1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19#define pr_fmt(fmt) "trace_kprobe: " fmt
20
21#include <linux/module.h>
22#include <linux/uaccess.h>
23#include <linux/rculist.h>
24
25#include "trace_probe.h"
26
27#define KPROBE_EVENT_SYSTEM "kprobes"
28#define KRETPROBE_MAXACTIVE_MAX 4096
29
30
31
32
33struct trace_kprobe {
34 struct list_head list;
35 struct kretprobe rp;
36 unsigned long __percpu *nhit;
37 const char *symbol;
38 struct trace_probe tp;
39};
40
41#define SIZEOF_TRACE_KPROBE(n) \
42 (offsetof(struct trace_kprobe, tp.args) + \
43 (sizeof(struct probe_arg) * (n)))
44
45
46static nokprobe_inline bool trace_kprobe_is_return(struct trace_kprobe *tk)
47{
48 return tk->rp.handler != NULL;
49}
50
51static nokprobe_inline const char *trace_kprobe_symbol(struct trace_kprobe *tk)
52{
53 return tk->symbol ? tk->symbol : "unknown";
54}
55
56static nokprobe_inline unsigned long trace_kprobe_offset(struct trace_kprobe *tk)
57{
58 return tk->rp.kp.offset;
59}
60
61static nokprobe_inline bool trace_kprobe_has_gone(struct trace_kprobe *tk)
62{
63 return !!(kprobe_gone(&tk->rp.kp));
64}
65
66static nokprobe_inline bool trace_kprobe_within_module(struct trace_kprobe *tk,
67 struct module *mod)
68{
69 int len = strlen(mod->name);
70 const char *name = trace_kprobe_symbol(tk);
71 return strncmp(mod->name, name, len) == 0 && name[len] == ':';
72}
73
74static nokprobe_inline bool trace_kprobe_is_on_module(struct trace_kprobe *tk)
75{
76 return !!strchr(trace_kprobe_symbol(tk), ':');
77}
78
79static nokprobe_inline unsigned long trace_kprobe_nhit(struct trace_kprobe *tk)
80{
81 unsigned long nhit = 0;
82 int cpu;
83
84 for_each_possible_cpu(cpu)
85 nhit += *per_cpu_ptr(tk->nhit, cpu);
86
87 return nhit;
88}
89
90static int register_kprobe_event(struct trace_kprobe *tk);
91static int unregister_kprobe_event(struct trace_kprobe *tk);
92
93static DEFINE_MUTEX(probe_lock);
94static LIST_HEAD(probe_list);
95
96static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
97static int kretprobe_dispatcher(struct kretprobe_instance *ri,
98 struct pt_regs *regs);
99
100
101struct symbol_cache {
102 char *symbol;
103 long offset;
104 unsigned long addr;
105};
106
107unsigned long update_symbol_cache(struct symbol_cache *sc)
108{
109 sc->addr = (unsigned long)kallsyms_lookup_name(sc->symbol);
110
111 if (sc->addr)
112 sc->addr += sc->offset;
113
114 return sc->addr;
115}
116
117void free_symbol_cache(struct symbol_cache *sc)
118{
119 kfree(sc->symbol);
120 kfree(sc);
121}
122
123struct symbol_cache *alloc_symbol_cache(const char *sym, long offset)
124{
125 struct symbol_cache *sc;
126
127 if (!sym || strlen(sym) == 0)
128 return NULL;
129
130 sc = kzalloc(sizeof(struct symbol_cache), GFP_KERNEL);
131 if (!sc)
132 return NULL;
133
134 sc->symbol = kstrdup(sym, GFP_KERNEL);
135 if (!sc->symbol) {
136 kfree(sc);
137 return NULL;
138 }
139 sc->offset = offset;
140 update_symbol_cache(sc);
141
142 return sc;
143}
144
145
146
147
148#define DEFINE_FETCH_stack(type) \
149static void FETCH_FUNC_NAME(stack, type)(struct pt_regs *regs, \
150 void *offset, void *dest) \
151{ \
152 *(type *)dest = (type)regs_get_kernel_stack_nth(regs, \
153 (unsigned int)((unsigned long)offset)); \
154} \
155NOKPROBE_SYMBOL(FETCH_FUNC_NAME(stack, type));
156
157DEFINE_BASIC_FETCH_FUNCS(stack)
158
159#define fetch_stack_string NULL
160#define fetch_stack_string_size NULL
161
162#define DEFINE_FETCH_memory(type) \
163static void FETCH_FUNC_NAME(memory, type)(struct pt_regs *regs, \
164 void *addr, void *dest) \
165{ \
166 type retval; \
167 if (probe_kernel_address(addr, retval)) \
168 *(type *)dest = 0; \
169 else \
170 *(type *)dest = retval; \
171} \
172NOKPROBE_SYMBOL(FETCH_FUNC_NAME(memory, type));
173
174DEFINE_BASIC_FETCH_FUNCS(memory)
175
176
177
178
179static void FETCH_FUNC_NAME(memory, string)(struct pt_regs *regs,
180 void *addr, void *dest)
181{
182 int maxlen = get_rloc_len(*(u32 *)dest);
183 u8 *dst = get_rloc_data(dest);
184 long ret;
185
186 if (!maxlen)
187 return;
188
189
190
191
192
193 ret = strncpy_from_unsafe(dst, addr, maxlen);
194
195 if (ret < 0) {
196 dst[0] = '\0';
197 *(u32 *)dest = make_data_rloc(0, get_rloc_offs(*(u32 *)dest));
198 } else {
199 *(u32 *)dest = make_data_rloc(ret, get_rloc_offs(*(u32 *)dest));
200 }
201}
202NOKPROBE_SYMBOL(FETCH_FUNC_NAME(memory, string));
203
204
205static void FETCH_FUNC_NAME(memory, string_size)(struct pt_regs *regs,
206 void *addr, void *dest)
207{
208 mm_segment_t old_fs;
209 int ret, len = 0;
210 u8 c;
211
212 old_fs = get_fs();
213 set_fs(KERNEL_DS);
214 pagefault_disable();
215
216 do {
217 ret = __copy_from_user_inatomic(&c, (u8 *)addr + len, 1);
218 len++;
219 } while (c && ret == 0 && len < MAX_STRING_SIZE);
220
221 pagefault_enable();
222 set_fs(old_fs);
223
224 if (ret < 0)
225 *(u32 *)dest = 0;
226 else
227 *(u32 *)dest = len;
228}
229NOKPROBE_SYMBOL(FETCH_FUNC_NAME(memory, string_size));
230
231#define DEFINE_FETCH_symbol(type) \
232void FETCH_FUNC_NAME(symbol, type)(struct pt_regs *regs, void *data, void *dest)\
233{ \
234 struct symbol_cache *sc = data; \
235 if (sc->addr) \
236 fetch_memory_##type(regs, (void *)sc->addr, dest); \
237 else \
238 *(type *)dest = 0; \
239} \
240NOKPROBE_SYMBOL(FETCH_FUNC_NAME(symbol, type));
241
242DEFINE_BASIC_FETCH_FUNCS(symbol)
243DEFINE_FETCH_symbol(string)
244DEFINE_FETCH_symbol(string_size)
245
246
247#define fetch_file_offset_u8 NULL
248#define fetch_file_offset_u16 NULL
249#define fetch_file_offset_u32 NULL
250#define fetch_file_offset_u64 NULL
251#define fetch_file_offset_string NULL
252#define fetch_file_offset_string_size NULL
253
254
255static const struct fetch_type kprobes_fetch_type_table[] = {
256
257 [FETCH_TYPE_STRING] = __ASSIGN_FETCH_TYPE("string", string, string,
258 sizeof(u32), 1, "__data_loc char[]"),
259 [FETCH_TYPE_STRSIZE] = __ASSIGN_FETCH_TYPE("string_size", u32,
260 string_size, sizeof(u32), 0, "u32"),
261
262 ASSIGN_FETCH_TYPE(u8, u8, 0),
263 ASSIGN_FETCH_TYPE(u16, u16, 0),
264 ASSIGN_FETCH_TYPE(u32, u32, 0),
265 ASSIGN_FETCH_TYPE(u64, u64, 0),
266 ASSIGN_FETCH_TYPE(s8, u8, 1),
267 ASSIGN_FETCH_TYPE(s16, u16, 1),
268 ASSIGN_FETCH_TYPE(s32, u32, 1),
269 ASSIGN_FETCH_TYPE(s64, u64, 1),
270 ASSIGN_FETCH_TYPE_ALIAS(x8, u8, u8, 0),
271 ASSIGN_FETCH_TYPE_ALIAS(x16, u16, u16, 0),
272 ASSIGN_FETCH_TYPE_ALIAS(x32, u32, u32, 0),
273 ASSIGN_FETCH_TYPE_ALIAS(x64, u64, u64, 0),
274
275 ASSIGN_FETCH_TYPE_END
276};
277
278
279
280
281static struct trace_kprobe *alloc_trace_kprobe(const char *group,
282 const char *event,
283 void *addr,
284 const char *symbol,
285 unsigned long offs,
286 int maxactive,
287 int nargs, bool is_return)
288{
289 struct trace_kprobe *tk;
290 int ret = -ENOMEM;
291
292 tk = kzalloc(SIZEOF_TRACE_KPROBE(nargs), GFP_KERNEL);
293 if (!tk)
294 return ERR_PTR(ret);
295
296 tk->nhit = alloc_percpu(unsigned long);
297 if (!tk->nhit)
298 goto error;
299
300 if (symbol) {
301 tk->symbol = kstrdup(symbol, GFP_KERNEL);
302 if (!tk->symbol)
303 goto error;
304 tk->rp.kp.symbol_name = tk->symbol;
305 tk->rp.kp.offset = offs;
306 } else
307 tk->rp.kp.addr = addr;
308
309 if (is_return)
310 tk->rp.handler = kretprobe_dispatcher;
311 else
312 tk->rp.kp.pre_handler = kprobe_dispatcher;
313
314 tk->rp.maxactive = maxactive;
315
316 if (!event || !is_good_name(event)) {
317 ret = -EINVAL;
318 goto error;
319 }
320
321 tk->tp.call.class = &tk->tp.class;
322 tk->tp.call.name = kstrdup(event, GFP_KERNEL);
323 if (!tk->tp.call.name)
324 goto error;
325
326 if (!group || !is_good_name(group)) {
327 ret = -EINVAL;
328 goto error;
329 }
330
331 tk->tp.class.system = kstrdup(group, GFP_KERNEL);
332 if (!tk->tp.class.system)
333 goto error;
334
335 INIT_LIST_HEAD(&tk->list);
336 INIT_LIST_HEAD(&tk->tp.files);
337 return tk;
338error:
339 kfree(tk->tp.call.name);
340 kfree(tk->symbol);
341 free_percpu(tk->nhit);
342 kfree(tk);
343 return ERR_PTR(ret);
344}
345
346static void free_trace_kprobe(struct trace_kprobe *tk)
347{
348 int i;
349
350 for (i = 0; i < tk->tp.nr_args; i++)
351 traceprobe_free_probe_arg(&tk->tp.args[i]);
352
353 kfree(tk->tp.call.class->system);
354 kfree(tk->tp.call.name);
355 kfree(tk->symbol);
356 free_percpu(tk->nhit);
357 kfree(tk);
358}
359
360static struct trace_kprobe *find_trace_kprobe(const char *event,
361 const char *group)
362{
363 struct trace_kprobe *tk;
364
365 list_for_each_entry(tk, &probe_list, list)
366 if (strcmp(trace_event_name(&tk->tp.call), event) == 0 &&
367 strcmp(tk->tp.call.class->system, group) == 0)
368 return tk;
369 return NULL;
370}
371
372
373
374
375
376static int
377enable_trace_kprobe(struct trace_kprobe *tk, struct trace_event_file *file)
378{
379 int ret = 0;
380
381 if (file) {
382 struct event_file_link *link;
383
384 link = kmalloc(sizeof(*link), GFP_KERNEL);
385 if (!link) {
386 ret = -ENOMEM;
387 goto out;
388 }
389
390 link->file = file;
391 list_add_tail_rcu(&link->list, &tk->tp.files);
392
393 tk->tp.flags |= TP_FLAG_TRACE;
394 } else
395 tk->tp.flags |= TP_FLAG_PROFILE;
396
397 if (trace_probe_is_registered(&tk->tp) && !trace_kprobe_has_gone(tk)) {
398 if (trace_kprobe_is_return(tk))
399 ret = enable_kretprobe(&tk->rp);
400 else
401 ret = enable_kprobe(&tk->rp.kp);
402 }
403 out:
404 return ret;
405}
406
407
408
409
410
411static int
412disable_trace_kprobe(struct trace_kprobe *tk, struct trace_event_file *file)
413{
414 struct event_file_link *link = NULL;
415 int wait = 0;
416 int ret = 0;
417
418 if (file) {
419 link = find_event_file_link(&tk->tp, file);
420 if (!link) {
421 ret = -EINVAL;
422 goto out;
423 }
424
425 list_del_rcu(&link->list);
426 wait = 1;
427 if (!list_empty(&tk->tp.files))
428 goto out;
429
430 tk->tp.flags &= ~TP_FLAG_TRACE;
431 } else
432 tk->tp.flags &= ~TP_FLAG_PROFILE;
433
434 if (!trace_probe_is_enabled(&tk->tp) && trace_probe_is_registered(&tk->tp)) {
435 if (trace_kprobe_is_return(tk))
436 disable_kretprobe(&tk->rp);
437 else
438 disable_kprobe(&tk->rp.kp);
439 wait = 1;
440 }
441 out:
442 if (wait) {
443
444
445
446
447
448
449
450
451 synchronize_sched();
452 kfree(link);
453 }
454
455 return ret;
456}
457
458
459static int __register_trace_kprobe(struct trace_kprobe *tk)
460{
461 int i, ret;
462
463 if (trace_probe_is_registered(&tk->tp))
464 return -EINVAL;
465
466 for (i = 0; i < tk->tp.nr_args; i++)
467 traceprobe_update_arg(&tk->tp.args[i]);
468
469
470 if (trace_probe_is_enabled(&tk->tp))
471 tk->rp.kp.flags &= ~KPROBE_FLAG_DISABLED;
472 else
473 tk->rp.kp.flags |= KPROBE_FLAG_DISABLED;
474
475 if (trace_kprobe_is_return(tk))
476 ret = register_kretprobe(&tk->rp);
477 else
478 ret = register_kprobe(&tk->rp.kp);
479
480 if (ret == 0)
481 tk->tp.flags |= TP_FLAG_REGISTERED;
482 else {
483 pr_warn("Could not insert probe at %s+%lu: %d\n",
484 trace_kprobe_symbol(tk), trace_kprobe_offset(tk), ret);
485 if (ret == -ENOENT && trace_kprobe_is_on_module(tk)) {
486 pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
487 ret = 0;
488 } else if (ret == -EILSEQ) {
489 pr_warn("Probing address(0x%p) is not an instruction boundary.\n",
490 tk->rp.kp.addr);
491 ret = -EINVAL;
492 }
493 }
494
495 return ret;
496}
497
498
499static void __unregister_trace_kprobe(struct trace_kprobe *tk)
500{
501 if (trace_probe_is_registered(&tk->tp)) {
502 if (trace_kprobe_is_return(tk))
503 unregister_kretprobe(&tk->rp);
504 else
505 unregister_kprobe(&tk->rp.kp);
506 tk->tp.flags &= ~TP_FLAG_REGISTERED;
507
508 if (tk->rp.kp.symbol_name)
509 tk->rp.kp.addr = NULL;
510 }
511}
512
513
514static int unregister_trace_kprobe(struct trace_kprobe *tk)
515{
516
517 if (trace_probe_is_enabled(&tk->tp))
518 return -EBUSY;
519
520
521 if (unregister_kprobe_event(tk))
522 return -EBUSY;
523
524 __unregister_trace_kprobe(tk);
525 list_del(&tk->list);
526
527 return 0;
528}
529
530
531static int register_trace_kprobe(struct trace_kprobe *tk)
532{
533 struct trace_kprobe *old_tk;
534 int ret;
535
536 mutex_lock(&probe_lock);
537
538
539 old_tk = find_trace_kprobe(trace_event_name(&tk->tp.call),
540 tk->tp.call.class->system);
541 if (old_tk) {
542 ret = unregister_trace_kprobe(old_tk);
543 if (ret < 0)
544 goto end;
545 free_trace_kprobe(old_tk);
546 }
547
548
549 ret = register_kprobe_event(tk);
550 if (ret) {
551 pr_warn("Failed to register probe event(%d)\n", ret);
552 goto end;
553 }
554
555
556 ret = __register_trace_kprobe(tk);
557 if (ret < 0)
558 unregister_kprobe_event(tk);
559 else
560 list_add_tail(&tk->list, &probe_list);
561
562end:
563 mutex_unlock(&probe_lock);
564 return ret;
565}
566
567
568static int trace_kprobe_module_callback(struct notifier_block *nb,
569 unsigned long val, void *data)
570{
571 struct module *mod = data;
572 struct trace_kprobe *tk;
573 int ret;
574
575 if (val != MODULE_STATE_COMING)
576 return NOTIFY_DONE;
577
578
579 mutex_lock(&probe_lock);
580 list_for_each_entry(tk, &probe_list, list) {
581 if (trace_kprobe_within_module(tk, mod)) {
582
583 __unregister_trace_kprobe(tk);
584 ret = __register_trace_kprobe(tk);
585 if (ret)
586 pr_warn("Failed to re-register probe %s on %s: %d\n",
587 trace_event_name(&tk->tp.call),
588 mod->name, ret);
589 }
590 }
591 mutex_unlock(&probe_lock);
592
593 return NOTIFY_DONE;
594}
595
596static struct notifier_block trace_kprobe_module_nb = {
597 .notifier_call = trace_kprobe_module_callback,
598 .priority = 1
599};
600
601
602static inline void sanitize_event_name(char *name)
603{
604 while (*name++ != '\0')
605 if (*name == ':' || *name == '.')
606 *name = '_';
607}
608
609static int create_trace_kprobe(int argc, char **argv)
610{
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632 struct trace_kprobe *tk;
633 int i, ret = 0;
634 bool is_return = false, is_delete = false;
635 char *symbol = NULL, *event = NULL, *group = NULL;
636 int maxactive = 0;
637 char *arg;
638 unsigned long offset = 0;
639 void *addr = NULL;
640 char buf[MAX_EVENT_NAME_LEN];
641
642
643 if (argv[0][0] == 'p')
644 is_return = false;
645 else if (argv[0][0] == 'r')
646 is_return = true;
647 else if (argv[0][0] == '-')
648 is_delete = true;
649 else {
650 pr_info("Probe definition must be started with 'p', 'r' or"
651 " '-'.\n");
652 return -EINVAL;
653 }
654
655 event = strchr(&argv[0][1], ':');
656 if (event) {
657 event[0] = '\0';
658 event++;
659 }
660 if (is_return && isdigit(argv[0][1])) {
661 ret = kstrtouint(&argv[0][1], 0, &maxactive);
662 if (ret) {
663 pr_info("Failed to parse maxactive.\n");
664 return ret;
665 }
666
667
668
669 if (maxactive > KRETPROBE_MAXACTIVE_MAX) {
670 pr_info("Maxactive is too big (%d > %d).\n",
671 maxactive, KRETPROBE_MAXACTIVE_MAX);
672 return -E2BIG;
673 }
674 }
675
676 if (event) {
677 if (strchr(event, '/')) {
678 group = event;
679 event = strchr(group, '/') + 1;
680 event[-1] = '\0';
681 if (strlen(group) == 0) {
682 pr_info("Group name is not specified\n");
683 return -EINVAL;
684 }
685 }
686 if (strlen(event) == 0) {
687 pr_info("Event name is not specified\n");
688 return -EINVAL;
689 }
690 }
691 if (!group)
692 group = KPROBE_EVENT_SYSTEM;
693
694 if (is_delete) {
695 if (!event) {
696 pr_info("Delete command needs an event name.\n");
697 return -EINVAL;
698 }
699 mutex_lock(&probe_lock);
700 tk = find_trace_kprobe(event, group);
701 if (!tk) {
702 mutex_unlock(&probe_lock);
703 pr_info("Event %s/%s doesn't exist.\n", group, event);
704 return -ENOENT;
705 }
706
707 ret = unregister_trace_kprobe(tk);
708 if (ret == 0)
709 free_trace_kprobe(tk);
710 mutex_unlock(&probe_lock);
711 return ret;
712 }
713
714 if (argc < 2) {
715 pr_info("Probe point is not specified.\n");
716 return -EINVAL;
717 }
718
719
720
721 if (kstrtoul(argv[1], 0, (unsigned long *)&addr)) {
722
723 symbol = argv[1];
724
725 ret = traceprobe_split_symbol_offset(symbol, &offset);
726 if (ret) {
727 pr_info("Failed to parse either an address or a symbol.\n");
728 return ret;
729 }
730 if (offset && is_return &&
731 !kprobe_on_func_entry(NULL, symbol, offset)) {
732 pr_info("Given offset is not valid for return probe.\n");
733 return -EINVAL;
734 }
735 }
736 argc -= 2; argv += 2;
737
738
739 if (!event) {
740
741 if (symbol)
742 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
743 is_return ? 'r' : 'p', symbol, offset);
744 else
745 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
746 is_return ? 'r' : 'p', addr);
747 sanitize_event_name(buf);
748 event = buf;
749 }
750 tk = alloc_trace_kprobe(group, event, addr, symbol, offset, maxactive,
751 argc, is_return);
752 if (IS_ERR(tk)) {
753 pr_info("Failed to allocate trace_probe.(%d)\n",
754 (int)PTR_ERR(tk));
755 return PTR_ERR(tk);
756 }
757
758
759 ret = 0;
760 for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
761 struct probe_arg *parg = &tk->tp.args[i];
762
763
764 tk->tp.nr_args++;
765
766
767 arg = strchr(argv[i], '=');
768 if (arg) {
769 *arg++ = '\0';
770 parg->name = kstrdup(argv[i], GFP_KERNEL);
771 } else {
772 arg = argv[i];
773
774 snprintf(buf, MAX_EVENT_NAME_LEN, "arg%d", i + 1);
775 parg->name = kstrdup(buf, GFP_KERNEL);
776 }
777
778 if (!parg->name) {
779 pr_info("Failed to allocate argument[%d] name.\n", i);
780 ret = -ENOMEM;
781 goto error;
782 }
783
784 if (!is_good_name(parg->name)) {
785 pr_info("Invalid argument[%d] name: %s\n",
786 i, parg->name);
787 ret = -EINVAL;
788 goto error;
789 }
790
791 if (traceprobe_conflict_field_name(parg->name,
792 tk->tp.args, i)) {
793 pr_info("Argument[%d] name '%s' conflicts with "
794 "another field.\n", i, argv[i]);
795 ret = -EINVAL;
796 goto error;
797 }
798
799
800 ret = traceprobe_parse_probe_arg(arg, &tk->tp.size, parg,
801 is_return, true,
802 kprobes_fetch_type_table);
803 if (ret) {
804 pr_info("Parse error at argument[%d]. (%d)\n", i, ret);
805 goto error;
806 }
807 }
808
809 ret = register_trace_kprobe(tk);
810 if (ret)
811 goto error;
812 return 0;
813
814error:
815 free_trace_kprobe(tk);
816 return ret;
817}
818
819static int release_all_trace_kprobes(void)
820{
821 struct trace_kprobe *tk;
822 int ret = 0;
823
824 mutex_lock(&probe_lock);
825
826 list_for_each_entry(tk, &probe_list, list)
827 if (trace_probe_is_enabled(&tk->tp)) {
828 ret = -EBUSY;
829 goto end;
830 }
831
832 while (!list_empty(&probe_list)) {
833 tk = list_entry(probe_list.next, struct trace_kprobe, list);
834 ret = unregister_trace_kprobe(tk);
835 if (ret)
836 goto end;
837 free_trace_kprobe(tk);
838 }
839
840end:
841 mutex_unlock(&probe_lock);
842
843 return ret;
844}
845
846
847static void *probes_seq_start(struct seq_file *m, loff_t *pos)
848{
849 mutex_lock(&probe_lock);
850 return seq_list_start(&probe_list, *pos);
851}
852
853static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
854{
855 return seq_list_next(v, &probe_list, pos);
856}
857
858static void probes_seq_stop(struct seq_file *m, void *v)
859{
860 mutex_unlock(&probe_lock);
861}
862
863static int probes_seq_show(struct seq_file *m, void *v)
864{
865 struct trace_kprobe *tk = v;
866 int i;
867
868 seq_putc(m, trace_kprobe_is_return(tk) ? 'r' : 'p');
869 seq_printf(m, ":%s/%s", tk->tp.call.class->system,
870 trace_event_name(&tk->tp.call));
871
872 if (!tk->symbol)
873 seq_printf(m, " 0x%p", tk->rp.kp.addr);
874 else if (tk->rp.kp.offset)
875 seq_printf(m, " %s+%u", trace_kprobe_symbol(tk),
876 tk->rp.kp.offset);
877 else
878 seq_printf(m, " %s", trace_kprobe_symbol(tk));
879
880 for (i = 0; i < tk->tp.nr_args; i++)
881 seq_printf(m, " %s=%s", tk->tp.args[i].name, tk->tp.args[i].comm);
882 seq_putc(m, '\n');
883
884 return 0;
885}
886
887static const struct seq_operations probes_seq_op = {
888 .start = probes_seq_start,
889 .next = probes_seq_next,
890 .stop = probes_seq_stop,
891 .show = probes_seq_show
892};
893
894static int probes_open(struct inode *inode, struct file *file)
895{
896 int ret;
897
898 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
899 ret = release_all_trace_kprobes();
900 if (ret < 0)
901 return ret;
902 }
903
904 return seq_open(file, &probes_seq_op);
905}
906
907static ssize_t probes_write(struct file *file, const char __user *buffer,
908 size_t count, loff_t *ppos)
909{
910 return traceprobe_probes_write(file, buffer, count, ppos,
911 create_trace_kprobe);
912}
913
914static const struct file_operations kprobe_events_ops = {
915 .owner = THIS_MODULE,
916 .open = probes_open,
917 .read = seq_read,
918 .llseek = seq_lseek,
919 .release = seq_release,
920 .write = probes_write,
921};
922
923
924static int probes_profile_seq_show(struct seq_file *m, void *v)
925{
926 struct trace_kprobe *tk = v;
927
928 seq_printf(m, " %-44s %15lu %15lu\n",
929 trace_event_name(&tk->tp.call),
930 trace_kprobe_nhit(tk),
931 tk->rp.kp.nmissed);
932
933 return 0;
934}
935
936static const struct seq_operations profile_seq_op = {
937 .start = probes_seq_start,
938 .next = probes_seq_next,
939 .stop = probes_seq_stop,
940 .show = probes_profile_seq_show
941};
942
943static int profile_open(struct inode *inode, struct file *file)
944{
945 return seq_open(file, &profile_seq_op);
946}
947
948static const struct file_operations kprobe_profile_ops = {
949 .owner = THIS_MODULE,
950 .open = profile_open,
951 .read = seq_read,
952 .llseek = seq_lseek,
953 .release = seq_release,
954};
955
956
957static nokprobe_inline void
958__kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs,
959 struct trace_event_file *trace_file)
960{
961 struct kprobe_trace_entry_head *entry;
962 struct ring_buffer_event *event;
963 struct ring_buffer *buffer;
964 int size, dsize, pc;
965 unsigned long irq_flags;
966 struct trace_event_call *call = &tk->tp.call;
967
968 WARN_ON(call != trace_file->event_call);
969
970 if (trace_trigger_soft_disabled(trace_file))
971 return;
972
973 local_save_flags(irq_flags);
974 pc = preempt_count();
975
976 dsize = __get_data_size(&tk->tp, regs);
977 size = sizeof(*entry) + tk->tp.size + dsize;
978
979 event = trace_event_buffer_lock_reserve(&buffer, trace_file,
980 call->event.type,
981 size, irq_flags, pc);
982 if (!event)
983 return;
984
985 entry = ring_buffer_event_data(event);
986 entry->ip = (unsigned long)tk->rp.kp.addr;
987 store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
988
989 event_trigger_unlock_commit_regs(trace_file, buffer, event,
990 entry, irq_flags, pc, regs);
991}
992
993static void
994kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs)
995{
996 struct event_file_link *link;
997
998 list_for_each_entry_rcu(link, &tk->tp.files, list)
999 __kprobe_trace_func(tk, regs, link->file);
1000}
1001NOKPROBE_SYMBOL(kprobe_trace_func);
1002
1003
1004static nokprobe_inline void
1005__kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1006 struct pt_regs *regs,
1007 struct trace_event_file *trace_file)
1008{
1009 struct kretprobe_trace_entry_head *entry;
1010 struct ring_buffer_event *event;
1011 struct ring_buffer *buffer;
1012 int size, pc, dsize;
1013 unsigned long irq_flags;
1014 struct trace_event_call *call = &tk->tp.call;
1015
1016 WARN_ON(call != trace_file->event_call);
1017
1018 if (trace_trigger_soft_disabled(trace_file))
1019 return;
1020
1021 local_save_flags(irq_flags);
1022 pc = preempt_count();
1023
1024 dsize = __get_data_size(&tk->tp, regs);
1025 size = sizeof(*entry) + tk->tp.size + dsize;
1026
1027 event = trace_event_buffer_lock_reserve(&buffer, trace_file,
1028 call->event.type,
1029 size, irq_flags, pc);
1030 if (!event)
1031 return;
1032
1033 entry = ring_buffer_event_data(event);
1034 entry->func = (unsigned long)tk->rp.kp.addr;
1035 entry->ret_ip = (unsigned long)ri->ret_addr;
1036 store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1037
1038 event_trigger_unlock_commit_regs(trace_file, buffer, event,
1039 entry, irq_flags, pc, regs);
1040}
1041
1042static void
1043kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1044 struct pt_regs *regs)
1045{
1046 struct event_file_link *link;
1047
1048 list_for_each_entry_rcu(link, &tk->tp.files, list)
1049 __kretprobe_trace_func(tk, ri, regs, link->file);
1050}
1051NOKPROBE_SYMBOL(kretprobe_trace_func);
1052
1053
1054static enum print_line_t
1055print_kprobe_event(struct trace_iterator *iter, int flags,
1056 struct trace_event *event)
1057{
1058 struct kprobe_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 kprobe_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->ip, flags | TRACE_ITER_SYM_OFFSET))
1070 goto out;
1071
1072 trace_seq_putc(s, ')');
1073
1074 data = (u8 *)&field[1];
1075 for (i = 0; i < tp->nr_args; i++)
1076 if (!tp->args[i].type->print(s, tp->args[i].name,
1077 data + tp->args[i].offset, field))
1078 goto out;
1079
1080 trace_seq_putc(s, '\n');
1081 out:
1082 return trace_handle_return(s);
1083}
1084
1085static enum print_line_t
1086print_kretprobe_event(struct trace_iterator *iter, int flags,
1087 struct trace_event *event)
1088{
1089 struct kretprobe_trace_entry_head *field;
1090 struct trace_seq *s = &iter->seq;
1091 struct trace_probe *tp;
1092 u8 *data;
1093 int i;
1094
1095 field = (struct kretprobe_trace_entry_head *)iter->ent;
1096 tp = container_of(event, struct trace_probe, call.event);
1097
1098 trace_seq_printf(s, "%s: (", trace_event_name(&tp->call));
1099
1100 if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1101 goto out;
1102
1103 trace_seq_puts(s, " <- ");
1104
1105 if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1106 goto out;
1107
1108 trace_seq_putc(s, ')');
1109
1110 data = (u8 *)&field[1];
1111 for (i = 0; i < tp->nr_args; i++)
1112 if (!tp->args[i].type->print(s, tp->args[i].name,
1113 data + tp->args[i].offset, field))
1114 goto out;
1115
1116 trace_seq_putc(s, '\n');
1117
1118 out:
1119 return trace_handle_return(s);
1120}
1121
1122
1123static int kprobe_event_define_fields(struct trace_event_call *event_call)
1124{
1125 int ret, i;
1126 struct kprobe_trace_entry_head field;
1127 struct trace_kprobe *tk = (struct trace_kprobe *)event_call->data;
1128
1129 DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1130
1131 for (i = 0; i < tk->tp.nr_args; i++) {
1132 struct probe_arg *parg = &tk->tp.args[i];
1133
1134 ret = trace_define_field(event_call, parg->type->fmttype,
1135 parg->name,
1136 sizeof(field) + parg->offset,
1137 parg->type->size,
1138 parg->type->is_signed,
1139 FILTER_OTHER);
1140 if (ret)
1141 return ret;
1142 }
1143 return 0;
1144}
1145
1146static int kretprobe_event_define_fields(struct trace_event_call *event_call)
1147{
1148 int ret, i;
1149 struct kretprobe_trace_entry_head field;
1150 struct trace_kprobe *tk = (struct trace_kprobe *)event_call->data;
1151
1152 DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1153 DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1154
1155 for (i = 0; i < tk->tp.nr_args; i++) {
1156 struct probe_arg *parg = &tk->tp.args[i];
1157
1158 ret = trace_define_field(event_call, parg->type->fmttype,
1159 parg->name,
1160 sizeof(field) + parg->offset,
1161 parg->type->size,
1162 parg->type->is_signed,
1163 FILTER_OTHER);
1164 if (ret)
1165 return ret;
1166 }
1167 return 0;
1168}
1169
1170#ifdef CONFIG_PERF_EVENTS
1171
1172
1173static void
1174kprobe_perf_func(struct trace_kprobe *tk, struct pt_regs *regs)
1175{
1176 struct trace_event_call *call = &tk->tp.call;
1177 struct bpf_prog *prog = call->prog;
1178 struct kprobe_trace_entry_head *entry;
1179 struct hlist_head *head;
1180 int size, __size, dsize;
1181 int rctx;
1182
1183 if (prog && !trace_call_bpf(prog, regs))
1184 return;
1185
1186 head = this_cpu_ptr(call->perf_events);
1187 if (hlist_empty(head))
1188 return;
1189
1190 dsize = __get_data_size(&tk->tp, regs);
1191 __size = sizeof(*entry) + tk->tp.size + dsize;
1192 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1193 size -= sizeof(u32);
1194
1195 entry = perf_trace_buf_alloc(size, NULL, &rctx);
1196 if (!entry)
1197 return;
1198
1199 entry->ip = (unsigned long)tk->rp.kp.addr;
1200 memset(&entry[1], 0, dsize);
1201 store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1202 perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1203 head, NULL, NULL);
1204}
1205NOKPROBE_SYMBOL(kprobe_perf_func);
1206
1207
1208static void
1209kretprobe_perf_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1210 struct pt_regs *regs)
1211{
1212 struct trace_event_call *call = &tk->tp.call;
1213 struct bpf_prog *prog = call->prog;
1214 struct kretprobe_trace_entry_head *entry;
1215 struct hlist_head *head;
1216 int size, __size, dsize;
1217 int rctx;
1218
1219 if (prog && !trace_call_bpf(prog, regs))
1220 return;
1221
1222 head = this_cpu_ptr(call->perf_events);
1223 if (hlist_empty(head))
1224 return;
1225
1226 dsize = __get_data_size(&tk->tp, regs);
1227 __size = sizeof(*entry) + tk->tp.size + dsize;
1228 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1229 size -= sizeof(u32);
1230
1231 entry = perf_trace_buf_alloc(size, NULL, &rctx);
1232 if (!entry)
1233 return;
1234
1235 entry->func = (unsigned long)tk->rp.kp.addr;
1236 entry->ret_ip = (unsigned long)ri->ret_addr;
1237 store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1238 perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1239 head, NULL, NULL);
1240}
1241NOKPROBE_SYMBOL(kretprobe_perf_func);
1242#endif
1243
1244
1245
1246
1247
1248
1249
1250static int kprobe_register(struct trace_event_call *event,
1251 enum trace_reg type, void *data)
1252{
1253 struct trace_kprobe *tk = (struct trace_kprobe *)event->data;
1254 struct trace_event_file *file = data;
1255
1256 switch (type) {
1257 case TRACE_REG_REGISTER:
1258 return enable_trace_kprobe(tk, file);
1259 case TRACE_REG_UNREGISTER:
1260 return disable_trace_kprobe(tk, file);
1261
1262#ifdef CONFIG_PERF_EVENTS
1263 case TRACE_REG_PERF_REGISTER:
1264 return enable_trace_kprobe(tk, NULL);
1265 case TRACE_REG_PERF_UNREGISTER:
1266 return disable_trace_kprobe(tk, NULL);
1267 case TRACE_REG_PERF_OPEN:
1268 case TRACE_REG_PERF_CLOSE:
1269 case TRACE_REG_PERF_ADD:
1270 case TRACE_REG_PERF_DEL:
1271 return 0;
1272#endif
1273 }
1274 return 0;
1275}
1276
1277static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1278{
1279 struct trace_kprobe *tk = container_of(kp, struct trace_kprobe, rp.kp);
1280
1281 raw_cpu_inc(*tk->nhit);
1282
1283 if (tk->tp.flags & TP_FLAG_TRACE)
1284 kprobe_trace_func(tk, regs);
1285#ifdef CONFIG_PERF_EVENTS
1286 if (tk->tp.flags & TP_FLAG_PROFILE)
1287 kprobe_perf_func(tk, regs);
1288#endif
1289 return 0;
1290}
1291NOKPROBE_SYMBOL(kprobe_dispatcher);
1292
1293static int
1294kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1295{
1296 struct trace_kprobe *tk = container_of(ri->rp, struct trace_kprobe, rp);
1297
1298 raw_cpu_inc(*tk->nhit);
1299
1300 if (tk->tp.flags & TP_FLAG_TRACE)
1301 kretprobe_trace_func(tk, ri, regs);
1302#ifdef CONFIG_PERF_EVENTS
1303 if (tk->tp.flags & TP_FLAG_PROFILE)
1304 kretprobe_perf_func(tk, ri, regs);
1305#endif
1306 return 0;
1307}
1308NOKPROBE_SYMBOL(kretprobe_dispatcher);
1309
1310static struct trace_event_functions kretprobe_funcs = {
1311 .trace = print_kretprobe_event
1312};
1313
1314static struct trace_event_functions kprobe_funcs = {
1315 .trace = print_kprobe_event
1316};
1317
1318static int register_kprobe_event(struct trace_kprobe *tk)
1319{
1320 struct trace_event_call *call = &tk->tp.call;
1321 int ret;
1322
1323
1324 INIT_LIST_HEAD(&call->class->fields);
1325 if (trace_kprobe_is_return(tk)) {
1326 call->event.funcs = &kretprobe_funcs;
1327 call->class->define_fields = kretprobe_event_define_fields;
1328 } else {
1329 call->event.funcs = &kprobe_funcs;
1330 call->class->define_fields = kprobe_event_define_fields;
1331 }
1332 if (set_print_fmt(&tk->tp, trace_kprobe_is_return(tk)) < 0)
1333 return -ENOMEM;
1334 ret = register_trace_event(&call->event);
1335 if (!ret) {
1336 kfree(call->print_fmt);
1337 return -ENODEV;
1338 }
1339 call->flags = TRACE_EVENT_FL_KPROBE;
1340 call->class->reg = kprobe_register;
1341 call->data = tk;
1342 ret = trace_add_event_call(call);
1343 if (ret) {
1344 pr_info("Failed to register kprobe event: %s\n",
1345 trace_event_name(call));
1346 kfree(call->print_fmt);
1347 unregister_trace_event(&call->event);
1348 }
1349 return ret;
1350}
1351
1352static int unregister_kprobe_event(struct trace_kprobe *tk)
1353{
1354 int ret;
1355
1356
1357 ret = trace_remove_event_call(&tk->tp.call);
1358 if (!ret)
1359 kfree(tk->tp.call.print_fmt);
1360 return ret;
1361}
1362
1363
1364static __init int init_kprobe_trace(void)
1365{
1366 struct dentry *d_tracer;
1367 struct dentry *entry;
1368
1369 if (register_module_notifier(&trace_kprobe_module_nb))
1370 return -EINVAL;
1371
1372 d_tracer = tracing_init_dentry();
1373 if (IS_ERR(d_tracer))
1374 return 0;
1375
1376 entry = tracefs_create_file("kprobe_events", 0644, d_tracer,
1377 NULL, &kprobe_events_ops);
1378
1379
1380 if (!entry)
1381 pr_warn("Could not create tracefs 'kprobe_events' entry\n");
1382
1383
1384 entry = tracefs_create_file("kprobe_profile", 0444, d_tracer,
1385 NULL, &kprobe_profile_ops);
1386
1387 if (!entry)
1388 pr_warn("Could not create tracefs 'kprobe_profile' entry\n");
1389 return 0;
1390}
1391fs_initcall(init_kprobe_trace);
1392
1393
1394#ifdef CONFIG_FTRACE_STARTUP_TEST
1395
1396
1397
1398
1399
1400static __used __init noinline int
1401kprobe_trace_selftest_target(int a1, int a2, int a3, int a4, int a5, int a6)
1402{
1403 return a1 + a2 + a3 + a4 + a5 + a6;
1404}
1405
1406static __init struct trace_event_file *
1407find_trace_probe_file(struct trace_kprobe *tk, struct trace_array *tr)
1408{
1409 struct trace_event_file *file;
1410
1411 list_for_each_entry(file, &tr->events, list)
1412 if (file->event_call == &tk->tp.call)
1413 return file;
1414
1415 return NULL;
1416}
1417
1418
1419
1420
1421
1422static __init int kprobe_trace_self_tests_init(void)
1423{
1424 int ret, warn = 0;
1425 int (*target)(int, int, int, int, int, int);
1426 struct trace_kprobe *tk;
1427 struct trace_event_file *file;
1428
1429 if (tracing_is_disabled())
1430 return -ENODEV;
1431
1432 target = kprobe_trace_selftest_target;
1433
1434 pr_info("Testing kprobe tracing: ");
1435
1436 ret = traceprobe_command("p:testprobe kprobe_trace_selftest_target "
1437 "$stack $stack0 +0($stack)",
1438 create_trace_kprobe);
1439 if (WARN_ON_ONCE(ret)) {
1440 pr_warn("error on probing function entry.\n");
1441 warn++;
1442 } else {
1443
1444 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1445 if (WARN_ON_ONCE(tk == NULL)) {
1446 pr_warn("error on getting new probe.\n");
1447 warn++;
1448 } else {
1449 file = find_trace_probe_file(tk, top_trace_array());
1450 if (WARN_ON_ONCE(file == NULL)) {
1451 pr_warn("error on getting probe file.\n");
1452 warn++;
1453 } else
1454 enable_trace_kprobe(tk, file);
1455 }
1456 }
1457
1458 ret = traceprobe_command("r:testprobe2 kprobe_trace_selftest_target "
1459 "$retval", create_trace_kprobe);
1460 if (WARN_ON_ONCE(ret)) {
1461 pr_warn("error on probing function return.\n");
1462 warn++;
1463 } else {
1464
1465 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1466 if (WARN_ON_ONCE(tk == NULL)) {
1467 pr_warn("error on getting 2nd new probe.\n");
1468 warn++;
1469 } else {
1470 file = find_trace_probe_file(tk, top_trace_array());
1471 if (WARN_ON_ONCE(file == NULL)) {
1472 pr_warn("error on getting probe file.\n");
1473 warn++;
1474 } else
1475 enable_trace_kprobe(tk, file);
1476 }
1477 }
1478
1479 if (warn)
1480 goto end;
1481
1482 ret = target(1, 2, 3, 4, 5, 6);
1483
1484
1485
1486
1487
1488
1489 if (ret != 21)
1490 warn++;
1491
1492
1493 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1494 if (WARN_ON_ONCE(tk == NULL)) {
1495 pr_warn("error on getting test probe.\n");
1496 warn++;
1497 } else {
1498 if (trace_kprobe_nhit(tk) != 1) {
1499 pr_warn("incorrect number of testprobe hits\n");
1500 warn++;
1501 }
1502
1503 file = find_trace_probe_file(tk, top_trace_array());
1504 if (WARN_ON_ONCE(file == NULL)) {
1505 pr_warn("error on getting probe file.\n");
1506 warn++;
1507 } else
1508 disable_trace_kprobe(tk, file);
1509 }
1510
1511 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1512 if (WARN_ON_ONCE(tk == NULL)) {
1513 pr_warn("error on getting 2nd test probe.\n");
1514 warn++;
1515 } else {
1516 if (trace_kprobe_nhit(tk) != 1) {
1517 pr_warn("incorrect number of testprobe2 hits\n");
1518 warn++;
1519 }
1520
1521 file = find_trace_probe_file(tk, top_trace_array());
1522 if (WARN_ON_ONCE(file == NULL)) {
1523 pr_warn("error on getting probe file.\n");
1524 warn++;
1525 } else
1526 disable_trace_kprobe(tk, file);
1527 }
1528
1529 ret = traceprobe_command("-:testprobe", create_trace_kprobe);
1530 if (WARN_ON_ONCE(ret)) {
1531 pr_warn("error on deleting a probe.\n");
1532 warn++;
1533 }
1534
1535 ret = traceprobe_command("-:testprobe2", create_trace_kprobe);
1536 if (WARN_ON_ONCE(ret)) {
1537 pr_warn("error on deleting a probe.\n");
1538 warn++;
1539 }
1540
1541end:
1542 release_all_trace_kprobes();
1543
1544
1545
1546
1547 wait_for_kprobe_optimizer();
1548 if (warn)
1549 pr_cont("NG: Some tests are failed. Please check them.\n");
1550 else
1551 pr_cont("OK\n");
1552 return 0;
1553}
1554
1555late_initcall(kprobe_trace_self_tests_init);
1556
1557#endif
1558
