ARM: dts: r8a7790: Remove obsolete vsp1 properties
[linux/fpc-iii.git] / kernel / trace / trace_kprobe.c
blob9aedb0b06683765f4293eb750465ed5849c2902f
1 /*
2 * Kprobes-based tracing events
4 * Created by Masami Hiramatsu <mhiramat@redhat.com>
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.
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.
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
20 #include <linux/module.h>
21 #include <linux/uaccess.h>
23 #include "trace_probe.h"
25 #define KPROBE_EVENT_SYSTEM "kprobes"
27 /**
28 * Kprobe event core functions
30 struct 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;
38 #define SIZEOF_TRACE_KPROBE(n) \
39 (offsetof(struct trace_kprobe, tp.args) + \
40 (sizeof(struct probe_arg) * (n)))
43 static nokprobe_inline bool trace_kprobe_is_return(struct trace_kprobe *tk)
45 return tk->rp.handler != NULL;
48 static nokprobe_inline const char *trace_kprobe_symbol(struct trace_kprobe *tk)
50 return tk->symbol ? tk->symbol : "unknown";
53 static nokprobe_inline unsigned long trace_kprobe_offset(struct trace_kprobe *tk)
55 return tk->rp.kp.offset;
58 static nokprobe_inline bool trace_kprobe_has_gone(struct trace_kprobe *tk)
60 return !!(kprobe_gone(&tk->rp.kp));
63 static nokprobe_inline bool trace_kprobe_within_module(struct trace_kprobe *tk,
64 struct module *mod)
66 int len = strlen(mod->name);
67 const char *name = trace_kprobe_symbol(tk);
68 return strncmp(mod->name, name, len) == 0 && name[len] == ':';
71 static nokprobe_inline bool trace_kprobe_is_on_module(struct trace_kprobe *tk)
73 return !!strchr(trace_kprobe_symbol(tk), ':');
76 static int register_kprobe_event(struct trace_kprobe *tk);
77 static int unregister_kprobe_event(struct trace_kprobe *tk);
79 static DEFINE_MUTEX(probe_lock);
80 static LIST_HEAD(probe_list);
82 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
83 static int kretprobe_dispatcher(struct kretprobe_instance *ri,
84 struct pt_regs *regs);
86 /* Memory fetching by symbol */
87 struct symbol_cache {
88 char *symbol;
89 long offset;
90 unsigned long addr;
93 unsigned long update_symbol_cache(struct symbol_cache *sc)
95 sc->addr = (unsigned long)kallsyms_lookup_name(sc->symbol);
97 if (sc->addr)
98 sc->addr += sc->offset;
100 return sc->addr;
103 void free_symbol_cache(struct symbol_cache *sc)
105 kfree(sc->symbol);
106 kfree(sc);
109 struct symbol_cache *alloc_symbol_cache(const char *sym, long offset)
111 struct symbol_cache *sc;
113 if (!sym || strlen(sym) == 0)
114 return NULL;
116 sc = kzalloc(sizeof(struct symbol_cache), GFP_KERNEL);
117 if (!sc)
118 return NULL;
120 sc->symbol = kstrdup(sym, GFP_KERNEL);
121 if (!sc->symbol) {
122 kfree(sc);
123 return NULL;
125 sc->offset = offset;
126 update_symbol_cache(sc);
128 return sc;
132 * Kprobes-specific fetch functions
134 #define DEFINE_FETCH_stack(type) \
135 static void FETCH_FUNC_NAME(stack, type)(struct pt_regs *regs, \
136 void *offset, void *dest) \
138 *(type *)dest = (type)regs_get_kernel_stack_nth(regs, \
139 (unsigned int)((unsigned long)offset)); \
141 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(stack, type));
143 DEFINE_BASIC_FETCH_FUNCS(stack)
144 /* No string on the stack entry */
145 #define fetch_stack_string NULL
146 #define fetch_stack_string_size NULL
148 #define DEFINE_FETCH_memory(type) \
149 static void FETCH_FUNC_NAME(memory, type)(struct pt_regs *regs, \
150 void *addr, void *dest) \
152 type retval; \
153 if (probe_kernel_address(addr, retval)) \
154 *(type *)dest = 0; \
155 else \
156 *(type *)dest = retval; \
158 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(memory, type));
160 DEFINE_BASIC_FETCH_FUNCS(memory)
162 * Fetch a null-terminated string. Caller MUST set *(u32 *)dest with max
163 * length and relative data location.
165 static void FETCH_FUNC_NAME(memory, string)(struct pt_regs *regs,
166 void *addr, void *dest)
168 int maxlen = get_rloc_len(*(u32 *)dest);
169 u8 *dst = get_rloc_data(dest);
170 long ret;
172 if (!maxlen)
173 return;
176 * Try to get string again, since the string can be changed while
177 * probing.
179 ret = strncpy_from_unsafe(dst, addr, maxlen);
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));
188 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(memory, string));
190 /* Return the length of string -- including null terminal byte */
191 static void FETCH_FUNC_NAME(memory, string_size)(struct pt_regs *regs,
192 void *addr, void *dest)
194 mm_segment_t old_fs;
195 int ret, len = 0;
196 u8 c;
198 old_fs = get_fs();
199 set_fs(KERNEL_DS);
200 pagefault_disable();
202 do {
203 ret = __copy_from_user_inatomic(&c, (u8 *)addr + len, 1);
204 len++;
205 } while (c && ret == 0 && len < MAX_STRING_SIZE);
207 pagefault_enable();
208 set_fs(old_fs);
210 if (ret < 0) /* Failed to check the length */
211 *(u32 *)dest = 0;
212 else
213 *(u32 *)dest = len;
215 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(memory, string_size));
217 #define DEFINE_FETCH_symbol(type) \
218 void FETCH_FUNC_NAME(symbol, type)(struct pt_regs *regs, void *data, void *dest)\
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; \
226 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(symbol, type));
228 DEFINE_BASIC_FETCH_FUNCS(symbol)
229 DEFINE_FETCH_symbol(string)
230 DEFINE_FETCH_symbol(string_size)
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
240 /* Fetch type information table */
241 static 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),
257 ASSIGN_FETCH_TYPE_END
261 * Allocate new trace_probe and initialize it (including kprobes).
263 static 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)
270 struct trace_kprobe *tk;
271 int ret = -ENOMEM;
273 tk = kzalloc(SIZEOF_TRACE_KPROBE(nargs), GFP_KERNEL);
274 if (!tk)
275 return ERR_PTR(ret);
277 tk->nhit = alloc_percpu(unsigned long);
278 if (!tk->nhit)
279 goto error;
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;
290 if (is_return)
291 tk->rp.handler = kretprobe_dispatcher;
292 else
293 tk->rp.kp.pre_handler = kprobe_dispatcher;
295 if (!event || !is_good_name(event)) {
296 ret = -EINVAL;
297 goto error;
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;
305 if (!group || !is_good_name(group)) {
306 ret = -EINVAL;
307 goto error;
310 tk->tp.class.system = kstrdup(group, GFP_KERNEL);
311 if (!tk->tp.class.system)
312 goto error;
314 INIT_LIST_HEAD(&tk->list);
315 INIT_LIST_HEAD(&tk->tp.files);
316 return tk;
317 error:
318 kfree(tk->tp.call.name);
319 kfree(tk->symbol);
320 free_percpu(tk->nhit);
321 kfree(tk);
322 return ERR_PTR(ret);
325 static void free_trace_kprobe(struct trace_kprobe *tk)
327 int i;
329 for (i = 0; i < tk->tp.nr_args; i++)
330 traceprobe_free_probe_arg(&tk->tp.args[i]);
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);
339 static struct trace_kprobe *find_trace_kprobe(const char *event,
340 const char *group)
342 struct trace_kprobe *tk;
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;
352 * Enable trace_probe
353 * if the file is NULL, enable "perf" handler, or enable "trace" handler.
355 static int
356 enable_trace_kprobe(struct trace_kprobe *tk, struct trace_event_file *file)
358 int ret = 0;
360 if (file) {
361 struct event_file_link *link;
363 link = kmalloc(sizeof(*link), GFP_KERNEL);
364 if (!link) {
365 ret = -ENOMEM;
366 goto out;
369 link->file = file;
370 list_add_tail_rcu(&link->list, &tk->tp.files);
372 tk->tp.flags |= TP_FLAG_TRACE;
373 } else
374 tk->tp.flags |= TP_FLAG_PROFILE;
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);
382 out:
383 return ret;
387 * Disable trace_probe
388 * if the file is NULL, disable "perf" handler, or disable "trace" handler.
390 static int
391 disable_trace_kprobe(struct trace_kprobe *tk, struct trace_event_file *file)
393 struct event_file_link *link = NULL;
394 int wait = 0;
395 int ret = 0;
397 if (file) {
398 link = find_event_file_link(&tk->tp, file);
399 if (!link) {
400 ret = -EINVAL;
401 goto out;
404 list_del_rcu(&link->list);
405 wait = 1;
406 if (!list_empty(&tk->tp.files))
407 goto out;
409 tk->tp.flags &= ~TP_FLAG_TRACE;
410 } else
411 tk->tp.flags &= ~TP_FLAG_PROFILE;
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;
420 out:
421 if (wait) {
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.
430 synchronize_sched();
431 kfree(link); /* Ignored if link == NULL */
434 return ret;
437 /* Internal register function - just handle k*probes and flags */
438 static int __register_trace_kprobe(struct trace_kprobe *tk)
440 int i, ret;
442 if (trace_probe_is_registered(&tk->tp))
443 return -EINVAL;
445 for (i = 0; i < tk->tp.nr_args; i++)
446 traceprobe_update_arg(&tk->tp.args[i]);
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;
454 if (trace_kprobe_is_return(tk))
455 ret = register_kretprobe(&tk->rp);
456 else
457 ret = register_kprobe(&tk->rp.kp);
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;
474 return ret;
477 /* Internal unregister function - just handle k*probes and flags */
478 static void __unregister_trace_kprobe(struct trace_kprobe *tk)
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;
492 /* Unregister a trace_probe and probe_event: call with locking probe_lock */
493 static int unregister_trace_kprobe(struct trace_kprobe *tk)
495 /* Enabled event can not be unregistered */
496 if (trace_probe_is_enabled(&tk->tp))
497 return -EBUSY;
499 /* Will fail if probe is being used by ftrace or perf */
500 if (unregister_kprobe_event(tk))
501 return -EBUSY;
503 __unregister_trace_kprobe(tk);
504 list_del(&tk->list);
506 return 0;
509 /* Register a trace_probe and probe_event */
510 static int register_trace_kprobe(struct trace_kprobe *tk)
512 struct trace_kprobe *old_tk;
513 int ret;
515 mutex_lock(&probe_lock);
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);
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;
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);
541 end:
542 mutex_unlock(&probe_lock);
543 return ret;
546 /* Module notifier call back, checking event on the module */
547 static int trace_kprobe_module_callback(struct notifier_block *nb,
548 unsigned long val, void *data)
550 struct module *mod = data;
551 struct trace_kprobe *tk;
552 int ret;
554 if (val != MODULE_STATE_COMING)
555 return NOTIFY_DONE;
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);
570 mutex_unlock(&probe_lock);
572 return NOTIFY_DONE;
575 static struct notifier_block trace_kprobe_module_nb = {
576 .notifier_call = trace_kprobe_module_callback,
577 .priority = 1 /* Invoked after kprobe module callback */
580 static int create_trace_kprobe(int argc, char **argv)
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 * $comm : fetch current task comm
591 * @ADDR : fetch memory at ADDR (ADDR should be in kernel)
592 * @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
593 * %REG : fetch register REG
594 * Dereferencing memory fetch:
595 * +|-offs(ARG) : fetch memory at ARG +|- offs address.
596 * Alias name of args:
597 * NAME=FETCHARG : set NAME as alias of FETCHARG.
598 * Type of args:
599 * FETCHARG:TYPE : use TYPE instead of unsigned long.
601 struct trace_kprobe *tk;
602 int i, ret = 0;
603 bool is_return = false, is_delete = false;
604 char *symbol = NULL, *event = NULL, *group = NULL;
605 char *arg;
606 unsigned long offset = 0;
607 void *addr = NULL;
608 char buf[MAX_EVENT_NAME_LEN];
610 /* argc must be >= 1 */
611 if (argv[0][0] == 'p')
612 is_return = false;
613 else if (argv[0][0] == 'r')
614 is_return = true;
615 else if (argv[0][0] == '-')
616 is_delete = true;
617 else {
618 pr_info("Probe definition must be started with 'p', 'r' or"
619 " '-'.\n");
620 return -EINVAL;
623 if (argv[0][1] == ':') {
624 event = &argv[0][2];
625 if (strchr(event, '/')) {
626 group = event;
627 event = strchr(group, '/') + 1;
628 event[-1] = '\0';
629 if (strlen(group) == 0) {
630 pr_info("Group name is not specified\n");
631 return -EINVAL;
634 if (strlen(event) == 0) {
635 pr_info("Event name is not specified\n");
636 return -EINVAL;
639 if (!group)
640 group = KPROBE_EVENT_SYSTEM;
642 if (is_delete) {
643 if (!event) {
644 pr_info("Delete command needs an event name.\n");
645 return -EINVAL;
647 mutex_lock(&probe_lock);
648 tk = find_trace_kprobe(event, group);
649 if (!tk) {
650 mutex_unlock(&probe_lock);
651 pr_info("Event %s/%s doesn't exist.\n", group, event);
652 return -ENOENT;
654 /* delete an event */
655 ret = unregister_trace_kprobe(tk);
656 if (ret == 0)
657 free_trace_kprobe(tk);
658 mutex_unlock(&probe_lock);
659 return ret;
662 if (argc < 2) {
663 pr_info("Probe point is not specified.\n");
664 return -EINVAL;
666 if (isdigit(argv[1][0])) {
667 if (is_return) {
668 pr_info("Return probe point must be a symbol.\n");
669 return -EINVAL;
671 /* an address specified */
672 ret = kstrtoul(&argv[1][0], 0, (unsigned long *)&addr);
673 if (ret) {
674 pr_info("Failed to parse address.\n");
675 return ret;
677 } else {
678 /* a symbol specified */
679 symbol = argv[1];
680 /* TODO: support .init module functions */
681 ret = traceprobe_split_symbol_offset(symbol, &offset);
682 if (ret) {
683 pr_info("Failed to parse symbol.\n");
684 return ret;
686 if (offset && is_return) {
687 pr_info("Return probe must be used without offset.\n");
688 return -EINVAL;
691 argc -= 2; argv += 2;
693 /* setup a probe */
694 if (!event) {
695 /* Make a new event name */
696 if (symbol)
697 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
698 is_return ? 'r' : 'p', symbol, offset);
699 else
700 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
701 is_return ? 'r' : 'p', addr);
702 event = buf;
704 tk = alloc_trace_kprobe(group, event, addr, symbol, offset, argc,
705 is_return);
706 if (IS_ERR(tk)) {
707 pr_info("Failed to allocate trace_probe.(%d)\n",
708 (int)PTR_ERR(tk));
709 return PTR_ERR(tk);
712 /* parse arguments */
713 ret = 0;
714 for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
715 struct probe_arg *parg = &tk->tp.args[i];
717 /* Increment count for freeing args in error case */
718 tk->tp.nr_args++;
720 /* Parse argument name */
721 arg = strchr(argv[i], '=');
722 if (arg) {
723 *arg++ = '\0';
724 parg->name = kstrdup(argv[i], GFP_KERNEL);
725 } else {
726 arg = argv[i];
727 /* If argument name is omitted, set "argN" */
728 snprintf(buf, MAX_EVENT_NAME_LEN, "arg%d", i + 1);
729 parg->name = kstrdup(buf, GFP_KERNEL);
732 if (!parg->name) {
733 pr_info("Failed to allocate argument[%d] name.\n", i);
734 ret = -ENOMEM;
735 goto error;
738 if (!is_good_name(parg->name)) {
739 pr_info("Invalid argument[%d] name: %s\n",
740 i, parg->name);
741 ret = -EINVAL;
742 goto error;
745 if (traceprobe_conflict_field_name(parg->name,
746 tk->tp.args, i)) {
747 pr_info("Argument[%d] name '%s' conflicts with "
748 "another field.\n", i, argv[i]);
749 ret = -EINVAL;
750 goto error;
753 /* Parse fetch argument */
754 ret = traceprobe_parse_probe_arg(arg, &tk->tp.size, parg,
755 is_return, true,
756 kprobes_fetch_type_table);
757 if (ret) {
758 pr_info("Parse error at argument[%d]. (%d)\n", i, ret);
759 goto error;
763 ret = register_trace_kprobe(tk);
764 if (ret)
765 goto error;
766 return 0;
768 error:
769 free_trace_kprobe(tk);
770 return ret;
773 static int release_all_trace_kprobes(void)
775 struct trace_kprobe *tk;
776 int ret = 0;
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;
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);
794 end:
795 mutex_unlock(&probe_lock);
797 return ret;
800 /* Probes listing interfaces */
801 static void *probes_seq_start(struct seq_file *m, loff_t *pos)
803 mutex_lock(&probe_lock);
804 return seq_list_start(&probe_list, *pos);
807 static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
809 return seq_list_next(v, &probe_list, pos);
812 static void probes_seq_stop(struct seq_file *m, void *v)
814 mutex_unlock(&probe_lock);
817 static int probes_seq_show(struct seq_file *m, void *v)
819 struct trace_kprobe *tk = v;
820 int i;
822 seq_putc(m, trace_kprobe_is_return(tk) ? 'r' : 'p');
823 seq_printf(m, ":%s/%s", tk->tp.call.class->system,
824 trace_event_name(&tk->tp.call));
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));
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_putc(m, '\n');
838 return 0;
841 static 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
848 static int probes_open(struct inode *inode, struct file *file)
850 int ret;
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;
858 return seq_open(file, &probes_seq_op);
861 static ssize_t probes_write(struct file *file, const char __user *buffer,
862 size_t count, loff_t *ppos)
864 return traceprobe_probes_write(file, buffer, count, ppos,
865 create_trace_kprobe);
868 static 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,
877 /* Probes profiling interfaces */
878 static int probes_profile_seq_show(struct seq_file *m, void *v)
880 struct trace_kprobe *tk = v;
881 unsigned long nhit = 0;
882 int cpu;
884 for_each_possible_cpu(cpu)
885 nhit += *per_cpu_ptr(tk->nhit, cpu);
887 seq_printf(m, " %-44s %15lu %15lu\n",
888 trace_event_name(&tk->tp.call), nhit,
889 tk->rp.kp.nmissed);
891 return 0;
894 static const struct seq_operations profile_seq_op = {
895 .start = probes_seq_start,
896 .next = probes_seq_next,
897 .stop = probes_seq_stop,
898 .show = probes_profile_seq_show
901 static int profile_open(struct inode *inode, struct file *file)
903 return seq_open(file, &profile_seq_op);
906 static const struct file_operations kprobe_profile_ops = {
907 .owner = THIS_MODULE,
908 .open = profile_open,
909 .read = seq_read,
910 .llseek = seq_lseek,
911 .release = seq_release,
914 /* Kprobe handler */
915 static nokprobe_inline void
916 __kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs,
917 struct trace_event_file *trace_file)
919 struct kprobe_trace_entry_head *entry;
920 struct ring_buffer_event *event;
921 struct ring_buffer *buffer;
922 int size, dsize, pc;
923 unsigned long irq_flags;
924 struct trace_event_call *call = &tk->tp.call;
926 WARN_ON(call != trace_file->event_call);
928 if (trace_trigger_soft_disabled(trace_file))
929 return;
931 local_save_flags(irq_flags);
932 pc = preempt_count();
934 dsize = __get_data_size(&tk->tp, regs);
935 size = sizeof(*entry) + tk->tp.size + dsize;
937 event = trace_event_buffer_lock_reserve(&buffer, trace_file,
938 call->event.type,
939 size, irq_flags, pc);
940 if (!event)
941 return;
943 entry = ring_buffer_event_data(event);
944 entry->ip = (unsigned long)tk->rp.kp.addr;
945 store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
947 event_trigger_unlock_commit_regs(trace_file, buffer, event,
948 entry, irq_flags, pc, regs);
951 static void
952 kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs)
954 struct event_file_link *link;
956 list_for_each_entry_rcu(link, &tk->tp.files, list)
957 __kprobe_trace_func(tk, regs, link->file);
959 NOKPROBE_SYMBOL(kprobe_trace_func);
961 /* Kretprobe handler */
962 static nokprobe_inline void
963 __kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
964 struct pt_regs *regs,
965 struct trace_event_file *trace_file)
967 struct kretprobe_trace_entry_head *entry;
968 struct ring_buffer_event *event;
969 struct ring_buffer *buffer;
970 int size, pc, dsize;
971 unsigned long irq_flags;
972 struct trace_event_call *call = &tk->tp.call;
974 WARN_ON(call != trace_file->event_call);
976 if (trace_trigger_soft_disabled(trace_file))
977 return;
979 local_save_flags(irq_flags);
980 pc = preempt_count();
982 dsize = __get_data_size(&tk->tp, regs);
983 size = sizeof(*entry) + tk->tp.size + dsize;
985 event = trace_event_buffer_lock_reserve(&buffer, trace_file,
986 call->event.type,
987 size, irq_flags, pc);
988 if (!event)
989 return;
991 entry = ring_buffer_event_data(event);
992 entry->func = (unsigned long)tk->rp.kp.addr;
993 entry->ret_ip = (unsigned long)ri->ret_addr;
994 store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
996 event_trigger_unlock_commit_regs(trace_file, buffer, event,
997 entry, irq_flags, pc, regs);
1000 static void
1001 kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1002 struct pt_regs *regs)
1004 struct event_file_link *link;
1006 list_for_each_entry_rcu(link, &tk->tp.files, list)
1007 __kretprobe_trace_func(tk, ri, regs, link->file);
1009 NOKPROBE_SYMBOL(kretprobe_trace_func);
1011 /* Event entry printers */
1012 static enum print_line_t
1013 print_kprobe_event(struct trace_iterator *iter, int flags,
1014 struct trace_event *event)
1016 struct kprobe_trace_entry_head *field;
1017 struct trace_seq *s = &iter->seq;
1018 struct trace_probe *tp;
1019 u8 *data;
1020 int i;
1022 field = (struct kprobe_trace_entry_head *)iter->ent;
1023 tp = container_of(event, struct trace_probe, call.event);
1025 trace_seq_printf(s, "%s: (", trace_event_name(&tp->call));
1027 if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
1028 goto out;
1030 trace_seq_putc(s, ')');
1032 data = (u8 *)&field[1];
1033 for (i = 0; i < tp->nr_args; i++)
1034 if (!tp->args[i].type->print(s, tp->args[i].name,
1035 data + tp->args[i].offset, field))
1036 goto out;
1038 trace_seq_putc(s, '\n');
1039 out:
1040 return trace_handle_return(s);
1043 static enum print_line_t
1044 print_kretprobe_event(struct trace_iterator *iter, int flags,
1045 struct trace_event *event)
1047 struct kretprobe_trace_entry_head *field;
1048 struct trace_seq *s = &iter->seq;
1049 struct trace_probe *tp;
1050 u8 *data;
1051 int i;
1053 field = (struct kretprobe_trace_entry_head *)iter->ent;
1054 tp = container_of(event, struct trace_probe, call.event);
1056 trace_seq_printf(s, "%s: (", trace_event_name(&tp->call));
1058 if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1059 goto out;
1061 trace_seq_puts(s, " <- ");
1063 if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1064 goto out;
1066 trace_seq_putc(s, ')');
1068 data = (u8 *)&field[1];
1069 for (i = 0; i < tp->nr_args; i++)
1070 if (!tp->args[i].type->print(s, tp->args[i].name,
1071 data + tp->args[i].offset, field))
1072 goto out;
1074 trace_seq_putc(s, '\n');
1076 out:
1077 return trace_handle_return(s);
1081 static int kprobe_event_define_fields(struct trace_event_call *event_call)
1083 int ret, i;
1084 struct kprobe_trace_entry_head field;
1085 struct trace_kprobe *tk = (struct trace_kprobe *)event_call->data;
1087 DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1088 /* Set argument names as fields */
1089 for (i = 0; i < tk->tp.nr_args; i++) {
1090 struct probe_arg *parg = &tk->tp.args[i];
1092 ret = trace_define_field(event_call, parg->type->fmttype,
1093 parg->name,
1094 sizeof(field) + parg->offset,
1095 parg->type->size,
1096 parg->type->is_signed,
1097 FILTER_OTHER);
1098 if (ret)
1099 return ret;
1101 return 0;
1104 static int kretprobe_event_define_fields(struct trace_event_call *event_call)
1106 int ret, i;
1107 struct kretprobe_trace_entry_head field;
1108 struct trace_kprobe *tk = (struct trace_kprobe *)event_call->data;
1110 DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1111 DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1112 /* Set argument names as fields */
1113 for (i = 0; i < tk->tp.nr_args; i++) {
1114 struct probe_arg *parg = &tk->tp.args[i];
1116 ret = trace_define_field(event_call, parg->type->fmttype,
1117 parg->name,
1118 sizeof(field) + parg->offset,
1119 parg->type->size,
1120 parg->type->is_signed,
1121 FILTER_OTHER);
1122 if (ret)
1123 return ret;
1125 return 0;
1128 #ifdef CONFIG_PERF_EVENTS
1130 /* Kprobe profile handler */
1131 static void
1132 kprobe_perf_func(struct trace_kprobe *tk, struct pt_regs *regs)
1134 struct trace_event_call *call = &tk->tp.call;
1135 struct bpf_prog *prog = call->prog;
1136 struct kprobe_trace_entry_head *entry;
1137 struct hlist_head *head;
1138 int size, __size, dsize;
1139 int rctx;
1141 if (prog && !trace_call_bpf(prog, regs))
1142 return;
1144 head = this_cpu_ptr(call->perf_events);
1145 if (hlist_empty(head))
1146 return;
1148 dsize = __get_data_size(&tk->tp, regs);
1149 __size = sizeof(*entry) + tk->tp.size + dsize;
1150 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1151 size -= sizeof(u32);
1153 entry = perf_trace_buf_alloc(size, NULL, &rctx);
1154 if (!entry)
1155 return;
1157 entry->ip = (unsigned long)tk->rp.kp.addr;
1158 memset(&entry[1], 0, dsize);
1159 store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1160 perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1161 head, NULL);
1163 NOKPROBE_SYMBOL(kprobe_perf_func);
1165 /* Kretprobe profile handler */
1166 static void
1167 kretprobe_perf_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1168 struct pt_regs *regs)
1170 struct trace_event_call *call = &tk->tp.call;
1171 struct bpf_prog *prog = call->prog;
1172 struct kretprobe_trace_entry_head *entry;
1173 struct hlist_head *head;
1174 int size, __size, dsize;
1175 int rctx;
1177 if (prog && !trace_call_bpf(prog, regs))
1178 return;
1180 head = this_cpu_ptr(call->perf_events);
1181 if (hlist_empty(head))
1182 return;
1184 dsize = __get_data_size(&tk->tp, regs);
1185 __size = sizeof(*entry) + tk->tp.size + dsize;
1186 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1187 size -= sizeof(u32);
1189 entry = perf_trace_buf_alloc(size, NULL, &rctx);
1190 if (!entry)
1191 return;
1193 entry->func = (unsigned long)tk->rp.kp.addr;
1194 entry->ret_ip = (unsigned long)ri->ret_addr;
1195 store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1196 perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1197 head, NULL);
1199 NOKPROBE_SYMBOL(kretprobe_perf_func);
1200 #endif /* CONFIG_PERF_EVENTS */
1203 * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex.
1205 * kprobe_trace_self_tests_init() does enable_trace_probe/disable_trace_probe
1206 * lockless, but we can't race with this __init function.
1208 static int kprobe_register(struct trace_event_call *event,
1209 enum trace_reg type, void *data)
1211 struct trace_kprobe *tk = (struct trace_kprobe *)event->data;
1212 struct trace_event_file *file = data;
1214 switch (type) {
1215 case TRACE_REG_REGISTER:
1216 return enable_trace_kprobe(tk, file);
1217 case TRACE_REG_UNREGISTER:
1218 return disable_trace_kprobe(tk, file);
1220 #ifdef CONFIG_PERF_EVENTS
1221 case TRACE_REG_PERF_REGISTER:
1222 return enable_trace_kprobe(tk, NULL);
1223 case TRACE_REG_PERF_UNREGISTER:
1224 return disable_trace_kprobe(tk, NULL);
1225 case TRACE_REG_PERF_OPEN:
1226 case TRACE_REG_PERF_CLOSE:
1227 case TRACE_REG_PERF_ADD:
1228 case TRACE_REG_PERF_DEL:
1229 return 0;
1230 #endif
1232 return 0;
1235 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1237 struct trace_kprobe *tk = container_of(kp, struct trace_kprobe, rp.kp);
1239 raw_cpu_inc(*tk->nhit);
1241 if (tk->tp.flags & TP_FLAG_TRACE)
1242 kprobe_trace_func(tk, regs);
1243 #ifdef CONFIG_PERF_EVENTS
1244 if (tk->tp.flags & TP_FLAG_PROFILE)
1245 kprobe_perf_func(tk, regs);
1246 #endif
1247 return 0; /* We don't tweek kernel, so just return 0 */
1249 NOKPROBE_SYMBOL(kprobe_dispatcher);
1251 static int
1252 kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1254 struct trace_kprobe *tk = container_of(ri->rp, struct trace_kprobe, rp);
1256 raw_cpu_inc(*tk->nhit);
1258 if (tk->tp.flags & TP_FLAG_TRACE)
1259 kretprobe_trace_func(tk, ri, regs);
1260 #ifdef CONFIG_PERF_EVENTS
1261 if (tk->tp.flags & TP_FLAG_PROFILE)
1262 kretprobe_perf_func(tk, ri, regs);
1263 #endif
1264 return 0; /* We don't tweek kernel, so just return 0 */
1266 NOKPROBE_SYMBOL(kretprobe_dispatcher);
1268 static struct trace_event_functions kretprobe_funcs = {
1269 .trace = print_kretprobe_event
1272 static struct trace_event_functions kprobe_funcs = {
1273 .trace = print_kprobe_event
1276 static int register_kprobe_event(struct trace_kprobe *tk)
1278 struct trace_event_call *call = &tk->tp.call;
1279 int ret;
1281 /* Initialize trace_event_call */
1282 INIT_LIST_HEAD(&call->class->fields);
1283 if (trace_kprobe_is_return(tk)) {
1284 call->event.funcs = &kretprobe_funcs;
1285 call->class->define_fields = kretprobe_event_define_fields;
1286 } else {
1287 call->event.funcs = &kprobe_funcs;
1288 call->class->define_fields = kprobe_event_define_fields;
1290 if (set_print_fmt(&tk->tp, trace_kprobe_is_return(tk)) < 0)
1291 return -ENOMEM;
1292 ret = register_trace_event(&call->event);
1293 if (!ret) {
1294 kfree(call->print_fmt);
1295 return -ENODEV;
1297 call->flags = TRACE_EVENT_FL_KPROBE;
1298 call->class->reg = kprobe_register;
1299 call->data = tk;
1300 ret = trace_add_event_call(call);
1301 if (ret) {
1302 pr_info("Failed to register kprobe event: %s\n",
1303 trace_event_name(call));
1304 kfree(call->print_fmt);
1305 unregister_trace_event(&call->event);
1307 return ret;
1310 static int unregister_kprobe_event(struct trace_kprobe *tk)
1312 int ret;
1314 /* tp->event is unregistered in trace_remove_event_call() */
1315 ret = trace_remove_event_call(&tk->tp.call);
1316 if (!ret)
1317 kfree(tk->tp.call.print_fmt);
1318 return ret;
1321 /* Make a tracefs interface for controlling probe points */
1322 static __init int init_kprobe_trace(void)
1324 struct dentry *d_tracer;
1325 struct dentry *entry;
1327 if (register_module_notifier(&trace_kprobe_module_nb))
1328 return -EINVAL;
1330 d_tracer = tracing_init_dentry();
1331 if (IS_ERR(d_tracer))
1332 return 0;
1334 entry = tracefs_create_file("kprobe_events", 0644, d_tracer,
1335 NULL, &kprobe_events_ops);
1337 /* Event list interface */
1338 if (!entry)
1339 pr_warn("Could not create tracefs 'kprobe_events' entry\n");
1341 /* Profile interface */
1342 entry = tracefs_create_file("kprobe_profile", 0444, d_tracer,
1343 NULL, &kprobe_profile_ops);
1345 if (!entry)
1346 pr_warn("Could not create tracefs 'kprobe_profile' entry\n");
1347 return 0;
1349 fs_initcall(init_kprobe_trace);
1352 #ifdef CONFIG_FTRACE_STARTUP_TEST
1355 * The "__used" keeps gcc from removing the function symbol
1356 * from the kallsyms table.
1358 static __used int kprobe_trace_selftest_target(int a1, int a2, int a3,
1359 int a4, int a5, int a6)
1361 return a1 + a2 + a3 + a4 + a5 + a6;
1364 static struct trace_event_file *
1365 find_trace_probe_file(struct trace_kprobe *tk, struct trace_array *tr)
1367 struct trace_event_file *file;
1369 list_for_each_entry(file, &tr->events, list)
1370 if (file->event_call == &tk->tp.call)
1371 return file;
1373 return NULL;
1377 * Nobody but us can call enable_trace_kprobe/disable_trace_kprobe at this
1378 * stage, we can do this lockless.
1380 static __init int kprobe_trace_self_tests_init(void)
1382 int ret, warn = 0;
1383 int (*target)(int, int, int, int, int, int);
1384 struct trace_kprobe *tk;
1385 struct trace_event_file *file;
1387 if (tracing_is_disabled())
1388 return -ENODEV;
1390 target = kprobe_trace_selftest_target;
1392 pr_info("Testing kprobe tracing: ");
1394 ret = traceprobe_command("p:testprobe kprobe_trace_selftest_target "
1395 "$stack $stack0 +0($stack)",
1396 create_trace_kprobe);
1397 if (WARN_ON_ONCE(ret)) {
1398 pr_warn("error on probing function entry.\n");
1399 warn++;
1400 } else {
1401 /* Enable trace point */
1402 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1403 if (WARN_ON_ONCE(tk == NULL)) {
1404 pr_warn("error on getting new probe.\n");
1405 warn++;
1406 } else {
1407 file = find_trace_probe_file(tk, top_trace_array());
1408 if (WARN_ON_ONCE(file == NULL)) {
1409 pr_warn("error on getting probe file.\n");
1410 warn++;
1411 } else
1412 enable_trace_kprobe(tk, file);
1416 ret = traceprobe_command("r:testprobe2 kprobe_trace_selftest_target "
1417 "$retval", create_trace_kprobe);
1418 if (WARN_ON_ONCE(ret)) {
1419 pr_warn("error on probing function return.\n");
1420 warn++;
1421 } else {
1422 /* Enable trace point */
1423 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1424 if (WARN_ON_ONCE(tk == NULL)) {
1425 pr_warn("error on getting 2nd new probe.\n");
1426 warn++;
1427 } else {
1428 file = find_trace_probe_file(tk, top_trace_array());
1429 if (WARN_ON_ONCE(file == NULL)) {
1430 pr_warn("error on getting probe file.\n");
1431 warn++;
1432 } else
1433 enable_trace_kprobe(tk, file);
1437 if (warn)
1438 goto end;
1440 ret = target(1, 2, 3, 4, 5, 6);
1442 /* Disable trace points before removing it */
1443 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1444 if (WARN_ON_ONCE(tk == NULL)) {
1445 pr_warn("error on getting test probe.\n");
1446 warn++;
1447 } else {
1448 file = find_trace_probe_file(tk, top_trace_array());
1449 if (WARN_ON_ONCE(file == NULL)) {
1450 pr_warn("error on getting probe file.\n");
1451 warn++;
1452 } else
1453 disable_trace_kprobe(tk, file);
1456 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1457 if (WARN_ON_ONCE(tk == NULL)) {
1458 pr_warn("error on getting 2nd test probe.\n");
1459 warn++;
1460 } else {
1461 file = find_trace_probe_file(tk, top_trace_array());
1462 if (WARN_ON_ONCE(file == NULL)) {
1463 pr_warn("error on getting probe file.\n");
1464 warn++;
1465 } else
1466 disable_trace_kprobe(tk, file);
1469 ret = traceprobe_command("-:testprobe", create_trace_kprobe);
1470 if (WARN_ON_ONCE(ret)) {
1471 pr_warn("error on deleting a probe.\n");
1472 warn++;
1475 ret = traceprobe_command("-:testprobe2", create_trace_kprobe);
1476 if (WARN_ON_ONCE(ret)) {
1477 pr_warn("error on deleting a probe.\n");
1478 warn++;
1481 end:
1482 release_all_trace_kprobes();
1483 if (warn)
1484 pr_cont("NG: Some tests are failed. Please check them.\n");
1485 else
1486 pr_cont("OK\n");
1487 return 0;
1490 late_initcall(kprobe_trace_self_tests_init);
1492 #endif