Loading...
1/*
2 * Kprobes-based tracing events
3 *
4 * Created by Masami Hiramatsu <mhiramat@redhat.com>
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 as
8 * published by the Free Software Foundation.
9 *
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
14 *
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write to the Free Software
17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
18 */
19
20#include <linux/module.h>
21#include <linux/uaccess.h>
22
23#include "trace_probe.h"
24
25#define KPROBE_EVENT_SYSTEM "kprobes"
26
27/**
28 * Kprobe event core functions
29 */
30struct trace_kprobe {
31 struct list_head list;
32 struct kretprobe rp; /* Use rp.kp for kprobe use */
33 unsigned long nhit;
34 const char *symbol; /* symbol name */
35 struct trace_probe tp;
36};
37
38#define SIZEOF_TRACE_KPROBE(n) \
39 (offsetof(struct trace_kprobe, tp.args) + \
40 (sizeof(struct probe_arg) * (n)))
41
42
43static __kprobes bool trace_kprobe_is_return(struct trace_kprobe *tk)
44{
45 return tk->rp.handler != NULL;
46}
47
48static __kprobes const char *trace_kprobe_symbol(struct trace_kprobe *tk)
49{
50 return tk->symbol ? tk->symbol : "unknown";
51}
52
53static __kprobes unsigned long trace_kprobe_offset(struct trace_kprobe *tk)
54{
55 return tk->rp.kp.offset;
56}
57
58static __kprobes bool trace_kprobe_has_gone(struct trace_kprobe *tk)
59{
60 return !!(kprobe_gone(&tk->rp.kp));
61}
62
63static __kprobes bool trace_kprobe_within_module(struct trace_kprobe *tk,
64 struct module *mod)
65{
66 int len = strlen(mod->name);
67 const char *name = trace_kprobe_symbol(tk);
68 return strncmp(mod->name, name, len) == 0 && name[len] == ':';
69}
70
71static __kprobes bool trace_kprobe_is_on_module(struct trace_kprobe *tk)
72{
73 return !!strchr(trace_kprobe_symbol(tk), ':');
74}
75
76static int register_kprobe_event(struct trace_kprobe *tk);
77static int unregister_kprobe_event(struct trace_kprobe *tk);
78
79static DEFINE_MUTEX(probe_lock);
80static LIST_HEAD(probe_list);
81
82static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
83static int kretprobe_dispatcher(struct kretprobe_instance *ri,
84 struct pt_regs *regs);
85
86/* Memory fetching by symbol */
87struct symbol_cache {
88 char *symbol;
89 long offset;
90 unsigned long addr;
91};
92
93unsigned long update_symbol_cache(struct symbol_cache *sc)
94{
95 sc->addr = (unsigned long)kallsyms_lookup_name(sc->symbol);
96
97 if (sc->addr)
98 sc->addr += sc->offset;
99
100 return sc->addr;
101}
102
103void free_symbol_cache(struct symbol_cache *sc)
104{
105 kfree(sc->symbol);
106 kfree(sc);
107}
108
109struct symbol_cache *alloc_symbol_cache(const char *sym, long offset)
110{
111 struct symbol_cache *sc;
112
113 if (!sym || strlen(sym) == 0)
114 return NULL;
115
116 sc = kzalloc(sizeof(struct symbol_cache), GFP_KERNEL);
117 if (!sc)
118 return NULL;
119
120 sc->symbol = kstrdup(sym, GFP_KERNEL);
121 if (!sc->symbol) {
122 kfree(sc);
123 return NULL;
124 }
125 sc->offset = offset;
126 update_symbol_cache(sc);
127
128 return sc;
129}
130
131/*
132 * Kprobes-specific fetch functions
133 */
134#define DEFINE_FETCH_stack(type) \
135static __kprobes void FETCH_FUNC_NAME(stack, type)(struct pt_regs *regs,\
136 void *offset, void *dest) \
137{ \
138 *(type *)dest = (type)regs_get_kernel_stack_nth(regs, \
139 (unsigned int)((unsigned long)offset)); \
140}
141DEFINE_BASIC_FETCH_FUNCS(stack)
142/* No string on the stack entry */
143#define fetch_stack_string NULL
144#define fetch_stack_string_size NULL
145
146#define DEFINE_FETCH_memory(type) \
147static __kprobes void FETCH_FUNC_NAME(memory, type)(struct pt_regs *regs,\
148 void *addr, void *dest) \
149{ \
150 type retval; \
151 if (probe_kernel_address(addr, retval)) \
152 *(type *)dest = 0; \
153 else \
154 *(type *)dest = retval; \
155}
156DEFINE_BASIC_FETCH_FUNCS(memory)
157/*
158 * Fetch a null-terminated string. Caller MUST set *(u32 *)dest with max
159 * length and relative data location.
160 */
161static __kprobes void FETCH_FUNC_NAME(memory, string)(struct pt_regs *regs,
162 void *addr, void *dest)
163{
164 long ret;
165 int maxlen = get_rloc_len(*(u32 *)dest);
166 u8 *dst = get_rloc_data(dest);
167 u8 *src = addr;
168 mm_segment_t old_fs = get_fs();
169
170 if (!maxlen)
171 return;
172
173 /*
174 * Try to get string again, since the string can be changed while
175 * probing.
176 */
177 set_fs(KERNEL_DS);
178 pagefault_disable();
179
180 do
181 ret = __copy_from_user_inatomic(dst++, src++, 1);
182 while (dst[-1] && ret == 0 && src - (u8 *)addr < maxlen);
183
184 dst[-1] = '\0';
185 pagefault_enable();
186 set_fs(old_fs);
187
188 if (ret < 0) { /* Failed to fetch string */
189 ((u8 *)get_rloc_data(dest))[0] = '\0';
190 *(u32 *)dest = make_data_rloc(0, get_rloc_offs(*(u32 *)dest));
191 } else {
192 *(u32 *)dest = make_data_rloc(src - (u8 *)addr,
193 get_rloc_offs(*(u32 *)dest));
194 }
195}
196
197/* Return the length of string -- including null terminal byte */
198static __kprobes void FETCH_FUNC_NAME(memory, string_size)(struct pt_regs *regs,
199 void *addr, void *dest)
200{
201 mm_segment_t old_fs;
202 int ret, len = 0;
203 u8 c;
204
205 old_fs = get_fs();
206 set_fs(KERNEL_DS);
207 pagefault_disable();
208
209 do {
210 ret = __copy_from_user_inatomic(&c, (u8 *)addr + len, 1);
211 len++;
212 } while (c && ret == 0 && len < MAX_STRING_SIZE);
213
214 pagefault_enable();
215 set_fs(old_fs);
216
217 if (ret < 0) /* Failed to check the length */
218 *(u32 *)dest = 0;
219 else
220 *(u32 *)dest = len;
221}
222
223#define DEFINE_FETCH_symbol(type) \
224__kprobes void FETCH_FUNC_NAME(symbol, type)(struct pt_regs *regs, \
225 void *data, void *dest) \
226{ \
227 struct symbol_cache *sc = data; \
228 if (sc->addr) \
229 fetch_memory_##type(regs, (void *)sc->addr, dest); \
230 else \
231 *(type *)dest = 0; \
232}
233DEFINE_BASIC_FETCH_FUNCS(symbol)
234DEFINE_FETCH_symbol(string)
235DEFINE_FETCH_symbol(string_size)
236
237/* kprobes don't support file_offset fetch methods */
238#define fetch_file_offset_u8 NULL
239#define fetch_file_offset_u16 NULL
240#define fetch_file_offset_u32 NULL
241#define fetch_file_offset_u64 NULL
242#define fetch_file_offset_string NULL
243#define fetch_file_offset_string_size NULL
244
245/* Fetch type information table */
246const struct fetch_type kprobes_fetch_type_table[] = {
247 /* Special types */
248 [FETCH_TYPE_STRING] = __ASSIGN_FETCH_TYPE("string", string, string,
249 sizeof(u32), 1, "__data_loc char[]"),
250 [FETCH_TYPE_STRSIZE] = __ASSIGN_FETCH_TYPE("string_size", u32,
251 string_size, sizeof(u32), 0, "u32"),
252 /* Basic types */
253 ASSIGN_FETCH_TYPE(u8, u8, 0),
254 ASSIGN_FETCH_TYPE(u16, u16, 0),
255 ASSIGN_FETCH_TYPE(u32, u32, 0),
256 ASSIGN_FETCH_TYPE(u64, u64, 0),
257 ASSIGN_FETCH_TYPE(s8, u8, 1),
258 ASSIGN_FETCH_TYPE(s16, u16, 1),
259 ASSIGN_FETCH_TYPE(s32, u32, 1),
260 ASSIGN_FETCH_TYPE(s64, u64, 1),
261
262 ASSIGN_FETCH_TYPE_END
263};
264
265/*
266 * Allocate new trace_probe and initialize it (including kprobes).
267 */
268static struct trace_kprobe *alloc_trace_kprobe(const char *group,
269 const char *event,
270 void *addr,
271 const char *symbol,
272 unsigned long offs,
273 int nargs, bool is_return)
274{
275 struct trace_kprobe *tk;
276 int ret = -ENOMEM;
277
278 tk = kzalloc(SIZEOF_TRACE_KPROBE(nargs), GFP_KERNEL);
279 if (!tk)
280 return ERR_PTR(ret);
281
282 if (symbol) {
283 tk->symbol = kstrdup(symbol, GFP_KERNEL);
284 if (!tk->symbol)
285 goto error;
286 tk->rp.kp.symbol_name = tk->symbol;
287 tk->rp.kp.offset = offs;
288 } else
289 tk->rp.kp.addr = addr;
290
291 if (is_return)
292 tk->rp.handler = kretprobe_dispatcher;
293 else
294 tk->rp.kp.pre_handler = kprobe_dispatcher;
295
296 if (!event || !is_good_name(event)) {
297 ret = -EINVAL;
298 goto error;
299 }
300
301 tk->tp.call.class = &tk->tp.class;
302 tk->tp.call.name = kstrdup(event, GFP_KERNEL);
303 if (!tk->tp.call.name)
304 goto error;
305
306 if (!group || !is_good_name(group)) {
307 ret = -EINVAL;
308 goto error;
309 }
310
311 tk->tp.class.system = kstrdup(group, GFP_KERNEL);
312 if (!tk->tp.class.system)
313 goto error;
314
315 INIT_LIST_HEAD(&tk->list);
316 INIT_LIST_HEAD(&tk->tp.files);
317 return tk;
318error:
319 kfree(tk->tp.call.name);
320 kfree(tk->symbol);
321 kfree(tk);
322 return ERR_PTR(ret);
323}
324
325static void free_trace_kprobe(struct trace_kprobe *tk)
326{
327 int i;
328
329 for (i = 0; i < tk->tp.nr_args; i++)
330 traceprobe_free_probe_arg(&tk->tp.args[i]);
331
332 kfree(tk->tp.call.class->system);
333 kfree(tk->tp.call.name);
334 kfree(tk->symbol);
335 kfree(tk);
336}
337
338static struct trace_kprobe *find_trace_kprobe(const char *event,
339 const char *group)
340{
341 struct trace_kprobe *tk;
342
343 list_for_each_entry(tk, &probe_list, list)
344 if (strcmp(ftrace_event_name(&tk->tp.call), event) == 0 &&
345 strcmp(tk->tp.call.class->system, group) == 0)
346 return tk;
347 return NULL;
348}
349
350/*
351 * Enable trace_probe
352 * if the file is NULL, enable "perf" handler, or enable "trace" handler.
353 */
354static int
355enable_trace_kprobe(struct trace_kprobe *tk, struct ftrace_event_file *file)
356{
357 int ret = 0;
358
359 if (file) {
360 struct event_file_link *link;
361
362 link = kmalloc(sizeof(*link), GFP_KERNEL);
363 if (!link) {
364 ret = -ENOMEM;
365 goto out;
366 }
367
368 link->file = file;
369 list_add_tail_rcu(&link->list, &tk->tp.files);
370
371 tk->tp.flags |= TP_FLAG_TRACE;
372 } else
373 tk->tp.flags |= TP_FLAG_PROFILE;
374
375 if (trace_probe_is_registered(&tk->tp) && !trace_kprobe_has_gone(tk)) {
376 if (trace_kprobe_is_return(tk))
377 ret = enable_kretprobe(&tk->rp);
378 else
379 ret = enable_kprobe(&tk->rp.kp);
380 }
381 out:
382 return ret;
383}
384
385/*
386 * Disable trace_probe
387 * if the file is NULL, disable "perf" handler, or disable "trace" handler.
388 */
389static int
390disable_trace_kprobe(struct trace_kprobe *tk, struct ftrace_event_file *file)
391{
392 struct event_file_link *link = NULL;
393 int wait = 0;
394 int ret = 0;
395
396 if (file) {
397 link = find_event_file_link(&tk->tp, file);
398 if (!link) {
399 ret = -EINVAL;
400 goto out;
401 }
402
403 list_del_rcu(&link->list);
404 wait = 1;
405 if (!list_empty(&tk->tp.files))
406 goto out;
407
408 tk->tp.flags &= ~TP_FLAG_TRACE;
409 } else
410 tk->tp.flags &= ~TP_FLAG_PROFILE;
411
412 if (!trace_probe_is_enabled(&tk->tp) && trace_probe_is_registered(&tk->tp)) {
413 if (trace_kprobe_is_return(tk))
414 disable_kretprobe(&tk->rp);
415 else
416 disable_kprobe(&tk->rp.kp);
417 wait = 1;
418 }
419 out:
420 if (wait) {
421 /*
422 * Synchronize with kprobe_trace_func/kretprobe_trace_func
423 * to ensure disabled (all running handlers are finished).
424 * This is not only for kfree(), but also the caller,
425 * trace_remove_event_call() supposes it for releasing
426 * event_call related objects, which will be accessed in
427 * the kprobe_trace_func/kretprobe_trace_func.
428 */
429 synchronize_sched();
430 kfree(link); /* Ignored if link == NULL */
431 }
432
433 return ret;
434}
435
436/* Internal register function - just handle k*probes and flags */
437static int __register_trace_kprobe(struct trace_kprobe *tk)
438{
439 int i, ret;
440
441 if (trace_probe_is_registered(&tk->tp))
442 return -EINVAL;
443
444 for (i = 0; i < tk->tp.nr_args; i++)
445 traceprobe_update_arg(&tk->tp.args[i]);
446
447 /* Set/clear disabled flag according to tp->flag */
448 if (trace_probe_is_enabled(&tk->tp))
449 tk->rp.kp.flags &= ~KPROBE_FLAG_DISABLED;
450 else
451 tk->rp.kp.flags |= KPROBE_FLAG_DISABLED;
452
453 if (trace_kprobe_is_return(tk))
454 ret = register_kretprobe(&tk->rp);
455 else
456 ret = register_kprobe(&tk->rp.kp);
457
458 if (ret == 0)
459 tk->tp.flags |= TP_FLAG_REGISTERED;
460 else {
461 pr_warning("Could not insert probe at %s+%lu: %d\n",
462 trace_kprobe_symbol(tk), trace_kprobe_offset(tk), ret);
463 if (ret == -ENOENT && trace_kprobe_is_on_module(tk)) {
464 pr_warning("This probe might be able to register after"
465 "target module is loaded. Continue.\n");
466 ret = 0;
467 } else if (ret == -EILSEQ) {
468 pr_warning("Probing address(0x%p) is not an "
469 "instruction boundary.\n",
470 tk->rp.kp.addr);
471 ret = -EINVAL;
472 }
473 }
474
475 return ret;
476}
477
478/* Internal unregister function - just handle k*probes and flags */
479static void __unregister_trace_kprobe(struct trace_kprobe *tk)
480{
481 if (trace_probe_is_registered(&tk->tp)) {
482 if (trace_kprobe_is_return(tk))
483 unregister_kretprobe(&tk->rp);
484 else
485 unregister_kprobe(&tk->rp.kp);
486 tk->tp.flags &= ~TP_FLAG_REGISTERED;
487 /* Cleanup kprobe for reuse */
488 if (tk->rp.kp.symbol_name)
489 tk->rp.kp.addr = NULL;
490 }
491}
492
493/* Unregister a trace_probe and probe_event: call with locking probe_lock */
494static int unregister_trace_kprobe(struct trace_kprobe *tk)
495{
496 /* Enabled event can not be unregistered */
497 if (trace_probe_is_enabled(&tk->tp))
498 return -EBUSY;
499
500 /* Will fail if probe is being used by ftrace or perf */
501 if (unregister_kprobe_event(tk))
502 return -EBUSY;
503
504 __unregister_trace_kprobe(tk);
505 list_del(&tk->list);
506
507 return 0;
508}
509
510/* Register a trace_probe and probe_event */
511static int register_trace_kprobe(struct trace_kprobe *tk)
512{
513 struct trace_kprobe *old_tk;
514 int ret;
515
516 mutex_lock(&probe_lock);
517
518 /* Delete old (same name) event if exist */
519 old_tk = find_trace_kprobe(ftrace_event_name(&tk->tp.call),
520 tk->tp.call.class->system);
521 if (old_tk) {
522 ret = unregister_trace_kprobe(old_tk);
523 if (ret < 0)
524 goto end;
525 free_trace_kprobe(old_tk);
526 }
527
528 /* Register new event */
529 ret = register_kprobe_event(tk);
530 if (ret) {
531 pr_warning("Failed to register probe event(%d)\n", ret);
532 goto end;
533 }
534
535 /* Register k*probe */
536 ret = __register_trace_kprobe(tk);
537 if (ret < 0)
538 unregister_kprobe_event(tk);
539 else
540 list_add_tail(&tk->list, &probe_list);
541
542end:
543 mutex_unlock(&probe_lock);
544 return ret;
545}
546
547/* Module notifier call back, checking event on the module */
548static int trace_kprobe_module_callback(struct notifier_block *nb,
549 unsigned long val, void *data)
550{
551 struct module *mod = data;
552 struct trace_kprobe *tk;
553 int ret;
554
555 if (val != MODULE_STATE_COMING)
556 return NOTIFY_DONE;
557
558 /* Update probes on coming module */
559 mutex_lock(&probe_lock);
560 list_for_each_entry(tk, &probe_list, list) {
561 if (trace_kprobe_within_module(tk, mod)) {
562 /* Don't need to check busy - this should have gone. */
563 __unregister_trace_kprobe(tk);
564 ret = __register_trace_kprobe(tk);
565 if (ret)
566 pr_warning("Failed to re-register probe %s on"
567 "%s: %d\n",
568 ftrace_event_name(&tk->tp.call),
569 mod->name, ret);
570 }
571 }
572 mutex_unlock(&probe_lock);
573
574 return NOTIFY_DONE;
575}
576
577static struct notifier_block trace_kprobe_module_nb = {
578 .notifier_call = trace_kprobe_module_callback,
579 .priority = 1 /* Invoked after kprobe module callback */
580};
581
582static int create_trace_kprobe(int argc, char **argv)
583{
584 /*
585 * Argument syntax:
586 * - Add kprobe: p[:[GRP/]EVENT] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
587 * - Add kretprobe: r[:[GRP/]EVENT] [MOD:]KSYM[+0] [FETCHARGS]
588 * Fetch args:
589 * $retval : fetch return value
590 * $stack : fetch stack address
591 * $stackN : fetch Nth of stack (N:0-)
592 * @ADDR : fetch memory at ADDR (ADDR should be in kernel)
593 * @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
594 * %REG : fetch register REG
595 * Dereferencing memory fetch:
596 * +|-offs(ARG) : fetch memory at ARG +|- offs address.
597 * Alias name of args:
598 * NAME=FETCHARG : set NAME as alias of FETCHARG.
599 * Type of args:
600 * FETCHARG:TYPE : use TYPE instead of unsigned long.
601 */
602 struct trace_kprobe *tk;
603 int i, ret = 0;
604 bool is_return = false, is_delete = false;
605 char *symbol = NULL, *event = NULL, *group = NULL;
606 char *arg;
607 unsigned long offset = 0;
608 void *addr = NULL;
609 char buf[MAX_EVENT_NAME_LEN];
610
611 /* argc must be >= 1 */
612 if (argv[0][0] == 'p')
613 is_return = false;
614 else if (argv[0][0] == 'r')
615 is_return = true;
616 else if (argv[0][0] == '-')
617 is_delete = true;
618 else {
619 pr_info("Probe definition must be started with 'p', 'r' or"
620 " '-'.\n");
621 return -EINVAL;
622 }
623
624 if (argv[0][1] == ':') {
625 event = &argv[0][2];
626 if (strchr(event, '/')) {
627 group = event;
628 event = strchr(group, '/') + 1;
629 event[-1] = '\0';
630 if (strlen(group) == 0) {
631 pr_info("Group name is not specified\n");
632 return -EINVAL;
633 }
634 }
635 if (strlen(event) == 0) {
636 pr_info("Event name is not specified\n");
637 return -EINVAL;
638 }
639 }
640 if (!group)
641 group = KPROBE_EVENT_SYSTEM;
642
643 if (is_delete) {
644 if (!event) {
645 pr_info("Delete command needs an event name.\n");
646 return -EINVAL;
647 }
648 mutex_lock(&probe_lock);
649 tk = find_trace_kprobe(event, group);
650 if (!tk) {
651 mutex_unlock(&probe_lock);
652 pr_info("Event %s/%s doesn't exist.\n", group, event);
653 return -ENOENT;
654 }
655 /* delete an event */
656 ret = unregister_trace_kprobe(tk);
657 if (ret == 0)
658 free_trace_kprobe(tk);
659 mutex_unlock(&probe_lock);
660 return ret;
661 }
662
663 if (argc < 2) {
664 pr_info("Probe point is not specified.\n");
665 return -EINVAL;
666 }
667 if (isdigit(argv[1][0])) {
668 if (is_return) {
669 pr_info("Return probe point must be a symbol.\n");
670 return -EINVAL;
671 }
672 /* an address specified */
673 ret = kstrtoul(&argv[1][0], 0, (unsigned long *)&addr);
674 if (ret) {
675 pr_info("Failed to parse address.\n");
676 return ret;
677 }
678 } else {
679 /* a symbol specified */
680 symbol = argv[1];
681 /* TODO: support .init module functions */
682 ret = traceprobe_split_symbol_offset(symbol, &offset);
683 if (ret) {
684 pr_info("Failed to parse symbol.\n");
685 return ret;
686 }
687 if (offset && is_return) {
688 pr_info("Return probe must be used without offset.\n");
689 return -EINVAL;
690 }
691 }
692 argc -= 2; argv += 2;
693
694 /* setup a probe */
695 if (!event) {
696 /* Make a new event name */
697 if (symbol)
698 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
699 is_return ? 'r' : 'p', symbol, offset);
700 else
701 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
702 is_return ? 'r' : 'p', addr);
703 event = buf;
704 }
705 tk = alloc_trace_kprobe(group, event, addr, symbol, offset, argc,
706 is_return);
707 if (IS_ERR(tk)) {
708 pr_info("Failed to allocate trace_probe.(%d)\n",
709 (int)PTR_ERR(tk));
710 return PTR_ERR(tk);
711 }
712
713 /* parse arguments */
714 ret = 0;
715 for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
716 struct probe_arg *parg = &tk->tp.args[i];
717
718 /* Increment count for freeing args in error case */
719 tk->tp.nr_args++;
720
721 /* Parse argument name */
722 arg = strchr(argv[i], '=');
723 if (arg) {
724 *arg++ = '\0';
725 parg->name = kstrdup(argv[i], GFP_KERNEL);
726 } else {
727 arg = argv[i];
728 /* If argument name is omitted, set "argN" */
729 snprintf(buf, MAX_EVENT_NAME_LEN, "arg%d", i + 1);
730 parg->name = kstrdup(buf, GFP_KERNEL);
731 }
732
733 if (!parg->name) {
734 pr_info("Failed to allocate argument[%d] name.\n", i);
735 ret = -ENOMEM;
736 goto error;
737 }
738
739 if (!is_good_name(parg->name)) {
740 pr_info("Invalid argument[%d] name: %s\n",
741 i, parg->name);
742 ret = -EINVAL;
743 goto error;
744 }
745
746 if (traceprobe_conflict_field_name(parg->name,
747 tk->tp.args, i)) {
748 pr_info("Argument[%d] name '%s' conflicts with "
749 "another field.\n", i, argv[i]);
750 ret = -EINVAL;
751 goto error;
752 }
753
754 /* Parse fetch argument */
755 ret = traceprobe_parse_probe_arg(arg, &tk->tp.size, parg,
756 is_return, true);
757 if (ret) {
758 pr_info("Parse error at argument[%d]. (%d)\n", i, ret);
759 goto error;
760 }
761 }
762
763 ret = register_trace_kprobe(tk);
764 if (ret)
765 goto error;
766 return 0;
767
768error:
769 free_trace_kprobe(tk);
770 return ret;
771}
772
773static int release_all_trace_kprobes(void)
774{
775 struct trace_kprobe *tk;
776 int ret = 0;
777
778 mutex_lock(&probe_lock);
779 /* Ensure no probe is in use. */
780 list_for_each_entry(tk, &probe_list, list)
781 if (trace_probe_is_enabled(&tk->tp)) {
782 ret = -EBUSY;
783 goto end;
784 }
785 /* TODO: Use batch unregistration */
786 while (!list_empty(&probe_list)) {
787 tk = list_entry(probe_list.next, struct trace_kprobe, list);
788 ret = unregister_trace_kprobe(tk);
789 if (ret)
790 goto end;
791 free_trace_kprobe(tk);
792 }
793
794end:
795 mutex_unlock(&probe_lock);
796
797 return ret;
798}
799
800/* Probes listing interfaces */
801static void *probes_seq_start(struct seq_file *m, loff_t *pos)
802{
803 mutex_lock(&probe_lock);
804 return seq_list_start(&probe_list, *pos);
805}
806
807static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
808{
809 return seq_list_next(v, &probe_list, pos);
810}
811
812static void probes_seq_stop(struct seq_file *m, void *v)
813{
814 mutex_unlock(&probe_lock);
815}
816
817static int probes_seq_show(struct seq_file *m, void *v)
818{
819 struct trace_kprobe *tk = v;
820 int i;
821
822 seq_printf(m, "%c", trace_kprobe_is_return(tk) ? 'r' : 'p');
823 seq_printf(m, ":%s/%s", tk->tp.call.class->system,
824 ftrace_event_name(&tk->tp.call));
825
826 if (!tk->symbol)
827 seq_printf(m, " 0x%p", tk->rp.kp.addr);
828 else if (tk->rp.kp.offset)
829 seq_printf(m, " %s+%u", trace_kprobe_symbol(tk),
830 tk->rp.kp.offset);
831 else
832 seq_printf(m, " %s", trace_kprobe_symbol(tk));
833
834 for (i = 0; i < tk->tp.nr_args; i++)
835 seq_printf(m, " %s=%s", tk->tp.args[i].name, tk->tp.args[i].comm);
836 seq_printf(m, "\n");
837
838 return 0;
839}
840
841static const struct seq_operations probes_seq_op = {
842 .start = probes_seq_start,
843 .next = probes_seq_next,
844 .stop = probes_seq_stop,
845 .show = probes_seq_show
846};
847
848static int probes_open(struct inode *inode, struct file *file)
849{
850 int ret;
851
852 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
853 ret = release_all_trace_kprobes();
854 if (ret < 0)
855 return ret;
856 }
857
858 return seq_open(file, &probes_seq_op);
859}
860
861static ssize_t probes_write(struct file *file, const char __user *buffer,
862 size_t count, loff_t *ppos)
863{
864 return traceprobe_probes_write(file, buffer, count, ppos,
865 create_trace_kprobe);
866}
867
868static const struct file_operations kprobe_events_ops = {
869 .owner = THIS_MODULE,
870 .open = probes_open,
871 .read = seq_read,
872 .llseek = seq_lseek,
873 .release = seq_release,
874 .write = probes_write,
875};
876
877/* Probes profiling interfaces */
878static int probes_profile_seq_show(struct seq_file *m, void *v)
879{
880 struct trace_kprobe *tk = v;
881
882 seq_printf(m, " %-44s %15lu %15lu\n",
883 ftrace_event_name(&tk->tp.call), tk->nhit,
884 tk->rp.kp.nmissed);
885
886 return 0;
887}
888
889static const struct seq_operations profile_seq_op = {
890 .start = probes_seq_start,
891 .next = probes_seq_next,
892 .stop = probes_seq_stop,
893 .show = probes_profile_seq_show
894};
895
896static int profile_open(struct inode *inode, struct file *file)
897{
898 return seq_open(file, &profile_seq_op);
899}
900
901static const struct file_operations kprobe_profile_ops = {
902 .owner = THIS_MODULE,
903 .open = profile_open,
904 .read = seq_read,
905 .llseek = seq_lseek,
906 .release = seq_release,
907};
908
909/* Kprobe handler */
910static __kprobes void
911__kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs,
912 struct ftrace_event_file *ftrace_file)
913{
914 struct kprobe_trace_entry_head *entry;
915 struct ring_buffer_event *event;
916 struct ring_buffer *buffer;
917 int size, dsize, pc;
918 unsigned long irq_flags;
919 struct ftrace_event_call *call = &tk->tp.call;
920
921 WARN_ON(call != ftrace_file->event_call);
922
923 if (ftrace_trigger_soft_disabled(ftrace_file))
924 return;
925
926 local_save_flags(irq_flags);
927 pc = preempt_count();
928
929 dsize = __get_data_size(&tk->tp, regs);
930 size = sizeof(*entry) + tk->tp.size + dsize;
931
932 event = trace_event_buffer_lock_reserve(&buffer, ftrace_file,
933 call->event.type,
934 size, irq_flags, pc);
935 if (!event)
936 return;
937
938 entry = ring_buffer_event_data(event);
939 entry->ip = (unsigned long)tk->rp.kp.addr;
940 store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
941
942 event_trigger_unlock_commit_regs(ftrace_file, buffer, event,
943 entry, irq_flags, pc, regs);
944}
945
946static __kprobes void
947kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs)
948{
949 struct event_file_link *link;
950
951 list_for_each_entry_rcu(link, &tk->tp.files, list)
952 __kprobe_trace_func(tk, regs, link->file);
953}
954
955/* Kretprobe handler */
956static __kprobes void
957__kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
958 struct pt_regs *regs,
959 struct ftrace_event_file *ftrace_file)
960{
961 struct kretprobe_trace_entry_head *entry;
962 struct ring_buffer_event *event;
963 struct ring_buffer *buffer;
964 int size, pc, dsize;
965 unsigned long irq_flags;
966 struct ftrace_event_call *call = &tk->tp.call;
967
968 WARN_ON(call != ftrace_file->event_call);
969
970 if (ftrace_trigger_soft_disabled(ftrace_file))
971 return;
972
973 local_save_flags(irq_flags);
974 pc = preempt_count();
975
976 dsize = __get_data_size(&tk->tp, regs);
977 size = sizeof(*entry) + tk->tp.size + dsize;
978
979 event = trace_event_buffer_lock_reserve(&buffer, ftrace_file,
980 call->event.type,
981 size, irq_flags, pc);
982 if (!event)
983 return;
984
985 entry = ring_buffer_event_data(event);
986 entry->func = (unsigned long)tk->rp.kp.addr;
987 entry->ret_ip = (unsigned long)ri->ret_addr;
988 store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
989
990 event_trigger_unlock_commit_regs(ftrace_file, buffer, event,
991 entry, irq_flags, pc, regs);
992}
993
994static __kprobes void
995kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
996 struct pt_regs *regs)
997{
998 struct event_file_link *link;
999
1000 list_for_each_entry_rcu(link, &tk->tp.files, list)
1001 __kretprobe_trace_func(tk, ri, regs, link->file);
1002}
1003
1004/* Event entry printers */
1005static enum print_line_t
1006print_kprobe_event(struct trace_iterator *iter, int flags,
1007 struct trace_event *event)
1008{
1009 struct kprobe_trace_entry_head *field;
1010 struct trace_seq *s = &iter->seq;
1011 struct trace_probe *tp;
1012 u8 *data;
1013 int i;
1014
1015 field = (struct kprobe_trace_entry_head *)iter->ent;
1016 tp = container_of(event, struct trace_probe, call.event);
1017
1018 if (!trace_seq_printf(s, "%s: (", ftrace_event_name(&tp->call)))
1019 goto partial;
1020
1021 if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
1022 goto partial;
1023
1024 if (!trace_seq_puts(s, ")"))
1025 goto partial;
1026
1027 data = (u8 *)&field[1];
1028 for (i = 0; i < tp->nr_args; i++)
1029 if (!tp->args[i].type->print(s, tp->args[i].name,
1030 data + tp->args[i].offset, field))
1031 goto partial;
1032
1033 if (!trace_seq_puts(s, "\n"))
1034 goto partial;
1035
1036 return TRACE_TYPE_HANDLED;
1037partial:
1038 return TRACE_TYPE_PARTIAL_LINE;
1039}
1040
1041static enum print_line_t
1042print_kretprobe_event(struct trace_iterator *iter, int flags,
1043 struct trace_event *event)
1044{
1045 struct kretprobe_trace_entry_head *field;
1046 struct trace_seq *s = &iter->seq;
1047 struct trace_probe *tp;
1048 u8 *data;
1049 int i;
1050
1051 field = (struct kretprobe_trace_entry_head *)iter->ent;
1052 tp = container_of(event, struct trace_probe, call.event);
1053
1054 if (!trace_seq_printf(s, "%s: (", ftrace_event_name(&tp->call)))
1055 goto partial;
1056
1057 if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1058 goto partial;
1059
1060 if (!trace_seq_puts(s, " <- "))
1061 goto partial;
1062
1063 if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1064 goto partial;
1065
1066 if (!trace_seq_puts(s, ")"))
1067 goto partial;
1068
1069 data = (u8 *)&field[1];
1070 for (i = 0; i < tp->nr_args; i++)
1071 if (!tp->args[i].type->print(s, tp->args[i].name,
1072 data + tp->args[i].offset, field))
1073 goto partial;
1074
1075 if (!trace_seq_puts(s, "\n"))
1076 goto partial;
1077
1078 return TRACE_TYPE_HANDLED;
1079partial:
1080 return TRACE_TYPE_PARTIAL_LINE;
1081}
1082
1083
1084static int kprobe_event_define_fields(struct ftrace_event_call *event_call)
1085{
1086 int ret, i;
1087 struct kprobe_trace_entry_head field;
1088 struct trace_kprobe *tk = (struct trace_kprobe *)event_call->data;
1089
1090 DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1091 /* Set argument names as fields */
1092 for (i = 0; i < tk->tp.nr_args; i++) {
1093 struct probe_arg *parg = &tk->tp.args[i];
1094
1095 ret = trace_define_field(event_call, parg->type->fmttype,
1096 parg->name,
1097 sizeof(field) + parg->offset,
1098 parg->type->size,
1099 parg->type->is_signed,
1100 FILTER_OTHER);
1101 if (ret)
1102 return ret;
1103 }
1104 return 0;
1105}
1106
1107static int kretprobe_event_define_fields(struct ftrace_event_call *event_call)
1108{
1109 int ret, i;
1110 struct kretprobe_trace_entry_head field;
1111 struct trace_kprobe *tk = (struct trace_kprobe *)event_call->data;
1112
1113 DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1114 DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1115 /* Set argument names as fields */
1116 for (i = 0; i < tk->tp.nr_args; i++) {
1117 struct probe_arg *parg = &tk->tp.args[i];
1118
1119 ret = trace_define_field(event_call, parg->type->fmttype,
1120 parg->name,
1121 sizeof(field) + parg->offset,
1122 parg->type->size,
1123 parg->type->is_signed,
1124 FILTER_OTHER);
1125 if (ret)
1126 return ret;
1127 }
1128 return 0;
1129}
1130
1131#ifdef CONFIG_PERF_EVENTS
1132
1133/* Kprobe profile handler */
1134static __kprobes void
1135kprobe_perf_func(struct trace_kprobe *tk, struct pt_regs *regs)
1136{
1137 struct ftrace_event_call *call = &tk->tp.call;
1138 struct kprobe_trace_entry_head *entry;
1139 struct hlist_head *head;
1140 int size, __size, dsize;
1141 int rctx;
1142
1143 head = this_cpu_ptr(call->perf_events);
1144 if (hlist_empty(head))
1145 return;
1146
1147 dsize = __get_data_size(&tk->tp, regs);
1148 __size = sizeof(*entry) + tk->tp.size + dsize;
1149 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1150 size -= sizeof(u32);
1151
1152 entry = perf_trace_buf_prepare(size, call->event.type, regs, &rctx);
1153 if (!entry)
1154 return;
1155
1156 entry->ip = (unsigned long)tk->rp.kp.addr;
1157 memset(&entry[1], 0, dsize);
1158 store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1159 perf_trace_buf_submit(entry, size, rctx, 0, 1, regs, head, NULL);
1160}
1161
1162/* Kretprobe profile handler */
1163static __kprobes void
1164kretprobe_perf_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1165 struct pt_regs *regs)
1166{
1167 struct ftrace_event_call *call = &tk->tp.call;
1168 struct kretprobe_trace_entry_head *entry;
1169 struct hlist_head *head;
1170 int size, __size, dsize;
1171 int rctx;
1172
1173 head = this_cpu_ptr(call->perf_events);
1174 if (hlist_empty(head))
1175 return;
1176
1177 dsize = __get_data_size(&tk->tp, regs);
1178 __size = sizeof(*entry) + tk->tp.size + dsize;
1179 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1180 size -= sizeof(u32);
1181
1182 entry = perf_trace_buf_prepare(size, call->event.type, regs, &rctx);
1183 if (!entry)
1184 return;
1185
1186 entry->func = (unsigned long)tk->rp.kp.addr;
1187 entry->ret_ip = (unsigned long)ri->ret_addr;
1188 store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1189 perf_trace_buf_submit(entry, size, rctx, 0, 1, regs, head, NULL);
1190}
1191#endif /* CONFIG_PERF_EVENTS */
1192
1193/*
1194 * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex.
1195 *
1196 * kprobe_trace_self_tests_init() does enable_trace_probe/disable_trace_probe
1197 * lockless, but we can't race with this __init function.
1198 */
1199static __kprobes
1200int kprobe_register(struct ftrace_event_call *event,
1201 enum trace_reg type, void *data)
1202{
1203 struct trace_kprobe *tk = (struct trace_kprobe *)event->data;
1204 struct ftrace_event_file *file = data;
1205
1206 switch (type) {
1207 case TRACE_REG_REGISTER:
1208 return enable_trace_kprobe(tk, file);
1209 case TRACE_REG_UNREGISTER:
1210 return disable_trace_kprobe(tk, file);
1211
1212#ifdef CONFIG_PERF_EVENTS
1213 case TRACE_REG_PERF_REGISTER:
1214 return enable_trace_kprobe(tk, NULL);
1215 case TRACE_REG_PERF_UNREGISTER:
1216 return disable_trace_kprobe(tk, NULL);
1217 case TRACE_REG_PERF_OPEN:
1218 case TRACE_REG_PERF_CLOSE:
1219 case TRACE_REG_PERF_ADD:
1220 case TRACE_REG_PERF_DEL:
1221 return 0;
1222#endif
1223 }
1224 return 0;
1225}
1226
1227static __kprobes
1228int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1229{
1230 struct trace_kprobe *tk = container_of(kp, struct trace_kprobe, rp.kp);
1231
1232 tk->nhit++;
1233
1234 if (tk->tp.flags & TP_FLAG_TRACE)
1235 kprobe_trace_func(tk, regs);
1236#ifdef CONFIG_PERF_EVENTS
1237 if (tk->tp.flags & TP_FLAG_PROFILE)
1238 kprobe_perf_func(tk, regs);
1239#endif
1240 return 0; /* We don't tweek kernel, so just return 0 */
1241}
1242
1243static __kprobes
1244int kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1245{
1246 struct trace_kprobe *tk = container_of(ri->rp, struct trace_kprobe, rp);
1247
1248 tk->nhit++;
1249
1250 if (tk->tp.flags & TP_FLAG_TRACE)
1251 kretprobe_trace_func(tk, ri, regs);
1252#ifdef CONFIG_PERF_EVENTS
1253 if (tk->tp.flags & TP_FLAG_PROFILE)
1254 kretprobe_perf_func(tk, ri, regs);
1255#endif
1256 return 0; /* We don't tweek kernel, so just return 0 */
1257}
1258
1259static struct trace_event_functions kretprobe_funcs = {
1260 .trace = print_kretprobe_event
1261};
1262
1263static struct trace_event_functions kprobe_funcs = {
1264 .trace = print_kprobe_event
1265};
1266
1267static int register_kprobe_event(struct trace_kprobe *tk)
1268{
1269 struct ftrace_event_call *call = &tk->tp.call;
1270 int ret;
1271
1272 /* Initialize ftrace_event_call */
1273 INIT_LIST_HEAD(&call->class->fields);
1274 if (trace_kprobe_is_return(tk)) {
1275 call->event.funcs = &kretprobe_funcs;
1276 call->class->define_fields = kretprobe_event_define_fields;
1277 } else {
1278 call->event.funcs = &kprobe_funcs;
1279 call->class->define_fields = kprobe_event_define_fields;
1280 }
1281 if (set_print_fmt(&tk->tp, trace_kprobe_is_return(tk)) < 0)
1282 return -ENOMEM;
1283 ret = register_ftrace_event(&call->event);
1284 if (!ret) {
1285 kfree(call->print_fmt);
1286 return -ENODEV;
1287 }
1288 call->flags = 0;
1289 call->class->reg = kprobe_register;
1290 call->data = tk;
1291 ret = trace_add_event_call(call);
1292 if (ret) {
1293 pr_info("Failed to register kprobe event: %s\n",
1294 ftrace_event_name(call));
1295 kfree(call->print_fmt);
1296 unregister_ftrace_event(&call->event);
1297 }
1298 return ret;
1299}
1300
1301static int unregister_kprobe_event(struct trace_kprobe *tk)
1302{
1303 int ret;
1304
1305 /* tp->event is unregistered in trace_remove_event_call() */
1306 ret = trace_remove_event_call(&tk->tp.call);
1307 if (!ret)
1308 kfree(tk->tp.call.print_fmt);
1309 return ret;
1310}
1311
1312/* Make a debugfs interface for controlling probe points */
1313static __init int init_kprobe_trace(void)
1314{
1315 struct dentry *d_tracer;
1316 struct dentry *entry;
1317
1318 if (register_module_notifier(&trace_kprobe_module_nb))
1319 return -EINVAL;
1320
1321 d_tracer = tracing_init_dentry();
1322 if (!d_tracer)
1323 return 0;
1324
1325 entry = debugfs_create_file("kprobe_events", 0644, d_tracer,
1326 NULL, &kprobe_events_ops);
1327
1328 /* Event list interface */
1329 if (!entry)
1330 pr_warning("Could not create debugfs "
1331 "'kprobe_events' entry\n");
1332
1333 /* Profile interface */
1334 entry = debugfs_create_file("kprobe_profile", 0444, d_tracer,
1335 NULL, &kprobe_profile_ops);
1336
1337 if (!entry)
1338 pr_warning("Could not create debugfs "
1339 "'kprobe_profile' entry\n");
1340 return 0;
1341}
1342fs_initcall(init_kprobe_trace);
1343
1344
1345#ifdef CONFIG_FTRACE_STARTUP_TEST
1346
1347/*
1348 * The "__used" keeps gcc from removing the function symbol
1349 * from the kallsyms table.
1350 */
1351static __used int kprobe_trace_selftest_target(int a1, int a2, int a3,
1352 int a4, int a5, int a6)
1353{
1354 return a1 + a2 + a3 + a4 + a5 + a6;
1355}
1356
1357static struct ftrace_event_file *
1358find_trace_probe_file(struct trace_kprobe *tk, struct trace_array *tr)
1359{
1360 struct ftrace_event_file *file;
1361
1362 list_for_each_entry(file, &tr->events, list)
1363 if (file->event_call == &tk->tp.call)
1364 return file;
1365
1366 return NULL;
1367}
1368
1369/*
1370 * Nobody but us can call enable_trace_kprobe/disable_trace_kprobe at this
1371 * stage, we can do this lockless.
1372 */
1373static __init int kprobe_trace_self_tests_init(void)
1374{
1375 int ret, warn = 0;
1376 int (*target)(int, int, int, int, int, int);
1377 struct trace_kprobe *tk;
1378 struct ftrace_event_file *file;
1379
1380 target = kprobe_trace_selftest_target;
1381
1382 pr_info("Testing kprobe tracing: ");
1383
1384 ret = traceprobe_command("p:testprobe kprobe_trace_selftest_target "
1385 "$stack $stack0 +0($stack)",
1386 create_trace_kprobe);
1387 if (WARN_ON_ONCE(ret)) {
1388 pr_warn("error on probing function entry.\n");
1389 warn++;
1390 } else {
1391 /* Enable trace point */
1392 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1393 if (WARN_ON_ONCE(tk == NULL)) {
1394 pr_warn("error on getting new probe.\n");
1395 warn++;
1396 } else {
1397 file = find_trace_probe_file(tk, top_trace_array());
1398 if (WARN_ON_ONCE(file == NULL)) {
1399 pr_warn("error on getting probe file.\n");
1400 warn++;
1401 } else
1402 enable_trace_kprobe(tk, file);
1403 }
1404 }
1405
1406 ret = traceprobe_command("r:testprobe2 kprobe_trace_selftest_target "
1407 "$retval", create_trace_kprobe);
1408 if (WARN_ON_ONCE(ret)) {
1409 pr_warn("error on probing function return.\n");
1410 warn++;
1411 } else {
1412 /* Enable trace point */
1413 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1414 if (WARN_ON_ONCE(tk == NULL)) {
1415 pr_warn("error on getting 2nd new probe.\n");
1416 warn++;
1417 } else {
1418 file = find_trace_probe_file(tk, top_trace_array());
1419 if (WARN_ON_ONCE(file == NULL)) {
1420 pr_warn("error on getting probe file.\n");
1421 warn++;
1422 } else
1423 enable_trace_kprobe(tk, file);
1424 }
1425 }
1426
1427 if (warn)
1428 goto end;
1429
1430 ret = target(1, 2, 3, 4, 5, 6);
1431
1432 /* Disable trace points before removing it */
1433 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1434 if (WARN_ON_ONCE(tk == NULL)) {
1435 pr_warn("error on getting test probe.\n");
1436 warn++;
1437 } else {
1438 file = find_trace_probe_file(tk, top_trace_array());
1439 if (WARN_ON_ONCE(file == NULL)) {
1440 pr_warn("error on getting probe file.\n");
1441 warn++;
1442 } else
1443 disable_trace_kprobe(tk, file);
1444 }
1445
1446 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1447 if (WARN_ON_ONCE(tk == NULL)) {
1448 pr_warn("error on getting 2nd test probe.\n");
1449 warn++;
1450 } else {
1451 file = find_trace_probe_file(tk, top_trace_array());
1452 if (WARN_ON_ONCE(file == NULL)) {
1453 pr_warn("error on getting probe file.\n");
1454 warn++;
1455 } else
1456 disable_trace_kprobe(tk, file);
1457 }
1458
1459 ret = traceprobe_command("-:testprobe", create_trace_kprobe);
1460 if (WARN_ON_ONCE(ret)) {
1461 pr_warn("error on deleting a probe.\n");
1462 warn++;
1463 }
1464
1465 ret = traceprobe_command("-:testprobe2", create_trace_kprobe);
1466 if (WARN_ON_ONCE(ret)) {
1467 pr_warn("error on deleting a probe.\n");
1468 warn++;
1469 }
1470
1471end:
1472 release_all_trace_kprobes();
1473 if (warn)
1474 pr_cont("NG: Some tests are failed. Please check them.\n");
1475 else
1476 pr_cont("OK\n");
1477 return 0;
1478}
1479
1480late_initcall(kprobe_trace_self_tests_init);
1481
1482#endif
1/*
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 */
30
31struct trace_probe {
32 struct list_head list;
33 struct kretprobe rp; /* Use rp.kp for kprobe use */
34 unsigned long nhit;
35 unsigned int flags; /* For TP_FLAG_* */
36 const char *symbol; /* symbol name */
37 struct ftrace_event_class class;
38 struct ftrace_event_call call;
39 ssize_t size; /* trace entry size */
40 unsigned int nr_args;
41 struct probe_arg args[];
42};
43
44#define SIZEOF_TRACE_PROBE(n) \
45 (offsetof(struct trace_probe, args) + \
46 (sizeof(struct probe_arg) * (n)))
47
48
49static __kprobes int trace_probe_is_return(struct trace_probe *tp)
50{
51 return tp->rp.handler != NULL;
52}
53
54static __kprobes const char *trace_probe_symbol(struct trace_probe *tp)
55{
56 return tp->symbol ? tp->symbol : "unknown";
57}
58
59static __kprobes unsigned long trace_probe_offset(struct trace_probe *tp)
60{
61 return tp->rp.kp.offset;
62}
63
64static __kprobes bool trace_probe_is_enabled(struct trace_probe *tp)
65{
66 return !!(tp->flags & (TP_FLAG_TRACE | TP_FLAG_PROFILE));
67}
68
69static __kprobes bool trace_probe_is_registered(struct trace_probe *tp)
70{
71 return !!(tp->flags & TP_FLAG_REGISTERED);
72}
73
74static __kprobes bool trace_probe_has_gone(struct trace_probe *tp)
75{
76 return !!(kprobe_gone(&tp->rp.kp));
77}
78
79static __kprobes bool trace_probe_within_module(struct trace_probe *tp,
80 struct module *mod)
81{
82 int len = strlen(mod->name);
83 const char *name = trace_probe_symbol(tp);
84 return strncmp(mod->name, name, len) == 0 && name[len] == ':';
85}
86
87static __kprobes bool trace_probe_is_on_module(struct trace_probe *tp)
88{
89 return !!strchr(trace_probe_symbol(tp), ':');
90}
91
92static int register_probe_event(struct trace_probe *tp);
93static void unregister_probe_event(struct trace_probe *tp);
94
95static DEFINE_MUTEX(probe_lock);
96static LIST_HEAD(probe_list);
97
98static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
99static int kretprobe_dispatcher(struct kretprobe_instance *ri,
100 struct pt_regs *regs);
101
102/*
103 * Allocate new trace_probe and initialize it (including kprobes).
104 */
105static struct trace_probe *alloc_trace_probe(const char *group,
106 const char *event,
107 void *addr,
108 const char *symbol,
109 unsigned long offs,
110 int nargs, bool is_return)
111{
112 struct trace_probe *tp;
113 int ret = -ENOMEM;
114
115 tp = kzalloc(SIZEOF_TRACE_PROBE(nargs), GFP_KERNEL);
116 if (!tp)
117 return ERR_PTR(ret);
118
119 if (symbol) {
120 tp->symbol = kstrdup(symbol, GFP_KERNEL);
121 if (!tp->symbol)
122 goto error;
123 tp->rp.kp.symbol_name = tp->symbol;
124 tp->rp.kp.offset = offs;
125 } else
126 tp->rp.kp.addr = addr;
127
128 if (is_return)
129 tp->rp.handler = kretprobe_dispatcher;
130 else
131 tp->rp.kp.pre_handler = kprobe_dispatcher;
132
133 if (!event || !is_good_name(event)) {
134 ret = -EINVAL;
135 goto error;
136 }
137
138 tp->call.class = &tp->class;
139 tp->call.name = kstrdup(event, GFP_KERNEL);
140 if (!tp->call.name)
141 goto error;
142
143 if (!group || !is_good_name(group)) {
144 ret = -EINVAL;
145 goto error;
146 }
147
148 tp->class.system = kstrdup(group, GFP_KERNEL);
149 if (!tp->class.system)
150 goto error;
151
152 INIT_LIST_HEAD(&tp->list);
153 return tp;
154error:
155 kfree(tp->call.name);
156 kfree(tp->symbol);
157 kfree(tp);
158 return ERR_PTR(ret);
159}
160
161static void free_trace_probe(struct trace_probe *tp)
162{
163 int i;
164
165 for (i = 0; i < tp->nr_args; i++)
166 traceprobe_free_probe_arg(&tp->args[i]);
167
168 kfree(tp->call.class->system);
169 kfree(tp->call.name);
170 kfree(tp->symbol);
171 kfree(tp);
172}
173
174static struct trace_probe *find_trace_probe(const char *event,
175 const char *group)
176{
177 struct trace_probe *tp;
178
179 list_for_each_entry(tp, &probe_list, list)
180 if (strcmp(tp->call.name, event) == 0 &&
181 strcmp(tp->call.class->system, group) == 0)
182 return tp;
183 return NULL;
184}
185
186/* Enable trace_probe - @flag must be TP_FLAG_TRACE or TP_FLAG_PROFILE */
187static int enable_trace_probe(struct trace_probe *tp, int flag)
188{
189 int ret = 0;
190
191 tp->flags |= flag;
192 if (trace_probe_is_enabled(tp) && trace_probe_is_registered(tp) &&
193 !trace_probe_has_gone(tp)) {
194 if (trace_probe_is_return(tp))
195 ret = enable_kretprobe(&tp->rp);
196 else
197 ret = enable_kprobe(&tp->rp.kp);
198 }
199
200 return ret;
201}
202
203/* Disable trace_probe - @flag must be TP_FLAG_TRACE or TP_FLAG_PROFILE */
204static void disable_trace_probe(struct trace_probe *tp, int flag)
205{
206 tp->flags &= ~flag;
207 if (!trace_probe_is_enabled(tp) && trace_probe_is_registered(tp)) {
208 if (trace_probe_is_return(tp))
209 disable_kretprobe(&tp->rp);
210 else
211 disable_kprobe(&tp->rp.kp);
212 }
213}
214
215/* Internal register function - just handle k*probes and flags */
216static int __register_trace_probe(struct trace_probe *tp)
217{
218 int i, ret;
219
220 if (trace_probe_is_registered(tp))
221 return -EINVAL;
222
223 for (i = 0; i < tp->nr_args; i++)
224 traceprobe_update_arg(&tp->args[i]);
225
226 /* Set/clear disabled flag according to tp->flag */
227 if (trace_probe_is_enabled(tp))
228 tp->rp.kp.flags &= ~KPROBE_FLAG_DISABLED;
229 else
230 tp->rp.kp.flags |= KPROBE_FLAG_DISABLED;
231
232 if (trace_probe_is_return(tp))
233 ret = register_kretprobe(&tp->rp);
234 else
235 ret = register_kprobe(&tp->rp.kp);
236
237 if (ret == 0)
238 tp->flags |= TP_FLAG_REGISTERED;
239 else {
240 pr_warning("Could not insert probe at %s+%lu: %d\n",
241 trace_probe_symbol(tp), trace_probe_offset(tp), ret);
242 if (ret == -ENOENT && trace_probe_is_on_module(tp)) {
243 pr_warning("This probe might be able to register after"
244 "target module is loaded. Continue.\n");
245 ret = 0;
246 } else if (ret == -EILSEQ) {
247 pr_warning("Probing address(0x%p) is not an "
248 "instruction boundary.\n",
249 tp->rp.kp.addr);
250 ret = -EINVAL;
251 }
252 }
253
254 return ret;
255}
256
257/* Internal unregister function - just handle k*probes and flags */
258static void __unregister_trace_probe(struct trace_probe *tp)
259{
260 if (trace_probe_is_registered(tp)) {
261 if (trace_probe_is_return(tp))
262 unregister_kretprobe(&tp->rp);
263 else
264 unregister_kprobe(&tp->rp.kp);
265 tp->flags &= ~TP_FLAG_REGISTERED;
266 /* Cleanup kprobe for reuse */
267 if (tp->rp.kp.symbol_name)
268 tp->rp.kp.addr = NULL;
269 }
270}
271
272/* Unregister a trace_probe and probe_event: call with locking probe_lock */
273static int unregister_trace_probe(struct trace_probe *tp)
274{
275 /* Enabled event can not be unregistered */
276 if (trace_probe_is_enabled(tp))
277 return -EBUSY;
278
279 __unregister_trace_probe(tp);
280 list_del(&tp->list);
281 unregister_probe_event(tp);
282
283 return 0;
284}
285
286/* Register a trace_probe and probe_event */
287static int register_trace_probe(struct trace_probe *tp)
288{
289 struct trace_probe *old_tp;
290 int ret;
291
292 mutex_lock(&probe_lock);
293
294 /* Delete old (same name) event if exist */
295 old_tp = find_trace_probe(tp->call.name, tp->call.class->system);
296 if (old_tp) {
297 ret = unregister_trace_probe(old_tp);
298 if (ret < 0)
299 goto end;
300 free_trace_probe(old_tp);
301 }
302
303 /* Register new event */
304 ret = register_probe_event(tp);
305 if (ret) {
306 pr_warning("Failed to register probe event(%d)\n", ret);
307 goto end;
308 }
309
310 /* Register k*probe */
311 ret = __register_trace_probe(tp);
312 if (ret < 0)
313 unregister_probe_event(tp);
314 else
315 list_add_tail(&tp->list, &probe_list);
316
317end:
318 mutex_unlock(&probe_lock);
319 return ret;
320}
321
322/* Module notifier call back, checking event on the module */
323static int trace_probe_module_callback(struct notifier_block *nb,
324 unsigned long val, void *data)
325{
326 struct module *mod = data;
327 struct trace_probe *tp;
328 int ret;
329
330 if (val != MODULE_STATE_COMING)
331 return NOTIFY_DONE;
332
333 /* Update probes on coming module */
334 mutex_lock(&probe_lock);
335 list_for_each_entry(tp, &probe_list, list) {
336 if (trace_probe_within_module(tp, mod)) {
337 /* Don't need to check busy - this should have gone. */
338 __unregister_trace_probe(tp);
339 ret = __register_trace_probe(tp);
340 if (ret)
341 pr_warning("Failed to re-register probe %s on"
342 "%s: %d\n",
343 tp->call.name, mod->name, ret);
344 }
345 }
346 mutex_unlock(&probe_lock);
347
348 return NOTIFY_DONE;
349}
350
351static struct notifier_block trace_probe_module_nb = {
352 .notifier_call = trace_probe_module_callback,
353 .priority = 1 /* Invoked after kprobe module callback */
354};
355
356static int create_trace_probe(int argc, char **argv)
357{
358 /*
359 * Argument syntax:
360 * - Add kprobe: p[:[GRP/]EVENT] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
361 * - Add kretprobe: r[:[GRP/]EVENT] [MOD:]KSYM[+0] [FETCHARGS]
362 * Fetch args:
363 * $retval : fetch return value
364 * $stack : fetch stack address
365 * $stackN : fetch Nth of stack (N:0-)
366 * @ADDR : fetch memory at ADDR (ADDR should be in kernel)
367 * @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
368 * %REG : fetch register REG
369 * Dereferencing memory fetch:
370 * +|-offs(ARG) : fetch memory at ARG +|- offs address.
371 * Alias name of args:
372 * NAME=FETCHARG : set NAME as alias of FETCHARG.
373 * Type of args:
374 * FETCHARG:TYPE : use TYPE instead of unsigned long.
375 */
376 struct trace_probe *tp;
377 int i, ret = 0;
378 bool is_return = false, is_delete = false;
379 char *symbol = NULL, *event = NULL, *group = NULL;
380 char *arg;
381 unsigned long offset = 0;
382 void *addr = NULL;
383 char buf[MAX_EVENT_NAME_LEN];
384
385 /* argc must be >= 1 */
386 if (argv[0][0] == 'p')
387 is_return = false;
388 else if (argv[0][0] == 'r')
389 is_return = true;
390 else if (argv[0][0] == '-')
391 is_delete = true;
392 else {
393 pr_info("Probe definition must be started with 'p', 'r' or"
394 " '-'.\n");
395 return -EINVAL;
396 }
397
398 if (argv[0][1] == ':') {
399 event = &argv[0][2];
400 if (strchr(event, '/')) {
401 group = event;
402 event = strchr(group, '/') + 1;
403 event[-1] = '\0';
404 if (strlen(group) == 0) {
405 pr_info("Group name is not specified\n");
406 return -EINVAL;
407 }
408 }
409 if (strlen(event) == 0) {
410 pr_info("Event name is not specified\n");
411 return -EINVAL;
412 }
413 }
414 if (!group)
415 group = KPROBE_EVENT_SYSTEM;
416
417 if (is_delete) {
418 if (!event) {
419 pr_info("Delete command needs an event name.\n");
420 return -EINVAL;
421 }
422 mutex_lock(&probe_lock);
423 tp = find_trace_probe(event, group);
424 if (!tp) {
425 mutex_unlock(&probe_lock);
426 pr_info("Event %s/%s doesn't exist.\n", group, event);
427 return -ENOENT;
428 }
429 /* delete an event */
430 ret = unregister_trace_probe(tp);
431 if (ret == 0)
432 free_trace_probe(tp);
433 mutex_unlock(&probe_lock);
434 return ret;
435 }
436
437 if (argc < 2) {
438 pr_info("Probe point is not specified.\n");
439 return -EINVAL;
440 }
441 if (isdigit(argv[1][0])) {
442 if (is_return) {
443 pr_info("Return probe point must be a symbol.\n");
444 return -EINVAL;
445 }
446 /* an address specified */
447 ret = strict_strtoul(&argv[1][0], 0, (unsigned long *)&addr);
448 if (ret) {
449 pr_info("Failed to parse address.\n");
450 return ret;
451 }
452 } else {
453 /* a symbol specified */
454 symbol = argv[1];
455 /* TODO: support .init module functions */
456 ret = traceprobe_split_symbol_offset(symbol, &offset);
457 if (ret) {
458 pr_info("Failed to parse symbol.\n");
459 return ret;
460 }
461 if (offset && is_return) {
462 pr_info("Return probe must be used without offset.\n");
463 return -EINVAL;
464 }
465 }
466 argc -= 2; argv += 2;
467
468 /* setup a probe */
469 if (!event) {
470 /* Make a new event name */
471 if (symbol)
472 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
473 is_return ? 'r' : 'p', symbol, offset);
474 else
475 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
476 is_return ? 'r' : 'p', addr);
477 event = buf;
478 }
479 tp = alloc_trace_probe(group, event, addr, symbol, offset, argc,
480 is_return);
481 if (IS_ERR(tp)) {
482 pr_info("Failed to allocate trace_probe.(%d)\n",
483 (int)PTR_ERR(tp));
484 return PTR_ERR(tp);
485 }
486
487 /* parse arguments */
488 ret = 0;
489 for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
490 /* Increment count for freeing args in error case */
491 tp->nr_args++;
492
493 /* Parse argument name */
494 arg = strchr(argv[i], '=');
495 if (arg) {
496 *arg++ = '\0';
497 tp->args[i].name = kstrdup(argv[i], GFP_KERNEL);
498 } else {
499 arg = argv[i];
500 /* If argument name is omitted, set "argN" */
501 snprintf(buf, MAX_EVENT_NAME_LEN, "arg%d", i + 1);
502 tp->args[i].name = kstrdup(buf, GFP_KERNEL);
503 }
504
505 if (!tp->args[i].name) {
506 pr_info("Failed to allocate argument[%d] name.\n", i);
507 ret = -ENOMEM;
508 goto error;
509 }
510
511 if (!is_good_name(tp->args[i].name)) {
512 pr_info("Invalid argument[%d] name: %s\n",
513 i, tp->args[i].name);
514 ret = -EINVAL;
515 goto error;
516 }
517
518 if (traceprobe_conflict_field_name(tp->args[i].name,
519 tp->args, i)) {
520 pr_info("Argument[%d] name '%s' conflicts with "
521 "another field.\n", i, argv[i]);
522 ret = -EINVAL;
523 goto error;
524 }
525
526 /* Parse fetch argument */
527 ret = traceprobe_parse_probe_arg(arg, &tp->size, &tp->args[i],
528 is_return, true);
529 if (ret) {
530 pr_info("Parse error at argument[%d]. (%d)\n", i, ret);
531 goto error;
532 }
533 }
534
535 ret = register_trace_probe(tp);
536 if (ret)
537 goto error;
538 return 0;
539
540error:
541 free_trace_probe(tp);
542 return ret;
543}
544
545static int release_all_trace_probes(void)
546{
547 struct trace_probe *tp;
548 int ret = 0;
549
550 mutex_lock(&probe_lock);
551 /* Ensure no probe is in use. */
552 list_for_each_entry(tp, &probe_list, list)
553 if (trace_probe_is_enabled(tp)) {
554 ret = -EBUSY;
555 goto end;
556 }
557 /* TODO: Use batch unregistration */
558 while (!list_empty(&probe_list)) {
559 tp = list_entry(probe_list.next, struct trace_probe, list);
560 unregister_trace_probe(tp);
561 free_trace_probe(tp);
562 }
563
564end:
565 mutex_unlock(&probe_lock);
566
567 return ret;
568}
569
570/* Probes listing interfaces */
571static void *probes_seq_start(struct seq_file *m, loff_t *pos)
572{
573 mutex_lock(&probe_lock);
574 return seq_list_start(&probe_list, *pos);
575}
576
577static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
578{
579 return seq_list_next(v, &probe_list, pos);
580}
581
582static void probes_seq_stop(struct seq_file *m, void *v)
583{
584 mutex_unlock(&probe_lock);
585}
586
587static int probes_seq_show(struct seq_file *m, void *v)
588{
589 struct trace_probe *tp = v;
590 int i;
591
592 seq_printf(m, "%c", trace_probe_is_return(tp) ? 'r' : 'p');
593 seq_printf(m, ":%s/%s", tp->call.class->system, tp->call.name);
594
595 if (!tp->symbol)
596 seq_printf(m, " 0x%p", tp->rp.kp.addr);
597 else if (tp->rp.kp.offset)
598 seq_printf(m, " %s+%u", trace_probe_symbol(tp),
599 tp->rp.kp.offset);
600 else
601 seq_printf(m, " %s", trace_probe_symbol(tp));
602
603 for (i = 0; i < tp->nr_args; i++)
604 seq_printf(m, " %s=%s", tp->args[i].name, tp->args[i].comm);
605 seq_printf(m, "\n");
606
607 return 0;
608}
609
610static const struct seq_operations probes_seq_op = {
611 .start = probes_seq_start,
612 .next = probes_seq_next,
613 .stop = probes_seq_stop,
614 .show = probes_seq_show
615};
616
617static int probes_open(struct inode *inode, struct file *file)
618{
619 int ret;
620
621 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
622 ret = release_all_trace_probes();
623 if (ret < 0)
624 return ret;
625 }
626
627 return seq_open(file, &probes_seq_op);
628}
629
630static ssize_t probes_write(struct file *file, const char __user *buffer,
631 size_t count, loff_t *ppos)
632{
633 return traceprobe_probes_write(file, buffer, count, ppos,
634 create_trace_probe);
635}
636
637static const struct file_operations kprobe_events_ops = {
638 .owner = THIS_MODULE,
639 .open = probes_open,
640 .read = seq_read,
641 .llseek = seq_lseek,
642 .release = seq_release,
643 .write = probes_write,
644};
645
646/* Probes profiling interfaces */
647static int probes_profile_seq_show(struct seq_file *m, void *v)
648{
649 struct trace_probe *tp = v;
650
651 seq_printf(m, " %-44s %15lu %15lu\n", tp->call.name, tp->nhit,
652 tp->rp.kp.nmissed);
653
654 return 0;
655}
656
657static const struct seq_operations profile_seq_op = {
658 .start = probes_seq_start,
659 .next = probes_seq_next,
660 .stop = probes_seq_stop,
661 .show = probes_profile_seq_show
662};
663
664static int profile_open(struct inode *inode, struct file *file)
665{
666 return seq_open(file, &profile_seq_op);
667}
668
669static const struct file_operations kprobe_profile_ops = {
670 .owner = THIS_MODULE,
671 .open = profile_open,
672 .read = seq_read,
673 .llseek = seq_lseek,
674 .release = seq_release,
675};
676
677/* Sum up total data length for dynamic arraies (strings) */
678static __kprobes int __get_data_size(struct trace_probe *tp,
679 struct pt_regs *regs)
680{
681 int i, ret = 0;
682 u32 len;
683
684 for (i = 0; i < tp->nr_args; i++)
685 if (unlikely(tp->args[i].fetch_size.fn)) {
686 call_fetch(&tp->args[i].fetch_size, regs, &len);
687 ret += len;
688 }
689
690 return ret;
691}
692
693/* Store the value of each argument */
694static __kprobes void store_trace_args(int ent_size, struct trace_probe *tp,
695 struct pt_regs *regs,
696 u8 *data, int maxlen)
697{
698 int i;
699 u32 end = tp->size;
700 u32 *dl; /* Data (relative) location */
701
702 for (i = 0; i < tp->nr_args; i++) {
703 if (unlikely(tp->args[i].fetch_size.fn)) {
704 /*
705 * First, we set the relative location and
706 * maximum data length to *dl
707 */
708 dl = (u32 *)(data + tp->args[i].offset);
709 *dl = make_data_rloc(maxlen, end - tp->args[i].offset);
710 /* Then try to fetch string or dynamic array data */
711 call_fetch(&tp->args[i].fetch, regs, dl);
712 /* Reduce maximum length */
713 end += get_rloc_len(*dl);
714 maxlen -= get_rloc_len(*dl);
715 /* Trick here, convert data_rloc to data_loc */
716 *dl = convert_rloc_to_loc(*dl,
717 ent_size + tp->args[i].offset);
718 } else
719 /* Just fetching data normally */
720 call_fetch(&tp->args[i].fetch, regs,
721 data + tp->args[i].offset);
722 }
723}
724
725/* Kprobe handler */
726static __kprobes void kprobe_trace_func(struct kprobe *kp, struct pt_regs *regs)
727{
728 struct trace_probe *tp = container_of(kp, struct trace_probe, rp.kp);
729 struct kprobe_trace_entry_head *entry;
730 struct ring_buffer_event *event;
731 struct ring_buffer *buffer;
732 int size, dsize, pc;
733 unsigned long irq_flags;
734 struct ftrace_event_call *call = &tp->call;
735
736 tp->nhit++;
737
738 local_save_flags(irq_flags);
739 pc = preempt_count();
740
741 dsize = __get_data_size(tp, regs);
742 size = sizeof(*entry) + tp->size + dsize;
743
744 event = trace_current_buffer_lock_reserve(&buffer, call->event.type,
745 size, irq_flags, pc);
746 if (!event)
747 return;
748
749 entry = ring_buffer_event_data(event);
750 entry->ip = (unsigned long)kp->addr;
751 store_trace_args(sizeof(*entry), tp, regs, (u8 *)&entry[1], dsize);
752
753 if (!filter_current_check_discard(buffer, call, entry, event))
754 trace_nowake_buffer_unlock_commit_regs(buffer, event,
755 irq_flags, pc, regs);
756}
757
758/* Kretprobe handler */
759static __kprobes void kretprobe_trace_func(struct kretprobe_instance *ri,
760 struct pt_regs *regs)
761{
762 struct trace_probe *tp = container_of(ri->rp, struct trace_probe, rp);
763 struct kretprobe_trace_entry_head *entry;
764 struct ring_buffer_event *event;
765 struct ring_buffer *buffer;
766 int size, pc, dsize;
767 unsigned long irq_flags;
768 struct ftrace_event_call *call = &tp->call;
769
770 local_save_flags(irq_flags);
771 pc = preempt_count();
772
773 dsize = __get_data_size(tp, regs);
774 size = sizeof(*entry) + tp->size + dsize;
775
776 event = trace_current_buffer_lock_reserve(&buffer, call->event.type,
777 size, irq_flags, pc);
778 if (!event)
779 return;
780
781 entry = ring_buffer_event_data(event);
782 entry->func = (unsigned long)tp->rp.kp.addr;
783 entry->ret_ip = (unsigned long)ri->ret_addr;
784 store_trace_args(sizeof(*entry), tp, regs, (u8 *)&entry[1], dsize);
785
786 if (!filter_current_check_discard(buffer, call, entry, event))
787 trace_nowake_buffer_unlock_commit_regs(buffer, event,
788 irq_flags, pc, regs);
789}
790
791/* Event entry printers */
792enum print_line_t
793print_kprobe_event(struct trace_iterator *iter, int flags,
794 struct trace_event *event)
795{
796 struct kprobe_trace_entry_head *field;
797 struct trace_seq *s = &iter->seq;
798 struct trace_probe *tp;
799 u8 *data;
800 int i;
801
802 field = (struct kprobe_trace_entry_head *)iter->ent;
803 tp = container_of(event, struct trace_probe, call.event);
804
805 if (!trace_seq_printf(s, "%s: (", tp->call.name))
806 goto partial;
807
808 if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
809 goto partial;
810
811 if (!trace_seq_puts(s, ")"))
812 goto partial;
813
814 data = (u8 *)&field[1];
815 for (i = 0; i < tp->nr_args; i++)
816 if (!tp->args[i].type->print(s, tp->args[i].name,
817 data + tp->args[i].offset, field))
818 goto partial;
819
820 if (!trace_seq_puts(s, "\n"))
821 goto partial;
822
823 return TRACE_TYPE_HANDLED;
824partial:
825 return TRACE_TYPE_PARTIAL_LINE;
826}
827
828enum print_line_t
829print_kretprobe_event(struct trace_iterator *iter, int flags,
830 struct trace_event *event)
831{
832 struct kretprobe_trace_entry_head *field;
833 struct trace_seq *s = &iter->seq;
834 struct trace_probe *tp;
835 u8 *data;
836 int i;
837
838 field = (struct kretprobe_trace_entry_head *)iter->ent;
839 tp = container_of(event, struct trace_probe, call.event);
840
841 if (!trace_seq_printf(s, "%s: (", tp->call.name))
842 goto partial;
843
844 if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
845 goto partial;
846
847 if (!trace_seq_puts(s, " <- "))
848 goto partial;
849
850 if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
851 goto partial;
852
853 if (!trace_seq_puts(s, ")"))
854 goto partial;
855
856 data = (u8 *)&field[1];
857 for (i = 0; i < tp->nr_args; i++)
858 if (!tp->args[i].type->print(s, tp->args[i].name,
859 data + tp->args[i].offset, field))
860 goto partial;
861
862 if (!trace_seq_puts(s, "\n"))
863 goto partial;
864
865 return TRACE_TYPE_HANDLED;
866partial:
867 return TRACE_TYPE_PARTIAL_LINE;
868}
869
870
871static int kprobe_event_define_fields(struct ftrace_event_call *event_call)
872{
873 int ret, i;
874 struct kprobe_trace_entry_head field;
875 struct trace_probe *tp = (struct trace_probe *)event_call->data;
876
877 DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
878 /* Set argument names as fields */
879 for (i = 0; i < tp->nr_args; i++) {
880 ret = trace_define_field(event_call, tp->args[i].type->fmttype,
881 tp->args[i].name,
882 sizeof(field) + tp->args[i].offset,
883 tp->args[i].type->size,
884 tp->args[i].type->is_signed,
885 FILTER_OTHER);
886 if (ret)
887 return ret;
888 }
889 return 0;
890}
891
892static int kretprobe_event_define_fields(struct ftrace_event_call *event_call)
893{
894 int ret, i;
895 struct kretprobe_trace_entry_head field;
896 struct trace_probe *tp = (struct trace_probe *)event_call->data;
897
898 DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
899 DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
900 /* Set argument names as fields */
901 for (i = 0; i < tp->nr_args; i++) {
902 ret = trace_define_field(event_call, tp->args[i].type->fmttype,
903 tp->args[i].name,
904 sizeof(field) + tp->args[i].offset,
905 tp->args[i].type->size,
906 tp->args[i].type->is_signed,
907 FILTER_OTHER);
908 if (ret)
909 return ret;
910 }
911 return 0;
912}
913
914static int __set_print_fmt(struct trace_probe *tp, char *buf, int len)
915{
916 int i;
917 int pos = 0;
918
919 const char *fmt, *arg;
920
921 if (!trace_probe_is_return(tp)) {
922 fmt = "(%lx)";
923 arg = "REC->" FIELD_STRING_IP;
924 } else {
925 fmt = "(%lx <- %lx)";
926 arg = "REC->" FIELD_STRING_FUNC ", REC->" FIELD_STRING_RETIP;
927 }
928
929 /* When len=0, we just calculate the needed length */
930#define LEN_OR_ZERO (len ? len - pos : 0)
931
932 pos += snprintf(buf + pos, LEN_OR_ZERO, "\"%s", fmt);
933
934 for (i = 0; i < tp->nr_args; i++) {
935 pos += snprintf(buf + pos, LEN_OR_ZERO, " %s=%s",
936 tp->args[i].name, tp->args[i].type->fmt);
937 }
938
939 pos += snprintf(buf + pos, LEN_OR_ZERO, "\", %s", arg);
940
941 for (i = 0; i < tp->nr_args; i++) {
942 if (strcmp(tp->args[i].type->name, "string") == 0)
943 pos += snprintf(buf + pos, LEN_OR_ZERO,
944 ", __get_str(%s)",
945 tp->args[i].name);
946 else
947 pos += snprintf(buf + pos, LEN_OR_ZERO, ", REC->%s",
948 tp->args[i].name);
949 }
950
951#undef LEN_OR_ZERO
952
953 /* return the length of print_fmt */
954 return pos;
955}
956
957static int set_print_fmt(struct trace_probe *tp)
958{
959 int len;
960 char *print_fmt;
961
962 /* First: called with 0 length to calculate the needed length */
963 len = __set_print_fmt(tp, NULL, 0);
964 print_fmt = kmalloc(len + 1, GFP_KERNEL);
965 if (!print_fmt)
966 return -ENOMEM;
967
968 /* Second: actually write the @print_fmt */
969 __set_print_fmt(tp, print_fmt, len + 1);
970 tp->call.print_fmt = print_fmt;
971
972 return 0;
973}
974
975#ifdef CONFIG_PERF_EVENTS
976
977/* Kprobe profile handler */
978static __kprobes void kprobe_perf_func(struct kprobe *kp,
979 struct pt_regs *regs)
980{
981 struct trace_probe *tp = container_of(kp, struct trace_probe, rp.kp);
982 struct ftrace_event_call *call = &tp->call;
983 struct kprobe_trace_entry_head *entry;
984 struct hlist_head *head;
985 int size, __size, dsize;
986 int rctx;
987
988 dsize = __get_data_size(tp, regs);
989 __size = sizeof(*entry) + tp->size + dsize;
990 size = ALIGN(__size + sizeof(u32), sizeof(u64));
991 size -= sizeof(u32);
992 if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE,
993 "profile buffer not large enough"))
994 return;
995
996 entry = perf_trace_buf_prepare(size, call->event.type, regs, &rctx);
997 if (!entry)
998 return;
999
1000 entry->ip = (unsigned long)kp->addr;
1001 memset(&entry[1], 0, dsize);
1002 store_trace_args(sizeof(*entry), tp, regs, (u8 *)&entry[1], dsize);
1003
1004 head = this_cpu_ptr(call->perf_events);
1005 perf_trace_buf_submit(entry, size, rctx, entry->ip, 1, regs, head);
1006}
1007
1008/* Kretprobe profile handler */
1009static __kprobes void kretprobe_perf_func(struct kretprobe_instance *ri,
1010 struct pt_regs *regs)
1011{
1012 struct trace_probe *tp = container_of(ri->rp, struct trace_probe, rp);
1013 struct ftrace_event_call *call = &tp->call;
1014 struct kretprobe_trace_entry_head *entry;
1015 struct hlist_head *head;
1016 int size, __size, dsize;
1017 int rctx;
1018
1019 dsize = __get_data_size(tp, regs);
1020 __size = sizeof(*entry) + tp->size + dsize;
1021 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1022 size -= sizeof(u32);
1023 if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE,
1024 "profile buffer not large enough"))
1025 return;
1026
1027 entry = perf_trace_buf_prepare(size, call->event.type, regs, &rctx);
1028 if (!entry)
1029 return;
1030
1031 entry->func = (unsigned long)tp->rp.kp.addr;
1032 entry->ret_ip = (unsigned long)ri->ret_addr;
1033 store_trace_args(sizeof(*entry), tp, regs, (u8 *)&entry[1], dsize);
1034
1035 head = this_cpu_ptr(call->perf_events);
1036 perf_trace_buf_submit(entry, size, rctx, entry->ret_ip, 1, regs, head);
1037}
1038#endif /* CONFIG_PERF_EVENTS */
1039
1040static __kprobes
1041int kprobe_register(struct ftrace_event_call *event,
1042 enum trace_reg type, void *data)
1043{
1044 struct trace_probe *tp = (struct trace_probe *)event->data;
1045
1046 switch (type) {
1047 case TRACE_REG_REGISTER:
1048 return enable_trace_probe(tp, TP_FLAG_TRACE);
1049 case TRACE_REG_UNREGISTER:
1050 disable_trace_probe(tp, TP_FLAG_TRACE);
1051 return 0;
1052
1053#ifdef CONFIG_PERF_EVENTS
1054 case TRACE_REG_PERF_REGISTER:
1055 return enable_trace_probe(tp, TP_FLAG_PROFILE);
1056 case TRACE_REG_PERF_UNREGISTER:
1057 disable_trace_probe(tp, TP_FLAG_PROFILE);
1058 return 0;
1059 case TRACE_REG_PERF_OPEN:
1060 case TRACE_REG_PERF_CLOSE:
1061 case TRACE_REG_PERF_ADD:
1062 case TRACE_REG_PERF_DEL:
1063 return 0;
1064#endif
1065 }
1066 return 0;
1067}
1068
1069static __kprobes
1070int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1071{
1072 struct trace_probe *tp = container_of(kp, struct trace_probe, rp.kp);
1073
1074 if (tp->flags & TP_FLAG_TRACE)
1075 kprobe_trace_func(kp, regs);
1076#ifdef CONFIG_PERF_EVENTS
1077 if (tp->flags & TP_FLAG_PROFILE)
1078 kprobe_perf_func(kp, regs);
1079#endif
1080 return 0; /* We don't tweek kernel, so just return 0 */
1081}
1082
1083static __kprobes
1084int kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1085{
1086 struct trace_probe *tp = container_of(ri->rp, struct trace_probe, rp);
1087
1088 if (tp->flags & TP_FLAG_TRACE)
1089 kretprobe_trace_func(ri, regs);
1090#ifdef CONFIG_PERF_EVENTS
1091 if (tp->flags & TP_FLAG_PROFILE)
1092 kretprobe_perf_func(ri, regs);
1093#endif
1094 return 0; /* We don't tweek kernel, so just return 0 */
1095}
1096
1097static struct trace_event_functions kretprobe_funcs = {
1098 .trace = print_kretprobe_event
1099};
1100
1101static struct trace_event_functions kprobe_funcs = {
1102 .trace = print_kprobe_event
1103};
1104
1105static int register_probe_event(struct trace_probe *tp)
1106{
1107 struct ftrace_event_call *call = &tp->call;
1108 int ret;
1109
1110 /* Initialize ftrace_event_call */
1111 INIT_LIST_HEAD(&call->class->fields);
1112 if (trace_probe_is_return(tp)) {
1113 call->event.funcs = &kretprobe_funcs;
1114 call->class->define_fields = kretprobe_event_define_fields;
1115 } else {
1116 call->event.funcs = &kprobe_funcs;
1117 call->class->define_fields = kprobe_event_define_fields;
1118 }
1119 if (set_print_fmt(tp) < 0)
1120 return -ENOMEM;
1121 ret = register_ftrace_event(&call->event);
1122 if (!ret) {
1123 kfree(call->print_fmt);
1124 return -ENODEV;
1125 }
1126 call->flags = 0;
1127 call->class->reg = kprobe_register;
1128 call->data = tp;
1129 ret = trace_add_event_call(call);
1130 if (ret) {
1131 pr_info("Failed to register kprobe event: %s\n", call->name);
1132 kfree(call->print_fmt);
1133 unregister_ftrace_event(&call->event);
1134 }
1135 return ret;
1136}
1137
1138static void unregister_probe_event(struct trace_probe *tp)
1139{
1140 /* tp->event is unregistered in trace_remove_event_call() */
1141 trace_remove_event_call(&tp->call);
1142 kfree(tp->call.print_fmt);
1143}
1144
1145/* Make a debugfs interface for controlling probe points */
1146static __init int init_kprobe_trace(void)
1147{
1148 struct dentry *d_tracer;
1149 struct dentry *entry;
1150
1151 if (register_module_notifier(&trace_probe_module_nb))
1152 return -EINVAL;
1153
1154 d_tracer = tracing_init_dentry();
1155 if (!d_tracer)
1156 return 0;
1157
1158 entry = debugfs_create_file("kprobe_events", 0644, d_tracer,
1159 NULL, &kprobe_events_ops);
1160
1161 /* Event list interface */
1162 if (!entry)
1163 pr_warning("Could not create debugfs "
1164 "'kprobe_events' entry\n");
1165
1166 /* Profile interface */
1167 entry = debugfs_create_file("kprobe_profile", 0444, d_tracer,
1168 NULL, &kprobe_profile_ops);
1169
1170 if (!entry)
1171 pr_warning("Could not create debugfs "
1172 "'kprobe_profile' entry\n");
1173 return 0;
1174}
1175fs_initcall(init_kprobe_trace);
1176
1177
1178#ifdef CONFIG_FTRACE_STARTUP_TEST
1179
1180/*
1181 * The "__used" keeps gcc from removing the function symbol
1182 * from the kallsyms table.
1183 */
1184static __used int kprobe_trace_selftest_target(int a1, int a2, int a3,
1185 int a4, int a5, int a6)
1186{
1187 return a1 + a2 + a3 + a4 + a5 + a6;
1188}
1189
1190static __init int kprobe_trace_self_tests_init(void)
1191{
1192 int ret, warn = 0;
1193 int (*target)(int, int, int, int, int, int);
1194 struct trace_probe *tp;
1195
1196 target = kprobe_trace_selftest_target;
1197
1198 pr_info("Testing kprobe tracing: ");
1199
1200 ret = traceprobe_command("p:testprobe kprobe_trace_selftest_target "
1201 "$stack $stack0 +0($stack)",
1202 create_trace_probe);
1203 if (WARN_ON_ONCE(ret)) {
1204 pr_warning("error on probing function entry.\n");
1205 warn++;
1206 } else {
1207 /* Enable trace point */
1208 tp = find_trace_probe("testprobe", KPROBE_EVENT_SYSTEM);
1209 if (WARN_ON_ONCE(tp == NULL)) {
1210 pr_warning("error on getting new probe.\n");
1211 warn++;
1212 } else
1213 enable_trace_probe(tp, TP_FLAG_TRACE);
1214 }
1215
1216 ret = traceprobe_command("r:testprobe2 kprobe_trace_selftest_target "
1217 "$retval", create_trace_probe);
1218 if (WARN_ON_ONCE(ret)) {
1219 pr_warning("error on probing function return.\n");
1220 warn++;
1221 } else {
1222 /* Enable trace point */
1223 tp = find_trace_probe("testprobe2", KPROBE_EVENT_SYSTEM);
1224 if (WARN_ON_ONCE(tp == NULL)) {
1225 pr_warning("error on getting new probe.\n");
1226 warn++;
1227 } else
1228 enable_trace_probe(tp, TP_FLAG_TRACE);
1229 }
1230
1231 if (warn)
1232 goto end;
1233
1234 ret = target(1, 2, 3, 4, 5, 6);
1235
1236 /* Disable trace points before removing it */
1237 tp = find_trace_probe("testprobe", KPROBE_EVENT_SYSTEM);
1238 if (WARN_ON_ONCE(tp == NULL)) {
1239 pr_warning("error on getting test probe.\n");
1240 warn++;
1241 } else
1242 disable_trace_probe(tp, TP_FLAG_TRACE);
1243
1244 tp = find_trace_probe("testprobe2", KPROBE_EVENT_SYSTEM);
1245 if (WARN_ON_ONCE(tp == NULL)) {
1246 pr_warning("error on getting 2nd test probe.\n");
1247 warn++;
1248 } else
1249 disable_trace_probe(tp, TP_FLAG_TRACE);
1250
1251 ret = traceprobe_command("-:testprobe", create_trace_probe);
1252 if (WARN_ON_ONCE(ret)) {
1253 pr_warning("error on deleting a probe.\n");
1254 warn++;
1255 }
1256
1257 ret = traceprobe_command("-:testprobe2", create_trace_probe);
1258 if (WARN_ON_ONCE(ret)) {
1259 pr_warning("error on deleting a probe.\n");
1260 warn++;
1261 }
1262
1263end:
1264 release_all_trace_probes();
1265 if (warn)
1266 pr_cont("NG: Some tests are failed. Please check them.\n");
1267 else
1268 pr_cont("OK\n");
1269 return 0;
1270}
1271
1272late_initcall(kprobe_trace_self_tests_init);
1273
1274#endif