Linux Audio

Check our new training course

Loading...
v3.5.6
 
  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
 
 26#include "trace_probe.h"
 
 27
 28#define UPROBE_EVENT_SYSTEM	"uprobes"
 29
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 30/*
 31 * uprobe event core functions
 32 */
 33struct trace_uprobe;
 34struct uprobe_trace_consumer {
 35	struct uprobe_consumer		cons;
 36	struct trace_uprobe		*tu;
 37};
 38
 39struct trace_uprobe {
 40	struct list_head		list;
 41	struct ftrace_event_class	class;
 42	struct ftrace_event_call	call;
 43	struct uprobe_trace_consumer	*consumer;
 44	struct inode			*inode;
 45	char				*filename;
 46	unsigned long			offset;
 
 47	unsigned long			nhit;
 48	unsigned int			flags;	/* For TP_FLAG_* */
 49	ssize_t				size;	/* trace entry size */
 50	unsigned int			nr_args;
 51	struct probe_arg		args[];
 52};
 53
 54#define SIZEOF_TRACE_UPROBE(n)			\
 55	(offsetof(struct trace_uprobe, args) +	\
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 56	(sizeof(struct probe_arg) * (n)))
 57
 58static int register_uprobe_event(struct trace_uprobe *tu);
 59static void unregister_uprobe_event(struct trace_uprobe *tu);
 60
 61static DEFINE_MUTEX(uprobe_lock);
 62static LIST_HEAD(uprobe_list);
 
 
 63
 64static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 65
 66/*
 67 * Allocate new trace_uprobe and initialize it (including uprobes).
 68 */
 69static struct trace_uprobe *
 70alloc_trace_uprobe(const char *group, const char *event, int nargs)
 71{
 72	struct trace_uprobe *tu;
 73
 74	if (!event || !is_good_name(event))
 75		return ERR_PTR(-EINVAL);
 76
 77	if (!group || !is_good_name(group))
 78		return ERR_PTR(-EINVAL);
 79
 80	tu = kzalloc(SIZEOF_TRACE_UPROBE(nargs), GFP_KERNEL);
 81	if (!tu)
 82		return ERR_PTR(-ENOMEM);
 83
 84	tu->call.class = &tu->class;
 85	tu->call.name = kstrdup(event, GFP_KERNEL);
 86	if (!tu->call.name)
 87		goto error;
 88
 89	tu->class.system = kstrdup(group, GFP_KERNEL);
 90	if (!tu->class.system)
 91		goto error;
 92
 93	INIT_LIST_HEAD(&tu->list);
 
 
 
 
 94	return tu;
 95
 96error:
 97	kfree(tu->call.name);
 98	kfree(tu);
 99
100	return ERR_PTR(-ENOMEM);
101}
102
103static void free_trace_uprobe(struct trace_uprobe *tu)
104{
105	int i;
106
107	for (i = 0; i < tu->nr_args; i++)
108		traceprobe_free_probe_arg(&tu->args[i]);
109
110	iput(tu->inode);
111	kfree(tu->call.class->system);
112	kfree(tu->call.name);
113	kfree(tu->filename);
114	kfree(tu);
115}
116
117static struct trace_uprobe *find_probe_event(const char *event, const char *group)
118{
 
119	struct trace_uprobe *tu;
120
121	list_for_each_entry(tu, &uprobe_list, list)
122		if (strcmp(tu->call.name, event) == 0 &&
123		    strcmp(tu->call.class->system, group) == 0)
124			return tu;
125
126	return NULL;
127}
128
129/* Unregister a trace_uprobe and probe_event: call with locking uprobe_lock */
130static void unregister_trace_uprobe(struct trace_uprobe *tu)
131{
132	list_del(&tu->list);
133	unregister_uprobe_event(tu);
 
 
 
 
 
 
 
 
 
 
134	free_trace_uprobe(tu);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
135}
136
137/* Register a trace_uprobe and probe_event */
138static int register_trace_uprobe(struct trace_uprobe *tu)
139{
140	struct trace_uprobe *old_tp;
141	int ret;
142
143	mutex_lock(&uprobe_lock);
 
 
 
 
144
145	/* register as an event */
146	old_tp = find_probe_event(tu->call.name, tu->call.class->system);
147	if (old_tp)
148		/* delete old event */
149		unregister_trace_uprobe(old_tp);
 
 
 
 
 
 
 
 
150
151	ret = register_uprobe_event(tu);
152	if (ret) {
153		pr_warning("Failed to register probe event(%d)\n", ret);
 
 
 
 
154		goto end;
155	}
156
157	list_add_tail(&tu->list, &uprobe_list);
158
159end:
160	mutex_unlock(&uprobe_lock);
161
162	return ret;
163}
164
165/*
166 * Argument syntax:
167 *  - Add uprobe: p[:[GRP/]EVENT] PATH:SYMBOL[+offs] [FETCHARGS]
168 *
169 *  - Remove uprobe: -:[GRP/]EVENT
170 */
171static int create_trace_uprobe(int argc, char **argv)
172{
173	struct trace_uprobe *tu;
174	struct inode *inode;
175	char *arg, *event, *group, *filename;
176	char buf[MAX_EVENT_NAME_LEN];
177	struct path path;
178	unsigned long offset;
179	bool is_delete;
180	int i, ret;
181
182	inode = NULL;
183	ret = 0;
184	is_delete = false;
185	event = NULL;
186	group = NULL;
187
188	/* argc must be >= 1 */
189	if (argv[0][0] == '-')
190		is_delete = true;
191	else if (argv[0][0] != 'p') {
192		pr_info("Probe definition must be started with 'p', 'r' or" " '-'.\n");
193		return -EINVAL;
194	}
195
196	if (argv[0][1] == ':') {
 
 
 
197		event = &argv[0][2];
198		arg = strchr(event, '/');
199
200		if (arg) {
201			group = event;
202			event = arg + 1;
203			event[-1] = '\0';
204
205			if (strlen(group) == 0) {
206				pr_info("Group name is not specified\n");
207				return -EINVAL;
208			}
209		}
210		if (strlen(event) == 0) {
211			pr_info("Event name is not specified\n");
212			return -EINVAL;
213		}
214	}
215	if (!group)
216		group = UPROBE_EVENT_SYSTEM;
217
218	if (is_delete) {
219		if (!event) {
220			pr_info("Delete command needs an event name.\n");
221			return -EINVAL;
222		}
223		mutex_lock(&uprobe_lock);
224		tu = find_probe_event(event, group);
225
226		if (!tu) {
227			mutex_unlock(&uprobe_lock);
228			pr_info("Event %s/%s doesn't exist.\n", group, event);
229			return -ENOENT;
230		}
231		/* delete an event */
232		unregister_trace_uprobe(tu);
233		mutex_unlock(&uprobe_lock);
234		return 0;
235	}
236
237	if (argc < 2) {
238		pr_info("Probe point is not specified.\n");
239		return -EINVAL;
240	}
241	if (isdigit(argv[1][0])) {
242		pr_info("probe point must be have a filename.\n");
243		return -EINVAL;
244	}
245	arg = strchr(argv[1], ':');
246	if (!arg)
247		goto fail_address_parse;
248
249	*arg++ = '\0';
250	filename = argv[1];
251	ret = kern_path(filename, LOOKUP_FOLLOW, &path);
252	if (ret)
 
 
 
 
 
 
 
 
253		goto fail_address_parse;
 
254
255	ret = strict_strtoul(arg, 0, &offset);
256	if (ret)
257		goto fail_address_parse;
 
 
 
 
 
 
 
 
 
 
 
 
 
258
259	inode = igrab(path.dentry->d_inode);
 
 
 
 
 
 
 
260
261	argc -= 2;
262	argv += 2;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
263
264	/* setup a probe */
265	if (!event) {
266		char *tail = strrchr(filename, '/');
 
 
 
 
 
 
267		char *ptr;
268
269		ptr = kstrdup((tail ? tail + 1 : filename), GFP_KERNEL);
270		if (!ptr) {
271			ret = -ENOMEM;
272			goto fail_address_parse;
273		}
274
275		tail = ptr;
276		ptr = strpbrk(tail, ".-_");
277		if (ptr)
278			*ptr = '\0';
279
280		snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_0x%lx", 'p', tail, offset);
281		event = buf;
282		kfree(tail);
283	}
284
285	tu = alloc_trace_uprobe(group, event, argc);
 
 
 
286	if (IS_ERR(tu)) {
287		pr_info("Failed to allocate trace_uprobe.(%d)\n", (int)PTR_ERR(tu));
288		ret = PTR_ERR(tu);
 
 
289		goto fail_address_parse;
290	}
291	tu->offset = offset;
292	tu->inode = inode;
293	tu->filename = kstrdup(filename, GFP_KERNEL);
294
295	if (!tu->filename) {
296		pr_info("Failed to allocate filename.\n");
297		ret = -ENOMEM;
298		goto error;
299	}
300
301	/* parse arguments */
302	ret = 0;
303	for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
304		/* Increment count for freeing args in error case */
305		tu->nr_args++;
306
307		/* Parse argument name */
308		arg = strchr(argv[i], '=');
309		if (arg) {
310			*arg++ = '\0';
311			tu->args[i].name = kstrdup(argv[i], GFP_KERNEL);
312		} else {
313			arg = argv[i];
314			/* If argument name is omitted, set "argN" */
315			snprintf(buf, MAX_EVENT_NAME_LEN, "arg%d", i + 1);
316			tu->args[i].name = kstrdup(buf, GFP_KERNEL);
317		}
318
319		if (!tu->args[i].name) {
320			pr_info("Failed to allocate argument[%d] name.\n", i);
321			ret = -ENOMEM;
322			goto error;
323		}
324
325		if (!is_good_name(tu->args[i].name)) {
326			pr_info("Invalid argument[%d] name: %s\n", i, tu->args[i].name);
327			ret = -EINVAL;
328			goto error;
329		}
330
331		if (traceprobe_conflict_field_name(tu->args[i].name, tu->args, i)) {
332			pr_info("Argument[%d] name '%s' conflicts with "
333				"another field.\n", i, argv[i]);
334			ret = -EINVAL;
335			goto error;
336		}
337
338		/* Parse fetch argument */
339		ret = traceprobe_parse_probe_arg(arg, &tu->size, &tu->args[i], false, false);
340		if (ret) {
341			pr_info("Parse error at argument[%d]. (%d)\n", i, ret);
342			goto error;
343		}
344	}
345
346	ret = register_trace_uprobe(tu);
347	if (ret)
348		goto error;
349	return 0;
 
 
 
350
351error:
352	free_trace_uprobe(tu);
 
 
353	return ret;
354
355fail_address_parse:
356	if (inode)
357		iput(inode);
358
359	pr_info("Failed to parse address.\n");
360
361	return ret;
362}
363
364static void cleanup_all_probes(void)
365{
366	struct trace_uprobe *tu;
367
368	mutex_lock(&uprobe_lock);
369	while (!list_empty(&uprobe_list)) {
370		tu = list_entry(uprobe_list.next, struct trace_uprobe, list);
371		unregister_trace_uprobe(tu);
372	}
373	mutex_unlock(&uprobe_lock);
374}
375
376/* Probes listing interfaces */
377static void *probes_seq_start(struct seq_file *m, loff_t *pos)
378{
379	mutex_lock(&uprobe_lock);
380	return seq_list_start(&uprobe_list, *pos);
381}
382
383static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
384{
385	return seq_list_next(v, &uprobe_list, pos);
 
 
386}
387
388static void probes_seq_stop(struct seq_file *m, void *v)
389{
390	mutex_unlock(&uprobe_lock);
 
 
391}
392
393static int probes_seq_show(struct seq_file *m, void *v)
 
394{
395	struct trace_uprobe *tu = v;
 
396	int i;
397
398	seq_printf(m, "p:%s/%s", tu->call.class->system, tu->call.name);
399	seq_printf(m, " %s:0x%p", tu->filename, (void *)tu->offset);
 
 
 
 
400
401	for (i = 0; i < tu->nr_args; i++)
402		seq_printf(m, " %s=%s", tu->args[i].name, tu->args[i].comm);
403
404	seq_printf(m, "\n");
405	return 0;
406}
407
 
 
 
 
 
 
 
 
 
 
408static const struct seq_operations probes_seq_op = {
409	.start	= probes_seq_start,
410	.next	= probes_seq_next,
411	.stop	= probes_seq_stop,
412	.show	= probes_seq_show
413};
414
415static int probes_open(struct inode *inode, struct file *file)
416{
417	if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC))
418		cleanup_all_probes();
 
 
 
 
 
 
 
 
 
419
420	return seq_open(file, &probes_seq_op);
421}
422
423static ssize_t probes_write(struct file *file, const char __user *buffer,
424			    size_t count, loff_t *ppos)
425{
426	return traceprobe_probes_write(file, buffer, count, ppos, create_trace_uprobe);
 
427}
428
429static const struct file_operations uprobe_events_ops = {
430	.owner		= THIS_MODULE,
431	.open		= probes_open,
432	.read		= seq_read,
433	.llseek		= seq_lseek,
434	.release	= seq_release,
435	.write		= probes_write,
436};
437
438/* Probes profiling interfaces */
439static int probes_profile_seq_show(struct seq_file *m, void *v)
440{
441	struct trace_uprobe *tu = v;
 
 
 
 
442
443	seq_printf(m, "  %s %-44s %15lu\n", tu->filename, tu->call.name, tu->nhit);
 
 
444	return 0;
445}
446
447static const struct seq_operations profile_seq_op = {
448	.start	= probes_seq_start,
449	.next	= probes_seq_next,
450	.stop	= probes_seq_stop,
451	.show	= probes_profile_seq_show
452};
453
454static int profile_open(struct inode *inode, struct file *file)
455{
 
 
 
 
 
 
456	return seq_open(file, &profile_seq_op);
457}
458
459static const struct file_operations uprobe_profile_ops = {
460	.owner		= THIS_MODULE,
461	.open		= profile_open,
462	.read		= seq_read,
463	.llseek		= seq_lseek,
464	.release	= seq_release,
465};
466
467/* uprobe handler */
468static void uprobe_trace_func(struct trace_uprobe *tu, struct pt_regs *regs)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
469{
470	struct uprobe_trace_entry_head *entry;
 
471	struct ring_buffer_event *event;
472	struct ring_buffer *buffer;
473	u8 *data;
474	int size, i, pc;
475	unsigned long irq_flags;
476	struct ftrace_event_call *call = &tu->call;
477
478	tu->nhit++;
479
480	local_save_flags(irq_flags);
481	pc = preempt_count();
482
483	size = sizeof(*entry) + tu->size;
 
484
485	event = trace_current_buffer_lock_reserve(&buffer, call->event.type,
486						  size, irq_flags, pc);
 
 
487	if (!event)
488		return;
489
490	entry = ring_buffer_event_data(event);
491	entry->ip = uprobe_get_swbp_addr(task_pt_regs(current));
492	data = (u8 *)&entry[1];
493	for (i = 0; i < tu->nr_args; i++)
494		call_fetch(&tu->args[i].fetch, regs, data + tu->args[i].offset);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
495
496	if (!filter_current_check_discard(buffer, call, entry, event))
497		trace_buffer_unlock_commit(buffer, event, irq_flags, pc);
 
 
 
 
 
 
 
 
498}
499
500/* Event entry printers */
501static enum print_line_t
502print_uprobe_event(struct trace_iterator *iter, int flags, struct trace_event *event)
503{
504	struct uprobe_trace_entry_head *field;
505	struct trace_seq *s = &iter->seq;
506	struct trace_uprobe *tu;
507	u8 *data;
508	int i;
509
510	field = (struct uprobe_trace_entry_head *)iter->ent;
511	tu = container_of(event, struct trace_uprobe, call.event);
 
 
 
512
513	if (!trace_seq_printf(s, "%s: (", tu->call.name))
514		goto partial;
 
 
 
 
 
 
 
 
 
515
516	if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
517		goto partial;
518
519	if (!trace_seq_puts(s, ")"))
520		goto partial;
521
522	data = (u8 *)&field[1];
523	for (i = 0; i < tu->nr_args; i++) {
524		if (!tu->args[i].type->print(s, tu->args[i].name,
525					     data + tu->args[i].offset, field))
526			goto partial;
527	}
 
528
529	if (trace_seq_puts(s, "\n"))
530		return TRACE_TYPE_HANDLED;
 
 
 
 
 
 
 
 
 
 
 
 
 
531
532partial:
533	return TRACE_TYPE_PARTIAL_LINE;
534}
535
536static int probe_event_enable(struct trace_uprobe *tu, int flag)
537{
538	struct uprobe_trace_consumer *utc;
539	int ret = 0;
540
541	if (!tu->inode || tu->consumer)
542		return -EINTR;
543
544	utc = kzalloc(sizeof(struct uprobe_trace_consumer), GFP_KERNEL);
545	if (!utc)
546		return -EINTR;
547
548	utc->cons.handler = uprobe_dispatcher;
549	utc->cons.filter = NULL;
550	ret = uprobe_register(tu->inode, tu->offset, &utc->cons);
551	if (ret) {
552		kfree(utc);
553		return ret;
554	}
 
555
556	tu->flags |= flag;
557	utc->tu = tu;
558	tu->consumer = utc;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
559
560	return 0;
 
 
 
 
 
 
 
 
 
 
 
561}
562
563static void probe_event_disable(struct trace_uprobe *tu, int flag)
 
564{
565	if (!tu->inode || !tu->consumer)
 
 
 
 
 
 
566		return;
567
568	uprobe_unregister(tu->inode, tu->offset, &tu->consumer->cons);
569	tu->flags &= ~flag;
570	kfree(tu->consumer);
571	tu->consumer = NULL;
 
 
 
 
 
 
 
572}
573
574static int uprobe_event_define_fields(struct ftrace_event_call *event_call)
575{
576	int ret, i;
577	struct uprobe_trace_entry_head field;
578	struct trace_uprobe *tu = (struct trace_uprobe *)event_call->data;
579
580	DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
581	/* Set argument names as fields */
582	for (i = 0; i < tu->nr_args; i++) {
583		ret = trace_define_field(event_call, tu->args[i].type->fmttype,
584					 tu->args[i].name,
585					 sizeof(field) + tu->args[i].offset,
586					 tu->args[i].type->size,
587					 tu->args[i].type->is_signed,
588					 FILTER_OTHER);
589
590		if (ret)
591			return ret;
 
 
 
 
 
592	}
593	return 0;
 
594}
595
596#define LEN_OR_ZERO		(len ? len - pos : 0)
597static int __set_print_fmt(struct trace_uprobe *tu, char *buf, int len)
 
598{
599	const char *fmt, *arg;
600	int i;
601	int pos = 0;
 
602
603	fmt = "(%lx)";
604	arg = "REC->" FIELD_STRING_IP;
 
 
 
 
 
605
606	/* When len=0, we just calculate the needed length */
 
 
 
 
 
607
608	pos += snprintf(buf + pos, LEN_OR_ZERO, "\"%s", fmt);
 
 
 
609
610	for (i = 0; i < tu->nr_args; i++) {
611		pos += snprintf(buf + pos, LEN_OR_ZERO, " %s=%s",
612				tu->args[i].name, tu->args[i].type->fmt);
 
 
 
 
 
 
613	}
 
614
615	pos += snprintf(buf + pos, LEN_OR_ZERO, "\", %s", arg);
 
616
617	for (i = 0; i < tu->nr_args; i++) {
618		pos += snprintf(buf + pos, LEN_OR_ZERO, ", REC->%s",
619				tu->args[i].name);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
620	}
 
621
622	return pos;	/* return the length of print_fmt */
623}
624#undef LEN_OR_ZERO
625
626static int set_print_fmt(struct trace_uprobe *tu)
 
627{
628	char *print_fmt;
629	int len;
 
630
631	/* First: called with 0 length to calculate the needed length */
632	len = __set_print_fmt(tu, NULL, 0);
633	print_fmt = kmalloc(len + 1, GFP_KERNEL);
634	if (!print_fmt)
635		return -ENOMEM;
636
637	/* Second: actually write the @print_fmt */
638	__set_print_fmt(tu, print_fmt, len + 1);
639	tu->call.print_fmt = print_fmt;
640
641	return 0;
 
 
 
 
 
 
 
642}
643
644#ifdef CONFIG_PERF_EVENTS
645/* uprobe profile handler */
646static void uprobe_perf_func(struct trace_uprobe *tu, struct pt_regs *regs)
647{
648	struct ftrace_event_call *call = &tu->call;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
649	struct uprobe_trace_entry_head *entry;
650	struct hlist_head *head;
651	u8 *data;
652	int size, __size, i;
653	int rctx;
654
655	__size = sizeof(*entry) + tu->size;
656	size = ALIGN(__size + sizeof(u32), sizeof(u64));
657	size -= sizeof(u32);
 
 
 
 
 
 
 
 
 
 
 
658	if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE, "profile buffer not large enough"))
659		return;
660
661	preempt_disable();
 
 
 
662
663	entry = perf_trace_buf_prepare(size, call->event.type, regs, &rctx);
664	if (!entry)
665		goto out;
666
667	entry->ip = uprobe_get_swbp_addr(task_pt_regs(current));
668	data = (u8 *)&entry[1];
669	for (i = 0; i < tu->nr_args; i++)
670		call_fetch(&tu->args[i].fetch, regs, data + tu->args[i].offset);
 
 
 
 
 
 
671
672	head = this_cpu_ptr(call->perf_events);
673	perf_trace_buf_submit(entry, size, rctx, entry->ip, 1, regs, head);
 
 
 
674
 
 
675 out:
676	preempt_enable();
677}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
678#endif	/* CONFIG_PERF_EVENTS */
679
680static
681int trace_uprobe_register(struct ftrace_event_call *event, enum trace_reg type, void *data)
 
682{
683	struct trace_uprobe *tu = (struct trace_uprobe *)event->data;
684
685	switch (type) {
686	case TRACE_REG_REGISTER:
687		return probe_event_enable(tu, TP_FLAG_TRACE);
688
689	case TRACE_REG_UNREGISTER:
690		probe_event_disable(tu, TP_FLAG_TRACE);
691		return 0;
692
693#ifdef CONFIG_PERF_EVENTS
694	case TRACE_REG_PERF_REGISTER:
695		return probe_event_enable(tu, TP_FLAG_PROFILE);
696
697	case TRACE_REG_PERF_UNREGISTER:
698		probe_event_disable(tu, TP_FLAG_PROFILE);
699		return 0;
 
 
 
 
 
 
 
700#endif
701	default:
702		return 0;
703	}
704	return 0;
705}
706
707static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs)
708{
709	struct uprobe_trace_consumer *utc;
710	struct trace_uprobe *tu;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
711
712	utc = container_of(con, struct uprobe_trace_consumer, cons);
713	tu = utc->tu;
714	if (!tu || tu->consumer != utc)
 
 
 
 
 
715		return 0;
716
717	if (tu->flags & TP_FLAG_TRACE)
718		uprobe_trace_func(tu, regs);
 
 
 
 
 
 
719
720#ifdef CONFIG_PERF_EVENTS
721	if (tu->flags & TP_FLAG_PROFILE)
722		uprobe_perf_func(tu, regs);
723#endif
 
724	return 0;
725}
726
727static struct trace_event_functions uprobe_funcs = {
728	.trace		= print_uprobe_event
729};
730
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
731static int register_uprobe_event(struct trace_uprobe *tu)
732{
733	struct ftrace_event_call *call = &tu->call;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
734	int ret;
735
736	/* Initialize ftrace_event_call */
737	INIT_LIST_HEAD(&call->class->fields);
738	call->event.funcs = &uprobe_funcs;
739	call->class->define_fields = uprobe_event_define_fields;
740
741	if (set_print_fmt(tu) < 0)
742		return -ENOMEM;
 
 
743
744	ret = register_ftrace_event(&call->event);
745	if (!ret) {
746		kfree(call->print_fmt);
747		return -ENODEV;
 
 
 
 
 
 
 
 
 
748	}
749	call->flags = 0;
750	call->class->reg = trace_uprobe_register;
751	call->data = tu;
752	ret = trace_add_event_call(call);
753
754	if (ret) {
755		pr_info("Failed to register uprobe event: %s\n", call->name);
756		kfree(call->print_fmt);
757		unregister_ftrace_event(&call->event);
 
 
 
 
 
758	}
759
760	return ret;
 
 
 
761}
762
763static void unregister_uprobe_event(struct trace_uprobe *tu)
764{
765	/* tu->event is unregistered in trace_remove_event_call() */
766	trace_remove_event_call(&tu->call);
767	kfree(tu->call.print_fmt);
768	tu->call.print_fmt = NULL;
 
769}
 
770
771/* Make a trace interface for controling probe points */
772static __init int init_uprobe_trace(void)
773{
774	struct dentry *d_tracer;
775
776	d_tracer = tracing_init_dentry();
777	if (!d_tracer)
 
 
 
 
778		return 0;
779
780	trace_create_file("uprobe_events", 0644, d_tracer,
781				    NULL, &uprobe_events_ops);
782	/* Profile interface */
783	trace_create_file("uprobe_profile", 0444, d_tracer,
784				    NULL, &uprobe_profile_ops);
785	return 0;
786}
787
788fs_initcall(init_uprobe_trace);
v5.14.15
   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
  25struct 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
  37static int trace_uprobe_create(const char *raw_command);
  38static int trace_uprobe_show(struct seq_file *m, struct dyn_event *ev);
  39static int trace_uprobe_release(struct dyn_event *ev);
  40static bool trace_uprobe_is_busy(struct dyn_event *ev);
  41static bool trace_uprobe_match(const char *system, const char *event,
  42			int argc, const char **argv, struct dyn_event *ev);
  43
  44static 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 */
 
 
 
 
 
 
  55struct 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
  67static bool is_trace_uprobe(struct dyn_event *ev)
  68{
  69	return ev->ops == &trace_uprobe_ops;
  70}
  71
  72static 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#define SIZEOF_TRACE_UPROBE(n)				\
  87	(offsetof(struct trace_uprobe, tp.args) +	\
  88	(sizeof(struct probe_arg) * (n)))
  89
  90static int register_uprobe_event(struct trace_uprobe *tu);
  91static int unregister_uprobe_event(struct trace_uprobe *tu);
  92
  93struct uprobe_dispatch_data {
  94	struct trace_uprobe	*tu;
  95	unsigned long		bp_addr;
  96};
  97
  98static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs);
  99static int uretprobe_dispatcher(struct uprobe_consumer *con,
 100				unsigned long func, struct pt_regs *regs);
 101
 102#ifdef CONFIG_STACK_GROWSUP
 103static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
 104{
 105	return addr - (n * sizeof(long));
 106}
 107#else
 108static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
 109{
 110	return addr + (n * sizeof(long));
 111}
 112#endif
 113
 114static unsigned long get_user_stack_nth(struct pt_regs *regs, unsigned int n)
 115{
 116	unsigned long ret;
 117	unsigned long addr = user_stack_pointer(regs);
 118
 119	addr = adjust_stack_addr(addr, n);
 120
 121	if (copy_from_user(&ret, (void __force __user *) addr, sizeof(ret)))
 122		return 0;
 123
 124	return ret;
 125}
 126
 127/*
 128 * Uprobes-specific fetch functions
 129 */
 130static nokprobe_inline int
 131probe_mem_read(void *dest, void *src, size_t size)
 132{
 133	void __user *vaddr = (void __force __user *)src;
 134
 135	return copy_from_user(dest, vaddr, size) ? -EFAULT : 0;
 136}
 137
 138static nokprobe_inline int
 139probe_mem_read_user(void *dest, void *src, size_t size)
 140{
 141	return probe_mem_read(dest, src, size);
 142}
 143
 144/*
 145 * Fetch a null-terminated string. Caller MUST set *(u32 *)dest with max
 146 * length and relative data location.
 147 */
 148static nokprobe_inline int
 149fetch_store_string(unsigned long addr, void *dest, void *base)
 150{
 151	long ret;
 152	u32 loc = *(u32 *)dest;
 153	int maxlen  = get_loc_len(loc);
 154	u8 *dst = get_loc_data(dest, base);
 155	void __user *src = (void __force __user *) addr;
 156
 157	if (unlikely(!maxlen))
 158		return -ENOMEM;
 159
 160	if (addr == FETCH_TOKEN_COMM)
 161		ret = strlcpy(dst, current->comm, maxlen);
 162	else
 163		ret = strncpy_from_user(dst, src, maxlen);
 164	if (ret >= 0) {
 165		if (ret == maxlen)
 166			dst[ret - 1] = '\0';
 167		else
 168			/*
 169			 * Include the terminating null byte. In this case it
 170			 * was copied by strncpy_from_user but not accounted
 171			 * for in ret.
 172			 */
 173			ret++;
 174		*(u32 *)dest = make_data_loc(ret, (void *)dst - base);
 175	}
 176
 177	return ret;
 178}
 179
 180static nokprobe_inline int
 181fetch_store_string_user(unsigned long addr, void *dest, void *base)
 182{
 183	return fetch_store_string(addr, dest, base);
 184}
 185
 186/* Return the length of string -- including null terminal byte */
 187static nokprobe_inline int
 188fetch_store_strlen(unsigned long addr)
 189{
 190	int len;
 191	void __user *vaddr = (void __force __user *) addr;
 192
 193	if (addr == FETCH_TOKEN_COMM)
 194		len = strlen(current->comm) + 1;
 195	else
 196		len = strnlen_user(vaddr, MAX_STRING_SIZE);
 197
 198	return (len > MAX_STRING_SIZE) ? 0 : len;
 199}
 200
 201static nokprobe_inline int
 202fetch_store_strlen_user(unsigned long addr)
 203{
 204	return fetch_store_strlen(addr);
 205}
 206
 207static unsigned long translate_user_vaddr(unsigned long file_offset)
 208{
 209	unsigned long base_addr;
 210	struct uprobe_dispatch_data *udd;
 211
 212	udd = (void *) current->utask->vaddr;
 213
 214	base_addr = udd->bp_addr - udd->tu->offset;
 215	return base_addr + file_offset;
 216}
 217
 218/* Note that we don't verify it, since the code does not come from user space */
 219static int
 220process_fetch_insn(struct fetch_insn *code, struct pt_regs *regs, void *dest,
 221		   void *base)
 222{
 223	unsigned long val;
 224
 225	/* 1st stage: get value from context */
 226	switch (code->op) {
 227	case FETCH_OP_REG:
 228		val = regs_get_register(regs, code->param);
 229		break;
 230	case FETCH_OP_STACK:
 231		val = get_user_stack_nth(regs, code->param);
 232		break;
 233	case FETCH_OP_STACKP:
 234		val = user_stack_pointer(regs);
 235		break;
 236	case FETCH_OP_RETVAL:
 237		val = regs_return_value(regs);
 238		break;
 239	case FETCH_OP_IMM:
 240		val = code->immediate;
 241		break;
 242	case FETCH_OP_COMM:
 243		val = FETCH_TOKEN_COMM;
 244		break;
 245	case FETCH_OP_DATA:
 246		val = (unsigned long)code->data;
 247		break;
 248	case FETCH_OP_FOFFS:
 249		val = translate_user_vaddr(code->immediate);
 250		break;
 251	default:
 252		return -EILSEQ;
 253	}
 254	code++;
 255
 256	return process_fetch_insn_bottom(code, val, dest, base);
 257}
 258NOKPROBE_SYMBOL(process_fetch_insn)
 259
 260static inline void init_trace_uprobe_filter(struct trace_uprobe_filter *filter)
 261{
 262	rwlock_init(&filter->rwlock);
 263	filter->nr_systemwide = 0;
 264	INIT_LIST_HEAD(&filter->perf_events);
 265}
 266
 267static inline bool uprobe_filter_is_empty(struct trace_uprobe_filter *filter)
 268{
 269	return !filter->nr_systemwide && list_empty(&filter->perf_events);
 270}
 271
 272static inline bool is_ret_probe(struct trace_uprobe *tu)
 273{
 274	return tu->consumer.ret_handler != NULL;
 275}
 276
 277static bool trace_uprobe_is_busy(struct dyn_event *ev)
 278{
 279	struct trace_uprobe *tu = to_trace_uprobe(ev);
 280
 281	return trace_probe_is_enabled(&tu->tp);
 282}
 283
 284static bool trace_uprobe_match_command_head(struct trace_uprobe *tu,
 285					    int argc, const char **argv)
 286{
 287	char buf[MAX_ARGSTR_LEN + 1];
 288	int len;
 289
 290	if (!argc)
 291		return true;
 292
 293	len = strlen(tu->filename);
 294	if (strncmp(tu->filename, argv[0], len) || argv[0][len] != ':')
 295		return false;
 296
 297	if (tu->ref_ctr_offset == 0)
 298		snprintf(buf, sizeof(buf), "0x%0*lx",
 299				(int)(sizeof(void *) * 2), tu->offset);
 300	else
 301		snprintf(buf, sizeof(buf), "0x%0*lx(0x%lx)",
 302				(int)(sizeof(void *) * 2), tu->offset,
 303				tu->ref_ctr_offset);
 304	if (strcmp(buf, &argv[0][len + 1]))
 305		return false;
 306
 307	argc--; argv++;
 308
 309	return trace_probe_match_command_args(&tu->tp, argc, argv);
 310}
 311
 312static bool trace_uprobe_match(const char *system, const char *event,
 313			int argc, const char **argv, struct dyn_event *ev)
 314{
 315	struct trace_uprobe *tu = to_trace_uprobe(ev);
 316
 317	return strcmp(trace_probe_name(&tu->tp), event) == 0 &&
 318	   (!system || strcmp(trace_probe_group_name(&tu->tp), system) == 0) &&
 319	   trace_uprobe_match_command_head(tu, argc, argv);
 320}
 321
 322static nokprobe_inline struct trace_uprobe *
 323trace_uprobe_primary_from_call(struct trace_event_call *call)
 324{
 325	struct trace_probe *tp;
 326
 327	tp = trace_probe_primary_from_call(call);
 328	if (WARN_ON_ONCE(!tp))
 329		return NULL;
 330
 331	return container_of(tp, struct trace_uprobe, tp);
 332}
 333
 334/*
 335 * Allocate new trace_uprobe and initialize it (including uprobes).
 336 */
 337static struct trace_uprobe *
 338alloc_trace_uprobe(const char *group, const char *event, int nargs, bool is_ret)
 339{
 340	struct trace_uprobe *tu;
 341	int ret;
 
 
 
 
 
 342
 343	tu = kzalloc(SIZEOF_TRACE_UPROBE(nargs), GFP_KERNEL);
 344	if (!tu)
 345		return ERR_PTR(-ENOMEM);
 346
 347	ret = trace_probe_init(&tu->tp, event, group, true);
 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
 358error:
 
 359	kfree(tu);
 360
 361	return ERR_PTR(ret);
 362}
 363
 364static void free_trace_uprobe(struct trace_uprobe *tu)
 365{
 366	if (!tu)
 367		return;
 
 
 368
 369	path_put(&tu->path);
 370	trace_probe_cleanup(&tu->tp);
 
 371	kfree(tu->filename);
 372	kfree(tu);
 373}
 374
 375static struct trace_uprobe *find_probe_event(const char *event, const char *group)
 376{
 377	struct dyn_event *pos;
 378	struct trace_uprobe *tu;
 379
 380	for_each_trace_uprobe(tu, pos)
 381		if (strcmp(trace_probe_name(&tu->tp), event) == 0 &&
 382		    strcmp(trace_probe_group_name(&tu->tp), group) == 0)
 383			return tu;
 384
 385	return NULL;
 386}
 387
 388/* Unregister a trace_uprobe and probe_event */
 389static int unregister_trace_uprobe(struct trace_uprobe *tu)
 390{
 391	int ret;
 392
 393	if (trace_probe_has_sibling(&tu->tp))
 394		goto unreg;
 395
 396	ret = unregister_uprobe_event(tu);
 397	if (ret)
 398		return ret;
 399
 400unreg:
 401	dyn_event_remove(&tu->devent);
 402	trace_probe_unlink(&tu->tp);
 403	free_trace_uprobe(tu);
 404	return 0;
 405}
 406
 407static bool trace_uprobe_has_same_uprobe(struct trace_uprobe *orig,
 408					 struct trace_uprobe *comp)
 409{
 410	struct trace_probe_event *tpe = orig->tp.event;
 411	struct trace_probe *pos;
 412	struct inode *comp_inode = d_real_inode(comp->path.dentry);
 413	int i;
 414
 415	list_for_each_entry(pos, &tpe->probes, list) {
 416		orig = container_of(pos, struct trace_uprobe, tp);
 417		if (comp_inode != d_real_inode(orig->path.dentry) ||
 418		    comp->offset != orig->offset)
 419			continue;
 420
 421		/*
 422		 * trace_probe_compare_arg_type() ensured that nr_args and
 423		 * each argument name and type are same. Let's compare comm.
 424		 */
 425		for (i = 0; i < orig->tp.nr_args; i++) {
 426			if (strcmp(orig->tp.args[i].comm,
 427				   comp->tp.args[i].comm))
 428				break;
 429		}
 430
 431		if (i == orig->tp.nr_args)
 432			return true;
 433	}
 434
 435	return false;
 436}
 437
 438static int append_trace_uprobe(struct trace_uprobe *tu, struct trace_uprobe *to)
 439{
 440	int ret;
 441
 442	ret = trace_probe_compare_arg_type(&tu->tp, &to->tp);
 443	if (ret) {
 444		/* Note that argument starts index = 2 */
 445		trace_probe_log_set_index(ret + 1);
 446		trace_probe_log_err(0, DIFF_ARG_TYPE);
 447		return -EEXIST;
 448	}
 449	if (trace_uprobe_has_same_uprobe(to, tu)) {
 450		trace_probe_log_set_index(0);
 451		trace_probe_log_err(0, SAME_PROBE);
 452		return -EEXIST;
 453	}
 454
 455	/* Append to existing event */
 456	ret = trace_probe_append(&tu->tp, &to->tp);
 457	if (!ret)
 458		dyn_event_add(&tu->devent);
 459
 460	return ret;
 461}
 462
 463/*
 464 * Uprobe with multiple reference counter is not allowed. i.e.
 465 * If inode and offset matches, reference counter offset *must*
 466 * match as well. Though, there is one exception: If user is
 467 * replacing old trace_uprobe with new one(same group/event),
 468 * then we allow same uprobe with new reference counter as far
 469 * as the new one does not conflict with any other existing
 470 * ones.
 471 */
 472static int validate_ref_ctr_offset(struct trace_uprobe *new)
 473{
 474	struct dyn_event *pos;
 475	struct trace_uprobe *tmp;
 476	struct inode *new_inode = d_real_inode(new->path.dentry);
 477
 478	for_each_trace_uprobe(tmp, pos) {
 479		if (new_inode == d_real_inode(tmp->path.dentry) &&
 480		    new->offset == tmp->offset &&
 481		    new->ref_ctr_offset != tmp->ref_ctr_offset) {
 482			pr_warn("Reference counter offset mismatch.");
 483			return -EINVAL;
 484		}
 485	}
 486	return 0;
 487}
 488
 489/* Register a trace_uprobe and probe_event */
 490static int register_trace_uprobe(struct trace_uprobe *tu)
 491{
 492	struct trace_uprobe *old_tu;
 493	int ret;
 494
 495	mutex_lock(&event_mutex);
 496
 497	ret = validate_ref_ctr_offset(tu);
 498	if (ret)
 499		goto end;
 500
 501	/* register as an event */
 502	old_tu = find_probe_event(trace_probe_name(&tu->tp),
 503				  trace_probe_group_name(&tu->tp));
 504	if (old_tu) {
 505		if (is_ret_probe(tu) != is_ret_probe(old_tu)) {
 506			trace_probe_log_set_index(0);
 507			trace_probe_log_err(0, DIFF_PROBE_TYPE);
 508			ret = -EEXIST;
 509		} else {
 510			ret = append_trace_uprobe(tu, old_tu);
 511		}
 512		goto end;
 513	}
 514
 515	ret = register_uprobe_event(tu);
 516	if (ret) {
 517		if (ret == -EEXIST) {
 518			trace_probe_log_set_index(0);
 519			trace_probe_log_err(0, EVENT_EXIST);
 520		} else
 521			pr_warn("Failed to register probe event(%d)\n", ret);
 522		goto end;
 523	}
 524
 525	dyn_event_add(&tu->devent);
 526
 527end:
 528	mutex_unlock(&event_mutex);
 529
 530	return ret;
 531}
 532
 533/*
 534 * Argument syntax:
 535 *  - Add uprobe: p|r[:[GRP/]EVENT] PATH:OFFSET[%return][(REF)] [FETCHARGS]
 
 
 536 */
 537static int __trace_uprobe_create(int argc, const char **argv)
 538{
 539	struct trace_uprobe *tu;
 540	const char *event = NULL, *group = UPROBE_EVENT_SYSTEM;
 541	char *arg, *filename, *rctr, *rctr_end, *tmp;
 542	char buf[MAX_EVENT_NAME_LEN];
 543	struct path path;
 544	unsigned long offset, ref_ctr_offset;
 545	bool is_return = false;
 546	int i, ret;
 547
 
 548	ret = 0;
 549	ref_ctr_offset = 0;
 550
 551	switch (argv[0][0]) {
 552	case 'r':
 553		is_return = true;
 554		break;
 555	case 'p':
 556		break;
 557	default:
 558		return -ECANCELED;
 559	}
 560
 561	if (argc < 2)
 562		return -ECANCELED;
 563
 564	if (argv[0][1] == ':')
 565		event = &argv[0][2];
 
 566
 567	if (!strchr(argv[1], '/'))
 568		return -ECANCELED;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 569
 570	filename = kstrdup(argv[1], GFP_KERNEL);
 571	if (!filename)
 572		return -ENOMEM;
 
 
 
 
 573
 574	/* Find the last occurrence, in case the path contains ':' too. */
 575	arg = strrchr(filename, ':');
 576	if (!arg || !isdigit(arg[1])) {
 577		kfree(filename);
 578		return -ECANCELED;
 
 
 
 
 579	}
 580
 581	trace_probe_log_init("trace_uprobe", argc, argv);
 582	trace_probe_log_set_index(1);	/* filename is the 2nd argument */
 
 
 
 
 
 
 
 
 
 583
 584	*arg++ = '\0';
 
 585	ret = kern_path(filename, LOOKUP_FOLLOW, &path);
 586	if (ret) {
 587		trace_probe_log_err(0, FILE_NOT_FOUND);
 588		kfree(filename);
 589		trace_probe_log_clear();
 590		return ret;
 591	}
 592	if (!d_is_reg(path.dentry)) {
 593		trace_probe_log_err(0, NO_REGULAR_FILE);
 594		ret = -EINVAL;
 595		goto fail_address_parse;
 596	}
 597
 598	/* Parse reference counter offset if specified. */
 599	rctr = strchr(arg, '(');
 600	if (rctr) {
 601		rctr_end = strchr(rctr, ')');
 602		if (!rctr_end) {
 603			ret = -EINVAL;
 604			rctr_end = rctr + strlen(rctr);
 605			trace_probe_log_err(rctr_end - filename,
 606					    REFCNT_OPEN_BRACE);
 607			goto fail_address_parse;
 608		} else if (rctr_end[1] != '\0') {
 609			ret = -EINVAL;
 610			trace_probe_log_err(rctr_end + 1 - filename,
 611					    BAD_REFCNT_SUFFIX);
 612			goto fail_address_parse;
 613		}
 614
 615		*rctr++ = '\0';
 616		*rctr_end = '\0';
 617		ret = kstrtoul(rctr, 0, &ref_ctr_offset);
 618		if (ret) {
 619			trace_probe_log_err(rctr - filename, BAD_REFCNT);
 620			goto fail_address_parse;
 621		}
 622	}
 623
 624	/* Check if there is %return suffix */
 625	tmp = strchr(arg, '%');
 626	if (tmp) {
 627		if (!strcmp(tmp, "%return")) {
 628			*tmp = '\0';
 629			is_return = true;
 630		} else {
 631			trace_probe_log_err(tmp - filename, BAD_ADDR_SUFFIX);
 632			ret = -EINVAL;
 633			goto fail_address_parse;
 634		}
 635	}
 636
 637	/* Parse uprobe offset. */
 638	ret = kstrtoul(arg, 0, &offset);
 639	if (ret) {
 640		trace_probe_log_err(arg - filename, BAD_UPROBE_OFFS);
 641		goto fail_address_parse;
 642	}
 643
 644	/* setup a probe */
 645	trace_probe_log_set_index(0);
 646	if (event) {
 647		ret = traceprobe_parse_event_name(&event, &group, buf,
 648						  event - argv[0]);
 649		if (ret)
 650			goto fail_address_parse;
 651	} else {
 652		char *tail;
 653		char *ptr;
 654
 655		tail = kstrdup(kbasename(filename), GFP_KERNEL);
 656		if (!tail) {
 657			ret = -ENOMEM;
 658			goto fail_address_parse;
 659		}
 660
 
 661		ptr = strpbrk(tail, ".-_");
 662		if (ptr)
 663			*ptr = '\0';
 664
 665		snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_0x%lx", 'p', tail, offset);
 666		event = buf;
 667		kfree(tail);
 668	}
 669
 670	argc -= 2;
 671	argv += 2;
 672
 673	tu = alloc_trace_uprobe(group, event, argc, is_return);
 674	if (IS_ERR(tu)) {
 
 675		ret = PTR_ERR(tu);
 676		/* This must return -ENOMEM otherwise there is a bug */
 677		WARN_ON_ONCE(ret != -ENOMEM);
 678		goto fail_address_parse;
 679	}
 680	tu->offset = offset;
 681	tu->ref_ctr_offset = ref_ctr_offset;
 682	tu->path = path;
 683	tu->filename = filename;
 
 
 
 
 
 684
 685	/* parse arguments */
 
 686	for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
 687		tmp = kstrdup(argv[i], GFP_KERNEL);
 688		if (!tmp) {
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 689			ret = -ENOMEM;
 690			goto error;
 691		}
 692
 693		trace_probe_log_set_index(i + 2);
 694		ret = traceprobe_parse_probe_arg(&tu->tp, i, tmp,
 695					is_return ? TPARG_FL_RETURN : 0);
 696		kfree(tmp);
 697		if (ret)
 
 
 
 
 
 
 
 
 
 
 
 
 698			goto error;
 
 699	}
 700
 701	ret = traceprobe_set_print_fmt(&tu->tp, is_ret_probe(tu));
 702	if (ret < 0)
 703		goto error;
 704
 705	ret = register_trace_uprobe(tu);
 706	if (!ret)
 707		goto out;
 708
 709error:
 710	free_trace_uprobe(tu);
 711out:
 712	trace_probe_log_clear();
 713	return ret;
 714
 715fail_address_parse:
 716	trace_probe_log_clear();
 717	path_put(&path);
 718	kfree(filename);
 
 719
 720	return ret;
 721}
 722
 723int trace_uprobe_create(const char *raw_command)
 724{
 725	return trace_probe_create(raw_command, __trace_uprobe_create);
 
 
 
 
 
 
 
 726}
 727
 728static int create_or_delete_trace_uprobe(const char *raw_command)
 
 729{
 730	int ret;
 
 
 731
 732	if (raw_command[0] == '-')
 733		return dyn_event_release(raw_command, &trace_uprobe_ops);
 734
 735	ret = trace_uprobe_create(raw_command);
 736	return ret == -ECANCELED ? -EINVAL : ret;
 737}
 738
 739static int trace_uprobe_release(struct dyn_event *ev)
 740{
 741	struct trace_uprobe *tu = to_trace_uprobe(ev);
 742
 743	return unregister_trace_uprobe(tu);
 744}
 745
 746/* Probes listing interfaces */
 747static int trace_uprobe_show(struct seq_file *m, struct dyn_event *ev)
 748{
 749	struct trace_uprobe *tu = to_trace_uprobe(ev);
 750	char c = is_ret_probe(tu) ? 'r' : 'p';
 751	int i;
 752
 753	seq_printf(m, "%c:%s/%s %s:0x%0*lx", c, trace_probe_group_name(&tu->tp),
 754			trace_probe_name(&tu->tp), tu->filename,
 755			(int)(sizeof(void *) * 2), tu->offset);
 756
 757	if (tu->ref_ctr_offset)
 758		seq_printf(m, "(0x%lx)", tu->ref_ctr_offset);
 759
 760	for (i = 0; i < tu->tp.nr_args; i++)
 761		seq_printf(m, " %s=%s", tu->tp.args[i].name, tu->tp.args[i].comm);
 762
 763	seq_putc(m, '\n');
 764	return 0;
 765}
 766
 767static int probes_seq_show(struct seq_file *m, void *v)
 768{
 769	struct dyn_event *ev = v;
 770
 771	if (!is_trace_uprobe(ev))
 772		return 0;
 773
 774	return trace_uprobe_show(m, ev);
 775}
 776
 777static const struct seq_operations probes_seq_op = {
 778	.start  = dyn_event_seq_start,
 779	.next   = dyn_event_seq_next,
 780	.stop   = dyn_event_seq_stop,
 781	.show   = probes_seq_show
 782};
 783
 784static int probes_open(struct inode *inode, struct file *file)
 785{
 786	int ret;
 787
 788	ret = security_locked_down(LOCKDOWN_TRACEFS);
 789	if (ret)
 790		return ret;
 791
 792	if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
 793		ret = dyn_events_release_all(&trace_uprobe_ops);
 794		if (ret)
 795			return ret;
 796	}
 797
 798	return seq_open(file, &probes_seq_op);
 799}
 800
 801static ssize_t probes_write(struct file *file, const char __user *buffer,
 802			    size_t count, loff_t *ppos)
 803{
 804	return trace_parse_run_command(file, buffer, count, ppos,
 805					create_or_delete_trace_uprobe);
 806}
 807
 808static const struct file_operations uprobe_events_ops = {
 809	.owner		= THIS_MODULE,
 810	.open		= probes_open,
 811	.read		= seq_read,
 812	.llseek		= seq_lseek,
 813	.release	= seq_release,
 814	.write		= probes_write,
 815};
 816
 817/* Probes profiling interfaces */
 818static int probes_profile_seq_show(struct seq_file *m, void *v)
 819{
 820	struct dyn_event *ev = v;
 821	struct trace_uprobe *tu;
 822
 823	if (!is_trace_uprobe(ev))
 824		return 0;
 825
 826	tu = to_trace_uprobe(ev);
 827	seq_printf(m, "  %s %-44s %15lu\n", tu->filename,
 828			trace_probe_name(&tu->tp), tu->nhit);
 829	return 0;
 830}
 831
 832static const struct seq_operations profile_seq_op = {
 833	.start  = dyn_event_seq_start,
 834	.next   = dyn_event_seq_next,
 835	.stop   = dyn_event_seq_stop,
 836	.show	= probes_profile_seq_show
 837};
 838
 839static int profile_open(struct inode *inode, struct file *file)
 840{
 841	int ret;
 842
 843	ret = security_locked_down(LOCKDOWN_TRACEFS);
 844	if (ret)
 845		return ret;
 846
 847	return seq_open(file, &profile_seq_op);
 848}
 849
 850static const struct file_operations uprobe_profile_ops = {
 851	.owner		= THIS_MODULE,
 852	.open		= profile_open,
 853	.read		= seq_read,
 854	.llseek		= seq_lseek,
 855	.release	= seq_release,
 856};
 857
 858struct uprobe_cpu_buffer {
 859	struct mutex mutex;
 860	void *buf;
 861};
 862static struct uprobe_cpu_buffer __percpu *uprobe_cpu_buffer;
 863static int uprobe_buffer_refcnt;
 864
 865static int uprobe_buffer_init(void)
 866{
 867	int cpu, err_cpu;
 868
 869	uprobe_cpu_buffer = alloc_percpu(struct uprobe_cpu_buffer);
 870	if (uprobe_cpu_buffer == NULL)
 871		return -ENOMEM;
 872
 873	for_each_possible_cpu(cpu) {
 874		struct page *p = alloc_pages_node(cpu_to_node(cpu),
 875						  GFP_KERNEL, 0);
 876		if (p == NULL) {
 877			err_cpu = cpu;
 878			goto err;
 879		}
 880		per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf = page_address(p);
 881		mutex_init(&per_cpu_ptr(uprobe_cpu_buffer, cpu)->mutex);
 882	}
 883
 884	return 0;
 885
 886err:
 887	for_each_possible_cpu(cpu) {
 888		if (cpu == err_cpu)
 889			break;
 890		free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf);
 891	}
 892
 893	free_percpu(uprobe_cpu_buffer);
 894	return -ENOMEM;
 895}
 896
 897static int uprobe_buffer_enable(void)
 898{
 899	int ret = 0;
 900
 901	BUG_ON(!mutex_is_locked(&event_mutex));
 902
 903	if (uprobe_buffer_refcnt++ == 0) {
 904		ret = uprobe_buffer_init();
 905		if (ret < 0)
 906			uprobe_buffer_refcnt--;
 907	}
 908
 909	return ret;
 910}
 911
 912static void uprobe_buffer_disable(void)
 913{
 914	int cpu;
 915
 916	BUG_ON(!mutex_is_locked(&event_mutex));
 917
 918	if (--uprobe_buffer_refcnt == 0) {
 919		for_each_possible_cpu(cpu)
 920			free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer,
 921							     cpu)->buf);
 922
 923		free_percpu(uprobe_cpu_buffer);
 924		uprobe_cpu_buffer = NULL;
 925	}
 926}
 927
 928static struct uprobe_cpu_buffer *uprobe_buffer_get(void)
 929{
 930	struct uprobe_cpu_buffer *ucb;
 931	int cpu;
 932
 933	cpu = raw_smp_processor_id();
 934	ucb = per_cpu_ptr(uprobe_cpu_buffer, cpu);
 935
 936	/*
 937	 * Use per-cpu buffers for fastest access, but we might migrate
 938	 * so the mutex makes sure we have sole access to it.
 939	 */
 940	mutex_lock(&ucb->mutex);
 941
 942	return ucb;
 943}
 944
 945static void uprobe_buffer_put(struct uprobe_cpu_buffer *ucb)
 946{
 947	mutex_unlock(&ucb->mutex);
 948}
 949
 950static void __uprobe_trace_func(struct trace_uprobe *tu,
 951				unsigned long func, struct pt_regs *regs,
 952				struct uprobe_cpu_buffer *ucb, int dsize,
 953				struct trace_event_file *trace_file)
 954{
 955	struct uprobe_trace_entry_head *entry;
 956	struct trace_buffer *buffer;
 957	struct ring_buffer_event *event;
 958	void *data;
 959	int size, esize;
 960	struct trace_event_call *call = trace_probe_event_call(&tu->tp);
 
 
 961
 962	WARN_ON(call != trace_file->event_call);
 963
 964	if (WARN_ON_ONCE(tu->tp.size + dsize > PAGE_SIZE))
 965		return;
 966
 967	if (trace_trigger_soft_disabled(trace_file))
 968		return;
 969
 970	esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
 971	size = esize + tu->tp.size + dsize;
 972	event = trace_event_buffer_lock_reserve(&buffer, trace_file,
 973						call->event.type, size, 0);
 974	if (!event)
 975		return;
 976
 977	entry = ring_buffer_event_data(event);
 978	if (is_ret_probe(tu)) {
 979		entry->vaddr[0] = func;
 980		entry->vaddr[1] = instruction_pointer(regs);
 981		data = DATAOF_TRACE_ENTRY(entry, true);
 982	} else {
 983		entry->vaddr[0] = instruction_pointer(regs);
 984		data = DATAOF_TRACE_ENTRY(entry, false);
 985	}
 986
 987	memcpy(data, ucb->buf, tu->tp.size + dsize);
 988
 989	event_trigger_unlock_commit(trace_file, buffer, event, entry, 0);
 990}
 991
 992/* uprobe handler */
 993static int uprobe_trace_func(struct trace_uprobe *tu, struct pt_regs *regs,
 994			     struct uprobe_cpu_buffer *ucb, int dsize)
 995{
 996	struct event_file_link *link;
 997
 998	if (is_ret_probe(tu))
 999		return 0;
1000
1001	rcu_read_lock();
1002	trace_probe_for_each_link_rcu(link, &tu->tp)
1003		__uprobe_trace_func(tu, 0, regs, ucb, dsize, link->file);
1004	rcu_read_unlock();
1005
1006	return 0;
1007}
1008
1009static void uretprobe_trace_func(struct trace_uprobe *tu, unsigned long func,
1010				 struct pt_regs *regs,
1011				 struct uprobe_cpu_buffer *ucb, int dsize)
1012{
1013	struct event_file_link *link;
1014
1015	rcu_read_lock();
1016	trace_probe_for_each_link_rcu(link, &tu->tp)
1017		__uprobe_trace_func(tu, func, regs, ucb, dsize, link->file);
1018	rcu_read_unlock();
1019}
1020
1021/* Event entry printers */
1022static enum print_line_t
1023print_uprobe_event(struct trace_iterator *iter, int flags, struct trace_event *event)
1024{
1025	struct uprobe_trace_entry_head *entry;
1026	struct trace_seq *s = &iter->seq;
1027	struct trace_uprobe *tu;
1028	u8 *data;
 
1029
1030	entry = (struct uprobe_trace_entry_head *)iter->ent;
1031	tu = trace_uprobe_primary_from_call(
1032		container_of(event, struct trace_event_call, event));
1033	if (unlikely(!tu))
1034		goto out;
1035
1036	if (is_ret_probe(tu)) {
1037		trace_seq_printf(s, "%s: (0x%lx <- 0x%lx)",
1038				 trace_probe_name(&tu->tp),
1039				 entry->vaddr[1], entry->vaddr[0]);
1040		data = DATAOF_TRACE_ENTRY(entry, true);
1041	} else {
1042		trace_seq_printf(s, "%s: (0x%lx)",
1043				 trace_probe_name(&tu->tp),
1044				 entry->vaddr[0]);
1045		data = DATAOF_TRACE_ENTRY(entry, false);
1046	}
1047
1048	if (print_probe_args(s, tu->tp.args, tu->tp.nr_args, data, entry) < 0)
1049		goto out;
1050
1051	trace_seq_putc(s, '\n');
 
1052
1053 out:
1054	return trace_handle_return(s);
1055}
1056
1057typedef bool (*filter_func_t)(struct uprobe_consumer *self,
1058				enum uprobe_filter_ctx ctx,
1059				struct mm_struct *mm);
1060
1061static int trace_uprobe_enable(struct trace_uprobe *tu, filter_func_t filter)
1062{
1063	int ret;
1064
1065	tu->consumer.filter = filter;
1066	tu->inode = d_real_inode(tu->path.dentry);
1067
1068	if (tu->ref_ctr_offset)
1069		ret = uprobe_register_refctr(tu->inode, tu->offset,
1070				tu->ref_ctr_offset, &tu->consumer);
1071	else
1072		ret = uprobe_register(tu->inode, tu->offset, &tu->consumer);
1073
1074	if (ret)
1075		tu->inode = NULL;
1076
1077	return ret;
 
1078}
1079
1080static void __probe_event_disable(struct trace_probe *tp)
1081{
1082	struct trace_probe *pos;
1083	struct trace_uprobe *tu;
1084
1085	tu = container_of(tp, struct trace_uprobe, tp);
1086	WARN_ON(!uprobe_filter_is_empty(tu->tp.event->filter));
1087
1088	list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
1089		tu = container_of(pos, struct trace_uprobe, tp);
1090		if (!tu->inode)
1091			continue;
1092
1093		uprobe_unregister(tu->inode, tu->offset, &tu->consumer);
1094		tu->inode = NULL;
 
 
 
1095	}
1096}
1097
1098static int probe_event_enable(struct trace_event_call *call,
1099			struct trace_event_file *file, filter_func_t filter)
1100{
1101	struct trace_probe *pos, *tp;
1102	struct trace_uprobe *tu;
1103	bool enabled;
1104	int ret;
1105
1106	tp = trace_probe_primary_from_call(call);
1107	if (WARN_ON_ONCE(!tp))
1108		return -ENODEV;
1109	enabled = trace_probe_is_enabled(tp);
1110
1111	/* This may also change "enabled" state */
1112	if (file) {
1113		if (trace_probe_test_flag(tp, TP_FLAG_PROFILE))
1114			return -EINTR;
1115
1116		ret = trace_probe_add_file(tp, file);
1117		if (ret < 0)
1118			return ret;
1119	} else {
1120		if (trace_probe_test_flag(tp, TP_FLAG_TRACE))
1121			return -EINTR;
1122
1123		trace_probe_set_flag(tp, TP_FLAG_PROFILE);
1124	}
1125
1126	tu = container_of(tp, struct trace_uprobe, tp);
1127	WARN_ON(!uprobe_filter_is_empty(tu->tp.event->filter));
1128
1129	if (enabled)
1130		return 0;
1131
1132	ret = uprobe_buffer_enable();
1133	if (ret)
1134		goto err_flags;
1135
1136	list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
1137		tu = container_of(pos, struct trace_uprobe, tp);
1138		ret = trace_uprobe_enable(tu, filter);
1139		if (ret) {
1140			__probe_event_disable(tp);
1141			goto err_buffer;
1142		}
1143	}
1144
1145	return 0;
1146
1147 err_buffer:
1148	uprobe_buffer_disable();
1149
1150 err_flags:
1151	if (file)
1152		trace_probe_remove_file(tp, file);
1153	else
1154		trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
1155
1156	return ret;
1157}
1158
1159static void probe_event_disable(struct trace_event_call *call,
1160				struct trace_event_file *file)
1161{
1162	struct trace_probe *tp;
1163
1164	tp = trace_probe_primary_from_call(call);
1165	if (WARN_ON_ONCE(!tp))
1166		return;
1167
1168	if (!trace_probe_is_enabled(tp))
1169		return;
1170
1171	if (file) {
1172		if (trace_probe_remove_file(tp, file) < 0)
1173			return;
1174
1175		if (trace_probe_is_enabled(tp))
1176			return;
1177	} else
1178		trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
1179
1180	__probe_event_disable(tp);
1181	uprobe_buffer_disable();
1182}
1183
1184static int uprobe_event_define_fields(struct trace_event_call *event_call)
1185{
1186	int ret, size;
1187	struct uprobe_trace_entry_head field;
1188	struct trace_uprobe *tu;
1189
1190	tu = trace_uprobe_primary_from_call(event_call);
1191	if (unlikely(!tu))
1192		return -ENODEV;
 
 
 
 
 
 
1193
1194	if (is_ret_probe(tu)) {
1195		DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_FUNC, 0);
1196		DEFINE_FIELD(unsigned long, vaddr[1], FIELD_STRING_RETIP, 0);
1197		size = SIZEOF_TRACE_ENTRY(true);
1198	} else {
1199		DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_IP, 0);
1200		size = SIZEOF_TRACE_ENTRY(false);
1201	}
1202
1203	return traceprobe_define_arg_fields(event_call, size, &tu->tp);
1204}
1205
1206#ifdef CONFIG_PERF_EVENTS
1207static bool
1208__uprobe_perf_filter(struct trace_uprobe_filter *filter, struct mm_struct *mm)
1209{
1210	struct perf_event *event;
1211
1212	if (filter->nr_systemwide)
1213		return true;
1214
1215	list_for_each_entry(event, &filter->perf_events, hw.tp_list) {
1216		if (event->hw.target->mm == mm)
1217			return true;
1218	}
1219
1220	return false;
1221}
1222
1223static inline bool
1224trace_uprobe_filter_event(struct trace_uprobe_filter *filter,
1225			  struct perf_event *event)
1226{
1227	return __uprobe_perf_filter(filter, event->hw.target->mm);
1228}
1229
1230static bool trace_uprobe_filter_remove(struct trace_uprobe_filter *filter,
1231				       struct perf_event *event)
1232{
1233	bool done;
1234
1235	write_lock(&filter->rwlock);
1236	if (event->hw.target) {
1237		list_del(&event->hw.tp_list);
1238		done = filter->nr_systemwide ||
1239			(event->hw.target->flags & PF_EXITING) ||
1240			trace_uprobe_filter_event(filter, event);
1241	} else {
1242		filter->nr_systemwide--;
1243		done = filter->nr_systemwide;
1244	}
1245	write_unlock(&filter->rwlock);
1246
1247	return done;
1248}
1249
1250/* This returns true if the filter always covers target mm */
1251static bool trace_uprobe_filter_add(struct trace_uprobe_filter *filter,
1252				    struct perf_event *event)
1253{
1254	bool done;
1255
1256	write_lock(&filter->rwlock);
1257	if (event->hw.target) {
1258		/*
1259		 * event->parent != NULL means copy_process(), we can avoid
1260		 * uprobe_apply(). current->mm must be probed and we can rely
1261		 * on dup_mmap() which preserves the already installed bp's.
1262		 *
1263		 * attr.enable_on_exec means that exec/mmap will install the
1264		 * breakpoints we need.
1265		 */
1266		done = filter->nr_systemwide ||
1267			event->parent || event->attr.enable_on_exec ||
1268			trace_uprobe_filter_event(filter, event);
1269		list_add(&event->hw.tp_list, &filter->perf_events);
1270	} else {
1271		done = filter->nr_systemwide;
1272		filter->nr_systemwide++;
1273	}
1274	write_unlock(&filter->rwlock);
1275
1276	return done;
1277}
 
1278
1279static int uprobe_perf_close(struct trace_event_call *call,
1280			     struct perf_event *event)
1281{
1282	struct trace_probe *pos, *tp;
1283	struct trace_uprobe *tu;
1284	int ret = 0;
1285
1286	tp = trace_probe_primary_from_call(call);
1287	if (WARN_ON_ONCE(!tp))
1288		return -ENODEV;
 
 
1289
1290	tu = container_of(tp, struct trace_uprobe, tp);
1291	if (trace_uprobe_filter_remove(tu->tp.event->filter, event))
1292		return 0;
1293
1294	list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
1295		tu = container_of(pos, struct trace_uprobe, tp);
1296		ret = uprobe_apply(tu->inode, tu->offset, &tu->consumer, false);
1297		if (ret)
1298			break;
1299	}
1300
1301	return ret;
1302}
1303
1304static int uprobe_perf_open(struct trace_event_call *call,
1305			    struct perf_event *event)
 
1306{
1307	struct trace_probe *pos, *tp;
1308	struct trace_uprobe *tu;
1309	int err = 0;
1310
1311	tp = trace_probe_primary_from_call(call);
1312	if (WARN_ON_ONCE(!tp))
1313		return -ENODEV;
1314
1315	tu = container_of(tp, struct trace_uprobe, tp);
1316	if (trace_uprobe_filter_add(tu->tp.event->filter, event))
1317		return 0;
1318
1319	list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
1320		err = uprobe_apply(tu->inode, tu->offset, &tu->consumer, true);
1321		if (err) {
1322			uprobe_perf_close(call, event);
1323			break;
1324		}
1325	}
1326
1327	return err;
1328}
1329
1330static bool uprobe_perf_filter(struct uprobe_consumer *uc,
1331				enum uprobe_filter_ctx ctx, struct mm_struct *mm)
1332{
1333	struct trace_uprobe_filter *filter;
1334	struct trace_uprobe *tu;
1335	int ret;
1336
1337	tu = container_of(uc, struct trace_uprobe, consumer);
1338	filter = tu->tp.event->filter;
1339
1340	read_lock(&filter->rwlock);
1341	ret = __uprobe_perf_filter(filter, mm);
1342	read_unlock(&filter->rwlock);
1343
1344	return ret;
1345}
1346
1347static void __uprobe_perf_func(struct trace_uprobe *tu,
1348			       unsigned long func, struct pt_regs *regs,
1349			       struct uprobe_cpu_buffer *ucb, int dsize)
1350{
1351	struct trace_event_call *call = trace_probe_event_call(&tu->tp);
1352	struct uprobe_trace_entry_head *entry;
1353	struct hlist_head *head;
1354	void *data;
1355	int size, esize;
1356	int rctx;
1357
1358	if (bpf_prog_array_valid(call)) {
1359		u32 ret;
1360
1361		preempt_disable();
1362		ret = trace_call_bpf(call, regs);
1363		preempt_enable();
1364		if (!ret)
1365			return;
1366	}
1367
1368	esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1369
1370	size = esize + tu->tp.size + dsize;
1371	size = ALIGN(size + sizeof(u32), sizeof(u64)) - sizeof(u32);
1372	if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE, "profile buffer not large enough"))
1373		return;
1374
1375	preempt_disable();
1376	head = this_cpu_ptr(call->perf_events);
1377	if (hlist_empty(head))
1378		goto out;
1379
1380	entry = perf_trace_buf_alloc(size, NULL, &rctx);
1381	if (!entry)
1382		goto out;
1383
1384	if (is_ret_probe(tu)) {
1385		entry->vaddr[0] = func;
1386		entry->vaddr[1] = instruction_pointer(regs);
1387		data = DATAOF_TRACE_ENTRY(entry, true);
1388	} else {
1389		entry->vaddr[0] = instruction_pointer(regs);
1390		data = DATAOF_TRACE_ENTRY(entry, false);
1391	}
1392
1393	memcpy(data, ucb->buf, tu->tp.size + dsize);
1394
1395	if (size - esize > tu->tp.size + dsize) {
1396		int len = tu->tp.size + dsize;
1397
1398		memset(data + len, 0, size - esize - len);
1399	}
1400
1401	perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1402			      head, NULL);
1403 out:
1404	preempt_enable();
1405}
1406
1407/* uprobe profile handler */
1408static int uprobe_perf_func(struct trace_uprobe *tu, struct pt_regs *regs,
1409			    struct uprobe_cpu_buffer *ucb, int dsize)
1410{
1411	if (!uprobe_perf_filter(&tu->consumer, 0, current->mm))
1412		return UPROBE_HANDLER_REMOVE;
1413
1414	if (!is_ret_probe(tu))
1415		__uprobe_perf_func(tu, 0, regs, ucb, dsize);
1416	return 0;
1417}
1418
1419static void uretprobe_perf_func(struct trace_uprobe *tu, unsigned long func,
1420				struct pt_regs *regs,
1421				struct uprobe_cpu_buffer *ucb, int dsize)
1422{
1423	__uprobe_perf_func(tu, func, regs, ucb, dsize);
1424}
1425
1426int bpf_get_uprobe_info(const struct perf_event *event, u32 *fd_type,
1427			const char **filename, u64 *probe_offset,
1428			bool perf_type_tracepoint)
1429{
1430	const char *pevent = trace_event_name(event->tp_event);
1431	const char *group = event->tp_event->class->system;
1432	struct trace_uprobe *tu;
1433
1434	if (perf_type_tracepoint)
1435		tu = find_probe_event(pevent, group);
1436	else
1437		tu = trace_uprobe_primary_from_call(event->tp_event);
1438	if (!tu)
1439		return -EINVAL;
1440
1441	*fd_type = is_ret_probe(tu) ? BPF_FD_TYPE_URETPROBE
1442				    : BPF_FD_TYPE_UPROBE;
1443	*filename = tu->filename;
1444	*probe_offset = tu->offset;
1445	return 0;
1446}
1447#endif	/* CONFIG_PERF_EVENTS */
1448
1449static int
1450trace_uprobe_register(struct trace_event_call *event, enum trace_reg type,
1451		      void *data)
1452{
1453	struct trace_event_file *file = data;
1454
1455	switch (type) {
1456	case TRACE_REG_REGISTER:
1457		return probe_event_enable(event, file, NULL);
1458
1459	case TRACE_REG_UNREGISTER:
1460		probe_event_disable(event, file);
1461		return 0;
1462
1463#ifdef CONFIG_PERF_EVENTS
1464	case TRACE_REG_PERF_REGISTER:
1465		return probe_event_enable(event, NULL, uprobe_perf_filter);
1466
1467	case TRACE_REG_PERF_UNREGISTER:
1468		probe_event_disable(event, NULL);
1469		return 0;
1470
1471	case TRACE_REG_PERF_OPEN:
1472		return uprobe_perf_open(event, data);
1473
1474	case TRACE_REG_PERF_CLOSE:
1475		return uprobe_perf_close(event, data);
1476
1477#endif
1478	default:
1479		return 0;
1480	}
 
1481}
1482
1483static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs)
1484{
 
1485	struct trace_uprobe *tu;
1486	struct uprobe_dispatch_data udd;
1487	struct uprobe_cpu_buffer *ucb;
1488	int dsize, esize;
1489	int ret = 0;
1490
1491
1492	tu = container_of(con, struct trace_uprobe, consumer);
1493	tu->nhit++;
1494
1495	udd.tu = tu;
1496	udd.bp_addr = instruction_pointer(regs);
1497
1498	current->utask->vaddr = (unsigned long) &udd;
1499
1500	if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1501		return 0;
1502
1503	dsize = __get_data_size(&tu->tp, regs);
1504	esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1505
1506	ucb = uprobe_buffer_get();
1507	store_trace_args(ucb->buf, &tu->tp, regs, esize, dsize);
1508
1509	if (trace_probe_test_flag(&tu->tp, TP_FLAG_TRACE))
1510		ret |= uprobe_trace_func(tu, regs, ucb, dsize);
1511
1512#ifdef CONFIG_PERF_EVENTS
1513	if (trace_probe_test_flag(&tu->tp, TP_FLAG_PROFILE))
1514		ret |= uprobe_perf_func(tu, regs, ucb, dsize);
1515#endif
1516	uprobe_buffer_put(ucb);
1517	return ret;
1518}
1519
1520static int uretprobe_dispatcher(struct uprobe_consumer *con,
1521				unsigned long func, struct pt_regs *regs)
1522{
1523	struct trace_uprobe *tu;
1524	struct uprobe_dispatch_data udd;
1525	struct uprobe_cpu_buffer *ucb;
1526	int dsize, esize;
1527
1528	tu = container_of(con, struct trace_uprobe, consumer);
1529
1530	udd.tu = tu;
1531	udd.bp_addr = func;
1532
1533	current->utask->vaddr = (unsigned long) &udd;
1534
1535	if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1536		return 0;
1537
1538	dsize = __get_data_size(&tu->tp, regs);
1539	esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1540
1541	ucb = uprobe_buffer_get();
1542	store_trace_args(ucb->buf, &tu->tp, regs, esize, dsize);
1543
1544	if (trace_probe_test_flag(&tu->tp, TP_FLAG_TRACE))
1545		uretprobe_trace_func(tu, func, regs, ucb, dsize);
1546
1547#ifdef CONFIG_PERF_EVENTS
1548	if (trace_probe_test_flag(&tu->tp, TP_FLAG_PROFILE))
1549		uretprobe_perf_func(tu, func, regs, ucb, dsize);
1550#endif
1551	uprobe_buffer_put(ucb);
1552	return 0;
1553}
1554
1555static struct trace_event_functions uprobe_funcs = {
1556	.trace		= print_uprobe_event
1557};
1558
1559static struct trace_event_fields uprobe_fields_array[] = {
1560	{ .type = TRACE_FUNCTION_TYPE,
1561	  .define_fields = uprobe_event_define_fields },
1562	{}
1563};
1564
1565static inline void init_trace_event_call(struct trace_uprobe *tu)
1566{
1567	struct trace_event_call *call = trace_probe_event_call(&tu->tp);
1568	call->event.funcs = &uprobe_funcs;
1569	call->class->fields_array = uprobe_fields_array;
1570
1571	call->flags = TRACE_EVENT_FL_UPROBE | TRACE_EVENT_FL_CAP_ANY;
1572	call->class->reg = trace_uprobe_register;
1573}
1574
1575static int register_uprobe_event(struct trace_uprobe *tu)
1576{
1577	init_trace_event_call(tu);
1578
1579	return trace_probe_register_event_call(&tu->tp);
1580}
1581
1582static int unregister_uprobe_event(struct trace_uprobe *tu)
1583{
1584	return trace_probe_unregister_event_call(&tu->tp);
1585}
1586
1587#ifdef CONFIG_PERF_EVENTS
1588struct trace_event_call *
1589create_local_trace_uprobe(char *name, unsigned long offs,
1590			  unsigned long ref_ctr_offset, bool is_return)
1591{
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	init_trace_event_call(tu);
1625
1626	if (traceprobe_set_print_fmt(&tu->tp, is_ret_probe(tu)) < 0) {
1627		ret = -ENOMEM;
1628		goto error;
1629	}
1630
1631	return trace_probe_event_call(&tu->tp);
1632error:
1633	free_trace_uprobe(tu);
1634	return ERR_PTR(ret);
1635}
1636
1637void destroy_local_trace_uprobe(struct trace_event_call *event_call)
1638{
1639	struct trace_uprobe *tu;
1640
1641	tu = trace_uprobe_primary_from_call(event_call);
1642
1643	free_trace_uprobe(tu);
1644}
1645#endif /* CONFIG_PERF_EVENTS */
1646
1647/* Make a trace interface for controlling probe points */
1648static __init int init_uprobe_trace(void)
1649{
1650	int ret;
1651
1652	ret = dyn_event_register(&trace_uprobe_ops);
1653	if (ret)
1654		return ret;
1655
1656	ret = tracing_init_dentry();
1657	if (ret)
1658		return 0;
1659
1660	trace_create_file("uprobe_events", 0644, NULL,
1661				    NULL, &uprobe_events_ops);
1662	/* Profile interface */
1663	trace_create_file("uprobe_profile", 0444, NULL,
1664				    NULL, &uprobe_profile_ops);
1665	return 0;
1666}
1667
1668fs_initcall(init_uprobe_trace);