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/*
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