xref: /openbmc/linux/kernel/trace/trace_uprobe.c (revision 206204a1)
1 /*
2  * uprobes-based tracing events
3  *
4  * This program is free software; you can redistribute it and/or modify
5  * it under the terms of the GNU General Public License version 2 as
6  * published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
11  * GNU General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public License
14  * along with this program; if not, write to the Free Software
15  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
16  *
17  * Copyright (C) IBM Corporation, 2010-2012
18  * Author:	Srikar Dronamraju <srikar@linux.vnet.ibm.com>
19  */
20 
21 #include <linux/module.h>
22 #include <linux/uaccess.h>
23 #include <linux/uprobes.h>
24 #include <linux/namei.h>
25 #include <linux/string.h>
26 
27 #include "trace_probe.h"
28 
29 #define UPROBE_EVENT_SYSTEM	"uprobes"
30 
31 struct uprobe_trace_entry_head {
32 	struct trace_entry	ent;
33 	unsigned long		vaddr[];
34 };
35 
36 #define SIZEOF_TRACE_ENTRY(is_return)			\
37 	(sizeof(struct uprobe_trace_entry_head) +	\
38 	 sizeof(unsigned long) * (is_return ? 2 : 1))
39 
40 #define DATAOF_TRACE_ENTRY(entry, is_return)		\
41 	((void*)(entry) + SIZEOF_TRACE_ENTRY(is_return))
42 
43 struct trace_uprobe_filter {
44 	rwlock_t		rwlock;
45 	int			nr_systemwide;
46 	struct list_head	perf_events;
47 };
48 
49 /*
50  * uprobe event core functions
51  */
52 struct trace_uprobe {
53 	struct list_head		list;
54 	struct trace_uprobe_filter	filter;
55 	struct uprobe_consumer		consumer;
56 	struct inode			*inode;
57 	char				*filename;
58 	unsigned long			offset;
59 	unsigned long			nhit;
60 	struct trace_probe		tp;
61 };
62 
63 #define SIZEOF_TRACE_UPROBE(n)				\
64 	(offsetof(struct trace_uprobe, tp.args) +	\
65 	(sizeof(struct probe_arg) * (n)))
66 
67 static int register_uprobe_event(struct trace_uprobe *tu);
68 static int unregister_uprobe_event(struct trace_uprobe *tu);
69 
70 static DEFINE_MUTEX(uprobe_lock);
71 static LIST_HEAD(uprobe_list);
72 
73 struct uprobe_dispatch_data {
74 	struct trace_uprobe	*tu;
75 	unsigned long		bp_addr;
76 };
77 
78 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs);
79 static int uretprobe_dispatcher(struct uprobe_consumer *con,
80 				unsigned long func, struct pt_regs *regs);
81 
82 #ifdef CONFIG_STACK_GROWSUP
83 static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
84 {
85 	return addr - (n * sizeof(long));
86 }
87 #else
88 static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
89 {
90 	return addr + (n * sizeof(long));
91 }
92 #endif
93 
94 static unsigned long get_user_stack_nth(struct pt_regs *regs, unsigned int n)
95 {
96 	unsigned long ret;
97 	unsigned long addr = user_stack_pointer(regs);
98 
99 	addr = adjust_stack_addr(addr, n);
100 
101 	if (copy_from_user(&ret, (void __force __user *) addr, sizeof(ret)))
102 		return 0;
103 
104 	return ret;
105 }
106 
107 /*
108  * Uprobes-specific fetch functions
109  */
110 #define DEFINE_FETCH_stack(type)					\
111 static void FETCH_FUNC_NAME(stack, type)(struct pt_regs *regs,		\
112 					 void *offset, void *dest)	\
113 {									\
114 	*(type *)dest = (type)get_user_stack_nth(regs,			\
115 					      ((unsigned long)offset)); \
116 }
117 DEFINE_BASIC_FETCH_FUNCS(stack)
118 /* No string on the stack entry */
119 #define fetch_stack_string	NULL
120 #define fetch_stack_string_size	NULL
121 
122 #define DEFINE_FETCH_memory(type)					\
123 static void FETCH_FUNC_NAME(memory, type)(struct pt_regs *regs,		\
124 					  void *addr, void *dest)	\
125 {									\
126 	type retval;							\
127 	void __user *vaddr = (void __force __user *) addr;		\
128 									\
129 	if (copy_from_user(&retval, vaddr, sizeof(type)))		\
130 		*(type *)dest = 0;					\
131 	else								\
132 		*(type *) dest = retval;				\
133 }
134 DEFINE_BASIC_FETCH_FUNCS(memory)
135 /*
136  * Fetch a null-terminated string. Caller MUST set *(u32 *)dest with max
137  * length and relative data location.
138  */
139 static void FETCH_FUNC_NAME(memory, string)(struct pt_regs *regs,
140 					    void *addr, void *dest)
141 {
142 	long ret;
143 	u32 rloc = *(u32 *)dest;
144 	int maxlen  = get_rloc_len(rloc);
145 	u8 *dst = get_rloc_data(dest);
146 	void __user *src = (void __force __user *) addr;
147 
148 	if (!maxlen)
149 		return;
150 
151 	ret = strncpy_from_user(dst, src, maxlen);
152 
153 	if (ret < 0) {	/* Failed to fetch string */
154 		((u8 *)get_rloc_data(dest))[0] = '\0';
155 		*(u32 *)dest = make_data_rloc(0, get_rloc_offs(rloc));
156 	} else {
157 		*(u32 *)dest = make_data_rloc(ret, get_rloc_offs(rloc));
158 	}
159 }
160 
161 static void FETCH_FUNC_NAME(memory, string_size)(struct pt_regs *regs,
162 						 void *addr, void *dest)
163 {
164 	int len;
165 	void __user *vaddr = (void __force __user *) addr;
166 
167 	len = strnlen_user(vaddr, MAX_STRING_SIZE);
168 
169 	if (len == 0 || len > MAX_STRING_SIZE)  /* Failed to check length */
170 		*(u32 *)dest = 0;
171 	else
172 		*(u32 *)dest = len;
173 }
174 
175 static unsigned long translate_user_vaddr(void *file_offset)
176 {
177 	unsigned long base_addr;
178 	struct uprobe_dispatch_data *udd;
179 
180 	udd = (void *) current->utask->vaddr;
181 
182 	base_addr = udd->bp_addr - udd->tu->offset;
183 	return base_addr + (unsigned long)file_offset;
184 }
185 
186 #define DEFINE_FETCH_file_offset(type)					\
187 static void FETCH_FUNC_NAME(file_offset, type)(struct pt_regs *regs,	\
188 					       void *offset, void *dest)\
189 {									\
190 	void *vaddr = (void *)translate_user_vaddr(offset);		\
191 									\
192 	FETCH_FUNC_NAME(memory, type)(regs, vaddr, dest);		\
193 }
194 DEFINE_BASIC_FETCH_FUNCS(file_offset)
195 DEFINE_FETCH_file_offset(string)
196 DEFINE_FETCH_file_offset(string_size)
197 
198 /* Fetch type information table */
199 const struct fetch_type uprobes_fetch_type_table[] = {
200 	/* Special types */
201 	[FETCH_TYPE_STRING] = __ASSIGN_FETCH_TYPE("string", string, string,
202 					sizeof(u32), 1, "__data_loc char[]"),
203 	[FETCH_TYPE_STRSIZE] = __ASSIGN_FETCH_TYPE("string_size", u32,
204 					string_size, sizeof(u32), 0, "u32"),
205 	/* Basic types */
206 	ASSIGN_FETCH_TYPE(u8,  u8,  0),
207 	ASSIGN_FETCH_TYPE(u16, u16, 0),
208 	ASSIGN_FETCH_TYPE(u32, u32, 0),
209 	ASSIGN_FETCH_TYPE(u64, u64, 0),
210 	ASSIGN_FETCH_TYPE(s8,  u8,  1),
211 	ASSIGN_FETCH_TYPE(s16, u16, 1),
212 	ASSIGN_FETCH_TYPE(s32, u32, 1),
213 	ASSIGN_FETCH_TYPE(s64, u64, 1),
214 
215 	ASSIGN_FETCH_TYPE_END
216 };
217 
218 static inline void init_trace_uprobe_filter(struct trace_uprobe_filter *filter)
219 {
220 	rwlock_init(&filter->rwlock);
221 	filter->nr_systemwide = 0;
222 	INIT_LIST_HEAD(&filter->perf_events);
223 }
224 
225 static inline bool uprobe_filter_is_empty(struct trace_uprobe_filter *filter)
226 {
227 	return !filter->nr_systemwide && list_empty(&filter->perf_events);
228 }
229 
230 static inline bool is_ret_probe(struct trace_uprobe *tu)
231 {
232 	return tu->consumer.ret_handler != NULL;
233 }
234 
235 /*
236  * Allocate new trace_uprobe and initialize it (including uprobes).
237  */
238 static struct trace_uprobe *
239 alloc_trace_uprobe(const char *group, const char *event, int nargs, bool is_ret)
240 {
241 	struct trace_uprobe *tu;
242 
243 	if (!event || !is_good_name(event))
244 		return ERR_PTR(-EINVAL);
245 
246 	if (!group || !is_good_name(group))
247 		return ERR_PTR(-EINVAL);
248 
249 	tu = kzalloc(SIZEOF_TRACE_UPROBE(nargs), GFP_KERNEL);
250 	if (!tu)
251 		return ERR_PTR(-ENOMEM);
252 
253 	tu->tp.call.class = &tu->tp.class;
254 	tu->tp.call.name = kstrdup(event, GFP_KERNEL);
255 	if (!tu->tp.call.name)
256 		goto error;
257 
258 	tu->tp.class.system = kstrdup(group, GFP_KERNEL);
259 	if (!tu->tp.class.system)
260 		goto error;
261 
262 	INIT_LIST_HEAD(&tu->list);
263 	INIT_LIST_HEAD(&tu->tp.files);
264 	tu->consumer.handler = uprobe_dispatcher;
265 	if (is_ret)
266 		tu->consumer.ret_handler = uretprobe_dispatcher;
267 	init_trace_uprobe_filter(&tu->filter);
268 	tu->tp.call.flags |= TRACE_EVENT_FL_USE_CALL_FILTER;
269 	return tu;
270 
271 error:
272 	kfree(tu->tp.call.name);
273 	kfree(tu);
274 
275 	return ERR_PTR(-ENOMEM);
276 }
277 
278 static void free_trace_uprobe(struct trace_uprobe *tu)
279 {
280 	int i;
281 
282 	for (i = 0; i < tu->tp.nr_args; i++)
283 		traceprobe_free_probe_arg(&tu->tp.args[i]);
284 
285 	iput(tu->inode);
286 	kfree(tu->tp.call.class->system);
287 	kfree(tu->tp.call.name);
288 	kfree(tu->filename);
289 	kfree(tu);
290 }
291 
292 static struct trace_uprobe *find_probe_event(const char *event, const char *group)
293 {
294 	struct trace_uprobe *tu;
295 
296 	list_for_each_entry(tu, &uprobe_list, list)
297 		if (strcmp(ftrace_event_name(&tu->tp.call), event) == 0 &&
298 		    strcmp(tu->tp.call.class->system, group) == 0)
299 			return tu;
300 
301 	return NULL;
302 }
303 
304 /* Unregister a trace_uprobe and probe_event: call with locking uprobe_lock */
305 static int unregister_trace_uprobe(struct trace_uprobe *tu)
306 {
307 	int ret;
308 
309 	ret = unregister_uprobe_event(tu);
310 	if (ret)
311 		return ret;
312 
313 	list_del(&tu->list);
314 	free_trace_uprobe(tu);
315 	return 0;
316 }
317 
318 /* Register a trace_uprobe and probe_event */
319 static int register_trace_uprobe(struct trace_uprobe *tu)
320 {
321 	struct trace_uprobe *old_tu;
322 	int ret;
323 
324 	mutex_lock(&uprobe_lock);
325 
326 	/* register as an event */
327 	old_tu = find_probe_event(ftrace_event_name(&tu->tp.call),
328 			tu->tp.call.class->system);
329 	if (old_tu) {
330 		/* delete old event */
331 		ret = unregister_trace_uprobe(old_tu);
332 		if (ret)
333 			goto end;
334 	}
335 
336 	ret = register_uprobe_event(tu);
337 	if (ret) {
338 		pr_warning("Failed to register probe event(%d)\n", ret);
339 		goto end;
340 	}
341 
342 	list_add_tail(&tu->list, &uprobe_list);
343 
344 end:
345 	mutex_unlock(&uprobe_lock);
346 
347 	return ret;
348 }
349 
350 /*
351  * Argument syntax:
352  *  - Add uprobe: p|r[:[GRP/]EVENT] PATH:OFFSET [FETCHARGS]
353  *
354  *  - Remove uprobe: -:[GRP/]EVENT
355  */
356 static int create_trace_uprobe(int argc, char **argv)
357 {
358 	struct trace_uprobe *tu;
359 	struct inode *inode;
360 	char *arg, *event, *group, *filename;
361 	char buf[MAX_EVENT_NAME_LEN];
362 	struct path path;
363 	unsigned long offset;
364 	bool is_delete, is_return;
365 	int i, ret;
366 
367 	inode = NULL;
368 	ret = 0;
369 	is_delete = false;
370 	is_return = false;
371 	event = NULL;
372 	group = NULL;
373 
374 	/* argc must be >= 1 */
375 	if (argv[0][0] == '-')
376 		is_delete = true;
377 	else if (argv[0][0] == 'r')
378 		is_return = true;
379 	else if (argv[0][0] != 'p') {
380 		pr_info("Probe definition must be started with 'p', 'r' or '-'.\n");
381 		return -EINVAL;
382 	}
383 
384 	if (argv[0][1] == ':') {
385 		event = &argv[0][2];
386 		arg = strchr(event, '/');
387 
388 		if (arg) {
389 			group = event;
390 			event = arg + 1;
391 			event[-1] = '\0';
392 
393 			if (strlen(group) == 0) {
394 				pr_info("Group name is not specified\n");
395 				return -EINVAL;
396 			}
397 		}
398 		if (strlen(event) == 0) {
399 			pr_info("Event name is not specified\n");
400 			return -EINVAL;
401 		}
402 	}
403 	if (!group)
404 		group = UPROBE_EVENT_SYSTEM;
405 
406 	if (is_delete) {
407 		int ret;
408 
409 		if (!event) {
410 			pr_info("Delete command needs an event name.\n");
411 			return -EINVAL;
412 		}
413 		mutex_lock(&uprobe_lock);
414 		tu = find_probe_event(event, group);
415 
416 		if (!tu) {
417 			mutex_unlock(&uprobe_lock);
418 			pr_info("Event %s/%s doesn't exist.\n", group, event);
419 			return -ENOENT;
420 		}
421 		/* delete an event */
422 		ret = unregister_trace_uprobe(tu);
423 		mutex_unlock(&uprobe_lock);
424 		return ret;
425 	}
426 
427 	if (argc < 2) {
428 		pr_info("Probe point is not specified.\n");
429 		return -EINVAL;
430 	}
431 	if (isdigit(argv[1][0])) {
432 		pr_info("probe point must be have a filename.\n");
433 		return -EINVAL;
434 	}
435 	arg = strchr(argv[1], ':');
436 	if (!arg) {
437 		ret = -EINVAL;
438 		goto fail_address_parse;
439 	}
440 
441 	*arg++ = '\0';
442 	filename = argv[1];
443 	ret = kern_path(filename, LOOKUP_FOLLOW, &path);
444 	if (ret)
445 		goto fail_address_parse;
446 
447 	inode = igrab(path.dentry->d_inode);
448 	path_put(&path);
449 
450 	if (!inode || !S_ISREG(inode->i_mode)) {
451 		ret = -EINVAL;
452 		goto fail_address_parse;
453 	}
454 
455 	ret = kstrtoul(arg, 0, &offset);
456 	if (ret)
457 		goto fail_address_parse;
458 
459 	argc -= 2;
460 	argv += 2;
461 
462 	/* setup a probe */
463 	if (!event) {
464 		char *tail;
465 		char *ptr;
466 
467 		tail = kstrdup(kbasename(filename), GFP_KERNEL);
468 		if (!tail) {
469 			ret = -ENOMEM;
470 			goto fail_address_parse;
471 		}
472 
473 		ptr = strpbrk(tail, ".-_");
474 		if (ptr)
475 			*ptr = '\0';
476 
477 		snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_0x%lx", 'p', tail, offset);
478 		event = buf;
479 		kfree(tail);
480 	}
481 
482 	tu = alloc_trace_uprobe(group, event, argc, is_return);
483 	if (IS_ERR(tu)) {
484 		pr_info("Failed to allocate trace_uprobe.(%d)\n", (int)PTR_ERR(tu));
485 		ret = PTR_ERR(tu);
486 		goto fail_address_parse;
487 	}
488 	tu->offset = offset;
489 	tu->inode = inode;
490 	tu->filename = kstrdup(filename, GFP_KERNEL);
491 
492 	if (!tu->filename) {
493 		pr_info("Failed to allocate filename.\n");
494 		ret = -ENOMEM;
495 		goto error;
496 	}
497 
498 	/* parse arguments */
499 	ret = 0;
500 	for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
501 		struct probe_arg *parg = &tu->tp.args[i];
502 
503 		/* Increment count for freeing args in error case */
504 		tu->tp.nr_args++;
505 
506 		/* Parse argument name */
507 		arg = strchr(argv[i], '=');
508 		if (arg) {
509 			*arg++ = '\0';
510 			parg->name = kstrdup(argv[i], GFP_KERNEL);
511 		} else {
512 			arg = argv[i];
513 			/* If argument name is omitted, set "argN" */
514 			snprintf(buf, MAX_EVENT_NAME_LEN, "arg%d", i + 1);
515 			parg->name = kstrdup(buf, GFP_KERNEL);
516 		}
517 
518 		if (!parg->name) {
519 			pr_info("Failed to allocate argument[%d] name.\n", i);
520 			ret = -ENOMEM;
521 			goto error;
522 		}
523 
524 		if (!is_good_name(parg->name)) {
525 			pr_info("Invalid argument[%d] name: %s\n", i, parg->name);
526 			ret = -EINVAL;
527 			goto error;
528 		}
529 
530 		if (traceprobe_conflict_field_name(parg->name, tu->tp.args, i)) {
531 			pr_info("Argument[%d] name '%s' conflicts with "
532 				"another field.\n", i, argv[i]);
533 			ret = -EINVAL;
534 			goto error;
535 		}
536 
537 		/* Parse fetch argument */
538 		ret = traceprobe_parse_probe_arg(arg, &tu->tp.size, parg,
539 						 is_return, false);
540 		if (ret) {
541 			pr_info("Parse error at argument[%d]. (%d)\n", i, ret);
542 			goto error;
543 		}
544 	}
545 
546 	ret = register_trace_uprobe(tu);
547 	if (ret)
548 		goto error;
549 	return 0;
550 
551 error:
552 	free_trace_uprobe(tu);
553 	return ret;
554 
555 fail_address_parse:
556 	if (inode)
557 		iput(inode);
558 
559 	pr_info("Failed to parse address or file.\n");
560 
561 	return ret;
562 }
563 
564 static int cleanup_all_probes(void)
565 {
566 	struct trace_uprobe *tu;
567 	int ret = 0;
568 
569 	mutex_lock(&uprobe_lock);
570 	while (!list_empty(&uprobe_list)) {
571 		tu = list_entry(uprobe_list.next, struct trace_uprobe, list);
572 		ret = unregister_trace_uprobe(tu);
573 		if (ret)
574 			break;
575 	}
576 	mutex_unlock(&uprobe_lock);
577 	return ret;
578 }
579 
580 /* Probes listing interfaces */
581 static void *probes_seq_start(struct seq_file *m, loff_t *pos)
582 {
583 	mutex_lock(&uprobe_lock);
584 	return seq_list_start(&uprobe_list, *pos);
585 }
586 
587 static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
588 {
589 	return seq_list_next(v, &uprobe_list, pos);
590 }
591 
592 static void probes_seq_stop(struct seq_file *m, void *v)
593 {
594 	mutex_unlock(&uprobe_lock);
595 }
596 
597 static int probes_seq_show(struct seq_file *m, void *v)
598 {
599 	struct trace_uprobe *tu = v;
600 	char c = is_ret_probe(tu) ? 'r' : 'p';
601 	int i;
602 
603 	seq_printf(m, "%c:%s/%s", c, tu->tp.call.class->system,
604 			ftrace_event_name(&tu->tp.call));
605 	seq_printf(m, " %s:0x%p", tu->filename, (void *)tu->offset);
606 
607 	for (i = 0; i < tu->tp.nr_args; i++)
608 		seq_printf(m, " %s=%s", tu->tp.args[i].name, tu->tp.args[i].comm);
609 
610 	seq_printf(m, "\n");
611 	return 0;
612 }
613 
614 static const struct seq_operations probes_seq_op = {
615 	.start	= probes_seq_start,
616 	.next	= probes_seq_next,
617 	.stop	= probes_seq_stop,
618 	.show	= probes_seq_show
619 };
620 
621 static int probes_open(struct inode *inode, struct file *file)
622 {
623 	int ret;
624 
625 	if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
626 		ret = cleanup_all_probes();
627 		if (ret)
628 			return ret;
629 	}
630 
631 	return seq_open(file, &probes_seq_op);
632 }
633 
634 static ssize_t probes_write(struct file *file, const char __user *buffer,
635 			    size_t count, loff_t *ppos)
636 {
637 	return traceprobe_probes_write(file, buffer, count, ppos, create_trace_uprobe);
638 }
639 
640 static const struct file_operations uprobe_events_ops = {
641 	.owner		= THIS_MODULE,
642 	.open		= probes_open,
643 	.read		= seq_read,
644 	.llseek		= seq_lseek,
645 	.release	= seq_release,
646 	.write		= probes_write,
647 };
648 
649 /* Probes profiling interfaces */
650 static int probes_profile_seq_show(struct seq_file *m, void *v)
651 {
652 	struct trace_uprobe *tu = v;
653 
654 	seq_printf(m, "  %s %-44s %15lu\n", tu->filename,
655 			ftrace_event_name(&tu->tp.call), tu->nhit);
656 	return 0;
657 }
658 
659 static const struct seq_operations profile_seq_op = {
660 	.start	= probes_seq_start,
661 	.next	= probes_seq_next,
662 	.stop	= probes_seq_stop,
663 	.show	= probes_profile_seq_show
664 };
665 
666 static int profile_open(struct inode *inode, struct file *file)
667 {
668 	return seq_open(file, &profile_seq_op);
669 }
670 
671 static const struct file_operations uprobe_profile_ops = {
672 	.owner		= THIS_MODULE,
673 	.open		= profile_open,
674 	.read		= seq_read,
675 	.llseek		= seq_lseek,
676 	.release	= seq_release,
677 };
678 
679 struct uprobe_cpu_buffer {
680 	struct mutex mutex;
681 	void *buf;
682 };
683 static struct uprobe_cpu_buffer __percpu *uprobe_cpu_buffer;
684 static int uprobe_buffer_refcnt;
685 
686 static int uprobe_buffer_init(void)
687 {
688 	int cpu, err_cpu;
689 
690 	uprobe_cpu_buffer = alloc_percpu(struct uprobe_cpu_buffer);
691 	if (uprobe_cpu_buffer == NULL)
692 		return -ENOMEM;
693 
694 	for_each_possible_cpu(cpu) {
695 		struct page *p = alloc_pages_node(cpu_to_node(cpu),
696 						  GFP_KERNEL, 0);
697 		if (p == NULL) {
698 			err_cpu = cpu;
699 			goto err;
700 		}
701 		per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf = page_address(p);
702 		mutex_init(&per_cpu_ptr(uprobe_cpu_buffer, cpu)->mutex);
703 	}
704 
705 	return 0;
706 
707 err:
708 	for_each_possible_cpu(cpu) {
709 		if (cpu == err_cpu)
710 			break;
711 		free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf);
712 	}
713 
714 	free_percpu(uprobe_cpu_buffer);
715 	return -ENOMEM;
716 }
717 
718 static int uprobe_buffer_enable(void)
719 {
720 	int ret = 0;
721 
722 	BUG_ON(!mutex_is_locked(&event_mutex));
723 
724 	if (uprobe_buffer_refcnt++ == 0) {
725 		ret = uprobe_buffer_init();
726 		if (ret < 0)
727 			uprobe_buffer_refcnt--;
728 	}
729 
730 	return ret;
731 }
732 
733 static void uprobe_buffer_disable(void)
734 {
735 	int cpu;
736 
737 	BUG_ON(!mutex_is_locked(&event_mutex));
738 
739 	if (--uprobe_buffer_refcnt == 0) {
740 		for_each_possible_cpu(cpu)
741 			free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer,
742 							     cpu)->buf);
743 
744 		free_percpu(uprobe_cpu_buffer);
745 		uprobe_cpu_buffer = NULL;
746 	}
747 }
748 
749 static struct uprobe_cpu_buffer *uprobe_buffer_get(void)
750 {
751 	struct uprobe_cpu_buffer *ucb;
752 	int cpu;
753 
754 	cpu = raw_smp_processor_id();
755 	ucb = per_cpu_ptr(uprobe_cpu_buffer, cpu);
756 
757 	/*
758 	 * Use per-cpu buffers for fastest access, but we might migrate
759 	 * so the mutex makes sure we have sole access to it.
760 	 */
761 	mutex_lock(&ucb->mutex);
762 
763 	return ucb;
764 }
765 
766 static void uprobe_buffer_put(struct uprobe_cpu_buffer *ucb)
767 {
768 	mutex_unlock(&ucb->mutex);
769 }
770 
771 static void __uprobe_trace_func(struct trace_uprobe *tu,
772 				unsigned long func, struct pt_regs *regs,
773 				struct uprobe_cpu_buffer *ucb, int dsize,
774 				struct ftrace_event_file *ftrace_file)
775 {
776 	struct uprobe_trace_entry_head *entry;
777 	struct ring_buffer_event *event;
778 	struct ring_buffer *buffer;
779 	void *data;
780 	int size, esize;
781 	struct ftrace_event_call *call = &tu->tp.call;
782 
783 	WARN_ON(call != ftrace_file->event_call);
784 
785 	if (WARN_ON_ONCE(tu->tp.size + dsize > PAGE_SIZE))
786 		return;
787 
788 	if (ftrace_trigger_soft_disabled(ftrace_file))
789 		return;
790 
791 	esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
792 	size = esize + tu->tp.size + dsize;
793 	event = trace_event_buffer_lock_reserve(&buffer, ftrace_file,
794 						call->event.type, size, 0, 0);
795 	if (!event)
796 		return;
797 
798 	entry = ring_buffer_event_data(event);
799 	if (is_ret_probe(tu)) {
800 		entry->vaddr[0] = func;
801 		entry->vaddr[1] = instruction_pointer(regs);
802 		data = DATAOF_TRACE_ENTRY(entry, true);
803 	} else {
804 		entry->vaddr[0] = instruction_pointer(regs);
805 		data = DATAOF_TRACE_ENTRY(entry, false);
806 	}
807 
808 	memcpy(data, ucb->buf, tu->tp.size + dsize);
809 
810 	event_trigger_unlock_commit(ftrace_file, buffer, event, entry, 0, 0);
811 }
812 
813 /* uprobe handler */
814 static int uprobe_trace_func(struct trace_uprobe *tu, struct pt_regs *regs,
815 			     struct uprobe_cpu_buffer *ucb, int dsize)
816 {
817 	struct event_file_link *link;
818 
819 	if (is_ret_probe(tu))
820 		return 0;
821 
822 	rcu_read_lock();
823 	list_for_each_entry_rcu(link, &tu->tp.files, list)
824 		__uprobe_trace_func(tu, 0, regs, ucb, dsize, link->file);
825 	rcu_read_unlock();
826 
827 	return 0;
828 }
829 
830 static void uretprobe_trace_func(struct trace_uprobe *tu, unsigned long func,
831 				 struct pt_regs *regs,
832 				 struct uprobe_cpu_buffer *ucb, int dsize)
833 {
834 	struct event_file_link *link;
835 
836 	rcu_read_lock();
837 	list_for_each_entry_rcu(link, &tu->tp.files, list)
838 		__uprobe_trace_func(tu, func, regs, ucb, dsize, link->file);
839 	rcu_read_unlock();
840 }
841 
842 /* Event entry printers */
843 static enum print_line_t
844 print_uprobe_event(struct trace_iterator *iter, int flags, struct trace_event *event)
845 {
846 	struct uprobe_trace_entry_head *entry;
847 	struct trace_seq *s = &iter->seq;
848 	struct trace_uprobe *tu;
849 	u8 *data;
850 	int i;
851 
852 	entry = (struct uprobe_trace_entry_head *)iter->ent;
853 	tu = container_of(event, struct trace_uprobe, tp.call.event);
854 
855 	if (is_ret_probe(tu)) {
856 		if (!trace_seq_printf(s, "%s: (0x%lx <- 0x%lx)",
857 					ftrace_event_name(&tu->tp.call),
858 					entry->vaddr[1], entry->vaddr[0]))
859 			goto partial;
860 		data = DATAOF_TRACE_ENTRY(entry, true);
861 	} else {
862 		if (!trace_seq_printf(s, "%s: (0x%lx)",
863 					ftrace_event_name(&tu->tp.call),
864 					entry->vaddr[0]))
865 			goto partial;
866 		data = DATAOF_TRACE_ENTRY(entry, false);
867 	}
868 
869 	for (i = 0; i < tu->tp.nr_args; i++) {
870 		struct probe_arg *parg = &tu->tp.args[i];
871 
872 		if (!parg->type->print(s, parg->name, data + parg->offset, entry))
873 			goto partial;
874 	}
875 
876 	if (trace_seq_puts(s, "\n"))
877 		return TRACE_TYPE_HANDLED;
878 
879 partial:
880 	return TRACE_TYPE_PARTIAL_LINE;
881 }
882 
883 typedef bool (*filter_func_t)(struct uprobe_consumer *self,
884 				enum uprobe_filter_ctx ctx,
885 				struct mm_struct *mm);
886 
887 static int
888 probe_event_enable(struct trace_uprobe *tu, struct ftrace_event_file *file,
889 		   filter_func_t filter)
890 {
891 	bool enabled = trace_probe_is_enabled(&tu->tp);
892 	struct event_file_link *link = NULL;
893 	int ret;
894 
895 	if (file) {
896 		link = kmalloc(sizeof(*link), GFP_KERNEL);
897 		if (!link)
898 			return -ENOMEM;
899 
900 		link->file = file;
901 		list_add_tail_rcu(&link->list, &tu->tp.files);
902 
903 		tu->tp.flags |= TP_FLAG_TRACE;
904 	} else
905 		tu->tp.flags |= TP_FLAG_PROFILE;
906 
907 	ret = uprobe_buffer_enable();
908 	if (ret < 0)
909 		return ret;
910 
911 	WARN_ON(!uprobe_filter_is_empty(&tu->filter));
912 
913 	if (enabled)
914 		return 0;
915 
916 	tu->consumer.filter = filter;
917 	ret = uprobe_register(tu->inode, tu->offset, &tu->consumer);
918 	if (ret) {
919 		if (file) {
920 			list_del(&link->list);
921 			kfree(link);
922 			tu->tp.flags &= ~TP_FLAG_TRACE;
923 		} else
924 			tu->tp.flags &= ~TP_FLAG_PROFILE;
925 	}
926 
927 	return ret;
928 }
929 
930 static void
931 probe_event_disable(struct trace_uprobe *tu, struct ftrace_event_file *file)
932 {
933 	if (!trace_probe_is_enabled(&tu->tp))
934 		return;
935 
936 	if (file) {
937 		struct event_file_link *link;
938 
939 		link = find_event_file_link(&tu->tp, file);
940 		if (!link)
941 			return;
942 
943 		list_del_rcu(&link->list);
944 		/* synchronize with u{,ret}probe_trace_func */
945 		synchronize_sched();
946 		kfree(link);
947 
948 		if (!list_empty(&tu->tp.files))
949 			return;
950 	}
951 
952 	WARN_ON(!uprobe_filter_is_empty(&tu->filter));
953 
954 	uprobe_unregister(tu->inode, tu->offset, &tu->consumer);
955 	tu->tp.flags &= file ? ~TP_FLAG_TRACE : ~TP_FLAG_PROFILE;
956 
957 	uprobe_buffer_disable();
958 }
959 
960 static int uprobe_event_define_fields(struct ftrace_event_call *event_call)
961 {
962 	int ret, i, size;
963 	struct uprobe_trace_entry_head field;
964 	struct trace_uprobe *tu = event_call->data;
965 
966 	if (is_ret_probe(tu)) {
967 		DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_FUNC, 0);
968 		DEFINE_FIELD(unsigned long, vaddr[1], FIELD_STRING_RETIP, 0);
969 		size = SIZEOF_TRACE_ENTRY(true);
970 	} else {
971 		DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_IP, 0);
972 		size = SIZEOF_TRACE_ENTRY(false);
973 	}
974 	/* Set argument names as fields */
975 	for (i = 0; i < tu->tp.nr_args; i++) {
976 		struct probe_arg *parg = &tu->tp.args[i];
977 
978 		ret = trace_define_field(event_call, parg->type->fmttype,
979 					 parg->name, size + parg->offset,
980 					 parg->type->size, parg->type->is_signed,
981 					 FILTER_OTHER);
982 
983 		if (ret)
984 			return ret;
985 	}
986 	return 0;
987 }
988 
989 #ifdef CONFIG_PERF_EVENTS
990 static bool
991 __uprobe_perf_filter(struct trace_uprobe_filter *filter, struct mm_struct *mm)
992 {
993 	struct perf_event *event;
994 
995 	if (filter->nr_systemwide)
996 		return true;
997 
998 	list_for_each_entry(event, &filter->perf_events, hw.tp_list) {
999 		if (event->hw.tp_target->mm == mm)
1000 			return true;
1001 	}
1002 
1003 	return false;
1004 }
1005 
1006 static inline bool
1007 uprobe_filter_event(struct trace_uprobe *tu, struct perf_event *event)
1008 {
1009 	return __uprobe_perf_filter(&tu->filter, event->hw.tp_target->mm);
1010 }
1011 
1012 static int uprobe_perf_close(struct trace_uprobe *tu, struct perf_event *event)
1013 {
1014 	bool done;
1015 
1016 	write_lock(&tu->filter.rwlock);
1017 	if (event->hw.tp_target) {
1018 		list_del(&event->hw.tp_list);
1019 		done = tu->filter.nr_systemwide ||
1020 			(event->hw.tp_target->flags & PF_EXITING) ||
1021 			uprobe_filter_event(tu, event);
1022 	} else {
1023 		tu->filter.nr_systemwide--;
1024 		done = tu->filter.nr_systemwide;
1025 	}
1026 	write_unlock(&tu->filter.rwlock);
1027 
1028 	if (!done)
1029 		return uprobe_apply(tu->inode, tu->offset, &tu->consumer, false);
1030 
1031 	return 0;
1032 }
1033 
1034 static int uprobe_perf_open(struct trace_uprobe *tu, struct perf_event *event)
1035 {
1036 	bool done;
1037 	int err;
1038 
1039 	write_lock(&tu->filter.rwlock);
1040 	if (event->hw.tp_target) {
1041 		/*
1042 		 * event->parent != NULL means copy_process(), we can avoid
1043 		 * uprobe_apply(). current->mm must be probed and we can rely
1044 		 * on dup_mmap() which preserves the already installed bp's.
1045 		 *
1046 		 * attr.enable_on_exec means that exec/mmap will install the
1047 		 * breakpoints we need.
1048 		 */
1049 		done = tu->filter.nr_systemwide ||
1050 			event->parent || event->attr.enable_on_exec ||
1051 			uprobe_filter_event(tu, event);
1052 		list_add(&event->hw.tp_list, &tu->filter.perf_events);
1053 	} else {
1054 		done = tu->filter.nr_systemwide;
1055 		tu->filter.nr_systemwide++;
1056 	}
1057 	write_unlock(&tu->filter.rwlock);
1058 
1059 	err = 0;
1060 	if (!done) {
1061 		err = uprobe_apply(tu->inode, tu->offset, &tu->consumer, true);
1062 		if (err)
1063 			uprobe_perf_close(tu, event);
1064 	}
1065 	return err;
1066 }
1067 
1068 static bool uprobe_perf_filter(struct uprobe_consumer *uc,
1069 				enum uprobe_filter_ctx ctx, struct mm_struct *mm)
1070 {
1071 	struct trace_uprobe *tu;
1072 	int ret;
1073 
1074 	tu = container_of(uc, struct trace_uprobe, consumer);
1075 	read_lock(&tu->filter.rwlock);
1076 	ret = __uprobe_perf_filter(&tu->filter, mm);
1077 	read_unlock(&tu->filter.rwlock);
1078 
1079 	return ret;
1080 }
1081 
1082 static void __uprobe_perf_func(struct trace_uprobe *tu,
1083 			       unsigned long func, struct pt_regs *regs,
1084 			       struct uprobe_cpu_buffer *ucb, int dsize)
1085 {
1086 	struct ftrace_event_call *call = &tu->tp.call;
1087 	struct uprobe_trace_entry_head *entry;
1088 	struct hlist_head *head;
1089 	void *data;
1090 	int size, esize;
1091 	int rctx;
1092 
1093 	esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1094 
1095 	size = esize + tu->tp.size + dsize;
1096 	size = ALIGN(size + sizeof(u32), sizeof(u64)) - sizeof(u32);
1097 	if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE, "profile buffer not large enough"))
1098 		return;
1099 
1100 	preempt_disable();
1101 	head = this_cpu_ptr(call->perf_events);
1102 	if (hlist_empty(head))
1103 		goto out;
1104 
1105 	entry = perf_trace_buf_prepare(size, call->event.type, regs, &rctx);
1106 	if (!entry)
1107 		goto out;
1108 
1109 	if (is_ret_probe(tu)) {
1110 		entry->vaddr[0] = func;
1111 		entry->vaddr[1] = instruction_pointer(regs);
1112 		data = DATAOF_TRACE_ENTRY(entry, true);
1113 	} else {
1114 		entry->vaddr[0] = instruction_pointer(regs);
1115 		data = DATAOF_TRACE_ENTRY(entry, false);
1116 	}
1117 
1118 	memcpy(data, ucb->buf, tu->tp.size + dsize);
1119 
1120 	if (size - esize > tu->tp.size + dsize) {
1121 		int len = tu->tp.size + dsize;
1122 
1123 		memset(data + len, 0, size - esize - len);
1124 	}
1125 
1126 	perf_trace_buf_submit(entry, size, rctx, 0, 1, regs, head, NULL);
1127  out:
1128 	preempt_enable();
1129 }
1130 
1131 /* uprobe profile handler */
1132 static int uprobe_perf_func(struct trace_uprobe *tu, struct pt_regs *regs,
1133 			    struct uprobe_cpu_buffer *ucb, int dsize)
1134 {
1135 	if (!uprobe_perf_filter(&tu->consumer, 0, current->mm))
1136 		return UPROBE_HANDLER_REMOVE;
1137 
1138 	if (!is_ret_probe(tu))
1139 		__uprobe_perf_func(tu, 0, regs, ucb, dsize);
1140 	return 0;
1141 }
1142 
1143 static void uretprobe_perf_func(struct trace_uprobe *tu, unsigned long func,
1144 				struct pt_regs *regs,
1145 				struct uprobe_cpu_buffer *ucb, int dsize)
1146 {
1147 	__uprobe_perf_func(tu, func, regs, ucb, dsize);
1148 }
1149 #endif	/* CONFIG_PERF_EVENTS */
1150 
1151 static int
1152 trace_uprobe_register(struct ftrace_event_call *event, enum trace_reg type,
1153 		      void *data)
1154 {
1155 	struct trace_uprobe *tu = event->data;
1156 	struct ftrace_event_file *file = data;
1157 
1158 	switch (type) {
1159 	case TRACE_REG_REGISTER:
1160 		return probe_event_enable(tu, file, NULL);
1161 
1162 	case TRACE_REG_UNREGISTER:
1163 		probe_event_disable(tu, file);
1164 		return 0;
1165 
1166 #ifdef CONFIG_PERF_EVENTS
1167 	case TRACE_REG_PERF_REGISTER:
1168 		return probe_event_enable(tu, NULL, uprobe_perf_filter);
1169 
1170 	case TRACE_REG_PERF_UNREGISTER:
1171 		probe_event_disable(tu, NULL);
1172 		return 0;
1173 
1174 	case TRACE_REG_PERF_OPEN:
1175 		return uprobe_perf_open(tu, data);
1176 
1177 	case TRACE_REG_PERF_CLOSE:
1178 		return uprobe_perf_close(tu, data);
1179 
1180 #endif
1181 	default:
1182 		return 0;
1183 	}
1184 	return 0;
1185 }
1186 
1187 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs)
1188 {
1189 	struct trace_uprobe *tu;
1190 	struct uprobe_dispatch_data udd;
1191 	struct uprobe_cpu_buffer *ucb;
1192 	int dsize, esize;
1193 	int ret = 0;
1194 
1195 
1196 	tu = container_of(con, struct trace_uprobe, consumer);
1197 	tu->nhit++;
1198 
1199 	udd.tu = tu;
1200 	udd.bp_addr = instruction_pointer(regs);
1201 
1202 	current->utask->vaddr = (unsigned long) &udd;
1203 
1204 #ifdef CONFIG_PERF_EVENTS
1205 	if ((tu->tp.flags & TP_FLAG_TRACE) == 0 &&
1206 	    !uprobe_perf_filter(&tu->consumer, 0, current->mm))
1207 		return UPROBE_HANDLER_REMOVE;
1208 #endif
1209 
1210 	if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1211 		return 0;
1212 
1213 	dsize = __get_data_size(&tu->tp, regs);
1214 	esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1215 
1216 	ucb = uprobe_buffer_get();
1217 	store_trace_args(esize, &tu->tp, regs, ucb->buf, dsize);
1218 
1219 	if (tu->tp.flags & TP_FLAG_TRACE)
1220 		ret |= uprobe_trace_func(tu, regs, ucb, dsize);
1221 
1222 #ifdef CONFIG_PERF_EVENTS
1223 	if (tu->tp.flags & TP_FLAG_PROFILE)
1224 		ret |= uprobe_perf_func(tu, regs, ucb, dsize);
1225 #endif
1226 	uprobe_buffer_put(ucb);
1227 	return ret;
1228 }
1229 
1230 static int uretprobe_dispatcher(struct uprobe_consumer *con,
1231 				unsigned long func, struct pt_regs *regs)
1232 {
1233 	struct trace_uprobe *tu;
1234 	struct uprobe_dispatch_data udd;
1235 	struct uprobe_cpu_buffer *ucb;
1236 	int dsize, esize;
1237 
1238 	tu = container_of(con, struct trace_uprobe, consumer);
1239 
1240 	udd.tu = tu;
1241 	udd.bp_addr = func;
1242 
1243 	current->utask->vaddr = (unsigned long) &udd;
1244 
1245 	if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1246 		return 0;
1247 
1248 	dsize = __get_data_size(&tu->tp, regs);
1249 	esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1250 
1251 	ucb = uprobe_buffer_get();
1252 	store_trace_args(esize, &tu->tp, regs, ucb->buf, dsize);
1253 
1254 	if (tu->tp.flags & TP_FLAG_TRACE)
1255 		uretprobe_trace_func(tu, func, regs, ucb, dsize);
1256 
1257 #ifdef CONFIG_PERF_EVENTS
1258 	if (tu->tp.flags & TP_FLAG_PROFILE)
1259 		uretprobe_perf_func(tu, func, regs, ucb, dsize);
1260 #endif
1261 	uprobe_buffer_put(ucb);
1262 	return 0;
1263 }
1264 
1265 static struct trace_event_functions uprobe_funcs = {
1266 	.trace		= print_uprobe_event
1267 };
1268 
1269 static int register_uprobe_event(struct trace_uprobe *tu)
1270 {
1271 	struct ftrace_event_call *call = &tu->tp.call;
1272 	int ret;
1273 
1274 	/* Initialize ftrace_event_call */
1275 	INIT_LIST_HEAD(&call->class->fields);
1276 	call->event.funcs = &uprobe_funcs;
1277 	call->class->define_fields = uprobe_event_define_fields;
1278 
1279 	if (set_print_fmt(&tu->tp, is_ret_probe(tu)) < 0)
1280 		return -ENOMEM;
1281 
1282 	ret = register_ftrace_event(&call->event);
1283 	if (!ret) {
1284 		kfree(call->print_fmt);
1285 		return -ENODEV;
1286 	}
1287 	call->flags = 0;
1288 	call->class->reg = trace_uprobe_register;
1289 	call->data = tu;
1290 	ret = trace_add_event_call(call);
1291 
1292 	if (ret) {
1293 		pr_info("Failed to register uprobe event: %s\n",
1294 			ftrace_event_name(call));
1295 		kfree(call->print_fmt);
1296 		unregister_ftrace_event(&call->event);
1297 	}
1298 
1299 	return ret;
1300 }
1301 
1302 static int unregister_uprobe_event(struct trace_uprobe *tu)
1303 {
1304 	int ret;
1305 
1306 	/* tu->event is unregistered in trace_remove_event_call() */
1307 	ret = trace_remove_event_call(&tu->tp.call);
1308 	if (ret)
1309 		return ret;
1310 	kfree(tu->tp.call.print_fmt);
1311 	tu->tp.call.print_fmt = NULL;
1312 	return 0;
1313 }
1314 
1315 /* Make a trace interface for controling probe points */
1316 static __init int init_uprobe_trace(void)
1317 {
1318 	struct dentry *d_tracer;
1319 
1320 	d_tracer = tracing_init_dentry();
1321 	if (!d_tracer)
1322 		return 0;
1323 
1324 	trace_create_file("uprobe_events", 0644, d_tracer,
1325 				    NULL, &uprobe_events_ops);
1326 	/* Profile interface */
1327 	trace_create_file("uprobe_profile", 0444, d_tracer,
1328 				    NULL, &uprobe_profile_ops);
1329 	return 0;
1330 }
1331 
1332 fs_initcall(init_uprobe_trace);
1333