Loading...
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
31struct trace_probe {
32 struct list_head list;
33 struct kretprobe rp; /* Use rp.kp for kprobe use */
34 unsigned long nhit;
35 unsigned int flags; /* For TP_FLAG_* */
36 const char *symbol; /* symbol name */
37 struct ftrace_event_class class;
38 struct ftrace_event_call call;
39 ssize_t size; /* trace entry size */
40 unsigned int nr_args;
41 struct probe_arg args[];
42};
43
44#define SIZEOF_TRACE_PROBE(n) \
45 (offsetof(struct trace_probe, args) + \
46 (sizeof(struct probe_arg) * (n)))
47
48
49static __kprobes int trace_probe_is_return(struct trace_probe *tp)
50{
51 return tp->rp.handler != NULL;
52}
53
54static __kprobes const char *trace_probe_symbol(struct trace_probe *tp)
55{
56 return tp->symbol ? tp->symbol : "unknown";
57}
58
59static __kprobes unsigned long trace_probe_offset(struct trace_probe *tp)
60{
61 return tp->rp.kp.offset;
62}
63
64static __kprobes bool trace_probe_is_enabled(struct trace_probe *tp)
65{
66 return !!(tp->flags & (TP_FLAG_TRACE | TP_FLAG_PROFILE));
67}
68
69static __kprobes bool trace_probe_is_registered(struct trace_probe *tp)
70{
71 return !!(tp->flags & TP_FLAG_REGISTERED);
72}
73
74static __kprobes bool trace_probe_has_gone(struct trace_probe *tp)
75{
76 return !!(kprobe_gone(&tp->rp.kp));
77}
78
79static __kprobes bool trace_probe_within_module(struct trace_probe *tp,
80 struct module *mod)
81{
82 int len = strlen(mod->name);
83 const char *name = trace_probe_symbol(tp);
84 return strncmp(mod->name, name, len) == 0 && name[len] == ':';
85}
86
87static __kprobes bool trace_probe_is_on_module(struct trace_probe *tp)
88{
89 return !!strchr(trace_probe_symbol(tp), ':');
90}
91
92static int register_probe_event(struct trace_probe *tp);
93static void unregister_probe_event(struct trace_probe *tp);
94
95static DEFINE_MUTEX(probe_lock);
96static LIST_HEAD(probe_list);
97
98static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
99static int kretprobe_dispatcher(struct kretprobe_instance *ri,
100 struct pt_regs *regs);
101
102/*
103 * Allocate new trace_probe and initialize it (including kprobes).
104 */
105static struct trace_probe *alloc_trace_probe(const char *group,
106 const char *event,
107 void *addr,
108 const char *symbol,
109 unsigned long offs,
110 int nargs, bool is_return)
111{
112 struct trace_probe *tp;
113 int ret = -ENOMEM;
114
115 tp = kzalloc(SIZEOF_TRACE_PROBE(nargs), GFP_KERNEL);
116 if (!tp)
117 return ERR_PTR(ret);
118
119 if (symbol) {
120 tp->symbol = kstrdup(symbol, GFP_KERNEL);
121 if (!tp->symbol)
122 goto error;
123 tp->rp.kp.symbol_name = tp->symbol;
124 tp->rp.kp.offset = offs;
125 } else
126 tp->rp.kp.addr = addr;
127
128 if (is_return)
129 tp->rp.handler = kretprobe_dispatcher;
130 else
131 tp->rp.kp.pre_handler = kprobe_dispatcher;
132
133 if (!event || !is_good_name(event)) {
134 ret = -EINVAL;
135 goto error;
136 }
137
138 tp->call.class = &tp->class;
139 tp->call.name = kstrdup(event, GFP_KERNEL);
140 if (!tp->call.name)
141 goto error;
142
143 if (!group || !is_good_name(group)) {
144 ret = -EINVAL;
145 goto error;
146 }
147
148 tp->class.system = kstrdup(group, GFP_KERNEL);
149 if (!tp->class.system)
150 goto error;
151
152 INIT_LIST_HEAD(&tp->list);
153 return tp;
154error:
155 kfree(tp->call.name);
156 kfree(tp->symbol);
157 kfree(tp);
158 return ERR_PTR(ret);
159}
160
161static void free_trace_probe(struct trace_probe *tp)
162{
163 int i;
164
165 for (i = 0; i < tp->nr_args; i++)
166 traceprobe_free_probe_arg(&tp->args[i]);
167
168 kfree(tp->call.class->system);
169 kfree(tp->call.name);
170 kfree(tp->symbol);
171 kfree(tp);
172}
173
174static struct trace_probe *find_trace_probe(const char *event,
175 const char *group)
176{
177 struct trace_probe *tp;
178
179 list_for_each_entry(tp, &probe_list, list)
180 if (strcmp(tp->call.name, event) == 0 &&
181 strcmp(tp->call.class->system, group) == 0)
182 return tp;
183 return NULL;
184}
185
186/* Enable trace_probe - @flag must be TP_FLAG_TRACE or TP_FLAG_PROFILE */
187static int enable_trace_probe(struct trace_probe *tp, int flag)
188{
189 int ret = 0;
190
191 tp->flags |= flag;
192 if (trace_probe_is_enabled(tp) && trace_probe_is_registered(tp) &&
193 !trace_probe_has_gone(tp)) {
194 if (trace_probe_is_return(tp))
195 ret = enable_kretprobe(&tp->rp);
196 else
197 ret = enable_kprobe(&tp->rp.kp);
198 }
199
200 return ret;
201}
202
203/* Disable trace_probe - @flag must be TP_FLAG_TRACE or TP_FLAG_PROFILE */
204static void disable_trace_probe(struct trace_probe *tp, int flag)
205{
206 tp->flags &= ~flag;
207 if (!trace_probe_is_enabled(tp) && trace_probe_is_registered(tp)) {
208 if (trace_probe_is_return(tp))
209 disable_kretprobe(&tp->rp);
210 else
211 disable_kprobe(&tp->rp.kp);
212 }
213}
214
215/* Internal register function - just handle k*probes and flags */
216static int __register_trace_probe(struct trace_probe *tp)
217{
218 int i, ret;
219
220 if (trace_probe_is_registered(tp))
221 return -EINVAL;
222
223 for (i = 0; i < tp->nr_args; i++)
224 traceprobe_update_arg(&tp->args[i]);
225
226 /* Set/clear disabled flag according to tp->flag */
227 if (trace_probe_is_enabled(tp))
228 tp->rp.kp.flags &= ~KPROBE_FLAG_DISABLED;
229 else
230 tp->rp.kp.flags |= KPROBE_FLAG_DISABLED;
231
232 if (trace_probe_is_return(tp))
233 ret = register_kretprobe(&tp->rp);
234 else
235 ret = register_kprobe(&tp->rp.kp);
236
237 if (ret == 0)
238 tp->flags |= TP_FLAG_REGISTERED;
239 else {
240 pr_warning("Could not insert probe at %s+%lu: %d\n",
241 trace_probe_symbol(tp), trace_probe_offset(tp), ret);
242 if (ret == -ENOENT && trace_probe_is_on_module(tp)) {
243 pr_warning("This probe might be able to register after"
244 "target module is loaded. Continue.\n");
245 ret = 0;
246 } else if (ret == -EILSEQ) {
247 pr_warning("Probing address(0x%p) is not an "
248 "instruction boundary.\n",
249 tp->rp.kp.addr);
250 ret = -EINVAL;
251 }
252 }
253
254 return ret;
255}
256
257/* Internal unregister function - just handle k*probes and flags */
258static void __unregister_trace_probe(struct trace_probe *tp)
259{
260 if (trace_probe_is_registered(tp)) {
261 if (trace_probe_is_return(tp))
262 unregister_kretprobe(&tp->rp);
263 else
264 unregister_kprobe(&tp->rp.kp);
265 tp->flags &= ~TP_FLAG_REGISTERED;
266 /* Cleanup kprobe for reuse */
267 if (tp->rp.kp.symbol_name)
268 tp->rp.kp.addr = NULL;
269 }
270}
271
272/* Unregister a trace_probe and probe_event: call with locking probe_lock */
273static int unregister_trace_probe(struct trace_probe *tp)
274{
275 /* Enabled event can not be unregistered */
276 if (trace_probe_is_enabled(tp))
277 return -EBUSY;
278
279 __unregister_trace_probe(tp);
280 list_del(&tp->list);
281 unregister_probe_event(tp);
282
283 return 0;
284}
285
286/* Register a trace_probe and probe_event */
287static int register_trace_probe(struct trace_probe *tp)
288{
289 struct trace_probe *old_tp;
290 int ret;
291
292 mutex_lock(&probe_lock);
293
294 /* Delete old (same name) event if exist */
295 old_tp = find_trace_probe(tp->call.name, tp->call.class->system);
296 if (old_tp) {
297 ret = unregister_trace_probe(old_tp);
298 if (ret < 0)
299 goto end;
300 free_trace_probe(old_tp);
301 }
302
303 /* Register new event */
304 ret = register_probe_event(tp);
305 if (ret) {
306 pr_warning("Failed to register probe event(%d)\n", ret);
307 goto end;
308 }
309
310 /* Register k*probe */
311 ret = __register_trace_probe(tp);
312 if (ret < 0)
313 unregister_probe_event(tp);
314 else
315 list_add_tail(&tp->list, &probe_list);
316
317end:
318 mutex_unlock(&probe_lock);
319 return ret;
320}
321
322/* Module notifier call back, checking event on the module */
323static int trace_probe_module_callback(struct notifier_block *nb,
324 unsigned long val, void *data)
325{
326 struct module *mod = data;
327 struct trace_probe *tp;
328 int ret;
329
330 if (val != MODULE_STATE_COMING)
331 return NOTIFY_DONE;
332
333 /* Update probes on coming module */
334 mutex_lock(&probe_lock);
335 list_for_each_entry(tp, &probe_list, list) {
336 if (trace_probe_within_module(tp, mod)) {
337 /* Don't need to check busy - this should have gone. */
338 __unregister_trace_probe(tp);
339 ret = __register_trace_probe(tp);
340 if (ret)
341 pr_warning("Failed to re-register probe %s on"
342 "%s: %d\n",
343 tp->call.name, mod->name, ret);
344 }
345 }
346 mutex_unlock(&probe_lock);
347
348 return NOTIFY_DONE;
349}
350
351static struct notifier_block trace_probe_module_nb = {
352 .notifier_call = trace_probe_module_callback,
353 .priority = 1 /* Invoked after kprobe module callback */
354};
355
356static int create_trace_probe(int argc, char **argv)
357{
358 /*
359 * Argument syntax:
360 * - Add kprobe: p[:[GRP/]EVENT] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
361 * - Add kretprobe: r[:[GRP/]EVENT] [MOD:]KSYM[+0] [FETCHARGS]
362 * Fetch args:
363 * $retval : fetch return value
364 * $stack : fetch stack address
365 * $stackN : fetch Nth of stack (N:0-)
366 * @ADDR : fetch memory at ADDR (ADDR should be in kernel)
367 * @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
368 * %REG : fetch register REG
369 * Dereferencing memory fetch:
370 * +|-offs(ARG) : fetch memory at ARG +|- offs address.
371 * Alias name of args:
372 * NAME=FETCHARG : set NAME as alias of FETCHARG.
373 * Type of args:
374 * FETCHARG:TYPE : use TYPE instead of unsigned long.
375 */
376 struct trace_probe *tp;
377 int i, ret = 0;
378 bool is_return = false, is_delete = false;
379 char *symbol = NULL, *event = NULL, *group = NULL;
380 char *arg;
381 unsigned long offset = 0;
382 void *addr = NULL;
383 char buf[MAX_EVENT_NAME_LEN];
384
385 /* argc must be >= 1 */
386 if (argv[0][0] == 'p')
387 is_return = false;
388 else if (argv[0][0] == 'r')
389 is_return = true;
390 else if (argv[0][0] == '-')
391 is_delete = true;
392 else {
393 pr_info("Probe definition must be started with 'p', 'r' or"
394 " '-'.\n");
395 return -EINVAL;
396 }
397
398 if (argv[0][1] == ':') {
399 event = &argv[0][2];
400 if (strchr(event, '/')) {
401 group = event;
402 event = strchr(group, '/') + 1;
403 event[-1] = '\0';
404 if (strlen(group) == 0) {
405 pr_info("Group name is not specified\n");
406 return -EINVAL;
407 }
408 }
409 if (strlen(event) == 0) {
410 pr_info("Event name is not specified\n");
411 return -EINVAL;
412 }
413 }
414 if (!group)
415 group = KPROBE_EVENT_SYSTEM;
416
417 if (is_delete) {
418 if (!event) {
419 pr_info("Delete command needs an event name.\n");
420 return -EINVAL;
421 }
422 mutex_lock(&probe_lock);
423 tp = find_trace_probe(event, group);
424 if (!tp) {
425 mutex_unlock(&probe_lock);
426 pr_info("Event %s/%s doesn't exist.\n", group, event);
427 return -ENOENT;
428 }
429 /* delete an event */
430 ret = unregister_trace_probe(tp);
431 if (ret == 0)
432 free_trace_probe(tp);
433 mutex_unlock(&probe_lock);
434 return ret;
435 }
436
437 if (argc < 2) {
438 pr_info("Probe point is not specified.\n");
439 return -EINVAL;
440 }
441 if (isdigit(argv[1][0])) {
442 if (is_return) {
443 pr_info("Return probe point must be a symbol.\n");
444 return -EINVAL;
445 }
446 /* an address specified */
447 ret = strict_strtoul(&argv[1][0], 0, (unsigned long *)&addr);
448 if (ret) {
449 pr_info("Failed to parse address.\n");
450 return ret;
451 }
452 } else {
453 /* a symbol specified */
454 symbol = argv[1];
455 /* TODO: support .init module functions */
456 ret = traceprobe_split_symbol_offset(symbol, &offset);
457 if (ret) {
458 pr_info("Failed to parse symbol.\n");
459 return ret;
460 }
461 if (offset && is_return) {
462 pr_info("Return probe must be used without offset.\n");
463 return -EINVAL;
464 }
465 }
466 argc -= 2; argv += 2;
467
468 /* setup a probe */
469 if (!event) {
470 /* Make a new event name */
471 if (symbol)
472 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
473 is_return ? 'r' : 'p', symbol, offset);
474 else
475 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
476 is_return ? 'r' : 'p', addr);
477 event = buf;
478 }
479 tp = alloc_trace_probe(group, event, addr, symbol, offset, argc,
480 is_return);
481 if (IS_ERR(tp)) {
482 pr_info("Failed to allocate trace_probe.(%d)\n",
483 (int)PTR_ERR(tp));
484 return PTR_ERR(tp);
485 }
486
487 /* parse arguments */
488 ret = 0;
489 for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
490 /* Increment count for freeing args in error case */
491 tp->nr_args++;
492
493 /* Parse argument name */
494 arg = strchr(argv[i], '=');
495 if (arg) {
496 *arg++ = '\0';
497 tp->args[i].name = kstrdup(argv[i], GFP_KERNEL);
498 } else {
499 arg = argv[i];
500 /* If argument name is omitted, set "argN" */
501 snprintf(buf, MAX_EVENT_NAME_LEN, "arg%d", i + 1);
502 tp->args[i].name = kstrdup(buf, GFP_KERNEL);
503 }
504
505 if (!tp->args[i].name) {
506 pr_info("Failed to allocate argument[%d] name.\n", i);
507 ret = -ENOMEM;
508 goto error;
509 }
510
511 if (!is_good_name(tp->args[i].name)) {
512 pr_info("Invalid argument[%d] name: %s\n",
513 i, tp->args[i].name);
514 ret = -EINVAL;
515 goto error;
516 }
517
518 if (traceprobe_conflict_field_name(tp->args[i].name,
519 tp->args, i)) {
520 pr_info("Argument[%d] name '%s' conflicts with "
521 "another field.\n", i, argv[i]);
522 ret = -EINVAL;
523 goto error;
524 }
525
526 /* Parse fetch argument */
527 ret = traceprobe_parse_probe_arg(arg, &tp->size, &tp->args[i],
528 is_return, true);
529 if (ret) {
530 pr_info("Parse error at argument[%d]. (%d)\n", i, ret);
531 goto error;
532 }
533 }
534
535 ret = register_trace_probe(tp);
536 if (ret)
537 goto error;
538 return 0;
539
540error:
541 free_trace_probe(tp);
542 return ret;
543}
544
545static int release_all_trace_probes(void)
546{
547 struct trace_probe *tp;
548 int ret = 0;
549
550 mutex_lock(&probe_lock);
551 /* Ensure no probe is in use. */
552 list_for_each_entry(tp, &probe_list, list)
553 if (trace_probe_is_enabled(tp)) {
554 ret = -EBUSY;
555 goto end;
556 }
557 /* TODO: Use batch unregistration */
558 while (!list_empty(&probe_list)) {
559 tp = list_entry(probe_list.next, struct trace_probe, list);
560 unregister_trace_probe(tp);
561 free_trace_probe(tp);
562 }
563
564end:
565 mutex_unlock(&probe_lock);
566
567 return ret;
568}
569
570/* Probes listing interfaces */
571static void *probes_seq_start(struct seq_file *m, loff_t *pos)
572{
573 mutex_lock(&probe_lock);
574 return seq_list_start(&probe_list, *pos);
575}
576
577static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
578{
579 return seq_list_next(v, &probe_list, pos);
580}
581
582static void probes_seq_stop(struct seq_file *m, void *v)
583{
584 mutex_unlock(&probe_lock);
585}
586
587static int probes_seq_show(struct seq_file *m, void *v)
588{
589 struct trace_probe *tp = v;
590 int i;
591
592 seq_printf(m, "%c", trace_probe_is_return(tp) ? 'r' : 'p');
593 seq_printf(m, ":%s/%s", tp->call.class->system, tp->call.name);
594
595 if (!tp->symbol)
596 seq_printf(m, " 0x%p", tp->rp.kp.addr);
597 else if (tp->rp.kp.offset)
598 seq_printf(m, " %s+%u", trace_probe_symbol(tp),
599 tp->rp.kp.offset);
600 else
601 seq_printf(m, " %s", trace_probe_symbol(tp));
602
603 for (i = 0; i < tp->nr_args; i++)
604 seq_printf(m, " %s=%s", tp->args[i].name, tp->args[i].comm);
605 seq_printf(m, "\n");
606
607 return 0;
608}
609
610static const struct seq_operations probes_seq_op = {
611 .start = probes_seq_start,
612 .next = probes_seq_next,
613 .stop = probes_seq_stop,
614 .show = probes_seq_show
615};
616
617static int probes_open(struct inode *inode, struct file *file)
618{
619 int ret;
620
621 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
622 ret = release_all_trace_probes();
623 if (ret < 0)
624 return ret;
625 }
626
627 return seq_open(file, &probes_seq_op);
628}
629
630static ssize_t probes_write(struct file *file, const char __user *buffer,
631 size_t count, loff_t *ppos)
632{
633 return traceprobe_probes_write(file, buffer, count, ppos,
634 create_trace_probe);
635}
636
637static const struct file_operations kprobe_events_ops = {
638 .owner = THIS_MODULE,
639 .open = probes_open,
640 .read = seq_read,
641 .llseek = seq_lseek,
642 .release = seq_release,
643 .write = probes_write,
644};
645
646/* Probes profiling interfaces */
647static int probes_profile_seq_show(struct seq_file *m, void *v)
648{
649 struct trace_probe *tp = v;
650
651 seq_printf(m, " %-44s %15lu %15lu\n", tp->call.name, tp->nhit,
652 tp->rp.kp.nmissed);
653
654 return 0;
655}
656
657static const struct seq_operations profile_seq_op = {
658 .start = probes_seq_start,
659 .next = probes_seq_next,
660 .stop = probes_seq_stop,
661 .show = probes_profile_seq_show
662};
663
664static int profile_open(struct inode *inode, struct file *file)
665{
666 return seq_open(file, &profile_seq_op);
667}
668
669static const struct file_operations kprobe_profile_ops = {
670 .owner = THIS_MODULE,
671 .open = profile_open,
672 .read = seq_read,
673 .llseek = seq_lseek,
674 .release = seq_release,
675};
676
677/* Sum up total data length for dynamic arraies (strings) */
678static __kprobes int __get_data_size(struct trace_probe *tp,
679 struct pt_regs *regs)
680{
681 int i, ret = 0;
682 u32 len;
683
684 for (i = 0; i < tp->nr_args; i++)
685 if (unlikely(tp->args[i].fetch_size.fn)) {
686 call_fetch(&tp->args[i].fetch_size, regs, &len);
687 ret += len;
688 }
689
690 return ret;
691}
692
693/* Store the value of each argument */
694static __kprobes void store_trace_args(int ent_size, struct trace_probe *tp,
695 struct pt_regs *regs,
696 u8 *data, int maxlen)
697{
698 int i;
699 u32 end = tp->size;
700 u32 *dl; /* Data (relative) location */
701
702 for (i = 0; i < tp->nr_args; i++) {
703 if (unlikely(tp->args[i].fetch_size.fn)) {
704 /*
705 * First, we set the relative location and
706 * maximum data length to *dl
707 */
708 dl = (u32 *)(data + tp->args[i].offset);
709 *dl = make_data_rloc(maxlen, end - tp->args[i].offset);
710 /* Then try to fetch string or dynamic array data */
711 call_fetch(&tp->args[i].fetch, regs, dl);
712 /* Reduce maximum length */
713 end += get_rloc_len(*dl);
714 maxlen -= get_rloc_len(*dl);
715 /* Trick here, convert data_rloc to data_loc */
716 *dl = convert_rloc_to_loc(*dl,
717 ent_size + tp->args[i].offset);
718 } else
719 /* Just fetching data normally */
720 call_fetch(&tp->args[i].fetch, regs,
721 data + tp->args[i].offset);
722 }
723}
724
725/* Kprobe handler */
726static __kprobes void kprobe_trace_func(struct kprobe *kp, struct pt_regs *regs)
727{
728 struct trace_probe *tp = container_of(kp, struct trace_probe, rp.kp);
729 struct kprobe_trace_entry_head *entry;
730 struct ring_buffer_event *event;
731 struct ring_buffer *buffer;
732 int size, dsize, pc;
733 unsigned long irq_flags;
734 struct ftrace_event_call *call = &tp->call;
735
736 tp->nhit++;
737
738 local_save_flags(irq_flags);
739 pc = preempt_count();
740
741 dsize = __get_data_size(tp, regs);
742 size = sizeof(*entry) + tp->size + dsize;
743
744 event = trace_current_buffer_lock_reserve(&buffer, call->event.type,
745 size, irq_flags, pc);
746 if (!event)
747 return;
748
749 entry = ring_buffer_event_data(event);
750 entry->ip = (unsigned long)kp->addr;
751 store_trace_args(sizeof(*entry), tp, regs, (u8 *)&entry[1], dsize);
752
753 if (!filter_current_check_discard(buffer, call, entry, event))
754 trace_nowake_buffer_unlock_commit_regs(buffer, event,
755 irq_flags, pc, regs);
756}
757
758/* Kretprobe handler */
759static __kprobes void kretprobe_trace_func(struct kretprobe_instance *ri,
760 struct pt_regs *regs)
761{
762 struct trace_probe *tp = container_of(ri->rp, struct trace_probe, rp);
763 struct kretprobe_trace_entry_head *entry;
764 struct ring_buffer_event *event;
765 struct ring_buffer *buffer;
766 int size, pc, dsize;
767 unsigned long irq_flags;
768 struct ftrace_event_call *call = &tp->call;
769
770 local_save_flags(irq_flags);
771 pc = preempt_count();
772
773 dsize = __get_data_size(tp, regs);
774 size = sizeof(*entry) + tp->size + dsize;
775
776 event = trace_current_buffer_lock_reserve(&buffer, call->event.type,
777 size, irq_flags, pc);
778 if (!event)
779 return;
780
781 entry = ring_buffer_event_data(event);
782 entry->func = (unsigned long)tp->rp.kp.addr;
783 entry->ret_ip = (unsigned long)ri->ret_addr;
784 store_trace_args(sizeof(*entry), tp, regs, (u8 *)&entry[1], dsize);
785
786 if (!filter_current_check_discard(buffer, call, entry, event))
787 trace_nowake_buffer_unlock_commit_regs(buffer, event,
788 irq_flags, pc, regs);
789}
790
791/* Event entry printers */
792enum print_line_t
793print_kprobe_event(struct trace_iterator *iter, int flags,
794 struct trace_event *event)
795{
796 struct kprobe_trace_entry_head *field;
797 struct trace_seq *s = &iter->seq;
798 struct trace_probe *tp;
799 u8 *data;
800 int i;
801
802 field = (struct kprobe_trace_entry_head *)iter->ent;
803 tp = container_of(event, struct trace_probe, call.event);
804
805 if (!trace_seq_printf(s, "%s: (", tp->call.name))
806 goto partial;
807
808 if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
809 goto partial;
810
811 if (!trace_seq_puts(s, ")"))
812 goto partial;
813
814 data = (u8 *)&field[1];
815 for (i = 0; i < tp->nr_args; i++)
816 if (!tp->args[i].type->print(s, tp->args[i].name,
817 data + tp->args[i].offset, field))
818 goto partial;
819
820 if (!trace_seq_puts(s, "\n"))
821 goto partial;
822
823 return TRACE_TYPE_HANDLED;
824partial:
825 return TRACE_TYPE_PARTIAL_LINE;
826}
827
828enum print_line_t
829print_kretprobe_event(struct trace_iterator *iter, int flags,
830 struct trace_event *event)
831{
832 struct kretprobe_trace_entry_head *field;
833 struct trace_seq *s = &iter->seq;
834 struct trace_probe *tp;
835 u8 *data;
836 int i;
837
838 field = (struct kretprobe_trace_entry_head *)iter->ent;
839 tp = container_of(event, struct trace_probe, call.event);
840
841 if (!trace_seq_printf(s, "%s: (", tp->call.name))
842 goto partial;
843
844 if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
845 goto partial;
846
847 if (!trace_seq_puts(s, " <- "))
848 goto partial;
849
850 if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
851 goto partial;
852
853 if (!trace_seq_puts(s, ")"))
854 goto partial;
855
856 data = (u8 *)&field[1];
857 for (i = 0; i < tp->nr_args; i++)
858 if (!tp->args[i].type->print(s, tp->args[i].name,
859 data + tp->args[i].offset, field))
860 goto partial;
861
862 if (!trace_seq_puts(s, "\n"))
863 goto partial;
864
865 return TRACE_TYPE_HANDLED;
866partial:
867 return TRACE_TYPE_PARTIAL_LINE;
868}
869
870
871static int kprobe_event_define_fields(struct ftrace_event_call *event_call)
872{
873 int ret, i;
874 struct kprobe_trace_entry_head field;
875 struct trace_probe *tp = (struct trace_probe *)event_call->data;
876
877 DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
878 /* Set argument names as fields */
879 for (i = 0; i < tp->nr_args; i++) {
880 ret = trace_define_field(event_call, tp->args[i].type->fmttype,
881 tp->args[i].name,
882 sizeof(field) + tp->args[i].offset,
883 tp->args[i].type->size,
884 tp->args[i].type->is_signed,
885 FILTER_OTHER);
886 if (ret)
887 return ret;
888 }
889 return 0;
890}
891
892static int kretprobe_event_define_fields(struct ftrace_event_call *event_call)
893{
894 int ret, i;
895 struct kretprobe_trace_entry_head field;
896 struct trace_probe *tp = (struct trace_probe *)event_call->data;
897
898 DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
899 DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
900 /* Set argument names as fields */
901 for (i = 0; i < tp->nr_args; i++) {
902 ret = trace_define_field(event_call, tp->args[i].type->fmttype,
903 tp->args[i].name,
904 sizeof(field) + tp->args[i].offset,
905 tp->args[i].type->size,
906 tp->args[i].type->is_signed,
907 FILTER_OTHER);
908 if (ret)
909 return ret;
910 }
911 return 0;
912}
913
914static int __set_print_fmt(struct trace_probe *tp, char *buf, int len)
915{
916 int i;
917 int pos = 0;
918
919 const char *fmt, *arg;
920
921 if (!trace_probe_is_return(tp)) {
922 fmt = "(%lx)";
923 arg = "REC->" FIELD_STRING_IP;
924 } else {
925 fmt = "(%lx <- %lx)";
926 arg = "REC->" FIELD_STRING_FUNC ", REC->" FIELD_STRING_RETIP;
927 }
928
929 /* When len=0, we just calculate the needed length */
930#define LEN_OR_ZERO (len ? len - pos : 0)
931
932 pos += snprintf(buf + pos, LEN_OR_ZERO, "\"%s", fmt);
933
934 for (i = 0; i < tp->nr_args; i++) {
935 pos += snprintf(buf + pos, LEN_OR_ZERO, " %s=%s",
936 tp->args[i].name, tp->args[i].type->fmt);
937 }
938
939 pos += snprintf(buf + pos, LEN_OR_ZERO, "\", %s", arg);
940
941 for (i = 0; i < tp->nr_args; i++) {
942 if (strcmp(tp->args[i].type->name, "string") == 0)
943 pos += snprintf(buf + pos, LEN_OR_ZERO,
944 ", __get_str(%s)",
945 tp->args[i].name);
946 else
947 pos += snprintf(buf + pos, LEN_OR_ZERO, ", REC->%s",
948 tp->args[i].name);
949 }
950
951#undef LEN_OR_ZERO
952
953 /* return the length of print_fmt */
954 return pos;
955}
956
957static int set_print_fmt(struct trace_probe *tp)
958{
959 int len;
960 char *print_fmt;
961
962 /* First: called with 0 length to calculate the needed length */
963 len = __set_print_fmt(tp, NULL, 0);
964 print_fmt = kmalloc(len + 1, GFP_KERNEL);
965 if (!print_fmt)
966 return -ENOMEM;
967
968 /* Second: actually write the @print_fmt */
969 __set_print_fmt(tp, print_fmt, len + 1);
970 tp->call.print_fmt = print_fmt;
971
972 return 0;
973}
974
975#ifdef CONFIG_PERF_EVENTS
976
977/* Kprobe profile handler */
978static __kprobes void kprobe_perf_func(struct kprobe *kp,
979 struct pt_regs *regs)
980{
981 struct trace_probe *tp = container_of(kp, struct trace_probe, rp.kp);
982 struct ftrace_event_call *call = &tp->call;
983 struct kprobe_trace_entry_head *entry;
984 struct hlist_head *head;
985 int size, __size, dsize;
986 int rctx;
987
988 dsize = __get_data_size(tp, regs);
989 __size = sizeof(*entry) + tp->size + dsize;
990 size = ALIGN(__size + sizeof(u32), sizeof(u64));
991 size -= sizeof(u32);
992 if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE,
993 "profile buffer not large enough"))
994 return;
995
996 entry = perf_trace_buf_prepare(size, call->event.type, regs, &rctx);
997 if (!entry)
998 return;
999
1000 entry->ip = (unsigned long)kp->addr;
1001 memset(&entry[1], 0, dsize);
1002 store_trace_args(sizeof(*entry), tp, regs, (u8 *)&entry[1], dsize);
1003
1004 head = this_cpu_ptr(call->perf_events);
1005 perf_trace_buf_submit(entry, size, rctx, entry->ip, 1, regs, head);
1006}
1007
1008/* Kretprobe profile handler */
1009static __kprobes void kretprobe_perf_func(struct kretprobe_instance *ri,
1010 struct pt_regs *regs)
1011{
1012 struct trace_probe *tp = container_of(ri->rp, struct trace_probe, rp);
1013 struct ftrace_event_call *call = &tp->call;
1014 struct kretprobe_trace_entry_head *entry;
1015 struct hlist_head *head;
1016 int size, __size, dsize;
1017 int rctx;
1018
1019 dsize = __get_data_size(tp, regs);
1020 __size = sizeof(*entry) + tp->size + dsize;
1021 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1022 size -= sizeof(u32);
1023 if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE,
1024 "profile buffer not large enough"))
1025 return;
1026
1027 entry = perf_trace_buf_prepare(size, call->event.type, regs, &rctx);
1028 if (!entry)
1029 return;
1030
1031 entry->func = (unsigned long)tp->rp.kp.addr;
1032 entry->ret_ip = (unsigned long)ri->ret_addr;
1033 store_trace_args(sizeof(*entry), tp, regs, (u8 *)&entry[1], dsize);
1034
1035 head = this_cpu_ptr(call->perf_events);
1036 perf_trace_buf_submit(entry, size, rctx, entry->ret_ip, 1, regs, head);
1037}
1038#endif /* CONFIG_PERF_EVENTS */
1039
1040static __kprobes
1041int kprobe_register(struct ftrace_event_call *event,
1042 enum trace_reg type, void *data)
1043{
1044 struct trace_probe *tp = (struct trace_probe *)event->data;
1045
1046 switch (type) {
1047 case TRACE_REG_REGISTER:
1048 return enable_trace_probe(tp, TP_FLAG_TRACE);
1049 case TRACE_REG_UNREGISTER:
1050 disable_trace_probe(tp, TP_FLAG_TRACE);
1051 return 0;
1052
1053#ifdef CONFIG_PERF_EVENTS
1054 case TRACE_REG_PERF_REGISTER:
1055 return enable_trace_probe(tp, TP_FLAG_PROFILE);
1056 case TRACE_REG_PERF_UNREGISTER:
1057 disable_trace_probe(tp, TP_FLAG_PROFILE);
1058 return 0;
1059 case TRACE_REG_PERF_OPEN:
1060 case TRACE_REG_PERF_CLOSE:
1061 case TRACE_REG_PERF_ADD:
1062 case TRACE_REG_PERF_DEL:
1063 return 0;
1064#endif
1065 }
1066 return 0;
1067}
1068
1069static __kprobes
1070int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1071{
1072 struct trace_probe *tp = container_of(kp, struct trace_probe, rp.kp);
1073
1074 if (tp->flags & TP_FLAG_TRACE)
1075 kprobe_trace_func(kp, regs);
1076#ifdef CONFIG_PERF_EVENTS
1077 if (tp->flags & TP_FLAG_PROFILE)
1078 kprobe_perf_func(kp, regs);
1079#endif
1080 return 0; /* We don't tweek kernel, so just return 0 */
1081}
1082
1083static __kprobes
1084int kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1085{
1086 struct trace_probe *tp = container_of(ri->rp, struct trace_probe, rp);
1087
1088 if (tp->flags & TP_FLAG_TRACE)
1089 kretprobe_trace_func(ri, regs);
1090#ifdef CONFIG_PERF_EVENTS
1091 if (tp->flags & TP_FLAG_PROFILE)
1092 kretprobe_perf_func(ri, regs);
1093#endif
1094 return 0; /* We don't tweek kernel, so just return 0 */
1095}
1096
1097static struct trace_event_functions kretprobe_funcs = {
1098 .trace = print_kretprobe_event
1099};
1100
1101static struct trace_event_functions kprobe_funcs = {
1102 .trace = print_kprobe_event
1103};
1104
1105static int register_probe_event(struct trace_probe *tp)
1106{
1107 struct ftrace_event_call *call = &tp->call;
1108 int ret;
1109
1110 /* Initialize ftrace_event_call */
1111 INIT_LIST_HEAD(&call->class->fields);
1112 if (trace_probe_is_return(tp)) {
1113 call->event.funcs = &kretprobe_funcs;
1114 call->class->define_fields = kretprobe_event_define_fields;
1115 } else {
1116 call->event.funcs = &kprobe_funcs;
1117 call->class->define_fields = kprobe_event_define_fields;
1118 }
1119 if (set_print_fmt(tp) < 0)
1120 return -ENOMEM;
1121 ret = register_ftrace_event(&call->event);
1122 if (!ret) {
1123 kfree(call->print_fmt);
1124 return -ENODEV;
1125 }
1126 call->flags = 0;
1127 call->class->reg = kprobe_register;
1128 call->data = tp;
1129 ret = trace_add_event_call(call);
1130 if (ret) {
1131 pr_info("Failed to register kprobe event: %s\n", call->name);
1132 kfree(call->print_fmt);
1133 unregister_ftrace_event(&call->event);
1134 }
1135 return ret;
1136}
1137
1138static void unregister_probe_event(struct trace_probe *tp)
1139{
1140 /* tp->event is unregistered in trace_remove_event_call() */
1141 trace_remove_event_call(&tp->call);
1142 kfree(tp->call.print_fmt);
1143}
1144
1145/* Make a debugfs interface for controlling probe points */
1146static __init int init_kprobe_trace(void)
1147{
1148 struct dentry *d_tracer;
1149 struct dentry *entry;
1150
1151 if (register_module_notifier(&trace_probe_module_nb))
1152 return -EINVAL;
1153
1154 d_tracer = tracing_init_dentry();
1155 if (!d_tracer)
1156 return 0;
1157
1158 entry = debugfs_create_file("kprobe_events", 0644, d_tracer,
1159 NULL, &kprobe_events_ops);
1160
1161 /* Event list interface */
1162 if (!entry)
1163 pr_warning("Could not create debugfs "
1164 "'kprobe_events' entry\n");
1165
1166 /* Profile interface */
1167 entry = debugfs_create_file("kprobe_profile", 0444, d_tracer,
1168 NULL, &kprobe_profile_ops);
1169
1170 if (!entry)
1171 pr_warning("Could not create debugfs "
1172 "'kprobe_profile' entry\n");
1173 return 0;
1174}
1175fs_initcall(init_kprobe_trace);
1176
1177
1178#ifdef CONFIG_FTRACE_STARTUP_TEST
1179
1180/*
1181 * The "__used" keeps gcc from removing the function symbol
1182 * from the kallsyms table.
1183 */
1184static __used int kprobe_trace_selftest_target(int a1, int a2, int a3,
1185 int a4, int a5, int a6)
1186{
1187 return a1 + a2 + a3 + a4 + a5 + a6;
1188}
1189
1190static __init int kprobe_trace_self_tests_init(void)
1191{
1192 int ret, warn = 0;
1193 int (*target)(int, int, int, int, int, int);
1194 struct trace_probe *tp;
1195
1196 target = kprobe_trace_selftest_target;
1197
1198 pr_info("Testing kprobe tracing: ");
1199
1200 ret = traceprobe_command("p:testprobe kprobe_trace_selftest_target "
1201 "$stack $stack0 +0($stack)",
1202 create_trace_probe);
1203 if (WARN_ON_ONCE(ret)) {
1204 pr_warning("error on probing function entry.\n");
1205 warn++;
1206 } else {
1207 /* Enable trace point */
1208 tp = find_trace_probe("testprobe", KPROBE_EVENT_SYSTEM);
1209 if (WARN_ON_ONCE(tp == NULL)) {
1210 pr_warning("error on getting new probe.\n");
1211 warn++;
1212 } else
1213 enable_trace_probe(tp, TP_FLAG_TRACE);
1214 }
1215
1216 ret = traceprobe_command("r:testprobe2 kprobe_trace_selftest_target "
1217 "$retval", create_trace_probe);
1218 if (WARN_ON_ONCE(ret)) {
1219 pr_warning("error on probing function return.\n");
1220 warn++;
1221 } else {
1222 /* Enable trace point */
1223 tp = find_trace_probe("testprobe2", KPROBE_EVENT_SYSTEM);
1224 if (WARN_ON_ONCE(tp == NULL)) {
1225 pr_warning("error on getting new probe.\n");
1226 warn++;
1227 } else
1228 enable_trace_probe(tp, TP_FLAG_TRACE);
1229 }
1230
1231 if (warn)
1232 goto end;
1233
1234 ret = target(1, 2, 3, 4, 5, 6);
1235
1236 /* Disable trace points before removing it */
1237 tp = find_trace_probe("testprobe", KPROBE_EVENT_SYSTEM);
1238 if (WARN_ON_ONCE(tp == NULL)) {
1239 pr_warning("error on getting test probe.\n");
1240 warn++;
1241 } else
1242 disable_trace_probe(tp, TP_FLAG_TRACE);
1243
1244 tp = find_trace_probe("testprobe2", KPROBE_EVENT_SYSTEM);
1245 if (WARN_ON_ONCE(tp == NULL)) {
1246 pr_warning("error on getting 2nd test probe.\n");
1247 warn++;
1248 } else
1249 disable_trace_probe(tp, TP_FLAG_TRACE);
1250
1251 ret = traceprobe_command("-:testprobe", create_trace_probe);
1252 if (WARN_ON_ONCE(ret)) {
1253 pr_warning("error on deleting a probe.\n");
1254 warn++;
1255 }
1256
1257 ret = traceprobe_command("-:testprobe2", create_trace_probe);
1258 if (WARN_ON_ONCE(ret)) {
1259 pr_warning("error on deleting a probe.\n");
1260 warn++;
1261 }
1262
1263end:
1264 release_all_trace_probes();
1265 if (warn)
1266 pr_cont("NG: Some tests are failed. Please check them.\n");
1267 else
1268 pr_cont("OK\n");
1269 return 0;
1270}
1271
1272late_initcall(kprobe_trace_self_tests_init);
1273
1274#endif
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 */
30struct trace_kprobe {
31 struct list_head list;
32 struct kretprobe rp; /* Use rp.kp for kprobe use */
33 unsigned long __percpu *nhit;
34 const char *symbol; /* symbol name */
35 struct trace_probe tp;
36};
37
38#define SIZEOF_TRACE_KPROBE(n) \
39 (offsetof(struct trace_kprobe, tp.args) + \
40 (sizeof(struct probe_arg) * (n)))
41
42
43static nokprobe_inline bool trace_kprobe_is_return(struct trace_kprobe *tk)
44{
45 return tk->rp.handler != NULL;
46}
47
48static nokprobe_inline const char *trace_kprobe_symbol(struct trace_kprobe *tk)
49{
50 return tk->symbol ? tk->symbol : "unknown";
51}
52
53static nokprobe_inline unsigned long trace_kprobe_offset(struct trace_kprobe *tk)
54{
55 return tk->rp.kp.offset;
56}
57
58static nokprobe_inline bool trace_kprobe_has_gone(struct trace_kprobe *tk)
59{
60 return !!(kprobe_gone(&tk->rp.kp));
61}
62
63static nokprobe_inline bool trace_kprobe_within_module(struct trace_kprobe *tk,
64 struct module *mod)
65{
66 int len = strlen(mod->name);
67 const char *name = trace_kprobe_symbol(tk);
68 return strncmp(mod->name, name, len) == 0 && name[len] == ':';
69}
70
71static nokprobe_inline bool trace_kprobe_is_on_module(struct trace_kprobe *tk)
72{
73 return !!strchr(trace_kprobe_symbol(tk), ':');
74}
75
76static nokprobe_inline unsigned long trace_kprobe_nhit(struct trace_kprobe *tk)
77{
78 unsigned long nhit = 0;
79 int cpu;
80
81 for_each_possible_cpu(cpu)
82 nhit += *per_cpu_ptr(tk->nhit, cpu);
83
84 return nhit;
85}
86
87static int register_kprobe_event(struct trace_kprobe *tk);
88static int unregister_kprobe_event(struct trace_kprobe *tk);
89
90static DEFINE_MUTEX(probe_lock);
91static LIST_HEAD(probe_list);
92
93static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
94static int kretprobe_dispatcher(struct kretprobe_instance *ri,
95 struct pt_regs *regs);
96
97/* Memory fetching by symbol */
98struct symbol_cache {
99 char *symbol;
100 long offset;
101 unsigned long addr;
102};
103
104unsigned long update_symbol_cache(struct symbol_cache *sc)
105{
106 sc->addr = (unsigned long)kallsyms_lookup_name(sc->symbol);
107
108 if (sc->addr)
109 sc->addr += sc->offset;
110
111 return sc->addr;
112}
113
114void free_symbol_cache(struct symbol_cache *sc)
115{
116 kfree(sc->symbol);
117 kfree(sc);
118}
119
120struct symbol_cache *alloc_symbol_cache(const char *sym, long offset)
121{
122 struct symbol_cache *sc;
123
124 if (!sym || strlen(sym) == 0)
125 return NULL;
126
127 sc = kzalloc(sizeof(struct symbol_cache), GFP_KERNEL);
128 if (!sc)
129 return NULL;
130
131 sc->symbol = kstrdup(sym, GFP_KERNEL);
132 if (!sc->symbol) {
133 kfree(sc);
134 return NULL;
135 }
136 sc->offset = offset;
137 update_symbol_cache(sc);
138
139 return sc;
140}
141
142/*
143 * Kprobes-specific fetch functions
144 */
145#define DEFINE_FETCH_stack(type) \
146static void FETCH_FUNC_NAME(stack, type)(struct pt_regs *regs, \
147 void *offset, void *dest) \
148{ \
149 *(type *)dest = (type)regs_get_kernel_stack_nth(regs, \
150 (unsigned int)((unsigned long)offset)); \
151} \
152NOKPROBE_SYMBOL(FETCH_FUNC_NAME(stack, type));
153
154DEFINE_BASIC_FETCH_FUNCS(stack)
155/* No string on the stack entry */
156#define fetch_stack_string NULL
157#define fetch_stack_string_size NULL
158
159#define DEFINE_FETCH_memory(type) \
160static void FETCH_FUNC_NAME(memory, type)(struct pt_regs *regs, \
161 void *addr, void *dest) \
162{ \
163 type retval; \
164 if (probe_kernel_address(addr, retval)) \
165 *(type *)dest = 0; \
166 else \
167 *(type *)dest = retval; \
168} \
169NOKPROBE_SYMBOL(FETCH_FUNC_NAME(memory, type));
170
171DEFINE_BASIC_FETCH_FUNCS(memory)
172/*
173 * Fetch a null-terminated string. Caller MUST set *(u32 *)dest with max
174 * length and relative data location.
175 */
176static void FETCH_FUNC_NAME(memory, string)(struct pt_regs *regs,
177 void *addr, void *dest)
178{
179 int maxlen = get_rloc_len(*(u32 *)dest);
180 u8 *dst = get_rloc_data(dest);
181 long ret;
182
183 if (!maxlen)
184 return;
185
186 /*
187 * Try to get string again, since the string can be changed while
188 * probing.
189 */
190 ret = strncpy_from_unsafe(dst, addr, maxlen);
191
192 if (ret < 0) { /* Failed to fetch string */
193 dst[0] = '\0';
194 *(u32 *)dest = make_data_rloc(0, get_rloc_offs(*(u32 *)dest));
195 } else {
196 *(u32 *)dest = make_data_rloc(ret, get_rloc_offs(*(u32 *)dest));
197 }
198}
199NOKPROBE_SYMBOL(FETCH_FUNC_NAME(memory, string));
200
201/* Return the length of string -- including null terminal byte */
202static void FETCH_FUNC_NAME(memory, string_size)(struct pt_regs *regs,
203 void *addr, void *dest)
204{
205 mm_segment_t old_fs;
206 int ret, len = 0;
207 u8 c;
208
209 old_fs = get_fs();
210 set_fs(KERNEL_DS);
211 pagefault_disable();
212
213 do {
214 ret = __copy_from_user_inatomic(&c, (u8 *)addr + len, 1);
215 len++;
216 } while (c && ret == 0 && len < MAX_STRING_SIZE);
217
218 pagefault_enable();
219 set_fs(old_fs);
220
221 if (ret < 0) /* Failed to check the length */
222 *(u32 *)dest = 0;
223 else
224 *(u32 *)dest = len;
225}
226NOKPROBE_SYMBOL(FETCH_FUNC_NAME(memory, string_size));
227
228#define DEFINE_FETCH_symbol(type) \
229void FETCH_FUNC_NAME(symbol, type)(struct pt_regs *regs, void *data, void *dest)\
230{ \
231 struct symbol_cache *sc = data; \
232 if (sc->addr) \
233 fetch_memory_##type(regs, (void *)sc->addr, dest); \
234 else \
235 *(type *)dest = 0; \
236} \
237NOKPROBE_SYMBOL(FETCH_FUNC_NAME(symbol, type));
238
239DEFINE_BASIC_FETCH_FUNCS(symbol)
240DEFINE_FETCH_symbol(string)
241DEFINE_FETCH_symbol(string_size)
242
243/* kprobes don't support file_offset fetch methods */
244#define fetch_file_offset_u8 NULL
245#define fetch_file_offset_u16 NULL
246#define fetch_file_offset_u32 NULL
247#define fetch_file_offset_u64 NULL
248#define fetch_file_offset_string NULL
249#define fetch_file_offset_string_size NULL
250
251/* Fetch type information table */
252static const struct fetch_type kprobes_fetch_type_table[] = {
253 /* Special types */
254 [FETCH_TYPE_STRING] = __ASSIGN_FETCH_TYPE("string", string, string,
255 sizeof(u32), 1, "__data_loc char[]"),
256 [FETCH_TYPE_STRSIZE] = __ASSIGN_FETCH_TYPE("string_size", u32,
257 string_size, sizeof(u32), 0, "u32"),
258 /* Basic types */
259 ASSIGN_FETCH_TYPE(u8, u8, 0),
260 ASSIGN_FETCH_TYPE(u16, u16, 0),
261 ASSIGN_FETCH_TYPE(u32, u32, 0),
262 ASSIGN_FETCH_TYPE(u64, u64, 0),
263 ASSIGN_FETCH_TYPE(s8, u8, 1),
264 ASSIGN_FETCH_TYPE(s16, u16, 1),
265 ASSIGN_FETCH_TYPE(s32, u32, 1),
266 ASSIGN_FETCH_TYPE(s64, u64, 1),
267 ASSIGN_FETCH_TYPE_ALIAS(x8, u8, u8, 0),
268 ASSIGN_FETCH_TYPE_ALIAS(x16, u16, u16, 0),
269 ASSIGN_FETCH_TYPE_ALIAS(x32, u32, u32, 0),
270 ASSIGN_FETCH_TYPE_ALIAS(x64, u64, u64, 0),
271
272 ASSIGN_FETCH_TYPE_END
273};
274
275/*
276 * Allocate new trace_probe and initialize it (including kprobes).
277 */
278static struct trace_kprobe *alloc_trace_kprobe(const char *group,
279 const char *event,
280 void *addr,
281 const char *symbol,
282 unsigned long offs,
283 int nargs, bool is_return)
284{
285 struct trace_kprobe *tk;
286 int ret = -ENOMEM;
287
288 tk = kzalloc(SIZEOF_TRACE_KPROBE(nargs), GFP_KERNEL);
289 if (!tk)
290 return ERR_PTR(ret);
291
292 tk->nhit = alloc_percpu(unsigned long);
293 if (!tk->nhit)
294 goto error;
295
296 if (symbol) {
297 tk->symbol = kstrdup(symbol, GFP_KERNEL);
298 if (!tk->symbol)
299 goto error;
300 tk->rp.kp.symbol_name = tk->symbol;
301 tk->rp.kp.offset = offs;
302 } else
303 tk->rp.kp.addr = addr;
304
305 if (is_return)
306 tk->rp.handler = kretprobe_dispatcher;
307 else
308 tk->rp.kp.pre_handler = kprobe_dispatcher;
309
310 if (!event || !is_good_name(event)) {
311 ret = -EINVAL;
312 goto error;
313 }
314
315 tk->tp.call.class = &tk->tp.class;
316 tk->tp.call.name = kstrdup(event, GFP_KERNEL);
317 if (!tk->tp.call.name)
318 goto error;
319
320 if (!group || !is_good_name(group)) {
321 ret = -EINVAL;
322 goto error;
323 }
324
325 tk->tp.class.system = kstrdup(group, GFP_KERNEL);
326 if (!tk->tp.class.system)
327 goto error;
328
329 INIT_LIST_HEAD(&tk->list);
330 INIT_LIST_HEAD(&tk->tp.files);
331 return tk;
332error:
333 kfree(tk->tp.call.name);
334 kfree(tk->symbol);
335 free_percpu(tk->nhit);
336 kfree(tk);
337 return ERR_PTR(ret);
338}
339
340static void free_trace_kprobe(struct trace_kprobe *tk)
341{
342 int i;
343
344 for (i = 0; i < tk->tp.nr_args; i++)
345 traceprobe_free_probe_arg(&tk->tp.args[i]);
346
347 kfree(tk->tp.call.class->system);
348 kfree(tk->tp.call.name);
349 kfree(tk->symbol);
350 free_percpu(tk->nhit);
351 kfree(tk);
352}
353
354static struct trace_kprobe *find_trace_kprobe(const char *event,
355 const char *group)
356{
357 struct trace_kprobe *tk;
358
359 list_for_each_entry(tk, &probe_list, list)
360 if (strcmp(trace_event_name(&tk->tp.call), event) == 0 &&
361 strcmp(tk->tp.call.class->system, group) == 0)
362 return tk;
363 return NULL;
364}
365
366/*
367 * Enable trace_probe
368 * if the file is NULL, enable "perf" handler, or enable "trace" handler.
369 */
370static int
371enable_trace_kprobe(struct trace_kprobe *tk, struct trace_event_file *file)
372{
373 int ret = 0;
374
375 if (file) {
376 struct event_file_link *link;
377
378 link = kmalloc(sizeof(*link), GFP_KERNEL);
379 if (!link) {
380 ret = -ENOMEM;
381 goto out;
382 }
383
384 link->file = file;
385 list_add_tail_rcu(&link->list, &tk->tp.files);
386
387 tk->tp.flags |= TP_FLAG_TRACE;
388 } else
389 tk->tp.flags |= TP_FLAG_PROFILE;
390
391 if (trace_probe_is_registered(&tk->tp) && !trace_kprobe_has_gone(tk)) {
392 if (trace_kprobe_is_return(tk))
393 ret = enable_kretprobe(&tk->rp);
394 else
395 ret = enable_kprobe(&tk->rp.kp);
396 }
397 out:
398 return ret;
399}
400
401/*
402 * Disable trace_probe
403 * if the file is NULL, disable "perf" handler, or disable "trace" handler.
404 */
405static int
406disable_trace_kprobe(struct trace_kprobe *tk, struct trace_event_file *file)
407{
408 struct event_file_link *link = NULL;
409 int wait = 0;
410 int ret = 0;
411
412 if (file) {
413 link = find_event_file_link(&tk->tp, file);
414 if (!link) {
415 ret = -EINVAL;
416 goto out;
417 }
418
419 list_del_rcu(&link->list);
420 wait = 1;
421 if (!list_empty(&tk->tp.files))
422 goto out;
423
424 tk->tp.flags &= ~TP_FLAG_TRACE;
425 } else
426 tk->tp.flags &= ~TP_FLAG_PROFILE;
427
428 if (!trace_probe_is_enabled(&tk->tp) && trace_probe_is_registered(&tk->tp)) {
429 if (trace_kprobe_is_return(tk))
430 disable_kretprobe(&tk->rp);
431 else
432 disable_kprobe(&tk->rp.kp);
433 wait = 1;
434 }
435 out:
436 if (wait) {
437 /*
438 * Synchronize with kprobe_trace_func/kretprobe_trace_func
439 * to ensure disabled (all running handlers are finished).
440 * This is not only for kfree(), but also the caller,
441 * trace_remove_event_call() supposes it for releasing
442 * event_call related objects, which will be accessed in
443 * the kprobe_trace_func/kretprobe_trace_func.
444 */
445 synchronize_sched();
446 kfree(link); /* Ignored if link == NULL */
447 }
448
449 return ret;
450}
451
452/* Internal register function - just handle k*probes and flags */
453static int __register_trace_kprobe(struct trace_kprobe *tk)
454{
455 int i, ret;
456
457 if (trace_probe_is_registered(&tk->tp))
458 return -EINVAL;
459
460 for (i = 0; i < tk->tp.nr_args; i++)
461 traceprobe_update_arg(&tk->tp.args[i]);
462
463 /* Set/clear disabled flag according to tp->flag */
464 if (trace_probe_is_enabled(&tk->tp))
465 tk->rp.kp.flags &= ~KPROBE_FLAG_DISABLED;
466 else
467 tk->rp.kp.flags |= KPROBE_FLAG_DISABLED;
468
469 if (trace_kprobe_is_return(tk))
470 ret = register_kretprobe(&tk->rp);
471 else
472 ret = register_kprobe(&tk->rp.kp);
473
474 if (ret == 0)
475 tk->tp.flags |= TP_FLAG_REGISTERED;
476 else {
477 pr_warn("Could not insert probe at %s+%lu: %d\n",
478 trace_kprobe_symbol(tk), trace_kprobe_offset(tk), ret);
479 if (ret == -ENOENT && trace_kprobe_is_on_module(tk)) {
480 pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
481 ret = 0;
482 } else if (ret == -EILSEQ) {
483 pr_warn("Probing address(0x%p) is not an instruction boundary.\n",
484 tk->rp.kp.addr);
485 ret = -EINVAL;
486 }
487 }
488
489 return ret;
490}
491
492/* Internal unregister function - just handle k*probes and flags */
493static void __unregister_trace_kprobe(struct trace_kprobe *tk)
494{
495 if (trace_probe_is_registered(&tk->tp)) {
496 if (trace_kprobe_is_return(tk))
497 unregister_kretprobe(&tk->rp);
498 else
499 unregister_kprobe(&tk->rp.kp);
500 tk->tp.flags &= ~TP_FLAG_REGISTERED;
501 /* Cleanup kprobe for reuse */
502 if (tk->rp.kp.symbol_name)
503 tk->rp.kp.addr = NULL;
504 }
505}
506
507/* Unregister a trace_probe and probe_event: call with locking probe_lock */
508static int unregister_trace_kprobe(struct trace_kprobe *tk)
509{
510 /* Enabled event can not be unregistered */
511 if (trace_probe_is_enabled(&tk->tp))
512 return -EBUSY;
513
514 /* Will fail if probe is being used by ftrace or perf */
515 if (unregister_kprobe_event(tk))
516 return -EBUSY;
517
518 __unregister_trace_kprobe(tk);
519 list_del(&tk->list);
520
521 return 0;
522}
523
524/* Register a trace_probe and probe_event */
525static int register_trace_kprobe(struct trace_kprobe *tk)
526{
527 struct trace_kprobe *old_tk;
528 int ret;
529
530 mutex_lock(&probe_lock);
531
532 /* Delete old (same name) event if exist */
533 old_tk = find_trace_kprobe(trace_event_name(&tk->tp.call),
534 tk->tp.call.class->system);
535 if (old_tk) {
536 ret = unregister_trace_kprobe(old_tk);
537 if (ret < 0)
538 goto end;
539 free_trace_kprobe(old_tk);
540 }
541
542 /* Register new event */
543 ret = register_kprobe_event(tk);
544 if (ret) {
545 pr_warn("Failed to register probe event(%d)\n", ret);
546 goto end;
547 }
548
549 /* Register k*probe */
550 ret = __register_trace_kprobe(tk);
551 if (ret < 0)
552 unregister_kprobe_event(tk);
553 else
554 list_add_tail(&tk->list, &probe_list);
555
556end:
557 mutex_unlock(&probe_lock);
558 return ret;
559}
560
561/* Module notifier call back, checking event on the module */
562static int trace_kprobe_module_callback(struct notifier_block *nb,
563 unsigned long val, void *data)
564{
565 struct module *mod = data;
566 struct trace_kprobe *tk;
567 int ret;
568
569 if (val != MODULE_STATE_COMING)
570 return NOTIFY_DONE;
571
572 /* Update probes on coming module */
573 mutex_lock(&probe_lock);
574 list_for_each_entry(tk, &probe_list, list) {
575 if (trace_kprobe_within_module(tk, mod)) {
576 /* Don't need to check busy - this should have gone. */
577 __unregister_trace_kprobe(tk);
578 ret = __register_trace_kprobe(tk);
579 if (ret)
580 pr_warn("Failed to re-register probe %s on %s: %d\n",
581 trace_event_name(&tk->tp.call),
582 mod->name, ret);
583 }
584 }
585 mutex_unlock(&probe_lock);
586
587 return NOTIFY_DONE;
588}
589
590static struct notifier_block trace_kprobe_module_nb = {
591 .notifier_call = trace_kprobe_module_callback,
592 .priority = 1 /* Invoked after kprobe module callback */
593};
594
595static int create_trace_kprobe(int argc, char **argv)
596{
597 /*
598 * Argument syntax:
599 * - Add kprobe: p[:[GRP/]EVENT] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
600 * - Add kretprobe: r[:[GRP/]EVENT] [MOD:]KSYM[+0] [FETCHARGS]
601 * Fetch args:
602 * $retval : fetch return value
603 * $stack : fetch stack address
604 * $stackN : fetch Nth of stack (N:0-)
605 * $comm : fetch current task comm
606 * @ADDR : fetch memory at ADDR (ADDR should be in kernel)
607 * @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
608 * %REG : fetch register REG
609 * Dereferencing memory fetch:
610 * +|-offs(ARG) : fetch memory at ARG +|- offs address.
611 * Alias name of args:
612 * NAME=FETCHARG : set NAME as alias of FETCHARG.
613 * Type of args:
614 * FETCHARG:TYPE : use TYPE instead of unsigned long.
615 */
616 struct trace_kprobe *tk;
617 int i, ret = 0;
618 bool is_return = false, is_delete = false;
619 char *symbol = NULL, *event = NULL, *group = NULL;
620 char *arg;
621 unsigned long offset = 0;
622 void *addr = NULL;
623 char buf[MAX_EVENT_NAME_LEN];
624
625 /* argc must be >= 1 */
626 if (argv[0][0] == 'p')
627 is_return = false;
628 else if (argv[0][0] == 'r')
629 is_return = true;
630 else if (argv[0][0] == '-')
631 is_delete = true;
632 else {
633 pr_info("Probe definition must be started with 'p', 'r' or"
634 " '-'.\n");
635 return -EINVAL;
636 }
637
638 if (argv[0][1] == ':') {
639 event = &argv[0][2];
640 if (strchr(event, '/')) {
641 group = event;
642 event = strchr(group, '/') + 1;
643 event[-1] = '\0';
644 if (strlen(group) == 0) {
645 pr_info("Group name is not specified\n");
646 return -EINVAL;
647 }
648 }
649 if (strlen(event) == 0) {
650 pr_info("Event name is not specified\n");
651 return -EINVAL;
652 }
653 }
654 if (!group)
655 group = KPROBE_EVENT_SYSTEM;
656
657 if (is_delete) {
658 if (!event) {
659 pr_info("Delete command needs an event name.\n");
660 return -EINVAL;
661 }
662 mutex_lock(&probe_lock);
663 tk = find_trace_kprobe(event, group);
664 if (!tk) {
665 mutex_unlock(&probe_lock);
666 pr_info("Event %s/%s doesn't exist.\n", group, event);
667 return -ENOENT;
668 }
669 /* delete an event */
670 ret = unregister_trace_kprobe(tk);
671 if (ret == 0)
672 free_trace_kprobe(tk);
673 mutex_unlock(&probe_lock);
674 return ret;
675 }
676
677 if (argc < 2) {
678 pr_info("Probe point is not specified.\n");
679 return -EINVAL;
680 }
681 if (isdigit(argv[1][0])) {
682 if (is_return) {
683 pr_info("Return probe point must be a symbol.\n");
684 return -EINVAL;
685 }
686 /* an address specified */
687 ret = kstrtoul(&argv[1][0], 0, (unsigned long *)&addr);
688 if (ret) {
689 pr_info("Failed to parse address.\n");
690 return ret;
691 }
692 } else {
693 /* a symbol specified */
694 symbol = argv[1];
695 /* TODO: support .init module functions */
696 ret = traceprobe_split_symbol_offset(symbol, &offset);
697 if (ret) {
698 pr_info("Failed to parse symbol.\n");
699 return ret;
700 }
701 if (offset && is_return) {
702 pr_info("Return probe must be used without offset.\n");
703 return -EINVAL;
704 }
705 }
706 argc -= 2; argv += 2;
707
708 /* setup a probe */
709 if (!event) {
710 /* Make a new event name */
711 if (symbol)
712 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
713 is_return ? 'r' : 'p', symbol, offset);
714 else
715 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
716 is_return ? 'r' : 'p', addr);
717 event = buf;
718 }
719 tk = alloc_trace_kprobe(group, event, addr, symbol, offset, argc,
720 is_return);
721 if (IS_ERR(tk)) {
722 pr_info("Failed to allocate trace_probe.(%d)\n",
723 (int)PTR_ERR(tk));
724 return PTR_ERR(tk);
725 }
726
727 /* parse arguments */
728 ret = 0;
729 for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
730 struct probe_arg *parg = &tk->tp.args[i];
731
732 /* Increment count for freeing args in error case */
733 tk->tp.nr_args++;
734
735 /* Parse argument name */
736 arg = strchr(argv[i], '=');
737 if (arg) {
738 *arg++ = '\0';
739 parg->name = kstrdup(argv[i], GFP_KERNEL);
740 } else {
741 arg = argv[i];
742 /* If argument name is omitted, set "argN" */
743 snprintf(buf, MAX_EVENT_NAME_LEN, "arg%d", i + 1);
744 parg->name = kstrdup(buf, GFP_KERNEL);
745 }
746
747 if (!parg->name) {
748 pr_info("Failed to allocate argument[%d] name.\n", i);
749 ret = -ENOMEM;
750 goto error;
751 }
752
753 if (!is_good_name(parg->name)) {
754 pr_info("Invalid argument[%d] name: %s\n",
755 i, parg->name);
756 ret = -EINVAL;
757 goto error;
758 }
759
760 if (traceprobe_conflict_field_name(parg->name,
761 tk->tp.args, i)) {
762 pr_info("Argument[%d] name '%s' conflicts with "
763 "another field.\n", i, argv[i]);
764 ret = -EINVAL;
765 goto error;
766 }
767
768 /* Parse fetch argument */
769 ret = traceprobe_parse_probe_arg(arg, &tk->tp.size, parg,
770 is_return, true,
771 kprobes_fetch_type_table);
772 if (ret) {
773 pr_info("Parse error at argument[%d]. (%d)\n", i, ret);
774 goto error;
775 }
776 }
777
778 ret = register_trace_kprobe(tk);
779 if (ret)
780 goto error;
781 return 0;
782
783error:
784 free_trace_kprobe(tk);
785 return ret;
786}
787
788static int release_all_trace_kprobes(void)
789{
790 struct trace_kprobe *tk;
791 int ret = 0;
792
793 mutex_lock(&probe_lock);
794 /* Ensure no probe is in use. */
795 list_for_each_entry(tk, &probe_list, list)
796 if (trace_probe_is_enabled(&tk->tp)) {
797 ret = -EBUSY;
798 goto end;
799 }
800 /* TODO: Use batch unregistration */
801 while (!list_empty(&probe_list)) {
802 tk = list_entry(probe_list.next, struct trace_kprobe, list);
803 ret = unregister_trace_kprobe(tk);
804 if (ret)
805 goto end;
806 free_trace_kprobe(tk);
807 }
808
809end:
810 mutex_unlock(&probe_lock);
811
812 return ret;
813}
814
815/* Probes listing interfaces */
816static void *probes_seq_start(struct seq_file *m, loff_t *pos)
817{
818 mutex_lock(&probe_lock);
819 return seq_list_start(&probe_list, *pos);
820}
821
822static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
823{
824 return seq_list_next(v, &probe_list, pos);
825}
826
827static void probes_seq_stop(struct seq_file *m, void *v)
828{
829 mutex_unlock(&probe_lock);
830}
831
832static int probes_seq_show(struct seq_file *m, void *v)
833{
834 struct trace_kprobe *tk = v;
835 int i;
836
837 seq_putc(m, trace_kprobe_is_return(tk) ? 'r' : 'p');
838 seq_printf(m, ":%s/%s", tk->tp.call.class->system,
839 trace_event_name(&tk->tp.call));
840
841 if (!tk->symbol)
842 seq_printf(m, " 0x%p", tk->rp.kp.addr);
843 else if (tk->rp.kp.offset)
844 seq_printf(m, " %s+%u", trace_kprobe_symbol(tk),
845 tk->rp.kp.offset);
846 else
847 seq_printf(m, " %s", trace_kprobe_symbol(tk));
848
849 for (i = 0; i < tk->tp.nr_args; i++)
850 seq_printf(m, " %s=%s", tk->tp.args[i].name, tk->tp.args[i].comm);
851 seq_putc(m, '\n');
852
853 return 0;
854}
855
856static const struct seq_operations probes_seq_op = {
857 .start = probes_seq_start,
858 .next = probes_seq_next,
859 .stop = probes_seq_stop,
860 .show = probes_seq_show
861};
862
863static int probes_open(struct inode *inode, struct file *file)
864{
865 int ret;
866
867 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
868 ret = release_all_trace_kprobes();
869 if (ret < 0)
870 return ret;
871 }
872
873 return seq_open(file, &probes_seq_op);
874}
875
876static ssize_t probes_write(struct file *file, const char __user *buffer,
877 size_t count, loff_t *ppos)
878{
879 return traceprobe_probes_write(file, buffer, count, ppos,
880 create_trace_kprobe);
881}
882
883static const struct file_operations kprobe_events_ops = {
884 .owner = THIS_MODULE,
885 .open = probes_open,
886 .read = seq_read,
887 .llseek = seq_lseek,
888 .release = seq_release,
889 .write = probes_write,
890};
891
892/* Probes profiling interfaces */
893static int probes_profile_seq_show(struct seq_file *m, void *v)
894{
895 struct trace_kprobe *tk = v;
896
897 seq_printf(m, " %-44s %15lu %15lu\n",
898 trace_event_name(&tk->tp.call),
899 trace_kprobe_nhit(tk),
900 tk->rp.kp.nmissed);
901
902 return 0;
903}
904
905static const struct seq_operations profile_seq_op = {
906 .start = probes_seq_start,
907 .next = probes_seq_next,
908 .stop = probes_seq_stop,
909 .show = probes_profile_seq_show
910};
911
912static int profile_open(struct inode *inode, struct file *file)
913{
914 return seq_open(file, &profile_seq_op);
915}
916
917static const struct file_operations kprobe_profile_ops = {
918 .owner = THIS_MODULE,
919 .open = profile_open,
920 .read = seq_read,
921 .llseek = seq_lseek,
922 .release = seq_release,
923};
924
925/* Kprobe handler */
926static nokprobe_inline void
927__kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs,
928 struct trace_event_file *trace_file)
929{
930 struct kprobe_trace_entry_head *entry;
931 struct ring_buffer_event *event;
932 struct ring_buffer *buffer;
933 int size, dsize, pc;
934 unsigned long irq_flags;
935 struct trace_event_call *call = &tk->tp.call;
936
937 WARN_ON(call != trace_file->event_call);
938
939 if (trace_trigger_soft_disabled(trace_file))
940 return;
941
942 local_save_flags(irq_flags);
943 pc = preempt_count();
944
945 dsize = __get_data_size(&tk->tp, regs);
946 size = sizeof(*entry) + tk->tp.size + dsize;
947
948 event = trace_event_buffer_lock_reserve(&buffer, trace_file,
949 call->event.type,
950 size, irq_flags, pc);
951 if (!event)
952 return;
953
954 entry = ring_buffer_event_data(event);
955 entry->ip = (unsigned long)tk->rp.kp.addr;
956 store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
957
958 event_trigger_unlock_commit_regs(trace_file, buffer, event,
959 entry, irq_flags, pc, regs);
960}
961
962static void
963kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs)
964{
965 struct event_file_link *link;
966
967 list_for_each_entry_rcu(link, &tk->tp.files, list)
968 __kprobe_trace_func(tk, regs, link->file);
969}
970NOKPROBE_SYMBOL(kprobe_trace_func);
971
972/* Kretprobe handler */
973static nokprobe_inline void
974__kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
975 struct pt_regs *regs,
976 struct trace_event_file *trace_file)
977{
978 struct kretprobe_trace_entry_head *entry;
979 struct ring_buffer_event *event;
980 struct ring_buffer *buffer;
981 int size, pc, dsize;
982 unsigned long irq_flags;
983 struct trace_event_call *call = &tk->tp.call;
984
985 WARN_ON(call != trace_file->event_call);
986
987 if (trace_trigger_soft_disabled(trace_file))
988 return;
989
990 local_save_flags(irq_flags);
991 pc = preempt_count();
992
993 dsize = __get_data_size(&tk->tp, regs);
994 size = sizeof(*entry) + tk->tp.size + dsize;
995
996 event = trace_event_buffer_lock_reserve(&buffer, trace_file,
997 call->event.type,
998 size, irq_flags, pc);
999 if (!event)
1000 return;
1001
1002 entry = ring_buffer_event_data(event);
1003 entry->func = (unsigned long)tk->rp.kp.addr;
1004 entry->ret_ip = (unsigned long)ri->ret_addr;
1005 store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1006
1007 event_trigger_unlock_commit_regs(trace_file, buffer, event,
1008 entry, irq_flags, pc, regs);
1009}
1010
1011static void
1012kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1013 struct pt_regs *regs)
1014{
1015 struct event_file_link *link;
1016
1017 list_for_each_entry_rcu(link, &tk->tp.files, list)
1018 __kretprobe_trace_func(tk, ri, regs, link->file);
1019}
1020NOKPROBE_SYMBOL(kretprobe_trace_func);
1021
1022/* Event entry printers */
1023static enum print_line_t
1024print_kprobe_event(struct trace_iterator *iter, int flags,
1025 struct trace_event *event)
1026{
1027 struct kprobe_trace_entry_head *field;
1028 struct trace_seq *s = &iter->seq;
1029 struct trace_probe *tp;
1030 u8 *data;
1031 int i;
1032
1033 field = (struct kprobe_trace_entry_head *)iter->ent;
1034 tp = container_of(event, struct trace_probe, call.event);
1035
1036 trace_seq_printf(s, "%s: (", trace_event_name(&tp->call));
1037
1038 if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
1039 goto out;
1040
1041 trace_seq_putc(s, ')');
1042
1043 data = (u8 *)&field[1];
1044 for (i = 0; i < tp->nr_args; i++)
1045 if (!tp->args[i].type->print(s, tp->args[i].name,
1046 data + tp->args[i].offset, field))
1047 goto out;
1048
1049 trace_seq_putc(s, '\n');
1050 out:
1051 return trace_handle_return(s);
1052}
1053
1054static enum print_line_t
1055print_kretprobe_event(struct trace_iterator *iter, int flags,
1056 struct trace_event *event)
1057{
1058 struct kretprobe_trace_entry_head *field;
1059 struct trace_seq *s = &iter->seq;
1060 struct trace_probe *tp;
1061 u8 *data;
1062 int i;
1063
1064 field = (struct kretprobe_trace_entry_head *)iter->ent;
1065 tp = container_of(event, struct trace_probe, call.event);
1066
1067 trace_seq_printf(s, "%s: (", trace_event_name(&tp->call));
1068
1069 if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1070 goto out;
1071
1072 trace_seq_puts(s, " <- ");
1073
1074 if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1075 goto out;
1076
1077 trace_seq_putc(s, ')');
1078
1079 data = (u8 *)&field[1];
1080 for (i = 0; i < tp->nr_args; i++)
1081 if (!tp->args[i].type->print(s, tp->args[i].name,
1082 data + tp->args[i].offset, field))
1083 goto out;
1084
1085 trace_seq_putc(s, '\n');
1086
1087 out:
1088 return trace_handle_return(s);
1089}
1090
1091
1092static int kprobe_event_define_fields(struct trace_event_call *event_call)
1093{
1094 int ret, i;
1095 struct kprobe_trace_entry_head field;
1096 struct trace_kprobe *tk = (struct trace_kprobe *)event_call->data;
1097
1098 DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1099 /* Set argument names as fields */
1100 for (i = 0; i < tk->tp.nr_args; i++) {
1101 struct probe_arg *parg = &tk->tp.args[i];
1102
1103 ret = trace_define_field(event_call, parg->type->fmttype,
1104 parg->name,
1105 sizeof(field) + parg->offset,
1106 parg->type->size,
1107 parg->type->is_signed,
1108 FILTER_OTHER);
1109 if (ret)
1110 return ret;
1111 }
1112 return 0;
1113}
1114
1115static int kretprobe_event_define_fields(struct trace_event_call *event_call)
1116{
1117 int ret, i;
1118 struct kretprobe_trace_entry_head field;
1119 struct trace_kprobe *tk = (struct trace_kprobe *)event_call->data;
1120
1121 DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1122 DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1123 /* Set argument names as fields */
1124 for (i = 0; i < tk->tp.nr_args; i++) {
1125 struct probe_arg *parg = &tk->tp.args[i];
1126
1127 ret = trace_define_field(event_call, parg->type->fmttype,
1128 parg->name,
1129 sizeof(field) + parg->offset,
1130 parg->type->size,
1131 parg->type->is_signed,
1132 FILTER_OTHER);
1133 if (ret)
1134 return ret;
1135 }
1136 return 0;
1137}
1138
1139#ifdef CONFIG_PERF_EVENTS
1140
1141/* Kprobe profile handler */
1142static void
1143kprobe_perf_func(struct trace_kprobe *tk, struct pt_regs *regs)
1144{
1145 struct trace_event_call *call = &tk->tp.call;
1146 struct bpf_prog *prog = call->prog;
1147 struct kprobe_trace_entry_head *entry;
1148 struct hlist_head *head;
1149 int size, __size, dsize;
1150 int rctx;
1151
1152 if (prog && !trace_call_bpf(prog, regs))
1153 return;
1154
1155 head = this_cpu_ptr(call->perf_events);
1156 if (hlist_empty(head))
1157 return;
1158
1159 dsize = __get_data_size(&tk->tp, regs);
1160 __size = sizeof(*entry) + tk->tp.size + dsize;
1161 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1162 size -= sizeof(u32);
1163
1164 entry = perf_trace_buf_alloc(size, NULL, &rctx);
1165 if (!entry)
1166 return;
1167
1168 entry->ip = (unsigned long)tk->rp.kp.addr;
1169 memset(&entry[1], 0, dsize);
1170 store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1171 perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1172 head, NULL);
1173}
1174NOKPROBE_SYMBOL(kprobe_perf_func);
1175
1176/* Kretprobe profile handler */
1177static void
1178kretprobe_perf_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1179 struct pt_regs *regs)
1180{
1181 struct trace_event_call *call = &tk->tp.call;
1182 struct bpf_prog *prog = call->prog;
1183 struct kretprobe_trace_entry_head *entry;
1184 struct hlist_head *head;
1185 int size, __size, dsize;
1186 int rctx;
1187
1188 if (prog && !trace_call_bpf(prog, regs))
1189 return;
1190
1191 head = this_cpu_ptr(call->perf_events);
1192 if (hlist_empty(head))
1193 return;
1194
1195 dsize = __get_data_size(&tk->tp, regs);
1196 __size = sizeof(*entry) + tk->tp.size + dsize;
1197 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1198 size -= sizeof(u32);
1199
1200 entry = perf_trace_buf_alloc(size, NULL, &rctx);
1201 if (!entry)
1202 return;
1203
1204 entry->func = (unsigned long)tk->rp.kp.addr;
1205 entry->ret_ip = (unsigned long)ri->ret_addr;
1206 store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1207 perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1208 head, NULL);
1209}
1210NOKPROBE_SYMBOL(kretprobe_perf_func);
1211#endif /* CONFIG_PERF_EVENTS */
1212
1213/*
1214 * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex.
1215 *
1216 * kprobe_trace_self_tests_init() does enable_trace_probe/disable_trace_probe
1217 * lockless, but we can't race with this __init function.
1218 */
1219static int kprobe_register(struct trace_event_call *event,
1220 enum trace_reg type, void *data)
1221{
1222 struct trace_kprobe *tk = (struct trace_kprobe *)event->data;
1223 struct trace_event_file *file = data;
1224
1225 switch (type) {
1226 case TRACE_REG_REGISTER:
1227 return enable_trace_kprobe(tk, file);
1228 case TRACE_REG_UNREGISTER:
1229 return disable_trace_kprobe(tk, file);
1230
1231#ifdef CONFIG_PERF_EVENTS
1232 case TRACE_REG_PERF_REGISTER:
1233 return enable_trace_kprobe(tk, NULL);
1234 case TRACE_REG_PERF_UNREGISTER:
1235 return disable_trace_kprobe(tk, NULL);
1236 case TRACE_REG_PERF_OPEN:
1237 case TRACE_REG_PERF_CLOSE:
1238 case TRACE_REG_PERF_ADD:
1239 case TRACE_REG_PERF_DEL:
1240 return 0;
1241#endif
1242 }
1243 return 0;
1244}
1245
1246static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1247{
1248 struct trace_kprobe *tk = container_of(kp, struct trace_kprobe, rp.kp);
1249
1250 raw_cpu_inc(*tk->nhit);
1251
1252 if (tk->tp.flags & TP_FLAG_TRACE)
1253 kprobe_trace_func(tk, regs);
1254#ifdef CONFIG_PERF_EVENTS
1255 if (tk->tp.flags & TP_FLAG_PROFILE)
1256 kprobe_perf_func(tk, regs);
1257#endif
1258 return 0; /* We don't tweek kernel, so just return 0 */
1259}
1260NOKPROBE_SYMBOL(kprobe_dispatcher);
1261
1262static int
1263kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1264{
1265 struct trace_kprobe *tk = container_of(ri->rp, struct trace_kprobe, rp);
1266
1267 raw_cpu_inc(*tk->nhit);
1268
1269 if (tk->tp.flags & TP_FLAG_TRACE)
1270 kretprobe_trace_func(tk, ri, regs);
1271#ifdef CONFIG_PERF_EVENTS
1272 if (tk->tp.flags & TP_FLAG_PROFILE)
1273 kretprobe_perf_func(tk, ri, regs);
1274#endif
1275 return 0; /* We don't tweek kernel, so just return 0 */
1276}
1277NOKPROBE_SYMBOL(kretprobe_dispatcher);
1278
1279static struct trace_event_functions kretprobe_funcs = {
1280 .trace = print_kretprobe_event
1281};
1282
1283static struct trace_event_functions kprobe_funcs = {
1284 .trace = print_kprobe_event
1285};
1286
1287static int register_kprobe_event(struct trace_kprobe *tk)
1288{
1289 struct trace_event_call *call = &tk->tp.call;
1290 int ret;
1291
1292 /* Initialize trace_event_call */
1293 INIT_LIST_HEAD(&call->class->fields);
1294 if (trace_kprobe_is_return(tk)) {
1295 call->event.funcs = &kretprobe_funcs;
1296 call->class->define_fields = kretprobe_event_define_fields;
1297 } else {
1298 call->event.funcs = &kprobe_funcs;
1299 call->class->define_fields = kprobe_event_define_fields;
1300 }
1301 if (set_print_fmt(&tk->tp, trace_kprobe_is_return(tk)) < 0)
1302 return -ENOMEM;
1303 ret = register_trace_event(&call->event);
1304 if (!ret) {
1305 kfree(call->print_fmt);
1306 return -ENODEV;
1307 }
1308 call->flags = TRACE_EVENT_FL_KPROBE;
1309 call->class->reg = kprobe_register;
1310 call->data = tk;
1311 ret = trace_add_event_call(call);
1312 if (ret) {
1313 pr_info("Failed to register kprobe event: %s\n",
1314 trace_event_name(call));
1315 kfree(call->print_fmt);
1316 unregister_trace_event(&call->event);
1317 }
1318 return ret;
1319}
1320
1321static int unregister_kprobe_event(struct trace_kprobe *tk)
1322{
1323 int ret;
1324
1325 /* tp->event is unregistered in trace_remove_event_call() */
1326 ret = trace_remove_event_call(&tk->tp.call);
1327 if (!ret)
1328 kfree(tk->tp.call.print_fmt);
1329 return ret;
1330}
1331
1332/* Make a tracefs interface for controlling probe points */
1333static __init int init_kprobe_trace(void)
1334{
1335 struct dentry *d_tracer;
1336 struct dentry *entry;
1337
1338 if (register_module_notifier(&trace_kprobe_module_nb))
1339 return -EINVAL;
1340
1341 d_tracer = tracing_init_dentry();
1342 if (IS_ERR(d_tracer))
1343 return 0;
1344
1345 entry = tracefs_create_file("kprobe_events", 0644, d_tracer,
1346 NULL, &kprobe_events_ops);
1347
1348 /* Event list interface */
1349 if (!entry)
1350 pr_warn("Could not create tracefs 'kprobe_events' entry\n");
1351
1352 /* Profile interface */
1353 entry = tracefs_create_file("kprobe_profile", 0444, d_tracer,
1354 NULL, &kprobe_profile_ops);
1355
1356 if (!entry)
1357 pr_warn("Could not create tracefs 'kprobe_profile' entry\n");
1358 return 0;
1359}
1360fs_initcall(init_kprobe_trace);
1361
1362
1363#ifdef CONFIG_FTRACE_STARTUP_TEST
1364/*
1365 * The "__used" keeps gcc from removing the function symbol
1366 * from the kallsyms table. 'noinline' makes sure that there
1367 * isn't an inlined version used by the test method below
1368 */
1369static __used __init noinline int
1370kprobe_trace_selftest_target(int a1, int a2, int a3, int a4, int a5, int a6)
1371{
1372 return a1 + a2 + a3 + a4 + a5 + a6;
1373}
1374
1375static __init struct trace_event_file *
1376find_trace_probe_file(struct trace_kprobe *tk, struct trace_array *tr)
1377{
1378 struct trace_event_file *file;
1379
1380 list_for_each_entry(file, &tr->events, list)
1381 if (file->event_call == &tk->tp.call)
1382 return file;
1383
1384 return NULL;
1385}
1386
1387/*
1388 * Nobody but us can call enable_trace_kprobe/disable_trace_kprobe at this
1389 * stage, we can do this lockless.
1390 */
1391static __init int kprobe_trace_self_tests_init(void)
1392{
1393 int ret, warn = 0;
1394 int (*target)(int, int, int, int, int, int);
1395 struct trace_kprobe *tk;
1396 struct trace_event_file *file;
1397
1398 if (tracing_is_disabled())
1399 return -ENODEV;
1400
1401 target = kprobe_trace_selftest_target;
1402
1403 pr_info("Testing kprobe tracing: ");
1404
1405 ret = traceprobe_command("p:testprobe kprobe_trace_selftest_target "
1406 "$stack $stack0 +0($stack)",
1407 create_trace_kprobe);
1408 if (WARN_ON_ONCE(ret)) {
1409 pr_warn("error on probing function entry.\n");
1410 warn++;
1411 } else {
1412 /* Enable trace point */
1413 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1414 if (WARN_ON_ONCE(tk == NULL)) {
1415 pr_warn("error on getting new probe.\n");
1416 warn++;
1417 } else {
1418 file = find_trace_probe_file(tk, top_trace_array());
1419 if (WARN_ON_ONCE(file == NULL)) {
1420 pr_warn("error on getting probe file.\n");
1421 warn++;
1422 } else
1423 enable_trace_kprobe(tk, file);
1424 }
1425 }
1426
1427 ret = traceprobe_command("r:testprobe2 kprobe_trace_selftest_target "
1428 "$retval", create_trace_kprobe);
1429 if (WARN_ON_ONCE(ret)) {
1430 pr_warn("error on probing function return.\n");
1431 warn++;
1432 } else {
1433 /* Enable trace point */
1434 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1435 if (WARN_ON_ONCE(tk == NULL)) {
1436 pr_warn("error on getting 2nd new probe.\n");
1437 warn++;
1438 } else {
1439 file = find_trace_probe_file(tk, top_trace_array());
1440 if (WARN_ON_ONCE(file == NULL)) {
1441 pr_warn("error on getting probe file.\n");
1442 warn++;
1443 } else
1444 enable_trace_kprobe(tk, file);
1445 }
1446 }
1447
1448 if (warn)
1449 goto end;
1450
1451 ret = target(1, 2, 3, 4, 5, 6);
1452
1453 /*
1454 * Not expecting an error here, the check is only to prevent the
1455 * optimizer from removing the call to target() as otherwise there
1456 * are no side-effects and the call is never performed.
1457 */
1458 if (ret != 21)
1459 warn++;
1460
1461 /* Disable trace points before removing it */
1462 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1463 if (WARN_ON_ONCE(tk == NULL)) {
1464 pr_warn("error on getting test probe.\n");
1465 warn++;
1466 } else {
1467 if (trace_kprobe_nhit(tk) != 1) {
1468 pr_warn("incorrect number of testprobe hits\n");
1469 warn++;
1470 }
1471
1472 file = find_trace_probe_file(tk, top_trace_array());
1473 if (WARN_ON_ONCE(file == NULL)) {
1474 pr_warn("error on getting probe file.\n");
1475 warn++;
1476 } else
1477 disable_trace_kprobe(tk, file);
1478 }
1479
1480 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1481 if (WARN_ON_ONCE(tk == NULL)) {
1482 pr_warn("error on getting 2nd test probe.\n");
1483 warn++;
1484 } else {
1485 if (trace_kprobe_nhit(tk) != 1) {
1486 pr_warn("incorrect number of testprobe2 hits\n");
1487 warn++;
1488 }
1489
1490 file = find_trace_probe_file(tk, top_trace_array());
1491 if (WARN_ON_ONCE(file == NULL)) {
1492 pr_warn("error on getting probe file.\n");
1493 warn++;
1494 } else
1495 disable_trace_kprobe(tk, file);
1496 }
1497
1498 ret = traceprobe_command("-:testprobe", create_trace_kprobe);
1499 if (WARN_ON_ONCE(ret)) {
1500 pr_warn("error on deleting a probe.\n");
1501 warn++;
1502 }
1503
1504 ret = traceprobe_command("-:testprobe2", create_trace_kprobe);
1505 if (WARN_ON_ONCE(ret)) {
1506 pr_warn("error on deleting a probe.\n");
1507 warn++;
1508 }
1509
1510end:
1511 release_all_trace_kprobes();
1512 if (warn)
1513 pr_cont("NG: Some tests are failed. Please check them.\n");
1514 else
1515 pr_cont("OK\n");
1516 return 0;
1517}
1518
1519late_initcall(kprobe_trace_self_tests_init);
1520
1521#endif