• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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/security.h>
11 #include <linux/ctype.h>
12 #include <linux/module.h>
13 #include <linux/uaccess.h>
14 #include <linux/uprobes.h>
15 #include <linux/namei.h>
16 #include <linux/string.h>
17 #include <linux/rculist.h>
18 
19 #include "trace_dynevent.h"
20 #include "trace_probe.h"
21 #include "trace_probe_tmpl.h"
22 
23 #define UPROBE_EVENT_SYSTEM	"uprobes"
24 
25 struct uprobe_trace_entry_head {
26 	struct trace_entry	ent;
27 	unsigned long		vaddr[];
28 };
29 
30 #define SIZEOF_TRACE_ENTRY(is_return)			\
31 	(sizeof(struct uprobe_trace_entry_head) +	\
32 	 sizeof(unsigned long) * (is_return ? 2 : 1))
33 
34 #define DATAOF_TRACE_ENTRY(entry, is_return)		\
35 	((void*)(entry) + SIZEOF_TRACE_ENTRY(is_return))
36 
37 static int trace_uprobe_create(const char *raw_command);
38 static int trace_uprobe_show(struct seq_file *m, struct dyn_event *ev);
39 static int trace_uprobe_release(struct dyn_event *ev);
40 static bool trace_uprobe_is_busy(struct dyn_event *ev);
41 static bool trace_uprobe_match(const char *system, const char *event,
42 			int argc, const char **argv, struct dyn_event *ev);
43 
44 static struct dyn_event_operations trace_uprobe_ops = {
45 	.create = trace_uprobe_create,
46 	.show = trace_uprobe_show,
47 	.is_busy = trace_uprobe_is_busy,
48 	.free = trace_uprobe_release,
49 	.match = trace_uprobe_match,
50 };
51 
52 /*
53  * uprobe event core functions
54  */
55 struct trace_uprobe {
56 	struct dyn_event		devent;
57 	struct uprobe_consumer		consumer;
58 	struct path			path;
59 	struct inode			*inode;
60 	char				*filename;
61 	unsigned long			offset;
62 	unsigned long			ref_ctr_offset;
63 	unsigned long			nhit;
64 	struct trace_probe		tp;
65 };
66 
is_trace_uprobe(struct dyn_event * ev)67 static bool is_trace_uprobe(struct dyn_event *ev)
68 {
69 	return ev->ops == &trace_uprobe_ops;
70 }
71 
to_trace_uprobe(struct dyn_event * ev)72 static struct trace_uprobe *to_trace_uprobe(struct dyn_event *ev)
73 {
74 	return container_of(ev, struct trace_uprobe, devent);
75 }
76 
77 /**
78  * for_each_trace_uprobe - iterate over the trace_uprobe list
79  * @pos:	the struct trace_uprobe * for each entry
80  * @dpos:	the struct dyn_event * to use as a loop cursor
81  */
82 #define for_each_trace_uprobe(pos, dpos)	\
83 	for_each_dyn_event(dpos)		\
84 		if (is_trace_uprobe(dpos) && (pos = to_trace_uprobe(dpos)))
85 
86 static int register_uprobe_event(struct trace_uprobe *tu);
87 static int unregister_uprobe_event(struct trace_uprobe *tu);
88 
89 struct uprobe_dispatch_data {
90 	struct trace_uprobe	*tu;
91 	unsigned long		bp_addr;
92 };
93 
94 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs);
95 static int uretprobe_dispatcher(struct uprobe_consumer *con,
96 				unsigned long func, struct pt_regs *regs);
97 
98 #ifdef CONFIG_STACK_GROWSUP
adjust_stack_addr(unsigned long addr,unsigned int n)99 static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
100 {
101 	return addr - (n * sizeof(long));
102 }
103 #else
adjust_stack_addr(unsigned long addr,unsigned int n)104 static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
105 {
106 	return addr + (n * sizeof(long));
107 }
108 #endif
109 
get_user_stack_nth(struct pt_regs * regs,unsigned int n)110 static unsigned long get_user_stack_nth(struct pt_regs *regs, unsigned int n)
111 {
112 	unsigned long ret;
113 	unsigned long addr = user_stack_pointer(regs);
114 
115 	addr = adjust_stack_addr(addr, n);
116 
117 	if (copy_from_user(&ret, (void __force __user *) addr, sizeof(ret)))
118 		return 0;
119 
120 	return ret;
121 }
122 
123 /*
124  * Uprobes-specific fetch functions
125  */
126 static nokprobe_inline int
probe_mem_read(void * dest,void * src,size_t size)127 probe_mem_read(void *dest, void *src, size_t size)
128 {
129 	void __user *vaddr = (void __force __user *)src;
130 
131 	return copy_from_user(dest, vaddr, size) ? -EFAULT : 0;
132 }
133 
134 static nokprobe_inline int
probe_mem_read_user(void * dest,void * src,size_t size)135 probe_mem_read_user(void *dest, void *src, size_t size)
136 {
137 	return probe_mem_read(dest, src, size);
138 }
139 
140 /*
141  * Fetch a null-terminated string. Caller MUST set *(u32 *)dest with max
142  * length and relative data location.
143  */
144 static nokprobe_inline int
fetch_store_string(unsigned long addr,void * dest,void * base)145 fetch_store_string(unsigned long addr, void *dest, void *base)
146 {
147 	long ret;
148 	u32 loc = *(u32 *)dest;
149 	int maxlen  = get_loc_len(loc);
150 	u8 *dst = get_loc_data(dest, base);
151 	void __user *src = (void __force __user *) addr;
152 
153 	if (unlikely(!maxlen))
154 		return -ENOMEM;
155 
156 	if (addr == FETCH_TOKEN_COMM)
157 		ret = strlcpy(dst, current->comm, maxlen);
158 	else
159 		ret = strncpy_from_user(dst, src, maxlen);
160 	if (ret >= 0) {
161 		if (ret == maxlen)
162 			dst[ret - 1] = '\0';
163 		else
164 			/*
165 			 * Include the terminating null byte. In this case it
166 			 * was copied by strncpy_from_user but not accounted
167 			 * for in ret.
168 			 */
169 			ret++;
170 		*(u32 *)dest = make_data_loc(ret, (void *)dst - base);
171 	} else
172 		*(u32 *)dest = make_data_loc(0, (void *)dst - base);
173 
174 	return ret;
175 }
176 
177 static nokprobe_inline int
fetch_store_string_user(unsigned long addr,void * dest,void * base)178 fetch_store_string_user(unsigned long addr, void *dest, void *base)
179 {
180 	return fetch_store_string(addr, dest, base);
181 }
182 
183 /* Return the length of string -- including null terminal byte */
184 static nokprobe_inline int
fetch_store_strlen(unsigned long addr)185 fetch_store_strlen(unsigned long addr)
186 {
187 	int len;
188 	void __user *vaddr = (void __force __user *) addr;
189 
190 	if (addr == FETCH_TOKEN_COMM)
191 		len = strlen(current->comm) + 1;
192 	else
193 		len = strnlen_user(vaddr, MAX_STRING_SIZE);
194 
195 	return (len > MAX_STRING_SIZE) ? 0 : len;
196 }
197 
198 static nokprobe_inline int
fetch_store_strlen_user(unsigned long addr)199 fetch_store_strlen_user(unsigned long addr)
200 {
201 	return fetch_store_strlen(addr);
202 }
203 
translate_user_vaddr(unsigned long file_offset)204 static unsigned long translate_user_vaddr(unsigned long file_offset)
205 {
206 	unsigned long base_addr;
207 	struct uprobe_dispatch_data *udd;
208 
209 	udd = (void *) current->utask->vaddr;
210 
211 	base_addr = udd->bp_addr - udd->tu->offset;
212 	return base_addr + file_offset;
213 }
214 
215 /* Note that we don't verify it, since the code does not come from user space */
216 static int
process_fetch_insn(struct fetch_insn * code,void * rec,void * dest,void * base)217 process_fetch_insn(struct fetch_insn *code, void *rec, void *dest,
218 		   void *base)
219 {
220 	struct pt_regs *regs = rec;
221 	unsigned long val;
222 
223 	/* 1st stage: get value from context */
224 	switch (code->op) {
225 	case FETCH_OP_REG:
226 		val = regs_get_register(regs, code->param);
227 		break;
228 	case FETCH_OP_STACK:
229 		val = get_user_stack_nth(regs, code->param);
230 		break;
231 	case FETCH_OP_STACKP:
232 		val = user_stack_pointer(regs);
233 		break;
234 	case FETCH_OP_RETVAL:
235 		val = regs_return_value(regs);
236 		break;
237 	case FETCH_OP_IMM:
238 		val = code->immediate;
239 		break;
240 	case FETCH_OP_COMM:
241 		val = FETCH_TOKEN_COMM;
242 		break;
243 	case FETCH_OP_DATA:
244 		val = (unsigned long)code->data;
245 		break;
246 	case FETCH_OP_FOFFS:
247 		val = translate_user_vaddr(code->immediate);
248 		break;
249 	default:
250 		return -EILSEQ;
251 	}
252 	code++;
253 
254 	return process_fetch_insn_bottom(code, val, dest, base);
255 }
NOKPROBE_SYMBOL(process_fetch_insn)256 NOKPROBE_SYMBOL(process_fetch_insn)
257 
258 static inline void init_trace_uprobe_filter(struct trace_uprobe_filter *filter)
259 {
260 	rwlock_init(&filter->rwlock);
261 	filter->nr_systemwide = 0;
262 	INIT_LIST_HEAD(&filter->perf_events);
263 }
264 
uprobe_filter_is_empty(struct trace_uprobe_filter * filter)265 static inline bool uprobe_filter_is_empty(struct trace_uprobe_filter *filter)
266 {
267 	return !filter->nr_systemwide && list_empty(&filter->perf_events);
268 }
269 
is_ret_probe(struct trace_uprobe * tu)270 static inline bool is_ret_probe(struct trace_uprobe *tu)
271 {
272 	return tu->consumer.ret_handler != NULL;
273 }
274 
trace_uprobe_is_busy(struct dyn_event * ev)275 static bool trace_uprobe_is_busy(struct dyn_event *ev)
276 {
277 	struct trace_uprobe *tu = to_trace_uprobe(ev);
278 
279 	return trace_probe_is_enabled(&tu->tp);
280 }
281 
trace_uprobe_match_command_head(struct trace_uprobe * tu,int argc,const char ** argv)282 static bool trace_uprobe_match_command_head(struct trace_uprobe *tu,
283 					    int argc, const char **argv)
284 {
285 	char buf[MAX_ARGSTR_LEN + 1];
286 	int len;
287 
288 	if (!argc)
289 		return true;
290 
291 	len = strlen(tu->filename);
292 	if (strncmp(tu->filename, argv[0], len) || argv[0][len] != ':')
293 		return false;
294 
295 	if (tu->ref_ctr_offset == 0)
296 		snprintf(buf, sizeof(buf), "0x%0*lx",
297 				(int)(sizeof(void *) * 2), tu->offset);
298 	else
299 		snprintf(buf, sizeof(buf), "0x%0*lx(0x%lx)",
300 				(int)(sizeof(void *) * 2), tu->offset,
301 				tu->ref_ctr_offset);
302 	if (strcmp(buf, &argv[0][len + 1]))
303 		return false;
304 
305 	argc--; argv++;
306 
307 	return trace_probe_match_command_args(&tu->tp, argc, argv);
308 }
309 
trace_uprobe_match(const char * system,const char * event,int argc,const char ** argv,struct dyn_event * ev)310 static bool trace_uprobe_match(const char *system, const char *event,
311 			int argc, const char **argv, struct dyn_event *ev)
312 {
313 	struct trace_uprobe *tu = to_trace_uprobe(ev);
314 
315 	return 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 *
trace_uprobe_primary_from_call(struct trace_event_call * call)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 *
alloc_trace_uprobe(const char * group,const char * event,int nargs,bool is_ret)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(struct_size(tu, tp.args, nargs), GFP_KERNEL);
342 	if (!tu)
343 		return ERR_PTR(-ENOMEM);
344 
345 	ret = trace_probe_init(&tu->tp, event, group, true);
346 	if (ret < 0)
347 		goto error;
348 
349 	dyn_event_init(&tu->devent, &trace_uprobe_ops);
350 	tu->consumer.handler = uprobe_dispatcher;
351 	if (is_ret)
352 		tu->consumer.ret_handler = uretprobe_dispatcher;
353 	init_trace_uprobe_filter(tu->tp.event->filter);
354 	return tu;
355 
356 error:
357 	kfree(tu);
358 
359 	return ERR_PTR(ret);
360 }
361 
free_trace_uprobe(struct trace_uprobe * tu)362 static void free_trace_uprobe(struct trace_uprobe *tu)
363 {
364 	if (!tu)
365 		return;
366 
367 	path_put(&tu->path);
368 	trace_probe_cleanup(&tu->tp);
369 	kfree(tu->filename);
370 	kfree(tu);
371 }
372 
find_probe_event(const char * event,const char * group)373 static struct trace_uprobe *find_probe_event(const char *event, const char *group)
374 {
375 	struct dyn_event *pos;
376 	struct trace_uprobe *tu;
377 
378 	for_each_trace_uprobe(tu, pos)
379 		if (strcmp(trace_probe_name(&tu->tp), event) == 0 &&
380 		    strcmp(trace_probe_group_name(&tu->tp), group) == 0)
381 			return tu;
382 
383 	return NULL;
384 }
385 
386 /* Unregister a trace_uprobe and probe_event */
unregister_trace_uprobe(struct trace_uprobe * tu)387 static int unregister_trace_uprobe(struct trace_uprobe *tu)
388 {
389 	int ret;
390 
391 	if (trace_probe_has_sibling(&tu->tp))
392 		goto unreg;
393 
394 	/* If there's a reference to the dynamic event */
395 	if (trace_event_dyn_busy(trace_probe_event_call(&tu->tp)))
396 		return -EBUSY;
397 
398 	ret = unregister_uprobe_event(tu);
399 	if (ret)
400 		return ret;
401 
402 unreg:
403 	dyn_event_remove(&tu->devent);
404 	trace_probe_unlink(&tu->tp);
405 	free_trace_uprobe(tu);
406 	return 0;
407 }
408 
trace_uprobe_has_same_uprobe(struct trace_uprobe * orig,struct trace_uprobe * comp)409 static bool trace_uprobe_has_same_uprobe(struct trace_uprobe *orig,
410 					 struct trace_uprobe *comp)
411 {
412 	struct trace_probe_event *tpe = orig->tp.event;
413 	struct trace_probe *pos;
414 	struct inode *comp_inode = d_real_inode(comp->path.dentry);
415 	int i;
416 
417 	list_for_each_entry(pos, &tpe->probes, list) {
418 		orig = container_of(pos, struct trace_uprobe, tp);
419 		if (comp_inode != d_real_inode(orig->path.dentry) ||
420 		    comp->offset != orig->offset)
421 			continue;
422 
423 		/*
424 		 * trace_probe_compare_arg_type() ensured that nr_args and
425 		 * each argument name and type are same. Let's compare comm.
426 		 */
427 		for (i = 0; i < orig->tp.nr_args; i++) {
428 			if (strcmp(orig->tp.args[i].comm,
429 				   comp->tp.args[i].comm))
430 				break;
431 		}
432 
433 		if (i == orig->tp.nr_args)
434 			return true;
435 	}
436 
437 	return false;
438 }
439 
append_trace_uprobe(struct trace_uprobe * tu,struct trace_uprobe * to)440 static int append_trace_uprobe(struct trace_uprobe *tu, struct trace_uprobe *to)
441 {
442 	int ret;
443 
444 	ret = trace_probe_compare_arg_type(&tu->tp, &to->tp);
445 	if (ret) {
446 		/* Note that argument starts index = 2 */
447 		trace_probe_log_set_index(ret + 1);
448 		trace_probe_log_err(0, DIFF_ARG_TYPE);
449 		return -EEXIST;
450 	}
451 	if (trace_uprobe_has_same_uprobe(to, tu)) {
452 		trace_probe_log_set_index(0);
453 		trace_probe_log_err(0, SAME_PROBE);
454 		return -EEXIST;
455 	}
456 
457 	/* Append to existing event */
458 	ret = trace_probe_append(&tu->tp, &to->tp);
459 	if (!ret)
460 		dyn_event_add(&tu->devent, trace_probe_event_call(&tu->tp));
461 
462 	return ret;
463 }
464 
465 /*
466  * Uprobe with multiple reference counter is not allowed. i.e.
467  * If inode and offset matches, reference counter offset *must*
468  * match as well. Though, there is one exception: If user is
469  * replacing old trace_uprobe with new one(same group/event),
470  * then we allow same uprobe with new reference counter as far
471  * as the new one does not conflict with any other existing
472  * ones.
473  */
validate_ref_ctr_offset(struct trace_uprobe * new)474 static int validate_ref_ctr_offset(struct trace_uprobe *new)
475 {
476 	struct dyn_event *pos;
477 	struct trace_uprobe *tmp;
478 	struct inode *new_inode = d_real_inode(new->path.dentry);
479 
480 	for_each_trace_uprobe(tmp, pos) {
481 		if (new_inode == d_real_inode(tmp->path.dentry) &&
482 		    new->offset == tmp->offset &&
483 		    new->ref_ctr_offset != tmp->ref_ctr_offset) {
484 			pr_warn("Reference counter offset mismatch.");
485 			return -EINVAL;
486 		}
487 	}
488 	return 0;
489 }
490 
491 /* Register a trace_uprobe and probe_event */
register_trace_uprobe(struct trace_uprobe * tu)492 static int register_trace_uprobe(struct trace_uprobe *tu)
493 {
494 	struct trace_uprobe *old_tu;
495 	int ret;
496 
497 	mutex_lock(&event_mutex);
498 
499 	ret = validate_ref_ctr_offset(tu);
500 	if (ret)
501 		goto end;
502 
503 	/* register as an event */
504 	old_tu = find_probe_event(trace_probe_name(&tu->tp),
505 				  trace_probe_group_name(&tu->tp));
506 	if (old_tu) {
507 		if (is_ret_probe(tu) != is_ret_probe(old_tu)) {
508 			trace_probe_log_set_index(0);
509 			trace_probe_log_err(0, DIFF_PROBE_TYPE);
510 			ret = -EEXIST;
511 		} else {
512 			ret = append_trace_uprobe(tu, old_tu);
513 		}
514 		goto end;
515 	}
516 
517 	ret = register_uprobe_event(tu);
518 	if (ret) {
519 		if (ret == -EEXIST) {
520 			trace_probe_log_set_index(0);
521 			trace_probe_log_err(0, EVENT_EXIST);
522 		} else
523 			pr_warn("Failed to register probe event(%d)\n", ret);
524 		goto end;
525 	}
526 
527 	dyn_event_add(&tu->devent, trace_probe_event_call(&tu->tp));
528 
529 end:
530 	mutex_unlock(&event_mutex);
531 
532 	return ret;
533 }
534 
535 /*
536  * Argument syntax:
537  *  - Add uprobe: p|r[:[GRP/]EVENT] PATH:OFFSET[%return][(REF)] [FETCHARGS]
538  */
__trace_uprobe_create(int argc,const char ** argv)539 static int __trace_uprobe_create(int argc, const char **argv)
540 {
541 	struct trace_uprobe *tu;
542 	const char *event = NULL, *group = UPROBE_EVENT_SYSTEM;
543 	char *arg, *filename, *rctr, *rctr_end, *tmp;
544 	char buf[MAX_EVENT_NAME_LEN];
545 	enum probe_print_type ptype;
546 	struct path path;
547 	unsigned long offset, ref_ctr_offset;
548 	bool is_return = false;
549 	int i, ret;
550 
551 	ret = 0;
552 	ref_ctr_offset = 0;
553 
554 	switch (argv[0][0]) {
555 	case 'r':
556 		is_return = true;
557 		break;
558 	case 'p':
559 		break;
560 	default:
561 		return -ECANCELED;
562 	}
563 
564 	if (argc < 2)
565 		return -ECANCELED;
566 
567 	if (argv[0][1] == ':')
568 		event = &argv[0][2];
569 
570 	if (!strchr(argv[1], '/'))
571 		return -ECANCELED;
572 
573 	filename = kstrdup(argv[1], GFP_KERNEL);
574 	if (!filename)
575 		return -ENOMEM;
576 
577 	/* Find the last occurrence, in case the path contains ':' too. */
578 	arg = strrchr(filename, ':');
579 	if (!arg || !isdigit(arg[1])) {
580 		kfree(filename);
581 		return -ECANCELED;
582 	}
583 
584 	trace_probe_log_init("trace_uprobe", argc, argv);
585 	trace_probe_log_set_index(1);	/* filename is the 2nd argument */
586 
587 	*arg++ = '\0';
588 	ret = kern_path(filename, LOOKUP_FOLLOW, &path);
589 	if (ret) {
590 		trace_probe_log_err(0, FILE_NOT_FOUND);
591 		kfree(filename);
592 		trace_probe_log_clear();
593 		return ret;
594 	}
595 	if (!d_is_reg(path.dentry)) {
596 		trace_probe_log_err(0, NO_REGULAR_FILE);
597 		ret = -EINVAL;
598 		goto fail_address_parse;
599 	}
600 
601 	/* Parse reference counter offset if specified. */
602 	rctr = strchr(arg, '(');
603 	if (rctr) {
604 		rctr_end = strchr(rctr, ')');
605 		if (!rctr_end) {
606 			ret = -EINVAL;
607 			rctr_end = rctr + strlen(rctr);
608 			trace_probe_log_err(rctr_end - filename,
609 					    REFCNT_OPEN_BRACE);
610 			goto fail_address_parse;
611 		} else if (rctr_end[1] != '\0') {
612 			ret = -EINVAL;
613 			trace_probe_log_err(rctr_end + 1 - filename,
614 					    BAD_REFCNT_SUFFIX);
615 			goto fail_address_parse;
616 		}
617 
618 		*rctr++ = '\0';
619 		*rctr_end = '\0';
620 		ret = kstrtoul(rctr, 0, &ref_ctr_offset);
621 		if (ret) {
622 			trace_probe_log_err(rctr - filename, BAD_REFCNT);
623 			goto fail_address_parse;
624 		}
625 	}
626 
627 	/* Check if there is %return suffix */
628 	tmp = strchr(arg, '%');
629 	if (tmp) {
630 		if (!strcmp(tmp, "%return")) {
631 			*tmp = '\0';
632 			is_return = true;
633 		} else {
634 			trace_probe_log_err(tmp - filename, BAD_ADDR_SUFFIX);
635 			ret = -EINVAL;
636 			goto fail_address_parse;
637 		}
638 	}
639 
640 	/* Parse uprobe offset. */
641 	ret = kstrtoul(arg, 0, &offset);
642 	if (ret) {
643 		trace_probe_log_err(arg - filename, BAD_UPROBE_OFFS);
644 		goto fail_address_parse;
645 	}
646 
647 	/* setup a probe */
648 	trace_probe_log_set_index(0);
649 	if (event) {
650 		ret = traceprobe_parse_event_name(&event, &group, buf,
651 						  event - argv[0]);
652 		if (ret)
653 			goto fail_address_parse;
654 	} else {
655 		char *tail;
656 		char *ptr;
657 
658 		tail = kstrdup(kbasename(filename), GFP_KERNEL);
659 		if (!tail) {
660 			ret = -ENOMEM;
661 			goto fail_address_parse;
662 		}
663 
664 		ptr = strpbrk(tail, ".-_");
665 		if (ptr)
666 			*ptr = '\0';
667 
668 		snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_0x%lx", 'p', tail, offset);
669 		event = buf;
670 		kfree(tail);
671 	}
672 
673 	argc -= 2;
674 	argv += 2;
675 
676 	tu = alloc_trace_uprobe(group, event, argc, is_return);
677 	if (IS_ERR(tu)) {
678 		ret = PTR_ERR(tu);
679 		/* This must return -ENOMEM otherwise there is a bug */
680 		WARN_ON_ONCE(ret != -ENOMEM);
681 		goto fail_address_parse;
682 	}
683 	tu->offset = offset;
684 	tu->ref_ctr_offset = ref_ctr_offset;
685 	tu->path = path;
686 	tu->filename = filename;
687 
688 	/* parse arguments */
689 	for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
690 		trace_probe_log_set_index(i + 2);
691 		ret = traceprobe_parse_probe_arg(&tu->tp, i, argv[i],
692 					is_return ? TPARG_FL_RETURN : 0);
693 		if (ret)
694 			goto error;
695 	}
696 
697 	ptype = is_ret_probe(tu) ? PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL;
698 	ret = traceprobe_set_print_fmt(&tu->tp, ptype);
699 	if (ret < 0)
700 		goto error;
701 
702 	ret = register_trace_uprobe(tu);
703 	if (!ret)
704 		goto out;
705 
706 error:
707 	free_trace_uprobe(tu);
708 out:
709 	trace_probe_log_clear();
710 	return ret;
711 
712 fail_address_parse:
713 	trace_probe_log_clear();
714 	path_put(&path);
715 	kfree(filename);
716 
717 	return ret;
718 }
719 
trace_uprobe_create(const char * raw_command)720 int trace_uprobe_create(const char *raw_command)
721 {
722 	return trace_probe_create(raw_command, __trace_uprobe_create);
723 }
724 
create_or_delete_trace_uprobe(const char * raw_command)725 static int create_or_delete_trace_uprobe(const char *raw_command)
726 {
727 	int ret;
728 
729 	if (raw_command[0] == '-')
730 		return dyn_event_release(raw_command, &trace_uprobe_ops);
731 
732 	ret = trace_uprobe_create(raw_command);
733 	return ret == -ECANCELED ? -EINVAL : ret;
734 }
735 
trace_uprobe_release(struct dyn_event * ev)736 static int trace_uprobe_release(struct dyn_event *ev)
737 {
738 	struct trace_uprobe *tu = to_trace_uprobe(ev);
739 
740 	return unregister_trace_uprobe(tu);
741 }
742 
743 /* Probes listing interfaces */
trace_uprobe_show(struct seq_file * m,struct dyn_event * ev)744 static int trace_uprobe_show(struct seq_file *m, struct dyn_event *ev)
745 {
746 	struct trace_uprobe *tu = to_trace_uprobe(ev);
747 	char c = is_ret_probe(tu) ? 'r' : 'p';
748 	int i;
749 
750 	seq_printf(m, "%c:%s/%s %s:0x%0*lx", c, trace_probe_group_name(&tu->tp),
751 			trace_probe_name(&tu->tp), tu->filename,
752 			(int)(sizeof(void *) * 2), tu->offset);
753 
754 	if (tu->ref_ctr_offset)
755 		seq_printf(m, "(0x%lx)", tu->ref_ctr_offset);
756 
757 	for (i = 0; i < tu->tp.nr_args; i++)
758 		seq_printf(m, " %s=%s", tu->tp.args[i].name, tu->tp.args[i].comm);
759 
760 	seq_putc(m, '\n');
761 	return 0;
762 }
763 
probes_seq_show(struct seq_file * m,void * v)764 static int probes_seq_show(struct seq_file *m, void *v)
765 {
766 	struct dyn_event *ev = v;
767 
768 	if (!is_trace_uprobe(ev))
769 		return 0;
770 
771 	return trace_uprobe_show(m, ev);
772 }
773 
774 static const struct seq_operations probes_seq_op = {
775 	.start  = dyn_event_seq_start,
776 	.next   = dyn_event_seq_next,
777 	.stop   = dyn_event_seq_stop,
778 	.show   = probes_seq_show
779 };
780 
probes_open(struct inode * inode,struct file * file)781 static int probes_open(struct inode *inode, struct file *file)
782 {
783 	int ret;
784 
785 	ret = security_locked_down(LOCKDOWN_TRACEFS);
786 	if (ret)
787 		return ret;
788 
789 	if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
790 		ret = dyn_events_release_all(&trace_uprobe_ops);
791 		if (ret)
792 			return ret;
793 	}
794 
795 	return seq_open(file, &probes_seq_op);
796 }
797 
probes_write(struct file * file,const char __user * buffer,size_t count,loff_t * ppos)798 static ssize_t probes_write(struct file *file, const char __user *buffer,
799 			    size_t count, loff_t *ppos)
800 {
801 	return trace_parse_run_command(file, buffer, count, ppos,
802 					create_or_delete_trace_uprobe);
803 }
804 
805 static const struct file_operations uprobe_events_ops = {
806 	.owner		= THIS_MODULE,
807 	.open		= probes_open,
808 	.read		= seq_read,
809 	.llseek		= seq_lseek,
810 	.release	= seq_release,
811 	.write		= probes_write,
812 };
813 
814 /* Probes profiling interfaces */
probes_profile_seq_show(struct seq_file * m,void * v)815 static int probes_profile_seq_show(struct seq_file *m, void *v)
816 {
817 	struct dyn_event *ev = v;
818 	struct trace_uprobe *tu;
819 
820 	if (!is_trace_uprobe(ev))
821 		return 0;
822 
823 	tu = to_trace_uprobe(ev);
824 	seq_printf(m, "  %s %-44s %15lu\n", tu->filename,
825 			trace_probe_name(&tu->tp), tu->nhit);
826 	return 0;
827 }
828 
829 static const struct seq_operations profile_seq_op = {
830 	.start  = dyn_event_seq_start,
831 	.next   = dyn_event_seq_next,
832 	.stop   = dyn_event_seq_stop,
833 	.show	= probes_profile_seq_show
834 };
835 
profile_open(struct inode * inode,struct file * file)836 static int profile_open(struct inode *inode, struct file *file)
837 {
838 	int ret;
839 
840 	ret = security_locked_down(LOCKDOWN_TRACEFS);
841 	if (ret)
842 		return ret;
843 
844 	return seq_open(file, &profile_seq_op);
845 }
846 
847 static const struct file_operations uprobe_profile_ops = {
848 	.owner		= THIS_MODULE,
849 	.open		= profile_open,
850 	.read		= seq_read,
851 	.llseek		= seq_lseek,
852 	.release	= seq_release,
853 };
854 
855 struct uprobe_cpu_buffer {
856 	struct mutex mutex;
857 	void *buf;
858 };
859 static struct uprobe_cpu_buffer __percpu *uprobe_cpu_buffer;
860 static int uprobe_buffer_refcnt;
861 
uprobe_buffer_init(void)862 static int uprobe_buffer_init(void)
863 {
864 	int cpu, err_cpu;
865 
866 	uprobe_cpu_buffer = alloc_percpu(struct uprobe_cpu_buffer);
867 	if (uprobe_cpu_buffer == NULL)
868 		return -ENOMEM;
869 
870 	for_each_possible_cpu(cpu) {
871 		struct page *p = alloc_pages_node(cpu_to_node(cpu),
872 						  GFP_KERNEL, 0);
873 		if (p == NULL) {
874 			err_cpu = cpu;
875 			goto err;
876 		}
877 		per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf = page_address(p);
878 		mutex_init(&per_cpu_ptr(uprobe_cpu_buffer, cpu)->mutex);
879 	}
880 
881 	return 0;
882 
883 err:
884 	for_each_possible_cpu(cpu) {
885 		if (cpu == err_cpu)
886 			break;
887 		free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf);
888 	}
889 
890 	free_percpu(uprobe_cpu_buffer);
891 	return -ENOMEM;
892 }
893 
uprobe_buffer_enable(void)894 static int uprobe_buffer_enable(void)
895 {
896 	int ret = 0;
897 
898 	BUG_ON(!mutex_is_locked(&event_mutex));
899 
900 	if (uprobe_buffer_refcnt++ == 0) {
901 		ret = uprobe_buffer_init();
902 		if (ret < 0)
903 			uprobe_buffer_refcnt--;
904 	}
905 
906 	return ret;
907 }
908 
uprobe_buffer_disable(void)909 static void uprobe_buffer_disable(void)
910 {
911 	int cpu;
912 
913 	BUG_ON(!mutex_is_locked(&event_mutex));
914 
915 	if (--uprobe_buffer_refcnt == 0) {
916 		for_each_possible_cpu(cpu)
917 			free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer,
918 							     cpu)->buf);
919 
920 		free_percpu(uprobe_cpu_buffer);
921 		uprobe_cpu_buffer = NULL;
922 	}
923 }
924 
uprobe_buffer_get(void)925 static struct uprobe_cpu_buffer *uprobe_buffer_get(void)
926 {
927 	struct uprobe_cpu_buffer *ucb;
928 	int cpu;
929 
930 	cpu = raw_smp_processor_id();
931 	ucb = per_cpu_ptr(uprobe_cpu_buffer, cpu);
932 
933 	/*
934 	 * Use per-cpu buffers for fastest access, but we might migrate
935 	 * so the mutex makes sure we have sole access to it.
936 	 */
937 	mutex_lock(&ucb->mutex);
938 
939 	return ucb;
940 }
941 
uprobe_buffer_put(struct uprobe_cpu_buffer * ucb)942 static void uprobe_buffer_put(struct uprobe_cpu_buffer *ucb)
943 {
944 	mutex_unlock(&ucb->mutex);
945 }
946 
__uprobe_trace_func(struct trace_uprobe * tu,unsigned long func,struct pt_regs * regs,struct uprobe_cpu_buffer * ucb,int dsize,struct trace_event_file * trace_file)947 static void __uprobe_trace_func(struct trace_uprobe *tu,
948 				unsigned long func, struct pt_regs *regs,
949 				struct uprobe_cpu_buffer *ucb, int dsize,
950 				struct trace_event_file *trace_file)
951 {
952 	struct uprobe_trace_entry_head *entry;
953 	struct trace_buffer *buffer;
954 	struct ring_buffer_event *event;
955 	void *data;
956 	int size, esize;
957 	struct trace_event_call *call = trace_probe_event_call(&tu->tp);
958 
959 	WARN_ON(call != trace_file->event_call);
960 
961 	if (WARN_ON_ONCE(tu->tp.size + dsize > PAGE_SIZE))
962 		return;
963 
964 	if (trace_trigger_soft_disabled(trace_file))
965 		return;
966 
967 	esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
968 	size = esize + tu->tp.size + dsize;
969 	event = trace_event_buffer_lock_reserve(&buffer, trace_file,
970 						call->event.type, size, 0);
971 	if (!event)
972 		return;
973 
974 	entry = ring_buffer_event_data(event);
975 	if (is_ret_probe(tu)) {
976 		entry->vaddr[0] = func;
977 		entry->vaddr[1] = instruction_pointer(regs);
978 		data = DATAOF_TRACE_ENTRY(entry, true);
979 	} else {
980 		entry->vaddr[0] = instruction_pointer(regs);
981 		data = DATAOF_TRACE_ENTRY(entry, false);
982 	}
983 
984 	memcpy(data, ucb->buf, tu->tp.size + dsize);
985 
986 	event_trigger_unlock_commit(trace_file, buffer, event, entry, 0);
987 }
988 
989 /* uprobe handler */
uprobe_trace_func(struct trace_uprobe * tu,struct pt_regs * regs,struct uprobe_cpu_buffer * ucb,int dsize)990 static int uprobe_trace_func(struct trace_uprobe *tu, struct pt_regs *regs,
991 			     struct uprobe_cpu_buffer *ucb, int dsize)
992 {
993 	struct event_file_link *link;
994 
995 	if (is_ret_probe(tu))
996 		return 0;
997 
998 	rcu_read_lock();
999 	trace_probe_for_each_link_rcu(link, &tu->tp)
1000 		__uprobe_trace_func(tu, 0, regs, ucb, dsize, link->file);
1001 	rcu_read_unlock();
1002 
1003 	return 0;
1004 }
1005 
uretprobe_trace_func(struct trace_uprobe * tu,unsigned long func,struct pt_regs * regs,struct uprobe_cpu_buffer * ucb,int dsize)1006 static void uretprobe_trace_func(struct trace_uprobe *tu, unsigned long func,
1007 				 struct pt_regs *regs,
1008 				 struct uprobe_cpu_buffer *ucb, int dsize)
1009 {
1010 	struct event_file_link *link;
1011 
1012 	rcu_read_lock();
1013 	trace_probe_for_each_link_rcu(link, &tu->tp)
1014 		__uprobe_trace_func(tu, func, regs, ucb, dsize, link->file);
1015 	rcu_read_unlock();
1016 }
1017 
1018 /* Event entry printers */
1019 static enum print_line_t
print_uprobe_event(struct trace_iterator * iter,int flags,struct trace_event * event)1020 print_uprobe_event(struct trace_iterator *iter, int flags, struct trace_event *event)
1021 {
1022 	struct uprobe_trace_entry_head *entry;
1023 	struct trace_seq *s = &iter->seq;
1024 	struct trace_uprobe *tu;
1025 	u8 *data;
1026 
1027 	entry = (struct uprobe_trace_entry_head *)iter->ent;
1028 	tu = trace_uprobe_primary_from_call(
1029 		container_of(event, struct trace_event_call, event));
1030 	if (unlikely(!tu))
1031 		goto out;
1032 
1033 	if (is_ret_probe(tu)) {
1034 		trace_seq_printf(s, "%s: (0x%lx <- 0x%lx)",
1035 				 trace_probe_name(&tu->tp),
1036 				 entry->vaddr[1], entry->vaddr[0]);
1037 		data = DATAOF_TRACE_ENTRY(entry, true);
1038 	} else {
1039 		trace_seq_printf(s, "%s: (0x%lx)",
1040 				 trace_probe_name(&tu->tp),
1041 				 entry->vaddr[0]);
1042 		data = DATAOF_TRACE_ENTRY(entry, false);
1043 	}
1044 
1045 	if (print_probe_args(s, tu->tp.args, tu->tp.nr_args, data, entry) < 0)
1046 		goto out;
1047 
1048 	trace_seq_putc(s, '\n');
1049 
1050  out:
1051 	return trace_handle_return(s);
1052 }
1053 
1054 typedef bool (*filter_func_t)(struct uprobe_consumer *self,
1055 				enum uprobe_filter_ctx ctx,
1056 				struct mm_struct *mm);
1057 
trace_uprobe_enable(struct trace_uprobe * tu,filter_func_t filter)1058 static int trace_uprobe_enable(struct trace_uprobe *tu, filter_func_t filter)
1059 {
1060 	int ret;
1061 
1062 	tu->consumer.filter = filter;
1063 	tu->inode = d_real_inode(tu->path.dentry);
1064 
1065 	if (tu->ref_ctr_offset)
1066 		ret = uprobe_register_refctr(tu->inode, tu->offset,
1067 				tu->ref_ctr_offset, &tu->consumer);
1068 	else
1069 		ret = uprobe_register(tu->inode, tu->offset, &tu->consumer);
1070 
1071 	if (ret)
1072 		tu->inode = NULL;
1073 
1074 	return ret;
1075 }
1076 
__probe_event_disable(struct trace_probe * tp)1077 static void __probe_event_disable(struct trace_probe *tp)
1078 {
1079 	struct trace_probe *pos;
1080 	struct trace_uprobe *tu;
1081 
1082 	tu = container_of(tp, struct trace_uprobe, tp);
1083 	WARN_ON(!uprobe_filter_is_empty(tu->tp.event->filter));
1084 
1085 	list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
1086 		tu = container_of(pos, struct trace_uprobe, tp);
1087 		if (!tu->inode)
1088 			continue;
1089 
1090 		uprobe_unregister(tu->inode, tu->offset, &tu->consumer);
1091 		tu->inode = NULL;
1092 	}
1093 }
1094 
probe_event_enable(struct trace_event_call * call,struct trace_event_file * file,filter_func_t filter)1095 static int probe_event_enable(struct trace_event_call *call,
1096 			struct trace_event_file *file, filter_func_t filter)
1097 {
1098 	struct trace_probe *pos, *tp;
1099 	struct trace_uprobe *tu;
1100 	bool enabled;
1101 	int ret;
1102 
1103 	tp = trace_probe_primary_from_call(call);
1104 	if (WARN_ON_ONCE(!tp))
1105 		return -ENODEV;
1106 	enabled = trace_probe_is_enabled(tp);
1107 
1108 	/* This may also change "enabled" state */
1109 	if (file) {
1110 		if (trace_probe_test_flag(tp, TP_FLAG_PROFILE))
1111 			return -EINTR;
1112 
1113 		ret = trace_probe_add_file(tp, file);
1114 		if (ret < 0)
1115 			return ret;
1116 	} else {
1117 		if (trace_probe_test_flag(tp, TP_FLAG_TRACE))
1118 			return -EINTR;
1119 
1120 		trace_probe_set_flag(tp, TP_FLAG_PROFILE);
1121 	}
1122 
1123 	tu = container_of(tp, struct trace_uprobe, tp);
1124 	WARN_ON(!uprobe_filter_is_empty(tu->tp.event->filter));
1125 
1126 	if (enabled)
1127 		return 0;
1128 
1129 	ret = uprobe_buffer_enable();
1130 	if (ret)
1131 		goto err_flags;
1132 
1133 	list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
1134 		tu = container_of(pos, struct trace_uprobe, tp);
1135 		ret = trace_uprobe_enable(tu, filter);
1136 		if (ret) {
1137 			__probe_event_disable(tp);
1138 			goto err_buffer;
1139 		}
1140 	}
1141 
1142 	return 0;
1143 
1144  err_buffer:
1145 	uprobe_buffer_disable();
1146 
1147  err_flags:
1148 	if (file)
1149 		trace_probe_remove_file(tp, file);
1150 	else
1151 		trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
1152 
1153 	return ret;
1154 }
1155 
probe_event_disable(struct trace_event_call * call,struct trace_event_file * file)1156 static void probe_event_disable(struct trace_event_call *call,
1157 				struct trace_event_file *file)
1158 {
1159 	struct trace_probe *tp;
1160 
1161 	tp = trace_probe_primary_from_call(call);
1162 	if (WARN_ON_ONCE(!tp))
1163 		return;
1164 
1165 	if (!trace_probe_is_enabled(tp))
1166 		return;
1167 
1168 	if (file) {
1169 		if (trace_probe_remove_file(tp, file) < 0)
1170 			return;
1171 
1172 		if (trace_probe_is_enabled(tp))
1173 			return;
1174 	} else
1175 		trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
1176 
1177 	__probe_event_disable(tp);
1178 	uprobe_buffer_disable();
1179 }
1180 
uprobe_event_define_fields(struct trace_event_call * event_call)1181 static int uprobe_event_define_fields(struct trace_event_call *event_call)
1182 {
1183 	int ret, size;
1184 	struct uprobe_trace_entry_head field;
1185 	struct trace_uprobe *tu;
1186 
1187 	tu = trace_uprobe_primary_from_call(event_call);
1188 	if (unlikely(!tu))
1189 		return -ENODEV;
1190 
1191 	if (is_ret_probe(tu)) {
1192 		DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_FUNC, 0);
1193 		DEFINE_FIELD(unsigned long, vaddr[1], FIELD_STRING_RETIP, 0);
1194 		size = SIZEOF_TRACE_ENTRY(true);
1195 	} else {
1196 		DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_IP, 0);
1197 		size = SIZEOF_TRACE_ENTRY(false);
1198 	}
1199 
1200 	return traceprobe_define_arg_fields(event_call, size, &tu->tp);
1201 }
1202 
1203 #ifdef CONFIG_PERF_EVENTS
1204 static bool
__uprobe_perf_filter(struct trace_uprobe_filter * filter,struct mm_struct * mm)1205 __uprobe_perf_filter(struct trace_uprobe_filter *filter, struct mm_struct *mm)
1206 {
1207 	struct perf_event *event;
1208 
1209 	if (filter->nr_systemwide)
1210 		return true;
1211 
1212 	list_for_each_entry(event, &filter->perf_events, hw.tp_list) {
1213 		if (event->hw.target->mm == mm)
1214 			return true;
1215 	}
1216 
1217 	return false;
1218 }
1219 
1220 static inline bool
trace_uprobe_filter_event(struct trace_uprobe_filter * filter,struct perf_event * event)1221 trace_uprobe_filter_event(struct trace_uprobe_filter *filter,
1222 			  struct perf_event *event)
1223 {
1224 	return __uprobe_perf_filter(filter, event->hw.target->mm);
1225 }
1226 
trace_uprobe_filter_remove(struct trace_uprobe_filter * filter,struct perf_event * event)1227 static bool trace_uprobe_filter_remove(struct trace_uprobe_filter *filter,
1228 				       struct perf_event *event)
1229 {
1230 	bool done;
1231 
1232 	write_lock(&filter->rwlock);
1233 	if (event->hw.target) {
1234 		list_del(&event->hw.tp_list);
1235 		done = filter->nr_systemwide ||
1236 			(event->hw.target->flags & PF_EXITING) ||
1237 			trace_uprobe_filter_event(filter, event);
1238 	} else {
1239 		filter->nr_systemwide--;
1240 		done = filter->nr_systemwide;
1241 	}
1242 	write_unlock(&filter->rwlock);
1243 
1244 	return done;
1245 }
1246 
1247 /* This returns true if the filter always covers target mm */
trace_uprobe_filter_add(struct trace_uprobe_filter * filter,struct perf_event * event)1248 static bool trace_uprobe_filter_add(struct trace_uprobe_filter *filter,
1249 				    struct perf_event *event)
1250 {
1251 	bool done;
1252 
1253 	write_lock(&filter->rwlock);
1254 	if (event->hw.target) {
1255 		/*
1256 		 * event->parent != NULL means copy_process(), we can avoid
1257 		 * uprobe_apply(). current->mm must be probed and we can rely
1258 		 * on dup_mmap() which preserves the already installed bp's.
1259 		 *
1260 		 * attr.enable_on_exec means that exec/mmap will install the
1261 		 * breakpoints we need.
1262 		 */
1263 		done = filter->nr_systemwide ||
1264 			event->parent || event->attr.enable_on_exec ||
1265 			trace_uprobe_filter_event(filter, event);
1266 		list_add(&event->hw.tp_list, &filter->perf_events);
1267 	} else {
1268 		done = filter->nr_systemwide;
1269 		filter->nr_systemwide++;
1270 	}
1271 	write_unlock(&filter->rwlock);
1272 
1273 	return done;
1274 }
1275 
uprobe_perf_close(struct trace_event_call * call,struct perf_event * event)1276 static int uprobe_perf_close(struct trace_event_call *call,
1277 			     struct perf_event *event)
1278 {
1279 	struct trace_probe *pos, *tp;
1280 	struct trace_uprobe *tu;
1281 	int ret = 0;
1282 
1283 	tp = trace_probe_primary_from_call(call);
1284 	if (WARN_ON_ONCE(!tp))
1285 		return -ENODEV;
1286 
1287 	tu = container_of(tp, struct trace_uprobe, tp);
1288 	if (trace_uprobe_filter_remove(tu->tp.event->filter, event))
1289 		return 0;
1290 
1291 	list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
1292 		tu = container_of(pos, struct trace_uprobe, tp);
1293 		ret = uprobe_apply(tu->inode, tu->offset, &tu->consumer, false);
1294 		if (ret)
1295 			break;
1296 	}
1297 
1298 	return ret;
1299 }
1300 
uprobe_perf_open(struct trace_event_call * call,struct perf_event * event)1301 static int uprobe_perf_open(struct trace_event_call *call,
1302 			    struct perf_event *event)
1303 {
1304 	struct trace_probe *pos, *tp;
1305 	struct trace_uprobe *tu;
1306 	int err = 0;
1307 
1308 	tp = trace_probe_primary_from_call(call);
1309 	if (WARN_ON_ONCE(!tp))
1310 		return -ENODEV;
1311 
1312 	tu = container_of(tp, struct trace_uprobe, tp);
1313 	if (trace_uprobe_filter_add(tu->tp.event->filter, event))
1314 		return 0;
1315 
1316 	list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
1317 		tu = container_of(pos, struct trace_uprobe, tp);
1318 		err = uprobe_apply(tu->inode, tu->offset, &tu->consumer, true);
1319 		if (err) {
1320 			uprobe_perf_close(call, event);
1321 			break;
1322 		}
1323 	}
1324 
1325 	return err;
1326 }
1327 
uprobe_perf_filter(struct uprobe_consumer * uc,enum uprobe_filter_ctx ctx,struct mm_struct * mm)1328 static bool uprobe_perf_filter(struct uprobe_consumer *uc,
1329 				enum uprobe_filter_ctx ctx, struct mm_struct *mm)
1330 {
1331 	struct trace_uprobe_filter *filter;
1332 	struct trace_uprobe *tu;
1333 	int ret;
1334 
1335 	tu = container_of(uc, struct trace_uprobe, consumer);
1336 	filter = tu->tp.event->filter;
1337 
1338 	read_lock(&filter->rwlock);
1339 	ret = __uprobe_perf_filter(filter, mm);
1340 	read_unlock(&filter->rwlock);
1341 
1342 	return ret;
1343 }
1344 
__uprobe_perf_func(struct trace_uprobe * tu,unsigned long func,struct pt_regs * regs,struct uprobe_cpu_buffer * ucb,int dsize)1345 static void __uprobe_perf_func(struct trace_uprobe *tu,
1346 			       unsigned long func, struct pt_regs *regs,
1347 			       struct uprobe_cpu_buffer *ucb, int dsize)
1348 {
1349 	struct trace_event_call *call = trace_probe_event_call(&tu->tp);
1350 	struct uprobe_trace_entry_head *entry;
1351 	struct hlist_head *head;
1352 	void *data;
1353 	int size, esize;
1354 	int rctx;
1355 
1356 	if (bpf_prog_array_valid(call)) {
1357 		u32 ret;
1358 
1359 		preempt_disable();
1360 		ret = trace_call_bpf(call, regs);
1361 		preempt_enable();
1362 		if (!ret)
1363 			return;
1364 	}
1365 
1366 	esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1367 
1368 	size = esize + tu->tp.size + dsize;
1369 	size = ALIGN(size + sizeof(u32), sizeof(u64)) - sizeof(u32);
1370 	if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE, "profile buffer not large enough"))
1371 		return;
1372 
1373 	preempt_disable();
1374 	head = this_cpu_ptr(call->perf_events);
1375 	if (hlist_empty(head))
1376 		goto out;
1377 
1378 	entry = perf_trace_buf_alloc(size, NULL, &rctx);
1379 	if (!entry)
1380 		goto out;
1381 
1382 	if (is_ret_probe(tu)) {
1383 		entry->vaddr[0] = func;
1384 		entry->vaddr[1] = instruction_pointer(regs);
1385 		data = DATAOF_TRACE_ENTRY(entry, true);
1386 	} else {
1387 		entry->vaddr[0] = instruction_pointer(regs);
1388 		data = DATAOF_TRACE_ENTRY(entry, false);
1389 	}
1390 
1391 	memcpy(data, ucb->buf, tu->tp.size + dsize);
1392 
1393 	if (size - esize > tu->tp.size + dsize) {
1394 		int len = tu->tp.size + dsize;
1395 
1396 		memset(data + len, 0, size - esize - len);
1397 	}
1398 
1399 	perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1400 			      head, NULL);
1401  out:
1402 	preempt_enable();
1403 }
1404 
1405 /* uprobe profile handler */
uprobe_perf_func(struct trace_uprobe * tu,struct pt_regs * regs,struct uprobe_cpu_buffer * ucb,int dsize)1406 static int uprobe_perf_func(struct trace_uprobe *tu, struct pt_regs *regs,
1407 			    struct uprobe_cpu_buffer *ucb, int dsize)
1408 {
1409 	if (!uprobe_perf_filter(&tu->consumer, 0, current->mm))
1410 		return UPROBE_HANDLER_REMOVE;
1411 
1412 	if (!is_ret_probe(tu))
1413 		__uprobe_perf_func(tu, 0, regs, ucb, dsize);
1414 	return 0;
1415 }
1416 
uretprobe_perf_func(struct trace_uprobe * tu,unsigned long func,struct pt_regs * regs,struct uprobe_cpu_buffer * ucb,int dsize)1417 static void uretprobe_perf_func(struct trace_uprobe *tu, unsigned long func,
1418 				struct pt_regs *regs,
1419 				struct uprobe_cpu_buffer *ucb, int dsize)
1420 {
1421 	__uprobe_perf_func(tu, func, regs, ucb, dsize);
1422 }
1423 
bpf_get_uprobe_info(const struct perf_event * event,u32 * fd_type,const char ** filename,u64 * probe_offset,u64 * probe_addr,bool perf_type_tracepoint)1424 int bpf_get_uprobe_info(const struct perf_event *event, u32 *fd_type,
1425 			const char **filename, u64 *probe_offset,
1426 			u64 *probe_addr, bool perf_type_tracepoint)
1427 {
1428 	const char *pevent = trace_event_name(event->tp_event);
1429 	const char *group = event->tp_event->class->system;
1430 	struct trace_uprobe *tu;
1431 
1432 	if (perf_type_tracepoint)
1433 		tu = find_probe_event(pevent, group);
1434 	else
1435 		tu = trace_uprobe_primary_from_call(event->tp_event);
1436 	if (!tu)
1437 		return -EINVAL;
1438 
1439 	*fd_type = is_ret_probe(tu) ? BPF_FD_TYPE_URETPROBE
1440 				    : BPF_FD_TYPE_UPROBE;
1441 	*filename = tu->filename;
1442 	*probe_offset = tu->offset;
1443 	*probe_addr = 0;
1444 	return 0;
1445 }
1446 #endif	/* CONFIG_PERF_EVENTS */
1447 
1448 static int
trace_uprobe_register(struct trace_event_call * event,enum trace_reg type,void * data)1449 trace_uprobe_register(struct trace_event_call *event, enum trace_reg type,
1450 		      void *data)
1451 {
1452 	struct trace_event_file *file = data;
1453 
1454 	switch (type) {
1455 	case TRACE_REG_REGISTER:
1456 		return probe_event_enable(event, file, NULL);
1457 
1458 	case TRACE_REG_UNREGISTER:
1459 		probe_event_disable(event, file);
1460 		return 0;
1461 
1462 #ifdef CONFIG_PERF_EVENTS
1463 	case TRACE_REG_PERF_REGISTER:
1464 		return probe_event_enable(event, NULL, uprobe_perf_filter);
1465 
1466 	case TRACE_REG_PERF_UNREGISTER:
1467 		probe_event_disable(event, NULL);
1468 		return 0;
1469 
1470 	case TRACE_REG_PERF_OPEN:
1471 		return uprobe_perf_open(event, data);
1472 
1473 	case TRACE_REG_PERF_CLOSE:
1474 		return uprobe_perf_close(event, data);
1475 
1476 #endif
1477 	default:
1478 		return 0;
1479 	}
1480 }
1481 
uprobe_dispatcher(struct uprobe_consumer * con,struct pt_regs * regs)1482 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs)
1483 {
1484 	struct trace_uprobe *tu;
1485 	struct uprobe_dispatch_data udd;
1486 	struct uprobe_cpu_buffer *ucb;
1487 	int dsize, esize;
1488 	int ret = 0;
1489 
1490 
1491 	tu = container_of(con, struct trace_uprobe, consumer);
1492 	tu->nhit++;
1493 
1494 	udd.tu = tu;
1495 	udd.bp_addr = instruction_pointer(regs);
1496 
1497 	current->utask->vaddr = (unsigned long) &udd;
1498 
1499 	if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1500 		return 0;
1501 
1502 	dsize = __get_data_size(&tu->tp, regs);
1503 	esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1504 
1505 	ucb = uprobe_buffer_get();
1506 	store_trace_args(ucb->buf, &tu->tp, regs, esize, dsize);
1507 
1508 	if (trace_probe_test_flag(&tu->tp, TP_FLAG_TRACE))
1509 		ret |= uprobe_trace_func(tu, regs, ucb, dsize);
1510 
1511 #ifdef CONFIG_PERF_EVENTS
1512 	if (trace_probe_test_flag(&tu->tp, TP_FLAG_PROFILE))
1513 		ret |= uprobe_perf_func(tu, regs, ucb, dsize);
1514 #endif
1515 	uprobe_buffer_put(ucb);
1516 	return ret;
1517 }
1518 
uretprobe_dispatcher(struct uprobe_consumer * con,unsigned long func,struct pt_regs * regs)1519 static int uretprobe_dispatcher(struct uprobe_consumer *con,
1520 				unsigned long func, struct pt_regs *regs)
1521 {
1522 	struct trace_uprobe *tu;
1523 	struct uprobe_dispatch_data udd;
1524 	struct uprobe_cpu_buffer *ucb;
1525 	int dsize, esize;
1526 
1527 	tu = container_of(con, struct trace_uprobe, consumer);
1528 
1529 	udd.tu = tu;
1530 	udd.bp_addr = func;
1531 
1532 	current->utask->vaddr = (unsigned long) &udd;
1533 
1534 	if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1535 		return 0;
1536 
1537 	dsize = __get_data_size(&tu->tp, regs);
1538 	esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1539 
1540 	ucb = uprobe_buffer_get();
1541 	store_trace_args(ucb->buf, &tu->tp, regs, esize, dsize);
1542 
1543 	if (trace_probe_test_flag(&tu->tp, TP_FLAG_TRACE))
1544 		uretprobe_trace_func(tu, func, regs, ucb, dsize);
1545 
1546 #ifdef CONFIG_PERF_EVENTS
1547 	if (trace_probe_test_flag(&tu->tp, TP_FLAG_PROFILE))
1548 		uretprobe_perf_func(tu, func, regs, ucb, dsize);
1549 #endif
1550 	uprobe_buffer_put(ucb);
1551 	return 0;
1552 }
1553 
1554 static struct trace_event_functions uprobe_funcs = {
1555 	.trace		= print_uprobe_event
1556 };
1557 
1558 static struct trace_event_fields uprobe_fields_array[] = {
1559 	{ .type = TRACE_FUNCTION_TYPE,
1560 	  .define_fields = uprobe_event_define_fields },
1561 	{}
1562 };
1563 
init_trace_event_call(struct trace_uprobe * tu)1564 static inline void init_trace_event_call(struct trace_uprobe *tu)
1565 {
1566 	struct trace_event_call *call = trace_probe_event_call(&tu->tp);
1567 	call->event.funcs = &uprobe_funcs;
1568 	call->class->fields_array = uprobe_fields_array;
1569 
1570 	call->flags = TRACE_EVENT_FL_UPROBE | TRACE_EVENT_FL_CAP_ANY;
1571 	call->class->reg = trace_uprobe_register;
1572 }
1573 
register_uprobe_event(struct trace_uprobe * tu)1574 static int register_uprobe_event(struct trace_uprobe *tu)
1575 {
1576 	init_trace_event_call(tu);
1577 
1578 	return trace_probe_register_event_call(&tu->tp);
1579 }
1580 
unregister_uprobe_event(struct trace_uprobe * tu)1581 static int unregister_uprobe_event(struct trace_uprobe *tu)
1582 {
1583 	return trace_probe_unregister_event_call(&tu->tp);
1584 }
1585 
1586 #ifdef CONFIG_PERF_EVENTS
1587 struct trace_event_call *
create_local_trace_uprobe(char * name,unsigned long offs,unsigned long ref_ctr_offset,bool is_return)1588 create_local_trace_uprobe(char *name, unsigned long offs,
1589 			  unsigned long ref_ctr_offset, bool is_return)
1590 {
1591 	enum probe_print_type ptype;
1592 	struct trace_uprobe *tu;
1593 	struct path path;
1594 	int ret;
1595 
1596 	ret = kern_path(name, LOOKUP_FOLLOW, &path);
1597 	if (ret)
1598 		return ERR_PTR(ret);
1599 
1600 	if (!d_is_reg(path.dentry)) {
1601 		path_put(&path);
1602 		return ERR_PTR(-EINVAL);
1603 	}
1604 
1605 	/*
1606 	 * local trace_kprobes are not added to dyn_event, so they are never
1607 	 * searched in find_trace_kprobe(). Therefore, there is no concern of
1608 	 * duplicated name "DUMMY_EVENT" here.
1609 	 */
1610 	tu = alloc_trace_uprobe(UPROBE_EVENT_SYSTEM, "DUMMY_EVENT", 0,
1611 				is_return);
1612 
1613 	if (IS_ERR(tu)) {
1614 		pr_info("Failed to allocate trace_uprobe.(%d)\n",
1615 			(int)PTR_ERR(tu));
1616 		path_put(&path);
1617 		return ERR_CAST(tu);
1618 	}
1619 
1620 	tu->offset = offs;
1621 	tu->path = path;
1622 	tu->ref_ctr_offset = ref_ctr_offset;
1623 	tu->filename = kstrdup(name, GFP_KERNEL);
1624 	if (!tu->filename) {
1625 		ret = -ENOMEM;
1626 		goto error;
1627 	}
1628 
1629 	init_trace_event_call(tu);
1630 
1631 	ptype = is_ret_probe(tu) ? PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL;
1632 	if (traceprobe_set_print_fmt(&tu->tp, ptype) < 0) {
1633 		ret = -ENOMEM;
1634 		goto error;
1635 	}
1636 
1637 	return trace_probe_event_call(&tu->tp);
1638 error:
1639 	free_trace_uprobe(tu);
1640 	return ERR_PTR(ret);
1641 }
1642 
destroy_local_trace_uprobe(struct trace_event_call * event_call)1643 void destroy_local_trace_uprobe(struct trace_event_call *event_call)
1644 {
1645 	struct trace_uprobe *tu;
1646 
1647 	tu = trace_uprobe_primary_from_call(event_call);
1648 
1649 	free_trace_uprobe(tu);
1650 }
1651 #endif /* CONFIG_PERF_EVENTS */
1652 
1653 /* Make a trace interface for controlling probe points */
init_uprobe_trace(void)1654 static __init int init_uprobe_trace(void)
1655 {
1656 	int ret;
1657 
1658 	ret = dyn_event_register(&trace_uprobe_ops);
1659 	if (ret)
1660 		return ret;
1661 
1662 	ret = tracing_init_dentry();
1663 	if (ret)
1664 		return 0;
1665 
1666 	trace_create_file("uprobe_events", 0644, NULL,
1667 				    NULL, &uprobe_events_ops);
1668 	/* Profile interface */
1669 	trace_create_file("uprobe_profile", 0444, NULL,
1670 				    NULL, &uprobe_profile_ops);
1671 	return 0;
1672 }
1673 
1674 fs_initcall(init_uprobe_trace);
1675