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 
is_trace_uprobe(struct dyn_event * ev)70 static bool is_trace_uprobe(struct dyn_event *ev)
71 {
72 	return ev->ops == &trace_uprobe_ops;
73 }
74 
to_trace_uprobe(struct dyn_event * ev)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
adjust_stack_addr(unsigned long addr,unsigned int n)97 static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
98 {
99 	return addr - (n * sizeof(long));
100 }
101 #else
adjust_stack_addr(unsigned long addr,unsigned int n)102 static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
103 {
104 	return addr + (n * sizeof(long));
105 }
106 #endif
107 
get_user_stack_nth(struct pt_regs * regs,unsigned int n)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
probe_mem_read(void * dest,void * src,size_t size)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
probe_mem_read_user(void * dest,void * src,size_t size)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
fetch_store_string(unsigned long addr,void * dest,void * base)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
fetch_store_string_user(unsigned long addr,void * dest,void * base)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
fetch_store_strlen(unsigned long addr)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
fetch_store_strlen_user(unsigned long addr)197 fetch_store_strlen_user(unsigned long addr)
198 {
199 	return fetch_store_strlen(addr);
200 }
201 
translate_user_vaddr(unsigned long file_offset)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
process_fetch_insn(struct fetch_insn * code,void * rec,void * edata,void * dest,void * base)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 }
NOKPROBE_SYMBOL(process_fetch_insn)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 
uprobe_filter_is_empty(struct trace_uprobe_filter * filter)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 
is_ret_probe(struct trace_uprobe * tu)265 static inline bool is_ret_probe(struct trace_uprobe *tu)
266 {
267 	return tu->consumer.ret_handler != NULL;
268 }
269 
trace_uprobe_is_busy(struct dyn_event * ev)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 
trace_uprobe_match_command_head(struct trace_uprobe * tu,int argc,const char ** argv)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 
trace_uprobe_match(const char * system,const char * event,int argc,const char ** argv,struct dyn_event * ev)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 *
trace_uprobe_primary_from_call(struct trace_event_call * call)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 *
alloc_trace_uprobe(const char * group,const char * event,int nargs,bool is_ret)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 
free_trace_uprobe(struct trace_uprobe * tu)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 
find_probe_event(const char * event,const char * group)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 */
unregister_trace_uprobe(struct trace_uprobe * tu)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 
trace_uprobe_has_same_uprobe(struct trace_uprobe * orig,struct trace_uprobe * comp)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 
append_trace_uprobe(struct trace_uprobe * tu,struct trace_uprobe * to)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  */
validate_ref_ctr_offset(struct trace_uprobe * new)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 */
register_trace_uprobe(struct trace_uprobe * tu)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  */
__trace_uprobe_create(int argc,const char ** argv)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 	if (argc - 2 > MAX_TRACE_ARGS)
569 		return -E2BIG;
570 
571 	if (argv[0][1] == ':')
572 		event = &argv[0][2];
573 
574 	if (!strchr(argv[1], '/'))
575 		return -ECANCELED;
576 
577 	filename = kstrdup(argv[1], GFP_KERNEL);
578 	if (!filename)
579 		return -ENOMEM;
580 
581 	/* Find the last occurrence, in case the path contains ':' too. */
582 	arg = strrchr(filename, ':');
583 	if (!arg || !isdigit(arg[1])) {
584 		kfree(filename);
585 		return -ECANCELED;
586 	}
587 
588 	trace_probe_log_init("trace_uprobe", argc, argv);
589 	trace_probe_log_set_index(1);	/* filename is the 2nd argument */
590 
591 	*arg++ = '\0';
592 	ret = kern_path(filename, LOOKUP_FOLLOW, &path);
593 	if (ret) {
594 		trace_probe_log_err(0, FILE_NOT_FOUND);
595 		kfree(filename);
596 		trace_probe_log_clear();
597 		return ret;
598 	}
599 	if (!d_is_reg(path.dentry)) {
600 		trace_probe_log_err(0, NO_REGULAR_FILE);
601 		ret = -EINVAL;
602 		goto fail_address_parse;
603 	}
604 
605 	/* Parse reference counter offset if specified. */
606 	rctr = strchr(arg, '(');
607 	if (rctr) {
608 		rctr_end = strchr(rctr, ')');
609 		if (!rctr_end) {
610 			ret = -EINVAL;
611 			rctr_end = rctr + strlen(rctr);
612 			trace_probe_log_err(rctr_end - filename,
613 					    REFCNT_OPEN_BRACE);
614 			goto fail_address_parse;
615 		} else if (rctr_end[1] != '\0') {
616 			ret = -EINVAL;
617 			trace_probe_log_err(rctr_end + 1 - filename,
618 					    BAD_REFCNT_SUFFIX);
619 			goto fail_address_parse;
620 		}
621 
622 		*rctr++ = '\0';
623 		*rctr_end = '\0';
624 		ret = kstrtoul(rctr, 0, &ref_ctr_offset);
625 		if (ret) {
626 			trace_probe_log_err(rctr - filename, BAD_REFCNT);
627 			goto fail_address_parse;
628 		}
629 	}
630 
631 	/* Check if there is %return suffix */
632 	tmp = strchr(arg, '%');
633 	if (tmp) {
634 		if (!strcmp(tmp, "%return")) {
635 			*tmp = '\0';
636 			is_return = true;
637 		} else {
638 			trace_probe_log_err(tmp - filename, BAD_ADDR_SUFFIX);
639 			ret = -EINVAL;
640 			goto fail_address_parse;
641 		}
642 	}
643 
644 	/* Parse uprobe offset. */
645 	ret = kstrtoul(arg, 0, &offset);
646 	if (ret) {
647 		trace_probe_log_err(arg - filename, BAD_UPROBE_OFFS);
648 		goto fail_address_parse;
649 	}
650 
651 	/* setup a probe */
652 	trace_probe_log_set_index(0);
653 	if (event) {
654 		ret = traceprobe_parse_event_name(&event, &group, gbuf,
655 						  event - argv[0]);
656 		if (ret)
657 			goto fail_address_parse;
658 	}
659 
660 	if (!event) {
661 		char *tail;
662 		char *ptr;
663 
664 		tail = kstrdup(kbasename(filename), GFP_KERNEL);
665 		if (!tail) {
666 			ret = -ENOMEM;
667 			goto fail_address_parse;
668 		}
669 
670 		ptr = strpbrk(tail, ".-_");
671 		if (ptr)
672 			*ptr = '\0';
673 
674 		snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_0x%lx", 'p', tail, offset);
675 		event = buf;
676 		kfree(tail);
677 	}
678 
679 	argc -= 2;
680 	argv += 2;
681 
682 	tu = alloc_trace_uprobe(group, event, argc, is_return);
683 	if (IS_ERR(tu)) {
684 		ret = PTR_ERR(tu);
685 		/* This must return -ENOMEM otherwise there is a bug */
686 		WARN_ON_ONCE(ret != -ENOMEM);
687 		goto fail_address_parse;
688 	}
689 	tu->offset = offset;
690 	tu->ref_ctr_offset = ref_ctr_offset;
691 	tu->path = path;
692 	tu->filename = filename;
693 
694 	/* parse arguments */
695 	for (i = 0; i < argc; i++) {
696 		struct traceprobe_parse_context ctx = {
697 			.flags = (is_return ? TPARG_FL_RETURN : 0) | TPARG_FL_USER,
698 		};
699 
700 		trace_probe_log_set_index(i + 2);
701 		ret = traceprobe_parse_probe_arg(&tu->tp, i, argv[i], &ctx);
702 		traceprobe_finish_parse(&ctx);
703 		if (ret)
704 			goto error;
705 	}
706 
707 	ptype = is_ret_probe(tu) ? PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL;
708 	ret = traceprobe_set_print_fmt(&tu->tp, ptype);
709 	if (ret < 0)
710 		goto error;
711 
712 	ret = register_trace_uprobe(tu);
713 	if (!ret)
714 		goto out;
715 
716 error:
717 	free_trace_uprobe(tu);
718 out:
719 	trace_probe_log_clear();
720 	return ret;
721 
722 fail_address_parse:
723 	trace_probe_log_clear();
724 	path_put(&path);
725 	kfree(filename);
726 
727 	return ret;
728 }
729 
trace_uprobe_create(const char * raw_command)730 int trace_uprobe_create(const char *raw_command)
731 {
732 	return trace_probe_create(raw_command, __trace_uprobe_create);
733 }
734 
create_or_delete_trace_uprobe(const char * raw_command)735 static int create_or_delete_trace_uprobe(const char *raw_command)
736 {
737 	int ret;
738 
739 	if (raw_command[0] == '-')
740 		return dyn_event_release(raw_command, &trace_uprobe_ops);
741 
742 	ret = trace_uprobe_create(raw_command);
743 	return ret == -ECANCELED ? -EINVAL : ret;
744 }
745 
trace_uprobe_release(struct dyn_event * ev)746 static int trace_uprobe_release(struct dyn_event *ev)
747 {
748 	struct trace_uprobe *tu = to_trace_uprobe(ev);
749 
750 	return unregister_trace_uprobe(tu);
751 }
752 
753 /* Probes listing interfaces */
trace_uprobe_show(struct seq_file * m,struct dyn_event * ev)754 static int trace_uprobe_show(struct seq_file *m, struct dyn_event *ev)
755 {
756 	struct trace_uprobe *tu = to_trace_uprobe(ev);
757 	char c = is_ret_probe(tu) ? 'r' : 'p';
758 	int i;
759 
760 	seq_printf(m, "%c:%s/%s %s:0x%0*lx", c, trace_probe_group_name(&tu->tp),
761 			trace_probe_name(&tu->tp), tu->filename,
762 			(int)(sizeof(void *) * 2), tu->offset);
763 
764 	if (tu->ref_ctr_offset)
765 		seq_printf(m, "(0x%lx)", tu->ref_ctr_offset);
766 
767 	for (i = 0; i < tu->tp.nr_args; i++)
768 		seq_printf(m, " %s=%s", tu->tp.args[i].name, tu->tp.args[i].comm);
769 
770 	seq_putc(m, '\n');
771 	return 0;
772 }
773 
probes_seq_show(struct seq_file * m,void * v)774 static int probes_seq_show(struct seq_file *m, void *v)
775 {
776 	struct dyn_event *ev = v;
777 
778 	if (!is_trace_uprobe(ev))
779 		return 0;
780 
781 	return trace_uprobe_show(m, ev);
782 }
783 
784 static const struct seq_operations probes_seq_op = {
785 	.start  = dyn_event_seq_start,
786 	.next   = dyn_event_seq_next,
787 	.stop   = dyn_event_seq_stop,
788 	.show   = probes_seq_show
789 };
790 
probes_open(struct inode * inode,struct file * file)791 static int probes_open(struct inode *inode, struct file *file)
792 {
793 	int ret;
794 
795 	ret = security_locked_down(LOCKDOWN_TRACEFS);
796 	if (ret)
797 		return ret;
798 
799 	if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
800 		ret = dyn_events_release_all(&trace_uprobe_ops);
801 		if (ret)
802 			return ret;
803 	}
804 
805 	return seq_open(file, &probes_seq_op);
806 }
807 
probes_write(struct file * file,const char __user * buffer,size_t count,loff_t * ppos)808 static ssize_t probes_write(struct file *file, const char __user *buffer,
809 			    size_t count, loff_t *ppos)
810 {
811 	return trace_parse_run_command(file, buffer, count, ppos,
812 					create_or_delete_trace_uprobe);
813 }
814 
815 static const struct file_operations uprobe_events_ops = {
816 	.owner		= THIS_MODULE,
817 	.open		= probes_open,
818 	.read		= seq_read,
819 	.llseek		= seq_lseek,
820 	.release	= seq_release,
821 	.write		= probes_write,
822 };
823 
824 /* Probes profiling interfaces */
probes_profile_seq_show(struct seq_file * m,void * v)825 static int probes_profile_seq_show(struct seq_file *m, void *v)
826 {
827 	struct dyn_event *ev = v;
828 	struct trace_uprobe *tu;
829 	unsigned long nhits;
830 	int cpu;
831 
832 	if (!is_trace_uprobe(ev))
833 		return 0;
834 
835 	tu = to_trace_uprobe(ev);
836 
837 	nhits = 0;
838 	for_each_possible_cpu(cpu) {
839 		nhits += per_cpu(*tu->nhits, cpu);
840 	}
841 
842 	seq_printf(m, "  %s %-44s %15lu\n", tu->filename,
843 		   trace_probe_name(&tu->tp), nhits);
844 	return 0;
845 }
846 
847 static const struct seq_operations profile_seq_op = {
848 	.start  = dyn_event_seq_start,
849 	.next   = dyn_event_seq_next,
850 	.stop   = dyn_event_seq_stop,
851 	.show	= probes_profile_seq_show
852 };
853 
profile_open(struct inode * inode,struct file * file)854 static int profile_open(struct inode *inode, struct file *file)
855 {
856 	int ret;
857 
858 	ret = security_locked_down(LOCKDOWN_TRACEFS);
859 	if (ret)
860 		return ret;
861 
862 	return seq_open(file, &profile_seq_op);
863 }
864 
865 static const struct file_operations uprobe_profile_ops = {
866 	.owner		= THIS_MODULE,
867 	.open		= profile_open,
868 	.read		= seq_read,
869 	.llseek		= seq_lseek,
870 	.release	= seq_release,
871 };
872 
873 struct uprobe_cpu_buffer {
874 	struct mutex mutex;
875 	void *buf;
876 	int dsize;
877 };
878 static struct uprobe_cpu_buffer __percpu *uprobe_cpu_buffer;
879 static int uprobe_buffer_refcnt;
880 #define MAX_UCB_BUFFER_SIZE PAGE_SIZE
881 
uprobe_buffer_init(void)882 static int uprobe_buffer_init(void)
883 {
884 	int cpu, err_cpu;
885 
886 	uprobe_cpu_buffer = alloc_percpu(struct uprobe_cpu_buffer);
887 	if (uprobe_cpu_buffer == NULL)
888 		return -ENOMEM;
889 
890 	for_each_possible_cpu(cpu) {
891 		struct page *p = alloc_pages_node(cpu_to_node(cpu),
892 						  GFP_KERNEL, 0);
893 		if (p == NULL) {
894 			err_cpu = cpu;
895 			goto err;
896 		}
897 		per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf = page_address(p);
898 		mutex_init(&per_cpu_ptr(uprobe_cpu_buffer, cpu)->mutex);
899 	}
900 
901 	return 0;
902 
903 err:
904 	for_each_possible_cpu(cpu) {
905 		if (cpu == err_cpu)
906 			break;
907 		free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf);
908 	}
909 
910 	free_percpu(uprobe_cpu_buffer);
911 	return -ENOMEM;
912 }
913 
uprobe_buffer_enable(void)914 static int uprobe_buffer_enable(void)
915 {
916 	int ret = 0;
917 
918 	BUG_ON(!mutex_is_locked(&event_mutex));
919 
920 	if (uprobe_buffer_refcnt++ == 0) {
921 		ret = uprobe_buffer_init();
922 		if (ret < 0)
923 			uprobe_buffer_refcnt--;
924 	}
925 
926 	return ret;
927 }
928 
uprobe_buffer_disable(void)929 static void uprobe_buffer_disable(void)
930 {
931 	int cpu;
932 
933 	BUG_ON(!mutex_is_locked(&event_mutex));
934 
935 	if (--uprobe_buffer_refcnt == 0) {
936 		for_each_possible_cpu(cpu)
937 			free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer,
938 							     cpu)->buf);
939 
940 		free_percpu(uprobe_cpu_buffer);
941 		uprobe_cpu_buffer = NULL;
942 	}
943 }
944 
uprobe_buffer_get(void)945 static struct uprobe_cpu_buffer *uprobe_buffer_get(void)
946 {
947 	struct uprobe_cpu_buffer *ucb;
948 	int cpu;
949 
950 	cpu = raw_smp_processor_id();
951 	ucb = per_cpu_ptr(uprobe_cpu_buffer, cpu);
952 
953 	/*
954 	 * Use per-cpu buffers for fastest access, but we might migrate
955 	 * so the mutex makes sure we have sole access to it.
956 	 */
957 	mutex_lock(&ucb->mutex);
958 
959 	return ucb;
960 }
961 
uprobe_buffer_put(struct uprobe_cpu_buffer * ucb)962 static void uprobe_buffer_put(struct uprobe_cpu_buffer *ucb)
963 {
964 	if (!ucb)
965 		return;
966 	mutex_unlock(&ucb->mutex);
967 }
968 
prepare_uprobe_buffer(struct trace_uprobe * tu,struct pt_regs * regs,struct uprobe_cpu_buffer ** ucbp)969 static struct uprobe_cpu_buffer *prepare_uprobe_buffer(struct trace_uprobe *tu,
970 						       struct pt_regs *regs,
971 						       struct uprobe_cpu_buffer **ucbp)
972 {
973 	struct uprobe_cpu_buffer *ucb;
974 	int dsize, esize;
975 
976 	if (*ucbp)
977 		return *ucbp;
978 
979 	esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
980 	dsize = __get_data_size(&tu->tp, regs, NULL);
981 
982 	ucb = uprobe_buffer_get();
983 	ucb->dsize = tu->tp.size + dsize;
984 
985 	if (WARN_ON_ONCE(ucb->dsize > MAX_UCB_BUFFER_SIZE)) {
986 		ucb->dsize = MAX_UCB_BUFFER_SIZE;
987 		dsize = MAX_UCB_BUFFER_SIZE - tu->tp.size;
988 	}
989 
990 	store_trace_args(ucb->buf, &tu->tp, regs, NULL, esize, dsize);
991 
992 	*ucbp = ucb;
993 	return ucb;
994 }
995 
__uprobe_trace_func(struct trace_uprobe * tu,unsigned long func,struct pt_regs * regs,struct uprobe_cpu_buffer * ucb,struct trace_event_file * trace_file)996 static void __uprobe_trace_func(struct trace_uprobe *tu,
997 				unsigned long func, struct pt_regs *regs,
998 				struct uprobe_cpu_buffer *ucb,
999 				struct trace_event_file *trace_file)
1000 {
1001 	struct uprobe_trace_entry_head *entry;
1002 	struct trace_event_buffer fbuffer;
1003 	void *data;
1004 	int size, esize;
1005 	struct trace_event_call *call = trace_probe_event_call(&tu->tp);
1006 
1007 	WARN_ON(call != trace_file->event_call);
1008 
1009 	if (trace_trigger_soft_disabled(trace_file))
1010 		return;
1011 
1012 	esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1013 	size = esize + ucb->dsize;
1014 	entry = trace_event_buffer_reserve(&fbuffer, trace_file, size);
1015 	if (!entry)
1016 		return;
1017 
1018 	if (is_ret_probe(tu)) {
1019 		entry->vaddr[0] = func;
1020 		entry->vaddr[1] = instruction_pointer(regs);
1021 		data = DATAOF_TRACE_ENTRY(entry, true);
1022 	} else {
1023 		entry->vaddr[0] = instruction_pointer(regs);
1024 		data = DATAOF_TRACE_ENTRY(entry, false);
1025 	}
1026 
1027 	memcpy(data, ucb->buf, ucb->dsize);
1028 
1029 	trace_event_buffer_commit(&fbuffer);
1030 }
1031 
1032 /* uprobe handler */
uprobe_trace_func(struct trace_uprobe * tu,struct pt_regs * regs,struct uprobe_cpu_buffer ** ucbp)1033 static int uprobe_trace_func(struct trace_uprobe *tu, struct pt_regs *regs,
1034 			     struct uprobe_cpu_buffer **ucbp)
1035 {
1036 	struct event_file_link *link;
1037 	struct uprobe_cpu_buffer *ucb;
1038 
1039 	if (is_ret_probe(tu))
1040 		return 0;
1041 
1042 	ucb = prepare_uprobe_buffer(tu, regs, ucbp);
1043 
1044 	rcu_read_lock();
1045 	trace_probe_for_each_link_rcu(link, &tu->tp)
1046 		__uprobe_trace_func(tu, 0, regs, ucb, link->file);
1047 	rcu_read_unlock();
1048 
1049 	return 0;
1050 }
1051 
uretprobe_trace_func(struct trace_uprobe * tu,unsigned long func,struct pt_regs * regs,struct uprobe_cpu_buffer ** ucbp)1052 static void uretprobe_trace_func(struct trace_uprobe *tu, unsigned long func,
1053 				 struct pt_regs *regs,
1054 				 struct uprobe_cpu_buffer **ucbp)
1055 {
1056 	struct event_file_link *link;
1057 	struct uprobe_cpu_buffer *ucb;
1058 
1059 	ucb = prepare_uprobe_buffer(tu, regs, ucbp);
1060 
1061 	rcu_read_lock();
1062 	trace_probe_for_each_link_rcu(link, &tu->tp)
1063 		__uprobe_trace_func(tu, func, regs, ucb, link->file);
1064 	rcu_read_unlock();
1065 }
1066 
1067 /* Event entry printers */
1068 static enum print_line_t
print_uprobe_event(struct trace_iterator * iter,int flags,struct trace_event * event)1069 print_uprobe_event(struct trace_iterator *iter, int flags, struct trace_event *event)
1070 {
1071 	struct uprobe_trace_entry_head *entry;
1072 	struct trace_seq *s = &iter->seq;
1073 	struct trace_uprobe *tu;
1074 	u8 *data;
1075 
1076 	entry = (struct uprobe_trace_entry_head *)iter->ent;
1077 	tu = trace_uprobe_primary_from_call(
1078 		container_of(event, struct trace_event_call, event));
1079 	if (unlikely(!tu))
1080 		goto out;
1081 
1082 	if (is_ret_probe(tu)) {
1083 		trace_seq_printf(s, "%s: (0x%lx <- 0x%lx)",
1084 				 trace_probe_name(&tu->tp),
1085 				 entry->vaddr[1], entry->vaddr[0]);
1086 		data = DATAOF_TRACE_ENTRY(entry, true);
1087 	} else {
1088 		trace_seq_printf(s, "%s: (0x%lx)",
1089 				 trace_probe_name(&tu->tp),
1090 				 entry->vaddr[0]);
1091 		data = DATAOF_TRACE_ENTRY(entry, false);
1092 	}
1093 
1094 	if (trace_probe_print_args(s, tu->tp.args, tu->tp.nr_args, data, entry) < 0)
1095 		goto out;
1096 
1097 	trace_seq_putc(s, '\n');
1098 
1099  out:
1100 	return trace_handle_return(s);
1101 }
1102 
1103 typedef bool (*filter_func_t)(struct uprobe_consumer *self, struct mm_struct *mm);
1104 
trace_uprobe_enable(struct trace_uprobe * tu,filter_func_t filter)1105 static int trace_uprobe_enable(struct trace_uprobe *tu, filter_func_t filter)
1106 {
1107 	struct inode *inode = d_real_inode(tu->path.dentry);
1108 	struct uprobe *uprobe;
1109 
1110 	tu->consumer.filter = filter;
1111 	uprobe = uprobe_register(inode, tu->offset, tu->ref_ctr_offset, &tu->consumer);
1112 	if (IS_ERR(uprobe))
1113 		return PTR_ERR(uprobe);
1114 
1115 	tu->uprobe = uprobe;
1116 	return 0;
1117 }
1118 
__probe_event_disable(struct trace_probe * tp)1119 static void __probe_event_disable(struct trace_probe *tp)
1120 {
1121 	struct trace_uprobe *tu;
1122 	bool sync = false;
1123 
1124 	tu = container_of(tp, struct trace_uprobe, tp);
1125 	WARN_ON(!uprobe_filter_is_empty(tu->tp.event->filter));
1126 
1127 	list_for_each_entry(tu, trace_probe_probe_list(tp), tp.list) {
1128 		if (!tu->uprobe)
1129 			continue;
1130 
1131 		uprobe_unregister_nosync(tu->uprobe, &tu->consumer);
1132 		sync = true;
1133 		tu->uprobe = NULL;
1134 	}
1135 	if (sync)
1136 		uprobe_unregister_sync();
1137 }
1138 
probe_event_enable(struct trace_event_call * call,struct trace_event_file * file,filter_func_t filter)1139 static int probe_event_enable(struct trace_event_call *call,
1140 			struct trace_event_file *file, filter_func_t filter)
1141 {
1142 	struct trace_probe *tp;
1143 	struct trace_uprobe *tu;
1144 	bool enabled;
1145 	int ret;
1146 
1147 	tp = trace_probe_primary_from_call(call);
1148 	if (WARN_ON_ONCE(!tp))
1149 		return -ENODEV;
1150 	enabled = trace_probe_is_enabled(tp);
1151 
1152 	/* This may also change "enabled" state */
1153 	if (file) {
1154 		if (trace_probe_test_flag(tp, TP_FLAG_PROFILE))
1155 			return -EINTR;
1156 
1157 		ret = trace_probe_add_file(tp, file);
1158 		if (ret < 0)
1159 			return ret;
1160 	} else {
1161 		if (trace_probe_test_flag(tp, TP_FLAG_TRACE))
1162 			return -EINTR;
1163 
1164 		trace_probe_set_flag(tp, TP_FLAG_PROFILE);
1165 	}
1166 
1167 	tu = container_of(tp, struct trace_uprobe, tp);
1168 	WARN_ON(!uprobe_filter_is_empty(tu->tp.event->filter));
1169 
1170 	if (enabled)
1171 		return 0;
1172 
1173 	ret = uprobe_buffer_enable();
1174 	if (ret)
1175 		goto err_flags;
1176 
1177 	list_for_each_entry(tu, trace_probe_probe_list(tp), tp.list) {
1178 		ret = trace_uprobe_enable(tu, filter);
1179 		if (ret) {
1180 			__probe_event_disable(tp);
1181 			goto err_buffer;
1182 		}
1183 	}
1184 
1185 	return 0;
1186 
1187  err_buffer:
1188 	uprobe_buffer_disable();
1189 
1190  err_flags:
1191 	if (file)
1192 		trace_probe_remove_file(tp, file);
1193 	else
1194 		trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
1195 
1196 	return ret;
1197 }
1198 
probe_event_disable(struct trace_event_call * call,struct trace_event_file * file)1199 static void probe_event_disable(struct trace_event_call *call,
1200 				struct trace_event_file *file)
1201 {
1202 	struct trace_probe *tp;
1203 
1204 	tp = trace_probe_primary_from_call(call);
1205 	if (WARN_ON_ONCE(!tp))
1206 		return;
1207 
1208 	if (!trace_probe_is_enabled(tp))
1209 		return;
1210 
1211 	if (file) {
1212 		if (trace_probe_remove_file(tp, file) < 0)
1213 			return;
1214 
1215 		if (trace_probe_is_enabled(tp))
1216 			return;
1217 	} else
1218 		trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
1219 
1220 	__probe_event_disable(tp);
1221 	uprobe_buffer_disable();
1222 }
1223 
uprobe_event_define_fields(struct trace_event_call * event_call)1224 static int uprobe_event_define_fields(struct trace_event_call *event_call)
1225 {
1226 	int ret, size;
1227 	struct uprobe_trace_entry_head field;
1228 	struct trace_uprobe *tu;
1229 
1230 	tu = trace_uprobe_primary_from_call(event_call);
1231 	if (unlikely(!tu))
1232 		return -ENODEV;
1233 
1234 	if (is_ret_probe(tu)) {
1235 		DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_FUNC, 0);
1236 		DEFINE_FIELD(unsigned long, vaddr[1], FIELD_STRING_RETIP, 0);
1237 		size = SIZEOF_TRACE_ENTRY(true);
1238 	} else {
1239 		DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_IP, 0);
1240 		size = SIZEOF_TRACE_ENTRY(false);
1241 	}
1242 
1243 	return traceprobe_define_arg_fields(event_call, size, &tu->tp);
1244 }
1245 
1246 #ifdef CONFIG_PERF_EVENTS
1247 static bool
__uprobe_perf_filter(struct trace_uprobe_filter * filter,struct mm_struct * mm)1248 __uprobe_perf_filter(struct trace_uprobe_filter *filter, struct mm_struct *mm)
1249 {
1250 	struct perf_event *event;
1251 
1252 	list_for_each_entry(event, &filter->perf_events, hw.tp_list) {
1253 		if (event->hw.target->mm == mm)
1254 			return true;
1255 	}
1256 
1257 	return false;
1258 }
1259 
1260 static inline bool
trace_uprobe_filter_event(struct trace_uprobe_filter * filter,struct perf_event * event)1261 trace_uprobe_filter_event(struct trace_uprobe_filter *filter,
1262 			  struct perf_event *event)
1263 {
1264 	return __uprobe_perf_filter(filter, event->hw.target->mm);
1265 }
1266 
trace_uprobe_filter_remove(struct trace_uprobe_filter * filter,struct perf_event * event)1267 static bool trace_uprobe_filter_remove(struct trace_uprobe_filter *filter,
1268 				       struct perf_event *event)
1269 {
1270 	bool done;
1271 
1272 	write_lock(&filter->rwlock);
1273 	if (event->hw.target) {
1274 		list_del(&event->hw.tp_list);
1275 		done = filter->nr_systemwide ||
1276 			(event->hw.target->flags & PF_EXITING) ||
1277 			trace_uprobe_filter_event(filter, event);
1278 	} else {
1279 		filter->nr_systemwide--;
1280 		done = filter->nr_systemwide;
1281 	}
1282 	write_unlock(&filter->rwlock);
1283 
1284 	return done;
1285 }
1286 
1287 /* This returns true if the filter always covers target mm */
trace_uprobe_filter_add(struct trace_uprobe_filter * filter,struct perf_event * event)1288 static bool trace_uprobe_filter_add(struct trace_uprobe_filter *filter,
1289 				    struct perf_event *event)
1290 {
1291 	bool done;
1292 
1293 	write_lock(&filter->rwlock);
1294 	if (event->hw.target) {
1295 		/*
1296 		 * event->parent != NULL means copy_process(), we can avoid
1297 		 * uprobe_apply(). current->mm must be probed and we can rely
1298 		 * on dup_mmap() which preserves the already installed bp's.
1299 		 *
1300 		 * attr.enable_on_exec means that exec/mmap will install the
1301 		 * breakpoints we need.
1302 		 */
1303 		done = filter->nr_systemwide ||
1304 			event->parent || event->attr.enable_on_exec ||
1305 			trace_uprobe_filter_event(filter, event);
1306 		list_add(&event->hw.tp_list, &filter->perf_events);
1307 	} else {
1308 		done = filter->nr_systemwide;
1309 		filter->nr_systemwide++;
1310 	}
1311 	write_unlock(&filter->rwlock);
1312 
1313 	return done;
1314 }
1315 
uprobe_perf_close(struct trace_event_call * call,struct perf_event * event)1316 static int uprobe_perf_close(struct trace_event_call *call,
1317 			     struct perf_event *event)
1318 {
1319 	struct trace_probe *tp;
1320 	struct trace_uprobe *tu;
1321 	int ret = 0;
1322 
1323 	tp = trace_probe_primary_from_call(call);
1324 	if (WARN_ON_ONCE(!tp))
1325 		return -ENODEV;
1326 
1327 	tu = container_of(tp, struct trace_uprobe, tp);
1328 	if (trace_uprobe_filter_remove(tu->tp.event->filter, event))
1329 		return 0;
1330 
1331 	list_for_each_entry(tu, trace_probe_probe_list(tp), tp.list) {
1332 		ret = uprobe_apply(tu->uprobe, &tu->consumer, false);
1333 		if (ret)
1334 			break;
1335 	}
1336 
1337 	return ret;
1338 }
1339 
uprobe_perf_open(struct trace_event_call * call,struct perf_event * event)1340 static int uprobe_perf_open(struct trace_event_call *call,
1341 			    struct perf_event *event)
1342 {
1343 	struct trace_probe *tp;
1344 	struct trace_uprobe *tu;
1345 	int err = 0;
1346 
1347 	tp = trace_probe_primary_from_call(call);
1348 	if (WARN_ON_ONCE(!tp))
1349 		return -ENODEV;
1350 
1351 	tu = container_of(tp, struct trace_uprobe, tp);
1352 	if (trace_uprobe_filter_add(tu->tp.event->filter, event))
1353 		return 0;
1354 
1355 	list_for_each_entry(tu, trace_probe_probe_list(tp), tp.list) {
1356 		err = uprobe_apply(tu->uprobe, &tu->consumer, true);
1357 		if (err) {
1358 			uprobe_perf_close(call, event);
1359 			break;
1360 		}
1361 	}
1362 
1363 	return err;
1364 }
1365 
uprobe_perf_filter(struct uprobe_consumer * uc,struct mm_struct * mm)1366 static bool uprobe_perf_filter(struct uprobe_consumer *uc, struct mm_struct *mm)
1367 {
1368 	struct trace_uprobe_filter *filter;
1369 	struct trace_uprobe *tu;
1370 	int ret;
1371 
1372 	tu = container_of(uc, struct trace_uprobe, consumer);
1373 	filter = tu->tp.event->filter;
1374 
1375 	/*
1376 	 * speculative short-circuiting check to avoid unnecessarily taking
1377 	 * filter->rwlock below, if the uprobe has system-wide consumer
1378 	 */
1379 	if (READ_ONCE(filter->nr_systemwide))
1380 		return true;
1381 
1382 	read_lock(&filter->rwlock);
1383 	ret = __uprobe_perf_filter(filter, mm);
1384 	read_unlock(&filter->rwlock);
1385 
1386 	return ret;
1387 }
1388 
__uprobe_perf_func(struct trace_uprobe * tu,unsigned long func,struct pt_regs * regs,struct uprobe_cpu_buffer ** ucbp)1389 static void __uprobe_perf_func(struct trace_uprobe *tu,
1390 			       unsigned long func, struct pt_regs *regs,
1391 			       struct uprobe_cpu_buffer **ucbp)
1392 {
1393 	struct trace_event_call *call = trace_probe_event_call(&tu->tp);
1394 	struct uprobe_trace_entry_head *entry;
1395 	struct uprobe_cpu_buffer *ucb;
1396 	struct hlist_head *head;
1397 	void *data;
1398 	int size, esize;
1399 	int rctx;
1400 
1401 #ifdef CONFIG_BPF_EVENTS
1402 	if (bpf_prog_array_valid(call)) {
1403 		u32 ret;
1404 
1405 		ret = bpf_prog_run_array_uprobe(call->prog_array, regs, bpf_prog_run);
1406 		if (!ret)
1407 			return;
1408 	}
1409 #endif /* CONFIG_BPF_EVENTS */
1410 
1411 	esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1412 
1413 	ucb = prepare_uprobe_buffer(tu, regs, ucbp);
1414 	size = esize + ucb->dsize;
1415 	size = ALIGN(size + sizeof(u32), sizeof(u64)) - sizeof(u32);
1416 	if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE, "profile buffer not large enough"))
1417 		return;
1418 
1419 	preempt_disable();
1420 	head = this_cpu_ptr(call->perf_events);
1421 	if (hlist_empty(head))
1422 		goto out;
1423 
1424 	entry = perf_trace_buf_alloc(size, NULL, &rctx);
1425 	if (!entry)
1426 		goto out;
1427 
1428 	if (is_ret_probe(tu)) {
1429 		entry->vaddr[0] = func;
1430 		entry->vaddr[1] = instruction_pointer(regs);
1431 		data = DATAOF_TRACE_ENTRY(entry, true);
1432 	} else {
1433 		entry->vaddr[0] = instruction_pointer(regs);
1434 		data = DATAOF_TRACE_ENTRY(entry, false);
1435 	}
1436 
1437 	memcpy(data, ucb->buf, ucb->dsize);
1438 
1439 	if (size - esize > ucb->dsize)
1440 		memset(data + ucb->dsize, 0, size - esize - ucb->dsize);
1441 
1442 	perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1443 			      head, NULL);
1444  out:
1445 	preempt_enable();
1446 }
1447 
1448 /* uprobe profile handler */
uprobe_perf_func(struct trace_uprobe * tu,struct pt_regs * regs,struct uprobe_cpu_buffer ** ucbp)1449 static int uprobe_perf_func(struct trace_uprobe *tu, struct pt_regs *regs,
1450 			    struct uprobe_cpu_buffer **ucbp)
1451 {
1452 	if (!uprobe_perf_filter(&tu->consumer, current->mm))
1453 		return UPROBE_HANDLER_REMOVE;
1454 
1455 	if (!is_ret_probe(tu))
1456 		__uprobe_perf_func(tu, 0, regs, ucbp);
1457 	return 0;
1458 }
1459 
uretprobe_perf_func(struct trace_uprobe * tu,unsigned long func,struct pt_regs * regs,struct uprobe_cpu_buffer ** ucbp)1460 static void uretprobe_perf_func(struct trace_uprobe *tu, unsigned long func,
1461 				struct pt_regs *regs,
1462 				struct uprobe_cpu_buffer **ucbp)
1463 {
1464 	__uprobe_perf_func(tu, func, regs, ucbp);
1465 }
1466 
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)1467 int bpf_get_uprobe_info(const struct perf_event *event, u32 *fd_type,
1468 			const char **filename, u64 *probe_offset,
1469 			u64 *probe_addr, bool perf_type_tracepoint)
1470 {
1471 	const char *pevent = trace_event_name(event->tp_event);
1472 	const char *group = event->tp_event->class->system;
1473 	struct trace_uprobe *tu;
1474 
1475 	if (perf_type_tracepoint)
1476 		tu = find_probe_event(pevent, group);
1477 	else
1478 		tu = trace_uprobe_primary_from_call(event->tp_event);
1479 	if (!tu)
1480 		return -EINVAL;
1481 
1482 	*fd_type = is_ret_probe(tu) ? BPF_FD_TYPE_URETPROBE
1483 				    : BPF_FD_TYPE_UPROBE;
1484 	*filename = tu->filename;
1485 	*probe_offset = tu->offset;
1486 	*probe_addr = 0;
1487 	return 0;
1488 }
1489 #endif	/* CONFIG_PERF_EVENTS */
1490 
1491 static int
trace_uprobe_register(struct trace_event_call * event,enum trace_reg type,void * data)1492 trace_uprobe_register(struct trace_event_call *event, enum trace_reg type,
1493 		      void *data)
1494 {
1495 	struct trace_event_file *file = data;
1496 
1497 	switch (type) {
1498 	case TRACE_REG_REGISTER:
1499 		return probe_event_enable(event, file, NULL);
1500 
1501 	case TRACE_REG_UNREGISTER:
1502 		probe_event_disable(event, file);
1503 		return 0;
1504 
1505 #ifdef CONFIG_PERF_EVENTS
1506 	case TRACE_REG_PERF_REGISTER:
1507 		return probe_event_enable(event, NULL, uprobe_perf_filter);
1508 
1509 	case TRACE_REG_PERF_UNREGISTER:
1510 		probe_event_disable(event, NULL);
1511 		return 0;
1512 
1513 	case TRACE_REG_PERF_OPEN:
1514 		return uprobe_perf_open(event, data);
1515 
1516 	case TRACE_REG_PERF_CLOSE:
1517 		return uprobe_perf_close(event, data);
1518 
1519 #endif
1520 	default:
1521 		return 0;
1522 	}
1523 }
1524 
uprobe_dispatcher(struct uprobe_consumer * con,struct pt_regs * regs)1525 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs)
1526 {
1527 	struct trace_uprobe *tu;
1528 	struct uprobe_dispatch_data udd;
1529 	struct uprobe_cpu_buffer *ucb = NULL;
1530 	int ret = 0;
1531 
1532 	tu = container_of(con, struct trace_uprobe, consumer);
1533 
1534 	this_cpu_inc(*tu->nhits);
1535 
1536 	udd.tu = tu;
1537 	udd.bp_addr = instruction_pointer(regs);
1538 
1539 	current->utask->vaddr = (unsigned long) &udd;
1540 
1541 	if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1542 		return 0;
1543 
1544 	if (trace_probe_test_flag(&tu->tp, TP_FLAG_TRACE))
1545 		ret |= uprobe_trace_func(tu, regs, &ucb);
1546 
1547 #ifdef CONFIG_PERF_EVENTS
1548 	if (trace_probe_test_flag(&tu->tp, TP_FLAG_PROFILE))
1549 		ret |= uprobe_perf_func(tu, regs, &ucb);
1550 #endif
1551 	uprobe_buffer_put(ucb);
1552 	return ret;
1553 }
1554 
uretprobe_dispatcher(struct uprobe_consumer * con,unsigned long func,struct pt_regs * regs)1555 static int uretprobe_dispatcher(struct uprobe_consumer *con,
1556 				unsigned long func, struct pt_regs *regs)
1557 {
1558 	struct trace_uprobe *tu;
1559 	struct uprobe_dispatch_data udd;
1560 	struct uprobe_cpu_buffer *ucb = NULL;
1561 
1562 	tu = container_of(con, struct trace_uprobe, consumer);
1563 
1564 	udd.tu = tu;
1565 	udd.bp_addr = func;
1566 
1567 	current->utask->vaddr = (unsigned long) &udd;
1568 
1569 	if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1570 		return 0;
1571 
1572 	if (trace_probe_test_flag(&tu->tp, TP_FLAG_TRACE))
1573 		uretprobe_trace_func(tu, func, regs, &ucb);
1574 
1575 #ifdef CONFIG_PERF_EVENTS
1576 	if (trace_probe_test_flag(&tu->tp, TP_FLAG_PROFILE))
1577 		uretprobe_perf_func(tu, func, regs, &ucb);
1578 #endif
1579 	uprobe_buffer_put(ucb);
1580 	return 0;
1581 }
1582 
1583 static struct trace_event_functions uprobe_funcs = {
1584 	.trace		= print_uprobe_event
1585 };
1586 
1587 static struct trace_event_fields uprobe_fields_array[] = {
1588 	{ .type = TRACE_FUNCTION_TYPE,
1589 	  .define_fields = uprobe_event_define_fields },
1590 	{}
1591 };
1592 
init_trace_event_call(struct trace_uprobe * tu)1593 static inline void init_trace_event_call(struct trace_uprobe *tu)
1594 {
1595 	struct trace_event_call *call = trace_probe_event_call(&tu->tp);
1596 	call->event.funcs = &uprobe_funcs;
1597 	call->class->fields_array = uprobe_fields_array;
1598 
1599 	call->flags = TRACE_EVENT_FL_UPROBE | TRACE_EVENT_FL_CAP_ANY;
1600 	call->class->reg = trace_uprobe_register;
1601 }
1602 
register_uprobe_event(struct trace_uprobe * tu)1603 static int register_uprobe_event(struct trace_uprobe *tu)
1604 {
1605 	init_trace_event_call(tu);
1606 
1607 	return trace_probe_register_event_call(&tu->tp);
1608 }
1609 
unregister_uprobe_event(struct trace_uprobe * tu)1610 static int unregister_uprobe_event(struct trace_uprobe *tu)
1611 {
1612 	return trace_probe_unregister_event_call(&tu->tp);
1613 }
1614 
1615 #ifdef CONFIG_PERF_EVENTS
1616 struct trace_event_call *
create_local_trace_uprobe(char * name,unsigned long offs,unsigned long ref_ctr_offset,bool is_return)1617 create_local_trace_uprobe(char *name, unsigned long offs,
1618 			  unsigned long ref_ctr_offset, bool is_return)
1619 {
1620 	enum probe_print_type ptype;
1621 	struct trace_uprobe *tu;
1622 	struct path path;
1623 	int ret;
1624 
1625 	ret = kern_path(name, LOOKUP_FOLLOW, &path);
1626 	if (ret)
1627 		return ERR_PTR(ret);
1628 
1629 	if (!d_is_reg(path.dentry)) {
1630 		path_put(&path);
1631 		return ERR_PTR(-EINVAL);
1632 	}
1633 
1634 	/*
1635 	 * local trace_kprobes are not added to dyn_event, so they are never
1636 	 * searched in find_trace_kprobe(). Therefore, there is no concern of
1637 	 * duplicated name "DUMMY_EVENT" here.
1638 	 */
1639 	tu = alloc_trace_uprobe(UPROBE_EVENT_SYSTEM, "DUMMY_EVENT", 0,
1640 				is_return);
1641 
1642 	if (IS_ERR(tu)) {
1643 		pr_info("Failed to allocate trace_uprobe.(%d)\n",
1644 			(int)PTR_ERR(tu));
1645 		path_put(&path);
1646 		return ERR_CAST(tu);
1647 	}
1648 
1649 	tu->offset = offs;
1650 	tu->path = path;
1651 	tu->ref_ctr_offset = ref_ctr_offset;
1652 	tu->filename = kstrdup(name, GFP_KERNEL);
1653 	if (!tu->filename) {
1654 		ret = -ENOMEM;
1655 		goto error;
1656 	}
1657 
1658 	init_trace_event_call(tu);
1659 
1660 	ptype = is_ret_probe(tu) ? PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL;
1661 	if (traceprobe_set_print_fmt(&tu->tp, ptype) < 0) {
1662 		ret = -ENOMEM;
1663 		goto error;
1664 	}
1665 
1666 	return trace_probe_event_call(&tu->tp);
1667 error:
1668 	free_trace_uprobe(tu);
1669 	return ERR_PTR(ret);
1670 }
1671 
destroy_local_trace_uprobe(struct trace_event_call * event_call)1672 void destroy_local_trace_uprobe(struct trace_event_call *event_call)
1673 {
1674 	struct trace_uprobe *tu;
1675 
1676 	tu = trace_uprobe_primary_from_call(event_call);
1677 
1678 	free_trace_uprobe(tu);
1679 }
1680 #endif /* CONFIG_PERF_EVENTS */
1681 
1682 /* Make a trace interface for controlling probe points */
init_uprobe_trace(void)1683 static __init int init_uprobe_trace(void)
1684 {
1685 	int ret;
1686 
1687 	ret = dyn_event_register(&trace_uprobe_ops);
1688 	if (ret)
1689 		return ret;
1690 
1691 	ret = tracing_init_dentry();
1692 	if (ret)
1693 		return 0;
1694 
1695 	trace_create_file("uprobe_events", TRACE_MODE_WRITE, NULL,
1696 				    NULL, &uprobe_events_ops);
1697 	/* Profile interface */
1698 	trace_create_file("uprobe_profile", TRACE_MODE_READ, NULL,
1699 				    NULL, &uprobe_profile_ops);
1700 	return 0;
1701 }
1702 
1703 fs_initcall(init_uprobe_trace);
1704