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