG_TRACE))
1540 		ret |= uprobe_trace_func(tu, regs, &ucb);
1541 
1542 #ifdef CONFIG_PERF_EVENTS
1543 	if (trace_probe_test_flag(&tu->tp, TP_FLAG_PROFILE))
1544 		ret |= uprobe_perf_func(tu, regs, &ucb);
1545 #endif
1546 	uprobe_buffer_put(ucb);
1547 	return ret;
1548 }
1549 
1550 static int uretprobe_dispatcher(struct uprobe_consumer *con,
1551 				unsigned long func, struct pt_regs *regs)
1552 {
1553 	struct trace_uprobe *tu;
1554 	struct uprobe_dispatch_data udd;
1555 	struct uprobe_cpu_buffer *ucb = NULL;
1556 
1557 	tu = container_of(con, struct trace_uprobe, consumer);
1558 
1559 	udd.tu = tu;
1560 	udd.bp_addr = func;
1561 
1562 	current->utask->vaddr = (unsigned long) &udd;
1563 
1564 	if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1565 		return 0;
1566 
1567 	if (trace_probe_test_flag(&tu->tp, TP_FLAG_TRACE))
1568 		uretprobe_trace_func(tu, func, regs, &ucb);
1569 
1570 #ifdef CONFIG_PERF_EVENTS
1571 	if (trace_probe_test_flag(&tu->tp, TP_FLAG_PROFILE))
1572 		uretprobe_perf_func(tu, func, regs, &ucb);
1573 #endif
1574 	uprobe_buffer_put(ucb);
1575 	return 0;
1576 }
1577 
1578 static struct trace_event_functions uprobe_funcs = {
1579 	.trace		= print_uprobe_event
1580 };
1581 
1582 static struct trace_event_fields uprobe_fields_array[] = {
1583 	{ .type = TRACE_FUNCTION_TYPE,
1584 	  .define_fields = uprobe_event_define_fields },
1585 	{}
1586 };
1587 
1588 static inline void init_trace_event_call(struct trace_uprobe *tu)
1589 {
1590 	struct trace_event_call *call = trace_probe_event_call(&tu->tp);
1591 	call->event.funcs = &uprobe_funcs;
1592 	call->class->fields_array = uprobe_fields_array;
1593 
1594 	call->flags = TRACE_EVENT_FL_UPROBE | TRACE_EVENT_FL_CAP_ANY;
1595 	call->class->reg = trace_uprobe_register;
1596 }
1597 
1598 static int register_uprobe_event(struct trace_uprobe *tu)
1599 {
1600 	init_trace_event_call(tu);
1601 
1602 	return trace_probe_register_event_call(&tu->tp);
1603 }
1604 
1605 static int unregister_uprobe_event(struct trace_uprobe *tu)
1606 {
1607 	return trace_probe_unregister_event_call(&tu->tp);
1608 }
1609 
1610 #ifdef CONFIG_PERF_EVENTS
1611 struct trace_event_call *
1612 create_local_trace_uprobe(char *name, unsigned long offs,
1613 			  unsigned long ref_ctr_offset, bool is_return)
1614 {
1615 	enum probe_print_type ptype;
1616 	struct trace_uprobe *tu;
1617 	struct path path;
1618 	int ret;
1619 
1620 	ret = kern_path(name, LOOKUP_FOLLOW, &path);
1621 	if (ret)
1622 		return ERR_PTR(ret);
1623 
1624 	if (!d_is_reg(path.dentry)) {
1625 		path_put(&path);
1626 		return ERR_PTR(-EINVAL);
1627 	}
1628 
1629 	/*
1630 	 * local trace_kprobes are not added to dyn_event, so they are never
1631 	 * searched in find_trace_kprobe(). Therefore, there is no concern of
1632 	 * duplicated name "DUMMY_EVENT" here.
1633 	 */
1634 	tu = alloc_trace_uprobe(UPROBE_EVENT_SYSTEM, "DUMMY_EVENT", 0,
1635 				is_return);
1636 
1637 	if (IS_ERR(tu)) {
1638 		pr_info("Failed to allocate trace_uprobe.(%d)\n",
1639 			(int)PTR_ERR(tu));
1640 		path_put(&path);
1641 		return ERR_CAST(tu);
1642 	}
1643 
1644 	tu->offset = offs;
1645 	tu->path = path;
1646 	tu->ref_ctr_offset = ref_ctr_offset;
1647 	tu->filename = kstrdup(name, GFP_KERNEL);
1648 	if (!tu->filename) {
1649 		ret = -ENOMEM;
1650 		goto error;
1651 	}
1652 
1653 	init_trace_event_call(tu);
1654 
1655 	ptype = is_ret_probe(tu) ? PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL;
1656 	if (traceprobe_set_print_fmt(&tu->tp, ptype) < 0) {
1657 		ret = -ENOMEM;
1658 		goto error;
1659 	}
1660 
1661 	return trace_probe_event_call(&tu->tp);
1662 error:
1663 	free_trace_uprobe(tu);
1664 	return ERR_PTR(ret);
1665 }
1666 
1667 void destroy_local_trace_uprobe(struct trace_event_call *event_call)
1668 {
1669 	struct trace_uprobe *tu;
1670 
1671 	tu = trace_uprobe_primary_from_call(event_call);
1672 
1673 	free_trace_uprobe(tu);
1674 }
1675 #endif /* CONFIG_PERF_EVENTS */
1676 
1677 /* Make a trace interface for controlling probe points */
1678 static __init int init_uprobe_trace(void)
1679 {
1680 	int ret;
1681 
1682 	ret = dyn_event_register(&trace_uprobe_ops);
1683 	if (ret)
1684 		return ret;
1685 
1686 	ret = tracing_init_dentry();
1687 	if (ret)
1688 		return 0;
1689 
1690 	trace_create_file("uprobe_events", TRACE_MODE_WRITE, NULL,
1691 				    NULL, &uprobe_events_ops);
1692 	/* Profile interface */
1693 	trace_create_file("uprobe_profile", TRACE_MODE_READ, NULL,
1694 				    NULL, &uprobe_profile_ops);
1695 	return 0;
1696 }
1697 
1698 fs_initcall(init_uprobe_trace);
1699