Loading...
1/*
2 * Kprobes-based tracing events
3 *
4 * Created by Masami Hiramatsu <mhiramat@redhat.com>
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 as
8 * published by the Free Software Foundation.
9 *
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
14 *
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write to the Free Software
17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
18 */
19
20#include <linux/module.h>
21#include <linux/uaccess.h>
22
23#include "trace_probe.h"
24
25#define KPROBE_EVENT_SYSTEM "kprobes"
26
27/**
28 * Kprobe event core functions
29 */
30struct trace_kprobe {
31 struct list_head list;
32 struct kretprobe rp; /* Use rp.kp for kprobe use */
33 unsigned long __percpu *nhit;
34 const char *symbol; /* symbol name */
35 struct trace_probe tp;
36};
37
38#define SIZEOF_TRACE_KPROBE(n) \
39 (offsetof(struct trace_kprobe, tp.args) + \
40 (sizeof(struct probe_arg) * (n)))
41
42
43static nokprobe_inline bool trace_kprobe_is_return(struct trace_kprobe *tk)
44{
45 return tk->rp.handler != NULL;
46}
47
48static nokprobe_inline const char *trace_kprobe_symbol(struct trace_kprobe *tk)
49{
50 return tk->symbol ? tk->symbol : "unknown";
51}
52
53static nokprobe_inline unsigned long trace_kprobe_offset(struct trace_kprobe *tk)
54{
55 return tk->rp.kp.offset;
56}
57
58static nokprobe_inline bool trace_kprobe_has_gone(struct trace_kprobe *tk)
59{
60 return !!(kprobe_gone(&tk->rp.kp));
61}
62
63static nokprobe_inline bool trace_kprobe_within_module(struct trace_kprobe *tk,
64 struct module *mod)
65{
66 int len = strlen(mod->name);
67 const char *name = trace_kprobe_symbol(tk);
68 return strncmp(mod->name, name, len) == 0 && name[len] == ':';
69}
70
71static nokprobe_inline bool trace_kprobe_is_on_module(struct trace_kprobe *tk)
72{
73 return !!strchr(trace_kprobe_symbol(tk), ':');
74}
75
76static int register_kprobe_event(struct trace_kprobe *tk);
77static int unregister_kprobe_event(struct trace_kprobe *tk);
78
79static DEFINE_MUTEX(probe_lock);
80static LIST_HEAD(probe_list);
81
82static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
83static int kretprobe_dispatcher(struct kretprobe_instance *ri,
84 struct pt_regs *regs);
85
86/* Memory fetching by symbol */
87struct symbol_cache {
88 char *symbol;
89 long offset;
90 unsigned long addr;
91};
92
93unsigned long update_symbol_cache(struct symbol_cache *sc)
94{
95 sc->addr = (unsigned long)kallsyms_lookup_name(sc->symbol);
96
97 if (sc->addr)
98 sc->addr += sc->offset;
99
100 return sc->addr;
101}
102
103void free_symbol_cache(struct symbol_cache *sc)
104{
105 kfree(sc->symbol);
106 kfree(sc);
107}
108
109struct symbol_cache *alloc_symbol_cache(const char *sym, long offset)
110{
111 struct symbol_cache *sc;
112
113 if (!sym || strlen(sym) == 0)
114 return NULL;
115
116 sc = kzalloc(sizeof(struct symbol_cache), GFP_KERNEL);
117 if (!sc)
118 return NULL;
119
120 sc->symbol = kstrdup(sym, GFP_KERNEL);
121 if (!sc->symbol) {
122 kfree(sc);
123 return NULL;
124 }
125 sc->offset = offset;
126 update_symbol_cache(sc);
127
128 return sc;
129}
130
131/*
132 * Kprobes-specific fetch functions
133 */
134#define DEFINE_FETCH_stack(type) \
135static void FETCH_FUNC_NAME(stack, type)(struct pt_regs *regs, \
136 void *offset, void *dest) \
137{ \
138 *(type *)dest = (type)regs_get_kernel_stack_nth(regs, \
139 (unsigned int)((unsigned long)offset)); \
140} \
141NOKPROBE_SYMBOL(FETCH_FUNC_NAME(stack, type));
142
143DEFINE_BASIC_FETCH_FUNCS(stack)
144/* No string on the stack entry */
145#define fetch_stack_string NULL
146#define fetch_stack_string_size NULL
147
148#define DEFINE_FETCH_memory(type) \
149static void FETCH_FUNC_NAME(memory, type)(struct pt_regs *regs, \
150 void *addr, void *dest) \
151{ \
152 type retval; \
153 if (probe_kernel_address(addr, retval)) \
154 *(type *)dest = 0; \
155 else \
156 *(type *)dest = retval; \
157} \
158NOKPROBE_SYMBOL(FETCH_FUNC_NAME(memory, type));
159
160DEFINE_BASIC_FETCH_FUNCS(memory)
161/*
162 * Fetch a null-terminated string. Caller MUST set *(u32 *)dest with max
163 * length and relative data location.
164 */
165static void FETCH_FUNC_NAME(memory, string)(struct pt_regs *regs,
166 void *addr, void *dest)
167{
168 int maxlen = get_rloc_len(*(u32 *)dest);
169 u8 *dst = get_rloc_data(dest);
170 long ret;
171
172 if (!maxlen)
173 return;
174
175 /*
176 * Try to get string again, since the string can be changed while
177 * probing.
178 */
179 ret = strncpy_from_unsafe(dst, addr, maxlen);
180
181 if (ret < 0) { /* Failed to fetch string */
182 dst[0] = '\0';
183 *(u32 *)dest = make_data_rloc(0, get_rloc_offs(*(u32 *)dest));
184 } else {
185 *(u32 *)dest = make_data_rloc(ret, get_rloc_offs(*(u32 *)dest));
186 }
187}
188NOKPROBE_SYMBOL(FETCH_FUNC_NAME(memory, string));
189
190/* Return the length of string -- including null terminal byte */
191static void FETCH_FUNC_NAME(memory, string_size)(struct pt_regs *regs,
192 void *addr, void *dest)
193{
194 mm_segment_t old_fs;
195 int ret, len = 0;
196 u8 c;
197
198 old_fs = get_fs();
199 set_fs(KERNEL_DS);
200 pagefault_disable();
201
202 do {
203 ret = __copy_from_user_inatomic(&c, (u8 *)addr + len, 1);
204 len++;
205 } while (c && ret == 0 && len < MAX_STRING_SIZE);
206
207 pagefault_enable();
208 set_fs(old_fs);
209
210 if (ret < 0) /* Failed to check the length */
211 *(u32 *)dest = 0;
212 else
213 *(u32 *)dest = len;
214}
215NOKPROBE_SYMBOL(FETCH_FUNC_NAME(memory, string_size));
216
217#define DEFINE_FETCH_symbol(type) \
218void FETCH_FUNC_NAME(symbol, type)(struct pt_regs *regs, void *data, void *dest)\
219{ \
220 struct symbol_cache *sc = data; \
221 if (sc->addr) \
222 fetch_memory_##type(regs, (void *)sc->addr, dest); \
223 else \
224 *(type *)dest = 0; \
225} \
226NOKPROBE_SYMBOL(FETCH_FUNC_NAME(symbol, type));
227
228DEFINE_BASIC_FETCH_FUNCS(symbol)
229DEFINE_FETCH_symbol(string)
230DEFINE_FETCH_symbol(string_size)
231
232/* kprobes don't support file_offset fetch methods */
233#define fetch_file_offset_u8 NULL
234#define fetch_file_offset_u16 NULL
235#define fetch_file_offset_u32 NULL
236#define fetch_file_offset_u64 NULL
237#define fetch_file_offset_string NULL
238#define fetch_file_offset_string_size NULL
239
240/* Fetch type information table */
241static const struct fetch_type kprobes_fetch_type_table[] = {
242 /* Special types */
243 [FETCH_TYPE_STRING] = __ASSIGN_FETCH_TYPE("string", string, string,
244 sizeof(u32), 1, "__data_loc char[]"),
245 [FETCH_TYPE_STRSIZE] = __ASSIGN_FETCH_TYPE("string_size", u32,
246 string_size, sizeof(u32), 0, "u32"),
247 /* Basic types */
248 ASSIGN_FETCH_TYPE(u8, u8, 0),
249 ASSIGN_FETCH_TYPE(u16, u16, 0),
250 ASSIGN_FETCH_TYPE(u32, u32, 0),
251 ASSIGN_FETCH_TYPE(u64, u64, 0),
252 ASSIGN_FETCH_TYPE(s8, u8, 1),
253 ASSIGN_FETCH_TYPE(s16, u16, 1),
254 ASSIGN_FETCH_TYPE(s32, u32, 1),
255 ASSIGN_FETCH_TYPE(s64, u64, 1),
256
257 ASSIGN_FETCH_TYPE_END
258};
259
260/*
261 * Allocate new trace_probe and initialize it (including kprobes).
262 */
263static struct trace_kprobe *alloc_trace_kprobe(const char *group,
264 const char *event,
265 void *addr,
266 const char *symbol,
267 unsigned long offs,
268 int nargs, bool is_return)
269{
270 struct trace_kprobe *tk;
271 int ret = -ENOMEM;
272
273 tk = kzalloc(SIZEOF_TRACE_KPROBE(nargs), GFP_KERNEL);
274 if (!tk)
275 return ERR_PTR(ret);
276
277 tk->nhit = alloc_percpu(unsigned long);
278 if (!tk->nhit)
279 goto error;
280
281 if (symbol) {
282 tk->symbol = kstrdup(symbol, GFP_KERNEL);
283 if (!tk->symbol)
284 goto error;
285 tk->rp.kp.symbol_name = tk->symbol;
286 tk->rp.kp.offset = offs;
287 } else
288 tk->rp.kp.addr = addr;
289
290 if (is_return)
291 tk->rp.handler = kretprobe_dispatcher;
292 else
293 tk->rp.kp.pre_handler = kprobe_dispatcher;
294
295 if (!event || !is_good_name(event)) {
296 ret = -EINVAL;
297 goto error;
298 }
299
300 tk->tp.call.class = &tk->tp.class;
301 tk->tp.call.name = kstrdup(event, GFP_KERNEL);
302 if (!tk->tp.call.name)
303 goto error;
304
305 if (!group || !is_good_name(group)) {
306 ret = -EINVAL;
307 goto error;
308 }
309
310 tk->tp.class.system = kstrdup(group, GFP_KERNEL);
311 if (!tk->tp.class.system)
312 goto error;
313
314 INIT_LIST_HEAD(&tk->list);
315 INIT_LIST_HEAD(&tk->tp.files);
316 return tk;
317error:
318 kfree(tk->tp.call.name);
319 kfree(tk->symbol);
320 free_percpu(tk->nhit);
321 kfree(tk);
322 return ERR_PTR(ret);
323}
324
325static void free_trace_kprobe(struct trace_kprobe *tk)
326{
327 int i;
328
329 for (i = 0; i < tk->tp.nr_args; i++)
330 traceprobe_free_probe_arg(&tk->tp.args[i]);
331
332 kfree(tk->tp.call.class->system);
333 kfree(tk->tp.call.name);
334 kfree(tk->symbol);
335 free_percpu(tk->nhit);
336 kfree(tk);
337}
338
339static struct trace_kprobe *find_trace_kprobe(const char *event,
340 const char *group)
341{
342 struct trace_kprobe *tk;
343
344 list_for_each_entry(tk, &probe_list, list)
345 if (strcmp(trace_event_name(&tk->tp.call), event) == 0 &&
346 strcmp(tk->tp.call.class->system, group) == 0)
347 return tk;
348 return NULL;
349}
350
351/*
352 * Enable trace_probe
353 * if the file is NULL, enable "perf" handler, or enable "trace" handler.
354 */
355static int
356enable_trace_kprobe(struct trace_kprobe *tk, struct trace_event_file *file)
357{
358 int ret = 0;
359
360 if (file) {
361 struct event_file_link *link;
362
363 link = kmalloc(sizeof(*link), GFP_KERNEL);
364 if (!link) {
365 ret = -ENOMEM;
366 goto out;
367 }
368
369 link->file = file;
370 list_add_tail_rcu(&link->list, &tk->tp.files);
371
372 tk->tp.flags |= TP_FLAG_TRACE;
373 } else
374 tk->tp.flags |= TP_FLAG_PROFILE;
375
376 if (trace_probe_is_registered(&tk->tp) && !trace_kprobe_has_gone(tk)) {
377 if (trace_kprobe_is_return(tk))
378 ret = enable_kretprobe(&tk->rp);
379 else
380 ret = enable_kprobe(&tk->rp.kp);
381 }
382 out:
383 return ret;
384}
385
386/*
387 * Disable trace_probe
388 * if the file is NULL, disable "perf" handler, or disable "trace" handler.
389 */
390static int
391disable_trace_kprobe(struct trace_kprobe *tk, struct trace_event_file *file)
392{
393 struct event_file_link *link = NULL;
394 int wait = 0;
395 int ret = 0;
396
397 if (file) {
398 link = find_event_file_link(&tk->tp, file);
399 if (!link) {
400 ret = -EINVAL;
401 goto out;
402 }
403
404 list_del_rcu(&link->list);
405 wait = 1;
406 if (!list_empty(&tk->tp.files))
407 goto out;
408
409 tk->tp.flags &= ~TP_FLAG_TRACE;
410 } else
411 tk->tp.flags &= ~TP_FLAG_PROFILE;
412
413 if (!trace_probe_is_enabled(&tk->tp) && trace_probe_is_registered(&tk->tp)) {
414 if (trace_kprobe_is_return(tk))
415 disable_kretprobe(&tk->rp);
416 else
417 disable_kprobe(&tk->rp.kp);
418 wait = 1;
419 }
420 out:
421 if (wait) {
422 /*
423 * Synchronize with kprobe_trace_func/kretprobe_trace_func
424 * to ensure disabled (all running handlers are finished).
425 * This is not only for kfree(), but also the caller,
426 * trace_remove_event_call() supposes it for releasing
427 * event_call related objects, which will be accessed in
428 * the kprobe_trace_func/kretprobe_trace_func.
429 */
430 synchronize_sched();
431 kfree(link); /* Ignored if link == NULL */
432 }
433
434 return ret;
435}
436
437/* Internal register function - just handle k*probes and flags */
438static int __register_trace_kprobe(struct trace_kprobe *tk)
439{
440 int i, ret;
441
442 if (trace_probe_is_registered(&tk->tp))
443 return -EINVAL;
444
445 for (i = 0; i < tk->tp.nr_args; i++)
446 traceprobe_update_arg(&tk->tp.args[i]);
447
448 /* Set/clear disabled flag according to tp->flag */
449 if (trace_probe_is_enabled(&tk->tp))
450 tk->rp.kp.flags &= ~KPROBE_FLAG_DISABLED;
451 else
452 tk->rp.kp.flags |= KPROBE_FLAG_DISABLED;
453
454 if (trace_kprobe_is_return(tk))
455 ret = register_kretprobe(&tk->rp);
456 else
457 ret = register_kprobe(&tk->rp.kp);
458
459 if (ret == 0)
460 tk->tp.flags |= TP_FLAG_REGISTERED;
461 else {
462 pr_warn("Could not insert probe at %s+%lu: %d\n",
463 trace_kprobe_symbol(tk), trace_kprobe_offset(tk), ret);
464 if (ret == -ENOENT && trace_kprobe_is_on_module(tk)) {
465 pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
466 ret = 0;
467 } else if (ret == -EILSEQ) {
468 pr_warn("Probing address(0x%p) is not an instruction boundary.\n",
469 tk->rp.kp.addr);
470 ret = -EINVAL;
471 }
472 }
473
474 return ret;
475}
476
477/* Internal unregister function - just handle k*probes and flags */
478static void __unregister_trace_kprobe(struct trace_kprobe *tk)
479{
480 if (trace_probe_is_registered(&tk->tp)) {
481 if (trace_kprobe_is_return(tk))
482 unregister_kretprobe(&tk->rp);
483 else
484 unregister_kprobe(&tk->rp.kp);
485 tk->tp.flags &= ~TP_FLAG_REGISTERED;
486 /* Cleanup kprobe for reuse */
487 if (tk->rp.kp.symbol_name)
488 tk->rp.kp.addr = NULL;
489 }
490}
491
492/* Unregister a trace_probe and probe_event: call with locking probe_lock */
493static int unregister_trace_kprobe(struct trace_kprobe *tk)
494{
495 /* Enabled event can not be unregistered */
496 if (trace_probe_is_enabled(&tk->tp))
497 return -EBUSY;
498
499 /* Will fail if probe is being used by ftrace or perf */
500 if (unregister_kprobe_event(tk))
501 return -EBUSY;
502
503 __unregister_trace_kprobe(tk);
504 list_del(&tk->list);
505
506 return 0;
507}
508
509/* Register a trace_probe and probe_event */
510static int register_trace_kprobe(struct trace_kprobe *tk)
511{
512 struct trace_kprobe *old_tk;
513 int ret;
514
515 mutex_lock(&probe_lock);
516
517 /* Delete old (same name) event if exist */
518 old_tk = find_trace_kprobe(trace_event_name(&tk->tp.call),
519 tk->tp.call.class->system);
520 if (old_tk) {
521 ret = unregister_trace_kprobe(old_tk);
522 if (ret < 0)
523 goto end;
524 free_trace_kprobe(old_tk);
525 }
526
527 /* Register new event */
528 ret = register_kprobe_event(tk);
529 if (ret) {
530 pr_warn("Failed to register probe event(%d)\n", ret);
531 goto end;
532 }
533
534 /* Register k*probe */
535 ret = __register_trace_kprobe(tk);
536 if (ret < 0)
537 unregister_kprobe_event(tk);
538 else
539 list_add_tail(&tk->list, &probe_list);
540
541end:
542 mutex_unlock(&probe_lock);
543 return ret;
544}
545
546/* Module notifier call back, checking event on the module */
547static int trace_kprobe_module_callback(struct notifier_block *nb,
548 unsigned long val, void *data)
549{
550 struct module *mod = data;
551 struct trace_kprobe *tk;
552 int ret;
553
554 if (val != MODULE_STATE_COMING)
555 return NOTIFY_DONE;
556
557 /* Update probes on coming module */
558 mutex_lock(&probe_lock);
559 list_for_each_entry(tk, &probe_list, list) {
560 if (trace_kprobe_within_module(tk, mod)) {
561 /* Don't need to check busy - this should have gone. */
562 __unregister_trace_kprobe(tk);
563 ret = __register_trace_kprobe(tk);
564 if (ret)
565 pr_warn("Failed to re-register probe %s on %s: %d\n",
566 trace_event_name(&tk->tp.call),
567 mod->name, ret);
568 }
569 }
570 mutex_unlock(&probe_lock);
571
572 return NOTIFY_DONE;
573}
574
575static struct notifier_block trace_kprobe_module_nb = {
576 .notifier_call = trace_kprobe_module_callback,
577 .priority = 1 /* Invoked after kprobe module callback */
578};
579
580static int create_trace_kprobe(int argc, char **argv)
581{
582 /*
583 * Argument syntax:
584 * - Add kprobe: p[:[GRP/]EVENT] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
585 * - Add kretprobe: r[:[GRP/]EVENT] [MOD:]KSYM[+0] [FETCHARGS]
586 * Fetch args:
587 * $retval : fetch return value
588 * $stack : fetch stack address
589 * $stackN : fetch Nth of stack (N:0-)
590 * @ADDR : fetch memory at ADDR (ADDR should be in kernel)
591 * @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
592 * %REG : fetch register REG
593 * Dereferencing memory fetch:
594 * +|-offs(ARG) : fetch memory at ARG +|- offs address.
595 * Alias name of args:
596 * NAME=FETCHARG : set NAME as alias of FETCHARG.
597 * Type of args:
598 * FETCHARG:TYPE : use TYPE instead of unsigned long.
599 */
600 struct trace_kprobe *tk;
601 int i, ret = 0;
602 bool is_return = false, is_delete = false;
603 char *symbol = NULL, *event = NULL, *group = NULL;
604 char *arg;
605 unsigned long offset = 0;
606 void *addr = NULL;
607 char buf[MAX_EVENT_NAME_LEN];
608
609 /* argc must be >= 1 */
610 if (argv[0][0] == 'p')
611 is_return = false;
612 else if (argv[0][0] == 'r')
613 is_return = true;
614 else if (argv[0][0] == '-')
615 is_delete = true;
616 else {
617 pr_info("Probe definition must be started with 'p', 'r' or"
618 " '-'.\n");
619 return -EINVAL;
620 }
621
622 if (argv[0][1] == ':') {
623 event = &argv[0][2];
624 if (strchr(event, '/')) {
625 group = event;
626 event = strchr(group, '/') + 1;
627 event[-1] = '\0';
628 if (strlen(group) == 0) {
629 pr_info("Group name is not specified\n");
630 return -EINVAL;
631 }
632 }
633 if (strlen(event) == 0) {
634 pr_info("Event name is not specified\n");
635 return -EINVAL;
636 }
637 }
638 if (!group)
639 group = KPROBE_EVENT_SYSTEM;
640
641 if (is_delete) {
642 if (!event) {
643 pr_info("Delete command needs an event name.\n");
644 return -EINVAL;
645 }
646 mutex_lock(&probe_lock);
647 tk = find_trace_kprobe(event, group);
648 if (!tk) {
649 mutex_unlock(&probe_lock);
650 pr_info("Event %s/%s doesn't exist.\n", group, event);
651 return -ENOENT;
652 }
653 /* delete an event */
654 ret = unregister_trace_kprobe(tk);
655 if (ret == 0)
656 free_trace_kprobe(tk);
657 mutex_unlock(&probe_lock);
658 return ret;
659 }
660
661 if (argc < 2) {
662 pr_info("Probe point is not specified.\n");
663 return -EINVAL;
664 }
665 if (isdigit(argv[1][0])) {
666 if (is_return) {
667 pr_info("Return probe point must be a symbol.\n");
668 return -EINVAL;
669 }
670 /* an address specified */
671 ret = kstrtoul(&argv[1][0], 0, (unsigned long *)&addr);
672 if (ret) {
673 pr_info("Failed to parse address.\n");
674 return ret;
675 }
676 } else {
677 /* a symbol specified */
678 symbol = argv[1];
679 /* TODO: support .init module functions */
680 ret = traceprobe_split_symbol_offset(symbol, &offset);
681 if (ret) {
682 pr_info("Failed to parse symbol.\n");
683 return ret;
684 }
685 if (offset && is_return) {
686 pr_info("Return probe must be used without offset.\n");
687 return -EINVAL;
688 }
689 }
690 argc -= 2; argv += 2;
691
692 /* setup a probe */
693 if (!event) {
694 /* Make a new event name */
695 if (symbol)
696 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
697 is_return ? 'r' : 'p', symbol, offset);
698 else
699 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
700 is_return ? 'r' : 'p', addr);
701 event = buf;
702 }
703 tk = alloc_trace_kprobe(group, event, addr, symbol, offset, argc,
704 is_return);
705 if (IS_ERR(tk)) {
706 pr_info("Failed to allocate trace_probe.(%d)\n",
707 (int)PTR_ERR(tk));
708 return PTR_ERR(tk);
709 }
710
711 /* parse arguments */
712 ret = 0;
713 for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
714 struct probe_arg *parg = &tk->tp.args[i];
715
716 /* Increment count for freeing args in error case */
717 tk->tp.nr_args++;
718
719 /* Parse argument name */
720 arg = strchr(argv[i], '=');
721 if (arg) {
722 *arg++ = '\0';
723 parg->name = kstrdup(argv[i], GFP_KERNEL);
724 } else {
725 arg = argv[i];
726 /* If argument name is omitted, set "argN" */
727 snprintf(buf, MAX_EVENT_NAME_LEN, "arg%d", i + 1);
728 parg->name = kstrdup(buf, GFP_KERNEL);
729 }
730
731 if (!parg->name) {
732 pr_info("Failed to allocate argument[%d] name.\n", i);
733 ret = -ENOMEM;
734 goto error;
735 }
736
737 if (!is_good_name(parg->name)) {
738 pr_info("Invalid argument[%d] name: %s\n",
739 i, parg->name);
740 ret = -EINVAL;
741 goto error;
742 }
743
744 if (traceprobe_conflict_field_name(parg->name,
745 tk->tp.args, i)) {
746 pr_info("Argument[%d] name '%s' conflicts with "
747 "another field.\n", i, argv[i]);
748 ret = -EINVAL;
749 goto error;
750 }
751
752 /* Parse fetch argument */
753 ret = traceprobe_parse_probe_arg(arg, &tk->tp.size, parg,
754 is_return, true,
755 kprobes_fetch_type_table);
756 if (ret) {
757 pr_info("Parse error at argument[%d]. (%d)\n", i, ret);
758 goto error;
759 }
760 }
761
762 ret = register_trace_kprobe(tk);
763 if (ret)
764 goto error;
765 return 0;
766
767error:
768 free_trace_kprobe(tk);
769 return ret;
770}
771
772static int release_all_trace_kprobes(void)
773{
774 struct trace_kprobe *tk;
775 int ret = 0;
776
777 mutex_lock(&probe_lock);
778 /* Ensure no probe is in use. */
779 list_for_each_entry(tk, &probe_list, list)
780 if (trace_probe_is_enabled(&tk->tp)) {
781 ret = -EBUSY;
782 goto end;
783 }
784 /* TODO: Use batch unregistration */
785 while (!list_empty(&probe_list)) {
786 tk = list_entry(probe_list.next, struct trace_kprobe, list);
787 ret = unregister_trace_kprobe(tk);
788 if (ret)
789 goto end;
790 free_trace_kprobe(tk);
791 }
792
793end:
794 mutex_unlock(&probe_lock);
795
796 return ret;
797}
798
799/* Probes listing interfaces */
800static void *probes_seq_start(struct seq_file *m, loff_t *pos)
801{
802 mutex_lock(&probe_lock);
803 return seq_list_start(&probe_list, *pos);
804}
805
806static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
807{
808 return seq_list_next(v, &probe_list, pos);
809}
810
811static void probes_seq_stop(struct seq_file *m, void *v)
812{
813 mutex_unlock(&probe_lock);
814}
815
816static int probes_seq_show(struct seq_file *m, void *v)
817{
818 struct trace_kprobe *tk = v;
819 int i;
820
821 seq_putc(m, trace_kprobe_is_return(tk) ? 'r' : 'p');
822 seq_printf(m, ":%s/%s", tk->tp.call.class->system,
823 trace_event_name(&tk->tp.call));
824
825 if (!tk->symbol)
826 seq_printf(m, " 0x%p", tk->rp.kp.addr);
827 else if (tk->rp.kp.offset)
828 seq_printf(m, " %s+%u", trace_kprobe_symbol(tk),
829 tk->rp.kp.offset);
830 else
831 seq_printf(m, " %s", trace_kprobe_symbol(tk));
832
833 for (i = 0; i < tk->tp.nr_args; i++)
834 seq_printf(m, " %s=%s", tk->tp.args[i].name, tk->tp.args[i].comm);
835 seq_putc(m, '\n');
836
837 return 0;
838}
839
840static const struct seq_operations probes_seq_op = {
841 .start = probes_seq_start,
842 .next = probes_seq_next,
843 .stop = probes_seq_stop,
844 .show = probes_seq_show
845};
846
847static int probes_open(struct inode *inode, struct file *file)
848{
849 int ret;
850
851 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
852 ret = release_all_trace_kprobes();
853 if (ret < 0)
854 return ret;
855 }
856
857 return seq_open(file, &probes_seq_op);
858}
859
860static ssize_t probes_write(struct file *file, const char __user *buffer,
861 size_t count, loff_t *ppos)
862{
863 return traceprobe_probes_write(file, buffer, count, ppos,
864 create_trace_kprobe);
865}
866
867static const struct file_operations kprobe_events_ops = {
868 .owner = THIS_MODULE,
869 .open = probes_open,
870 .read = seq_read,
871 .llseek = seq_lseek,
872 .release = seq_release,
873 .write = probes_write,
874};
875
876/* Probes profiling interfaces */
877static int probes_profile_seq_show(struct seq_file *m, void *v)
878{
879 struct trace_kprobe *tk = v;
880 unsigned long nhit = 0;
881 int cpu;
882
883 for_each_possible_cpu(cpu)
884 nhit += *per_cpu_ptr(tk->nhit, cpu);
885
886 seq_printf(m, " %-44s %15lu %15lu\n",
887 trace_event_name(&tk->tp.call), nhit,
888 tk->rp.kp.nmissed);
889
890 return 0;
891}
892
893static const struct seq_operations profile_seq_op = {
894 .start = probes_seq_start,
895 .next = probes_seq_next,
896 .stop = probes_seq_stop,
897 .show = probes_profile_seq_show
898};
899
900static int profile_open(struct inode *inode, struct file *file)
901{
902 return seq_open(file, &profile_seq_op);
903}
904
905static const struct file_operations kprobe_profile_ops = {
906 .owner = THIS_MODULE,
907 .open = profile_open,
908 .read = seq_read,
909 .llseek = seq_lseek,
910 .release = seq_release,
911};
912
913/* Kprobe handler */
914static nokprobe_inline void
915__kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs,
916 struct trace_event_file *trace_file)
917{
918 struct kprobe_trace_entry_head *entry;
919 struct ring_buffer_event *event;
920 struct ring_buffer *buffer;
921 int size, dsize, pc;
922 unsigned long irq_flags;
923 struct trace_event_call *call = &tk->tp.call;
924
925 WARN_ON(call != trace_file->event_call);
926
927 if (trace_trigger_soft_disabled(trace_file))
928 return;
929
930 local_save_flags(irq_flags);
931 pc = preempt_count();
932
933 dsize = __get_data_size(&tk->tp, regs);
934 size = sizeof(*entry) + tk->tp.size + dsize;
935
936 event = trace_event_buffer_lock_reserve(&buffer, trace_file,
937 call->event.type,
938 size, irq_flags, pc);
939 if (!event)
940 return;
941
942 entry = ring_buffer_event_data(event);
943 entry->ip = (unsigned long)tk->rp.kp.addr;
944 store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
945
946 event_trigger_unlock_commit_regs(trace_file, buffer, event,
947 entry, irq_flags, pc, regs);
948}
949
950static void
951kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs)
952{
953 struct event_file_link *link;
954
955 list_for_each_entry_rcu(link, &tk->tp.files, list)
956 __kprobe_trace_func(tk, regs, link->file);
957}
958NOKPROBE_SYMBOL(kprobe_trace_func);
959
960/* Kretprobe handler */
961static nokprobe_inline void
962__kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
963 struct pt_regs *regs,
964 struct trace_event_file *trace_file)
965{
966 struct kretprobe_trace_entry_head *entry;
967 struct ring_buffer_event *event;
968 struct ring_buffer *buffer;
969 int size, pc, dsize;
970 unsigned long irq_flags;
971 struct trace_event_call *call = &tk->tp.call;
972
973 WARN_ON(call != trace_file->event_call);
974
975 if (trace_trigger_soft_disabled(trace_file))
976 return;
977
978 local_save_flags(irq_flags);
979 pc = preempt_count();
980
981 dsize = __get_data_size(&tk->tp, regs);
982 size = sizeof(*entry) + tk->tp.size + dsize;
983
984 event = trace_event_buffer_lock_reserve(&buffer, trace_file,
985 call->event.type,
986 size, irq_flags, pc);
987 if (!event)
988 return;
989
990 entry = ring_buffer_event_data(event);
991 entry->func = (unsigned long)tk->rp.kp.addr;
992 entry->ret_ip = (unsigned long)ri->ret_addr;
993 store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
994
995 event_trigger_unlock_commit_regs(trace_file, buffer, event,
996 entry, irq_flags, pc, regs);
997}
998
999static void
1000kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1001 struct pt_regs *regs)
1002{
1003 struct event_file_link *link;
1004
1005 list_for_each_entry_rcu(link, &tk->tp.files, list)
1006 __kretprobe_trace_func(tk, ri, regs, link->file);
1007}
1008NOKPROBE_SYMBOL(kretprobe_trace_func);
1009
1010/* Event entry printers */
1011static enum print_line_t
1012print_kprobe_event(struct trace_iterator *iter, int flags,
1013 struct trace_event *event)
1014{
1015 struct kprobe_trace_entry_head *field;
1016 struct trace_seq *s = &iter->seq;
1017 struct trace_probe *tp;
1018 u8 *data;
1019 int i;
1020
1021 field = (struct kprobe_trace_entry_head *)iter->ent;
1022 tp = container_of(event, struct trace_probe, call.event);
1023
1024 trace_seq_printf(s, "%s: (", trace_event_name(&tp->call));
1025
1026 if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
1027 goto out;
1028
1029 trace_seq_putc(s, ')');
1030
1031 data = (u8 *)&field[1];
1032 for (i = 0; i < tp->nr_args; i++)
1033 if (!tp->args[i].type->print(s, tp->args[i].name,
1034 data + tp->args[i].offset, field))
1035 goto out;
1036
1037 trace_seq_putc(s, '\n');
1038 out:
1039 return trace_handle_return(s);
1040}
1041
1042static enum print_line_t
1043print_kretprobe_event(struct trace_iterator *iter, int flags,
1044 struct trace_event *event)
1045{
1046 struct kretprobe_trace_entry_head *field;
1047 struct trace_seq *s = &iter->seq;
1048 struct trace_probe *tp;
1049 u8 *data;
1050 int i;
1051
1052 field = (struct kretprobe_trace_entry_head *)iter->ent;
1053 tp = container_of(event, struct trace_probe, call.event);
1054
1055 trace_seq_printf(s, "%s: (", trace_event_name(&tp->call));
1056
1057 if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1058 goto out;
1059
1060 trace_seq_puts(s, " <- ");
1061
1062 if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1063 goto out;
1064
1065 trace_seq_putc(s, ')');
1066
1067 data = (u8 *)&field[1];
1068 for (i = 0; i < tp->nr_args; i++)
1069 if (!tp->args[i].type->print(s, tp->args[i].name,
1070 data + tp->args[i].offset, field))
1071 goto out;
1072
1073 trace_seq_putc(s, '\n');
1074
1075 out:
1076 return trace_handle_return(s);
1077}
1078
1079
1080static int kprobe_event_define_fields(struct trace_event_call *event_call)
1081{
1082 int ret, i;
1083 struct kprobe_trace_entry_head field;
1084 struct trace_kprobe *tk = (struct trace_kprobe *)event_call->data;
1085
1086 DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1087 /* Set argument names as fields */
1088 for (i = 0; i < tk->tp.nr_args; i++) {
1089 struct probe_arg *parg = &tk->tp.args[i];
1090
1091 ret = trace_define_field(event_call, parg->type->fmttype,
1092 parg->name,
1093 sizeof(field) + parg->offset,
1094 parg->type->size,
1095 parg->type->is_signed,
1096 FILTER_OTHER);
1097 if (ret)
1098 return ret;
1099 }
1100 return 0;
1101}
1102
1103static int kretprobe_event_define_fields(struct trace_event_call *event_call)
1104{
1105 int ret, i;
1106 struct kretprobe_trace_entry_head field;
1107 struct trace_kprobe *tk = (struct trace_kprobe *)event_call->data;
1108
1109 DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1110 DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1111 /* Set argument names as fields */
1112 for (i = 0; i < tk->tp.nr_args; i++) {
1113 struct probe_arg *parg = &tk->tp.args[i];
1114
1115 ret = trace_define_field(event_call, parg->type->fmttype,
1116 parg->name,
1117 sizeof(field) + parg->offset,
1118 parg->type->size,
1119 parg->type->is_signed,
1120 FILTER_OTHER);
1121 if (ret)
1122 return ret;
1123 }
1124 return 0;
1125}
1126
1127#ifdef CONFIG_PERF_EVENTS
1128
1129/* Kprobe profile handler */
1130static void
1131kprobe_perf_func(struct trace_kprobe *tk, struct pt_regs *regs)
1132{
1133 struct trace_event_call *call = &tk->tp.call;
1134 struct bpf_prog *prog = call->prog;
1135 struct kprobe_trace_entry_head *entry;
1136 struct hlist_head *head;
1137 int size, __size, dsize;
1138 int rctx;
1139
1140 if (prog && !trace_call_bpf(prog, regs))
1141 return;
1142
1143 head = this_cpu_ptr(call->perf_events);
1144 if (hlist_empty(head))
1145 return;
1146
1147 dsize = __get_data_size(&tk->tp, regs);
1148 __size = sizeof(*entry) + tk->tp.size + dsize;
1149 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1150 size -= sizeof(u32);
1151
1152 entry = perf_trace_buf_prepare(size, call->event.type, NULL, &rctx);
1153 if (!entry)
1154 return;
1155
1156 entry->ip = (unsigned long)tk->rp.kp.addr;
1157 memset(&entry[1], 0, dsize);
1158 store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1159 perf_trace_buf_submit(entry, size, rctx, 0, 1, regs, head, NULL);
1160}
1161NOKPROBE_SYMBOL(kprobe_perf_func);
1162
1163/* Kretprobe profile handler */
1164static void
1165kretprobe_perf_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1166 struct pt_regs *regs)
1167{
1168 struct trace_event_call *call = &tk->tp.call;
1169 struct bpf_prog *prog = call->prog;
1170 struct kretprobe_trace_entry_head *entry;
1171 struct hlist_head *head;
1172 int size, __size, dsize;
1173 int rctx;
1174
1175 if (prog && !trace_call_bpf(prog, regs))
1176 return;
1177
1178 head = this_cpu_ptr(call->perf_events);
1179 if (hlist_empty(head))
1180 return;
1181
1182 dsize = __get_data_size(&tk->tp, regs);
1183 __size = sizeof(*entry) + tk->tp.size + dsize;
1184 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1185 size -= sizeof(u32);
1186
1187 entry = perf_trace_buf_prepare(size, call->event.type, NULL, &rctx);
1188 if (!entry)
1189 return;
1190
1191 entry->func = (unsigned long)tk->rp.kp.addr;
1192 entry->ret_ip = (unsigned long)ri->ret_addr;
1193 store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1194 perf_trace_buf_submit(entry, size, rctx, 0, 1, regs, head, NULL);
1195}
1196NOKPROBE_SYMBOL(kretprobe_perf_func);
1197#endif /* CONFIG_PERF_EVENTS */
1198
1199/*
1200 * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex.
1201 *
1202 * kprobe_trace_self_tests_init() does enable_trace_probe/disable_trace_probe
1203 * lockless, but we can't race with this __init function.
1204 */
1205static int kprobe_register(struct trace_event_call *event,
1206 enum trace_reg type, void *data)
1207{
1208 struct trace_kprobe *tk = (struct trace_kprobe *)event->data;
1209 struct trace_event_file *file = data;
1210
1211 switch (type) {
1212 case TRACE_REG_REGISTER:
1213 return enable_trace_kprobe(tk, file);
1214 case TRACE_REG_UNREGISTER:
1215 return disable_trace_kprobe(tk, file);
1216
1217#ifdef CONFIG_PERF_EVENTS
1218 case TRACE_REG_PERF_REGISTER:
1219 return enable_trace_kprobe(tk, NULL);
1220 case TRACE_REG_PERF_UNREGISTER:
1221 return disable_trace_kprobe(tk, NULL);
1222 case TRACE_REG_PERF_OPEN:
1223 case TRACE_REG_PERF_CLOSE:
1224 case TRACE_REG_PERF_ADD:
1225 case TRACE_REG_PERF_DEL:
1226 return 0;
1227#endif
1228 }
1229 return 0;
1230}
1231
1232static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1233{
1234 struct trace_kprobe *tk = container_of(kp, struct trace_kprobe, rp.kp);
1235
1236 raw_cpu_inc(*tk->nhit);
1237
1238 if (tk->tp.flags & TP_FLAG_TRACE)
1239 kprobe_trace_func(tk, regs);
1240#ifdef CONFIG_PERF_EVENTS
1241 if (tk->tp.flags & TP_FLAG_PROFILE)
1242 kprobe_perf_func(tk, regs);
1243#endif
1244 return 0; /* We don't tweek kernel, so just return 0 */
1245}
1246NOKPROBE_SYMBOL(kprobe_dispatcher);
1247
1248static int
1249kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1250{
1251 struct trace_kprobe *tk = container_of(ri->rp, struct trace_kprobe, rp);
1252
1253 raw_cpu_inc(*tk->nhit);
1254
1255 if (tk->tp.flags & TP_FLAG_TRACE)
1256 kretprobe_trace_func(tk, ri, regs);
1257#ifdef CONFIG_PERF_EVENTS
1258 if (tk->tp.flags & TP_FLAG_PROFILE)
1259 kretprobe_perf_func(tk, ri, regs);
1260#endif
1261 return 0; /* We don't tweek kernel, so just return 0 */
1262}
1263NOKPROBE_SYMBOL(kretprobe_dispatcher);
1264
1265static struct trace_event_functions kretprobe_funcs = {
1266 .trace = print_kretprobe_event
1267};
1268
1269static struct trace_event_functions kprobe_funcs = {
1270 .trace = print_kprobe_event
1271};
1272
1273static int register_kprobe_event(struct trace_kprobe *tk)
1274{
1275 struct trace_event_call *call = &tk->tp.call;
1276 int ret;
1277
1278 /* Initialize trace_event_call */
1279 INIT_LIST_HEAD(&call->class->fields);
1280 if (trace_kprobe_is_return(tk)) {
1281 call->event.funcs = &kretprobe_funcs;
1282 call->class->define_fields = kretprobe_event_define_fields;
1283 } else {
1284 call->event.funcs = &kprobe_funcs;
1285 call->class->define_fields = kprobe_event_define_fields;
1286 }
1287 if (set_print_fmt(&tk->tp, trace_kprobe_is_return(tk)) < 0)
1288 return -ENOMEM;
1289 ret = register_trace_event(&call->event);
1290 if (!ret) {
1291 kfree(call->print_fmt);
1292 return -ENODEV;
1293 }
1294 call->flags = TRACE_EVENT_FL_KPROBE;
1295 call->class->reg = kprobe_register;
1296 call->data = tk;
1297 ret = trace_add_event_call(call);
1298 if (ret) {
1299 pr_info("Failed to register kprobe event: %s\n",
1300 trace_event_name(call));
1301 kfree(call->print_fmt);
1302 unregister_trace_event(&call->event);
1303 }
1304 return ret;
1305}
1306
1307static int unregister_kprobe_event(struct trace_kprobe *tk)
1308{
1309 int ret;
1310
1311 /* tp->event is unregistered in trace_remove_event_call() */
1312 ret = trace_remove_event_call(&tk->tp.call);
1313 if (!ret)
1314 kfree(tk->tp.call.print_fmt);
1315 return ret;
1316}
1317
1318/* Make a tracefs interface for controlling probe points */
1319static __init int init_kprobe_trace(void)
1320{
1321 struct dentry *d_tracer;
1322 struct dentry *entry;
1323
1324 if (register_module_notifier(&trace_kprobe_module_nb))
1325 return -EINVAL;
1326
1327 d_tracer = tracing_init_dentry();
1328 if (IS_ERR(d_tracer))
1329 return 0;
1330
1331 entry = tracefs_create_file("kprobe_events", 0644, d_tracer,
1332 NULL, &kprobe_events_ops);
1333
1334 /* Event list interface */
1335 if (!entry)
1336 pr_warn("Could not create tracefs 'kprobe_events' entry\n");
1337
1338 /* Profile interface */
1339 entry = tracefs_create_file("kprobe_profile", 0444, d_tracer,
1340 NULL, &kprobe_profile_ops);
1341
1342 if (!entry)
1343 pr_warn("Could not create tracefs 'kprobe_profile' entry\n");
1344 return 0;
1345}
1346fs_initcall(init_kprobe_trace);
1347
1348
1349#ifdef CONFIG_FTRACE_STARTUP_TEST
1350
1351/*
1352 * The "__used" keeps gcc from removing the function symbol
1353 * from the kallsyms table.
1354 */
1355static __used int kprobe_trace_selftest_target(int a1, int a2, int a3,
1356 int a4, int a5, int a6)
1357{
1358 return a1 + a2 + a3 + a4 + a5 + a6;
1359}
1360
1361static struct trace_event_file *
1362find_trace_probe_file(struct trace_kprobe *tk, struct trace_array *tr)
1363{
1364 struct trace_event_file *file;
1365
1366 list_for_each_entry(file, &tr->events, list)
1367 if (file->event_call == &tk->tp.call)
1368 return file;
1369
1370 return NULL;
1371}
1372
1373/*
1374 * Nobody but us can call enable_trace_kprobe/disable_trace_kprobe at this
1375 * stage, we can do this lockless.
1376 */
1377static __init int kprobe_trace_self_tests_init(void)
1378{
1379 int ret, warn = 0;
1380 int (*target)(int, int, int, int, int, int);
1381 struct trace_kprobe *tk;
1382 struct trace_event_file *file;
1383
1384 if (tracing_is_disabled())
1385 return -ENODEV;
1386
1387 target = kprobe_trace_selftest_target;
1388
1389 pr_info("Testing kprobe tracing: ");
1390
1391 ret = traceprobe_command("p:testprobe kprobe_trace_selftest_target "
1392 "$stack $stack0 +0($stack)",
1393 create_trace_kprobe);
1394 if (WARN_ON_ONCE(ret)) {
1395 pr_warn("error on probing function entry.\n");
1396 warn++;
1397 } else {
1398 /* Enable trace point */
1399 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1400 if (WARN_ON_ONCE(tk == NULL)) {
1401 pr_warn("error on getting new probe.\n");
1402 warn++;
1403 } else {
1404 file = find_trace_probe_file(tk, top_trace_array());
1405 if (WARN_ON_ONCE(file == NULL)) {
1406 pr_warn("error on getting probe file.\n");
1407 warn++;
1408 } else
1409 enable_trace_kprobe(tk, file);
1410 }
1411 }
1412
1413 ret = traceprobe_command("r:testprobe2 kprobe_trace_selftest_target "
1414 "$retval", create_trace_kprobe);
1415 if (WARN_ON_ONCE(ret)) {
1416 pr_warn("error on probing function return.\n");
1417 warn++;
1418 } else {
1419 /* Enable trace point */
1420 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1421 if (WARN_ON_ONCE(tk == NULL)) {
1422 pr_warn("error on getting 2nd new probe.\n");
1423 warn++;
1424 } else {
1425 file = find_trace_probe_file(tk, top_trace_array());
1426 if (WARN_ON_ONCE(file == NULL)) {
1427 pr_warn("error on getting probe file.\n");
1428 warn++;
1429 } else
1430 enable_trace_kprobe(tk, file);
1431 }
1432 }
1433
1434 if (warn)
1435 goto end;
1436
1437 ret = target(1, 2, 3, 4, 5, 6);
1438
1439 /* Disable trace points before removing it */
1440 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1441 if (WARN_ON_ONCE(tk == NULL)) {
1442 pr_warn("error on getting test probe.\n");
1443 warn++;
1444 } else {
1445 file = find_trace_probe_file(tk, top_trace_array());
1446 if (WARN_ON_ONCE(file == NULL)) {
1447 pr_warn("error on getting probe file.\n");
1448 warn++;
1449 } else
1450 disable_trace_kprobe(tk, file);
1451 }
1452
1453 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1454 if (WARN_ON_ONCE(tk == NULL)) {
1455 pr_warn("error on getting 2nd test probe.\n");
1456 warn++;
1457 } else {
1458 file = find_trace_probe_file(tk, top_trace_array());
1459 if (WARN_ON_ONCE(file == NULL)) {
1460 pr_warn("error on getting probe file.\n");
1461 warn++;
1462 } else
1463 disable_trace_kprobe(tk, file);
1464 }
1465
1466 ret = traceprobe_command("-:testprobe", create_trace_kprobe);
1467 if (WARN_ON_ONCE(ret)) {
1468 pr_warn("error on deleting a probe.\n");
1469 warn++;
1470 }
1471
1472 ret = traceprobe_command("-:testprobe2", create_trace_kprobe);
1473 if (WARN_ON_ONCE(ret)) {
1474 pr_warn("error on deleting a probe.\n");
1475 warn++;
1476 }
1477
1478end:
1479 release_all_trace_kprobes();
1480 if (warn)
1481 pr_cont("NG: Some tests are failed. Please check them.\n");
1482 else
1483 pr_cont("OK\n");
1484 return 0;
1485}
1486
1487late_initcall(kprobe_trace_self_tests_init);
1488
1489#endif
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Kprobes-based tracing events
4 *
5 * Created by Masami Hiramatsu <mhiramat@redhat.com>
6 *
7 */
8#define pr_fmt(fmt) "trace_kprobe: " fmt
9
10#include <linux/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 strlcpy(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 %s\n",
491 trace_kprobe_symbol(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 __trace_kprobe_create(int argc, const char *argv[])
709{
710 /*
711 * Argument syntax:
712 * - Add kprobe:
713 * p[:[GRP/][EVENT]] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
714 * - Add kretprobe:
715 * r[MAXACTIVE][:[GRP/][EVENT]] [MOD:]KSYM[+0] [FETCHARGS]
716 * Or
717 * p[:[GRP/][EVENT]] [MOD:]KSYM[+0]%return [FETCHARGS]
718 *
719 * Fetch args:
720 * $retval : fetch return value
721 * $stack : fetch stack address
722 * $stackN : fetch Nth of stack (N:0-)
723 * $comm : fetch current task comm
724 * @ADDR : fetch memory at ADDR (ADDR should be in kernel)
725 * @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
726 * %REG : fetch register REG
727 * Dereferencing memory fetch:
728 * +|-offs(ARG) : fetch memory at ARG +|- offs address.
729 * Alias name of args:
730 * NAME=FETCHARG : set NAME as alias of FETCHARG.
731 * Type of args:
732 * FETCHARG:TYPE : use TYPE instead of unsigned long.
733 */
734 struct trace_kprobe *tk = NULL;
735 int i, len, ret = 0;
736 bool is_return = false;
737 char *symbol = NULL, *tmp = NULL;
738 const char *event = NULL, *group = KPROBE_EVENT_SYSTEM;
739 enum probe_print_type ptype;
740 int maxactive = 0;
741 long offset = 0;
742 void *addr = NULL;
743 char buf[MAX_EVENT_NAME_LEN];
744 char gbuf[MAX_EVENT_NAME_LEN];
745 unsigned int flags = TPARG_FL_KERNEL;
746
747 switch (argv[0][0]) {
748 case 'r':
749 is_return = true;
750 break;
751 case 'p':
752 break;
753 default:
754 return -ECANCELED;
755 }
756 if (argc < 2)
757 return -ECANCELED;
758
759 trace_probe_log_init("trace_kprobe", argc, argv);
760
761 event = strchr(&argv[0][1], ':');
762 if (event)
763 event++;
764
765 if (isdigit(argv[0][1])) {
766 if (!is_return) {
767 trace_probe_log_err(1, MAXACT_NO_KPROBE);
768 goto parse_error;
769 }
770 if (event)
771 len = event - &argv[0][1] - 1;
772 else
773 len = strlen(&argv[0][1]);
774 if (len > MAX_EVENT_NAME_LEN - 1) {
775 trace_probe_log_err(1, BAD_MAXACT);
776 goto parse_error;
777 }
778 memcpy(buf, &argv[0][1], len);
779 buf[len] = '\0';
780 ret = kstrtouint(buf, 0, &maxactive);
781 if (ret || !maxactive) {
782 trace_probe_log_err(1, BAD_MAXACT);
783 goto parse_error;
784 }
785 /* kretprobes instances are iterated over via a list. The
786 * maximum should stay reasonable.
787 */
788 if (maxactive > KRETPROBE_MAXACTIVE_MAX) {
789 trace_probe_log_err(1, MAXACT_TOO_BIG);
790 goto parse_error;
791 }
792 }
793
794 /* try to parse an address. if that fails, try to read the
795 * input as a symbol. */
796 if (kstrtoul(argv[1], 0, (unsigned long *)&addr)) {
797 trace_probe_log_set_index(1);
798 /* Check whether uprobe event specified */
799 if (strchr(argv[1], '/') && strchr(argv[1], ':')) {
800 ret = -ECANCELED;
801 goto error;
802 }
803 /* a symbol specified */
804 symbol = kstrdup(argv[1], GFP_KERNEL);
805 if (!symbol)
806 return -ENOMEM;
807
808 tmp = strchr(symbol, '%');
809 if (tmp) {
810 if (!strcmp(tmp, "%return")) {
811 *tmp = '\0';
812 is_return = true;
813 } else {
814 trace_probe_log_err(tmp - symbol, BAD_ADDR_SUFFIX);
815 goto parse_error;
816 }
817 }
818
819 /* TODO: support .init module functions */
820 ret = traceprobe_split_symbol_offset(symbol, &offset);
821 if (ret || offset < 0 || offset > UINT_MAX) {
822 trace_probe_log_err(0, BAD_PROBE_ADDR);
823 goto parse_error;
824 }
825 if (is_return)
826 flags |= TPARG_FL_RETURN;
827 ret = kprobe_on_func_entry(NULL, symbol, offset);
828 if (ret == 0)
829 flags |= TPARG_FL_FENTRY;
830 /* Defer the ENOENT case until register kprobe */
831 if (ret == -EINVAL && is_return) {
832 trace_probe_log_err(0, BAD_RETPROBE);
833 goto parse_error;
834 }
835 }
836
837 trace_probe_log_set_index(0);
838 if (event) {
839 ret = traceprobe_parse_event_name(&event, &group, gbuf,
840 event - argv[0]);
841 if (ret)
842 goto parse_error;
843 }
844
845 if (!event) {
846 /* Make a new event name */
847 if (symbol)
848 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
849 is_return ? 'r' : 'p', symbol, offset);
850 else
851 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
852 is_return ? 'r' : 'p', addr);
853 sanitize_event_name(buf);
854 event = buf;
855 }
856
857 /* setup a probe */
858 tk = alloc_trace_kprobe(group, event, addr, symbol, offset, maxactive,
859 argc - 2, is_return);
860 if (IS_ERR(tk)) {
861 ret = PTR_ERR(tk);
862 /* This must return -ENOMEM, else there is a bug */
863 WARN_ON_ONCE(ret != -ENOMEM);
864 goto out; /* We know tk is not allocated */
865 }
866 argc -= 2; argv += 2;
867
868 /* parse arguments */
869 for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
870 trace_probe_log_set_index(i + 2);
871 ret = traceprobe_parse_probe_arg(&tk->tp, i, argv[i], flags);
872 if (ret)
873 goto error; /* This can be -ENOMEM */
874 }
875
876 ptype = is_return ? PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL;
877 ret = traceprobe_set_print_fmt(&tk->tp, ptype);
878 if (ret < 0)
879 goto error;
880
881 ret = register_trace_kprobe(tk);
882 if (ret) {
883 trace_probe_log_set_index(1);
884 if (ret == -EILSEQ)
885 trace_probe_log_err(0, BAD_INSN_BNDRY);
886 else if (ret == -ENOENT)
887 trace_probe_log_err(0, BAD_PROBE_ADDR);
888 else if (ret != -ENOMEM && ret != -EEXIST)
889 trace_probe_log_err(0, FAIL_REG_PROBE);
890 goto error;
891 }
892
893out:
894 trace_probe_log_clear();
895 kfree(symbol);
896 return ret;
897
898parse_error:
899 ret = -EINVAL;
900error:
901 free_trace_kprobe(tk);
902 goto out;
903}
904
905static int trace_kprobe_create(const char *raw_command)
906{
907 return trace_probe_create(raw_command, __trace_kprobe_create);
908}
909
910static int create_or_delete_trace_kprobe(const char *raw_command)
911{
912 int ret;
913
914 if (raw_command[0] == '-')
915 return dyn_event_release(raw_command, &trace_kprobe_ops);
916
917 ret = trace_kprobe_create(raw_command);
918 return ret == -ECANCELED ? -EINVAL : ret;
919}
920
921static int trace_kprobe_run_command(struct dynevent_cmd *cmd)
922{
923 return create_or_delete_trace_kprobe(cmd->seq.buffer);
924}
925
926/**
927 * kprobe_event_cmd_init - Initialize a kprobe event command object
928 * @cmd: A pointer to the dynevent_cmd struct representing the new event
929 * @buf: A pointer to the buffer used to build the command
930 * @maxlen: The length of the buffer passed in @buf
931 *
932 * Initialize a synthetic event command object. Use this before
933 * calling any of the other kprobe_event functions.
934 */
935void kprobe_event_cmd_init(struct dynevent_cmd *cmd, char *buf, int maxlen)
936{
937 dynevent_cmd_init(cmd, buf, maxlen, DYNEVENT_TYPE_KPROBE,
938 trace_kprobe_run_command);
939}
940EXPORT_SYMBOL_GPL(kprobe_event_cmd_init);
941
942/**
943 * __kprobe_event_gen_cmd_start - Generate a kprobe event command from arg list
944 * @cmd: A pointer to the dynevent_cmd struct representing the new event
945 * @name: The name of the kprobe event
946 * @loc: The location of the kprobe event
947 * @kretprobe: Is this a return probe?
948 * @args: Variable number of arg (pairs), one pair for each field
949 *
950 * NOTE: Users normally won't want to call this function directly, but
951 * rather use the kprobe_event_gen_cmd_start() wrapper, which automatically
952 * adds a NULL to the end of the arg list. If this function is used
953 * directly, make sure the last arg in the variable arg list is NULL.
954 *
955 * Generate a kprobe event command to be executed by
956 * kprobe_event_gen_cmd_end(). This function can be used to generate the
957 * complete command or only the first part of it; in the latter case,
958 * kprobe_event_add_fields() can be used to add more fields following this.
959 *
960 * Unlikely the synth_event_gen_cmd_start(), @loc must be specified. This
961 * returns -EINVAL if @loc == NULL.
962 *
963 * Return: 0 if successful, error otherwise.
964 */
965int __kprobe_event_gen_cmd_start(struct dynevent_cmd *cmd, bool kretprobe,
966 const char *name, const char *loc, ...)
967{
968 char buf[MAX_EVENT_NAME_LEN];
969 struct dynevent_arg arg;
970 va_list args;
971 int ret;
972
973 if (cmd->type != DYNEVENT_TYPE_KPROBE)
974 return -EINVAL;
975
976 if (!loc)
977 return -EINVAL;
978
979 if (kretprobe)
980 snprintf(buf, MAX_EVENT_NAME_LEN, "r:kprobes/%s", name);
981 else
982 snprintf(buf, MAX_EVENT_NAME_LEN, "p:kprobes/%s", name);
983
984 ret = dynevent_str_add(cmd, buf);
985 if (ret)
986 return ret;
987
988 dynevent_arg_init(&arg, 0);
989 arg.str = loc;
990 ret = dynevent_arg_add(cmd, &arg, NULL);
991 if (ret)
992 return ret;
993
994 va_start(args, loc);
995 for (;;) {
996 const char *field;
997
998 field = va_arg(args, const char *);
999 if (!field)
1000 break;
1001
1002 if (++cmd->n_fields > MAX_TRACE_ARGS) {
1003 ret = -EINVAL;
1004 break;
1005 }
1006
1007 arg.str = field;
1008 ret = dynevent_arg_add(cmd, &arg, NULL);
1009 if (ret)
1010 break;
1011 }
1012 va_end(args);
1013
1014 return ret;
1015}
1016EXPORT_SYMBOL_GPL(__kprobe_event_gen_cmd_start);
1017
1018/**
1019 * __kprobe_event_add_fields - Add probe fields to a kprobe command from arg list
1020 * @cmd: A pointer to the dynevent_cmd struct representing the new event
1021 * @args: Variable number of arg (pairs), one pair for each field
1022 *
1023 * NOTE: Users normally won't want to call this function directly, but
1024 * rather use the kprobe_event_add_fields() wrapper, which
1025 * automatically adds a NULL to the end of the arg list. If this
1026 * function is used directly, make sure the last arg in the variable
1027 * arg list is NULL.
1028 *
1029 * Add probe fields to an existing kprobe command using a variable
1030 * list of args. Fields are added in the same order they're listed.
1031 *
1032 * Return: 0 if successful, error otherwise.
1033 */
1034int __kprobe_event_add_fields(struct dynevent_cmd *cmd, ...)
1035{
1036 struct dynevent_arg arg;
1037 va_list args;
1038 int ret = 0;
1039
1040 if (cmd->type != DYNEVENT_TYPE_KPROBE)
1041 return -EINVAL;
1042
1043 dynevent_arg_init(&arg, 0);
1044
1045 va_start(args, cmd);
1046 for (;;) {
1047 const char *field;
1048
1049 field = va_arg(args, const char *);
1050 if (!field)
1051 break;
1052
1053 if (++cmd->n_fields > MAX_TRACE_ARGS) {
1054 ret = -EINVAL;
1055 break;
1056 }
1057
1058 arg.str = field;
1059 ret = dynevent_arg_add(cmd, &arg, NULL);
1060 if (ret)
1061 break;
1062 }
1063 va_end(args);
1064
1065 return ret;
1066}
1067EXPORT_SYMBOL_GPL(__kprobe_event_add_fields);
1068
1069/**
1070 * kprobe_event_delete - Delete a kprobe event
1071 * @name: The name of the kprobe event to delete
1072 *
1073 * Delete a kprobe event with the give @name from kernel code rather
1074 * than directly from the command line.
1075 *
1076 * Return: 0 if successful, error otherwise.
1077 */
1078int kprobe_event_delete(const char *name)
1079{
1080 char buf[MAX_EVENT_NAME_LEN];
1081
1082 snprintf(buf, MAX_EVENT_NAME_LEN, "-:%s", name);
1083
1084 return create_or_delete_trace_kprobe(buf);
1085}
1086EXPORT_SYMBOL_GPL(kprobe_event_delete);
1087
1088static int trace_kprobe_release(struct dyn_event *ev)
1089{
1090 struct trace_kprobe *tk = to_trace_kprobe(ev);
1091 int ret = unregister_trace_kprobe(tk);
1092
1093 if (!ret)
1094 free_trace_kprobe(tk);
1095 return ret;
1096}
1097
1098static int trace_kprobe_show(struct seq_file *m, struct dyn_event *ev)
1099{
1100 struct trace_kprobe *tk = to_trace_kprobe(ev);
1101 int i;
1102
1103 seq_putc(m, trace_kprobe_is_return(tk) ? 'r' : 'p');
1104 if (trace_kprobe_is_return(tk) && tk->rp.maxactive)
1105 seq_printf(m, "%d", tk->rp.maxactive);
1106 seq_printf(m, ":%s/%s", trace_probe_group_name(&tk->tp),
1107 trace_probe_name(&tk->tp));
1108
1109 if (!tk->symbol)
1110 seq_printf(m, " 0x%p", tk->rp.kp.addr);
1111 else if (tk->rp.kp.offset)
1112 seq_printf(m, " %s+%u", trace_kprobe_symbol(tk),
1113 tk->rp.kp.offset);
1114 else
1115 seq_printf(m, " %s", trace_kprobe_symbol(tk));
1116
1117 for (i = 0; i < tk->tp.nr_args; i++)
1118 seq_printf(m, " %s=%s", tk->tp.args[i].name, tk->tp.args[i].comm);
1119 seq_putc(m, '\n');
1120
1121 return 0;
1122}
1123
1124static int probes_seq_show(struct seq_file *m, void *v)
1125{
1126 struct dyn_event *ev = v;
1127
1128 if (!is_trace_kprobe(ev))
1129 return 0;
1130
1131 return trace_kprobe_show(m, ev);
1132}
1133
1134static const struct seq_operations probes_seq_op = {
1135 .start = dyn_event_seq_start,
1136 .next = dyn_event_seq_next,
1137 .stop = dyn_event_seq_stop,
1138 .show = probes_seq_show
1139};
1140
1141static int probes_open(struct inode *inode, struct file *file)
1142{
1143 int ret;
1144
1145 ret = security_locked_down(LOCKDOWN_TRACEFS);
1146 if (ret)
1147 return ret;
1148
1149 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
1150 ret = dyn_events_release_all(&trace_kprobe_ops);
1151 if (ret < 0)
1152 return ret;
1153 }
1154
1155 return seq_open(file, &probes_seq_op);
1156}
1157
1158static ssize_t probes_write(struct file *file, const char __user *buffer,
1159 size_t count, loff_t *ppos)
1160{
1161 return trace_parse_run_command(file, buffer, count, ppos,
1162 create_or_delete_trace_kprobe);
1163}
1164
1165static const struct file_operations kprobe_events_ops = {
1166 .owner = THIS_MODULE,
1167 .open = probes_open,
1168 .read = seq_read,
1169 .llseek = seq_lseek,
1170 .release = seq_release,
1171 .write = probes_write,
1172};
1173
1174/* Probes profiling interfaces */
1175static int probes_profile_seq_show(struct seq_file *m, void *v)
1176{
1177 struct dyn_event *ev = v;
1178 struct trace_kprobe *tk;
1179 unsigned long nmissed;
1180
1181 if (!is_trace_kprobe(ev))
1182 return 0;
1183
1184 tk = to_trace_kprobe(ev);
1185 nmissed = trace_kprobe_is_return(tk) ?
1186 tk->rp.kp.nmissed + tk->rp.nmissed : tk->rp.kp.nmissed;
1187 seq_printf(m, " %-44s %15lu %15lu\n",
1188 trace_probe_name(&tk->tp),
1189 trace_kprobe_nhit(tk),
1190 nmissed);
1191
1192 return 0;
1193}
1194
1195static const struct seq_operations profile_seq_op = {
1196 .start = dyn_event_seq_start,
1197 .next = dyn_event_seq_next,
1198 .stop = dyn_event_seq_stop,
1199 .show = probes_profile_seq_show
1200};
1201
1202static int profile_open(struct inode *inode, struct file *file)
1203{
1204 int ret;
1205
1206 ret = security_locked_down(LOCKDOWN_TRACEFS);
1207 if (ret)
1208 return ret;
1209
1210 return seq_open(file, &profile_seq_op);
1211}
1212
1213static const struct file_operations kprobe_profile_ops = {
1214 .owner = THIS_MODULE,
1215 .open = profile_open,
1216 .read = seq_read,
1217 .llseek = seq_lseek,
1218 .release = seq_release,
1219};
1220
1221/* Kprobe specific fetch functions */
1222
1223/* Return the length of string -- including null terminal byte */
1224static nokprobe_inline int
1225fetch_store_strlen_user(unsigned long addr)
1226{
1227 return kern_fetch_store_strlen_user(addr);
1228}
1229
1230/* Return the length of string -- including null terminal byte */
1231static nokprobe_inline int
1232fetch_store_strlen(unsigned long addr)
1233{
1234 return kern_fetch_store_strlen(addr);
1235}
1236
1237/*
1238 * Fetch a null-terminated string from user. Caller MUST set *(u32 *)buf
1239 * with max length and relative data location.
1240 */
1241static nokprobe_inline int
1242fetch_store_string_user(unsigned long addr, void *dest, void *base)
1243{
1244 return kern_fetch_store_string_user(addr, dest, base);
1245}
1246
1247/*
1248 * Fetch a null-terminated string. Caller MUST set *(u32 *)buf with max
1249 * length and relative data location.
1250 */
1251static nokprobe_inline int
1252fetch_store_string(unsigned long addr, void *dest, void *base)
1253{
1254 return kern_fetch_store_string(addr, dest, base);
1255}
1256
1257static nokprobe_inline int
1258probe_mem_read_user(void *dest, void *src, size_t size)
1259{
1260 const void __user *uaddr = (__force const void __user *)src;
1261
1262 return copy_from_user_nofault(dest, uaddr, size);
1263}
1264
1265static nokprobe_inline int
1266probe_mem_read(void *dest, void *src, size_t size)
1267{
1268#ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
1269 if ((unsigned long)src < TASK_SIZE)
1270 return probe_mem_read_user(dest, src, size);
1271#endif
1272 return copy_from_kernel_nofault(dest, src, size);
1273}
1274
1275/* Note that we don't verify it, since the code does not come from user space */
1276static int
1277process_fetch_insn(struct fetch_insn *code, void *rec, void *dest,
1278 void *base)
1279{
1280 struct pt_regs *regs = rec;
1281 unsigned long val;
1282
1283retry:
1284 /* 1st stage: get value from context */
1285 switch (code->op) {
1286 case FETCH_OP_REG:
1287 val = regs_get_register(regs, code->param);
1288 break;
1289 case FETCH_OP_STACK:
1290 val = regs_get_kernel_stack_nth(regs, code->param);
1291 break;
1292 case FETCH_OP_STACKP:
1293 val = kernel_stack_pointer(regs);
1294 break;
1295 case FETCH_OP_RETVAL:
1296 val = regs_return_value(regs);
1297 break;
1298 case FETCH_OP_IMM:
1299 val = code->immediate;
1300 break;
1301 case FETCH_OP_COMM:
1302 val = (unsigned long)current->comm;
1303 break;
1304 case FETCH_OP_DATA:
1305 val = (unsigned long)code->data;
1306 break;
1307#ifdef CONFIG_HAVE_FUNCTION_ARG_ACCESS_API
1308 case FETCH_OP_ARG:
1309 val = regs_get_kernel_argument(regs, code->param);
1310 break;
1311#endif
1312 case FETCH_NOP_SYMBOL: /* Ignore a place holder */
1313 code++;
1314 goto retry;
1315 default:
1316 return -EILSEQ;
1317 }
1318 code++;
1319
1320 return process_fetch_insn_bottom(code, val, dest, base);
1321}
1322NOKPROBE_SYMBOL(process_fetch_insn)
1323
1324/* Kprobe handler */
1325static nokprobe_inline void
1326__kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs,
1327 struct trace_event_file *trace_file)
1328{
1329 struct kprobe_trace_entry_head *entry;
1330 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1331 struct trace_event_buffer fbuffer;
1332 int dsize;
1333
1334 WARN_ON(call != trace_file->event_call);
1335
1336 if (trace_trigger_soft_disabled(trace_file))
1337 return;
1338
1339 dsize = __get_data_size(&tk->tp, regs);
1340
1341 entry = trace_event_buffer_reserve(&fbuffer, trace_file,
1342 sizeof(*entry) + tk->tp.size + dsize);
1343 if (!entry)
1344 return;
1345
1346 fbuffer.regs = regs;
1347 entry->ip = (unsigned long)tk->rp.kp.addr;
1348 store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1349
1350 trace_event_buffer_commit(&fbuffer);
1351}
1352
1353static void
1354kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs)
1355{
1356 struct event_file_link *link;
1357
1358 trace_probe_for_each_link_rcu(link, &tk->tp)
1359 __kprobe_trace_func(tk, regs, link->file);
1360}
1361NOKPROBE_SYMBOL(kprobe_trace_func);
1362
1363/* Kretprobe handler */
1364static nokprobe_inline void
1365__kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1366 struct pt_regs *regs,
1367 struct trace_event_file *trace_file)
1368{
1369 struct kretprobe_trace_entry_head *entry;
1370 struct trace_event_buffer fbuffer;
1371 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1372 int dsize;
1373
1374 WARN_ON(call != trace_file->event_call);
1375
1376 if (trace_trigger_soft_disabled(trace_file))
1377 return;
1378
1379 dsize = __get_data_size(&tk->tp, regs);
1380
1381 entry = trace_event_buffer_reserve(&fbuffer, trace_file,
1382 sizeof(*entry) + tk->tp.size + dsize);
1383 if (!entry)
1384 return;
1385
1386 fbuffer.regs = regs;
1387 entry->func = (unsigned long)tk->rp.kp.addr;
1388 entry->ret_ip = get_kretprobe_retaddr(ri);
1389 store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1390
1391 trace_event_buffer_commit(&fbuffer);
1392}
1393
1394static void
1395kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1396 struct pt_regs *regs)
1397{
1398 struct event_file_link *link;
1399
1400 trace_probe_for_each_link_rcu(link, &tk->tp)
1401 __kretprobe_trace_func(tk, ri, regs, link->file);
1402}
1403NOKPROBE_SYMBOL(kretprobe_trace_func);
1404
1405/* Event entry printers */
1406static enum print_line_t
1407print_kprobe_event(struct trace_iterator *iter, int flags,
1408 struct trace_event *event)
1409{
1410 struct kprobe_trace_entry_head *field;
1411 struct trace_seq *s = &iter->seq;
1412 struct trace_probe *tp;
1413
1414 field = (struct kprobe_trace_entry_head *)iter->ent;
1415 tp = trace_probe_primary_from_call(
1416 container_of(event, struct trace_event_call, event));
1417 if (WARN_ON_ONCE(!tp))
1418 goto out;
1419
1420 trace_seq_printf(s, "%s: (", trace_probe_name(tp));
1421
1422 if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
1423 goto out;
1424
1425 trace_seq_putc(s, ')');
1426
1427 if (print_probe_args(s, tp->args, tp->nr_args,
1428 (u8 *)&field[1], field) < 0)
1429 goto out;
1430
1431 trace_seq_putc(s, '\n');
1432 out:
1433 return trace_handle_return(s);
1434}
1435
1436static enum print_line_t
1437print_kretprobe_event(struct trace_iterator *iter, int flags,
1438 struct trace_event *event)
1439{
1440 struct kretprobe_trace_entry_head *field;
1441 struct trace_seq *s = &iter->seq;
1442 struct trace_probe *tp;
1443
1444 field = (struct kretprobe_trace_entry_head *)iter->ent;
1445 tp = trace_probe_primary_from_call(
1446 container_of(event, struct trace_event_call, event));
1447 if (WARN_ON_ONCE(!tp))
1448 goto out;
1449
1450 trace_seq_printf(s, "%s: (", trace_probe_name(tp));
1451
1452 if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1453 goto out;
1454
1455 trace_seq_puts(s, " <- ");
1456
1457 if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1458 goto out;
1459
1460 trace_seq_putc(s, ')');
1461
1462 if (print_probe_args(s, tp->args, tp->nr_args,
1463 (u8 *)&field[1], field) < 0)
1464 goto out;
1465
1466 trace_seq_putc(s, '\n');
1467
1468 out:
1469 return trace_handle_return(s);
1470}
1471
1472
1473static int kprobe_event_define_fields(struct trace_event_call *event_call)
1474{
1475 int ret;
1476 struct kprobe_trace_entry_head field;
1477 struct trace_probe *tp;
1478
1479 tp = trace_probe_primary_from_call(event_call);
1480 if (WARN_ON_ONCE(!tp))
1481 return -ENOENT;
1482
1483 DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1484
1485 return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
1486}
1487
1488static int kretprobe_event_define_fields(struct trace_event_call *event_call)
1489{
1490 int ret;
1491 struct kretprobe_trace_entry_head field;
1492 struct trace_probe *tp;
1493
1494 tp = trace_probe_primary_from_call(event_call);
1495 if (WARN_ON_ONCE(!tp))
1496 return -ENOENT;
1497
1498 DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1499 DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1500
1501 return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
1502}
1503
1504#ifdef CONFIG_PERF_EVENTS
1505
1506/* Kprobe profile handler */
1507static int
1508kprobe_perf_func(struct trace_kprobe *tk, struct pt_regs *regs)
1509{
1510 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1511 struct kprobe_trace_entry_head *entry;
1512 struct hlist_head *head;
1513 int size, __size, dsize;
1514 int rctx;
1515
1516 if (bpf_prog_array_valid(call)) {
1517 unsigned long orig_ip = instruction_pointer(regs);
1518 int ret;
1519
1520 ret = trace_call_bpf(call, regs);
1521
1522 /*
1523 * We need to check and see if we modified the pc of the
1524 * pt_regs, and if so return 1 so that we don't do the
1525 * single stepping.
1526 */
1527 if (orig_ip != instruction_pointer(regs))
1528 return 1;
1529 if (!ret)
1530 return 0;
1531 }
1532
1533 head = this_cpu_ptr(call->perf_events);
1534 if (hlist_empty(head))
1535 return 0;
1536
1537 dsize = __get_data_size(&tk->tp, regs);
1538 __size = sizeof(*entry) + tk->tp.size + dsize;
1539 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1540 size -= sizeof(u32);
1541
1542 entry = perf_trace_buf_alloc(size, NULL, &rctx);
1543 if (!entry)
1544 return 0;
1545
1546 entry->ip = (unsigned long)tk->rp.kp.addr;
1547 memset(&entry[1], 0, dsize);
1548 store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1549 perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1550 head, NULL);
1551 return 0;
1552}
1553NOKPROBE_SYMBOL(kprobe_perf_func);
1554
1555/* Kretprobe profile handler */
1556static void
1557kretprobe_perf_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1558 struct pt_regs *regs)
1559{
1560 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1561 struct kretprobe_trace_entry_head *entry;
1562 struct hlist_head *head;
1563 int size, __size, dsize;
1564 int rctx;
1565
1566 if (bpf_prog_array_valid(call) && !trace_call_bpf(call, regs))
1567 return;
1568
1569 head = this_cpu_ptr(call->perf_events);
1570 if (hlist_empty(head))
1571 return;
1572
1573 dsize = __get_data_size(&tk->tp, regs);
1574 __size = sizeof(*entry) + tk->tp.size + dsize;
1575 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1576 size -= sizeof(u32);
1577
1578 entry = perf_trace_buf_alloc(size, NULL, &rctx);
1579 if (!entry)
1580 return;
1581
1582 entry->func = (unsigned long)tk->rp.kp.addr;
1583 entry->ret_ip = get_kretprobe_retaddr(ri);
1584 store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1585 perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1586 head, NULL);
1587}
1588NOKPROBE_SYMBOL(kretprobe_perf_func);
1589
1590int bpf_get_kprobe_info(const struct perf_event *event, u32 *fd_type,
1591 const char **symbol, u64 *probe_offset,
1592 u64 *probe_addr, bool perf_type_tracepoint)
1593{
1594 const char *pevent = trace_event_name(event->tp_event);
1595 const char *group = event->tp_event->class->system;
1596 struct trace_kprobe *tk;
1597
1598 if (perf_type_tracepoint)
1599 tk = find_trace_kprobe(pevent, group);
1600 else
1601 tk = trace_kprobe_primary_from_call(event->tp_event);
1602 if (!tk)
1603 return -EINVAL;
1604
1605 *fd_type = trace_kprobe_is_return(tk) ? BPF_FD_TYPE_KRETPROBE
1606 : BPF_FD_TYPE_KPROBE;
1607 if (tk->symbol) {
1608 *symbol = tk->symbol;
1609 *probe_offset = tk->rp.kp.offset;
1610 *probe_addr = 0;
1611 } else {
1612 *symbol = NULL;
1613 *probe_offset = 0;
1614 *probe_addr = (unsigned long)tk->rp.kp.addr;
1615 }
1616 return 0;
1617}
1618#endif /* CONFIG_PERF_EVENTS */
1619
1620/*
1621 * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex.
1622 *
1623 * kprobe_trace_self_tests_init() does enable_trace_probe/disable_trace_probe
1624 * lockless, but we can't race with this __init function.
1625 */
1626static int kprobe_register(struct trace_event_call *event,
1627 enum trace_reg type, void *data)
1628{
1629 struct trace_event_file *file = data;
1630
1631 switch (type) {
1632 case TRACE_REG_REGISTER:
1633 return enable_trace_kprobe(event, file);
1634 case TRACE_REG_UNREGISTER:
1635 return disable_trace_kprobe(event, file);
1636
1637#ifdef CONFIG_PERF_EVENTS
1638 case TRACE_REG_PERF_REGISTER:
1639 return enable_trace_kprobe(event, NULL);
1640 case TRACE_REG_PERF_UNREGISTER:
1641 return disable_trace_kprobe(event, NULL);
1642 case TRACE_REG_PERF_OPEN:
1643 case TRACE_REG_PERF_CLOSE:
1644 case TRACE_REG_PERF_ADD:
1645 case TRACE_REG_PERF_DEL:
1646 return 0;
1647#endif
1648 }
1649 return 0;
1650}
1651
1652static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1653{
1654 struct trace_kprobe *tk = container_of(kp, struct trace_kprobe, rp.kp);
1655 int ret = 0;
1656
1657 raw_cpu_inc(*tk->nhit);
1658
1659 if (trace_probe_test_flag(&tk->tp, TP_FLAG_TRACE))
1660 kprobe_trace_func(tk, regs);
1661#ifdef CONFIG_PERF_EVENTS
1662 if (trace_probe_test_flag(&tk->tp, TP_FLAG_PROFILE))
1663 ret = kprobe_perf_func(tk, regs);
1664#endif
1665 return ret;
1666}
1667NOKPROBE_SYMBOL(kprobe_dispatcher);
1668
1669static int
1670kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1671{
1672 struct kretprobe *rp = get_kretprobe(ri);
1673 struct trace_kprobe *tk;
1674
1675 /*
1676 * There is a small chance that get_kretprobe(ri) returns NULL when
1677 * the kretprobe is unregister on another CPU between kretprobe's
1678 * trampoline_handler and this function.
1679 */
1680 if (unlikely(!rp))
1681 return 0;
1682
1683 tk = container_of(rp, struct trace_kprobe, rp);
1684 raw_cpu_inc(*tk->nhit);
1685
1686 if (trace_probe_test_flag(&tk->tp, TP_FLAG_TRACE))
1687 kretprobe_trace_func(tk, ri, regs);
1688#ifdef CONFIG_PERF_EVENTS
1689 if (trace_probe_test_flag(&tk->tp, TP_FLAG_PROFILE))
1690 kretprobe_perf_func(tk, ri, regs);
1691#endif
1692 return 0; /* We don't tweak kernel, so just return 0 */
1693}
1694NOKPROBE_SYMBOL(kretprobe_dispatcher);
1695
1696static struct trace_event_functions kretprobe_funcs = {
1697 .trace = print_kretprobe_event
1698};
1699
1700static struct trace_event_functions kprobe_funcs = {
1701 .trace = print_kprobe_event
1702};
1703
1704static struct trace_event_fields kretprobe_fields_array[] = {
1705 { .type = TRACE_FUNCTION_TYPE,
1706 .define_fields = kretprobe_event_define_fields },
1707 {}
1708};
1709
1710static struct trace_event_fields kprobe_fields_array[] = {
1711 { .type = TRACE_FUNCTION_TYPE,
1712 .define_fields = kprobe_event_define_fields },
1713 {}
1714};
1715
1716static inline void init_trace_event_call(struct trace_kprobe *tk)
1717{
1718 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1719
1720 if (trace_kprobe_is_return(tk)) {
1721 call->event.funcs = &kretprobe_funcs;
1722 call->class->fields_array = kretprobe_fields_array;
1723 } else {
1724 call->event.funcs = &kprobe_funcs;
1725 call->class->fields_array = kprobe_fields_array;
1726 }
1727
1728 call->flags = TRACE_EVENT_FL_KPROBE;
1729 call->class->reg = kprobe_register;
1730}
1731
1732static int register_kprobe_event(struct trace_kprobe *tk)
1733{
1734 init_trace_event_call(tk);
1735
1736 return trace_probe_register_event_call(&tk->tp);
1737}
1738
1739static int unregister_kprobe_event(struct trace_kprobe *tk)
1740{
1741 return trace_probe_unregister_event_call(&tk->tp);
1742}
1743
1744#ifdef CONFIG_PERF_EVENTS
1745/* create a trace_kprobe, but don't add it to global lists */
1746struct trace_event_call *
1747create_local_trace_kprobe(char *func, void *addr, unsigned long offs,
1748 bool is_return)
1749{
1750 enum probe_print_type ptype;
1751 struct trace_kprobe *tk;
1752 int ret;
1753 char *event;
1754
1755 /*
1756 * local trace_kprobes are not added to dyn_event, so they are never
1757 * searched in find_trace_kprobe(). Therefore, there is no concern of
1758 * duplicated name here.
1759 */
1760 event = func ? func : "DUMMY_EVENT";
1761
1762 tk = alloc_trace_kprobe(KPROBE_EVENT_SYSTEM, event, (void *)addr, func,
1763 offs, 0 /* maxactive */, 0 /* nargs */,
1764 is_return);
1765
1766 if (IS_ERR(tk)) {
1767 pr_info("Failed to allocate trace_probe.(%d)\n",
1768 (int)PTR_ERR(tk));
1769 return ERR_CAST(tk);
1770 }
1771
1772 init_trace_event_call(tk);
1773
1774 ptype = trace_kprobe_is_return(tk) ?
1775 PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL;
1776 if (traceprobe_set_print_fmt(&tk->tp, ptype) < 0) {
1777 ret = -ENOMEM;
1778 goto error;
1779 }
1780
1781 ret = __register_trace_kprobe(tk);
1782 if (ret < 0)
1783 goto error;
1784
1785 return trace_probe_event_call(&tk->tp);
1786error:
1787 free_trace_kprobe(tk);
1788 return ERR_PTR(ret);
1789}
1790
1791void destroy_local_trace_kprobe(struct trace_event_call *event_call)
1792{
1793 struct trace_kprobe *tk;
1794
1795 tk = trace_kprobe_primary_from_call(event_call);
1796 if (unlikely(!tk))
1797 return;
1798
1799 if (trace_probe_is_enabled(&tk->tp)) {
1800 WARN_ON(1);
1801 return;
1802 }
1803
1804 __unregister_trace_kprobe(tk);
1805
1806 free_trace_kprobe(tk);
1807}
1808#endif /* CONFIG_PERF_EVENTS */
1809
1810static __init void enable_boot_kprobe_events(void)
1811{
1812 struct trace_array *tr = top_trace_array();
1813 struct trace_event_file *file;
1814 struct trace_kprobe *tk;
1815 struct dyn_event *pos;
1816
1817 mutex_lock(&event_mutex);
1818 for_each_trace_kprobe(tk, pos) {
1819 list_for_each_entry(file, &tr->events, list)
1820 if (file->event_call == trace_probe_event_call(&tk->tp))
1821 trace_event_enable_disable(file, 1, 0);
1822 }
1823 mutex_unlock(&event_mutex);
1824}
1825
1826static __init void setup_boot_kprobe_events(void)
1827{
1828 char *p, *cmd = kprobe_boot_events_buf;
1829 int ret;
1830
1831 strreplace(kprobe_boot_events_buf, ',', ' ');
1832
1833 while (cmd && *cmd != '\0') {
1834 p = strchr(cmd, ';');
1835 if (p)
1836 *p++ = '\0';
1837
1838 ret = create_or_delete_trace_kprobe(cmd);
1839 if (ret)
1840 pr_warn("Failed to add event(%d): %s\n", ret, cmd);
1841
1842 cmd = p;
1843 }
1844
1845 enable_boot_kprobe_events();
1846}
1847
1848/*
1849 * Register dynevent at core_initcall. This allows kernel to setup kprobe
1850 * events in postcore_initcall without tracefs.
1851 */
1852static __init int init_kprobe_trace_early(void)
1853{
1854 int ret;
1855
1856 ret = dyn_event_register(&trace_kprobe_ops);
1857 if (ret)
1858 return ret;
1859
1860 if (register_module_notifier(&trace_kprobe_module_nb))
1861 return -EINVAL;
1862
1863 return 0;
1864}
1865core_initcall(init_kprobe_trace_early);
1866
1867/* Make a tracefs interface for controlling probe points */
1868static __init int init_kprobe_trace(void)
1869{
1870 int ret;
1871
1872 ret = tracing_init_dentry();
1873 if (ret)
1874 return 0;
1875
1876 /* Event list interface */
1877 trace_create_file("kprobe_events", TRACE_MODE_WRITE,
1878 NULL, NULL, &kprobe_events_ops);
1879
1880 /* Profile interface */
1881 trace_create_file("kprobe_profile", TRACE_MODE_READ,
1882 NULL, NULL, &kprobe_profile_ops);
1883
1884 setup_boot_kprobe_events();
1885
1886 return 0;
1887}
1888fs_initcall(init_kprobe_trace);
1889
1890
1891#ifdef CONFIG_FTRACE_STARTUP_TEST
1892static __init struct trace_event_file *
1893find_trace_probe_file(struct trace_kprobe *tk, struct trace_array *tr)
1894{
1895 struct trace_event_file *file;
1896
1897 list_for_each_entry(file, &tr->events, list)
1898 if (file->event_call == trace_probe_event_call(&tk->tp))
1899 return file;
1900
1901 return NULL;
1902}
1903
1904/*
1905 * Nobody but us can call enable_trace_kprobe/disable_trace_kprobe at this
1906 * stage, we can do this lockless.
1907 */
1908static __init int kprobe_trace_self_tests_init(void)
1909{
1910 int ret, warn = 0;
1911 int (*target)(int, int, int, int, int, int);
1912 struct trace_kprobe *tk;
1913 struct trace_event_file *file;
1914
1915 if (tracing_is_disabled())
1916 return -ENODEV;
1917
1918 if (tracing_selftest_disabled)
1919 return 0;
1920
1921 target = kprobe_trace_selftest_target;
1922
1923 pr_info("Testing kprobe tracing: ");
1924
1925 ret = create_or_delete_trace_kprobe("p:testprobe kprobe_trace_selftest_target $stack $stack0 +0($stack)");
1926 if (WARN_ON_ONCE(ret)) {
1927 pr_warn("error on probing function entry.\n");
1928 warn++;
1929 } else {
1930 /* Enable trace point */
1931 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1932 if (WARN_ON_ONCE(tk == NULL)) {
1933 pr_warn("error on getting new probe.\n");
1934 warn++;
1935 } else {
1936 file = find_trace_probe_file(tk, top_trace_array());
1937 if (WARN_ON_ONCE(file == NULL)) {
1938 pr_warn("error on getting probe file.\n");
1939 warn++;
1940 } else
1941 enable_trace_kprobe(
1942 trace_probe_event_call(&tk->tp), file);
1943 }
1944 }
1945
1946 ret = create_or_delete_trace_kprobe("r:testprobe2 kprobe_trace_selftest_target $retval");
1947 if (WARN_ON_ONCE(ret)) {
1948 pr_warn("error on probing function return.\n");
1949 warn++;
1950 } else {
1951 /* Enable trace point */
1952 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1953 if (WARN_ON_ONCE(tk == NULL)) {
1954 pr_warn("error on getting 2nd new probe.\n");
1955 warn++;
1956 } else {
1957 file = find_trace_probe_file(tk, top_trace_array());
1958 if (WARN_ON_ONCE(file == NULL)) {
1959 pr_warn("error on getting probe file.\n");
1960 warn++;
1961 } else
1962 enable_trace_kprobe(
1963 trace_probe_event_call(&tk->tp), file);
1964 }
1965 }
1966
1967 if (warn)
1968 goto end;
1969
1970 ret = target(1, 2, 3, 4, 5, 6);
1971
1972 /*
1973 * Not expecting an error here, the check is only to prevent the
1974 * optimizer from removing the call to target() as otherwise there
1975 * are no side-effects and the call is never performed.
1976 */
1977 if (ret != 21)
1978 warn++;
1979
1980 /* Disable trace points before removing it */
1981 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1982 if (WARN_ON_ONCE(tk == NULL)) {
1983 pr_warn("error on getting test probe.\n");
1984 warn++;
1985 } else {
1986 if (trace_kprobe_nhit(tk) != 1) {
1987 pr_warn("incorrect number of testprobe hits\n");
1988 warn++;
1989 }
1990
1991 file = find_trace_probe_file(tk, top_trace_array());
1992 if (WARN_ON_ONCE(file == NULL)) {
1993 pr_warn("error on getting probe file.\n");
1994 warn++;
1995 } else
1996 disable_trace_kprobe(
1997 trace_probe_event_call(&tk->tp), file);
1998 }
1999
2000 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
2001 if (WARN_ON_ONCE(tk == NULL)) {
2002 pr_warn("error on getting 2nd test probe.\n");
2003 warn++;
2004 } else {
2005 if (trace_kprobe_nhit(tk) != 1) {
2006 pr_warn("incorrect number of testprobe2 hits\n");
2007 warn++;
2008 }
2009
2010 file = find_trace_probe_file(tk, top_trace_array());
2011 if (WARN_ON_ONCE(file == NULL)) {
2012 pr_warn("error on getting probe file.\n");
2013 warn++;
2014 } else
2015 disable_trace_kprobe(
2016 trace_probe_event_call(&tk->tp), file);
2017 }
2018
2019 ret = create_or_delete_trace_kprobe("-:testprobe");
2020 if (WARN_ON_ONCE(ret)) {
2021 pr_warn("error on deleting a probe.\n");
2022 warn++;
2023 }
2024
2025 ret = create_or_delete_trace_kprobe("-:testprobe2");
2026 if (WARN_ON_ONCE(ret)) {
2027 pr_warn("error on deleting a probe.\n");
2028 warn++;
2029 }
2030
2031end:
2032 ret = dyn_events_release_all(&trace_kprobe_ops);
2033 if (WARN_ON_ONCE(ret)) {
2034 pr_warn("error on cleaning up probes.\n");
2035 warn++;
2036 }
2037 /*
2038 * Wait for the optimizer work to finish. Otherwise it might fiddle
2039 * with probes in already freed __init text.
2040 */
2041 wait_for_kprobe_optimizer();
2042 if (warn)
2043 pr_cont("NG: Some tests are failed. Please check them.\n");
2044 else
2045 pr_cont("OK\n");
2046 return 0;
2047}
2048
2049late_initcall(kprobe_trace_self_tests_init);
2050
2051#endif