Loading...
1/*
2 * Kprobes-based tracing events
3 *
4 * Created by Masami Hiramatsu <mhiramat@redhat.com>
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 as
8 * published by the Free Software Foundation.
9 *
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
14 *
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write to the Free Software
17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
18 */
19
20#include <linux/module.h>
21#include <linux/uaccess.h>
22
23#include "trace_probe.h"
24
25#define KPROBE_EVENT_SYSTEM "kprobes"
26
27/**
28 * Kprobe event core functions
29 */
30struct trace_kprobe {
31 struct list_head list;
32 struct kretprobe rp; /* Use rp.kp for kprobe use */
33 unsigned long __percpu *nhit;
34 const char *symbol; /* symbol name */
35 struct trace_probe tp;
36};
37
38#define SIZEOF_TRACE_KPROBE(n) \
39 (offsetof(struct trace_kprobe, tp.args) + \
40 (sizeof(struct probe_arg) * (n)))
41
42
43static nokprobe_inline bool trace_kprobe_is_return(struct trace_kprobe *tk)
44{
45 return tk->rp.handler != NULL;
46}
47
48static nokprobe_inline const char *trace_kprobe_symbol(struct trace_kprobe *tk)
49{
50 return tk->symbol ? tk->symbol : "unknown";
51}
52
53static nokprobe_inline unsigned long trace_kprobe_offset(struct trace_kprobe *tk)
54{
55 return tk->rp.kp.offset;
56}
57
58static nokprobe_inline bool trace_kprobe_has_gone(struct trace_kprobe *tk)
59{
60 return !!(kprobe_gone(&tk->rp.kp));
61}
62
63static nokprobe_inline bool trace_kprobe_within_module(struct trace_kprobe *tk,
64 struct module *mod)
65{
66 int len = strlen(mod->name);
67 const char *name = trace_kprobe_symbol(tk);
68 return strncmp(mod->name, name, len) == 0 && name[len] == ':';
69}
70
71static nokprobe_inline bool trace_kprobe_is_on_module(struct trace_kprobe *tk)
72{
73 return !!strchr(trace_kprobe_symbol(tk), ':');
74}
75
76static int register_kprobe_event(struct trace_kprobe *tk);
77static int unregister_kprobe_event(struct trace_kprobe *tk);
78
79static DEFINE_MUTEX(probe_lock);
80static LIST_HEAD(probe_list);
81
82static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
83static int kretprobe_dispatcher(struct kretprobe_instance *ri,
84 struct pt_regs *regs);
85
86/* Memory fetching by symbol */
87struct symbol_cache {
88 char *symbol;
89 long offset;
90 unsigned long addr;
91};
92
93unsigned long update_symbol_cache(struct symbol_cache *sc)
94{
95 sc->addr = (unsigned long)kallsyms_lookup_name(sc->symbol);
96
97 if (sc->addr)
98 sc->addr += sc->offset;
99
100 return sc->addr;
101}
102
103void free_symbol_cache(struct symbol_cache *sc)
104{
105 kfree(sc->symbol);
106 kfree(sc);
107}
108
109struct symbol_cache *alloc_symbol_cache(const char *sym, long offset)
110{
111 struct symbol_cache *sc;
112
113 if (!sym || strlen(sym) == 0)
114 return NULL;
115
116 sc = kzalloc(sizeof(struct symbol_cache), GFP_KERNEL);
117 if (!sc)
118 return NULL;
119
120 sc->symbol = kstrdup(sym, GFP_KERNEL);
121 if (!sc->symbol) {
122 kfree(sc);
123 return NULL;
124 }
125 sc->offset = offset;
126 update_symbol_cache(sc);
127
128 return sc;
129}
130
131/*
132 * Kprobes-specific fetch functions
133 */
134#define DEFINE_FETCH_stack(type) \
135static void FETCH_FUNC_NAME(stack, type)(struct pt_regs *regs, \
136 void *offset, void *dest) \
137{ \
138 *(type *)dest = (type)regs_get_kernel_stack_nth(regs, \
139 (unsigned int)((unsigned long)offset)); \
140} \
141NOKPROBE_SYMBOL(FETCH_FUNC_NAME(stack, type));
142
143DEFINE_BASIC_FETCH_FUNCS(stack)
144/* No string on the stack entry */
145#define fetch_stack_string NULL
146#define fetch_stack_string_size NULL
147
148#define DEFINE_FETCH_memory(type) \
149static void FETCH_FUNC_NAME(memory, type)(struct pt_regs *regs, \
150 void *addr, void *dest) \
151{ \
152 type retval; \
153 if (probe_kernel_address(addr, retval)) \
154 *(type *)dest = 0; \
155 else \
156 *(type *)dest = retval; \
157} \
158NOKPROBE_SYMBOL(FETCH_FUNC_NAME(memory, type));
159
160DEFINE_BASIC_FETCH_FUNCS(memory)
161/*
162 * Fetch a null-terminated string. Caller MUST set *(u32 *)dest with max
163 * length and relative data location.
164 */
165static void FETCH_FUNC_NAME(memory, string)(struct pt_regs *regs,
166 void *addr, void *dest)
167{
168 int maxlen = get_rloc_len(*(u32 *)dest);
169 u8 *dst = get_rloc_data(dest);
170 long ret;
171
172 if (!maxlen)
173 return;
174
175 /*
176 * Try to get string again, since the string can be changed while
177 * probing.
178 */
179 ret = strncpy_from_unsafe(dst, addr, maxlen);
180
181 if (ret < 0) { /* Failed to fetch string */
182 dst[0] = '\0';
183 *(u32 *)dest = make_data_rloc(0, get_rloc_offs(*(u32 *)dest));
184 } else {
185 *(u32 *)dest = make_data_rloc(ret, get_rloc_offs(*(u32 *)dest));
186 }
187}
188NOKPROBE_SYMBOL(FETCH_FUNC_NAME(memory, string));
189
190/* Return the length of string -- including null terminal byte */
191static void FETCH_FUNC_NAME(memory, string_size)(struct pt_regs *regs,
192 void *addr, void *dest)
193{
194 mm_segment_t old_fs;
195 int ret, len = 0;
196 u8 c;
197
198 old_fs = get_fs();
199 set_fs(KERNEL_DS);
200 pagefault_disable();
201
202 do {
203 ret = __copy_from_user_inatomic(&c, (u8 *)addr + len, 1);
204 len++;
205 } while (c && ret == 0 && len < MAX_STRING_SIZE);
206
207 pagefault_enable();
208 set_fs(old_fs);
209
210 if (ret < 0) /* Failed to check the length */
211 *(u32 *)dest = 0;
212 else
213 *(u32 *)dest = len;
214}
215NOKPROBE_SYMBOL(FETCH_FUNC_NAME(memory, string_size));
216
217#define DEFINE_FETCH_symbol(type) \
218void FETCH_FUNC_NAME(symbol, type)(struct pt_regs *regs, void *data, void *dest)\
219{ \
220 struct symbol_cache *sc = data; \
221 if (sc->addr) \
222 fetch_memory_##type(regs, (void *)sc->addr, dest); \
223 else \
224 *(type *)dest = 0; \
225} \
226NOKPROBE_SYMBOL(FETCH_FUNC_NAME(symbol, type));
227
228DEFINE_BASIC_FETCH_FUNCS(symbol)
229DEFINE_FETCH_symbol(string)
230DEFINE_FETCH_symbol(string_size)
231
232/* kprobes don't support file_offset fetch methods */
233#define fetch_file_offset_u8 NULL
234#define fetch_file_offset_u16 NULL
235#define fetch_file_offset_u32 NULL
236#define fetch_file_offset_u64 NULL
237#define fetch_file_offset_string NULL
238#define fetch_file_offset_string_size NULL
239
240/* Fetch type information table */
241static const struct fetch_type kprobes_fetch_type_table[] = {
242 /* Special types */
243 [FETCH_TYPE_STRING] = __ASSIGN_FETCH_TYPE("string", string, string,
244 sizeof(u32), 1, "__data_loc char[]"),
245 [FETCH_TYPE_STRSIZE] = __ASSIGN_FETCH_TYPE("string_size", u32,
246 string_size, sizeof(u32), 0, "u32"),
247 /* Basic types */
248 ASSIGN_FETCH_TYPE(u8, u8, 0),
249 ASSIGN_FETCH_TYPE(u16, u16, 0),
250 ASSIGN_FETCH_TYPE(u32, u32, 0),
251 ASSIGN_FETCH_TYPE(u64, u64, 0),
252 ASSIGN_FETCH_TYPE(s8, u8, 1),
253 ASSIGN_FETCH_TYPE(s16, u16, 1),
254 ASSIGN_FETCH_TYPE(s32, u32, 1),
255 ASSIGN_FETCH_TYPE(s64, u64, 1),
256
257 ASSIGN_FETCH_TYPE_END
258};
259
260/*
261 * Allocate new trace_probe and initialize it (including kprobes).
262 */
263static struct trace_kprobe *alloc_trace_kprobe(const char *group,
264 const char *event,
265 void *addr,
266 const char *symbol,
267 unsigned long offs,
268 int nargs, bool is_return)
269{
270 struct trace_kprobe *tk;
271 int ret = -ENOMEM;
272
273 tk = kzalloc(SIZEOF_TRACE_KPROBE(nargs), GFP_KERNEL);
274 if (!tk)
275 return ERR_PTR(ret);
276
277 tk->nhit = alloc_percpu(unsigned long);
278 if (!tk->nhit)
279 goto error;
280
281 if (symbol) {
282 tk->symbol = kstrdup(symbol, GFP_KERNEL);
283 if (!tk->symbol)
284 goto error;
285 tk->rp.kp.symbol_name = tk->symbol;
286 tk->rp.kp.offset = offs;
287 } else
288 tk->rp.kp.addr = addr;
289
290 if (is_return)
291 tk->rp.handler = kretprobe_dispatcher;
292 else
293 tk->rp.kp.pre_handler = kprobe_dispatcher;
294
295 if (!event || !is_good_name(event)) {
296 ret = -EINVAL;
297 goto error;
298 }
299
300 tk->tp.call.class = &tk->tp.class;
301 tk->tp.call.name = kstrdup(event, GFP_KERNEL);
302 if (!tk->tp.call.name)
303 goto error;
304
305 if (!group || !is_good_name(group)) {
306 ret = -EINVAL;
307 goto error;
308 }
309
310 tk->tp.class.system = kstrdup(group, GFP_KERNEL);
311 if (!tk->tp.class.system)
312 goto error;
313
314 INIT_LIST_HEAD(&tk->list);
315 INIT_LIST_HEAD(&tk->tp.files);
316 return tk;
317error:
318 kfree(tk->tp.call.name);
319 kfree(tk->symbol);
320 free_percpu(tk->nhit);
321 kfree(tk);
322 return ERR_PTR(ret);
323}
324
325static void free_trace_kprobe(struct trace_kprobe *tk)
326{
327 int i;
328
329 for (i = 0; i < tk->tp.nr_args; i++)
330 traceprobe_free_probe_arg(&tk->tp.args[i]);
331
332 kfree(tk->tp.call.class->system);
333 kfree(tk->tp.call.name);
334 kfree(tk->symbol);
335 free_percpu(tk->nhit);
336 kfree(tk);
337}
338
339static struct trace_kprobe *find_trace_kprobe(const char *event,
340 const char *group)
341{
342 struct trace_kprobe *tk;
343
344 list_for_each_entry(tk, &probe_list, list)
345 if (strcmp(trace_event_name(&tk->tp.call), event) == 0 &&
346 strcmp(tk->tp.call.class->system, group) == 0)
347 return tk;
348 return NULL;
349}
350
351/*
352 * Enable trace_probe
353 * if the file is NULL, enable "perf" handler, or enable "trace" handler.
354 */
355static int
356enable_trace_kprobe(struct trace_kprobe *tk, struct trace_event_file *file)
357{
358 int ret = 0;
359
360 if (file) {
361 struct event_file_link *link;
362
363 link = kmalloc(sizeof(*link), GFP_KERNEL);
364 if (!link) {
365 ret = -ENOMEM;
366 goto out;
367 }
368
369 link->file = file;
370 list_add_tail_rcu(&link->list, &tk->tp.files);
371
372 tk->tp.flags |= TP_FLAG_TRACE;
373 } else
374 tk->tp.flags |= TP_FLAG_PROFILE;
375
376 if (trace_probe_is_registered(&tk->tp) && !trace_kprobe_has_gone(tk)) {
377 if (trace_kprobe_is_return(tk))
378 ret = enable_kretprobe(&tk->rp);
379 else
380 ret = enable_kprobe(&tk->rp.kp);
381 }
382 out:
383 return ret;
384}
385
386/*
387 * Disable trace_probe
388 * if the file is NULL, disable "perf" handler, or disable "trace" handler.
389 */
390static int
391disable_trace_kprobe(struct trace_kprobe *tk, struct trace_event_file *file)
392{
393 struct event_file_link *link = NULL;
394 int wait = 0;
395 int ret = 0;
396
397 if (file) {
398 link = find_event_file_link(&tk->tp, file);
399 if (!link) {
400 ret = -EINVAL;
401 goto out;
402 }
403
404 list_del_rcu(&link->list);
405 wait = 1;
406 if (!list_empty(&tk->tp.files))
407 goto out;
408
409 tk->tp.flags &= ~TP_FLAG_TRACE;
410 } else
411 tk->tp.flags &= ~TP_FLAG_PROFILE;
412
413 if (!trace_probe_is_enabled(&tk->tp) && trace_probe_is_registered(&tk->tp)) {
414 if (trace_kprobe_is_return(tk))
415 disable_kretprobe(&tk->rp);
416 else
417 disable_kprobe(&tk->rp.kp);
418 wait = 1;
419 }
420 out:
421 if (wait) {
422 /*
423 * Synchronize with kprobe_trace_func/kretprobe_trace_func
424 * to ensure disabled (all running handlers are finished).
425 * This is not only for kfree(), but also the caller,
426 * trace_remove_event_call() supposes it for releasing
427 * event_call related objects, which will be accessed in
428 * the kprobe_trace_func/kretprobe_trace_func.
429 */
430 synchronize_sched();
431 kfree(link); /* Ignored if link == NULL */
432 }
433
434 return ret;
435}
436
437/* Internal register function - just handle k*probes and flags */
438static int __register_trace_kprobe(struct trace_kprobe *tk)
439{
440 int i, ret;
441
442 if (trace_probe_is_registered(&tk->tp))
443 return -EINVAL;
444
445 for (i = 0; i < tk->tp.nr_args; i++)
446 traceprobe_update_arg(&tk->tp.args[i]);
447
448 /* Set/clear disabled flag according to tp->flag */
449 if (trace_probe_is_enabled(&tk->tp))
450 tk->rp.kp.flags &= ~KPROBE_FLAG_DISABLED;
451 else
452 tk->rp.kp.flags |= KPROBE_FLAG_DISABLED;
453
454 if (trace_kprobe_is_return(tk))
455 ret = register_kretprobe(&tk->rp);
456 else
457 ret = register_kprobe(&tk->rp.kp);
458
459 if (ret == 0)
460 tk->tp.flags |= TP_FLAG_REGISTERED;
461 else {
462 pr_warn("Could not insert probe at %s+%lu: %d\n",
463 trace_kprobe_symbol(tk), trace_kprobe_offset(tk), ret);
464 if (ret == -ENOENT && trace_kprobe_is_on_module(tk)) {
465 pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
466 ret = 0;
467 } else if (ret == -EILSEQ) {
468 pr_warn("Probing address(0x%p) is not an instruction boundary.\n",
469 tk->rp.kp.addr);
470 ret = -EINVAL;
471 }
472 }
473
474 return ret;
475}
476
477/* Internal unregister function - just handle k*probes and flags */
478static void __unregister_trace_kprobe(struct trace_kprobe *tk)
479{
480 if (trace_probe_is_registered(&tk->tp)) {
481 if (trace_kprobe_is_return(tk))
482 unregister_kretprobe(&tk->rp);
483 else
484 unregister_kprobe(&tk->rp.kp);
485 tk->tp.flags &= ~TP_FLAG_REGISTERED;
486 /* Cleanup kprobe for reuse */
487 if (tk->rp.kp.symbol_name)
488 tk->rp.kp.addr = NULL;
489 }
490}
491
492/* Unregister a trace_probe and probe_event: call with locking probe_lock */
493static int unregister_trace_kprobe(struct trace_kprobe *tk)
494{
495 /* Enabled event can not be unregistered */
496 if (trace_probe_is_enabled(&tk->tp))
497 return -EBUSY;
498
499 /* Will fail if probe is being used by ftrace or perf */
500 if (unregister_kprobe_event(tk))
501 return -EBUSY;
502
503 __unregister_trace_kprobe(tk);
504 list_del(&tk->list);
505
506 return 0;
507}
508
509/* Register a trace_probe and probe_event */
510static int register_trace_kprobe(struct trace_kprobe *tk)
511{
512 struct trace_kprobe *old_tk;
513 int ret;
514
515 mutex_lock(&probe_lock);
516
517 /* Delete old (same name) event if exist */
518 old_tk = find_trace_kprobe(trace_event_name(&tk->tp.call),
519 tk->tp.call.class->system);
520 if (old_tk) {
521 ret = unregister_trace_kprobe(old_tk);
522 if (ret < 0)
523 goto end;
524 free_trace_kprobe(old_tk);
525 }
526
527 /* Register new event */
528 ret = register_kprobe_event(tk);
529 if (ret) {
530 pr_warn("Failed to register probe event(%d)\n", ret);
531 goto end;
532 }
533
534 /* Register k*probe */
535 ret = __register_trace_kprobe(tk);
536 if (ret < 0)
537 unregister_kprobe_event(tk);
538 else
539 list_add_tail(&tk->list, &probe_list);
540
541end:
542 mutex_unlock(&probe_lock);
543 return ret;
544}
545
546/* Module notifier call back, checking event on the module */
547static int trace_kprobe_module_callback(struct notifier_block *nb,
548 unsigned long val, void *data)
549{
550 struct module *mod = data;
551 struct trace_kprobe *tk;
552 int ret;
553
554 if (val != MODULE_STATE_COMING)
555 return NOTIFY_DONE;
556
557 /* Update probes on coming module */
558 mutex_lock(&probe_lock);
559 list_for_each_entry(tk, &probe_list, list) {
560 if (trace_kprobe_within_module(tk, mod)) {
561 /* Don't need to check busy - this should have gone. */
562 __unregister_trace_kprobe(tk);
563 ret = __register_trace_kprobe(tk);
564 if (ret)
565 pr_warn("Failed to re-register probe %s on %s: %d\n",
566 trace_event_name(&tk->tp.call),
567 mod->name, ret);
568 }
569 }
570 mutex_unlock(&probe_lock);
571
572 return NOTIFY_DONE;
573}
574
575static struct notifier_block trace_kprobe_module_nb = {
576 .notifier_call = trace_kprobe_module_callback,
577 .priority = 1 /* Invoked after kprobe module callback */
578};
579
580static int create_trace_kprobe(int argc, char **argv)
581{
582 /*
583 * Argument syntax:
584 * - Add kprobe: p[:[GRP/]EVENT] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
585 * - Add kretprobe: r[:[GRP/]EVENT] [MOD:]KSYM[+0] [FETCHARGS]
586 * Fetch args:
587 * $retval : fetch return value
588 * $stack : fetch stack address
589 * $stackN : fetch Nth of stack (N:0-)
590 * @ADDR : fetch memory at ADDR (ADDR should be in kernel)
591 * @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
592 * %REG : fetch register REG
593 * Dereferencing memory fetch:
594 * +|-offs(ARG) : fetch memory at ARG +|- offs address.
595 * Alias name of args:
596 * NAME=FETCHARG : set NAME as alias of FETCHARG.
597 * Type of args:
598 * FETCHARG:TYPE : use TYPE instead of unsigned long.
599 */
600 struct trace_kprobe *tk;
601 int i, ret = 0;
602 bool is_return = false, is_delete = false;
603 char *symbol = NULL, *event = NULL, *group = NULL;
604 char *arg;
605 unsigned long offset = 0;
606 void *addr = NULL;
607 char buf[MAX_EVENT_NAME_LEN];
608
609 /* argc must be >= 1 */
610 if (argv[0][0] == 'p')
611 is_return = false;
612 else if (argv[0][0] == 'r')
613 is_return = true;
614 else if (argv[0][0] == '-')
615 is_delete = true;
616 else {
617 pr_info("Probe definition must be started with 'p', 'r' or"
618 " '-'.\n");
619 return -EINVAL;
620 }
621
622 if (argv[0][1] == ':') {
623 event = &argv[0][2];
624 if (strchr(event, '/')) {
625 group = event;
626 event = strchr(group, '/') + 1;
627 event[-1] = '\0';
628 if (strlen(group) == 0) {
629 pr_info("Group name is not specified\n");
630 return -EINVAL;
631 }
632 }
633 if (strlen(event) == 0) {
634 pr_info("Event name is not specified\n");
635 return -EINVAL;
636 }
637 }
638 if (!group)
639 group = KPROBE_EVENT_SYSTEM;
640
641 if (is_delete) {
642 if (!event) {
643 pr_info("Delete command needs an event name.\n");
644 return -EINVAL;
645 }
646 mutex_lock(&probe_lock);
647 tk = find_trace_kprobe(event, group);
648 if (!tk) {
649 mutex_unlock(&probe_lock);
650 pr_info("Event %s/%s doesn't exist.\n", group, event);
651 return -ENOENT;
652 }
653 /* delete an event */
654 ret = unregister_trace_kprobe(tk);
655 if (ret == 0)
656 free_trace_kprobe(tk);
657 mutex_unlock(&probe_lock);
658 return ret;
659 }
660
661 if (argc < 2) {
662 pr_info("Probe point is not specified.\n");
663 return -EINVAL;
664 }
665 if (isdigit(argv[1][0])) {
666 if (is_return) {
667 pr_info("Return probe point must be a symbol.\n");
668 return -EINVAL;
669 }
670 /* an address specified */
671 ret = kstrtoul(&argv[1][0], 0, (unsigned long *)&addr);
672 if (ret) {
673 pr_info("Failed to parse address.\n");
674 return ret;
675 }
676 } else {
677 /* a symbol specified */
678 symbol = argv[1];
679 /* TODO: support .init module functions */
680 ret = traceprobe_split_symbol_offset(symbol, &offset);
681 if (ret) {
682 pr_info("Failed to parse symbol.\n");
683 return ret;
684 }
685 if (offset && is_return) {
686 pr_info("Return probe must be used without offset.\n");
687 return -EINVAL;
688 }
689 }
690 argc -= 2; argv += 2;
691
692 /* setup a probe */
693 if (!event) {
694 /* Make a new event name */
695 if (symbol)
696 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
697 is_return ? 'r' : 'p', symbol, offset);
698 else
699 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
700 is_return ? 'r' : 'p', addr);
701 event = buf;
702 }
703 tk = alloc_trace_kprobe(group, event, addr, symbol, offset, argc,
704 is_return);
705 if (IS_ERR(tk)) {
706 pr_info("Failed to allocate trace_probe.(%d)\n",
707 (int)PTR_ERR(tk));
708 return PTR_ERR(tk);
709 }
710
711 /* parse arguments */
712 ret = 0;
713 for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
714 struct probe_arg *parg = &tk->tp.args[i];
715
716 /* Increment count for freeing args in error case */
717 tk->tp.nr_args++;
718
719 /* Parse argument name */
720 arg = strchr(argv[i], '=');
721 if (arg) {
722 *arg++ = '\0';
723 parg->name = kstrdup(argv[i], GFP_KERNEL);
724 } else {
725 arg = argv[i];
726 /* If argument name is omitted, set "argN" */
727 snprintf(buf, MAX_EVENT_NAME_LEN, "arg%d", i + 1);
728 parg->name = kstrdup(buf, GFP_KERNEL);
729 }
730
731 if (!parg->name) {
732 pr_info("Failed to allocate argument[%d] name.\n", i);
733 ret = -ENOMEM;
734 goto error;
735 }
736
737 if (!is_good_name(parg->name)) {
738 pr_info("Invalid argument[%d] name: %s\n",
739 i, parg->name);
740 ret = -EINVAL;
741 goto error;
742 }
743
744 if (traceprobe_conflict_field_name(parg->name,
745 tk->tp.args, i)) {
746 pr_info("Argument[%d] name '%s' conflicts with "
747 "another field.\n", i, argv[i]);
748 ret = -EINVAL;
749 goto error;
750 }
751
752 /* Parse fetch argument */
753 ret = traceprobe_parse_probe_arg(arg, &tk->tp.size, parg,
754 is_return, true,
755 kprobes_fetch_type_table);
756 if (ret) {
757 pr_info("Parse error at argument[%d]. (%d)\n", i, ret);
758 goto error;
759 }
760 }
761
762 ret = register_trace_kprobe(tk);
763 if (ret)
764 goto error;
765 return 0;
766
767error:
768 free_trace_kprobe(tk);
769 return ret;
770}
771
772static int release_all_trace_kprobes(void)
773{
774 struct trace_kprobe *tk;
775 int ret = 0;
776
777 mutex_lock(&probe_lock);
778 /* Ensure no probe is in use. */
779 list_for_each_entry(tk, &probe_list, list)
780 if (trace_probe_is_enabled(&tk->tp)) {
781 ret = -EBUSY;
782 goto end;
783 }
784 /* TODO: Use batch unregistration */
785 while (!list_empty(&probe_list)) {
786 tk = list_entry(probe_list.next, struct trace_kprobe, list);
787 ret = unregister_trace_kprobe(tk);
788 if (ret)
789 goto end;
790 free_trace_kprobe(tk);
791 }
792
793end:
794 mutex_unlock(&probe_lock);
795
796 return ret;
797}
798
799/* Probes listing interfaces */
800static void *probes_seq_start(struct seq_file *m, loff_t *pos)
801{
802 mutex_lock(&probe_lock);
803 return seq_list_start(&probe_list, *pos);
804}
805
806static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
807{
808 return seq_list_next(v, &probe_list, pos);
809}
810
811static void probes_seq_stop(struct seq_file *m, void *v)
812{
813 mutex_unlock(&probe_lock);
814}
815
816static int probes_seq_show(struct seq_file *m, void *v)
817{
818 struct trace_kprobe *tk = v;
819 int i;
820
821 seq_putc(m, trace_kprobe_is_return(tk) ? 'r' : 'p');
822 seq_printf(m, ":%s/%s", tk->tp.call.class->system,
823 trace_event_name(&tk->tp.call));
824
825 if (!tk->symbol)
826 seq_printf(m, " 0x%p", tk->rp.kp.addr);
827 else if (tk->rp.kp.offset)
828 seq_printf(m, " %s+%u", trace_kprobe_symbol(tk),
829 tk->rp.kp.offset);
830 else
831 seq_printf(m, " %s", trace_kprobe_symbol(tk));
832
833 for (i = 0; i < tk->tp.nr_args; i++)
834 seq_printf(m, " %s=%s", tk->tp.args[i].name, tk->tp.args[i].comm);
835 seq_putc(m, '\n');
836
837 return 0;
838}
839
840static const struct seq_operations probes_seq_op = {
841 .start = probes_seq_start,
842 .next = probes_seq_next,
843 .stop = probes_seq_stop,
844 .show = probes_seq_show
845};
846
847static int probes_open(struct inode *inode, struct file *file)
848{
849 int ret;
850
851 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
852 ret = release_all_trace_kprobes();
853 if (ret < 0)
854 return ret;
855 }
856
857 return seq_open(file, &probes_seq_op);
858}
859
860static ssize_t probes_write(struct file *file, const char __user *buffer,
861 size_t count, loff_t *ppos)
862{
863 return traceprobe_probes_write(file, buffer, count, ppos,
864 create_trace_kprobe);
865}
866
867static const struct file_operations kprobe_events_ops = {
868 .owner = THIS_MODULE,
869 .open = probes_open,
870 .read = seq_read,
871 .llseek = seq_lseek,
872 .release = seq_release,
873 .write = probes_write,
874};
875
876/* Probes profiling interfaces */
877static int probes_profile_seq_show(struct seq_file *m, void *v)
878{
879 struct trace_kprobe *tk = v;
880 unsigned long nhit = 0;
881 int cpu;
882
883 for_each_possible_cpu(cpu)
884 nhit += *per_cpu_ptr(tk->nhit, cpu);
885
886 seq_printf(m, " %-44s %15lu %15lu\n",
887 trace_event_name(&tk->tp.call), nhit,
888 tk->rp.kp.nmissed);
889
890 return 0;
891}
892
893static const struct seq_operations profile_seq_op = {
894 .start = probes_seq_start,
895 .next = probes_seq_next,
896 .stop = probes_seq_stop,
897 .show = probes_profile_seq_show
898};
899
900static int profile_open(struct inode *inode, struct file *file)
901{
902 return seq_open(file, &profile_seq_op);
903}
904
905static const struct file_operations kprobe_profile_ops = {
906 .owner = THIS_MODULE,
907 .open = profile_open,
908 .read = seq_read,
909 .llseek = seq_lseek,
910 .release = seq_release,
911};
912
913/* Kprobe handler */
914static nokprobe_inline void
915__kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs,
916 struct trace_event_file *trace_file)
917{
918 struct kprobe_trace_entry_head *entry;
919 struct ring_buffer_event *event;
920 struct ring_buffer *buffer;
921 int size, dsize, pc;
922 unsigned long irq_flags;
923 struct trace_event_call *call = &tk->tp.call;
924
925 WARN_ON(call != trace_file->event_call);
926
927 if (trace_trigger_soft_disabled(trace_file))
928 return;
929
930 local_save_flags(irq_flags);
931 pc = preempt_count();
932
933 dsize = __get_data_size(&tk->tp, regs);
934 size = sizeof(*entry) + tk->tp.size + dsize;
935
936 event = trace_event_buffer_lock_reserve(&buffer, trace_file,
937 call->event.type,
938 size, irq_flags, pc);
939 if (!event)
940 return;
941
942 entry = ring_buffer_event_data(event);
943 entry->ip = (unsigned long)tk->rp.kp.addr;
944 store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
945
946 event_trigger_unlock_commit_regs(trace_file, buffer, event,
947 entry, irq_flags, pc, regs);
948}
949
950static void
951kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs)
952{
953 struct event_file_link *link;
954
955 list_for_each_entry_rcu(link, &tk->tp.files, list)
956 __kprobe_trace_func(tk, regs, link->file);
957}
958NOKPROBE_SYMBOL(kprobe_trace_func);
959
960/* Kretprobe handler */
961static nokprobe_inline void
962__kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
963 struct pt_regs *regs,
964 struct trace_event_file *trace_file)
965{
966 struct kretprobe_trace_entry_head *entry;
967 struct ring_buffer_event *event;
968 struct ring_buffer *buffer;
969 int size, pc, dsize;
970 unsigned long irq_flags;
971 struct trace_event_call *call = &tk->tp.call;
972
973 WARN_ON(call != trace_file->event_call);
974
975 if (trace_trigger_soft_disabled(trace_file))
976 return;
977
978 local_save_flags(irq_flags);
979 pc = preempt_count();
980
981 dsize = __get_data_size(&tk->tp, regs);
982 size = sizeof(*entry) + tk->tp.size + dsize;
983
984 event = trace_event_buffer_lock_reserve(&buffer, trace_file,
985 call->event.type,
986 size, irq_flags, pc);
987 if (!event)
988 return;
989
990 entry = ring_buffer_event_data(event);
991 entry->func = (unsigned long)tk->rp.kp.addr;
992 entry->ret_ip = (unsigned long)ri->ret_addr;
993 store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
994
995 event_trigger_unlock_commit_regs(trace_file, buffer, event,
996 entry, irq_flags, pc, regs);
997}
998
999static void
1000kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1001 struct pt_regs *regs)
1002{
1003 struct event_file_link *link;
1004
1005 list_for_each_entry_rcu(link, &tk->tp.files, list)
1006 __kretprobe_trace_func(tk, ri, regs, link->file);
1007}
1008NOKPROBE_SYMBOL(kretprobe_trace_func);
1009
1010/* Event entry printers */
1011static enum print_line_t
1012print_kprobe_event(struct trace_iterator *iter, int flags,
1013 struct trace_event *event)
1014{
1015 struct kprobe_trace_entry_head *field;
1016 struct trace_seq *s = &iter->seq;
1017 struct trace_probe *tp;
1018 u8 *data;
1019 int i;
1020
1021 field = (struct kprobe_trace_entry_head *)iter->ent;
1022 tp = container_of(event, struct trace_probe, call.event);
1023
1024 trace_seq_printf(s, "%s: (", trace_event_name(&tp->call));
1025
1026 if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
1027 goto out;
1028
1029 trace_seq_putc(s, ')');
1030
1031 data = (u8 *)&field[1];
1032 for (i = 0; i < tp->nr_args; i++)
1033 if (!tp->args[i].type->print(s, tp->args[i].name,
1034 data + tp->args[i].offset, field))
1035 goto out;
1036
1037 trace_seq_putc(s, '\n');
1038 out:
1039 return trace_handle_return(s);
1040}
1041
1042static enum print_line_t
1043print_kretprobe_event(struct trace_iterator *iter, int flags,
1044 struct trace_event *event)
1045{
1046 struct kretprobe_trace_entry_head *field;
1047 struct trace_seq *s = &iter->seq;
1048 struct trace_probe *tp;
1049 u8 *data;
1050 int i;
1051
1052 field = (struct kretprobe_trace_entry_head *)iter->ent;
1053 tp = container_of(event, struct trace_probe, call.event);
1054
1055 trace_seq_printf(s, "%s: (", trace_event_name(&tp->call));
1056
1057 if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1058 goto out;
1059
1060 trace_seq_puts(s, " <- ");
1061
1062 if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1063 goto out;
1064
1065 trace_seq_putc(s, ')');
1066
1067 data = (u8 *)&field[1];
1068 for (i = 0; i < tp->nr_args; i++)
1069 if (!tp->args[i].type->print(s, tp->args[i].name,
1070 data + tp->args[i].offset, field))
1071 goto out;
1072
1073 trace_seq_putc(s, '\n');
1074
1075 out:
1076 return trace_handle_return(s);
1077}
1078
1079
1080static int kprobe_event_define_fields(struct trace_event_call *event_call)
1081{
1082 int ret, i;
1083 struct kprobe_trace_entry_head field;
1084 struct trace_kprobe *tk = (struct trace_kprobe *)event_call->data;
1085
1086 DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1087 /* Set argument names as fields */
1088 for (i = 0; i < tk->tp.nr_args; i++) {
1089 struct probe_arg *parg = &tk->tp.args[i];
1090
1091 ret = trace_define_field(event_call, parg->type->fmttype,
1092 parg->name,
1093 sizeof(field) + parg->offset,
1094 parg->type->size,
1095 parg->type->is_signed,
1096 FILTER_OTHER);
1097 if (ret)
1098 return ret;
1099 }
1100 return 0;
1101}
1102
1103static int kretprobe_event_define_fields(struct trace_event_call *event_call)
1104{
1105 int ret, i;
1106 struct kretprobe_trace_entry_head field;
1107 struct trace_kprobe *tk = (struct trace_kprobe *)event_call->data;
1108
1109 DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1110 DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1111 /* Set argument names as fields */
1112 for (i = 0; i < tk->tp.nr_args; i++) {
1113 struct probe_arg *parg = &tk->tp.args[i];
1114
1115 ret = trace_define_field(event_call, parg->type->fmttype,
1116 parg->name,
1117 sizeof(field) + parg->offset,
1118 parg->type->size,
1119 parg->type->is_signed,
1120 FILTER_OTHER);
1121 if (ret)
1122 return ret;
1123 }
1124 return 0;
1125}
1126
1127#ifdef CONFIG_PERF_EVENTS
1128
1129/* Kprobe profile handler */
1130static void
1131kprobe_perf_func(struct trace_kprobe *tk, struct pt_regs *regs)
1132{
1133 struct trace_event_call *call = &tk->tp.call;
1134 struct bpf_prog *prog = call->prog;
1135 struct kprobe_trace_entry_head *entry;
1136 struct hlist_head *head;
1137 int size, __size, dsize;
1138 int rctx;
1139
1140 if (prog && !trace_call_bpf(prog, regs))
1141 return;
1142
1143 head = this_cpu_ptr(call->perf_events);
1144 if (hlist_empty(head))
1145 return;
1146
1147 dsize = __get_data_size(&tk->tp, regs);
1148 __size = sizeof(*entry) + tk->tp.size + dsize;
1149 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1150 size -= sizeof(u32);
1151
1152 entry = perf_trace_buf_prepare(size, call->event.type, NULL, &rctx);
1153 if (!entry)
1154 return;
1155
1156 entry->ip = (unsigned long)tk->rp.kp.addr;
1157 memset(&entry[1], 0, dsize);
1158 store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1159 perf_trace_buf_submit(entry, size, rctx, 0, 1, regs, head, NULL);
1160}
1161NOKPROBE_SYMBOL(kprobe_perf_func);
1162
1163/* Kretprobe profile handler */
1164static void
1165kretprobe_perf_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1166 struct pt_regs *regs)
1167{
1168 struct trace_event_call *call = &tk->tp.call;
1169 struct bpf_prog *prog = call->prog;
1170 struct kretprobe_trace_entry_head *entry;
1171 struct hlist_head *head;
1172 int size, __size, dsize;
1173 int rctx;
1174
1175 if (prog && !trace_call_bpf(prog, regs))
1176 return;
1177
1178 head = this_cpu_ptr(call->perf_events);
1179 if (hlist_empty(head))
1180 return;
1181
1182 dsize = __get_data_size(&tk->tp, regs);
1183 __size = sizeof(*entry) + tk->tp.size + dsize;
1184 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1185 size -= sizeof(u32);
1186
1187 entry = perf_trace_buf_prepare(size, call->event.type, NULL, &rctx);
1188 if (!entry)
1189 return;
1190
1191 entry->func = (unsigned long)tk->rp.kp.addr;
1192 entry->ret_ip = (unsigned long)ri->ret_addr;
1193 store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1194 perf_trace_buf_submit(entry, size, rctx, 0, 1, regs, head, NULL);
1195}
1196NOKPROBE_SYMBOL(kretprobe_perf_func);
1197#endif /* CONFIG_PERF_EVENTS */
1198
1199/*
1200 * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex.
1201 *
1202 * kprobe_trace_self_tests_init() does enable_trace_probe/disable_trace_probe
1203 * lockless, but we can't race with this __init function.
1204 */
1205static int kprobe_register(struct trace_event_call *event,
1206 enum trace_reg type, void *data)
1207{
1208 struct trace_kprobe *tk = (struct trace_kprobe *)event->data;
1209 struct trace_event_file *file = data;
1210
1211 switch (type) {
1212 case TRACE_REG_REGISTER:
1213 return enable_trace_kprobe(tk, file);
1214 case TRACE_REG_UNREGISTER:
1215 return disable_trace_kprobe(tk, file);
1216
1217#ifdef CONFIG_PERF_EVENTS
1218 case TRACE_REG_PERF_REGISTER:
1219 return enable_trace_kprobe(tk, NULL);
1220 case TRACE_REG_PERF_UNREGISTER:
1221 return disable_trace_kprobe(tk, NULL);
1222 case TRACE_REG_PERF_OPEN:
1223 case TRACE_REG_PERF_CLOSE:
1224 case TRACE_REG_PERF_ADD:
1225 case TRACE_REG_PERF_DEL:
1226 return 0;
1227#endif
1228 }
1229 return 0;
1230}
1231
1232static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1233{
1234 struct trace_kprobe *tk = container_of(kp, struct trace_kprobe, rp.kp);
1235
1236 raw_cpu_inc(*tk->nhit);
1237
1238 if (tk->tp.flags & TP_FLAG_TRACE)
1239 kprobe_trace_func(tk, regs);
1240#ifdef CONFIG_PERF_EVENTS
1241 if (tk->tp.flags & TP_FLAG_PROFILE)
1242 kprobe_perf_func(tk, regs);
1243#endif
1244 return 0; /* We don't tweek kernel, so just return 0 */
1245}
1246NOKPROBE_SYMBOL(kprobe_dispatcher);
1247
1248static int
1249kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1250{
1251 struct trace_kprobe *tk = container_of(ri->rp, struct trace_kprobe, rp);
1252
1253 raw_cpu_inc(*tk->nhit);
1254
1255 if (tk->tp.flags & TP_FLAG_TRACE)
1256 kretprobe_trace_func(tk, ri, regs);
1257#ifdef CONFIG_PERF_EVENTS
1258 if (tk->tp.flags & TP_FLAG_PROFILE)
1259 kretprobe_perf_func(tk, ri, regs);
1260#endif
1261 return 0; /* We don't tweek kernel, so just return 0 */
1262}
1263NOKPROBE_SYMBOL(kretprobe_dispatcher);
1264
1265static struct trace_event_functions kretprobe_funcs = {
1266 .trace = print_kretprobe_event
1267};
1268
1269static struct trace_event_functions kprobe_funcs = {
1270 .trace = print_kprobe_event
1271};
1272
1273static int register_kprobe_event(struct trace_kprobe *tk)
1274{
1275 struct trace_event_call *call = &tk->tp.call;
1276 int ret;
1277
1278 /* Initialize trace_event_call */
1279 INIT_LIST_HEAD(&call->class->fields);
1280 if (trace_kprobe_is_return(tk)) {
1281 call->event.funcs = &kretprobe_funcs;
1282 call->class->define_fields = kretprobe_event_define_fields;
1283 } else {
1284 call->event.funcs = &kprobe_funcs;
1285 call->class->define_fields = kprobe_event_define_fields;
1286 }
1287 if (set_print_fmt(&tk->tp, trace_kprobe_is_return(tk)) < 0)
1288 return -ENOMEM;
1289 ret = register_trace_event(&call->event);
1290 if (!ret) {
1291 kfree(call->print_fmt);
1292 return -ENODEV;
1293 }
1294 call->flags = TRACE_EVENT_FL_KPROBE;
1295 call->class->reg = kprobe_register;
1296 call->data = tk;
1297 ret = trace_add_event_call(call);
1298 if (ret) {
1299 pr_info("Failed to register kprobe event: %s\n",
1300 trace_event_name(call));
1301 kfree(call->print_fmt);
1302 unregister_trace_event(&call->event);
1303 }
1304 return ret;
1305}
1306
1307static int unregister_kprobe_event(struct trace_kprobe *tk)
1308{
1309 int ret;
1310
1311 /* tp->event is unregistered in trace_remove_event_call() */
1312 ret = trace_remove_event_call(&tk->tp.call);
1313 if (!ret)
1314 kfree(tk->tp.call.print_fmt);
1315 return ret;
1316}
1317
1318/* Make a tracefs interface for controlling probe points */
1319static __init int init_kprobe_trace(void)
1320{
1321 struct dentry *d_tracer;
1322 struct dentry *entry;
1323
1324 if (register_module_notifier(&trace_kprobe_module_nb))
1325 return -EINVAL;
1326
1327 d_tracer = tracing_init_dentry();
1328 if (IS_ERR(d_tracer))
1329 return 0;
1330
1331 entry = tracefs_create_file("kprobe_events", 0644, d_tracer,
1332 NULL, &kprobe_events_ops);
1333
1334 /* Event list interface */
1335 if (!entry)
1336 pr_warn("Could not create tracefs 'kprobe_events' entry\n");
1337
1338 /* Profile interface */
1339 entry = tracefs_create_file("kprobe_profile", 0444, d_tracer,
1340 NULL, &kprobe_profile_ops);
1341
1342 if (!entry)
1343 pr_warn("Could not create tracefs 'kprobe_profile' entry\n");
1344 return 0;
1345}
1346fs_initcall(init_kprobe_trace);
1347
1348
1349#ifdef CONFIG_FTRACE_STARTUP_TEST
1350
1351/*
1352 * The "__used" keeps gcc from removing the function symbol
1353 * from the kallsyms table.
1354 */
1355static __used int kprobe_trace_selftest_target(int a1, int a2, int a3,
1356 int a4, int a5, int a6)
1357{
1358 return a1 + a2 + a3 + a4 + a5 + a6;
1359}
1360
1361static struct trace_event_file *
1362find_trace_probe_file(struct trace_kprobe *tk, struct trace_array *tr)
1363{
1364 struct trace_event_file *file;
1365
1366 list_for_each_entry(file, &tr->events, list)
1367 if (file->event_call == &tk->tp.call)
1368 return file;
1369
1370 return NULL;
1371}
1372
1373/*
1374 * Nobody but us can call enable_trace_kprobe/disable_trace_kprobe at this
1375 * stage, we can do this lockless.
1376 */
1377static __init int kprobe_trace_self_tests_init(void)
1378{
1379 int ret, warn = 0;
1380 int (*target)(int, int, int, int, int, int);
1381 struct trace_kprobe *tk;
1382 struct trace_event_file *file;
1383
1384 if (tracing_is_disabled())
1385 return -ENODEV;
1386
1387 target = kprobe_trace_selftest_target;
1388
1389 pr_info("Testing kprobe tracing: ");
1390
1391 ret = traceprobe_command("p:testprobe kprobe_trace_selftest_target "
1392 "$stack $stack0 +0($stack)",
1393 create_trace_kprobe);
1394 if (WARN_ON_ONCE(ret)) {
1395 pr_warn("error on probing function entry.\n");
1396 warn++;
1397 } else {
1398 /* Enable trace point */
1399 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1400 if (WARN_ON_ONCE(tk == NULL)) {
1401 pr_warn("error on getting new probe.\n");
1402 warn++;
1403 } else {
1404 file = find_trace_probe_file(tk, top_trace_array());
1405 if (WARN_ON_ONCE(file == NULL)) {
1406 pr_warn("error on getting probe file.\n");
1407 warn++;
1408 } else
1409 enable_trace_kprobe(tk, file);
1410 }
1411 }
1412
1413 ret = traceprobe_command("r:testprobe2 kprobe_trace_selftest_target "
1414 "$retval", create_trace_kprobe);
1415 if (WARN_ON_ONCE(ret)) {
1416 pr_warn("error on probing function return.\n");
1417 warn++;
1418 } else {
1419 /* Enable trace point */
1420 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1421 if (WARN_ON_ONCE(tk == NULL)) {
1422 pr_warn("error on getting 2nd new probe.\n");
1423 warn++;
1424 } else {
1425 file = find_trace_probe_file(tk, top_trace_array());
1426 if (WARN_ON_ONCE(file == NULL)) {
1427 pr_warn("error on getting probe file.\n");
1428 warn++;
1429 } else
1430 enable_trace_kprobe(tk, file);
1431 }
1432 }
1433
1434 if (warn)
1435 goto end;
1436
1437 ret = target(1, 2, 3, 4, 5, 6);
1438
1439 /* Disable trace points before removing it */
1440 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1441 if (WARN_ON_ONCE(tk == NULL)) {
1442 pr_warn("error on getting test probe.\n");
1443 warn++;
1444 } else {
1445 file = find_trace_probe_file(tk, top_trace_array());
1446 if (WARN_ON_ONCE(file == NULL)) {
1447 pr_warn("error on getting probe file.\n");
1448 warn++;
1449 } else
1450 disable_trace_kprobe(tk, file);
1451 }
1452
1453 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1454 if (WARN_ON_ONCE(tk == NULL)) {
1455 pr_warn("error on getting 2nd test probe.\n");
1456 warn++;
1457 } else {
1458 file = find_trace_probe_file(tk, top_trace_array());
1459 if (WARN_ON_ONCE(file == NULL)) {
1460 pr_warn("error on getting probe file.\n");
1461 warn++;
1462 } else
1463 disable_trace_kprobe(tk, file);
1464 }
1465
1466 ret = traceprobe_command("-:testprobe", create_trace_kprobe);
1467 if (WARN_ON_ONCE(ret)) {
1468 pr_warn("error on deleting a probe.\n");
1469 warn++;
1470 }
1471
1472 ret = traceprobe_command("-:testprobe2", create_trace_kprobe);
1473 if (WARN_ON_ONCE(ret)) {
1474 pr_warn("error on deleting a probe.\n");
1475 warn++;
1476 }
1477
1478end:
1479 release_all_trace_kprobes();
1480 if (warn)
1481 pr_cont("NG: Some tests are failed. Please check them.\n");
1482 else
1483 pr_cont("OK\n");
1484 return 0;
1485}
1486
1487late_initcall(kprobe_trace_self_tests_init);
1488
1489#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#define MAX_KPROBE_CMDLINE_SIZE 1024
26
27/* Kprobe early definition from command line */
28static char kprobe_boot_events_buf[COMMAND_LINE_SIZE] __initdata;
29static bool kprobe_boot_events_enabled __initdata;
30
31static int __init set_kprobe_boot_events(char *str)
32{
33 strlcpy(kprobe_boot_events_buf, str, COMMAND_LINE_SIZE);
34 return 0;
35}
36__setup("kprobe_event=", set_kprobe_boot_events);
37
38static int trace_kprobe_create(int argc, const char **argv);
39static int trace_kprobe_show(struct seq_file *m, struct dyn_event *ev);
40static int trace_kprobe_release(struct dyn_event *ev);
41static bool trace_kprobe_is_busy(struct dyn_event *ev);
42static bool trace_kprobe_match(const char *system, const char *event,
43 int argc, const char **argv, struct dyn_event *ev);
44
45static struct dyn_event_operations trace_kprobe_ops = {
46 .create = trace_kprobe_create,
47 .show = trace_kprobe_show,
48 .is_busy = trace_kprobe_is_busy,
49 .free = trace_kprobe_release,
50 .match = trace_kprobe_match,
51};
52
53/*
54 * Kprobe event core functions
55 */
56struct trace_kprobe {
57 struct dyn_event devent;
58 struct kretprobe rp; /* Use rp.kp for kprobe use */
59 unsigned long __percpu *nhit;
60 const char *symbol; /* symbol name */
61 struct trace_probe tp;
62};
63
64static bool is_trace_kprobe(struct dyn_event *ev)
65{
66 return ev->ops == &trace_kprobe_ops;
67}
68
69static struct trace_kprobe *to_trace_kprobe(struct dyn_event *ev)
70{
71 return container_of(ev, struct trace_kprobe, devent);
72}
73
74/**
75 * for_each_trace_kprobe - iterate over the trace_kprobe list
76 * @pos: the struct trace_kprobe * for each entry
77 * @dpos: the struct dyn_event * to use as a loop cursor
78 */
79#define for_each_trace_kprobe(pos, dpos) \
80 for_each_dyn_event(dpos) \
81 if (is_trace_kprobe(dpos) && (pos = to_trace_kprobe(dpos)))
82
83#define SIZEOF_TRACE_KPROBE(n) \
84 (offsetof(struct trace_kprobe, tp.args) + \
85 (sizeof(struct probe_arg) * (n)))
86
87static nokprobe_inline bool trace_kprobe_is_return(struct trace_kprobe *tk)
88{
89 return tk->rp.handler != NULL;
90}
91
92static nokprobe_inline const char *trace_kprobe_symbol(struct trace_kprobe *tk)
93{
94 return tk->symbol ? tk->symbol : "unknown";
95}
96
97static nokprobe_inline unsigned long trace_kprobe_offset(struct trace_kprobe *tk)
98{
99 return tk->rp.kp.offset;
100}
101
102static nokprobe_inline bool trace_kprobe_has_gone(struct trace_kprobe *tk)
103{
104 return !!(kprobe_gone(&tk->rp.kp));
105}
106
107static nokprobe_inline bool trace_kprobe_within_module(struct trace_kprobe *tk,
108 struct module *mod)
109{
110 int len = strlen(mod->name);
111 const char *name = trace_kprobe_symbol(tk);
112 return strncmp(mod->name, name, len) == 0 && name[len] == ':';
113}
114
115static nokprobe_inline bool trace_kprobe_module_exist(struct trace_kprobe *tk)
116{
117 char *p;
118 bool ret;
119
120 if (!tk->symbol)
121 return false;
122 p = strchr(tk->symbol, ':');
123 if (!p)
124 return true;
125 *p = '\0';
126 mutex_lock(&module_mutex);
127 ret = !!find_module(tk->symbol);
128 mutex_unlock(&module_mutex);
129 *p = ':';
130
131 return ret;
132}
133
134static bool trace_kprobe_is_busy(struct dyn_event *ev)
135{
136 struct trace_kprobe *tk = to_trace_kprobe(ev);
137
138 return trace_probe_is_enabled(&tk->tp);
139}
140
141static bool trace_kprobe_match_command_head(struct trace_kprobe *tk,
142 int argc, const char **argv)
143{
144 char buf[MAX_ARGSTR_LEN + 1];
145
146 if (!argc)
147 return true;
148
149 if (!tk->symbol)
150 snprintf(buf, sizeof(buf), "0x%p", tk->rp.kp.addr);
151 else if (tk->rp.kp.offset)
152 snprintf(buf, sizeof(buf), "%s+%u",
153 trace_kprobe_symbol(tk), tk->rp.kp.offset);
154 else
155 snprintf(buf, sizeof(buf), "%s", trace_kprobe_symbol(tk));
156 if (strcmp(buf, argv[0]))
157 return false;
158 argc--; argv++;
159
160 return trace_probe_match_command_args(&tk->tp, argc, argv);
161}
162
163static bool trace_kprobe_match(const char *system, const char *event,
164 int argc, const char **argv, struct dyn_event *ev)
165{
166 struct trace_kprobe *tk = to_trace_kprobe(ev);
167
168 return strcmp(trace_probe_name(&tk->tp), event) == 0 &&
169 (!system || strcmp(trace_probe_group_name(&tk->tp), system) == 0) &&
170 trace_kprobe_match_command_head(tk, argc, argv);
171}
172
173static nokprobe_inline unsigned long trace_kprobe_nhit(struct trace_kprobe *tk)
174{
175 unsigned long nhit = 0;
176 int cpu;
177
178 for_each_possible_cpu(cpu)
179 nhit += *per_cpu_ptr(tk->nhit, cpu);
180
181 return nhit;
182}
183
184static nokprobe_inline bool trace_kprobe_is_registered(struct trace_kprobe *tk)
185{
186 return !(list_empty(&tk->rp.kp.list) &&
187 hlist_unhashed(&tk->rp.kp.hlist));
188}
189
190/* Return 0 if it fails to find the symbol address */
191static nokprobe_inline
192unsigned long trace_kprobe_address(struct trace_kprobe *tk)
193{
194 unsigned long addr;
195
196 if (tk->symbol) {
197 addr = (unsigned long)
198 kallsyms_lookup_name(trace_kprobe_symbol(tk));
199 if (addr)
200 addr += tk->rp.kp.offset;
201 } else {
202 addr = (unsigned long)tk->rp.kp.addr;
203 }
204 return addr;
205}
206
207static nokprobe_inline struct trace_kprobe *
208trace_kprobe_primary_from_call(struct trace_event_call *call)
209{
210 struct trace_probe *tp;
211
212 tp = trace_probe_primary_from_call(call);
213 if (WARN_ON_ONCE(!tp))
214 return NULL;
215
216 return container_of(tp, struct trace_kprobe, tp);
217}
218
219bool trace_kprobe_on_func_entry(struct trace_event_call *call)
220{
221 struct trace_kprobe *tk = trace_kprobe_primary_from_call(call);
222
223 return tk ? kprobe_on_func_entry(tk->rp.kp.addr,
224 tk->rp.kp.addr ? NULL : tk->rp.kp.symbol_name,
225 tk->rp.kp.addr ? 0 : tk->rp.kp.offset) : false;
226}
227
228bool trace_kprobe_error_injectable(struct trace_event_call *call)
229{
230 struct trace_kprobe *tk = trace_kprobe_primary_from_call(call);
231
232 return tk ? within_error_injection_list(trace_kprobe_address(tk)) :
233 false;
234}
235
236static int register_kprobe_event(struct trace_kprobe *tk);
237static int unregister_kprobe_event(struct trace_kprobe *tk);
238
239static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
240static int kretprobe_dispatcher(struct kretprobe_instance *ri,
241 struct pt_regs *regs);
242
243static void free_trace_kprobe(struct trace_kprobe *tk)
244{
245 if (tk) {
246 trace_probe_cleanup(&tk->tp);
247 kfree(tk->symbol);
248 free_percpu(tk->nhit);
249 kfree(tk);
250 }
251}
252
253/*
254 * Allocate new trace_probe and initialize it (including kprobes).
255 */
256static struct trace_kprobe *alloc_trace_kprobe(const char *group,
257 const char *event,
258 void *addr,
259 const char *symbol,
260 unsigned long offs,
261 int maxactive,
262 int nargs, bool is_return)
263{
264 struct trace_kprobe *tk;
265 int ret = -ENOMEM;
266
267 tk = kzalloc(SIZEOF_TRACE_KPROBE(nargs), GFP_KERNEL);
268 if (!tk)
269 return ERR_PTR(ret);
270
271 tk->nhit = alloc_percpu(unsigned long);
272 if (!tk->nhit)
273 goto error;
274
275 if (symbol) {
276 tk->symbol = kstrdup(symbol, GFP_KERNEL);
277 if (!tk->symbol)
278 goto error;
279 tk->rp.kp.symbol_name = tk->symbol;
280 tk->rp.kp.offset = offs;
281 } else
282 tk->rp.kp.addr = addr;
283
284 if (is_return)
285 tk->rp.handler = kretprobe_dispatcher;
286 else
287 tk->rp.kp.pre_handler = kprobe_dispatcher;
288
289 tk->rp.maxactive = maxactive;
290 INIT_HLIST_NODE(&tk->rp.kp.hlist);
291 INIT_LIST_HEAD(&tk->rp.kp.list);
292
293 ret = trace_probe_init(&tk->tp, event, group);
294 if (ret < 0)
295 goto error;
296
297 dyn_event_init(&tk->devent, &trace_kprobe_ops);
298 return tk;
299error:
300 free_trace_kprobe(tk);
301 return ERR_PTR(ret);
302}
303
304static struct trace_kprobe *find_trace_kprobe(const char *event,
305 const char *group)
306{
307 struct dyn_event *pos;
308 struct trace_kprobe *tk;
309
310 for_each_trace_kprobe(tk, pos)
311 if (strcmp(trace_probe_name(&tk->tp), event) == 0 &&
312 strcmp(trace_probe_group_name(&tk->tp), group) == 0)
313 return tk;
314 return NULL;
315}
316
317static inline int __enable_trace_kprobe(struct trace_kprobe *tk)
318{
319 int ret = 0;
320
321 if (trace_kprobe_is_registered(tk) && !trace_kprobe_has_gone(tk)) {
322 if (trace_kprobe_is_return(tk))
323 ret = enable_kretprobe(&tk->rp);
324 else
325 ret = enable_kprobe(&tk->rp.kp);
326 }
327
328 return ret;
329}
330
331static void __disable_trace_kprobe(struct trace_probe *tp)
332{
333 struct trace_probe *pos;
334 struct trace_kprobe *tk;
335
336 list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
337 tk = container_of(pos, struct trace_kprobe, tp);
338 if (!trace_kprobe_is_registered(tk))
339 continue;
340 if (trace_kprobe_is_return(tk))
341 disable_kretprobe(&tk->rp);
342 else
343 disable_kprobe(&tk->rp.kp);
344 }
345}
346
347/*
348 * Enable trace_probe
349 * if the file is NULL, enable "perf" handler, or enable "trace" handler.
350 */
351static int enable_trace_kprobe(struct trace_event_call *call,
352 struct trace_event_file *file)
353{
354 struct trace_probe *pos, *tp;
355 struct trace_kprobe *tk;
356 bool enabled;
357 int ret = 0;
358
359 tp = trace_probe_primary_from_call(call);
360 if (WARN_ON_ONCE(!tp))
361 return -ENODEV;
362 enabled = trace_probe_is_enabled(tp);
363
364 /* This also changes "enabled" state */
365 if (file) {
366 ret = trace_probe_add_file(tp, file);
367 if (ret)
368 return ret;
369 } else
370 trace_probe_set_flag(tp, TP_FLAG_PROFILE);
371
372 if (enabled)
373 return 0;
374
375 list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
376 tk = container_of(pos, struct trace_kprobe, tp);
377 if (trace_kprobe_has_gone(tk))
378 continue;
379 ret = __enable_trace_kprobe(tk);
380 if (ret)
381 break;
382 enabled = true;
383 }
384
385 if (ret) {
386 /* Failed to enable one of them. Roll back all */
387 if (enabled)
388 __disable_trace_kprobe(tp);
389 if (file)
390 trace_probe_remove_file(tp, file);
391 else
392 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
393 }
394
395 return ret;
396}
397
398/*
399 * Disable trace_probe
400 * if the file is NULL, disable "perf" handler, or disable "trace" handler.
401 */
402static int disable_trace_kprobe(struct trace_event_call *call,
403 struct trace_event_file *file)
404{
405 struct trace_probe *tp;
406
407 tp = trace_probe_primary_from_call(call);
408 if (WARN_ON_ONCE(!tp))
409 return -ENODEV;
410
411 if (file) {
412 if (!trace_probe_get_file_link(tp, file))
413 return -ENOENT;
414 if (!trace_probe_has_single_file(tp))
415 goto out;
416 trace_probe_clear_flag(tp, TP_FLAG_TRACE);
417 } else
418 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
419
420 if (!trace_probe_is_enabled(tp))
421 __disable_trace_kprobe(tp);
422
423 out:
424 if (file)
425 /*
426 * Synchronization is done in below function. For perf event,
427 * file == NULL and perf_trace_event_unreg() calls
428 * tracepoint_synchronize_unregister() to ensure synchronize
429 * event. We don't need to care about it.
430 */
431 trace_probe_remove_file(tp, file);
432
433 return 0;
434}
435
436#if defined(CONFIG_KPROBES_ON_FTRACE) && \
437 !defined(CONFIG_KPROBE_EVENTS_ON_NOTRACE)
438static bool within_notrace_func(struct trace_kprobe *tk)
439{
440 unsigned long offset, size, addr;
441
442 addr = trace_kprobe_address(tk);
443 if (!addr || !kallsyms_lookup_size_offset(addr, &size, &offset))
444 return false;
445
446 /* Get the entry address of the target function */
447 addr -= offset;
448
449 /*
450 * Since ftrace_location_range() does inclusive range check, we need
451 * to subtract 1 byte from the end address.
452 */
453 return !ftrace_location_range(addr, addr + size - 1);
454}
455#else
456#define within_notrace_func(tk) (false)
457#endif
458
459/* Internal register function - just handle k*probes and flags */
460static int __register_trace_kprobe(struct trace_kprobe *tk)
461{
462 int i, ret;
463
464 ret = security_locked_down(LOCKDOWN_KPROBES);
465 if (ret)
466 return ret;
467
468 if (trace_kprobe_is_registered(tk))
469 return -EINVAL;
470
471 if (within_notrace_func(tk)) {
472 pr_warn("Could not probe notrace function %s\n",
473 trace_kprobe_symbol(tk));
474 return -EINVAL;
475 }
476
477 for (i = 0; i < tk->tp.nr_args; i++) {
478 ret = traceprobe_update_arg(&tk->tp.args[i]);
479 if (ret)
480 return ret;
481 }
482
483 /* Set/clear disabled flag according to tp->flag */
484 if (trace_probe_is_enabled(&tk->tp))
485 tk->rp.kp.flags &= ~KPROBE_FLAG_DISABLED;
486 else
487 tk->rp.kp.flags |= KPROBE_FLAG_DISABLED;
488
489 if (trace_kprobe_is_return(tk))
490 ret = register_kretprobe(&tk->rp);
491 else
492 ret = register_kprobe(&tk->rp.kp);
493
494 return ret;
495}
496
497/* Internal unregister function - just handle k*probes and flags */
498static void __unregister_trace_kprobe(struct trace_kprobe *tk)
499{
500 if (trace_kprobe_is_registered(tk)) {
501 if (trace_kprobe_is_return(tk))
502 unregister_kretprobe(&tk->rp);
503 else
504 unregister_kprobe(&tk->rp.kp);
505 /* Cleanup kprobe for reuse and mark it unregistered */
506 INIT_HLIST_NODE(&tk->rp.kp.hlist);
507 INIT_LIST_HEAD(&tk->rp.kp.list);
508 if (tk->rp.kp.symbol_name)
509 tk->rp.kp.addr = NULL;
510 }
511}
512
513/* Unregister a trace_probe and probe_event */
514static int unregister_trace_kprobe(struct trace_kprobe *tk)
515{
516 /* If other probes are on the event, just unregister kprobe */
517 if (trace_probe_has_sibling(&tk->tp))
518 goto unreg;
519
520 /* Enabled event can not be unregistered */
521 if (trace_probe_is_enabled(&tk->tp))
522 return -EBUSY;
523
524 /* Will fail if probe is being used by ftrace or perf */
525 if (unregister_kprobe_event(tk))
526 return -EBUSY;
527
528unreg:
529 __unregister_trace_kprobe(tk);
530 dyn_event_remove(&tk->devent);
531 trace_probe_unlink(&tk->tp);
532
533 return 0;
534}
535
536static bool trace_kprobe_has_same_kprobe(struct trace_kprobe *orig,
537 struct trace_kprobe *comp)
538{
539 struct trace_probe_event *tpe = orig->tp.event;
540 struct trace_probe *pos;
541 int i;
542
543 list_for_each_entry(pos, &tpe->probes, list) {
544 orig = container_of(pos, struct trace_kprobe, tp);
545 if (strcmp(trace_kprobe_symbol(orig),
546 trace_kprobe_symbol(comp)) ||
547 trace_kprobe_offset(orig) != trace_kprobe_offset(comp))
548 continue;
549
550 /*
551 * trace_probe_compare_arg_type() ensured that nr_args and
552 * each argument name and type are same. Let's compare comm.
553 */
554 for (i = 0; i < orig->tp.nr_args; i++) {
555 if (strcmp(orig->tp.args[i].comm,
556 comp->tp.args[i].comm))
557 break;
558 }
559
560 if (i == orig->tp.nr_args)
561 return true;
562 }
563
564 return false;
565}
566
567static int append_trace_kprobe(struct trace_kprobe *tk, struct trace_kprobe *to)
568{
569 int ret;
570
571 ret = trace_probe_compare_arg_type(&tk->tp, &to->tp);
572 if (ret) {
573 /* Note that argument starts index = 2 */
574 trace_probe_log_set_index(ret + 1);
575 trace_probe_log_err(0, DIFF_ARG_TYPE);
576 return -EEXIST;
577 }
578 if (trace_kprobe_has_same_kprobe(to, tk)) {
579 trace_probe_log_set_index(0);
580 trace_probe_log_err(0, SAME_PROBE);
581 return -EEXIST;
582 }
583
584 /* Append to existing event */
585 ret = trace_probe_append(&tk->tp, &to->tp);
586 if (ret)
587 return ret;
588
589 /* Register k*probe */
590 ret = __register_trace_kprobe(tk);
591 if (ret == -ENOENT && !trace_kprobe_module_exist(tk)) {
592 pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
593 ret = 0;
594 }
595
596 if (ret)
597 trace_probe_unlink(&tk->tp);
598 else
599 dyn_event_add(&tk->devent);
600
601 return ret;
602}
603
604/* Register a trace_probe and probe_event */
605static int register_trace_kprobe(struct trace_kprobe *tk)
606{
607 struct trace_kprobe *old_tk;
608 int ret;
609
610 mutex_lock(&event_mutex);
611
612 old_tk = find_trace_kprobe(trace_probe_name(&tk->tp),
613 trace_probe_group_name(&tk->tp));
614 if (old_tk) {
615 if (trace_kprobe_is_return(tk) != trace_kprobe_is_return(old_tk)) {
616 trace_probe_log_set_index(0);
617 trace_probe_log_err(0, DIFF_PROBE_TYPE);
618 ret = -EEXIST;
619 } else {
620 ret = append_trace_kprobe(tk, old_tk);
621 }
622 goto end;
623 }
624
625 /* Register new event */
626 ret = register_kprobe_event(tk);
627 if (ret) {
628 pr_warn("Failed to register probe event(%d)\n", ret);
629 goto end;
630 }
631
632 /* Register k*probe */
633 ret = __register_trace_kprobe(tk);
634 if (ret == -ENOENT && !trace_kprobe_module_exist(tk)) {
635 pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
636 ret = 0;
637 }
638
639 if (ret < 0)
640 unregister_kprobe_event(tk);
641 else
642 dyn_event_add(&tk->devent);
643
644end:
645 mutex_unlock(&event_mutex);
646 return ret;
647}
648
649/* Module notifier call back, checking event on the module */
650static int trace_kprobe_module_callback(struct notifier_block *nb,
651 unsigned long val, void *data)
652{
653 struct module *mod = data;
654 struct dyn_event *pos;
655 struct trace_kprobe *tk;
656 int ret;
657
658 if (val != MODULE_STATE_COMING)
659 return NOTIFY_DONE;
660
661 /* Update probes on coming module */
662 mutex_lock(&event_mutex);
663 for_each_trace_kprobe(tk, pos) {
664 if (trace_kprobe_within_module(tk, mod)) {
665 /* Don't need to check busy - this should have gone. */
666 __unregister_trace_kprobe(tk);
667 ret = __register_trace_kprobe(tk);
668 if (ret)
669 pr_warn("Failed to re-register probe %s on %s: %d\n",
670 trace_probe_name(&tk->tp),
671 mod->name, ret);
672 }
673 }
674 mutex_unlock(&event_mutex);
675
676 return NOTIFY_DONE;
677}
678
679static struct notifier_block trace_kprobe_module_nb = {
680 .notifier_call = trace_kprobe_module_callback,
681 .priority = 1 /* Invoked after kprobe module callback */
682};
683
684/* Convert certain expected symbols into '_' when generating event names */
685static inline void sanitize_event_name(char *name)
686{
687 while (*name++ != '\0')
688 if (*name == ':' || *name == '.')
689 *name = '_';
690}
691
692static int trace_kprobe_create(int argc, const char *argv[])
693{
694 /*
695 * Argument syntax:
696 * - Add kprobe:
697 * p[:[GRP/]EVENT] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
698 * - Add kretprobe:
699 * r[MAXACTIVE][:[GRP/]EVENT] [MOD:]KSYM[+0] [FETCHARGS]
700 * Fetch args:
701 * $retval : fetch return value
702 * $stack : fetch stack address
703 * $stackN : fetch Nth of stack (N:0-)
704 * $comm : fetch current task comm
705 * @ADDR : fetch memory at ADDR (ADDR should be in kernel)
706 * @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
707 * %REG : fetch register REG
708 * Dereferencing memory fetch:
709 * +|-offs(ARG) : fetch memory at ARG +|- offs address.
710 * Alias name of args:
711 * NAME=FETCHARG : set NAME as alias of FETCHARG.
712 * Type of args:
713 * FETCHARG:TYPE : use TYPE instead of unsigned long.
714 */
715 struct trace_kprobe *tk = NULL;
716 int i, len, ret = 0;
717 bool is_return = false;
718 char *symbol = NULL, *tmp = NULL;
719 const char *event = NULL, *group = KPROBE_EVENT_SYSTEM;
720 int maxactive = 0;
721 long offset = 0;
722 void *addr = NULL;
723 char buf[MAX_EVENT_NAME_LEN];
724 unsigned int flags = TPARG_FL_KERNEL;
725
726 switch (argv[0][0]) {
727 case 'r':
728 is_return = true;
729 flags |= TPARG_FL_RETURN;
730 break;
731 case 'p':
732 break;
733 default:
734 return -ECANCELED;
735 }
736 if (argc < 2)
737 return -ECANCELED;
738
739 trace_probe_log_init("trace_kprobe", argc, argv);
740
741 event = strchr(&argv[0][1], ':');
742 if (event)
743 event++;
744
745 if (isdigit(argv[0][1])) {
746 if (!is_return) {
747 trace_probe_log_err(1, MAXACT_NO_KPROBE);
748 goto parse_error;
749 }
750 if (event)
751 len = event - &argv[0][1] - 1;
752 else
753 len = strlen(&argv[0][1]);
754 if (len > MAX_EVENT_NAME_LEN - 1) {
755 trace_probe_log_err(1, BAD_MAXACT);
756 goto parse_error;
757 }
758 memcpy(buf, &argv[0][1], len);
759 buf[len] = '\0';
760 ret = kstrtouint(buf, 0, &maxactive);
761 if (ret || !maxactive) {
762 trace_probe_log_err(1, BAD_MAXACT);
763 goto parse_error;
764 }
765 /* kretprobes instances are iterated over via a list. The
766 * maximum should stay reasonable.
767 */
768 if (maxactive > KRETPROBE_MAXACTIVE_MAX) {
769 trace_probe_log_err(1, MAXACT_TOO_BIG);
770 goto parse_error;
771 }
772 }
773
774 /* try to parse an address. if that fails, try to read the
775 * input as a symbol. */
776 if (kstrtoul(argv[1], 0, (unsigned long *)&addr)) {
777 trace_probe_log_set_index(1);
778 /* Check whether uprobe event specified */
779 if (strchr(argv[1], '/') && strchr(argv[1], ':')) {
780 ret = -ECANCELED;
781 goto error;
782 }
783 /* a symbol specified */
784 symbol = kstrdup(argv[1], GFP_KERNEL);
785 if (!symbol)
786 return -ENOMEM;
787 /* TODO: support .init module functions */
788 ret = traceprobe_split_symbol_offset(symbol, &offset);
789 if (ret || offset < 0 || offset > UINT_MAX) {
790 trace_probe_log_err(0, BAD_PROBE_ADDR);
791 goto parse_error;
792 }
793 if (kprobe_on_func_entry(NULL, symbol, offset))
794 flags |= TPARG_FL_FENTRY;
795 if (offset && is_return && !(flags & TPARG_FL_FENTRY)) {
796 trace_probe_log_err(0, BAD_RETPROBE);
797 goto parse_error;
798 }
799 }
800
801 trace_probe_log_set_index(0);
802 if (event) {
803 ret = traceprobe_parse_event_name(&event, &group, buf,
804 event - argv[0]);
805 if (ret)
806 goto parse_error;
807 } else {
808 /* Make a new event name */
809 if (symbol)
810 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
811 is_return ? 'r' : 'p', symbol, offset);
812 else
813 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
814 is_return ? 'r' : 'p', addr);
815 sanitize_event_name(buf);
816 event = buf;
817 }
818
819 /* setup a probe */
820 tk = alloc_trace_kprobe(group, event, addr, symbol, offset, maxactive,
821 argc - 2, is_return);
822 if (IS_ERR(tk)) {
823 ret = PTR_ERR(tk);
824 /* This must return -ENOMEM, else there is a bug */
825 WARN_ON_ONCE(ret != -ENOMEM);
826 goto out; /* We know tk is not allocated */
827 }
828 argc -= 2; argv += 2;
829
830 /* parse arguments */
831 for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
832 tmp = kstrdup(argv[i], GFP_KERNEL);
833 if (!tmp) {
834 ret = -ENOMEM;
835 goto error;
836 }
837
838 trace_probe_log_set_index(i + 2);
839 ret = traceprobe_parse_probe_arg(&tk->tp, i, tmp, flags);
840 kfree(tmp);
841 if (ret)
842 goto error; /* This can be -ENOMEM */
843 }
844
845 ret = traceprobe_set_print_fmt(&tk->tp, is_return);
846 if (ret < 0)
847 goto error;
848
849 ret = register_trace_kprobe(tk);
850 if (ret) {
851 trace_probe_log_set_index(1);
852 if (ret == -EILSEQ)
853 trace_probe_log_err(0, BAD_INSN_BNDRY);
854 else if (ret == -ENOENT)
855 trace_probe_log_err(0, BAD_PROBE_ADDR);
856 else if (ret != -ENOMEM && ret != -EEXIST)
857 trace_probe_log_err(0, FAIL_REG_PROBE);
858 goto error;
859 }
860
861out:
862 trace_probe_log_clear();
863 kfree(symbol);
864 return ret;
865
866parse_error:
867 ret = -EINVAL;
868error:
869 free_trace_kprobe(tk);
870 goto out;
871}
872
873static int create_or_delete_trace_kprobe(int argc, char **argv)
874{
875 int ret;
876
877 if (argv[0][0] == '-')
878 return dyn_event_release(argc, argv, &trace_kprobe_ops);
879
880 ret = trace_kprobe_create(argc, (const char **)argv);
881 return ret == -ECANCELED ? -EINVAL : ret;
882}
883
884static int trace_kprobe_release(struct dyn_event *ev)
885{
886 struct trace_kprobe *tk = to_trace_kprobe(ev);
887 int ret = unregister_trace_kprobe(tk);
888
889 if (!ret)
890 free_trace_kprobe(tk);
891 return ret;
892}
893
894static int trace_kprobe_show(struct seq_file *m, struct dyn_event *ev)
895{
896 struct trace_kprobe *tk = to_trace_kprobe(ev);
897 int i;
898
899 seq_putc(m, trace_kprobe_is_return(tk) ? 'r' : 'p');
900 seq_printf(m, ":%s/%s", trace_probe_group_name(&tk->tp),
901 trace_probe_name(&tk->tp));
902
903 if (!tk->symbol)
904 seq_printf(m, " 0x%p", tk->rp.kp.addr);
905 else if (tk->rp.kp.offset)
906 seq_printf(m, " %s+%u", trace_kprobe_symbol(tk),
907 tk->rp.kp.offset);
908 else
909 seq_printf(m, " %s", trace_kprobe_symbol(tk));
910
911 for (i = 0; i < tk->tp.nr_args; i++)
912 seq_printf(m, " %s=%s", tk->tp.args[i].name, tk->tp.args[i].comm);
913 seq_putc(m, '\n');
914
915 return 0;
916}
917
918static int probes_seq_show(struct seq_file *m, void *v)
919{
920 struct dyn_event *ev = v;
921
922 if (!is_trace_kprobe(ev))
923 return 0;
924
925 return trace_kprobe_show(m, ev);
926}
927
928static const struct seq_operations probes_seq_op = {
929 .start = dyn_event_seq_start,
930 .next = dyn_event_seq_next,
931 .stop = dyn_event_seq_stop,
932 .show = probes_seq_show
933};
934
935static int probes_open(struct inode *inode, struct file *file)
936{
937 int ret;
938
939 ret = security_locked_down(LOCKDOWN_TRACEFS);
940 if (ret)
941 return ret;
942
943 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
944 ret = dyn_events_release_all(&trace_kprobe_ops);
945 if (ret < 0)
946 return ret;
947 }
948
949 return seq_open(file, &probes_seq_op);
950}
951
952static ssize_t probes_write(struct file *file, const char __user *buffer,
953 size_t count, loff_t *ppos)
954{
955 return trace_parse_run_command(file, buffer, count, ppos,
956 create_or_delete_trace_kprobe);
957}
958
959static const struct file_operations kprobe_events_ops = {
960 .owner = THIS_MODULE,
961 .open = probes_open,
962 .read = seq_read,
963 .llseek = seq_lseek,
964 .release = seq_release,
965 .write = probes_write,
966};
967
968/* Probes profiling interfaces */
969static int probes_profile_seq_show(struct seq_file *m, void *v)
970{
971 struct dyn_event *ev = v;
972 struct trace_kprobe *tk;
973
974 if (!is_trace_kprobe(ev))
975 return 0;
976
977 tk = to_trace_kprobe(ev);
978 seq_printf(m, " %-44s %15lu %15lu\n",
979 trace_probe_name(&tk->tp),
980 trace_kprobe_nhit(tk),
981 tk->rp.kp.nmissed);
982
983 return 0;
984}
985
986static const struct seq_operations profile_seq_op = {
987 .start = dyn_event_seq_start,
988 .next = dyn_event_seq_next,
989 .stop = dyn_event_seq_stop,
990 .show = probes_profile_seq_show
991};
992
993static int profile_open(struct inode *inode, struct file *file)
994{
995 int ret;
996
997 ret = security_locked_down(LOCKDOWN_TRACEFS);
998 if (ret)
999 return ret;
1000
1001 return seq_open(file, &profile_seq_op);
1002}
1003
1004static const struct file_operations kprobe_profile_ops = {
1005 .owner = THIS_MODULE,
1006 .open = profile_open,
1007 .read = seq_read,
1008 .llseek = seq_lseek,
1009 .release = seq_release,
1010};
1011
1012/* Kprobe specific fetch functions */
1013
1014/* Return the length of string -- including null terminal byte */
1015static nokprobe_inline int
1016fetch_store_strlen(unsigned long addr)
1017{
1018 int ret, len = 0;
1019 u8 c;
1020
1021 do {
1022 ret = probe_kernel_read(&c, (u8 *)addr + len, 1);
1023 len++;
1024 } while (c && ret == 0 && len < MAX_STRING_SIZE);
1025
1026 return (ret < 0) ? ret : len;
1027}
1028
1029/* Return the length of string -- including null terminal byte */
1030static nokprobe_inline int
1031fetch_store_strlen_user(unsigned long addr)
1032{
1033 const void __user *uaddr = (__force const void __user *)addr;
1034
1035 return strnlen_unsafe_user(uaddr, MAX_STRING_SIZE);
1036}
1037
1038/*
1039 * Fetch a null-terminated string. Caller MUST set *(u32 *)buf with max
1040 * length and relative data location.
1041 */
1042static nokprobe_inline int
1043fetch_store_string(unsigned long addr, void *dest, void *base)
1044{
1045 int maxlen = get_loc_len(*(u32 *)dest);
1046 void *__dest;
1047 long ret;
1048
1049 if (unlikely(!maxlen))
1050 return -ENOMEM;
1051
1052 __dest = get_loc_data(dest, base);
1053
1054 /*
1055 * Try to get string again, since the string can be changed while
1056 * probing.
1057 */
1058 ret = strncpy_from_unsafe(__dest, (void *)addr, maxlen);
1059 if (ret >= 0)
1060 *(u32 *)dest = make_data_loc(ret, __dest - base);
1061
1062 return ret;
1063}
1064
1065/*
1066 * Fetch a null-terminated string from user. Caller MUST set *(u32 *)buf
1067 * with max length and relative data location.
1068 */
1069static nokprobe_inline int
1070fetch_store_string_user(unsigned long addr, void *dest, void *base)
1071{
1072 const void __user *uaddr = (__force const void __user *)addr;
1073 int maxlen = get_loc_len(*(u32 *)dest);
1074 void *__dest;
1075 long ret;
1076
1077 if (unlikely(!maxlen))
1078 return -ENOMEM;
1079
1080 __dest = get_loc_data(dest, base);
1081
1082 ret = strncpy_from_unsafe_user(__dest, uaddr, maxlen);
1083 if (ret >= 0)
1084 *(u32 *)dest = make_data_loc(ret, __dest - base);
1085
1086 return ret;
1087}
1088
1089static nokprobe_inline int
1090probe_mem_read(void *dest, void *src, size_t size)
1091{
1092 return probe_kernel_read(dest, src, size);
1093}
1094
1095static nokprobe_inline int
1096probe_mem_read_user(void *dest, void *src, size_t size)
1097{
1098 const void __user *uaddr = (__force const void __user *)src;
1099
1100 return probe_user_read(dest, uaddr, size);
1101}
1102
1103/* Note that we don't verify it, since the code does not come from user space */
1104static int
1105process_fetch_insn(struct fetch_insn *code, struct pt_regs *regs, void *dest,
1106 void *base)
1107{
1108 unsigned long val;
1109
1110retry:
1111 /* 1st stage: get value from context */
1112 switch (code->op) {
1113 case FETCH_OP_REG:
1114 val = regs_get_register(regs, code->param);
1115 break;
1116 case FETCH_OP_STACK:
1117 val = regs_get_kernel_stack_nth(regs, code->param);
1118 break;
1119 case FETCH_OP_STACKP:
1120 val = kernel_stack_pointer(regs);
1121 break;
1122 case FETCH_OP_RETVAL:
1123 val = regs_return_value(regs);
1124 break;
1125 case FETCH_OP_IMM:
1126 val = code->immediate;
1127 break;
1128 case FETCH_OP_COMM:
1129 val = (unsigned long)current->comm;
1130 break;
1131 case FETCH_OP_DATA:
1132 val = (unsigned long)code->data;
1133 break;
1134#ifdef CONFIG_HAVE_FUNCTION_ARG_ACCESS_API
1135 case FETCH_OP_ARG:
1136 val = regs_get_kernel_argument(regs, code->param);
1137 break;
1138#endif
1139 case FETCH_NOP_SYMBOL: /* Ignore a place holder */
1140 code++;
1141 goto retry;
1142 default:
1143 return -EILSEQ;
1144 }
1145 code++;
1146
1147 return process_fetch_insn_bottom(code, val, dest, base);
1148}
1149NOKPROBE_SYMBOL(process_fetch_insn)
1150
1151/* Kprobe handler */
1152static nokprobe_inline void
1153__kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs,
1154 struct trace_event_file *trace_file)
1155{
1156 struct kprobe_trace_entry_head *entry;
1157 struct ring_buffer_event *event;
1158 struct ring_buffer *buffer;
1159 int size, dsize, pc;
1160 unsigned long irq_flags;
1161 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1162
1163 WARN_ON(call != trace_file->event_call);
1164
1165 if (trace_trigger_soft_disabled(trace_file))
1166 return;
1167
1168 local_save_flags(irq_flags);
1169 pc = preempt_count();
1170
1171 dsize = __get_data_size(&tk->tp, regs);
1172 size = sizeof(*entry) + tk->tp.size + dsize;
1173
1174 event = trace_event_buffer_lock_reserve(&buffer, trace_file,
1175 call->event.type,
1176 size, irq_flags, pc);
1177 if (!event)
1178 return;
1179
1180 entry = ring_buffer_event_data(event);
1181 entry->ip = (unsigned long)tk->rp.kp.addr;
1182 store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1183
1184 event_trigger_unlock_commit_regs(trace_file, buffer, event,
1185 entry, irq_flags, pc, regs);
1186}
1187
1188static void
1189kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs)
1190{
1191 struct event_file_link *link;
1192
1193 trace_probe_for_each_link_rcu(link, &tk->tp)
1194 __kprobe_trace_func(tk, regs, link->file);
1195}
1196NOKPROBE_SYMBOL(kprobe_trace_func);
1197
1198/* Kretprobe handler */
1199static nokprobe_inline void
1200__kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1201 struct pt_regs *regs,
1202 struct trace_event_file *trace_file)
1203{
1204 struct kretprobe_trace_entry_head *entry;
1205 struct ring_buffer_event *event;
1206 struct ring_buffer *buffer;
1207 int size, pc, dsize;
1208 unsigned long irq_flags;
1209 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1210
1211 WARN_ON(call != trace_file->event_call);
1212
1213 if (trace_trigger_soft_disabled(trace_file))
1214 return;
1215
1216 local_save_flags(irq_flags);
1217 pc = preempt_count();
1218
1219 dsize = __get_data_size(&tk->tp, regs);
1220 size = sizeof(*entry) + tk->tp.size + dsize;
1221
1222 event = trace_event_buffer_lock_reserve(&buffer, trace_file,
1223 call->event.type,
1224 size, irq_flags, pc);
1225 if (!event)
1226 return;
1227
1228 entry = ring_buffer_event_data(event);
1229 entry->func = (unsigned long)tk->rp.kp.addr;
1230 entry->ret_ip = (unsigned long)ri->ret_addr;
1231 store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1232
1233 event_trigger_unlock_commit_regs(trace_file, buffer, event,
1234 entry, irq_flags, pc, regs);
1235}
1236
1237static void
1238kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1239 struct pt_regs *regs)
1240{
1241 struct event_file_link *link;
1242
1243 trace_probe_for_each_link_rcu(link, &tk->tp)
1244 __kretprobe_trace_func(tk, ri, regs, link->file);
1245}
1246NOKPROBE_SYMBOL(kretprobe_trace_func);
1247
1248/* Event entry printers */
1249static enum print_line_t
1250print_kprobe_event(struct trace_iterator *iter, int flags,
1251 struct trace_event *event)
1252{
1253 struct kprobe_trace_entry_head *field;
1254 struct trace_seq *s = &iter->seq;
1255 struct trace_probe *tp;
1256
1257 field = (struct kprobe_trace_entry_head *)iter->ent;
1258 tp = trace_probe_primary_from_call(
1259 container_of(event, struct trace_event_call, event));
1260 if (WARN_ON_ONCE(!tp))
1261 goto out;
1262
1263 trace_seq_printf(s, "%s: (", trace_probe_name(tp));
1264
1265 if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
1266 goto out;
1267
1268 trace_seq_putc(s, ')');
1269
1270 if (print_probe_args(s, tp->args, tp->nr_args,
1271 (u8 *)&field[1], field) < 0)
1272 goto out;
1273
1274 trace_seq_putc(s, '\n');
1275 out:
1276 return trace_handle_return(s);
1277}
1278
1279static enum print_line_t
1280print_kretprobe_event(struct trace_iterator *iter, int flags,
1281 struct trace_event *event)
1282{
1283 struct kretprobe_trace_entry_head *field;
1284 struct trace_seq *s = &iter->seq;
1285 struct trace_probe *tp;
1286
1287 field = (struct kretprobe_trace_entry_head *)iter->ent;
1288 tp = trace_probe_primary_from_call(
1289 container_of(event, struct trace_event_call, event));
1290 if (WARN_ON_ONCE(!tp))
1291 goto out;
1292
1293 trace_seq_printf(s, "%s: (", trace_probe_name(tp));
1294
1295 if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1296 goto out;
1297
1298 trace_seq_puts(s, " <- ");
1299
1300 if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1301 goto out;
1302
1303 trace_seq_putc(s, ')');
1304
1305 if (print_probe_args(s, tp->args, tp->nr_args,
1306 (u8 *)&field[1], field) < 0)
1307 goto out;
1308
1309 trace_seq_putc(s, '\n');
1310
1311 out:
1312 return trace_handle_return(s);
1313}
1314
1315
1316static int kprobe_event_define_fields(struct trace_event_call *event_call)
1317{
1318 int ret;
1319 struct kprobe_trace_entry_head field;
1320 struct trace_probe *tp;
1321
1322 tp = trace_probe_primary_from_call(event_call);
1323 if (WARN_ON_ONCE(!tp))
1324 return -ENOENT;
1325
1326 DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1327
1328 return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
1329}
1330
1331static int kretprobe_event_define_fields(struct trace_event_call *event_call)
1332{
1333 int ret;
1334 struct kretprobe_trace_entry_head field;
1335 struct trace_probe *tp;
1336
1337 tp = trace_probe_primary_from_call(event_call);
1338 if (WARN_ON_ONCE(!tp))
1339 return -ENOENT;
1340
1341 DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1342 DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1343
1344 return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
1345}
1346
1347#ifdef CONFIG_PERF_EVENTS
1348
1349/* Kprobe profile handler */
1350static int
1351kprobe_perf_func(struct trace_kprobe *tk, struct pt_regs *regs)
1352{
1353 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1354 struct kprobe_trace_entry_head *entry;
1355 struct hlist_head *head;
1356 int size, __size, dsize;
1357 int rctx;
1358
1359 if (bpf_prog_array_valid(call)) {
1360 unsigned long orig_ip = instruction_pointer(regs);
1361 int ret;
1362
1363 ret = trace_call_bpf(call, regs);
1364
1365 /*
1366 * We need to check and see if we modified the pc of the
1367 * pt_regs, and if so return 1 so that we don't do the
1368 * single stepping.
1369 */
1370 if (orig_ip != instruction_pointer(regs))
1371 return 1;
1372 if (!ret)
1373 return 0;
1374 }
1375
1376 head = this_cpu_ptr(call->perf_events);
1377 if (hlist_empty(head))
1378 return 0;
1379
1380 dsize = __get_data_size(&tk->tp, regs);
1381 __size = sizeof(*entry) + tk->tp.size + dsize;
1382 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1383 size -= sizeof(u32);
1384
1385 entry = perf_trace_buf_alloc(size, NULL, &rctx);
1386 if (!entry)
1387 return 0;
1388
1389 entry->ip = (unsigned long)tk->rp.kp.addr;
1390 memset(&entry[1], 0, dsize);
1391 store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1392 perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1393 head, NULL);
1394 return 0;
1395}
1396NOKPROBE_SYMBOL(kprobe_perf_func);
1397
1398/* Kretprobe profile handler */
1399static void
1400kretprobe_perf_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1401 struct pt_regs *regs)
1402{
1403 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1404 struct kretprobe_trace_entry_head *entry;
1405 struct hlist_head *head;
1406 int size, __size, dsize;
1407 int rctx;
1408
1409 if (bpf_prog_array_valid(call) && !trace_call_bpf(call, regs))
1410 return;
1411
1412 head = this_cpu_ptr(call->perf_events);
1413 if (hlist_empty(head))
1414 return;
1415
1416 dsize = __get_data_size(&tk->tp, regs);
1417 __size = sizeof(*entry) + tk->tp.size + dsize;
1418 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1419 size -= sizeof(u32);
1420
1421 entry = perf_trace_buf_alloc(size, NULL, &rctx);
1422 if (!entry)
1423 return;
1424
1425 entry->func = (unsigned long)tk->rp.kp.addr;
1426 entry->ret_ip = (unsigned long)ri->ret_addr;
1427 store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1428 perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1429 head, NULL);
1430}
1431NOKPROBE_SYMBOL(kretprobe_perf_func);
1432
1433int bpf_get_kprobe_info(const struct perf_event *event, u32 *fd_type,
1434 const char **symbol, u64 *probe_offset,
1435 u64 *probe_addr, bool perf_type_tracepoint)
1436{
1437 const char *pevent = trace_event_name(event->tp_event);
1438 const char *group = event->tp_event->class->system;
1439 struct trace_kprobe *tk;
1440
1441 if (perf_type_tracepoint)
1442 tk = find_trace_kprobe(pevent, group);
1443 else
1444 tk = event->tp_event->data;
1445 if (!tk)
1446 return -EINVAL;
1447
1448 *fd_type = trace_kprobe_is_return(tk) ? BPF_FD_TYPE_KRETPROBE
1449 : BPF_FD_TYPE_KPROBE;
1450 if (tk->symbol) {
1451 *symbol = tk->symbol;
1452 *probe_offset = tk->rp.kp.offset;
1453 *probe_addr = 0;
1454 } else {
1455 *symbol = NULL;
1456 *probe_offset = 0;
1457 *probe_addr = (unsigned long)tk->rp.kp.addr;
1458 }
1459 return 0;
1460}
1461#endif /* CONFIG_PERF_EVENTS */
1462
1463/*
1464 * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex.
1465 *
1466 * kprobe_trace_self_tests_init() does enable_trace_probe/disable_trace_probe
1467 * lockless, but we can't race with this __init function.
1468 */
1469static int kprobe_register(struct trace_event_call *event,
1470 enum trace_reg type, void *data)
1471{
1472 struct trace_event_file *file = data;
1473
1474 switch (type) {
1475 case TRACE_REG_REGISTER:
1476 return enable_trace_kprobe(event, file);
1477 case TRACE_REG_UNREGISTER:
1478 return disable_trace_kprobe(event, file);
1479
1480#ifdef CONFIG_PERF_EVENTS
1481 case TRACE_REG_PERF_REGISTER:
1482 return enable_trace_kprobe(event, NULL);
1483 case TRACE_REG_PERF_UNREGISTER:
1484 return disable_trace_kprobe(event, NULL);
1485 case TRACE_REG_PERF_OPEN:
1486 case TRACE_REG_PERF_CLOSE:
1487 case TRACE_REG_PERF_ADD:
1488 case TRACE_REG_PERF_DEL:
1489 return 0;
1490#endif
1491 }
1492 return 0;
1493}
1494
1495static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1496{
1497 struct trace_kprobe *tk = container_of(kp, struct trace_kprobe, rp.kp);
1498 int ret = 0;
1499
1500 raw_cpu_inc(*tk->nhit);
1501
1502 if (trace_probe_test_flag(&tk->tp, TP_FLAG_TRACE))
1503 kprobe_trace_func(tk, regs);
1504#ifdef CONFIG_PERF_EVENTS
1505 if (trace_probe_test_flag(&tk->tp, TP_FLAG_PROFILE))
1506 ret = kprobe_perf_func(tk, regs);
1507#endif
1508 return ret;
1509}
1510NOKPROBE_SYMBOL(kprobe_dispatcher);
1511
1512static int
1513kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1514{
1515 struct trace_kprobe *tk = container_of(ri->rp, struct trace_kprobe, rp);
1516
1517 raw_cpu_inc(*tk->nhit);
1518
1519 if (trace_probe_test_flag(&tk->tp, TP_FLAG_TRACE))
1520 kretprobe_trace_func(tk, ri, regs);
1521#ifdef CONFIG_PERF_EVENTS
1522 if (trace_probe_test_flag(&tk->tp, TP_FLAG_PROFILE))
1523 kretprobe_perf_func(tk, ri, regs);
1524#endif
1525 return 0; /* We don't tweek kernel, so just return 0 */
1526}
1527NOKPROBE_SYMBOL(kretprobe_dispatcher);
1528
1529static struct trace_event_functions kretprobe_funcs = {
1530 .trace = print_kretprobe_event
1531};
1532
1533static struct trace_event_functions kprobe_funcs = {
1534 .trace = print_kprobe_event
1535};
1536
1537static inline void init_trace_event_call(struct trace_kprobe *tk)
1538{
1539 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1540
1541 if (trace_kprobe_is_return(tk)) {
1542 call->event.funcs = &kretprobe_funcs;
1543 call->class->define_fields = kretprobe_event_define_fields;
1544 } else {
1545 call->event.funcs = &kprobe_funcs;
1546 call->class->define_fields = kprobe_event_define_fields;
1547 }
1548
1549 call->flags = TRACE_EVENT_FL_KPROBE;
1550 call->class->reg = kprobe_register;
1551}
1552
1553static int register_kprobe_event(struct trace_kprobe *tk)
1554{
1555 init_trace_event_call(tk);
1556
1557 return trace_probe_register_event_call(&tk->tp);
1558}
1559
1560static int unregister_kprobe_event(struct trace_kprobe *tk)
1561{
1562 return trace_probe_unregister_event_call(&tk->tp);
1563}
1564
1565#ifdef CONFIG_PERF_EVENTS
1566/* create a trace_kprobe, but don't add it to global lists */
1567struct trace_event_call *
1568create_local_trace_kprobe(char *func, void *addr, unsigned long offs,
1569 bool is_return)
1570{
1571 struct trace_kprobe *tk;
1572 int ret;
1573 char *event;
1574
1575 /*
1576 * local trace_kprobes are not added to dyn_event, so they are never
1577 * searched in find_trace_kprobe(). Therefore, there is no concern of
1578 * duplicated name here.
1579 */
1580 event = func ? func : "DUMMY_EVENT";
1581
1582 tk = alloc_trace_kprobe(KPROBE_EVENT_SYSTEM, event, (void *)addr, func,
1583 offs, 0 /* maxactive */, 0 /* nargs */,
1584 is_return);
1585
1586 if (IS_ERR(tk)) {
1587 pr_info("Failed to allocate trace_probe.(%d)\n",
1588 (int)PTR_ERR(tk));
1589 return ERR_CAST(tk);
1590 }
1591
1592 init_trace_event_call(tk);
1593
1594 if (traceprobe_set_print_fmt(&tk->tp, trace_kprobe_is_return(tk)) < 0) {
1595 ret = -ENOMEM;
1596 goto error;
1597 }
1598
1599 ret = __register_trace_kprobe(tk);
1600 if (ret < 0)
1601 goto error;
1602
1603 return trace_probe_event_call(&tk->tp);
1604error:
1605 free_trace_kprobe(tk);
1606 return ERR_PTR(ret);
1607}
1608
1609void destroy_local_trace_kprobe(struct trace_event_call *event_call)
1610{
1611 struct trace_kprobe *tk;
1612
1613 tk = trace_kprobe_primary_from_call(event_call);
1614 if (unlikely(!tk))
1615 return;
1616
1617 if (trace_probe_is_enabled(&tk->tp)) {
1618 WARN_ON(1);
1619 return;
1620 }
1621
1622 __unregister_trace_kprobe(tk);
1623
1624 free_trace_kprobe(tk);
1625}
1626#endif /* CONFIG_PERF_EVENTS */
1627
1628static __init void enable_boot_kprobe_events(void)
1629{
1630 struct trace_array *tr = top_trace_array();
1631 struct trace_event_file *file;
1632 struct trace_kprobe *tk;
1633 struct dyn_event *pos;
1634
1635 mutex_lock(&event_mutex);
1636 for_each_trace_kprobe(tk, pos) {
1637 list_for_each_entry(file, &tr->events, list)
1638 if (file->event_call == trace_probe_event_call(&tk->tp))
1639 trace_event_enable_disable(file, 1, 0);
1640 }
1641 mutex_unlock(&event_mutex);
1642}
1643
1644static __init void setup_boot_kprobe_events(void)
1645{
1646 char *p, *cmd = kprobe_boot_events_buf;
1647 int ret;
1648
1649 strreplace(kprobe_boot_events_buf, ',', ' ');
1650
1651 while (cmd && *cmd != '\0') {
1652 p = strchr(cmd, ';');
1653 if (p)
1654 *p++ = '\0';
1655
1656 ret = trace_run_command(cmd, create_or_delete_trace_kprobe);
1657 if (ret)
1658 pr_warn("Failed to add event(%d): %s\n", ret, cmd);
1659 else
1660 kprobe_boot_events_enabled = true;
1661
1662 cmd = p;
1663 }
1664
1665 enable_boot_kprobe_events();
1666}
1667
1668/* Make a tracefs interface for controlling probe points */
1669static __init int init_kprobe_trace(void)
1670{
1671 struct dentry *d_tracer;
1672 struct dentry *entry;
1673 int ret;
1674
1675 ret = dyn_event_register(&trace_kprobe_ops);
1676 if (ret)
1677 return ret;
1678
1679 if (register_module_notifier(&trace_kprobe_module_nb))
1680 return -EINVAL;
1681
1682 d_tracer = tracing_init_dentry();
1683 if (IS_ERR(d_tracer))
1684 return 0;
1685
1686 entry = tracefs_create_file("kprobe_events", 0644, d_tracer,
1687 NULL, &kprobe_events_ops);
1688
1689 /* Event list interface */
1690 if (!entry)
1691 pr_warn("Could not create tracefs 'kprobe_events' entry\n");
1692
1693 /* Profile interface */
1694 entry = tracefs_create_file("kprobe_profile", 0444, d_tracer,
1695 NULL, &kprobe_profile_ops);
1696
1697 if (!entry)
1698 pr_warn("Could not create tracefs 'kprobe_profile' entry\n");
1699
1700 setup_boot_kprobe_events();
1701
1702 return 0;
1703}
1704fs_initcall(init_kprobe_trace);
1705
1706
1707#ifdef CONFIG_FTRACE_STARTUP_TEST
1708static __init struct trace_event_file *
1709find_trace_probe_file(struct trace_kprobe *tk, struct trace_array *tr)
1710{
1711 struct trace_event_file *file;
1712
1713 list_for_each_entry(file, &tr->events, list)
1714 if (file->event_call == trace_probe_event_call(&tk->tp))
1715 return file;
1716
1717 return NULL;
1718}
1719
1720/*
1721 * Nobody but us can call enable_trace_kprobe/disable_trace_kprobe at this
1722 * stage, we can do this lockless.
1723 */
1724static __init int kprobe_trace_self_tests_init(void)
1725{
1726 int ret, warn = 0;
1727 int (*target)(int, int, int, int, int, int);
1728 struct trace_kprobe *tk;
1729 struct trace_event_file *file;
1730
1731 if (tracing_is_disabled())
1732 return -ENODEV;
1733
1734 if (kprobe_boot_events_enabled) {
1735 pr_info("Skipping kprobe tests due to kprobe_event on cmdline\n");
1736 return 0;
1737 }
1738
1739 target = kprobe_trace_selftest_target;
1740
1741 pr_info("Testing kprobe tracing: ");
1742
1743 ret = trace_run_command("p:testprobe kprobe_trace_selftest_target $stack $stack0 +0($stack)",
1744 create_or_delete_trace_kprobe);
1745 if (WARN_ON_ONCE(ret)) {
1746 pr_warn("error on probing function entry.\n");
1747 warn++;
1748 } else {
1749 /* Enable trace point */
1750 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1751 if (WARN_ON_ONCE(tk == NULL)) {
1752 pr_warn("error on getting new probe.\n");
1753 warn++;
1754 } else {
1755 file = find_trace_probe_file(tk, top_trace_array());
1756 if (WARN_ON_ONCE(file == NULL)) {
1757 pr_warn("error on getting probe file.\n");
1758 warn++;
1759 } else
1760 enable_trace_kprobe(
1761 trace_probe_event_call(&tk->tp), file);
1762 }
1763 }
1764
1765 ret = trace_run_command("r:testprobe2 kprobe_trace_selftest_target $retval",
1766 create_or_delete_trace_kprobe);
1767 if (WARN_ON_ONCE(ret)) {
1768 pr_warn("error on probing function return.\n");
1769 warn++;
1770 } else {
1771 /* Enable trace point */
1772 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1773 if (WARN_ON_ONCE(tk == NULL)) {
1774 pr_warn("error on getting 2nd new probe.\n");
1775 warn++;
1776 } else {
1777 file = find_trace_probe_file(tk, top_trace_array());
1778 if (WARN_ON_ONCE(file == NULL)) {
1779 pr_warn("error on getting probe file.\n");
1780 warn++;
1781 } else
1782 enable_trace_kprobe(
1783 trace_probe_event_call(&tk->tp), file);
1784 }
1785 }
1786
1787 if (warn)
1788 goto end;
1789
1790 ret = target(1, 2, 3, 4, 5, 6);
1791
1792 /*
1793 * Not expecting an error here, the check is only to prevent the
1794 * optimizer from removing the call to target() as otherwise there
1795 * are no side-effects and the call is never performed.
1796 */
1797 if (ret != 21)
1798 warn++;
1799
1800 /* Disable trace points before removing it */
1801 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1802 if (WARN_ON_ONCE(tk == NULL)) {
1803 pr_warn("error on getting test probe.\n");
1804 warn++;
1805 } else {
1806 if (trace_kprobe_nhit(tk) != 1) {
1807 pr_warn("incorrect number of testprobe hits\n");
1808 warn++;
1809 }
1810
1811 file = find_trace_probe_file(tk, top_trace_array());
1812 if (WARN_ON_ONCE(file == NULL)) {
1813 pr_warn("error on getting probe file.\n");
1814 warn++;
1815 } else
1816 disable_trace_kprobe(
1817 trace_probe_event_call(&tk->tp), file);
1818 }
1819
1820 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1821 if (WARN_ON_ONCE(tk == NULL)) {
1822 pr_warn("error on getting 2nd test probe.\n");
1823 warn++;
1824 } else {
1825 if (trace_kprobe_nhit(tk) != 1) {
1826 pr_warn("incorrect number of testprobe2 hits\n");
1827 warn++;
1828 }
1829
1830 file = find_trace_probe_file(tk, top_trace_array());
1831 if (WARN_ON_ONCE(file == NULL)) {
1832 pr_warn("error on getting probe file.\n");
1833 warn++;
1834 } else
1835 disable_trace_kprobe(
1836 trace_probe_event_call(&tk->tp), file);
1837 }
1838
1839 ret = trace_run_command("-:testprobe", create_or_delete_trace_kprobe);
1840 if (WARN_ON_ONCE(ret)) {
1841 pr_warn("error on deleting a probe.\n");
1842 warn++;
1843 }
1844
1845 ret = trace_run_command("-:testprobe2", create_or_delete_trace_kprobe);
1846 if (WARN_ON_ONCE(ret)) {
1847 pr_warn("error on deleting a probe.\n");
1848 warn++;
1849 }
1850
1851end:
1852 ret = dyn_events_release_all(&trace_kprobe_ops);
1853 if (WARN_ON_ONCE(ret)) {
1854 pr_warn("error on cleaning up probes.\n");
1855 warn++;
1856 }
1857 /*
1858 * Wait for the optimizer work to finish. Otherwise it might fiddle
1859 * with probes in already freed __init text.
1860 */
1861 wait_for_kprobe_optimizer();
1862 if (warn)
1863 pr_cont("NG: Some tests are failed. Please check them.\n");
1864 else
1865 pr_cont("OK\n");
1866 return 0;
1867}
1868
1869late_initcall(kprobe_trace_self_tests_init);
1870
1871#endif