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#define pr_fmt(fmt) "trace_kprobe: " fmt
20
21#include <linux/module.h>
22#include <linux/uaccess.h>
23#include <linux/rculist.h>
24#include <linux/error-injection.h>
25
26#include "trace_probe.h"
27
28#define KPROBE_EVENT_SYSTEM "kprobes"
29#define KRETPROBE_MAXACTIVE_MAX 4096
30
31/**
32 * Kprobe event core functions
33 */
34struct trace_kprobe {
35 struct list_head list;
36 struct kretprobe rp; /* Use rp.kp for kprobe use */
37 unsigned long __percpu *nhit;
38 const char *symbol; /* symbol name */
39 struct trace_probe tp;
40};
41
42#define SIZEOF_TRACE_KPROBE(n) \
43 (offsetof(struct trace_kprobe, tp.args) + \
44 (sizeof(struct probe_arg) * (n)))
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
90bool trace_kprobe_on_func_entry(struct trace_event_call *call)
91{
92 struct trace_kprobe *tk = (struct trace_kprobe *)call->data;
93
94 return kprobe_on_func_entry(tk->rp.kp.addr,
95 tk->rp.kp.addr ? NULL : tk->rp.kp.symbol_name,
96 tk->rp.kp.addr ? 0 : tk->rp.kp.offset);
97}
98
99bool trace_kprobe_error_injectable(struct trace_event_call *call)
100{
101 struct trace_kprobe *tk = (struct trace_kprobe *)call->data;
102 unsigned long addr;
103
104 if (tk->symbol) {
105 addr = (unsigned long)
106 kallsyms_lookup_name(trace_kprobe_symbol(tk));
107 addr += tk->rp.kp.offset;
108 } else {
109 addr = (unsigned long)tk->rp.kp.addr;
110 }
111 return within_error_injection_list(addr);
112}
113
114static int register_kprobe_event(struct trace_kprobe *tk);
115static int unregister_kprobe_event(struct trace_kprobe *tk);
116
117static DEFINE_MUTEX(probe_lock);
118static LIST_HEAD(probe_list);
119
120static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
121static int kretprobe_dispatcher(struct kretprobe_instance *ri,
122 struct pt_regs *regs);
123
124/* Memory fetching by symbol */
125struct symbol_cache {
126 char *symbol;
127 long offset;
128 unsigned long addr;
129};
130
131unsigned long update_symbol_cache(struct symbol_cache *sc)
132{
133 sc->addr = (unsigned long)kallsyms_lookup_name(sc->symbol);
134
135 if (sc->addr)
136 sc->addr += sc->offset;
137
138 return sc->addr;
139}
140
141void free_symbol_cache(struct symbol_cache *sc)
142{
143 kfree(sc->symbol);
144 kfree(sc);
145}
146
147struct symbol_cache *alloc_symbol_cache(const char *sym, long offset)
148{
149 struct symbol_cache *sc;
150
151 if (!sym || strlen(sym) == 0)
152 return NULL;
153
154 sc = kzalloc(sizeof(struct symbol_cache), GFP_KERNEL);
155 if (!sc)
156 return NULL;
157
158 sc->symbol = kstrdup(sym, GFP_KERNEL);
159 if (!sc->symbol) {
160 kfree(sc);
161 return NULL;
162 }
163 sc->offset = offset;
164 update_symbol_cache(sc);
165
166 return sc;
167}
168
169/*
170 * Kprobes-specific fetch functions
171 */
172#define DEFINE_FETCH_stack(type) \
173static void FETCH_FUNC_NAME(stack, type)(struct pt_regs *regs, \
174 void *offset, void *dest) \
175{ \
176 *(type *)dest = (type)regs_get_kernel_stack_nth(regs, \
177 (unsigned int)((unsigned long)offset)); \
178} \
179NOKPROBE_SYMBOL(FETCH_FUNC_NAME(stack, type));
180
181DEFINE_BASIC_FETCH_FUNCS(stack)
182/* No string on the stack entry */
183#define fetch_stack_string NULL
184#define fetch_stack_string_size NULL
185
186#define DEFINE_FETCH_memory(type) \
187static void FETCH_FUNC_NAME(memory, type)(struct pt_regs *regs, \
188 void *addr, void *dest) \
189{ \
190 type retval; \
191 if (probe_kernel_address(addr, retval)) \
192 *(type *)dest = 0; \
193 else \
194 *(type *)dest = retval; \
195} \
196NOKPROBE_SYMBOL(FETCH_FUNC_NAME(memory, type));
197
198DEFINE_BASIC_FETCH_FUNCS(memory)
199/*
200 * Fetch a null-terminated string. Caller MUST set *(u32 *)dest with max
201 * length and relative data location.
202 */
203static void FETCH_FUNC_NAME(memory, string)(struct pt_regs *regs,
204 void *addr, void *dest)
205{
206 int maxlen = get_rloc_len(*(u32 *)dest);
207 u8 *dst = get_rloc_data(dest);
208 long ret;
209
210 if (!maxlen)
211 return;
212
213 /*
214 * Try to get string again, since the string can be changed while
215 * probing.
216 */
217 ret = strncpy_from_unsafe(dst, addr, maxlen);
218
219 if (ret < 0) { /* Failed to fetch string */
220 dst[0] = '\0';
221 *(u32 *)dest = make_data_rloc(0, get_rloc_offs(*(u32 *)dest));
222 } else {
223 *(u32 *)dest = make_data_rloc(ret, get_rloc_offs(*(u32 *)dest));
224 }
225}
226NOKPROBE_SYMBOL(FETCH_FUNC_NAME(memory, string));
227
228/* Return the length of string -- including null terminal byte */
229static void FETCH_FUNC_NAME(memory, string_size)(struct pt_regs *regs,
230 void *addr, void *dest)
231{
232 mm_segment_t old_fs;
233 int ret, len = 0;
234 u8 c;
235
236 old_fs = get_fs();
237 set_fs(KERNEL_DS);
238 pagefault_disable();
239
240 do {
241 ret = __copy_from_user_inatomic(&c, (u8 *)addr + len, 1);
242 len++;
243 } while (c && ret == 0 && len < MAX_STRING_SIZE);
244
245 pagefault_enable();
246 set_fs(old_fs);
247
248 if (ret < 0) /* Failed to check the length */
249 *(u32 *)dest = 0;
250 else
251 *(u32 *)dest = len;
252}
253NOKPROBE_SYMBOL(FETCH_FUNC_NAME(memory, string_size));
254
255#define DEFINE_FETCH_symbol(type) \
256void FETCH_FUNC_NAME(symbol, type)(struct pt_regs *regs, void *data, void *dest)\
257{ \
258 struct symbol_cache *sc = data; \
259 if (sc->addr) \
260 fetch_memory_##type(regs, (void *)sc->addr, dest); \
261 else \
262 *(type *)dest = 0; \
263} \
264NOKPROBE_SYMBOL(FETCH_FUNC_NAME(symbol, type));
265
266DEFINE_BASIC_FETCH_FUNCS(symbol)
267DEFINE_FETCH_symbol(string)
268DEFINE_FETCH_symbol(string_size)
269
270/* kprobes don't support file_offset fetch methods */
271#define fetch_file_offset_u8 NULL
272#define fetch_file_offset_u16 NULL
273#define fetch_file_offset_u32 NULL
274#define fetch_file_offset_u64 NULL
275#define fetch_file_offset_string NULL
276#define fetch_file_offset_string_size NULL
277
278/* Fetch type information table */
279static const struct fetch_type kprobes_fetch_type_table[] = {
280 /* Special types */
281 [FETCH_TYPE_STRING] = __ASSIGN_FETCH_TYPE("string", string, string,
282 sizeof(u32), 1, "__data_loc char[]"),
283 [FETCH_TYPE_STRSIZE] = __ASSIGN_FETCH_TYPE("string_size", u32,
284 string_size, sizeof(u32), 0, "u32"),
285 /* Basic types */
286 ASSIGN_FETCH_TYPE(u8, u8, 0),
287 ASSIGN_FETCH_TYPE(u16, u16, 0),
288 ASSIGN_FETCH_TYPE(u32, u32, 0),
289 ASSIGN_FETCH_TYPE(u64, u64, 0),
290 ASSIGN_FETCH_TYPE(s8, u8, 1),
291 ASSIGN_FETCH_TYPE(s16, u16, 1),
292 ASSIGN_FETCH_TYPE(s32, u32, 1),
293 ASSIGN_FETCH_TYPE(s64, u64, 1),
294 ASSIGN_FETCH_TYPE_ALIAS(x8, u8, u8, 0),
295 ASSIGN_FETCH_TYPE_ALIAS(x16, u16, u16, 0),
296 ASSIGN_FETCH_TYPE_ALIAS(x32, u32, u32, 0),
297 ASSIGN_FETCH_TYPE_ALIAS(x64, u64, u64, 0),
298
299 ASSIGN_FETCH_TYPE_END
300};
301
302/*
303 * Allocate new trace_probe and initialize it (including kprobes).
304 */
305static struct trace_kprobe *alloc_trace_kprobe(const char *group,
306 const char *event,
307 void *addr,
308 const char *symbol,
309 unsigned long offs,
310 int maxactive,
311 int nargs, bool is_return)
312{
313 struct trace_kprobe *tk;
314 int ret = -ENOMEM;
315
316 tk = kzalloc(SIZEOF_TRACE_KPROBE(nargs), GFP_KERNEL);
317 if (!tk)
318 return ERR_PTR(ret);
319
320 tk->nhit = alloc_percpu(unsigned long);
321 if (!tk->nhit)
322 goto error;
323
324 if (symbol) {
325 tk->symbol = kstrdup(symbol, GFP_KERNEL);
326 if (!tk->symbol)
327 goto error;
328 tk->rp.kp.symbol_name = tk->symbol;
329 tk->rp.kp.offset = offs;
330 } else
331 tk->rp.kp.addr = addr;
332
333 if (is_return)
334 tk->rp.handler = kretprobe_dispatcher;
335 else
336 tk->rp.kp.pre_handler = kprobe_dispatcher;
337
338 tk->rp.maxactive = maxactive;
339
340 if (!event || !is_good_name(event)) {
341 ret = -EINVAL;
342 goto error;
343 }
344
345 tk->tp.call.class = &tk->tp.class;
346 tk->tp.call.name = kstrdup(event, GFP_KERNEL);
347 if (!tk->tp.call.name)
348 goto error;
349
350 if (!group || !is_good_name(group)) {
351 ret = -EINVAL;
352 goto error;
353 }
354
355 tk->tp.class.system = kstrdup(group, GFP_KERNEL);
356 if (!tk->tp.class.system)
357 goto error;
358
359 INIT_LIST_HEAD(&tk->list);
360 INIT_LIST_HEAD(&tk->tp.files);
361 return tk;
362error:
363 kfree(tk->tp.call.name);
364 kfree(tk->symbol);
365 free_percpu(tk->nhit);
366 kfree(tk);
367 return ERR_PTR(ret);
368}
369
370static void free_trace_kprobe(struct trace_kprobe *tk)
371{
372 int i;
373
374 for (i = 0; i < tk->tp.nr_args; i++)
375 traceprobe_free_probe_arg(&tk->tp.args[i]);
376
377 kfree(tk->tp.call.class->system);
378 kfree(tk->tp.call.name);
379 kfree(tk->symbol);
380 free_percpu(tk->nhit);
381 kfree(tk);
382}
383
384static struct trace_kprobe *find_trace_kprobe(const char *event,
385 const char *group)
386{
387 struct trace_kprobe *tk;
388
389 list_for_each_entry(tk, &probe_list, list)
390 if (strcmp(trace_event_name(&tk->tp.call), event) == 0 &&
391 strcmp(tk->tp.call.class->system, group) == 0)
392 return tk;
393 return NULL;
394}
395
396/*
397 * Enable trace_probe
398 * if the file is NULL, enable "perf" handler, or enable "trace" handler.
399 */
400static int
401enable_trace_kprobe(struct trace_kprobe *tk, struct trace_event_file *file)
402{
403 int ret = 0;
404
405 if (file) {
406 struct event_file_link *link;
407
408 link = kmalloc(sizeof(*link), GFP_KERNEL);
409 if (!link) {
410 ret = -ENOMEM;
411 goto out;
412 }
413
414 link->file = file;
415 list_add_tail_rcu(&link->list, &tk->tp.files);
416
417 tk->tp.flags |= TP_FLAG_TRACE;
418 } else
419 tk->tp.flags |= TP_FLAG_PROFILE;
420
421 if (trace_probe_is_registered(&tk->tp) && !trace_kprobe_has_gone(tk)) {
422 if (trace_kprobe_is_return(tk))
423 ret = enable_kretprobe(&tk->rp);
424 else
425 ret = enable_kprobe(&tk->rp.kp);
426 }
427 out:
428 return ret;
429}
430
431/*
432 * Disable trace_probe
433 * if the file is NULL, disable "perf" handler, or disable "trace" handler.
434 */
435static int
436disable_trace_kprobe(struct trace_kprobe *tk, struct trace_event_file *file)
437{
438 struct event_file_link *link = NULL;
439 int wait = 0;
440 int ret = 0;
441
442 if (file) {
443 link = find_event_file_link(&tk->tp, file);
444 if (!link) {
445 ret = -EINVAL;
446 goto out;
447 }
448
449 list_del_rcu(&link->list);
450 wait = 1;
451 if (!list_empty(&tk->tp.files))
452 goto out;
453
454 tk->tp.flags &= ~TP_FLAG_TRACE;
455 } else
456 tk->tp.flags &= ~TP_FLAG_PROFILE;
457
458 if (!trace_probe_is_enabled(&tk->tp) && trace_probe_is_registered(&tk->tp)) {
459 if (trace_kprobe_is_return(tk))
460 disable_kretprobe(&tk->rp);
461 else
462 disable_kprobe(&tk->rp.kp);
463 wait = 1;
464 }
465
466 /*
467 * if tk is not added to any list, it must be a local trace_kprobe
468 * created with perf_event_open. We don't need to wait for these
469 * trace_kprobes
470 */
471 if (list_empty(&tk->list))
472 wait = 0;
473 out:
474 if (wait) {
475 /*
476 * Synchronize with kprobe_trace_func/kretprobe_trace_func
477 * to ensure disabled (all running handlers are finished).
478 * This is not only for kfree(), but also the caller,
479 * trace_remove_event_call() supposes it for releasing
480 * event_call related objects, which will be accessed in
481 * the kprobe_trace_func/kretprobe_trace_func.
482 */
483 synchronize_sched();
484 kfree(link); /* Ignored if link == NULL */
485 }
486
487 return ret;
488}
489
490/* Internal register function - just handle k*probes and flags */
491static int __register_trace_kprobe(struct trace_kprobe *tk)
492{
493 int i, ret;
494
495 if (trace_probe_is_registered(&tk->tp))
496 return -EINVAL;
497
498 for (i = 0; i < tk->tp.nr_args; i++)
499 traceprobe_update_arg(&tk->tp.args[i]);
500
501 /* Set/clear disabled flag according to tp->flag */
502 if (trace_probe_is_enabled(&tk->tp))
503 tk->rp.kp.flags &= ~KPROBE_FLAG_DISABLED;
504 else
505 tk->rp.kp.flags |= KPROBE_FLAG_DISABLED;
506
507 if (trace_kprobe_is_return(tk))
508 ret = register_kretprobe(&tk->rp);
509 else
510 ret = register_kprobe(&tk->rp.kp);
511
512 if (ret == 0)
513 tk->tp.flags |= TP_FLAG_REGISTERED;
514 else {
515 if (ret == -ENOENT && trace_kprobe_is_on_module(tk)) {
516 pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
517 ret = 0;
518 } else if (ret == -EILSEQ) {
519 pr_warn("Probing address(0x%p) is not an instruction boundary.\n",
520 tk->rp.kp.addr);
521 ret = -EINVAL;
522 }
523 }
524
525 return ret;
526}
527
528/* Internal unregister function - just handle k*probes and flags */
529static void __unregister_trace_kprobe(struct trace_kprobe *tk)
530{
531 if (trace_probe_is_registered(&tk->tp)) {
532 if (trace_kprobe_is_return(tk))
533 unregister_kretprobe(&tk->rp);
534 else
535 unregister_kprobe(&tk->rp.kp);
536 tk->tp.flags &= ~TP_FLAG_REGISTERED;
537 /* Cleanup kprobe for reuse */
538 if (tk->rp.kp.symbol_name)
539 tk->rp.kp.addr = NULL;
540 }
541}
542
543/* Unregister a trace_probe and probe_event: call with locking probe_lock */
544static int unregister_trace_kprobe(struct trace_kprobe *tk)
545{
546 /* Enabled event can not be unregistered */
547 if (trace_probe_is_enabled(&tk->tp))
548 return -EBUSY;
549
550 /* Will fail if probe is being used by ftrace or perf */
551 if (unregister_kprobe_event(tk))
552 return -EBUSY;
553
554 __unregister_trace_kprobe(tk);
555 list_del(&tk->list);
556
557 return 0;
558}
559
560/* Register a trace_probe and probe_event */
561static int register_trace_kprobe(struct trace_kprobe *tk)
562{
563 struct trace_kprobe *old_tk;
564 int ret;
565
566 mutex_lock(&probe_lock);
567
568 /* Delete old (same name) event if exist */
569 old_tk = find_trace_kprobe(trace_event_name(&tk->tp.call),
570 tk->tp.call.class->system);
571 if (old_tk) {
572 ret = unregister_trace_kprobe(old_tk);
573 if (ret < 0)
574 goto end;
575 free_trace_kprobe(old_tk);
576 }
577
578 /* Register new event */
579 ret = register_kprobe_event(tk);
580 if (ret) {
581 pr_warn("Failed to register probe event(%d)\n", ret);
582 goto end;
583 }
584
585 /* Register k*probe */
586 ret = __register_trace_kprobe(tk);
587 if (ret < 0)
588 unregister_kprobe_event(tk);
589 else
590 list_add_tail(&tk->list, &probe_list);
591
592end:
593 mutex_unlock(&probe_lock);
594 return ret;
595}
596
597/* Module notifier call back, checking event on the module */
598static int trace_kprobe_module_callback(struct notifier_block *nb,
599 unsigned long val, void *data)
600{
601 struct module *mod = data;
602 struct trace_kprobe *tk;
603 int ret;
604
605 if (val != MODULE_STATE_COMING)
606 return NOTIFY_DONE;
607
608 /* Update probes on coming module */
609 mutex_lock(&probe_lock);
610 list_for_each_entry(tk, &probe_list, list) {
611 if (trace_kprobe_within_module(tk, mod)) {
612 /* Don't need to check busy - this should have gone. */
613 __unregister_trace_kprobe(tk);
614 ret = __register_trace_kprobe(tk);
615 if (ret)
616 pr_warn("Failed to re-register probe %s on %s: %d\n",
617 trace_event_name(&tk->tp.call),
618 mod->name, ret);
619 }
620 }
621 mutex_unlock(&probe_lock);
622
623 return NOTIFY_DONE;
624}
625
626static struct notifier_block trace_kprobe_module_nb = {
627 .notifier_call = trace_kprobe_module_callback,
628 .priority = 1 /* Invoked after kprobe module callback */
629};
630
631/* Convert certain expected symbols into '_' when generating event names */
632static inline void sanitize_event_name(char *name)
633{
634 while (*name++ != '\0')
635 if (*name == ':' || *name == '.')
636 *name = '_';
637}
638
639static int create_trace_kprobe(int argc, char **argv)
640{
641 /*
642 * Argument syntax:
643 * - Add kprobe:
644 * p[:[GRP/]EVENT] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
645 * - Add kretprobe:
646 * r[MAXACTIVE][:[GRP/]EVENT] [MOD:]KSYM[+0] [FETCHARGS]
647 * Fetch args:
648 * $retval : fetch return value
649 * $stack : fetch stack address
650 * $stackN : fetch Nth of stack (N:0-)
651 * $comm : fetch current task comm
652 * @ADDR : fetch memory at ADDR (ADDR should be in kernel)
653 * @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
654 * %REG : fetch register REG
655 * Dereferencing memory fetch:
656 * +|-offs(ARG) : fetch memory at ARG +|- offs address.
657 * Alias name of args:
658 * NAME=FETCHARG : set NAME as alias of FETCHARG.
659 * Type of args:
660 * FETCHARG:TYPE : use TYPE instead of unsigned long.
661 */
662 struct trace_kprobe *tk;
663 int i, ret = 0;
664 bool is_return = false, is_delete = false;
665 char *symbol = NULL, *event = NULL, *group = NULL;
666 int maxactive = 0;
667 char *arg;
668 long offset = 0;
669 void *addr = NULL;
670 char buf[MAX_EVENT_NAME_LEN];
671
672 /* argc must be >= 1 */
673 if (argv[0][0] == 'p')
674 is_return = false;
675 else if (argv[0][0] == 'r')
676 is_return = true;
677 else if (argv[0][0] == '-')
678 is_delete = true;
679 else {
680 pr_info("Probe definition must be started with 'p', 'r' or"
681 " '-'.\n");
682 return -EINVAL;
683 }
684
685 event = strchr(&argv[0][1], ':');
686 if (event) {
687 event[0] = '\0';
688 event++;
689 }
690 if (is_return && isdigit(argv[0][1])) {
691 ret = kstrtouint(&argv[0][1], 0, &maxactive);
692 if (ret) {
693 pr_info("Failed to parse maxactive.\n");
694 return ret;
695 }
696 /* kretprobes instances are iterated over via a list. The
697 * maximum should stay reasonable.
698 */
699 if (maxactive > KRETPROBE_MAXACTIVE_MAX) {
700 pr_info("Maxactive is too big (%d > %d).\n",
701 maxactive, KRETPROBE_MAXACTIVE_MAX);
702 return -E2BIG;
703 }
704 }
705
706 if (event) {
707 if (strchr(event, '/')) {
708 group = event;
709 event = strchr(group, '/') + 1;
710 event[-1] = '\0';
711 if (strlen(group) == 0) {
712 pr_info("Group name is not specified\n");
713 return -EINVAL;
714 }
715 }
716 if (strlen(event) == 0) {
717 pr_info("Event name is not specified\n");
718 return -EINVAL;
719 }
720 }
721 if (!group)
722 group = KPROBE_EVENT_SYSTEM;
723
724 if (is_delete) {
725 if (!event) {
726 pr_info("Delete command needs an event name.\n");
727 return -EINVAL;
728 }
729 mutex_lock(&probe_lock);
730 tk = find_trace_kprobe(event, group);
731 if (!tk) {
732 mutex_unlock(&probe_lock);
733 pr_info("Event %s/%s doesn't exist.\n", group, event);
734 return -ENOENT;
735 }
736 /* delete an event */
737 ret = unregister_trace_kprobe(tk);
738 if (ret == 0)
739 free_trace_kprobe(tk);
740 mutex_unlock(&probe_lock);
741 return ret;
742 }
743
744 if (argc < 2) {
745 pr_info("Probe point is not specified.\n");
746 return -EINVAL;
747 }
748
749 /* try to parse an address. if that fails, try to read the
750 * input as a symbol. */
751 if (kstrtoul(argv[1], 0, (unsigned long *)&addr)) {
752 /* a symbol specified */
753 symbol = argv[1];
754 /* TODO: support .init module functions */
755 ret = traceprobe_split_symbol_offset(symbol, &offset);
756 if (ret || offset < 0 || offset > UINT_MAX) {
757 pr_info("Failed to parse either an address or a symbol.\n");
758 return ret;
759 }
760 if (offset && is_return &&
761 !kprobe_on_func_entry(NULL, symbol, offset)) {
762 pr_info("Given offset is not valid for return probe.\n");
763 return -EINVAL;
764 }
765 }
766 argc -= 2; argv += 2;
767
768 /* setup a probe */
769 if (!event) {
770 /* Make a new event name */
771 if (symbol)
772 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
773 is_return ? 'r' : 'p', symbol, offset);
774 else
775 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
776 is_return ? 'r' : 'p', addr);
777 sanitize_event_name(buf);
778 event = buf;
779 }
780 tk = alloc_trace_kprobe(group, event, addr, symbol, offset, maxactive,
781 argc, is_return);
782 if (IS_ERR(tk)) {
783 pr_info("Failed to allocate trace_probe.(%d)\n",
784 (int)PTR_ERR(tk));
785 return PTR_ERR(tk);
786 }
787
788 /* parse arguments */
789 ret = 0;
790 for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
791 struct probe_arg *parg = &tk->tp.args[i];
792
793 /* Increment count for freeing args in error case */
794 tk->tp.nr_args++;
795
796 /* Parse argument name */
797 arg = strchr(argv[i], '=');
798 if (arg) {
799 *arg++ = '\0';
800 parg->name = kstrdup(argv[i], GFP_KERNEL);
801 } else {
802 arg = argv[i];
803 /* If argument name is omitted, set "argN" */
804 snprintf(buf, MAX_EVENT_NAME_LEN, "arg%d", i + 1);
805 parg->name = kstrdup(buf, GFP_KERNEL);
806 }
807
808 if (!parg->name) {
809 pr_info("Failed to allocate argument[%d] name.\n", i);
810 ret = -ENOMEM;
811 goto error;
812 }
813
814 if (!is_good_name(parg->name)) {
815 pr_info("Invalid argument[%d] name: %s\n",
816 i, parg->name);
817 ret = -EINVAL;
818 goto error;
819 }
820
821 if (traceprobe_conflict_field_name(parg->name,
822 tk->tp.args, i)) {
823 pr_info("Argument[%d] name '%s' conflicts with "
824 "another field.\n", i, argv[i]);
825 ret = -EINVAL;
826 goto error;
827 }
828
829 /* Parse fetch argument */
830 ret = traceprobe_parse_probe_arg(arg, &tk->tp.size, parg,
831 is_return, true,
832 kprobes_fetch_type_table);
833 if (ret) {
834 pr_info("Parse error at argument[%d]. (%d)\n", i, ret);
835 goto error;
836 }
837 }
838
839 ret = register_trace_kprobe(tk);
840 if (ret)
841 goto error;
842 return 0;
843
844error:
845 free_trace_kprobe(tk);
846 return ret;
847}
848
849static int release_all_trace_kprobes(void)
850{
851 struct trace_kprobe *tk;
852 int ret = 0;
853
854 mutex_lock(&probe_lock);
855 /* Ensure no probe is in use. */
856 list_for_each_entry(tk, &probe_list, list)
857 if (trace_probe_is_enabled(&tk->tp)) {
858 ret = -EBUSY;
859 goto end;
860 }
861 /* TODO: Use batch unregistration */
862 while (!list_empty(&probe_list)) {
863 tk = list_entry(probe_list.next, struct trace_kprobe, list);
864 ret = unregister_trace_kprobe(tk);
865 if (ret)
866 goto end;
867 free_trace_kprobe(tk);
868 }
869
870end:
871 mutex_unlock(&probe_lock);
872
873 return ret;
874}
875
876/* Probes listing interfaces */
877static void *probes_seq_start(struct seq_file *m, loff_t *pos)
878{
879 mutex_lock(&probe_lock);
880 return seq_list_start(&probe_list, *pos);
881}
882
883static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
884{
885 return seq_list_next(v, &probe_list, pos);
886}
887
888static void probes_seq_stop(struct seq_file *m, void *v)
889{
890 mutex_unlock(&probe_lock);
891}
892
893static int probes_seq_show(struct seq_file *m, void *v)
894{
895 struct trace_kprobe *tk = v;
896 int i;
897
898 seq_putc(m, trace_kprobe_is_return(tk) ? 'r' : 'p');
899 seq_printf(m, ":%s/%s", tk->tp.call.class->system,
900 trace_event_name(&tk->tp.call));
901
902 if (!tk->symbol)
903 seq_printf(m, " 0x%p", tk->rp.kp.addr);
904 else if (tk->rp.kp.offset)
905 seq_printf(m, " %s+%u", trace_kprobe_symbol(tk),
906 tk->rp.kp.offset);
907 else
908 seq_printf(m, " %s", trace_kprobe_symbol(tk));
909
910 for (i = 0; i < tk->tp.nr_args; i++)
911 seq_printf(m, " %s=%s", tk->tp.args[i].name, tk->tp.args[i].comm);
912 seq_putc(m, '\n');
913
914 return 0;
915}
916
917static const struct seq_operations probes_seq_op = {
918 .start = probes_seq_start,
919 .next = probes_seq_next,
920 .stop = probes_seq_stop,
921 .show = probes_seq_show
922};
923
924static int probes_open(struct inode *inode, struct file *file)
925{
926 int ret;
927
928 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
929 ret = release_all_trace_kprobes();
930 if (ret < 0)
931 return ret;
932 }
933
934 return seq_open(file, &probes_seq_op);
935}
936
937static ssize_t probes_write(struct file *file, const char __user *buffer,
938 size_t count, loff_t *ppos)
939{
940 return trace_parse_run_command(file, buffer, count, ppos,
941 create_trace_kprobe);
942}
943
944static const struct file_operations kprobe_events_ops = {
945 .owner = THIS_MODULE,
946 .open = probes_open,
947 .read = seq_read,
948 .llseek = seq_lseek,
949 .release = seq_release,
950 .write = probes_write,
951};
952
953/* Probes profiling interfaces */
954static int probes_profile_seq_show(struct seq_file *m, void *v)
955{
956 struct trace_kprobe *tk = v;
957
958 seq_printf(m, " %-44s %15lu %15lu\n",
959 trace_event_name(&tk->tp.call),
960 trace_kprobe_nhit(tk),
961 tk->rp.kp.nmissed);
962
963 return 0;
964}
965
966static const struct seq_operations profile_seq_op = {
967 .start = probes_seq_start,
968 .next = probes_seq_next,
969 .stop = probes_seq_stop,
970 .show = probes_profile_seq_show
971};
972
973static int profile_open(struct inode *inode, struct file *file)
974{
975 return seq_open(file, &profile_seq_op);
976}
977
978static const struct file_operations kprobe_profile_ops = {
979 .owner = THIS_MODULE,
980 .open = profile_open,
981 .read = seq_read,
982 .llseek = seq_lseek,
983 .release = seq_release,
984};
985
986/* Kprobe handler */
987static nokprobe_inline void
988__kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs,
989 struct trace_event_file *trace_file)
990{
991 struct kprobe_trace_entry_head *entry;
992 struct ring_buffer_event *event;
993 struct ring_buffer *buffer;
994 int size, dsize, pc;
995 unsigned long irq_flags;
996 struct trace_event_call *call = &tk->tp.call;
997
998 WARN_ON(call != trace_file->event_call);
999
1000 if (trace_trigger_soft_disabled(trace_file))
1001 return;
1002
1003 local_save_flags(irq_flags);
1004 pc = preempt_count();
1005
1006 dsize = __get_data_size(&tk->tp, regs);
1007 size = sizeof(*entry) + tk->tp.size + dsize;
1008
1009 event = trace_event_buffer_lock_reserve(&buffer, trace_file,
1010 call->event.type,
1011 size, irq_flags, pc);
1012 if (!event)
1013 return;
1014
1015 entry = ring_buffer_event_data(event);
1016 entry->ip = (unsigned long)tk->rp.kp.addr;
1017 store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1018
1019 event_trigger_unlock_commit_regs(trace_file, buffer, event,
1020 entry, irq_flags, pc, regs);
1021}
1022
1023static void
1024kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs)
1025{
1026 struct event_file_link *link;
1027
1028 list_for_each_entry_rcu(link, &tk->tp.files, list)
1029 __kprobe_trace_func(tk, regs, link->file);
1030}
1031NOKPROBE_SYMBOL(kprobe_trace_func);
1032
1033/* Kretprobe handler */
1034static nokprobe_inline void
1035__kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1036 struct pt_regs *regs,
1037 struct trace_event_file *trace_file)
1038{
1039 struct kretprobe_trace_entry_head *entry;
1040 struct ring_buffer_event *event;
1041 struct ring_buffer *buffer;
1042 int size, pc, dsize;
1043 unsigned long irq_flags;
1044 struct trace_event_call *call = &tk->tp.call;
1045
1046 WARN_ON(call != trace_file->event_call);
1047
1048 if (trace_trigger_soft_disabled(trace_file))
1049 return;
1050
1051 local_save_flags(irq_flags);
1052 pc = preempt_count();
1053
1054 dsize = __get_data_size(&tk->tp, regs);
1055 size = sizeof(*entry) + tk->tp.size + dsize;
1056
1057 event = trace_event_buffer_lock_reserve(&buffer, trace_file,
1058 call->event.type,
1059 size, irq_flags, pc);
1060 if (!event)
1061 return;
1062
1063 entry = ring_buffer_event_data(event);
1064 entry->func = (unsigned long)tk->rp.kp.addr;
1065 entry->ret_ip = (unsigned long)ri->ret_addr;
1066 store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1067
1068 event_trigger_unlock_commit_regs(trace_file, buffer, event,
1069 entry, irq_flags, pc, regs);
1070}
1071
1072static void
1073kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1074 struct pt_regs *regs)
1075{
1076 struct event_file_link *link;
1077
1078 list_for_each_entry_rcu(link, &tk->tp.files, list)
1079 __kretprobe_trace_func(tk, ri, regs, link->file);
1080}
1081NOKPROBE_SYMBOL(kretprobe_trace_func);
1082
1083/* Event entry printers */
1084static enum print_line_t
1085print_kprobe_event(struct trace_iterator *iter, int flags,
1086 struct trace_event *event)
1087{
1088 struct kprobe_trace_entry_head *field;
1089 struct trace_seq *s = &iter->seq;
1090 struct trace_probe *tp;
1091 u8 *data;
1092 int i;
1093
1094 field = (struct kprobe_trace_entry_head *)iter->ent;
1095 tp = container_of(event, struct trace_probe, call.event);
1096
1097 trace_seq_printf(s, "%s: (", trace_event_name(&tp->call));
1098
1099 if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
1100 goto out;
1101
1102 trace_seq_putc(s, ')');
1103
1104 data = (u8 *)&field[1];
1105 for (i = 0; i < tp->nr_args; i++)
1106 if (!tp->args[i].type->print(s, tp->args[i].name,
1107 data + tp->args[i].offset, field))
1108 goto out;
1109
1110 trace_seq_putc(s, '\n');
1111 out:
1112 return trace_handle_return(s);
1113}
1114
1115static enum print_line_t
1116print_kretprobe_event(struct trace_iterator *iter, int flags,
1117 struct trace_event *event)
1118{
1119 struct kretprobe_trace_entry_head *field;
1120 struct trace_seq *s = &iter->seq;
1121 struct trace_probe *tp;
1122 u8 *data;
1123 int i;
1124
1125 field = (struct kretprobe_trace_entry_head *)iter->ent;
1126 tp = container_of(event, struct trace_probe, call.event);
1127
1128 trace_seq_printf(s, "%s: (", trace_event_name(&tp->call));
1129
1130 if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1131 goto out;
1132
1133 trace_seq_puts(s, " <- ");
1134
1135 if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1136 goto out;
1137
1138 trace_seq_putc(s, ')');
1139
1140 data = (u8 *)&field[1];
1141 for (i = 0; i < tp->nr_args; i++)
1142 if (!tp->args[i].type->print(s, tp->args[i].name,
1143 data + tp->args[i].offset, field))
1144 goto out;
1145
1146 trace_seq_putc(s, '\n');
1147
1148 out:
1149 return trace_handle_return(s);
1150}
1151
1152
1153static int kprobe_event_define_fields(struct trace_event_call *event_call)
1154{
1155 int ret, i;
1156 struct kprobe_trace_entry_head field;
1157 struct trace_kprobe *tk = (struct trace_kprobe *)event_call->data;
1158
1159 DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1160 /* Set argument names as fields */
1161 for (i = 0; i < tk->tp.nr_args; i++) {
1162 struct probe_arg *parg = &tk->tp.args[i];
1163
1164 ret = trace_define_field(event_call, parg->type->fmttype,
1165 parg->name,
1166 sizeof(field) + parg->offset,
1167 parg->type->size,
1168 parg->type->is_signed,
1169 FILTER_OTHER);
1170 if (ret)
1171 return ret;
1172 }
1173 return 0;
1174}
1175
1176static int kretprobe_event_define_fields(struct trace_event_call *event_call)
1177{
1178 int ret, i;
1179 struct kretprobe_trace_entry_head field;
1180 struct trace_kprobe *tk = (struct trace_kprobe *)event_call->data;
1181
1182 DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1183 DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1184 /* Set argument names as fields */
1185 for (i = 0; i < tk->tp.nr_args; i++) {
1186 struct probe_arg *parg = &tk->tp.args[i];
1187
1188 ret = trace_define_field(event_call, parg->type->fmttype,
1189 parg->name,
1190 sizeof(field) + parg->offset,
1191 parg->type->size,
1192 parg->type->is_signed,
1193 FILTER_OTHER);
1194 if (ret)
1195 return ret;
1196 }
1197 return 0;
1198}
1199
1200#ifdef CONFIG_PERF_EVENTS
1201
1202/* Kprobe profile handler */
1203static int
1204kprobe_perf_func(struct trace_kprobe *tk, struct pt_regs *regs)
1205{
1206 struct trace_event_call *call = &tk->tp.call;
1207 struct kprobe_trace_entry_head *entry;
1208 struct hlist_head *head;
1209 int size, __size, dsize;
1210 int rctx;
1211
1212 if (bpf_prog_array_valid(call)) {
1213 unsigned long orig_ip = instruction_pointer(regs);
1214 int ret;
1215
1216 ret = trace_call_bpf(call, regs);
1217
1218 /*
1219 * We need to check and see if we modified the pc of the
1220 * pt_regs, and if so clear the kprobe and return 1 so that we
1221 * don't do the single stepping.
1222 * The ftrace kprobe handler leaves it up to us to re-enable
1223 * preemption here before returning if we've modified the ip.
1224 */
1225 if (orig_ip != instruction_pointer(regs)) {
1226 reset_current_kprobe();
1227 preempt_enable_no_resched();
1228 return 1;
1229 }
1230 if (!ret)
1231 return 0;
1232 }
1233
1234 head = this_cpu_ptr(call->perf_events);
1235 if (hlist_empty(head))
1236 return 0;
1237
1238 dsize = __get_data_size(&tk->tp, regs);
1239 __size = sizeof(*entry) + tk->tp.size + dsize;
1240 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1241 size -= sizeof(u32);
1242
1243 entry = perf_trace_buf_alloc(size, NULL, &rctx);
1244 if (!entry)
1245 return 0;
1246
1247 entry->ip = (unsigned long)tk->rp.kp.addr;
1248 memset(&entry[1], 0, dsize);
1249 store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1250 perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1251 head, NULL);
1252 return 0;
1253}
1254NOKPROBE_SYMBOL(kprobe_perf_func);
1255
1256/* Kretprobe profile handler */
1257static void
1258kretprobe_perf_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1259 struct pt_regs *regs)
1260{
1261 struct trace_event_call *call = &tk->tp.call;
1262 struct kretprobe_trace_entry_head *entry;
1263 struct hlist_head *head;
1264 int size, __size, dsize;
1265 int rctx;
1266
1267 if (bpf_prog_array_valid(call) && !trace_call_bpf(call, regs))
1268 return;
1269
1270 head = this_cpu_ptr(call->perf_events);
1271 if (hlist_empty(head))
1272 return;
1273
1274 dsize = __get_data_size(&tk->tp, regs);
1275 __size = sizeof(*entry) + tk->tp.size + dsize;
1276 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1277 size -= sizeof(u32);
1278
1279 entry = perf_trace_buf_alloc(size, NULL, &rctx);
1280 if (!entry)
1281 return;
1282
1283 entry->func = (unsigned long)tk->rp.kp.addr;
1284 entry->ret_ip = (unsigned long)ri->ret_addr;
1285 store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1286 perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1287 head, NULL);
1288}
1289NOKPROBE_SYMBOL(kretprobe_perf_func);
1290#endif /* CONFIG_PERF_EVENTS */
1291
1292/*
1293 * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex.
1294 *
1295 * kprobe_trace_self_tests_init() does enable_trace_probe/disable_trace_probe
1296 * lockless, but we can't race with this __init function.
1297 */
1298static int kprobe_register(struct trace_event_call *event,
1299 enum trace_reg type, void *data)
1300{
1301 struct trace_kprobe *tk = (struct trace_kprobe *)event->data;
1302 struct trace_event_file *file = data;
1303
1304 switch (type) {
1305 case TRACE_REG_REGISTER:
1306 return enable_trace_kprobe(tk, file);
1307 case TRACE_REG_UNREGISTER:
1308 return disable_trace_kprobe(tk, file);
1309
1310#ifdef CONFIG_PERF_EVENTS
1311 case TRACE_REG_PERF_REGISTER:
1312 return enable_trace_kprobe(tk, NULL);
1313 case TRACE_REG_PERF_UNREGISTER:
1314 return disable_trace_kprobe(tk, NULL);
1315 case TRACE_REG_PERF_OPEN:
1316 case TRACE_REG_PERF_CLOSE:
1317 case TRACE_REG_PERF_ADD:
1318 case TRACE_REG_PERF_DEL:
1319 return 0;
1320#endif
1321 }
1322 return 0;
1323}
1324
1325static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1326{
1327 struct trace_kprobe *tk = container_of(kp, struct trace_kprobe, rp.kp);
1328 int ret = 0;
1329
1330 raw_cpu_inc(*tk->nhit);
1331
1332 if (tk->tp.flags & TP_FLAG_TRACE)
1333 kprobe_trace_func(tk, regs);
1334#ifdef CONFIG_PERF_EVENTS
1335 if (tk->tp.flags & TP_FLAG_PROFILE)
1336 ret = kprobe_perf_func(tk, regs);
1337#endif
1338 return ret;
1339}
1340NOKPROBE_SYMBOL(kprobe_dispatcher);
1341
1342static int
1343kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1344{
1345 struct trace_kprobe *tk = container_of(ri->rp, struct trace_kprobe, rp);
1346
1347 raw_cpu_inc(*tk->nhit);
1348
1349 if (tk->tp.flags & TP_FLAG_TRACE)
1350 kretprobe_trace_func(tk, ri, regs);
1351#ifdef CONFIG_PERF_EVENTS
1352 if (tk->tp.flags & TP_FLAG_PROFILE)
1353 kretprobe_perf_func(tk, ri, regs);
1354#endif
1355 return 0; /* We don't tweek kernel, so just return 0 */
1356}
1357NOKPROBE_SYMBOL(kretprobe_dispatcher);
1358
1359static struct trace_event_functions kretprobe_funcs = {
1360 .trace = print_kretprobe_event
1361};
1362
1363static struct trace_event_functions kprobe_funcs = {
1364 .trace = print_kprobe_event
1365};
1366
1367static inline void init_trace_event_call(struct trace_kprobe *tk,
1368 struct trace_event_call *call)
1369{
1370 INIT_LIST_HEAD(&call->class->fields);
1371 if (trace_kprobe_is_return(tk)) {
1372 call->event.funcs = &kretprobe_funcs;
1373 call->class->define_fields = kretprobe_event_define_fields;
1374 } else {
1375 call->event.funcs = &kprobe_funcs;
1376 call->class->define_fields = kprobe_event_define_fields;
1377 }
1378
1379 call->flags = TRACE_EVENT_FL_KPROBE;
1380 call->class->reg = kprobe_register;
1381 call->data = tk;
1382}
1383
1384static int register_kprobe_event(struct trace_kprobe *tk)
1385{
1386 struct trace_event_call *call = &tk->tp.call;
1387 int ret = 0;
1388
1389 init_trace_event_call(tk, call);
1390
1391 if (set_print_fmt(&tk->tp, trace_kprobe_is_return(tk)) < 0)
1392 return -ENOMEM;
1393 ret = register_trace_event(&call->event);
1394 if (!ret) {
1395 kfree(call->print_fmt);
1396 return -ENODEV;
1397 }
1398 ret = trace_add_event_call(call);
1399 if (ret) {
1400 pr_info("Failed to register kprobe event: %s\n",
1401 trace_event_name(call));
1402 kfree(call->print_fmt);
1403 unregister_trace_event(&call->event);
1404 }
1405 return ret;
1406}
1407
1408static int unregister_kprobe_event(struct trace_kprobe *tk)
1409{
1410 int ret;
1411
1412 /* tp->event is unregistered in trace_remove_event_call() */
1413 ret = trace_remove_event_call(&tk->tp.call);
1414 if (!ret)
1415 kfree(tk->tp.call.print_fmt);
1416 return ret;
1417}
1418
1419#ifdef CONFIG_PERF_EVENTS
1420/* create a trace_kprobe, but don't add it to global lists */
1421struct trace_event_call *
1422create_local_trace_kprobe(char *func, void *addr, unsigned long offs,
1423 bool is_return)
1424{
1425 struct trace_kprobe *tk;
1426 int ret;
1427 char *event;
1428
1429 /*
1430 * local trace_kprobes are not added to probe_list, so they are never
1431 * searched in find_trace_kprobe(). Therefore, there is no concern of
1432 * duplicated name here.
1433 */
1434 event = func ? func : "DUMMY_EVENT";
1435
1436 tk = alloc_trace_kprobe(KPROBE_EVENT_SYSTEM, event, (void *)addr, func,
1437 offs, 0 /* maxactive */, 0 /* nargs */,
1438 is_return);
1439
1440 if (IS_ERR(tk)) {
1441 pr_info("Failed to allocate trace_probe.(%d)\n",
1442 (int)PTR_ERR(tk));
1443 return ERR_CAST(tk);
1444 }
1445
1446 init_trace_event_call(tk, &tk->tp.call);
1447
1448 if (set_print_fmt(&tk->tp, trace_kprobe_is_return(tk)) < 0) {
1449 ret = -ENOMEM;
1450 goto error;
1451 }
1452
1453 ret = __register_trace_kprobe(tk);
1454 if (ret < 0)
1455 goto error;
1456
1457 return &tk->tp.call;
1458error:
1459 free_trace_kprobe(tk);
1460 return ERR_PTR(ret);
1461}
1462
1463void destroy_local_trace_kprobe(struct trace_event_call *event_call)
1464{
1465 struct trace_kprobe *tk;
1466
1467 tk = container_of(event_call, struct trace_kprobe, tp.call);
1468
1469 if (trace_probe_is_enabled(&tk->tp)) {
1470 WARN_ON(1);
1471 return;
1472 }
1473
1474 __unregister_trace_kprobe(tk);
1475 free_trace_kprobe(tk);
1476}
1477#endif /* CONFIG_PERF_EVENTS */
1478
1479/* Make a tracefs interface for controlling probe points */
1480static __init int init_kprobe_trace(void)
1481{
1482 struct dentry *d_tracer;
1483 struct dentry *entry;
1484
1485 if (register_module_notifier(&trace_kprobe_module_nb))
1486 return -EINVAL;
1487
1488 d_tracer = tracing_init_dentry();
1489 if (IS_ERR(d_tracer))
1490 return 0;
1491
1492 entry = tracefs_create_file("kprobe_events", 0644, d_tracer,
1493 NULL, &kprobe_events_ops);
1494
1495 /* Event list interface */
1496 if (!entry)
1497 pr_warn("Could not create tracefs 'kprobe_events' entry\n");
1498
1499 /* Profile interface */
1500 entry = tracefs_create_file("kprobe_profile", 0444, d_tracer,
1501 NULL, &kprobe_profile_ops);
1502
1503 if (!entry)
1504 pr_warn("Could not create tracefs 'kprobe_profile' entry\n");
1505 return 0;
1506}
1507fs_initcall(init_kprobe_trace);
1508
1509
1510#ifdef CONFIG_FTRACE_STARTUP_TEST
1511/*
1512 * The "__used" keeps gcc from removing the function symbol
1513 * from the kallsyms table. 'noinline' makes sure that there
1514 * isn't an inlined version used by the test method below
1515 */
1516static __used __init noinline int
1517kprobe_trace_selftest_target(int a1, int a2, int a3, int a4, int a5, int a6)
1518{
1519 return a1 + a2 + a3 + a4 + a5 + a6;
1520}
1521
1522static __init struct trace_event_file *
1523find_trace_probe_file(struct trace_kprobe *tk, struct trace_array *tr)
1524{
1525 struct trace_event_file *file;
1526
1527 list_for_each_entry(file, &tr->events, list)
1528 if (file->event_call == &tk->tp.call)
1529 return file;
1530
1531 return NULL;
1532}
1533
1534/*
1535 * Nobody but us can call enable_trace_kprobe/disable_trace_kprobe at this
1536 * stage, we can do this lockless.
1537 */
1538static __init int kprobe_trace_self_tests_init(void)
1539{
1540 int ret, warn = 0;
1541 int (*target)(int, int, int, int, int, int);
1542 struct trace_kprobe *tk;
1543 struct trace_event_file *file;
1544
1545 if (tracing_is_disabled())
1546 return -ENODEV;
1547
1548 target = kprobe_trace_selftest_target;
1549
1550 pr_info("Testing kprobe tracing: ");
1551
1552 ret = trace_run_command("p:testprobe kprobe_trace_selftest_target "
1553 "$stack $stack0 +0($stack)",
1554 create_trace_kprobe);
1555 if (WARN_ON_ONCE(ret)) {
1556 pr_warn("error on probing function entry.\n");
1557 warn++;
1558 } else {
1559 /* Enable trace point */
1560 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1561 if (WARN_ON_ONCE(tk == NULL)) {
1562 pr_warn("error on getting new probe.\n");
1563 warn++;
1564 } else {
1565 file = find_trace_probe_file(tk, top_trace_array());
1566 if (WARN_ON_ONCE(file == NULL)) {
1567 pr_warn("error on getting probe file.\n");
1568 warn++;
1569 } else
1570 enable_trace_kprobe(tk, file);
1571 }
1572 }
1573
1574 ret = trace_run_command("r:testprobe2 kprobe_trace_selftest_target "
1575 "$retval", create_trace_kprobe);
1576 if (WARN_ON_ONCE(ret)) {
1577 pr_warn("error on probing function return.\n");
1578 warn++;
1579 } else {
1580 /* Enable trace point */
1581 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1582 if (WARN_ON_ONCE(tk == NULL)) {
1583 pr_warn("error on getting 2nd new probe.\n");
1584 warn++;
1585 } else {
1586 file = find_trace_probe_file(tk, top_trace_array());
1587 if (WARN_ON_ONCE(file == NULL)) {
1588 pr_warn("error on getting probe file.\n");
1589 warn++;
1590 } else
1591 enable_trace_kprobe(tk, file);
1592 }
1593 }
1594
1595 if (warn)
1596 goto end;
1597
1598 ret = target(1, 2, 3, 4, 5, 6);
1599
1600 /*
1601 * Not expecting an error here, the check is only to prevent the
1602 * optimizer from removing the call to target() as otherwise there
1603 * are no side-effects and the call is never performed.
1604 */
1605 if (ret != 21)
1606 warn++;
1607
1608 /* Disable trace points before removing it */
1609 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1610 if (WARN_ON_ONCE(tk == NULL)) {
1611 pr_warn("error on getting test probe.\n");
1612 warn++;
1613 } else {
1614 if (trace_kprobe_nhit(tk) != 1) {
1615 pr_warn("incorrect number of testprobe hits\n");
1616 warn++;
1617 }
1618
1619 file = find_trace_probe_file(tk, top_trace_array());
1620 if (WARN_ON_ONCE(file == NULL)) {
1621 pr_warn("error on getting probe file.\n");
1622 warn++;
1623 } else
1624 disable_trace_kprobe(tk, file);
1625 }
1626
1627 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1628 if (WARN_ON_ONCE(tk == NULL)) {
1629 pr_warn("error on getting 2nd test probe.\n");
1630 warn++;
1631 } else {
1632 if (trace_kprobe_nhit(tk) != 1) {
1633 pr_warn("incorrect number of testprobe2 hits\n");
1634 warn++;
1635 }
1636
1637 file = find_trace_probe_file(tk, top_trace_array());
1638 if (WARN_ON_ONCE(file == NULL)) {
1639 pr_warn("error on getting probe file.\n");
1640 warn++;
1641 } else
1642 disable_trace_kprobe(tk, file);
1643 }
1644
1645 ret = trace_run_command("-:testprobe", create_trace_kprobe);
1646 if (WARN_ON_ONCE(ret)) {
1647 pr_warn("error on deleting a probe.\n");
1648 warn++;
1649 }
1650
1651 ret = trace_run_command("-:testprobe2", create_trace_kprobe);
1652 if (WARN_ON_ONCE(ret)) {
1653 pr_warn("error on deleting a probe.\n");
1654 warn++;
1655 }
1656
1657end:
1658 release_all_trace_kprobes();
1659 /*
1660 * Wait for the optimizer work to finish. Otherwise it might fiddle
1661 * with probes in already freed __init text.
1662 */
1663 wait_for_kprobe_optimizer();
1664 if (warn)
1665 pr_cont("NG: Some tests are failed. Please check them.\n");
1666 else
1667 pr_cont("OK\n");
1668 return 0;
1669}
1670
1671late_initcall(kprobe_trace_self_tests_init);
1672
1673#endif
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Kprobes-based tracing events
4 *
5 * Created by Masami Hiramatsu <mhiramat@redhat.com>
6 *
7 */
8#define pr_fmt(fmt) "trace_kprobe: " fmt
9
10#include <linux/security.h>
11#include <linux/module.h>
12#include <linux/uaccess.h>
13#include <linux/rculist.h>
14#include <linux/error-injection.h>
15
16#include <asm/setup.h> /* for COMMAND_LINE_SIZE */
17
18#include "trace_dynevent.h"
19#include "trace_kprobe_selftest.h"
20#include "trace_probe.h"
21#include "trace_probe_tmpl.h"
22
23#define KPROBE_EVENT_SYSTEM "kprobes"
24#define KRETPROBE_MAXACTIVE_MAX 4096
25
26/* Kprobe early definition from command line */
27static char kprobe_boot_events_buf[COMMAND_LINE_SIZE] __initdata;
28static bool kprobe_boot_events_enabled __initdata;
29
30static int __init set_kprobe_boot_events(char *str)
31{
32 strlcpy(kprobe_boot_events_buf, str, COMMAND_LINE_SIZE);
33 return 0;
34}
35__setup("kprobe_event=", set_kprobe_boot_events);
36
37static int trace_kprobe_create(int argc, const char **argv);
38static int trace_kprobe_show(struct seq_file *m, struct dyn_event *ev);
39static int trace_kprobe_release(struct dyn_event *ev);
40static bool trace_kprobe_is_busy(struct dyn_event *ev);
41static bool trace_kprobe_match(const char *system, const char *event,
42 int argc, const char **argv, struct dyn_event *ev);
43
44static struct dyn_event_operations trace_kprobe_ops = {
45 .create = trace_kprobe_create,
46 .show = trace_kprobe_show,
47 .is_busy = trace_kprobe_is_busy,
48 .free = trace_kprobe_release,
49 .match = trace_kprobe_match,
50};
51
52/*
53 * Kprobe event core functions
54 */
55struct trace_kprobe {
56 struct dyn_event devent;
57 struct kretprobe rp; /* Use rp.kp for kprobe use */
58 unsigned long __percpu *nhit;
59 const char *symbol; /* symbol name */
60 struct trace_probe tp;
61};
62
63static bool is_trace_kprobe(struct dyn_event *ev)
64{
65 return ev->ops == &trace_kprobe_ops;
66}
67
68static struct trace_kprobe *to_trace_kprobe(struct dyn_event *ev)
69{
70 return container_of(ev, struct trace_kprobe, devent);
71}
72
73/**
74 * for_each_trace_kprobe - iterate over the trace_kprobe list
75 * @pos: the struct trace_kprobe * for each entry
76 * @dpos: the struct dyn_event * to use as a loop cursor
77 */
78#define for_each_trace_kprobe(pos, dpos) \
79 for_each_dyn_event(dpos) \
80 if (is_trace_kprobe(dpos) && (pos = to_trace_kprobe(dpos)))
81
82#define SIZEOF_TRACE_KPROBE(n) \
83 (offsetof(struct trace_kprobe, tp.args) + \
84 (sizeof(struct probe_arg) * (n)))
85
86static nokprobe_inline bool trace_kprobe_is_return(struct trace_kprobe *tk)
87{
88 return tk->rp.handler != NULL;
89}
90
91static nokprobe_inline const char *trace_kprobe_symbol(struct trace_kprobe *tk)
92{
93 return tk->symbol ? tk->symbol : "unknown";
94}
95
96static nokprobe_inline unsigned long trace_kprobe_offset(struct trace_kprobe *tk)
97{
98 return tk->rp.kp.offset;
99}
100
101static nokprobe_inline bool trace_kprobe_has_gone(struct trace_kprobe *tk)
102{
103 return !!(kprobe_gone(&tk->rp.kp));
104}
105
106static nokprobe_inline bool trace_kprobe_within_module(struct trace_kprobe *tk,
107 struct module *mod)
108{
109 int len = strlen(mod->name);
110 const char *name = trace_kprobe_symbol(tk);
111 return strncmp(mod->name, name, len) == 0 && name[len] == ':';
112}
113
114static nokprobe_inline bool trace_kprobe_module_exist(struct trace_kprobe *tk)
115{
116 char *p;
117 bool ret;
118
119 if (!tk->symbol)
120 return false;
121 p = strchr(tk->symbol, ':');
122 if (!p)
123 return true;
124 *p = '\0';
125 mutex_lock(&module_mutex);
126 ret = !!find_module(tk->symbol);
127 mutex_unlock(&module_mutex);
128 *p = ':';
129
130 return ret;
131}
132
133static bool trace_kprobe_is_busy(struct dyn_event *ev)
134{
135 struct trace_kprobe *tk = to_trace_kprobe(ev);
136
137 return trace_probe_is_enabled(&tk->tp);
138}
139
140static bool trace_kprobe_match_command_head(struct trace_kprobe *tk,
141 int argc, const char **argv)
142{
143 char buf[MAX_ARGSTR_LEN + 1];
144
145 if (!argc)
146 return true;
147
148 if (!tk->symbol)
149 snprintf(buf, sizeof(buf), "0x%p", tk->rp.kp.addr);
150 else if (tk->rp.kp.offset)
151 snprintf(buf, sizeof(buf), "%s+%u",
152 trace_kprobe_symbol(tk), tk->rp.kp.offset);
153 else
154 snprintf(buf, sizeof(buf), "%s", trace_kprobe_symbol(tk));
155 if (strcmp(buf, argv[0]))
156 return false;
157 argc--; argv++;
158
159 return trace_probe_match_command_args(&tk->tp, argc, argv);
160}
161
162static bool trace_kprobe_match(const char *system, const char *event,
163 int argc, const char **argv, struct dyn_event *ev)
164{
165 struct trace_kprobe *tk = to_trace_kprobe(ev);
166
167 return strcmp(trace_probe_name(&tk->tp), event) == 0 &&
168 (!system || strcmp(trace_probe_group_name(&tk->tp), system) == 0) &&
169 trace_kprobe_match_command_head(tk, argc, argv);
170}
171
172static nokprobe_inline unsigned long trace_kprobe_nhit(struct trace_kprobe *tk)
173{
174 unsigned long nhit = 0;
175 int cpu;
176
177 for_each_possible_cpu(cpu)
178 nhit += *per_cpu_ptr(tk->nhit, cpu);
179
180 return nhit;
181}
182
183static nokprobe_inline bool trace_kprobe_is_registered(struct trace_kprobe *tk)
184{
185 return !(list_empty(&tk->rp.kp.list) &&
186 hlist_unhashed(&tk->rp.kp.hlist));
187}
188
189/* Return 0 if it fails to find the symbol address */
190static nokprobe_inline
191unsigned long trace_kprobe_address(struct trace_kprobe *tk)
192{
193 unsigned long addr;
194
195 if (tk->symbol) {
196 addr = (unsigned long)
197 kallsyms_lookup_name(trace_kprobe_symbol(tk));
198 if (addr)
199 addr += tk->rp.kp.offset;
200 } else {
201 addr = (unsigned long)tk->rp.kp.addr;
202 }
203 return addr;
204}
205
206static nokprobe_inline struct trace_kprobe *
207trace_kprobe_primary_from_call(struct trace_event_call *call)
208{
209 struct trace_probe *tp;
210
211 tp = trace_probe_primary_from_call(call);
212 if (WARN_ON_ONCE(!tp))
213 return NULL;
214
215 return container_of(tp, struct trace_kprobe, tp);
216}
217
218bool trace_kprobe_on_func_entry(struct trace_event_call *call)
219{
220 struct trace_kprobe *tk = trace_kprobe_primary_from_call(call);
221
222 return tk ? kprobe_on_func_entry(tk->rp.kp.addr,
223 tk->rp.kp.addr ? NULL : tk->rp.kp.symbol_name,
224 tk->rp.kp.addr ? 0 : tk->rp.kp.offset) : false;
225}
226
227bool trace_kprobe_error_injectable(struct trace_event_call *call)
228{
229 struct trace_kprobe *tk = trace_kprobe_primary_from_call(call);
230
231 return tk ? within_error_injection_list(trace_kprobe_address(tk)) :
232 false;
233}
234
235static int register_kprobe_event(struct trace_kprobe *tk);
236static int unregister_kprobe_event(struct trace_kprobe *tk);
237
238static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
239static int kretprobe_dispatcher(struct kretprobe_instance *ri,
240 struct pt_regs *regs);
241
242static void free_trace_kprobe(struct trace_kprobe *tk)
243{
244 if (tk) {
245 trace_probe_cleanup(&tk->tp);
246 kfree(tk->symbol);
247 free_percpu(tk->nhit);
248 kfree(tk);
249 }
250}
251
252/*
253 * Allocate new trace_probe and initialize it (including kprobes).
254 */
255static struct trace_kprobe *alloc_trace_kprobe(const char *group,
256 const char *event,
257 void *addr,
258 const char *symbol,
259 unsigned long offs,
260 int maxactive,
261 int nargs, bool is_return)
262{
263 struct trace_kprobe *tk;
264 int ret = -ENOMEM;
265
266 tk = kzalloc(SIZEOF_TRACE_KPROBE(nargs), GFP_KERNEL);
267 if (!tk)
268 return ERR_PTR(ret);
269
270 tk->nhit = alloc_percpu(unsigned long);
271 if (!tk->nhit)
272 goto error;
273
274 if (symbol) {
275 tk->symbol = kstrdup(symbol, GFP_KERNEL);
276 if (!tk->symbol)
277 goto error;
278 tk->rp.kp.symbol_name = tk->symbol;
279 tk->rp.kp.offset = offs;
280 } else
281 tk->rp.kp.addr = addr;
282
283 if (is_return)
284 tk->rp.handler = kretprobe_dispatcher;
285 else
286 tk->rp.kp.pre_handler = kprobe_dispatcher;
287
288 tk->rp.maxactive = maxactive;
289 INIT_HLIST_NODE(&tk->rp.kp.hlist);
290 INIT_LIST_HEAD(&tk->rp.kp.list);
291
292 ret = trace_probe_init(&tk->tp, event, group, false);
293 if (ret < 0)
294 goto error;
295
296 dyn_event_init(&tk->devent, &trace_kprobe_ops);
297 return tk;
298error:
299 free_trace_kprobe(tk);
300 return ERR_PTR(ret);
301}
302
303static struct trace_kprobe *find_trace_kprobe(const char *event,
304 const char *group)
305{
306 struct dyn_event *pos;
307 struct trace_kprobe *tk;
308
309 for_each_trace_kprobe(tk, pos)
310 if (strcmp(trace_probe_name(&tk->tp), event) == 0 &&
311 strcmp(trace_probe_group_name(&tk->tp), group) == 0)
312 return tk;
313 return NULL;
314}
315
316static inline int __enable_trace_kprobe(struct trace_kprobe *tk)
317{
318 int ret = 0;
319
320 if (trace_kprobe_is_registered(tk) && !trace_kprobe_has_gone(tk)) {
321 if (trace_kprobe_is_return(tk))
322 ret = enable_kretprobe(&tk->rp);
323 else
324 ret = enable_kprobe(&tk->rp.kp);
325 }
326
327 return ret;
328}
329
330static void __disable_trace_kprobe(struct trace_probe *tp)
331{
332 struct trace_probe *pos;
333 struct trace_kprobe *tk;
334
335 list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
336 tk = container_of(pos, struct trace_kprobe, tp);
337 if (!trace_kprobe_is_registered(tk))
338 continue;
339 if (trace_kprobe_is_return(tk))
340 disable_kretprobe(&tk->rp);
341 else
342 disable_kprobe(&tk->rp.kp);
343 }
344}
345
346/*
347 * Enable trace_probe
348 * if the file is NULL, enable "perf" handler, or enable "trace" handler.
349 */
350static int enable_trace_kprobe(struct trace_event_call *call,
351 struct trace_event_file *file)
352{
353 struct trace_probe *pos, *tp;
354 struct trace_kprobe *tk;
355 bool enabled;
356 int ret = 0;
357
358 tp = trace_probe_primary_from_call(call);
359 if (WARN_ON_ONCE(!tp))
360 return -ENODEV;
361 enabled = trace_probe_is_enabled(tp);
362
363 /* This also changes "enabled" state */
364 if (file) {
365 ret = trace_probe_add_file(tp, file);
366 if (ret)
367 return ret;
368 } else
369 trace_probe_set_flag(tp, TP_FLAG_PROFILE);
370
371 if (enabled)
372 return 0;
373
374 list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
375 tk = container_of(pos, struct trace_kprobe, tp);
376 if (trace_kprobe_has_gone(tk))
377 continue;
378 ret = __enable_trace_kprobe(tk);
379 if (ret)
380 break;
381 enabled = true;
382 }
383
384 if (ret) {
385 /* Failed to enable one of them. Roll back all */
386 if (enabled)
387 __disable_trace_kprobe(tp);
388 if (file)
389 trace_probe_remove_file(tp, file);
390 else
391 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
392 }
393
394 return ret;
395}
396
397/*
398 * Disable trace_probe
399 * if the file is NULL, disable "perf" handler, or disable "trace" handler.
400 */
401static int disable_trace_kprobe(struct trace_event_call *call,
402 struct trace_event_file *file)
403{
404 struct trace_probe *tp;
405
406 tp = trace_probe_primary_from_call(call);
407 if (WARN_ON_ONCE(!tp))
408 return -ENODEV;
409
410 if (file) {
411 if (!trace_probe_get_file_link(tp, file))
412 return -ENOENT;
413 if (!trace_probe_has_single_file(tp))
414 goto out;
415 trace_probe_clear_flag(tp, TP_FLAG_TRACE);
416 } else
417 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
418
419 if (!trace_probe_is_enabled(tp))
420 __disable_trace_kprobe(tp);
421
422 out:
423 if (file)
424 /*
425 * Synchronization is done in below function. For perf event,
426 * file == NULL and perf_trace_event_unreg() calls
427 * tracepoint_synchronize_unregister() to ensure synchronize
428 * event. We don't need to care about it.
429 */
430 trace_probe_remove_file(tp, file);
431
432 return 0;
433}
434
435#if defined(CONFIG_KPROBES_ON_FTRACE) && \
436 !defined(CONFIG_KPROBE_EVENTS_ON_NOTRACE)
437static bool __within_notrace_func(unsigned long addr)
438{
439 unsigned long offset, size;
440
441 if (!addr || !kallsyms_lookup_size_offset(addr, &size, &offset))
442 return false;
443
444 /* Get the entry address of the target function */
445 addr -= offset;
446
447 /*
448 * Since ftrace_location_range() does inclusive range check, we need
449 * to subtract 1 byte from the end address.
450 */
451 return !ftrace_location_range(addr, addr + size - 1);
452}
453
454static bool within_notrace_func(struct trace_kprobe *tk)
455{
456 unsigned long addr = trace_kprobe_address(tk);
457 char symname[KSYM_NAME_LEN], *p;
458
459 if (!__within_notrace_func(addr))
460 return false;
461
462 /* Check if the address is on a suffixed-symbol */
463 if (!lookup_symbol_name(addr, symname)) {
464 p = strchr(symname, '.');
465 if (!p)
466 return true;
467 *p = '\0';
468 addr = (unsigned long)kprobe_lookup_name(symname, 0);
469 if (addr)
470 return __within_notrace_func(addr);
471 }
472
473 return true;
474}
475#else
476#define within_notrace_func(tk) (false)
477#endif
478
479/* Internal register function - just handle k*probes and flags */
480static int __register_trace_kprobe(struct trace_kprobe *tk)
481{
482 int i, ret;
483
484 ret = security_locked_down(LOCKDOWN_KPROBES);
485 if (ret)
486 return ret;
487
488 if (trace_kprobe_is_registered(tk))
489 return -EINVAL;
490
491 if (within_notrace_func(tk)) {
492 pr_warn("Could not probe notrace function %s\n",
493 trace_kprobe_symbol(tk));
494 return -EINVAL;
495 }
496
497 for (i = 0; i < tk->tp.nr_args; i++) {
498 ret = traceprobe_update_arg(&tk->tp.args[i]);
499 if (ret)
500 return ret;
501 }
502
503 /* Set/clear disabled flag according to tp->flag */
504 if (trace_probe_is_enabled(&tk->tp))
505 tk->rp.kp.flags &= ~KPROBE_FLAG_DISABLED;
506 else
507 tk->rp.kp.flags |= KPROBE_FLAG_DISABLED;
508
509 if (trace_kprobe_is_return(tk))
510 ret = register_kretprobe(&tk->rp);
511 else
512 ret = register_kprobe(&tk->rp.kp);
513
514 return ret;
515}
516
517/* Internal unregister function - just handle k*probes and flags */
518static void __unregister_trace_kprobe(struct trace_kprobe *tk)
519{
520 if (trace_kprobe_is_registered(tk)) {
521 if (trace_kprobe_is_return(tk))
522 unregister_kretprobe(&tk->rp);
523 else
524 unregister_kprobe(&tk->rp.kp);
525 /* Cleanup kprobe for reuse and mark it unregistered */
526 INIT_HLIST_NODE(&tk->rp.kp.hlist);
527 INIT_LIST_HEAD(&tk->rp.kp.list);
528 if (tk->rp.kp.symbol_name)
529 tk->rp.kp.addr = NULL;
530 }
531}
532
533/* Unregister a trace_probe and probe_event */
534static int unregister_trace_kprobe(struct trace_kprobe *tk)
535{
536 /* If other probes are on the event, just unregister kprobe */
537 if (trace_probe_has_sibling(&tk->tp))
538 goto unreg;
539
540 /* Enabled event can not be unregistered */
541 if (trace_probe_is_enabled(&tk->tp))
542 return -EBUSY;
543
544 /* Will fail if probe is being used by ftrace or perf */
545 if (unregister_kprobe_event(tk))
546 return -EBUSY;
547
548unreg:
549 __unregister_trace_kprobe(tk);
550 dyn_event_remove(&tk->devent);
551 trace_probe_unlink(&tk->tp);
552
553 return 0;
554}
555
556static bool trace_kprobe_has_same_kprobe(struct trace_kprobe *orig,
557 struct trace_kprobe *comp)
558{
559 struct trace_probe_event *tpe = orig->tp.event;
560 struct trace_probe *pos;
561 int i;
562
563 list_for_each_entry(pos, &tpe->probes, list) {
564 orig = container_of(pos, struct trace_kprobe, tp);
565 if (strcmp(trace_kprobe_symbol(orig),
566 trace_kprobe_symbol(comp)) ||
567 trace_kprobe_offset(orig) != trace_kprobe_offset(comp))
568 continue;
569
570 /*
571 * trace_probe_compare_arg_type() ensured that nr_args and
572 * each argument name and type are same. Let's compare comm.
573 */
574 for (i = 0; i < orig->tp.nr_args; i++) {
575 if (strcmp(orig->tp.args[i].comm,
576 comp->tp.args[i].comm))
577 break;
578 }
579
580 if (i == orig->tp.nr_args)
581 return true;
582 }
583
584 return false;
585}
586
587static int append_trace_kprobe(struct trace_kprobe *tk, struct trace_kprobe *to)
588{
589 int ret;
590
591 ret = trace_probe_compare_arg_type(&tk->tp, &to->tp);
592 if (ret) {
593 /* Note that argument starts index = 2 */
594 trace_probe_log_set_index(ret + 1);
595 trace_probe_log_err(0, DIFF_ARG_TYPE);
596 return -EEXIST;
597 }
598 if (trace_kprobe_has_same_kprobe(to, tk)) {
599 trace_probe_log_set_index(0);
600 trace_probe_log_err(0, SAME_PROBE);
601 return -EEXIST;
602 }
603
604 /* Append to existing event */
605 ret = trace_probe_append(&tk->tp, &to->tp);
606 if (ret)
607 return ret;
608
609 /* Register k*probe */
610 ret = __register_trace_kprobe(tk);
611 if (ret == -ENOENT && !trace_kprobe_module_exist(tk)) {
612 pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
613 ret = 0;
614 }
615
616 if (ret)
617 trace_probe_unlink(&tk->tp);
618 else
619 dyn_event_add(&tk->devent);
620
621 return ret;
622}
623
624/* Register a trace_probe and probe_event */
625static int register_trace_kprobe(struct trace_kprobe *tk)
626{
627 struct trace_kprobe *old_tk;
628 int ret;
629
630 mutex_lock(&event_mutex);
631
632 old_tk = find_trace_kprobe(trace_probe_name(&tk->tp),
633 trace_probe_group_name(&tk->tp));
634 if (old_tk) {
635 if (trace_kprobe_is_return(tk) != trace_kprobe_is_return(old_tk)) {
636 trace_probe_log_set_index(0);
637 trace_probe_log_err(0, DIFF_PROBE_TYPE);
638 ret = -EEXIST;
639 } else {
640 ret = append_trace_kprobe(tk, old_tk);
641 }
642 goto end;
643 }
644
645 /* Register new event */
646 ret = register_kprobe_event(tk);
647 if (ret) {
648 pr_warn("Failed to register probe event(%d)\n", ret);
649 goto end;
650 }
651
652 /* Register k*probe */
653 ret = __register_trace_kprobe(tk);
654 if (ret == -ENOENT && !trace_kprobe_module_exist(tk)) {
655 pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
656 ret = 0;
657 }
658
659 if (ret < 0)
660 unregister_kprobe_event(tk);
661 else
662 dyn_event_add(&tk->devent);
663
664end:
665 mutex_unlock(&event_mutex);
666 return ret;
667}
668
669/* Module notifier call back, checking event on the module */
670static int trace_kprobe_module_callback(struct notifier_block *nb,
671 unsigned long val, void *data)
672{
673 struct module *mod = data;
674 struct dyn_event *pos;
675 struct trace_kprobe *tk;
676 int ret;
677
678 if (val != MODULE_STATE_COMING)
679 return NOTIFY_DONE;
680
681 /* Update probes on coming module */
682 mutex_lock(&event_mutex);
683 for_each_trace_kprobe(tk, pos) {
684 if (trace_kprobe_within_module(tk, mod)) {
685 /* Don't need to check busy - this should have gone. */
686 __unregister_trace_kprobe(tk);
687 ret = __register_trace_kprobe(tk);
688 if (ret)
689 pr_warn("Failed to re-register probe %s on %s: %d\n",
690 trace_probe_name(&tk->tp),
691 mod->name, ret);
692 }
693 }
694 mutex_unlock(&event_mutex);
695
696 return NOTIFY_DONE;
697}
698
699static struct notifier_block trace_kprobe_module_nb = {
700 .notifier_call = trace_kprobe_module_callback,
701 .priority = 1 /* Invoked after kprobe module callback */
702};
703
704/* Convert certain expected symbols into '_' when generating event names */
705static inline void sanitize_event_name(char *name)
706{
707 while (*name++ != '\0')
708 if (*name == ':' || *name == '.')
709 *name = '_';
710}
711
712static int trace_kprobe_create(int argc, const char *argv[])
713{
714 /*
715 * Argument syntax:
716 * - Add kprobe:
717 * p[:[GRP/]EVENT] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
718 * - Add kretprobe:
719 * r[MAXACTIVE][:[GRP/]EVENT] [MOD:]KSYM[+0] [FETCHARGS]
720 * Fetch args:
721 * $retval : fetch return value
722 * $stack : fetch stack address
723 * $stackN : fetch Nth of stack (N:0-)
724 * $comm : fetch current task comm
725 * @ADDR : fetch memory at ADDR (ADDR should be in kernel)
726 * @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
727 * %REG : fetch register REG
728 * Dereferencing memory fetch:
729 * +|-offs(ARG) : fetch memory at ARG +|- offs address.
730 * Alias name of args:
731 * NAME=FETCHARG : set NAME as alias of FETCHARG.
732 * Type of args:
733 * FETCHARG:TYPE : use TYPE instead of unsigned long.
734 */
735 struct trace_kprobe *tk = NULL;
736 int i, len, ret = 0;
737 bool is_return = false;
738 char *symbol = NULL, *tmp = NULL;
739 const char *event = NULL, *group = KPROBE_EVENT_SYSTEM;
740 int maxactive = 0;
741 long offset = 0;
742 void *addr = NULL;
743 char buf[MAX_EVENT_NAME_LEN];
744 unsigned int flags = TPARG_FL_KERNEL;
745
746 switch (argv[0][0]) {
747 case 'r':
748 is_return = true;
749 flags |= TPARG_FL_RETURN;
750 break;
751 case 'p':
752 break;
753 default:
754 return -ECANCELED;
755 }
756 if (argc < 2)
757 return -ECANCELED;
758
759 trace_probe_log_init("trace_kprobe", argc, argv);
760
761 event = strchr(&argv[0][1], ':');
762 if (event)
763 event++;
764
765 if (isdigit(argv[0][1])) {
766 if (!is_return) {
767 trace_probe_log_err(1, MAXACT_NO_KPROBE);
768 goto parse_error;
769 }
770 if (event)
771 len = event - &argv[0][1] - 1;
772 else
773 len = strlen(&argv[0][1]);
774 if (len > MAX_EVENT_NAME_LEN - 1) {
775 trace_probe_log_err(1, BAD_MAXACT);
776 goto parse_error;
777 }
778 memcpy(buf, &argv[0][1], len);
779 buf[len] = '\0';
780 ret = kstrtouint(buf, 0, &maxactive);
781 if (ret || !maxactive) {
782 trace_probe_log_err(1, BAD_MAXACT);
783 goto parse_error;
784 }
785 /* kretprobes instances are iterated over via a list. The
786 * maximum should stay reasonable.
787 */
788 if (maxactive > KRETPROBE_MAXACTIVE_MAX) {
789 trace_probe_log_err(1, MAXACT_TOO_BIG);
790 goto parse_error;
791 }
792 }
793
794 /* try to parse an address. if that fails, try to read the
795 * input as a symbol. */
796 if (kstrtoul(argv[1], 0, (unsigned long *)&addr)) {
797 trace_probe_log_set_index(1);
798 /* Check whether uprobe event specified */
799 if (strchr(argv[1], '/') && strchr(argv[1], ':')) {
800 ret = -ECANCELED;
801 goto error;
802 }
803 /* a symbol specified */
804 symbol = kstrdup(argv[1], GFP_KERNEL);
805 if (!symbol)
806 return -ENOMEM;
807 /* TODO: support .init module functions */
808 ret = traceprobe_split_symbol_offset(symbol, &offset);
809 if (ret || offset < 0 || offset > UINT_MAX) {
810 trace_probe_log_err(0, BAD_PROBE_ADDR);
811 goto parse_error;
812 }
813 if (kprobe_on_func_entry(NULL, symbol, offset))
814 flags |= TPARG_FL_FENTRY;
815 if (offset && is_return && !(flags & TPARG_FL_FENTRY)) {
816 trace_probe_log_err(0, BAD_RETPROBE);
817 goto parse_error;
818 }
819 }
820
821 trace_probe_log_set_index(0);
822 if (event) {
823 ret = traceprobe_parse_event_name(&event, &group, buf,
824 event - argv[0]);
825 if (ret)
826 goto parse_error;
827 } else {
828 /* Make a new event name */
829 if (symbol)
830 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
831 is_return ? 'r' : 'p', symbol, offset);
832 else
833 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
834 is_return ? 'r' : 'p', addr);
835 sanitize_event_name(buf);
836 event = buf;
837 }
838
839 /* setup a probe */
840 tk = alloc_trace_kprobe(group, event, addr, symbol, offset, maxactive,
841 argc - 2, is_return);
842 if (IS_ERR(tk)) {
843 ret = PTR_ERR(tk);
844 /* This must return -ENOMEM, else there is a bug */
845 WARN_ON_ONCE(ret != -ENOMEM);
846 goto out; /* We know tk is not allocated */
847 }
848 argc -= 2; argv += 2;
849
850 /* parse arguments */
851 for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
852 tmp = kstrdup(argv[i], GFP_KERNEL);
853 if (!tmp) {
854 ret = -ENOMEM;
855 goto error;
856 }
857
858 trace_probe_log_set_index(i + 2);
859 ret = traceprobe_parse_probe_arg(&tk->tp, i, tmp, flags);
860 kfree(tmp);
861 if (ret)
862 goto error; /* This can be -ENOMEM */
863 }
864
865 ret = traceprobe_set_print_fmt(&tk->tp, is_return);
866 if (ret < 0)
867 goto error;
868
869 ret = register_trace_kprobe(tk);
870 if (ret) {
871 trace_probe_log_set_index(1);
872 if (ret == -EILSEQ)
873 trace_probe_log_err(0, BAD_INSN_BNDRY);
874 else if (ret == -ENOENT)
875 trace_probe_log_err(0, BAD_PROBE_ADDR);
876 else if (ret != -ENOMEM && ret != -EEXIST)
877 trace_probe_log_err(0, FAIL_REG_PROBE);
878 goto error;
879 }
880
881out:
882 trace_probe_log_clear();
883 kfree(symbol);
884 return ret;
885
886parse_error:
887 ret = -EINVAL;
888error:
889 free_trace_kprobe(tk);
890 goto out;
891}
892
893static int create_or_delete_trace_kprobe(int argc, char **argv)
894{
895 int ret;
896
897 if (argv[0][0] == '-')
898 return dyn_event_release(argc, argv, &trace_kprobe_ops);
899
900 ret = trace_kprobe_create(argc, (const char **)argv);
901 return ret == -ECANCELED ? -EINVAL : ret;
902}
903
904static int trace_kprobe_run_command(struct dynevent_cmd *cmd)
905{
906 return trace_run_command(cmd->seq.buffer, create_or_delete_trace_kprobe);
907}
908
909/**
910 * kprobe_event_cmd_init - Initialize a kprobe event command object
911 * @cmd: A pointer to the dynevent_cmd struct representing the new event
912 * @buf: A pointer to the buffer used to build the command
913 * @maxlen: The length of the buffer passed in @buf
914 *
915 * Initialize a synthetic event command object. Use this before
916 * calling any of the other kprobe_event functions.
917 */
918void kprobe_event_cmd_init(struct dynevent_cmd *cmd, char *buf, int maxlen)
919{
920 dynevent_cmd_init(cmd, buf, maxlen, DYNEVENT_TYPE_KPROBE,
921 trace_kprobe_run_command);
922}
923EXPORT_SYMBOL_GPL(kprobe_event_cmd_init);
924
925/**
926 * __kprobe_event_gen_cmd_start - Generate a kprobe event command from arg list
927 * @cmd: A pointer to the dynevent_cmd struct representing the new event
928 * @name: The name of the kprobe event
929 * @loc: The location of the kprobe event
930 * @kretprobe: Is this a return probe?
931 * @args: Variable number of arg (pairs), one pair for each field
932 *
933 * NOTE: Users normally won't want to call this function directly, but
934 * rather use the kprobe_event_gen_cmd_start() wrapper, which automatically
935 * adds a NULL to the end of the arg list. If this function is used
936 * directly, make sure the last arg in the variable arg list is NULL.
937 *
938 * Generate a kprobe event command to be executed by
939 * kprobe_event_gen_cmd_end(). This function can be used to generate the
940 * complete command or only the first part of it; in the latter case,
941 * kprobe_event_add_fields() can be used to add more fields following this.
942 *
943 * Unlikely the synth_event_gen_cmd_start(), @loc must be specified. This
944 * returns -EINVAL if @loc == NULL.
945 *
946 * Return: 0 if successful, error otherwise.
947 */
948int __kprobe_event_gen_cmd_start(struct dynevent_cmd *cmd, bool kretprobe,
949 const char *name, const char *loc, ...)
950{
951 char buf[MAX_EVENT_NAME_LEN];
952 struct dynevent_arg arg;
953 va_list args;
954 int ret;
955
956 if (cmd->type != DYNEVENT_TYPE_KPROBE)
957 return -EINVAL;
958
959 if (!loc)
960 return -EINVAL;
961
962 if (kretprobe)
963 snprintf(buf, MAX_EVENT_NAME_LEN, "r:kprobes/%s", name);
964 else
965 snprintf(buf, MAX_EVENT_NAME_LEN, "p:kprobes/%s", name);
966
967 ret = dynevent_str_add(cmd, buf);
968 if (ret)
969 return ret;
970
971 dynevent_arg_init(&arg, 0);
972 arg.str = loc;
973 ret = dynevent_arg_add(cmd, &arg, NULL);
974 if (ret)
975 return ret;
976
977 va_start(args, loc);
978 for (;;) {
979 const char *field;
980
981 field = va_arg(args, const char *);
982 if (!field)
983 break;
984
985 if (++cmd->n_fields > MAX_TRACE_ARGS) {
986 ret = -EINVAL;
987 break;
988 }
989
990 arg.str = field;
991 ret = dynevent_arg_add(cmd, &arg, NULL);
992 if (ret)
993 break;
994 }
995 va_end(args);
996
997 return ret;
998}
999EXPORT_SYMBOL_GPL(__kprobe_event_gen_cmd_start);
1000
1001/**
1002 * __kprobe_event_add_fields - Add probe fields to a kprobe command from arg list
1003 * @cmd: A pointer to the dynevent_cmd struct representing the new event
1004 * @args: Variable number of arg (pairs), one pair for each field
1005 *
1006 * NOTE: Users normally won't want to call this function directly, but
1007 * rather use the kprobe_event_add_fields() wrapper, which
1008 * automatically adds a NULL to the end of the arg list. If this
1009 * function is used directly, make sure the last arg in the variable
1010 * arg list is NULL.
1011 *
1012 * Add probe fields to an existing kprobe command using a variable
1013 * list of args. Fields are added in the same order they're listed.
1014 *
1015 * Return: 0 if successful, error otherwise.
1016 */
1017int __kprobe_event_add_fields(struct dynevent_cmd *cmd, ...)
1018{
1019 struct dynevent_arg arg;
1020 va_list args;
1021 int ret = 0;
1022
1023 if (cmd->type != DYNEVENT_TYPE_KPROBE)
1024 return -EINVAL;
1025
1026 dynevent_arg_init(&arg, 0);
1027
1028 va_start(args, cmd);
1029 for (;;) {
1030 const char *field;
1031
1032 field = va_arg(args, const char *);
1033 if (!field)
1034 break;
1035
1036 if (++cmd->n_fields > MAX_TRACE_ARGS) {
1037 ret = -EINVAL;
1038 break;
1039 }
1040
1041 arg.str = field;
1042 ret = dynevent_arg_add(cmd, &arg, NULL);
1043 if (ret)
1044 break;
1045 }
1046 va_end(args);
1047
1048 return ret;
1049}
1050EXPORT_SYMBOL_GPL(__kprobe_event_add_fields);
1051
1052/**
1053 * kprobe_event_delete - Delete a kprobe event
1054 * @name: The name of the kprobe event to delete
1055 *
1056 * Delete a kprobe event with the give @name from kernel code rather
1057 * than directly from the command line.
1058 *
1059 * Return: 0 if successful, error otherwise.
1060 */
1061int kprobe_event_delete(const char *name)
1062{
1063 char buf[MAX_EVENT_NAME_LEN];
1064
1065 snprintf(buf, MAX_EVENT_NAME_LEN, "-:%s", name);
1066
1067 return trace_run_command(buf, create_or_delete_trace_kprobe);
1068}
1069EXPORT_SYMBOL_GPL(kprobe_event_delete);
1070
1071static int trace_kprobe_release(struct dyn_event *ev)
1072{
1073 struct trace_kprobe *tk = to_trace_kprobe(ev);
1074 int ret = unregister_trace_kprobe(tk);
1075
1076 if (!ret)
1077 free_trace_kprobe(tk);
1078 return ret;
1079}
1080
1081static int trace_kprobe_show(struct seq_file *m, struct dyn_event *ev)
1082{
1083 struct trace_kprobe *tk = to_trace_kprobe(ev);
1084 int i;
1085
1086 seq_putc(m, trace_kprobe_is_return(tk) ? 'r' : 'p');
1087 if (trace_kprobe_is_return(tk) && tk->rp.maxactive)
1088 seq_printf(m, "%d", tk->rp.maxactive);
1089 seq_printf(m, ":%s/%s", trace_probe_group_name(&tk->tp),
1090 trace_probe_name(&tk->tp));
1091
1092 if (!tk->symbol)
1093 seq_printf(m, " 0x%p", tk->rp.kp.addr);
1094 else if (tk->rp.kp.offset)
1095 seq_printf(m, " %s+%u", trace_kprobe_symbol(tk),
1096 tk->rp.kp.offset);
1097 else
1098 seq_printf(m, " %s", trace_kprobe_symbol(tk));
1099
1100 for (i = 0; i < tk->tp.nr_args; i++)
1101 seq_printf(m, " %s=%s", tk->tp.args[i].name, tk->tp.args[i].comm);
1102 seq_putc(m, '\n');
1103
1104 return 0;
1105}
1106
1107static int probes_seq_show(struct seq_file *m, void *v)
1108{
1109 struct dyn_event *ev = v;
1110
1111 if (!is_trace_kprobe(ev))
1112 return 0;
1113
1114 return trace_kprobe_show(m, ev);
1115}
1116
1117static const struct seq_operations probes_seq_op = {
1118 .start = dyn_event_seq_start,
1119 .next = dyn_event_seq_next,
1120 .stop = dyn_event_seq_stop,
1121 .show = probes_seq_show
1122};
1123
1124static int probes_open(struct inode *inode, struct file *file)
1125{
1126 int ret;
1127
1128 ret = security_locked_down(LOCKDOWN_TRACEFS);
1129 if (ret)
1130 return ret;
1131
1132 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
1133 ret = dyn_events_release_all(&trace_kprobe_ops);
1134 if (ret < 0)
1135 return ret;
1136 }
1137
1138 return seq_open(file, &probes_seq_op);
1139}
1140
1141static ssize_t probes_write(struct file *file, const char __user *buffer,
1142 size_t count, loff_t *ppos)
1143{
1144 return trace_parse_run_command(file, buffer, count, ppos,
1145 create_or_delete_trace_kprobe);
1146}
1147
1148static const struct file_operations kprobe_events_ops = {
1149 .owner = THIS_MODULE,
1150 .open = probes_open,
1151 .read = seq_read,
1152 .llseek = seq_lseek,
1153 .release = seq_release,
1154 .write = probes_write,
1155};
1156
1157/* Probes profiling interfaces */
1158static int probes_profile_seq_show(struct seq_file *m, void *v)
1159{
1160 struct dyn_event *ev = v;
1161 struct trace_kprobe *tk;
1162
1163 if (!is_trace_kprobe(ev))
1164 return 0;
1165
1166 tk = to_trace_kprobe(ev);
1167 seq_printf(m, " %-44s %15lu %15lu\n",
1168 trace_probe_name(&tk->tp),
1169 trace_kprobe_nhit(tk),
1170 tk->rp.kp.nmissed);
1171
1172 return 0;
1173}
1174
1175static const struct seq_operations profile_seq_op = {
1176 .start = dyn_event_seq_start,
1177 .next = dyn_event_seq_next,
1178 .stop = dyn_event_seq_stop,
1179 .show = probes_profile_seq_show
1180};
1181
1182static int profile_open(struct inode *inode, struct file *file)
1183{
1184 int ret;
1185
1186 ret = security_locked_down(LOCKDOWN_TRACEFS);
1187 if (ret)
1188 return ret;
1189
1190 return seq_open(file, &profile_seq_op);
1191}
1192
1193static const struct file_operations kprobe_profile_ops = {
1194 .owner = THIS_MODULE,
1195 .open = profile_open,
1196 .read = seq_read,
1197 .llseek = seq_lseek,
1198 .release = seq_release,
1199};
1200
1201/* Kprobe specific fetch functions */
1202
1203/* Return the length of string -- including null terminal byte */
1204static nokprobe_inline int
1205fetch_store_strlen_user(unsigned long addr)
1206{
1207 const void __user *uaddr = (__force const void __user *)addr;
1208
1209 return strnlen_user_nofault(uaddr, MAX_STRING_SIZE);
1210}
1211
1212/* Return the length of string -- including null terminal byte */
1213static nokprobe_inline int
1214fetch_store_strlen(unsigned long addr)
1215{
1216 int ret, len = 0;
1217 u8 c;
1218
1219#ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
1220 if (addr < TASK_SIZE)
1221 return fetch_store_strlen_user(addr);
1222#endif
1223
1224 do {
1225 ret = copy_from_kernel_nofault(&c, (u8 *)addr + len, 1);
1226 len++;
1227 } while (c && ret == 0 && len < MAX_STRING_SIZE);
1228
1229 return (ret < 0) ? ret : len;
1230}
1231
1232/*
1233 * Fetch a null-terminated string from user. Caller MUST set *(u32 *)buf
1234 * with max length and relative data location.
1235 */
1236static nokprobe_inline int
1237fetch_store_string_user(unsigned long addr, void *dest, void *base)
1238{
1239 const void __user *uaddr = (__force const void __user *)addr;
1240 int maxlen = get_loc_len(*(u32 *)dest);
1241 void *__dest;
1242 long ret;
1243
1244 if (unlikely(!maxlen))
1245 return -ENOMEM;
1246
1247 __dest = get_loc_data(dest, base);
1248
1249 ret = strncpy_from_user_nofault(__dest, uaddr, maxlen);
1250 if (ret >= 0)
1251 *(u32 *)dest = make_data_loc(ret, __dest - base);
1252
1253 return ret;
1254}
1255
1256/*
1257 * Fetch a null-terminated string. Caller MUST set *(u32 *)buf with max
1258 * length and relative data location.
1259 */
1260static nokprobe_inline int
1261fetch_store_string(unsigned long addr, void *dest, void *base)
1262{
1263 int maxlen = get_loc_len(*(u32 *)dest);
1264 void *__dest;
1265 long ret;
1266
1267#ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
1268 if ((unsigned long)addr < TASK_SIZE)
1269 return fetch_store_string_user(addr, dest, base);
1270#endif
1271
1272 if (unlikely(!maxlen))
1273 return -ENOMEM;
1274
1275 __dest = get_loc_data(dest, base);
1276
1277 /*
1278 * Try to get string again, since the string can be changed while
1279 * probing.
1280 */
1281 ret = strncpy_from_kernel_nofault(__dest, (void *)addr, maxlen);
1282 if (ret >= 0)
1283 *(u32 *)dest = make_data_loc(ret, __dest - base);
1284
1285 return ret;
1286}
1287
1288static nokprobe_inline int
1289probe_mem_read_user(void *dest, void *src, size_t size)
1290{
1291 const void __user *uaddr = (__force const void __user *)src;
1292
1293 return copy_from_user_nofault(dest, uaddr, size);
1294}
1295
1296static nokprobe_inline int
1297probe_mem_read(void *dest, void *src, size_t size)
1298{
1299#ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
1300 if ((unsigned long)src < TASK_SIZE)
1301 return probe_mem_read_user(dest, src, size);
1302#endif
1303 return copy_from_kernel_nofault(dest, src, size);
1304}
1305
1306/* Note that we don't verify it, since the code does not come from user space */
1307static int
1308process_fetch_insn(struct fetch_insn *code, struct pt_regs *regs, void *dest,
1309 void *base)
1310{
1311 unsigned long val;
1312
1313retry:
1314 /* 1st stage: get value from context */
1315 switch (code->op) {
1316 case FETCH_OP_REG:
1317 val = regs_get_register(regs, code->param);
1318 break;
1319 case FETCH_OP_STACK:
1320 val = regs_get_kernel_stack_nth(regs, code->param);
1321 break;
1322 case FETCH_OP_STACKP:
1323 val = kernel_stack_pointer(regs);
1324 break;
1325 case FETCH_OP_RETVAL:
1326 val = regs_return_value(regs);
1327 break;
1328 case FETCH_OP_IMM:
1329 val = code->immediate;
1330 break;
1331 case FETCH_OP_COMM:
1332 val = (unsigned long)current->comm;
1333 break;
1334 case FETCH_OP_DATA:
1335 val = (unsigned long)code->data;
1336 break;
1337#ifdef CONFIG_HAVE_FUNCTION_ARG_ACCESS_API
1338 case FETCH_OP_ARG:
1339 val = regs_get_kernel_argument(regs, code->param);
1340 break;
1341#endif
1342 case FETCH_NOP_SYMBOL: /* Ignore a place holder */
1343 code++;
1344 goto retry;
1345 default:
1346 return -EILSEQ;
1347 }
1348 code++;
1349
1350 return process_fetch_insn_bottom(code, val, dest, base);
1351}
1352NOKPROBE_SYMBOL(process_fetch_insn)
1353
1354/* Kprobe handler */
1355static nokprobe_inline void
1356__kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs,
1357 struct trace_event_file *trace_file)
1358{
1359 struct kprobe_trace_entry_head *entry;
1360 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1361 struct trace_event_buffer fbuffer;
1362 int dsize;
1363
1364 WARN_ON(call != trace_file->event_call);
1365
1366 if (trace_trigger_soft_disabled(trace_file))
1367 return;
1368
1369 local_save_flags(fbuffer.flags);
1370 fbuffer.pc = preempt_count();
1371 fbuffer.trace_file = trace_file;
1372
1373 dsize = __get_data_size(&tk->tp, regs);
1374
1375 fbuffer.event =
1376 trace_event_buffer_lock_reserve(&fbuffer.buffer, trace_file,
1377 call->event.type,
1378 sizeof(*entry) + tk->tp.size + dsize,
1379 fbuffer.flags, fbuffer.pc);
1380 if (!fbuffer.event)
1381 return;
1382
1383 fbuffer.regs = regs;
1384 entry = fbuffer.entry = ring_buffer_event_data(fbuffer.event);
1385 entry->ip = (unsigned long)tk->rp.kp.addr;
1386 store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1387
1388 trace_event_buffer_commit(&fbuffer);
1389}
1390
1391static void
1392kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs)
1393{
1394 struct event_file_link *link;
1395
1396 trace_probe_for_each_link_rcu(link, &tk->tp)
1397 __kprobe_trace_func(tk, regs, link->file);
1398}
1399NOKPROBE_SYMBOL(kprobe_trace_func);
1400
1401/* Kretprobe handler */
1402static nokprobe_inline void
1403__kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1404 struct pt_regs *regs,
1405 struct trace_event_file *trace_file)
1406{
1407 struct kretprobe_trace_entry_head *entry;
1408 struct trace_event_buffer fbuffer;
1409 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1410 int dsize;
1411
1412 WARN_ON(call != trace_file->event_call);
1413
1414 if (trace_trigger_soft_disabled(trace_file))
1415 return;
1416
1417 local_save_flags(fbuffer.flags);
1418 fbuffer.pc = preempt_count();
1419 fbuffer.trace_file = trace_file;
1420
1421 dsize = __get_data_size(&tk->tp, regs);
1422 fbuffer.event =
1423 trace_event_buffer_lock_reserve(&fbuffer.buffer, trace_file,
1424 call->event.type,
1425 sizeof(*entry) + tk->tp.size + dsize,
1426 fbuffer.flags, fbuffer.pc);
1427 if (!fbuffer.event)
1428 return;
1429
1430 fbuffer.regs = regs;
1431 entry = fbuffer.entry = ring_buffer_event_data(fbuffer.event);
1432 entry->func = (unsigned long)tk->rp.kp.addr;
1433 entry->ret_ip = (unsigned long)ri->ret_addr;
1434 store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1435
1436 trace_event_buffer_commit(&fbuffer);
1437}
1438
1439static void
1440kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1441 struct pt_regs *regs)
1442{
1443 struct event_file_link *link;
1444
1445 trace_probe_for_each_link_rcu(link, &tk->tp)
1446 __kretprobe_trace_func(tk, ri, regs, link->file);
1447}
1448NOKPROBE_SYMBOL(kretprobe_trace_func);
1449
1450/* Event entry printers */
1451static enum print_line_t
1452print_kprobe_event(struct trace_iterator *iter, int flags,
1453 struct trace_event *event)
1454{
1455 struct kprobe_trace_entry_head *field;
1456 struct trace_seq *s = &iter->seq;
1457 struct trace_probe *tp;
1458
1459 field = (struct kprobe_trace_entry_head *)iter->ent;
1460 tp = trace_probe_primary_from_call(
1461 container_of(event, struct trace_event_call, event));
1462 if (WARN_ON_ONCE(!tp))
1463 goto out;
1464
1465 trace_seq_printf(s, "%s: (", trace_probe_name(tp));
1466
1467 if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
1468 goto out;
1469
1470 trace_seq_putc(s, ')');
1471
1472 if (print_probe_args(s, tp->args, tp->nr_args,
1473 (u8 *)&field[1], field) < 0)
1474 goto out;
1475
1476 trace_seq_putc(s, '\n');
1477 out:
1478 return trace_handle_return(s);
1479}
1480
1481static enum print_line_t
1482print_kretprobe_event(struct trace_iterator *iter, int flags,
1483 struct trace_event *event)
1484{
1485 struct kretprobe_trace_entry_head *field;
1486 struct trace_seq *s = &iter->seq;
1487 struct trace_probe *tp;
1488
1489 field = (struct kretprobe_trace_entry_head *)iter->ent;
1490 tp = trace_probe_primary_from_call(
1491 container_of(event, struct trace_event_call, event));
1492 if (WARN_ON_ONCE(!tp))
1493 goto out;
1494
1495 trace_seq_printf(s, "%s: (", trace_probe_name(tp));
1496
1497 if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1498 goto out;
1499
1500 trace_seq_puts(s, " <- ");
1501
1502 if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1503 goto out;
1504
1505 trace_seq_putc(s, ')');
1506
1507 if (print_probe_args(s, tp->args, tp->nr_args,
1508 (u8 *)&field[1], field) < 0)
1509 goto out;
1510
1511 trace_seq_putc(s, '\n');
1512
1513 out:
1514 return trace_handle_return(s);
1515}
1516
1517
1518static int kprobe_event_define_fields(struct trace_event_call *event_call)
1519{
1520 int ret;
1521 struct kprobe_trace_entry_head field;
1522 struct trace_probe *tp;
1523
1524 tp = trace_probe_primary_from_call(event_call);
1525 if (WARN_ON_ONCE(!tp))
1526 return -ENOENT;
1527
1528 DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1529
1530 return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
1531}
1532
1533static int kretprobe_event_define_fields(struct trace_event_call *event_call)
1534{
1535 int ret;
1536 struct kretprobe_trace_entry_head field;
1537 struct trace_probe *tp;
1538
1539 tp = trace_probe_primary_from_call(event_call);
1540 if (WARN_ON_ONCE(!tp))
1541 return -ENOENT;
1542
1543 DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1544 DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1545
1546 return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
1547}
1548
1549#ifdef CONFIG_PERF_EVENTS
1550
1551/* Kprobe profile handler */
1552static int
1553kprobe_perf_func(struct trace_kprobe *tk, struct pt_regs *regs)
1554{
1555 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1556 struct kprobe_trace_entry_head *entry;
1557 struct hlist_head *head;
1558 int size, __size, dsize;
1559 int rctx;
1560
1561 if (bpf_prog_array_valid(call)) {
1562 unsigned long orig_ip = instruction_pointer(regs);
1563 int ret;
1564
1565 ret = trace_call_bpf(call, regs);
1566
1567 /*
1568 * We need to check and see if we modified the pc of the
1569 * pt_regs, and if so return 1 so that we don't do the
1570 * single stepping.
1571 */
1572 if (orig_ip != instruction_pointer(regs))
1573 return 1;
1574 if (!ret)
1575 return 0;
1576 }
1577
1578 head = this_cpu_ptr(call->perf_events);
1579 if (hlist_empty(head))
1580 return 0;
1581
1582 dsize = __get_data_size(&tk->tp, regs);
1583 __size = sizeof(*entry) + tk->tp.size + dsize;
1584 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1585 size -= sizeof(u32);
1586
1587 entry = perf_trace_buf_alloc(size, NULL, &rctx);
1588 if (!entry)
1589 return 0;
1590
1591 entry->ip = (unsigned long)tk->rp.kp.addr;
1592 memset(&entry[1], 0, dsize);
1593 store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1594 perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1595 head, NULL);
1596 return 0;
1597}
1598NOKPROBE_SYMBOL(kprobe_perf_func);
1599
1600/* Kretprobe profile handler */
1601static void
1602kretprobe_perf_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1603 struct pt_regs *regs)
1604{
1605 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1606 struct kretprobe_trace_entry_head *entry;
1607 struct hlist_head *head;
1608 int size, __size, dsize;
1609 int rctx;
1610
1611 if (bpf_prog_array_valid(call) && !trace_call_bpf(call, regs))
1612 return;
1613
1614 head = this_cpu_ptr(call->perf_events);
1615 if (hlist_empty(head))
1616 return;
1617
1618 dsize = __get_data_size(&tk->tp, regs);
1619 __size = sizeof(*entry) + tk->tp.size + dsize;
1620 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1621 size -= sizeof(u32);
1622
1623 entry = perf_trace_buf_alloc(size, NULL, &rctx);
1624 if (!entry)
1625 return;
1626
1627 entry->func = (unsigned long)tk->rp.kp.addr;
1628 entry->ret_ip = (unsigned long)ri->ret_addr;
1629 store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1630 perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1631 head, NULL);
1632}
1633NOKPROBE_SYMBOL(kretprobe_perf_func);
1634
1635int bpf_get_kprobe_info(const struct perf_event *event, u32 *fd_type,
1636 const char **symbol, u64 *probe_offset,
1637 u64 *probe_addr, bool perf_type_tracepoint)
1638{
1639 const char *pevent = trace_event_name(event->tp_event);
1640 const char *group = event->tp_event->class->system;
1641 struct trace_kprobe *tk;
1642
1643 if (perf_type_tracepoint)
1644 tk = find_trace_kprobe(pevent, group);
1645 else
1646 tk = trace_kprobe_primary_from_call(event->tp_event);
1647 if (!tk)
1648 return -EINVAL;
1649
1650 *fd_type = trace_kprobe_is_return(tk) ? BPF_FD_TYPE_KRETPROBE
1651 : BPF_FD_TYPE_KPROBE;
1652 if (tk->symbol) {
1653 *symbol = tk->symbol;
1654 *probe_offset = tk->rp.kp.offset;
1655 *probe_addr = 0;
1656 } else {
1657 *symbol = NULL;
1658 *probe_offset = 0;
1659 *probe_addr = (unsigned long)tk->rp.kp.addr;
1660 }
1661 return 0;
1662}
1663#endif /* CONFIG_PERF_EVENTS */
1664
1665/*
1666 * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex.
1667 *
1668 * kprobe_trace_self_tests_init() does enable_trace_probe/disable_trace_probe
1669 * lockless, but we can't race with this __init function.
1670 */
1671static int kprobe_register(struct trace_event_call *event,
1672 enum trace_reg type, void *data)
1673{
1674 struct trace_event_file *file = data;
1675
1676 switch (type) {
1677 case TRACE_REG_REGISTER:
1678 return enable_trace_kprobe(event, file);
1679 case TRACE_REG_UNREGISTER:
1680 return disable_trace_kprobe(event, file);
1681
1682#ifdef CONFIG_PERF_EVENTS
1683 case TRACE_REG_PERF_REGISTER:
1684 return enable_trace_kprobe(event, NULL);
1685 case TRACE_REG_PERF_UNREGISTER:
1686 return disable_trace_kprobe(event, NULL);
1687 case TRACE_REG_PERF_OPEN:
1688 case TRACE_REG_PERF_CLOSE:
1689 case TRACE_REG_PERF_ADD:
1690 case TRACE_REG_PERF_DEL:
1691 return 0;
1692#endif
1693 }
1694 return 0;
1695}
1696
1697static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1698{
1699 struct trace_kprobe *tk = container_of(kp, struct trace_kprobe, rp.kp);
1700 int ret = 0;
1701
1702 raw_cpu_inc(*tk->nhit);
1703
1704 if (trace_probe_test_flag(&tk->tp, TP_FLAG_TRACE))
1705 kprobe_trace_func(tk, regs);
1706#ifdef CONFIG_PERF_EVENTS
1707 if (trace_probe_test_flag(&tk->tp, TP_FLAG_PROFILE))
1708 ret = kprobe_perf_func(tk, regs);
1709#endif
1710 return ret;
1711}
1712NOKPROBE_SYMBOL(kprobe_dispatcher);
1713
1714static int
1715kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1716{
1717 struct trace_kprobe *tk = container_of(ri->rp, struct trace_kprobe, rp);
1718
1719 raw_cpu_inc(*tk->nhit);
1720
1721 if (trace_probe_test_flag(&tk->tp, TP_FLAG_TRACE))
1722 kretprobe_trace_func(tk, ri, regs);
1723#ifdef CONFIG_PERF_EVENTS
1724 if (trace_probe_test_flag(&tk->tp, TP_FLAG_PROFILE))
1725 kretprobe_perf_func(tk, ri, regs);
1726#endif
1727 return 0; /* We don't tweek kernel, so just return 0 */
1728}
1729NOKPROBE_SYMBOL(kretprobe_dispatcher);
1730
1731static struct trace_event_functions kretprobe_funcs = {
1732 .trace = print_kretprobe_event
1733};
1734
1735static struct trace_event_functions kprobe_funcs = {
1736 .trace = print_kprobe_event
1737};
1738
1739static struct trace_event_fields kretprobe_fields_array[] = {
1740 { .type = TRACE_FUNCTION_TYPE,
1741 .define_fields = kretprobe_event_define_fields },
1742 {}
1743};
1744
1745static struct trace_event_fields kprobe_fields_array[] = {
1746 { .type = TRACE_FUNCTION_TYPE,
1747 .define_fields = kprobe_event_define_fields },
1748 {}
1749};
1750
1751static inline void init_trace_event_call(struct trace_kprobe *tk)
1752{
1753 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1754
1755 if (trace_kprobe_is_return(tk)) {
1756 call->event.funcs = &kretprobe_funcs;
1757 call->class->fields_array = kretprobe_fields_array;
1758 } else {
1759 call->event.funcs = &kprobe_funcs;
1760 call->class->fields_array = kprobe_fields_array;
1761 }
1762
1763 call->flags = TRACE_EVENT_FL_KPROBE;
1764 call->class->reg = kprobe_register;
1765}
1766
1767static int register_kprobe_event(struct trace_kprobe *tk)
1768{
1769 init_trace_event_call(tk);
1770
1771 return trace_probe_register_event_call(&tk->tp);
1772}
1773
1774static int unregister_kprobe_event(struct trace_kprobe *tk)
1775{
1776 return trace_probe_unregister_event_call(&tk->tp);
1777}
1778
1779#ifdef CONFIG_PERF_EVENTS
1780/* create a trace_kprobe, but don't add it to global lists */
1781struct trace_event_call *
1782create_local_trace_kprobe(char *func, void *addr, unsigned long offs,
1783 bool is_return)
1784{
1785 struct trace_kprobe *tk;
1786 int ret;
1787 char *event;
1788
1789 /*
1790 * local trace_kprobes are not added to dyn_event, so they are never
1791 * searched in find_trace_kprobe(). Therefore, there is no concern of
1792 * duplicated name here.
1793 */
1794 event = func ? func : "DUMMY_EVENT";
1795
1796 tk = alloc_trace_kprobe(KPROBE_EVENT_SYSTEM, event, (void *)addr, func,
1797 offs, 0 /* maxactive */, 0 /* nargs */,
1798 is_return);
1799
1800 if (IS_ERR(tk)) {
1801 pr_info("Failed to allocate trace_probe.(%d)\n",
1802 (int)PTR_ERR(tk));
1803 return ERR_CAST(tk);
1804 }
1805
1806 init_trace_event_call(tk);
1807
1808 if (traceprobe_set_print_fmt(&tk->tp, trace_kprobe_is_return(tk)) < 0) {
1809 ret = -ENOMEM;
1810 goto error;
1811 }
1812
1813 ret = __register_trace_kprobe(tk);
1814 if (ret < 0)
1815 goto error;
1816
1817 return trace_probe_event_call(&tk->tp);
1818error:
1819 free_trace_kprobe(tk);
1820 return ERR_PTR(ret);
1821}
1822
1823void destroy_local_trace_kprobe(struct trace_event_call *event_call)
1824{
1825 struct trace_kprobe *tk;
1826
1827 tk = trace_kprobe_primary_from_call(event_call);
1828 if (unlikely(!tk))
1829 return;
1830
1831 if (trace_probe_is_enabled(&tk->tp)) {
1832 WARN_ON(1);
1833 return;
1834 }
1835
1836 __unregister_trace_kprobe(tk);
1837
1838 free_trace_kprobe(tk);
1839}
1840#endif /* CONFIG_PERF_EVENTS */
1841
1842static __init void enable_boot_kprobe_events(void)
1843{
1844 struct trace_array *tr = top_trace_array();
1845 struct trace_event_file *file;
1846 struct trace_kprobe *tk;
1847 struct dyn_event *pos;
1848
1849 mutex_lock(&event_mutex);
1850 for_each_trace_kprobe(tk, pos) {
1851 list_for_each_entry(file, &tr->events, list)
1852 if (file->event_call == trace_probe_event_call(&tk->tp))
1853 trace_event_enable_disable(file, 1, 0);
1854 }
1855 mutex_unlock(&event_mutex);
1856}
1857
1858static __init void setup_boot_kprobe_events(void)
1859{
1860 char *p, *cmd = kprobe_boot_events_buf;
1861 int ret;
1862
1863 strreplace(kprobe_boot_events_buf, ',', ' ');
1864
1865 while (cmd && *cmd != '\0') {
1866 p = strchr(cmd, ';');
1867 if (p)
1868 *p++ = '\0';
1869
1870 ret = trace_run_command(cmd, create_or_delete_trace_kprobe);
1871 if (ret)
1872 pr_warn("Failed to add event(%d): %s\n", ret, cmd);
1873 else
1874 kprobe_boot_events_enabled = true;
1875
1876 cmd = p;
1877 }
1878
1879 enable_boot_kprobe_events();
1880}
1881
1882/*
1883 * Register dynevent at subsys_initcall. This allows kernel to setup kprobe
1884 * events in fs_initcall without tracefs.
1885 */
1886static __init int init_kprobe_trace_early(void)
1887{
1888 int ret;
1889
1890 ret = dyn_event_register(&trace_kprobe_ops);
1891 if (ret)
1892 return ret;
1893
1894 if (register_module_notifier(&trace_kprobe_module_nb))
1895 return -EINVAL;
1896
1897 return 0;
1898}
1899subsys_initcall(init_kprobe_trace_early);
1900
1901/* Make a tracefs interface for controlling probe points */
1902static __init int init_kprobe_trace(void)
1903{
1904 struct dentry *d_tracer;
1905 struct dentry *entry;
1906
1907 d_tracer = tracing_init_dentry();
1908 if (IS_ERR(d_tracer))
1909 return 0;
1910
1911 entry = tracefs_create_file("kprobe_events", 0644, d_tracer,
1912 NULL, &kprobe_events_ops);
1913
1914 /* Event list interface */
1915 if (!entry)
1916 pr_warn("Could not create tracefs 'kprobe_events' entry\n");
1917
1918 /* Profile interface */
1919 entry = tracefs_create_file("kprobe_profile", 0444, d_tracer,
1920 NULL, &kprobe_profile_ops);
1921
1922 if (!entry)
1923 pr_warn("Could not create tracefs 'kprobe_profile' entry\n");
1924
1925 setup_boot_kprobe_events();
1926
1927 return 0;
1928}
1929fs_initcall(init_kprobe_trace);
1930
1931
1932#ifdef CONFIG_FTRACE_STARTUP_TEST
1933static __init struct trace_event_file *
1934find_trace_probe_file(struct trace_kprobe *tk, struct trace_array *tr)
1935{
1936 struct trace_event_file *file;
1937
1938 list_for_each_entry(file, &tr->events, list)
1939 if (file->event_call == trace_probe_event_call(&tk->tp))
1940 return file;
1941
1942 return NULL;
1943}
1944
1945/*
1946 * Nobody but us can call enable_trace_kprobe/disable_trace_kprobe at this
1947 * stage, we can do this lockless.
1948 */
1949static __init int kprobe_trace_self_tests_init(void)
1950{
1951 int ret, warn = 0;
1952 int (*target)(int, int, int, int, int, int);
1953 struct trace_kprobe *tk;
1954 struct trace_event_file *file;
1955
1956 if (tracing_is_disabled())
1957 return -ENODEV;
1958
1959 if (kprobe_boot_events_enabled) {
1960 pr_info("Skipping kprobe tests due to kprobe_event on cmdline\n");
1961 return 0;
1962 }
1963
1964 target = kprobe_trace_selftest_target;
1965
1966 pr_info("Testing kprobe tracing: ");
1967
1968 ret = trace_run_command("p:testprobe kprobe_trace_selftest_target $stack $stack0 +0($stack)",
1969 create_or_delete_trace_kprobe);
1970 if (WARN_ON_ONCE(ret)) {
1971 pr_warn("error on probing function entry.\n");
1972 warn++;
1973 } else {
1974 /* Enable trace point */
1975 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1976 if (WARN_ON_ONCE(tk == NULL)) {
1977 pr_warn("error on getting new probe.\n");
1978 warn++;
1979 } else {
1980 file = find_trace_probe_file(tk, top_trace_array());
1981 if (WARN_ON_ONCE(file == NULL)) {
1982 pr_warn("error on getting probe file.\n");
1983 warn++;
1984 } else
1985 enable_trace_kprobe(
1986 trace_probe_event_call(&tk->tp), file);
1987 }
1988 }
1989
1990 ret = trace_run_command("r:testprobe2 kprobe_trace_selftest_target $retval",
1991 create_or_delete_trace_kprobe);
1992 if (WARN_ON_ONCE(ret)) {
1993 pr_warn("error on probing function return.\n");
1994 warn++;
1995 } else {
1996 /* Enable trace point */
1997 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1998 if (WARN_ON_ONCE(tk == NULL)) {
1999 pr_warn("error on getting 2nd new probe.\n");
2000 warn++;
2001 } else {
2002 file = find_trace_probe_file(tk, top_trace_array());
2003 if (WARN_ON_ONCE(file == NULL)) {
2004 pr_warn("error on getting probe file.\n");
2005 warn++;
2006 } else
2007 enable_trace_kprobe(
2008 trace_probe_event_call(&tk->tp), file);
2009 }
2010 }
2011
2012 if (warn)
2013 goto end;
2014
2015 ret = target(1, 2, 3, 4, 5, 6);
2016
2017 /*
2018 * Not expecting an error here, the check is only to prevent the
2019 * optimizer from removing the call to target() as otherwise there
2020 * are no side-effects and the call is never performed.
2021 */
2022 if (ret != 21)
2023 warn++;
2024
2025 /* Disable trace points before removing it */
2026 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
2027 if (WARN_ON_ONCE(tk == NULL)) {
2028 pr_warn("error on getting test probe.\n");
2029 warn++;
2030 } else {
2031 if (trace_kprobe_nhit(tk) != 1) {
2032 pr_warn("incorrect number of testprobe hits\n");
2033 warn++;
2034 }
2035
2036 file = find_trace_probe_file(tk, top_trace_array());
2037 if (WARN_ON_ONCE(file == NULL)) {
2038 pr_warn("error on getting probe file.\n");
2039 warn++;
2040 } else
2041 disable_trace_kprobe(
2042 trace_probe_event_call(&tk->tp), file);
2043 }
2044
2045 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
2046 if (WARN_ON_ONCE(tk == NULL)) {
2047 pr_warn("error on getting 2nd test probe.\n");
2048 warn++;
2049 } else {
2050 if (trace_kprobe_nhit(tk) != 1) {
2051 pr_warn("incorrect number of testprobe2 hits\n");
2052 warn++;
2053 }
2054
2055 file = find_trace_probe_file(tk, top_trace_array());
2056 if (WARN_ON_ONCE(file == NULL)) {
2057 pr_warn("error on getting probe file.\n");
2058 warn++;
2059 } else
2060 disable_trace_kprobe(
2061 trace_probe_event_call(&tk->tp), file);
2062 }
2063
2064 ret = trace_run_command("-:testprobe", create_or_delete_trace_kprobe);
2065 if (WARN_ON_ONCE(ret)) {
2066 pr_warn("error on deleting a probe.\n");
2067 warn++;
2068 }
2069
2070 ret = trace_run_command("-:testprobe2", create_or_delete_trace_kprobe);
2071 if (WARN_ON_ONCE(ret)) {
2072 pr_warn("error on deleting a probe.\n");
2073 warn++;
2074 }
2075
2076end:
2077 ret = dyn_events_release_all(&trace_kprobe_ops);
2078 if (WARN_ON_ONCE(ret)) {
2079 pr_warn("error on cleaning up probes.\n");
2080 warn++;
2081 }
2082 /*
2083 * Wait for the optimizer work to finish. Otherwise it might fiddle
2084 * with probes in already freed __init text.
2085 */
2086 wait_for_kprobe_optimizer();
2087 if (warn)
2088 pr_cont("NG: Some tests are failed. Please check them.\n");
2089 else
2090 pr_cont("OK\n");
2091 return 0;
2092}
2093
2094late_initcall(kprobe_trace_self_tests_init);
2095
2096#endif