projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
tracing/kprobe: Wait for disabling all running kprobe handlers
[deliverable/linux.git] / kernel / trace / trace_kprobe.c
1 /*
2 * Kprobes-based tracing events
3 *
4 * Created by Masami Hiramatsu <mhiramat@redhat.com>
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 as
8 * published by the Free Software Foundation.
9 *
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
14 *
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write to the Free Software
17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
18 */
19
20 #include <linux/module.h>
21 #include <linux/uaccess.h>
22
23 #include "trace_probe.h"
24
25 #define KPROBE_EVENT_SYSTEM "kprobes"
26
27 /**
28 * Kprobe event core functions
29 */
30 struct trace_probe {
31 struct list_head list;
32 struct kretprobe rp; /* Use rp.kp for kprobe use */
33 unsigned long nhit;
34 unsigned int flags; /* For TP_FLAG_* */
35 const char *symbol; /* symbol name */
36 struct ftrace_event_class class;
37 struct ftrace_event_call call;
38 struct list_head files;
39 ssize_t size; /* trace entry size */
40 unsigned int nr_args;
41 struct probe_arg args[];
42 };
43
44 struct event_file_link {
45 struct ftrace_event_file *file;
46 struct list_head list;
47 };
48
49 #define SIZEOF_TRACE_PROBE(n) \
50 (offsetof(struct trace_probe, args) + \
51 (sizeof(struct probe_arg) * (n)))
52
53
54 static __kprobes bool trace_probe_is_return(struct trace_probe *tp)
55 {
56 return tp->rp.handler != NULL;
57 }
58
59 static __kprobes const char *trace_probe_symbol(struct trace_probe *tp)
60 {
61 return tp->symbol ? tp->symbol : "unknown";
62 }
63
64 static __kprobes unsigned long trace_probe_offset(struct trace_probe *tp)
65 {
66 return tp->rp.kp.offset;
67 }
68
69 static __kprobes bool trace_probe_is_enabled(struct trace_probe *tp)
70 {
71 return !!(tp->flags & (TP_FLAG_TRACE | TP_FLAG_PROFILE));
72 }
73
74 static __kprobes bool trace_probe_is_registered(struct trace_probe *tp)
75 {
76 return !!(tp->flags & TP_FLAG_REGISTERED);
77 }
78
79 static __kprobes bool trace_probe_has_gone(struct trace_probe *tp)
80 {
81 return !!(kprobe_gone(&tp->rp.kp));
82 }
83
84 static __kprobes bool trace_probe_within_module(struct trace_probe *tp,
85 struct module *mod)
86 {
87 int len = strlen(mod->name);
88 const char *name = trace_probe_symbol(tp);
89 return strncmp(mod->name, name, len) == 0 && name[len] == ':';
90 }
91
92 static __kprobes bool trace_probe_is_on_module(struct trace_probe *tp)
93 {
94 return !!strchr(trace_probe_symbol(tp), ':');
95 }
96
97 static int register_probe_event(struct trace_probe *tp);
98 static void unregister_probe_event(struct trace_probe *tp);
99
100 static DEFINE_MUTEX(probe_lock);
101 static LIST_HEAD(probe_list);
102
103 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
104 static int kretprobe_dispatcher(struct kretprobe_instance *ri,
105 struct pt_regs *regs);
106
107 /*
108 * Allocate new trace_probe and initialize it (including kprobes).
109 */
110 static struct trace_probe *alloc_trace_probe(const char *group,
111 const char *event,
112 void *addr,
113 const char *symbol,
114 unsigned long offs,
115 int nargs, bool is_return)
116 {
117 struct trace_probe *tp;
118 int ret = -ENOMEM;
119
120 tp = kzalloc(SIZEOF_TRACE_PROBE(nargs), GFP_KERNEL);
121 if (!tp)
122 return ERR_PTR(ret);
123
124 if (symbol) {
125 tp->symbol = kstrdup(symbol, GFP_KERNEL);
126 if (!tp->symbol)
127 goto error;
128 tp->rp.kp.symbol_name = tp->symbol;
129 tp->rp.kp.offset = offs;
130 } else
131 tp->rp.kp.addr = addr;
132
133 if (is_return)
134 tp->rp.handler = kretprobe_dispatcher;
135 else
136 tp->rp.kp.pre_handler = kprobe_dispatcher;
137
138 if (!event || !is_good_name(event)) {
139 ret = -EINVAL;
140 goto error;
141 }
142
143 tp->call.class = &tp->class;
144 tp->call.name = kstrdup(event, GFP_KERNEL);
145 if (!tp->call.name)
146 goto error;
147
148 if (!group || !is_good_name(group)) {
149 ret = -EINVAL;
150 goto error;
151 }
152
153 tp->class.system = kstrdup(group, GFP_KERNEL);
154 if (!tp->class.system)
155 goto error;
156
157 INIT_LIST_HEAD(&tp->list);
158 INIT_LIST_HEAD(&tp->files);
159 return tp;
160 error:
161 kfree(tp->call.name);
162 kfree(tp->symbol);
163 kfree(tp);
164 return ERR_PTR(ret);
165 }
166
167 static void free_trace_probe(struct trace_probe *tp)
168 {
169 int i;
170
171 for (i = 0; i < tp->nr_args; i++)
172 traceprobe_free_probe_arg(&tp->args[i]);
173
174 kfree(tp->call.class->system);
175 kfree(tp->call.name);
176 kfree(tp->symbol);
177 kfree(tp);
178 }
179
180 static struct trace_probe *find_trace_probe(const char *event,
181 const char *group)
182 {
183 struct trace_probe *tp;
184
185 list_for_each_entry(tp, &probe_list, list)
186 if (strcmp(tp->call.name, event) == 0 &&
187 strcmp(tp->call.class->system, group) == 0)
188 return tp;
189 return NULL;
190 }
191
192 /*
193 * Enable trace_probe
194 * if the file is NULL, enable "perf" handler, or enable "trace" handler.
195 */
196 static int
197 enable_trace_probe(struct trace_probe *tp, struct ftrace_event_file *file)
198 {
199 int ret = 0;
200
201 if (file) {
202 struct event_file_link *link;
203
204 link = kmalloc(sizeof(*link), GFP_KERNEL);
205 if (!link) {
206 ret = -ENOMEM;
207 goto out;
208 }
209
210 link->file = file;
211 list_add_tail_rcu(&link->list, &tp->files);
212
213 tp->flags |= TP_FLAG_TRACE;
214 } else
215 tp->flags |= TP_FLAG_PROFILE;
216
217 if (trace_probe_is_registered(tp) && !trace_probe_has_gone(tp)) {
218 if (trace_probe_is_return(tp))
219 ret = enable_kretprobe(&tp->rp);
220 else
221 ret = enable_kprobe(&tp->rp.kp);
222 }
223 out:
224 return ret;
225 }
226
227 static struct event_file_link *
228 find_event_file_link(struct trace_probe *tp, struct ftrace_event_file *file)
229 {
230 struct event_file_link *link;
231
232 list_for_each_entry(link, &tp->files, list)
233 if (link->file == file)
234 return link;
235
236 return NULL;
237 }
238
239 /*
240 * Disable trace_probe
241 * if the file is NULL, disable "perf" handler, or disable "trace" handler.
242 */
243 static int
244 disable_trace_probe(struct trace_probe *tp, struct ftrace_event_file *file)
245 {
246 struct event_file_link *link = NULL;
247 int wait = 0;
248 int ret = 0;
249
250 if (file) {
251 link = find_event_file_link(tp, file);
252 if (!link) {
253 ret = -EINVAL;
254 goto out;
255 }
256
257 list_del_rcu(&link->list);
258 wait = 1;
259 if (!list_empty(&tp->files))
260 goto out;
261
262 tp->flags &= ~TP_FLAG_TRACE;
263 } else
264 tp->flags &= ~TP_FLAG_PROFILE;
265
266 if (!trace_probe_is_enabled(tp) && trace_probe_is_registered(tp)) {
267 if (trace_probe_is_return(tp))
268 disable_kretprobe(&tp->rp);
269 else
270 disable_kprobe(&tp->rp.kp);
271 wait = 1;
272 }
273 out:
274 if (wait) {
275 /*
276 * Synchronize with kprobe_trace_func/kretprobe_trace_func
277 * to ensure disabled (all running handlers are finished).
278 * This is not only for kfree(), but also the caller,
279 * trace_remove_event_call() supposes it for releasing
280 * event_call related objects, which will be accessed in
281 * the kprobe_trace_func/kretprobe_trace_func.
282 */
283 synchronize_sched();
284 kfree(link); /* Ignored if link == NULL */
285 }
286
287 return ret;
288 }
289
290 /* Internal register function - just handle k*probes and flags */
291 static int __register_trace_probe(struct trace_probe *tp)
292 {
293 int i, ret;
294
295 if (trace_probe_is_registered(tp))
296 return -EINVAL;
297
298 for (i = 0; i < tp->nr_args; i++)
299 traceprobe_update_arg(&tp->args[i]);
300
301 /* Set/clear disabled flag according to tp->flag */
302 if (trace_probe_is_enabled(tp))
303 tp->rp.kp.flags &= ~KPROBE_FLAG_DISABLED;
304 else
305 tp->rp.kp.flags |= KPROBE_FLAG_DISABLED;
306
307 if (trace_probe_is_return(tp))
308 ret = register_kretprobe(&tp->rp);
309 else
310 ret = register_kprobe(&tp->rp.kp);
311
312 if (ret == 0)
313 tp->flags |= TP_FLAG_REGISTERED;
314 else {
315 pr_warning("Could not insert probe at %s+%lu: %d\n",
316 trace_probe_symbol(tp), trace_probe_offset(tp), ret);
317 if (ret == -ENOENT && trace_probe_is_on_module(tp)) {
318 pr_warning("This probe might be able to register after"
319 "target module is loaded. Continue.\n");
320 ret = 0;
321 } else if (ret == -EILSEQ) {
322 pr_warning("Probing address(0x%p) is not an "
323 "instruction boundary.\n",
324 tp->rp.kp.addr);
325 ret = -EINVAL;
326 }
327 }
328
329 return ret;
330 }
331
332 /* Internal unregister function - just handle k*probes and flags */
333 static void __unregister_trace_probe(struct trace_probe *tp)
334 {
335 if (trace_probe_is_registered(tp)) {
336 if (trace_probe_is_return(tp))
337 unregister_kretprobe(&tp->rp);
338 else
339 unregister_kprobe(&tp->rp.kp);
340 tp->flags &= ~TP_FLAG_REGISTERED;
341 /* Cleanup kprobe for reuse */
342 if (tp->rp.kp.symbol_name)
343 tp->rp.kp.addr = NULL;
344 }
345 }
346
347 /* Unregister a trace_probe and probe_event: call with locking probe_lock */
348 static int unregister_trace_probe(struct trace_probe *tp)
349 {
350 /* Enabled event can not be unregistered */
351 if (trace_probe_is_enabled(tp))
352 return -EBUSY;
353
354 __unregister_trace_probe(tp);
355 list_del(&tp->list);
356 unregister_probe_event(tp);
357
358 return 0;
359 }
360
361 /* Register a trace_probe and probe_event */
362 static int register_trace_probe(struct trace_probe *tp)
363 {
364 struct trace_probe *old_tp;
365 int ret;
366
367 mutex_lock(&probe_lock);
368
369 /* Delete old (same name) event if exist */
370 old_tp = find_trace_probe(tp->call.name, tp->call.class->system);
371 if (old_tp) {
372 ret = unregister_trace_probe(old_tp);
373 if (ret < 0)
374 goto end;
375 free_trace_probe(old_tp);
376 }
377
378 /* Register new event */
379 ret = register_probe_event(tp);
380 if (ret) {
381 pr_warning("Failed to register probe event(%d)\n", ret);
382 goto end;
383 }
384
385 /* Register k*probe */
386 ret = __register_trace_probe(tp);
387 if (ret < 0)
388 unregister_probe_event(tp);
389 else
390 list_add_tail(&tp->list, &probe_list);
391
392 end:
393 mutex_unlock(&probe_lock);
394 return ret;
395 }
396
397 /* Module notifier call back, checking event on the module */
398 static int trace_probe_module_callback(struct notifier_block *nb,
399 unsigned long val, void *data)
400 {
401 struct module *mod = data;
402 struct trace_probe *tp;
403 int ret;
404
405 if (val != MODULE_STATE_COMING)
406 return NOTIFY_DONE;
407
408 /* Update probes on coming module */
409 mutex_lock(&probe_lock);
410 list_for_each_entry(tp, &probe_list, list) {
411 if (trace_probe_within_module(tp, mod)) {
412 /* Don't need to check busy - this should have gone. */
413 __unregister_trace_probe(tp);
414 ret = __register_trace_probe(tp);
415 if (ret)
416 pr_warning("Failed to re-register probe %s on"
417 "%s: %d\n",
418 tp->call.name, mod->name, ret);
419 }
420 }
421 mutex_unlock(&probe_lock);
422
423 return NOTIFY_DONE;
424 }
425
426 static struct notifier_block trace_probe_module_nb = {
427 .notifier_call = trace_probe_module_callback,
428 .priority = 1 /* Invoked after kprobe module callback */
429 };
430
431 static int create_trace_probe(int argc, char **argv)
432 {
433 /*
434 * Argument syntax:
435 * - Add kprobe: p[:[GRP/]EVENT] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
436 * - Add kretprobe: r[:[GRP/]EVENT] [MOD:]KSYM[+0] [FETCHARGS]
437 * Fetch args:
438 * $retval : fetch return value
439 * $stack : fetch stack address
440 * $stackN : fetch Nth of stack (N:0-)
441 * @ADDR : fetch memory at ADDR (ADDR should be in kernel)
442 * @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
443 * %REG : fetch register REG
444 * Dereferencing memory fetch:
445 * +|-offs(ARG) : fetch memory at ARG +|- offs address.
446 * Alias name of args:
447 * NAME=FETCHARG : set NAME as alias of FETCHARG.
448 * Type of args:
449 * FETCHARG:TYPE : use TYPE instead of unsigned long.
450 */
451 struct trace_probe *tp;
452 int i, ret = 0;
453 bool is_return = false, is_delete = false;
454 char *symbol = NULL, *event = NULL, *group = NULL;
455 char *arg;
456 unsigned long offset = 0;
457 void *addr = NULL;
458 char buf[MAX_EVENT_NAME_LEN];
459
460 /* argc must be >= 1 */
461 if (argv[0][0] == 'p')
462 is_return = false;
463 else if (argv[0][0] == 'r')
464 is_return = true;
465 else if (argv[0][0] == '-')
466 is_delete = true;
467 else {
468 pr_info("Probe definition must be started with 'p', 'r' or"
469 " '-'.\n");
470 return -EINVAL;
471 }
472
473 if (argv[0][1] == ':') {
474 event = &argv[0][2];
475 if (strchr(event, '/')) {
476 group = event;
477 event = strchr(group, '/') + 1;
478 event[-1] = '\0';
479 if (strlen(group) == 0) {
480 pr_info("Group name is not specified\n");
481 return -EINVAL;
482 }
483 }
484 if (strlen(event) == 0) {
485 pr_info("Event name is not specified\n");
486 return -EINVAL;
487 }
488 }
489 if (!group)
490 group = KPROBE_EVENT_SYSTEM;
491
492 if (is_delete) {
493 if (!event) {
494 pr_info("Delete command needs an event name.\n");
495 return -EINVAL;
496 }
497 mutex_lock(&probe_lock);
498 tp = find_trace_probe(event, group);
499 if (!tp) {
500 mutex_unlock(&probe_lock);
501 pr_info("Event %s/%s doesn't exist.\n", group, event);
502 return -ENOENT;
503 }
504 /* delete an event */
505 ret = unregister_trace_probe(tp);
506 if (ret == 0)
507 free_trace_probe(tp);
508 mutex_unlock(&probe_lock);
509 return ret;
510 }
511
512 if (argc < 2) {
513 pr_info("Probe point is not specified.\n");
514 return -EINVAL;
515 }
516 if (isdigit(argv[1][0])) {
517 if (is_return) {
518 pr_info("Return probe point must be a symbol.\n");
519 return -EINVAL;
520 }
521 /* an address specified */
522 ret = kstrtoul(&argv[1][0], 0, (unsigned long *)&addr);
523 if (ret) {
524 pr_info("Failed to parse address.\n");
525 return ret;
526 }
527 } else {
528 /* a symbol specified */
529 symbol = argv[1];
530 /* TODO: support .init module functions */
531 ret = traceprobe_split_symbol_offset(symbol, &offset);
532 if (ret) {
533 pr_info("Failed to parse symbol.\n");
534 return ret;
535 }
536 if (offset && is_return) {
537 pr_info("Return probe must be used without offset.\n");
538 return -EINVAL;
539 }
540 }
541 argc -= 2; argv += 2;
542
543 /* setup a probe */
544 if (!event) {
545 /* Make a new event name */
546 if (symbol)
547 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
548 is_return ? 'r' : 'p', symbol, offset);
549 else
550 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
551 is_return ? 'r' : 'p', addr);
552 event = buf;
553 }
554 tp = alloc_trace_probe(group, event, addr, symbol, offset, argc,
555 is_return);
556 if (IS_ERR(tp)) {
557 pr_info("Failed to allocate trace_probe.(%d)\n",
558 (int)PTR_ERR(tp));
559 return PTR_ERR(tp);
560 }
561
562 /* parse arguments */
563 ret = 0;
564 for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
565 /* Increment count for freeing args in error case */
566 tp->nr_args++;
567
568 /* Parse argument name */
569 arg = strchr(argv[i], '=');
570 if (arg) {
571 *arg++ = '\0';
572 tp->args[i].name = kstrdup(argv[i], GFP_KERNEL);
573 } else {
574 arg = argv[i];
575 /* If argument name is omitted, set "argN" */
576 snprintf(buf, MAX_EVENT_NAME_LEN, "arg%d", i + 1);
577 tp->args[i].name = kstrdup(buf, GFP_KERNEL);
578 }
579
580 if (!tp->args[i].name) {
581 pr_info("Failed to allocate argument[%d] name.\n", i);
582 ret = -ENOMEM;
583 goto error;
584 }
585
586 if (!is_good_name(tp->args[i].name)) {
587 pr_info("Invalid argument[%d] name: %s\n",
588 i, tp->args[i].name);
589 ret = -EINVAL;
590 goto error;
591 }
592
593 if (traceprobe_conflict_field_name(tp->args[i].name,
594 tp->args, i)) {
595 pr_info("Argument[%d] name '%s' conflicts with "
596 "another field.\n", i, argv[i]);
597 ret = -EINVAL;
598 goto error;
599 }
600
601 /* Parse fetch argument */
602 ret = traceprobe_parse_probe_arg(arg, &tp->size, &tp->args[i],
603 is_return, true);
604 if (ret) {
605 pr_info("Parse error at argument[%d]. (%d)\n", i, ret);
606 goto error;
607 }
608 }
609
610 ret = register_trace_probe(tp);
611 if (ret)
612 goto error;
613 return 0;
614
615 error:
616 free_trace_probe(tp);
617 return ret;
618 }
619
620 static int release_all_trace_probes(void)
621 {
622 struct trace_probe *tp;
623 int ret = 0;
624
625 mutex_lock(&probe_lock);
626 /* Ensure no probe is in use. */
627 list_for_each_entry(tp, &probe_list, list)
628 if (trace_probe_is_enabled(tp)) {
629 ret = -EBUSY;
630 goto end;
631 }
632 /* TODO: Use batch unregistration */
633 while (!list_empty(&probe_list)) {
634 tp = list_entry(probe_list.next, struct trace_probe, list);
635 unregister_trace_probe(tp);
636 free_trace_probe(tp);
637 }
638
639 end:
640 mutex_unlock(&probe_lock);
641
642 return ret;
643 }
644
645 /* Probes listing interfaces */
646 static void *probes_seq_start(struct seq_file *m, loff_t *pos)
647 {
648 mutex_lock(&probe_lock);
649 return seq_list_start(&probe_list, *pos);
650 }
651
652 static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
653 {
654 return seq_list_next(v, &probe_list, pos);
655 }
656
657 static void probes_seq_stop(struct seq_file *m, void *v)
658 {
659 mutex_unlock(&probe_lock);
660 }
661
662 static int probes_seq_show(struct seq_file *m, void *v)
663 {
664 struct trace_probe *tp = v;
665 int i;
666
667 seq_printf(m, "%c", trace_probe_is_return(tp) ? 'r' : 'p');
668 seq_printf(m, ":%s/%s", tp->call.class->system, tp->call.name);
669
670 if (!tp->symbol)
671 seq_printf(m, " 0x%p", tp->rp.kp.addr);
672 else if (tp->rp.kp.offset)
673 seq_printf(m, " %s+%u", trace_probe_symbol(tp),
674 tp->rp.kp.offset);
675 else
676 seq_printf(m, " %s", trace_probe_symbol(tp));
677
678 for (i = 0; i < tp->nr_args; i++)
679 seq_printf(m, " %s=%s", tp->args[i].name, tp->args[i].comm);
680 seq_printf(m, "\n");
681
682 return 0;
683 }
684
685 static const struct seq_operations probes_seq_op = {
686 .start = probes_seq_start,
687 .next = probes_seq_next,
688 .stop = probes_seq_stop,
689 .show = probes_seq_show
690 };
691
692 static int probes_open(struct inode *inode, struct file *file)
693 {
694 int ret;
695
696 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
697 ret = release_all_trace_probes();
698 if (ret < 0)
699 return ret;
700 }
701
702 return seq_open(file, &probes_seq_op);
703 }
704
705 static ssize_t probes_write(struct file *file, const char __user *buffer,
706 size_t count, loff_t *ppos)
707 {
708 return traceprobe_probes_write(file, buffer, count, ppos,
709 create_trace_probe);
710 }
711
712 static const struct file_operations kprobe_events_ops = {
713 .owner = THIS_MODULE,
714 .open = probes_open,
715 .read = seq_read,
716 .llseek = seq_lseek,
717 .release = seq_release,
718 .write = probes_write,
719 };
720
721 /* Probes profiling interfaces */
722 static int probes_profile_seq_show(struct seq_file *m, void *v)
723 {
724 struct trace_probe *tp = v;
725
726 seq_printf(m, " %-44s %15lu %15lu\n", tp->call.name, tp->nhit,
727 tp->rp.kp.nmissed);
728
729 return 0;
730 }
731
732 static const struct seq_operations profile_seq_op = {
733 .start = probes_seq_start,
734 .next = probes_seq_next,
735 .stop = probes_seq_stop,
736 .show = probes_profile_seq_show
737 };
738
739 static int profile_open(struct inode *inode, struct file *file)
740 {
741 return seq_open(file, &profile_seq_op);
742 }
743
744 static const struct file_operations kprobe_profile_ops = {
745 .owner = THIS_MODULE,
746 .open = profile_open,
747 .read = seq_read,
748 .llseek = seq_lseek,
749 .release = seq_release,
750 };
751
752 /* Sum up total data length for dynamic arraies (strings) */
753 static __kprobes int __get_data_size(struct trace_probe *tp,
754 struct pt_regs *regs)
755 {
756 int i, ret = 0;
757 u32 len;
758
759 for (i = 0; i < tp->nr_args; i++)
760 if (unlikely(tp->args[i].fetch_size.fn)) {
761 call_fetch(&tp->args[i].fetch_size, regs, &len);
762 ret += len;
763 }
764
765 return ret;
766 }
767
768 /* Store the value of each argument */
769 static __kprobes void store_trace_args(int ent_size, struct trace_probe *tp,
770 struct pt_regs *regs,
771 u8 *data, int maxlen)
772 {
773 int i;
774 u32 end = tp->size;
775 u32 *dl; /* Data (relative) location */
776
777 for (i = 0; i < tp->nr_args; i++) {
778 if (unlikely(tp->args[i].fetch_size.fn)) {
779 /*
780 * First, we set the relative location and
781 * maximum data length to *dl
782 */
783 dl = (u32 *)(data + tp->args[i].offset);
784 *dl = make_data_rloc(maxlen, end - tp->args[i].offset);
785 /* Then try to fetch string or dynamic array data */
786 call_fetch(&tp->args[i].fetch, regs, dl);
787 /* Reduce maximum length */
788 end += get_rloc_len(*dl);
789 maxlen -= get_rloc_len(*dl);
790 /* Trick here, convert data_rloc to data_loc */
791 *dl = convert_rloc_to_loc(*dl,
792 ent_size + tp->args[i].offset);
793 } else
794 /* Just fetching data normally */
795 call_fetch(&tp->args[i].fetch, regs,
796 data + tp->args[i].offset);
797 }
798 }
799
800 /* Kprobe handler */
801 static __kprobes void
802 __kprobe_trace_func(struct trace_probe *tp, struct pt_regs *regs,
803 struct ftrace_event_file *ftrace_file)
804 {
805 struct kprobe_trace_entry_head *entry;
806 struct ring_buffer_event *event;
807 struct ring_buffer *buffer;
808 int size, dsize, pc;
809 unsigned long irq_flags;
810 struct ftrace_event_call *call = &tp->call;
811
812 WARN_ON(call != ftrace_file->event_call);
813
814 if (test_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT, &ftrace_file->flags))
815 return;
816
817 local_save_flags(irq_flags);
818 pc = preempt_count();
819
820 dsize = __get_data_size(tp, regs);
821 size = sizeof(*entry) + tp->size + dsize;
822
823 event = trace_event_buffer_lock_reserve(&buffer, ftrace_file,
824 call->event.type,
825 size, irq_flags, pc);
826 if (!event)
827 return;
828
829 entry = ring_buffer_event_data(event);
830 entry->ip = (unsigned long)tp->rp.kp.addr;
831 store_trace_args(sizeof(*entry), tp, regs, (u8 *)&entry[1], dsize);
832
833 if (!filter_current_check_discard(buffer, call, entry, event))
834 trace_buffer_unlock_commit_regs(buffer, event,
835 irq_flags, pc, regs);
836 }
837
838 static __kprobes void
839 kprobe_trace_func(struct trace_probe *tp, struct pt_regs *regs)
840 {
841 struct event_file_link *link;
842
843 list_for_each_entry_rcu(link, &tp->files, list)
844 __kprobe_trace_func(tp, regs, link->file);
845 }
846
847 /* Kretprobe handler */
848 static __kprobes void
849 __kretprobe_trace_func(struct trace_probe *tp, struct kretprobe_instance *ri,
850 struct pt_regs *regs,
851 struct ftrace_event_file *ftrace_file)
852 {
853 struct kretprobe_trace_entry_head *entry;
854 struct ring_buffer_event *event;
855 struct ring_buffer *buffer;
856 int size, pc, dsize;
857 unsigned long irq_flags;
858 struct ftrace_event_call *call = &tp->call;
859
860 WARN_ON(call != ftrace_file->event_call);
861
862 if (test_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT, &ftrace_file->flags))
863 return;
864
865 local_save_flags(irq_flags);
866 pc = preempt_count();
867
868 dsize = __get_data_size(tp, regs);
869 size = sizeof(*entry) + tp->size + dsize;
870
871 event = trace_event_buffer_lock_reserve(&buffer, ftrace_file,
872 call->event.type,
873 size, irq_flags, pc);
874 if (!event)
875 return;
876
877 entry = ring_buffer_event_data(event);
878 entry->func = (unsigned long)tp->rp.kp.addr;
879 entry->ret_ip = (unsigned long)ri->ret_addr;
880 store_trace_args(sizeof(*entry), tp, regs, (u8 *)&entry[1], dsize);
881
882 if (!filter_current_check_discard(buffer, call, entry, event))
883 trace_buffer_unlock_commit_regs(buffer, event,
884 irq_flags, pc, regs);
885 }
886
887 static __kprobes void
888 kretprobe_trace_func(struct trace_probe *tp, struct kretprobe_instance *ri,
889 struct pt_regs *regs)
890 {
891 struct event_file_link *link;
892
893 list_for_each_entry_rcu(link, &tp->files, list)
894 __kretprobe_trace_func(tp, ri, regs, link->file);
895 }
896
897 /* Event entry printers */
898 static enum print_line_t
899 print_kprobe_event(struct trace_iterator *iter, int flags,
900 struct trace_event *event)
901 {
902 struct kprobe_trace_entry_head *field;
903 struct trace_seq *s = &iter->seq;
904 struct trace_probe *tp;
905 u8 *data;
906 int i;
907
908 field = (struct kprobe_trace_entry_head *)iter->ent;
909 tp = container_of(event, struct trace_probe, call.event);
910
911 if (!trace_seq_printf(s, "%s: (", tp->call.name))
912 goto partial;
913
914 if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
915 goto partial;
916
917 if (!trace_seq_puts(s, ")"))
918 goto partial;
919
920 data = (u8 *)&field[1];
921 for (i = 0; i < tp->nr_args; i++)
922 if (!tp->args[i].type->print(s, tp->args[i].name,
923 data + tp->args[i].offset, field))
924 goto partial;
925
926 if (!trace_seq_puts(s, "\n"))
927 goto partial;
928
929 return TRACE_TYPE_HANDLED;
930 partial:
931 return TRACE_TYPE_PARTIAL_LINE;
932 }
933
934 static enum print_line_t
935 print_kretprobe_event(struct trace_iterator *iter, int flags,
936 struct trace_event *event)
937 {
938 struct kretprobe_trace_entry_head *field;
939 struct trace_seq *s = &iter->seq;
940 struct trace_probe *tp;
941 u8 *data;
942 int i;
943
944 field = (struct kretprobe_trace_entry_head *)iter->ent;
945 tp = container_of(event, struct trace_probe, call.event);
946
947 if (!trace_seq_printf(s, "%s: (", tp->call.name))
948 goto partial;
949
950 if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
951 goto partial;
952
953 if (!trace_seq_puts(s, " <- "))
954 goto partial;
955
956 if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
957 goto partial;
958
959 if (!trace_seq_puts(s, ")"))
960 goto partial;
961
962 data = (u8 *)&field[1];
963 for (i = 0; i < tp->nr_args; i++)
964 if (!tp->args[i].type->print(s, tp->args[i].name,
965 data + tp->args[i].offset, field))
966 goto partial;
967
968 if (!trace_seq_puts(s, "\n"))
969 goto partial;
970
971 return TRACE_TYPE_HANDLED;
972 partial:
973 return TRACE_TYPE_PARTIAL_LINE;
974 }
975
976
977 static int kprobe_event_define_fields(struct ftrace_event_call *event_call)
978 {
979 int ret, i;
980 struct kprobe_trace_entry_head field;
981 struct trace_probe *tp = (struct trace_probe *)event_call->data;
982
983 DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
984 /* Set argument names as fields */
985 for (i = 0; i < tp->nr_args; i++) {
986 ret = trace_define_field(event_call, tp->args[i].type->fmttype,
987 tp->args[i].name,
988 sizeof(field) + tp->args[i].offset,
989 tp->args[i].type->size,
990 tp->args[i].type->is_signed,
991 FILTER_OTHER);
992 if (ret)
993 return ret;
994 }
995 return 0;
996 }
997
998 static int kretprobe_event_define_fields(struct ftrace_event_call *event_call)
999 {
1000 int ret, i;
1001 struct kretprobe_trace_entry_head field;
1002 struct trace_probe *tp = (struct trace_probe *)event_call->data;
1003
1004 DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1005 DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1006 /* Set argument names as fields */
1007 for (i = 0; i < tp->nr_args; i++) {
1008 ret = trace_define_field(event_call, tp->args[i].type->fmttype,
1009 tp->args[i].name,
1010 sizeof(field) + tp->args[i].offset,
1011 tp->args[i].type->size,
1012 tp->args[i].type->is_signed,
1013 FILTER_OTHER);
1014 if (ret)
1015 return ret;
1016 }
1017 return 0;
1018 }
1019
1020 static int __set_print_fmt(struct trace_probe *tp, char *buf, int len)
1021 {
1022 int i;
1023 int pos = 0;
1024
1025 const char *fmt, *arg;
1026
1027 if (!trace_probe_is_return(tp)) {
1028 fmt = "(%lx)";
1029 arg = "REC->" FIELD_STRING_IP;
1030 } else {
1031 fmt = "(%lx <- %lx)";
1032 arg = "REC->" FIELD_STRING_FUNC ", REC->" FIELD_STRING_RETIP;
1033 }
1034
1035 /* When len=0, we just calculate the needed length */
1036 #define LEN_OR_ZERO (len ? len - pos : 0)
1037
1038 pos += snprintf(buf + pos, LEN_OR_ZERO, "\"%s", fmt);
1039
1040 for (i = 0; i < tp->nr_args; i++) {
1041 pos += snprintf(buf + pos, LEN_OR_ZERO, " %s=%s",
1042 tp->args[i].name, tp->args[i].type->fmt);
1043 }
1044
1045 pos += snprintf(buf + pos, LEN_OR_ZERO, "\", %s", arg);
1046
1047 for (i = 0; i < tp->nr_args; i++) {
1048 if (strcmp(tp->args[i].type->name, "string") == 0)
1049 pos += snprintf(buf + pos, LEN_OR_ZERO,
1050 ", __get_str(%s)",
1051 tp->args[i].name);
1052 else
1053 pos += snprintf(buf + pos, LEN_OR_ZERO, ", REC->%s",
1054 tp->args[i].name);
1055 }
1056
1057 #undef LEN_OR_ZERO
1058
1059 /* return the length of print_fmt */
1060 return pos;
1061 }
1062
1063 static int set_print_fmt(struct trace_probe *tp)
1064 {
1065 int len;
1066 char *print_fmt;
1067
1068 /* First: called with 0 length to calculate the needed length */
1069 len = __set_print_fmt(tp, NULL, 0);
1070 print_fmt = kmalloc(len + 1, GFP_KERNEL);
1071 if (!print_fmt)
1072 return -ENOMEM;
1073
1074 /* Second: actually write the @print_fmt */
1075 __set_print_fmt(tp, print_fmt, len + 1);
1076 tp->call.print_fmt = print_fmt;
1077
1078 return 0;
1079 }
1080
1081 #ifdef CONFIG_PERF_EVENTS
1082
1083 /* Kprobe profile handler */
1084 static __kprobes void
1085 kprobe_perf_func(struct trace_probe *tp, struct pt_regs *regs)
1086 {
1087 struct ftrace_event_call *call = &tp->call;
1088 struct kprobe_trace_entry_head *entry;
1089 struct hlist_head *head;
1090 int size, __size, dsize;
1091 int rctx;
1092
1093 head = this_cpu_ptr(call->perf_events);
1094 if (hlist_empty(head))
1095 return;
1096
1097 dsize = __get_data_size(tp, regs);
1098 __size = sizeof(*entry) + tp->size + dsize;
1099 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1100 size -= sizeof(u32);
1101
1102 entry = perf_trace_buf_prepare(size, call->event.type, regs, &rctx);
1103 if (!entry)
1104 return;
1105
1106 entry->ip = (unsigned long)tp->rp.kp.addr;
1107 memset(&entry[1], 0, dsize);
1108 store_trace_args(sizeof(*entry), tp, regs, (u8 *)&entry[1], dsize);
1109 perf_trace_buf_submit(entry, size, rctx, 0, 1, regs, head, NULL);
1110 }
1111
1112 /* Kretprobe profile handler */
1113 static __kprobes void
1114 kretprobe_perf_func(struct trace_probe *tp, struct kretprobe_instance *ri,
1115 struct pt_regs *regs)
1116 {
1117 struct ftrace_event_call *call = &tp->call;
1118 struct kretprobe_trace_entry_head *entry;
1119 struct hlist_head *head;
1120 int size, __size, dsize;
1121 int rctx;
1122
1123 head = this_cpu_ptr(call->perf_events);
1124 if (hlist_empty(head))
1125 return;
1126
1127 dsize = __get_data_size(tp, regs);
1128 __size = sizeof(*entry) + tp->size + dsize;
1129 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1130 size -= sizeof(u32);
1131
1132 entry = perf_trace_buf_prepare(size, call->event.type, regs, &rctx);
1133 if (!entry)
1134 return;
1135
1136 entry->func = (unsigned long)tp->rp.kp.addr;
1137 entry->ret_ip = (unsigned long)ri->ret_addr;
1138 store_trace_args(sizeof(*entry), tp, regs, (u8 *)&entry[1], dsize);
1139 perf_trace_buf_submit(entry, size, rctx, 0, 1, regs, head, NULL);
1140 }
1141 #endif /* CONFIG_PERF_EVENTS */
1142
1143 /*
1144 * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex.
1145 *
1146 * kprobe_trace_self_tests_init() does enable_trace_probe/disable_trace_probe
1147 * lockless, but we can't race with this __init function.
1148 */
1149 static __kprobes
1150 int kprobe_register(struct ftrace_event_call *event,
1151 enum trace_reg type, void *data)
1152 {
1153 struct trace_probe *tp = (struct trace_probe *)event->data;
1154 struct ftrace_event_file *file = data;
1155
1156 switch (type) {
1157 case TRACE_REG_REGISTER:
1158 return enable_trace_probe(tp, file);
1159 case TRACE_REG_UNREGISTER:
1160 return disable_trace_probe(tp, file);
1161
1162 #ifdef CONFIG_PERF_EVENTS
1163 case TRACE_REG_PERF_REGISTER:
1164 return enable_trace_probe(tp, NULL);
1165 case TRACE_REG_PERF_UNREGISTER:
1166 return disable_trace_probe(tp, NULL);
1167 case TRACE_REG_PERF_OPEN:
1168 case TRACE_REG_PERF_CLOSE:
1169 case TRACE_REG_PERF_ADD:
1170 case TRACE_REG_PERF_DEL:
1171 return 0;
1172 #endif
1173 }
1174 return 0;
1175 }
1176
1177 static __kprobes
1178 int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1179 {
1180 struct trace_probe *tp = container_of(kp, struct trace_probe, rp.kp);
1181
1182 tp->nhit++;
1183
1184 if (tp->flags & TP_FLAG_TRACE)
1185 kprobe_trace_func(tp, regs);
1186 #ifdef CONFIG_PERF_EVENTS
1187 if (tp->flags & TP_FLAG_PROFILE)
1188 kprobe_perf_func(tp, regs);
1189 #endif
1190 return 0; /* We don't tweek kernel, so just return 0 */
1191 }
1192
1193 static __kprobes
1194 int kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1195 {
1196 struct trace_probe *tp = container_of(ri->rp, struct trace_probe, rp);
1197
1198 tp->nhit++;
1199
1200 if (tp->flags & TP_FLAG_TRACE)
1201 kretprobe_trace_func(tp, ri, regs);
1202 #ifdef CONFIG_PERF_EVENTS
1203 if (tp->flags & TP_FLAG_PROFILE)
1204 kretprobe_perf_func(tp, ri, regs);
1205 #endif
1206 return 0; /* We don't tweek kernel, so just return 0 */
1207 }
1208
1209 static struct trace_event_functions kretprobe_funcs = {
1210 .trace = print_kretprobe_event
1211 };
1212
1213 static struct trace_event_functions kprobe_funcs = {
1214 .trace = print_kprobe_event
1215 };
1216
1217 static int register_probe_event(struct trace_probe *tp)
1218 {
1219 struct ftrace_event_call *call = &tp->call;
1220 int ret;
1221
1222 /* Initialize ftrace_event_call */
1223 INIT_LIST_HEAD(&call->class->fields);
1224 if (trace_probe_is_return(tp)) {
1225 call->event.funcs = &kretprobe_funcs;
1226 call->class->define_fields = kretprobe_event_define_fields;
1227 } else {
1228 call->event.funcs = &kprobe_funcs;
1229 call->class->define_fields = kprobe_event_define_fields;
1230 }
1231 if (set_print_fmt(tp) < 0)
1232 return -ENOMEM;
1233 ret = register_ftrace_event(&call->event);
1234 if (!ret) {
1235 kfree(call->print_fmt);
1236 return -ENODEV;
1237 }
1238 call->flags = 0;
1239 call->class->reg = kprobe_register;
1240 call->data = tp;
1241 ret = trace_add_event_call(call);
1242 if (ret) {
1243 pr_info("Failed to register kprobe event: %s\n", call->name);
1244 kfree(call->print_fmt);
1245 unregister_ftrace_event(&call->event);
1246 }
1247 return ret;
1248 }
1249
1250 static void unregister_probe_event(struct trace_probe *tp)
1251 {
1252 /* tp->event is unregistered in trace_remove_event_call() */
1253 trace_remove_event_call(&tp->call);
1254 kfree(tp->call.print_fmt);
1255 }
1256
1257 /* Make a debugfs interface for controlling probe points */
1258 static __init int init_kprobe_trace(void)
1259 {
1260 struct dentry *d_tracer;
1261 struct dentry *entry;
1262
1263 if (register_module_notifier(&trace_probe_module_nb))
1264 return -EINVAL;
1265
1266 d_tracer = tracing_init_dentry();
1267 if (!d_tracer)
1268 return 0;
1269
1270 entry = debugfs_create_file("kprobe_events", 0644, d_tracer,
1271 NULL, &kprobe_events_ops);
1272
1273 /* Event list interface */
1274 if (!entry)
1275 pr_warning("Could not create debugfs "
1276 "'kprobe_events' entry\n");
1277
1278 /* Profile interface */
1279 entry = debugfs_create_file("kprobe_profile", 0444, d_tracer,
1280 NULL, &kprobe_profile_ops);
1281
1282 if (!entry)
1283 pr_warning("Could not create debugfs "
1284 "'kprobe_profile' entry\n");
1285 return 0;
1286 }
1287 fs_initcall(init_kprobe_trace);
1288
1289
1290 #ifdef CONFIG_FTRACE_STARTUP_TEST
1291
1292 /*
1293 * The "__used" keeps gcc from removing the function symbol
1294 * from the kallsyms table.
1295 */
1296 static __used int kprobe_trace_selftest_target(int a1, int a2, int a3,
1297 int a4, int a5, int a6)
1298 {
1299 return a1 + a2 + a3 + a4 + a5 + a6;
1300 }
1301
1302 static struct ftrace_event_file *
1303 find_trace_probe_file(struct trace_probe *tp, struct trace_array *tr)
1304 {
1305 struct ftrace_event_file *file;
1306
1307 list_for_each_entry(file, &tr->events, list)
1308 if (file->event_call == &tp->call)
1309 return file;
1310
1311 return NULL;
1312 }
1313
1314 /*
1315 * Nobody but us can call enable_trace_probe/disable_trace_probe at this
1316 * stage, we can do this lockless.
1317 */
1318 static __init int kprobe_trace_self_tests_init(void)
1319 {
1320 int ret, warn = 0;
1321 int (*target)(int, int, int, int, int, int);
1322 struct trace_probe *tp;
1323 struct ftrace_event_file *file;
1324
1325 target = kprobe_trace_selftest_target;
1326
1327 pr_info("Testing kprobe tracing: ");
1328
1329 ret = traceprobe_command("p:testprobe kprobe_trace_selftest_target "
1330 "$stack $stack0 +0($stack)",
1331 create_trace_probe);
1332 if (WARN_ON_ONCE(ret)) {
1333 pr_warn("error on probing function entry.\n");
1334 warn++;
1335 } else {
1336 /* Enable trace point */
1337 tp = find_trace_probe("testprobe", KPROBE_EVENT_SYSTEM);
1338 if (WARN_ON_ONCE(tp == NULL)) {
1339 pr_warn("error on getting new probe.\n");
1340 warn++;
1341 } else {
1342 file = find_trace_probe_file(tp, top_trace_array());
1343 if (WARN_ON_ONCE(file == NULL)) {
1344 pr_warn("error on getting probe file.\n");
1345 warn++;
1346 } else
1347 enable_trace_probe(tp, file);
1348 }
1349 }
1350
1351 ret = traceprobe_command("r:testprobe2 kprobe_trace_selftest_target "
1352 "$retval", create_trace_probe);
1353 if (WARN_ON_ONCE(ret)) {
1354 pr_warn("error on probing function return.\n");
1355 warn++;
1356 } else {
1357 /* Enable trace point */
1358 tp = find_trace_probe("testprobe2", KPROBE_EVENT_SYSTEM);
1359 if (WARN_ON_ONCE(tp == NULL)) {
1360 pr_warn("error on getting 2nd new probe.\n");
1361 warn++;
1362 } else {
1363 file = find_trace_probe_file(tp, top_trace_array());
1364 if (WARN_ON_ONCE(file == NULL)) {
1365 pr_warn("error on getting probe file.\n");
1366 warn++;
1367 } else
1368 enable_trace_probe(tp, file);
1369 }
1370 }
1371
1372 if (warn)
1373 goto end;
1374
1375 ret = target(1, 2, 3, 4, 5, 6);
1376
1377 /* Disable trace points before removing it */
1378 tp = find_trace_probe("testprobe", KPROBE_EVENT_SYSTEM);
1379 if (WARN_ON_ONCE(tp == NULL)) {
1380 pr_warn("error on getting test probe.\n");
1381 warn++;
1382 } else {
1383 file = find_trace_probe_file(tp, top_trace_array());
1384 if (WARN_ON_ONCE(file == NULL)) {
1385 pr_warn("error on getting probe file.\n");
1386 warn++;
1387 } else
1388 disable_trace_probe(tp, file);
1389 }
1390
1391 tp = find_trace_probe("testprobe2", KPROBE_EVENT_SYSTEM);
1392 if (WARN_ON_ONCE(tp == NULL)) {
1393 pr_warn("error on getting 2nd test probe.\n");
1394 warn++;
1395 } else {
1396 file = find_trace_probe_file(tp, top_trace_array());
1397 if (WARN_ON_ONCE(file == NULL)) {
1398 pr_warn("error on getting probe file.\n");
1399 warn++;
1400 } else
1401 disable_trace_probe(tp, file);
1402 }
1403
1404 ret = traceprobe_command("-:testprobe", create_trace_probe);
1405 if (WARN_ON_ONCE(ret)) {
1406 pr_warn("error on deleting a probe.\n");
1407 warn++;
1408 }
1409
1410 ret = traceprobe_command("-:testprobe2", create_trace_probe);
1411 if (WARN_ON_ONCE(ret)) {
1412 pr_warn("error on deleting a probe.\n");
1413 warn++;
1414 }
1415
1416 end:
1417 release_all_trace_probes();
1418 if (warn)
1419 pr_cont("NG: Some tests are failed. Please check them.\n");
1420 else
1421 pr_cont("OK\n");
1422 return 0;
1423 }
1424
1425 late_initcall(kprobe_trace_self_tests_init);
1426
1427 #endif
Linux kernel - Deliverables
This page took 0.060129 seconds and 5 git commands to generate.