libceph: clear r_req_lru_item in __unregister_linger_request()
[linux/fpc-iii.git] / kernel / trace / trace_kprobe.c
blob282f6e4e553988bb6b75a974bb62742d0a80a7e2
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 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 long ret;
169 int maxlen = get_rloc_len(*(u32 *)dest);
170 u8 *dst = get_rloc_data(dest);
171 u8 *src = addr;
172 mm_segment_t old_fs = get_fs();
174 if (!maxlen)
175 return;
178 * Try to get string again, since the string can be changed while
179 * probing.
181 set_fs(KERNEL_DS);
182 pagefault_disable();
185 ret = __copy_from_user_inatomic(dst++, src++, 1);
186 while (dst[-1] && ret == 0 && src - (u8 *)addr < maxlen);
188 dst[-1] = '\0';
189 pagefault_enable();
190 set_fs(old_fs);
192 if (ret < 0) { /* Failed to fetch string */
193 ((u8 *)get_rloc_data(dest))[0] = '\0';
194 *(u32 *)dest = make_data_rloc(0, get_rloc_offs(*(u32 *)dest));
195 } else {
196 *(u32 *)dest = make_data_rloc(src - (u8 *)addr,
197 get_rloc_offs(*(u32 *)dest));
200 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(memory, string));
202 /* Return the length of string -- including null terminal byte */
203 static void FETCH_FUNC_NAME(memory, string_size)(struct pt_regs *regs,
204 void *addr, void *dest)
206 mm_segment_t old_fs;
207 int ret, len = 0;
208 u8 c;
210 old_fs = get_fs();
211 set_fs(KERNEL_DS);
212 pagefault_disable();
214 do {
215 ret = __copy_from_user_inatomic(&c, (u8 *)addr + len, 1);
216 len++;
217 } while (c && ret == 0 && len < MAX_STRING_SIZE);
219 pagefault_enable();
220 set_fs(old_fs);
222 if (ret < 0) /* Failed to check the length */
223 *(u32 *)dest = 0;
224 else
225 *(u32 *)dest = len;
227 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(memory, string_size));
229 #define DEFINE_FETCH_symbol(type) \
230 void FETCH_FUNC_NAME(symbol, type)(struct pt_regs *regs, void *data, void *dest)\
232 struct symbol_cache *sc = data; \
233 if (sc->addr) \
234 fetch_memory_##type(regs, (void *)sc->addr, dest); \
235 else \
236 *(type *)dest = 0; \
238 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(symbol, type));
240 DEFINE_BASIC_FETCH_FUNCS(symbol)
241 DEFINE_FETCH_symbol(string)
242 DEFINE_FETCH_symbol(string_size)
244 /* kprobes don't support file_offset fetch methods */
245 #define fetch_file_offset_u8 NULL
246 #define fetch_file_offset_u16 NULL
247 #define fetch_file_offset_u32 NULL
248 #define fetch_file_offset_u64 NULL
249 #define fetch_file_offset_string NULL
250 #define fetch_file_offset_string_size NULL
252 /* Fetch type information table */
253 const struct fetch_type kprobes_fetch_type_table[] = {
254 /* Special types */
255 [FETCH_TYPE_STRING] = __ASSIGN_FETCH_TYPE("string", string, string,
256 sizeof(u32), 1, "__data_loc char[]"),
257 [FETCH_TYPE_STRSIZE] = __ASSIGN_FETCH_TYPE("string_size", u32,
258 string_size, sizeof(u32), 0, "u32"),
259 /* Basic types */
260 ASSIGN_FETCH_TYPE(u8, u8, 0),
261 ASSIGN_FETCH_TYPE(u16, u16, 0),
262 ASSIGN_FETCH_TYPE(u32, u32, 0),
263 ASSIGN_FETCH_TYPE(u64, u64, 0),
264 ASSIGN_FETCH_TYPE(s8, u8, 1),
265 ASSIGN_FETCH_TYPE(s16, u16, 1),
266 ASSIGN_FETCH_TYPE(s32, u32, 1),
267 ASSIGN_FETCH_TYPE(s64, u64, 1),
269 ASSIGN_FETCH_TYPE_END
273 * Allocate new trace_probe and initialize it (including kprobes).
275 static struct trace_kprobe *alloc_trace_kprobe(const char *group,
276 const char *event,
277 void *addr,
278 const char *symbol,
279 unsigned long offs,
280 int nargs, bool is_return)
282 struct trace_kprobe *tk;
283 int ret = -ENOMEM;
285 tk = kzalloc(SIZEOF_TRACE_KPROBE(nargs), GFP_KERNEL);
286 if (!tk)
287 return ERR_PTR(ret);
289 if (symbol) {
290 tk->symbol = kstrdup(symbol, GFP_KERNEL);
291 if (!tk->symbol)
292 goto error;
293 tk->rp.kp.symbol_name = tk->symbol;
294 tk->rp.kp.offset = offs;
295 } else
296 tk->rp.kp.addr = addr;
298 if (is_return)
299 tk->rp.handler = kretprobe_dispatcher;
300 else
301 tk->rp.kp.pre_handler = kprobe_dispatcher;
303 if (!event || !is_good_name(event)) {
304 ret = -EINVAL;
305 goto error;
308 tk->tp.call.class = &tk->tp.class;
309 tk->tp.call.name = kstrdup(event, GFP_KERNEL);
310 if (!tk->tp.call.name)
311 goto error;
313 if (!group || !is_good_name(group)) {
314 ret = -EINVAL;
315 goto error;
318 tk->tp.class.system = kstrdup(group, GFP_KERNEL);
319 if (!tk->tp.class.system)
320 goto error;
322 INIT_LIST_HEAD(&tk->list);
323 INIT_LIST_HEAD(&tk->tp.files);
324 return tk;
325 error:
326 kfree(tk->tp.call.name);
327 kfree(tk->symbol);
328 kfree(tk);
329 return ERR_PTR(ret);
332 static void free_trace_kprobe(struct trace_kprobe *tk)
334 int i;
336 for (i = 0; i < tk->tp.nr_args; i++)
337 traceprobe_free_probe_arg(&tk->tp.args[i]);
339 kfree(tk->tp.call.class->system);
340 kfree(tk->tp.call.name);
341 kfree(tk->symbol);
342 kfree(tk);
345 static struct trace_kprobe *find_trace_kprobe(const char *event,
346 const char *group)
348 struct trace_kprobe *tk;
350 list_for_each_entry(tk, &probe_list, list)
351 if (strcmp(ftrace_event_name(&tk->tp.call), event) == 0 &&
352 strcmp(tk->tp.call.class->system, group) == 0)
353 return tk;
354 return NULL;
358 * Enable trace_probe
359 * if the file is NULL, enable "perf" handler, or enable "trace" handler.
361 static int
362 enable_trace_kprobe(struct trace_kprobe *tk, struct ftrace_event_file *file)
364 int ret = 0;
366 if (file) {
367 struct event_file_link *link;
369 link = kmalloc(sizeof(*link), GFP_KERNEL);
370 if (!link) {
371 ret = -ENOMEM;
372 goto out;
375 link->file = file;
376 list_add_tail_rcu(&link->list, &tk->tp.files);
378 tk->tp.flags |= TP_FLAG_TRACE;
379 } else
380 tk->tp.flags |= TP_FLAG_PROFILE;
382 if (trace_probe_is_registered(&tk->tp) && !trace_kprobe_has_gone(tk)) {
383 if (trace_kprobe_is_return(tk))
384 ret = enable_kretprobe(&tk->rp);
385 else
386 ret = enable_kprobe(&tk->rp.kp);
388 out:
389 return ret;
393 * Disable trace_probe
394 * if the file is NULL, disable "perf" handler, or disable "trace" handler.
396 static int
397 disable_trace_kprobe(struct trace_kprobe *tk, struct ftrace_event_file *file)
399 struct event_file_link *link = NULL;
400 int wait = 0;
401 int ret = 0;
403 if (file) {
404 link = find_event_file_link(&tk->tp, file);
405 if (!link) {
406 ret = -EINVAL;
407 goto out;
410 list_del_rcu(&link->list);
411 wait = 1;
412 if (!list_empty(&tk->tp.files))
413 goto out;
415 tk->tp.flags &= ~TP_FLAG_TRACE;
416 } else
417 tk->tp.flags &= ~TP_FLAG_PROFILE;
419 if (!trace_probe_is_enabled(&tk->tp) && trace_probe_is_registered(&tk->tp)) {
420 if (trace_kprobe_is_return(tk))
421 disable_kretprobe(&tk->rp);
422 else
423 disable_kprobe(&tk->rp.kp);
424 wait = 1;
426 out:
427 if (wait) {
429 * Synchronize with kprobe_trace_func/kretprobe_trace_func
430 * to ensure disabled (all running handlers are finished).
431 * This is not only for kfree(), but also the caller,
432 * trace_remove_event_call() supposes it for releasing
433 * event_call related objects, which will be accessed in
434 * the kprobe_trace_func/kretprobe_trace_func.
436 synchronize_sched();
437 kfree(link); /* Ignored if link == NULL */
440 return ret;
443 /* Internal register function - just handle k*probes and flags */
444 static int __register_trace_kprobe(struct trace_kprobe *tk)
446 int i, ret;
448 if (trace_probe_is_registered(&tk->tp))
449 return -EINVAL;
451 for (i = 0; i < tk->tp.nr_args; i++)
452 traceprobe_update_arg(&tk->tp.args[i]);
454 /* Set/clear disabled flag according to tp->flag */
455 if (trace_probe_is_enabled(&tk->tp))
456 tk->rp.kp.flags &= ~KPROBE_FLAG_DISABLED;
457 else
458 tk->rp.kp.flags |= KPROBE_FLAG_DISABLED;
460 if (trace_kprobe_is_return(tk))
461 ret = register_kretprobe(&tk->rp);
462 else
463 ret = register_kprobe(&tk->rp.kp);
465 if (ret == 0)
466 tk->tp.flags |= TP_FLAG_REGISTERED;
467 else {
468 pr_warning("Could not insert probe at %s+%lu: %d\n",
469 trace_kprobe_symbol(tk), trace_kprobe_offset(tk), ret);
470 if (ret == -ENOENT && trace_kprobe_is_on_module(tk)) {
471 pr_warning("This probe might be able to register after"
472 "target module is loaded. Continue.\n");
473 ret = 0;
474 } else if (ret == -EILSEQ) {
475 pr_warning("Probing address(0x%p) is not an "
476 "instruction boundary.\n",
477 tk->rp.kp.addr);
478 ret = -EINVAL;
482 return ret;
485 /* Internal unregister function - just handle k*probes and flags */
486 static void __unregister_trace_kprobe(struct trace_kprobe *tk)
488 if (trace_probe_is_registered(&tk->tp)) {
489 if (trace_kprobe_is_return(tk))
490 unregister_kretprobe(&tk->rp);
491 else
492 unregister_kprobe(&tk->rp.kp);
493 tk->tp.flags &= ~TP_FLAG_REGISTERED;
494 /* Cleanup kprobe for reuse */
495 if (tk->rp.kp.symbol_name)
496 tk->rp.kp.addr = NULL;
500 /* Unregister a trace_probe and probe_event: call with locking probe_lock */
501 static int unregister_trace_kprobe(struct trace_kprobe *tk)
503 /* Enabled event can not be unregistered */
504 if (trace_probe_is_enabled(&tk->tp))
505 return -EBUSY;
507 /* Will fail if probe is being used by ftrace or perf */
508 if (unregister_kprobe_event(tk))
509 return -EBUSY;
511 __unregister_trace_kprobe(tk);
512 list_del(&tk->list);
514 return 0;
517 /* Register a trace_probe and probe_event */
518 static int register_trace_kprobe(struct trace_kprobe *tk)
520 struct trace_kprobe *old_tk;
521 int ret;
523 mutex_lock(&probe_lock);
525 /* Delete old (same name) event if exist */
526 old_tk = find_trace_kprobe(ftrace_event_name(&tk->tp.call),
527 tk->tp.call.class->system);
528 if (old_tk) {
529 ret = unregister_trace_kprobe(old_tk);
530 if (ret < 0)
531 goto end;
532 free_trace_kprobe(old_tk);
535 /* Register new event */
536 ret = register_kprobe_event(tk);
537 if (ret) {
538 pr_warning("Failed to register probe event(%d)\n", ret);
539 goto end;
542 /* Register k*probe */
543 ret = __register_trace_kprobe(tk);
544 if (ret < 0)
545 unregister_kprobe_event(tk);
546 else
547 list_add_tail(&tk->list, &probe_list);
549 end:
550 mutex_unlock(&probe_lock);
551 return ret;
554 /* Module notifier call back, checking event on the module */
555 static int trace_kprobe_module_callback(struct notifier_block *nb,
556 unsigned long val, void *data)
558 struct module *mod = data;
559 struct trace_kprobe *tk;
560 int ret;
562 if (val != MODULE_STATE_COMING)
563 return NOTIFY_DONE;
565 /* Update probes on coming module */
566 mutex_lock(&probe_lock);
567 list_for_each_entry(tk, &probe_list, list) {
568 if (trace_kprobe_within_module(tk, mod)) {
569 /* Don't need to check busy - this should have gone. */
570 __unregister_trace_kprobe(tk);
571 ret = __register_trace_kprobe(tk);
572 if (ret)
573 pr_warning("Failed to re-register probe %s on"
574 "%s: %d\n",
575 ftrace_event_name(&tk->tp.call),
576 mod->name, ret);
579 mutex_unlock(&probe_lock);
581 return NOTIFY_DONE;
584 static struct notifier_block trace_kprobe_module_nb = {
585 .notifier_call = trace_kprobe_module_callback,
586 .priority = 1 /* Invoked after kprobe module callback */
589 static int create_trace_kprobe(int argc, char **argv)
592 * Argument syntax:
593 * - Add kprobe: p[:[GRP/]EVENT] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
594 * - Add kretprobe: r[:[GRP/]EVENT] [MOD:]KSYM[+0] [FETCHARGS]
595 * Fetch args:
596 * $retval : fetch return value
597 * $stack : fetch stack address
598 * $stackN : fetch Nth of stack (N:0-)
599 * @ADDR : fetch memory at ADDR (ADDR should be in kernel)
600 * @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
601 * %REG : fetch register REG
602 * Dereferencing memory fetch:
603 * +|-offs(ARG) : fetch memory at ARG +|- offs address.
604 * Alias name of args:
605 * NAME=FETCHARG : set NAME as alias of FETCHARG.
606 * Type of args:
607 * FETCHARG:TYPE : use TYPE instead of unsigned long.
609 struct trace_kprobe *tk;
610 int i, ret = 0;
611 bool is_return = false, is_delete = false;
612 char *symbol = NULL, *event = NULL, *group = NULL;
613 char *arg;
614 unsigned long offset = 0;
615 void *addr = NULL;
616 char buf[MAX_EVENT_NAME_LEN];
618 /* argc must be >= 1 */
619 if (argv[0][0] == 'p')
620 is_return = false;
621 else if (argv[0][0] == 'r')
622 is_return = true;
623 else if (argv[0][0] == '-')
624 is_delete = true;
625 else {
626 pr_info("Probe definition must be started with 'p', 'r' or"
627 " '-'.\n");
628 return -EINVAL;
631 if (argv[0][1] == ':') {
632 event = &argv[0][2];
633 if (strchr(event, '/')) {
634 group = event;
635 event = strchr(group, '/') + 1;
636 event[-1] = '\0';
637 if (strlen(group) == 0) {
638 pr_info("Group name is not specified\n");
639 return -EINVAL;
642 if (strlen(event) == 0) {
643 pr_info("Event name is not specified\n");
644 return -EINVAL;
647 if (!group)
648 group = KPROBE_EVENT_SYSTEM;
650 if (is_delete) {
651 if (!event) {
652 pr_info("Delete command needs an event name.\n");
653 return -EINVAL;
655 mutex_lock(&probe_lock);
656 tk = find_trace_kprobe(event, group);
657 if (!tk) {
658 mutex_unlock(&probe_lock);
659 pr_info("Event %s/%s doesn't exist.\n", group, event);
660 return -ENOENT;
662 /* delete an event */
663 ret = unregister_trace_kprobe(tk);
664 if (ret == 0)
665 free_trace_kprobe(tk);
666 mutex_unlock(&probe_lock);
667 return ret;
670 if (argc < 2) {
671 pr_info("Probe point is not specified.\n");
672 return -EINVAL;
674 if (isdigit(argv[1][0])) {
675 if (is_return) {
676 pr_info("Return probe point must be a symbol.\n");
677 return -EINVAL;
679 /* an address specified */
680 ret = kstrtoul(&argv[1][0], 0, (unsigned long *)&addr);
681 if (ret) {
682 pr_info("Failed to parse address.\n");
683 return ret;
685 } else {
686 /* a symbol specified */
687 symbol = argv[1];
688 /* TODO: support .init module functions */
689 ret = traceprobe_split_symbol_offset(symbol, &offset);
690 if (ret) {
691 pr_info("Failed to parse symbol.\n");
692 return ret;
694 if (offset && is_return) {
695 pr_info("Return probe must be used without offset.\n");
696 return -EINVAL;
699 argc -= 2; argv += 2;
701 /* setup a probe */
702 if (!event) {
703 /* Make a new event name */
704 if (symbol)
705 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
706 is_return ? 'r' : 'p', symbol, offset);
707 else
708 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
709 is_return ? 'r' : 'p', addr);
710 event = buf;
712 tk = alloc_trace_kprobe(group, event, addr, symbol, offset, argc,
713 is_return);
714 if (IS_ERR(tk)) {
715 pr_info("Failed to allocate trace_probe.(%d)\n",
716 (int)PTR_ERR(tk));
717 return PTR_ERR(tk);
720 /* parse arguments */
721 ret = 0;
722 for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
723 struct probe_arg *parg = &tk->tp.args[i];
725 /* Increment count for freeing args in error case */
726 tk->tp.nr_args++;
728 /* Parse argument name */
729 arg = strchr(argv[i], '=');
730 if (arg) {
731 *arg++ = '\0';
732 parg->name = kstrdup(argv[i], GFP_KERNEL);
733 } else {
734 arg = argv[i];
735 /* If argument name is omitted, set "argN" */
736 snprintf(buf, MAX_EVENT_NAME_LEN, "arg%d", i + 1);
737 parg->name = kstrdup(buf, GFP_KERNEL);
740 if (!parg->name) {
741 pr_info("Failed to allocate argument[%d] name.\n", i);
742 ret = -ENOMEM;
743 goto error;
746 if (!is_good_name(parg->name)) {
747 pr_info("Invalid argument[%d] name: %s\n",
748 i, parg->name);
749 ret = -EINVAL;
750 goto error;
753 if (traceprobe_conflict_field_name(parg->name,
754 tk->tp.args, i)) {
755 pr_info("Argument[%d] name '%s' conflicts with "
756 "another field.\n", i, argv[i]);
757 ret = -EINVAL;
758 goto error;
761 /* Parse fetch argument */
762 ret = traceprobe_parse_probe_arg(arg, &tk->tp.size, parg,
763 is_return, true);
764 if (ret) {
765 pr_info("Parse error at argument[%d]. (%d)\n", i, ret);
766 goto error;
770 ret = register_trace_kprobe(tk);
771 if (ret)
772 goto error;
773 return 0;
775 error:
776 free_trace_kprobe(tk);
777 return ret;
780 static int release_all_trace_kprobes(void)
782 struct trace_kprobe *tk;
783 int ret = 0;
785 mutex_lock(&probe_lock);
786 /* Ensure no probe is in use. */
787 list_for_each_entry(tk, &probe_list, list)
788 if (trace_probe_is_enabled(&tk->tp)) {
789 ret = -EBUSY;
790 goto end;
792 /* TODO: Use batch unregistration */
793 while (!list_empty(&probe_list)) {
794 tk = list_entry(probe_list.next, struct trace_kprobe, list);
795 ret = unregister_trace_kprobe(tk);
796 if (ret)
797 goto end;
798 free_trace_kprobe(tk);
801 end:
802 mutex_unlock(&probe_lock);
804 return ret;
807 /* Probes listing interfaces */
808 static void *probes_seq_start(struct seq_file *m, loff_t *pos)
810 mutex_lock(&probe_lock);
811 return seq_list_start(&probe_list, *pos);
814 static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
816 return seq_list_next(v, &probe_list, pos);
819 static void probes_seq_stop(struct seq_file *m, void *v)
821 mutex_unlock(&probe_lock);
824 static int probes_seq_show(struct seq_file *m, void *v)
826 struct trace_kprobe *tk = v;
827 int i;
829 seq_printf(m, "%c", trace_kprobe_is_return(tk) ? 'r' : 'p');
830 seq_printf(m, ":%s/%s", tk->tp.call.class->system,
831 ftrace_event_name(&tk->tp.call));
833 if (!tk->symbol)
834 seq_printf(m, " 0x%p", tk->rp.kp.addr);
835 else if (tk->rp.kp.offset)
836 seq_printf(m, " %s+%u", trace_kprobe_symbol(tk),
837 tk->rp.kp.offset);
838 else
839 seq_printf(m, " %s", trace_kprobe_symbol(tk));
841 for (i = 0; i < tk->tp.nr_args; i++)
842 seq_printf(m, " %s=%s", tk->tp.args[i].name, tk->tp.args[i].comm);
843 seq_printf(m, "\n");
845 return 0;
848 static const struct seq_operations probes_seq_op = {
849 .start = probes_seq_start,
850 .next = probes_seq_next,
851 .stop = probes_seq_stop,
852 .show = probes_seq_show
855 static int probes_open(struct inode *inode, struct file *file)
857 int ret;
859 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
860 ret = release_all_trace_kprobes();
861 if (ret < 0)
862 return ret;
865 return seq_open(file, &probes_seq_op);
868 static ssize_t probes_write(struct file *file, const char __user *buffer,
869 size_t count, loff_t *ppos)
871 return traceprobe_probes_write(file, buffer, count, ppos,
872 create_trace_kprobe);
875 static const struct file_operations kprobe_events_ops = {
876 .owner = THIS_MODULE,
877 .open = probes_open,
878 .read = seq_read,
879 .llseek = seq_lseek,
880 .release = seq_release,
881 .write = probes_write,
884 /* Probes profiling interfaces */
885 static int probes_profile_seq_show(struct seq_file *m, void *v)
887 struct trace_kprobe *tk = v;
889 seq_printf(m, " %-44s %15lu %15lu\n",
890 ftrace_event_name(&tk->tp.call), tk->nhit,
891 tk->rp.kp.nmissed);
893 return 0;
896 static const struct seq_operations profile_seq_op = {
897 .start = probes_seq_start,
898 .next = probes_seq_next,
899 .stop = probes_seq_stop,
900 .show = probes_profile_seq_show
903 static int profile_open(struct inode *inode, struct file *file)
905 return seq_open(file, &profile_seq_op);
908 static const struct file_operations kprobe_profile_ops = {
909 .owner = THIS_MODULE,
910 .open = profile_open,
911 .read = seq_read,
912 .llseek = seq_lseek,
913 .release = seq_release,
916 /* Kprobe handler */
917 static nokprobe_inline void
918 __kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs,
919 struct ftrace_event_file *ftrace_file)
921 struct kprobe_trace_entry_head *entry;
922 struct ring_buffer_event *event;
923 struct ring_buffer *buffer;
924 int size, dsize, pc;
925 unsigned long irq_flags;
926 struct ftrace_event_call *call = &tk->tp.call;
928 WARN_ON(call != ftrace_file->event_call);
930 if (ftrace_trigger_soft_disabled(ftrace_file))
931 return;
933 local_save_flags(irq_flags);
934 pc = preempt_count();
936 dsize = __get_data_size(&tk->tp, regs);
937 size = sizeof(*entry) + tk->tp.size + dsize;
939 event = trace_event_buffer_lock_reserve(&buffer, ftrace_file,
940 call->event.type,
941 size, irq_flags, pc);
942 if (!event)
943 return;
945 entry = ring_buffer_event_data(event);
946 entry->ip = (unsigned long)tk->rp.kp.addr;
947 store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
949 event_trigger_unlock_commit_regs(ftrace_file, buffer, event,
950 entry, irq_flags, pc, regs);
953 static void
954 kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs)
956 struct event_file_link *link;
958 list_for_each_entry_rcu(link, &tk->tp.files, list)
959 __kprobe_trace_func(tk, regs, link->file);
961 NOKPROBE_SYMBOL(kprobe_trace_func);
963 /* Kretprobe handler */
964 static nokprobe_inline void
965 __kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
966 struct pt_regs *regs,
967 struct ftrace_event_file *ftrace_file)
969 struct kretprobe_trace_entry_head *entry;
970 struct ring_buffer_event *event;
971 struct ring_buffer *buffer;
972 int size, pc, dsize;
973 unsigned long irq_flags;
974 struct ftrace_event_call *call = &tk->tp.call;
976 WARN_ON(call != ftrace_file->event_call);
978 if (ftrace_trigger_soft_disabled(ftrace_file))
979 return;
981 local_save_flags(irq_flags);
982 pc = preempt_count();
984 dsize = __get_data_size(&tk->tp, regs);
985 size = sizeof(*entry) + tk->tp.size + dsize;
987 event = trace_event_buffer_lock_reserve(&buffer, ftrace_file,
988 call->event.type,
989 size, irq_flags, pc);
990 if (!event)
991 return;
993 entry = ring_buffer_event_data(event);
994 entry->func = (unsigned long)tk->rp.kp.addr;
995 entry->ret_ip = (unsigned long)ri->ret_addr;
996 store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
998 event_trigger_unlock_commit_regs(ftrace_file, buffer, event,
999 entry, irq_flags, pc, regs);
1002 static void
1003 kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1004 struct pt_regs *regs)
1006 struct event_file_link *link;
1008 list_for_each_entry_rcu(link, &tk->tp.files, list)
1009 __kretprobe_trace_func(tk, ri, regs, link->file);
1011 NOKPROBE_SYMBOL(kretprobe_trace_func);
1013 /* Event entry printers */
1014 static enum print_line_t
1015 print_kprobe_event(struct trace_iterator *iter, int flags,
1016 struct trace_event *event)
1018 struct kprobe_trace_entry_head *field;
1019 struct trace_seq *s = &iter->seq;
1020 struct trace_probe *tp;
1021 u8 *data;
1022 int i;
1024 field = (struct kprobe_trace_entry_head *)iter->ent;
1025 tp = container_of(event, struct trace_probe, call.event);
1027 if (!trace_seq_printf(s, "%s: (", ftrace_event_name(&tp->call)))
1028 goto partial;
1030 if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
1031 goto partial;
1033 if (!trace_seq_puts(s, ")"))
1034 goto partial;
1036 data = (u8 *)&field[1];
1037 for (i = 0; i < tp->nr_args; i++)
1038 if (!tp->args[i].type->print(s, tp->args[i].name,
1039 data + tp->args[i].offset, field))
1040 goto partial;
1042 if (!trace_seq_puts(s, "\n"))
1043 goto partial;
1045 return TRACE_TYPE_HANDLED;
1046 partial:
1047 return TRACE_TYPE_PARTIAL_LINE;
1050 static enum print_line_t
1051 print_kretprobe_event(struct trace_iterator *iter, int flags,
1052 struct trace_event *event)
1054 struct kretprobe_trace_entry_head *field;
1055 struct trace_seq *s = &iter->seq;
1056 struct trace_probe *tp;
1057 u8 *data;
1058 int i;
1060 field = (struct kretprobe_trace_entry_head *)iter->ent;
1061 tp = container_of(event, struct trace_probe, call.event);
1063 if (!trace_seq_printf(s, "%s: (", ftrace_event_name(&tp->call)))
1064 goto partial;
1066 if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1067 goto partial;
1069 if (!trace_seq_puts(s, " <- "))
1070 goto partial;
1072 if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1073 goto partial;
1075 if (!trace_seq_puts(s, ")"))
1076 goto partial;
1078 data = (u8 *)&field[1];
1079 for (i = 0; i < tp->nr_args; i++)
1080 if (!tp->args[i].type->print(s, tp->args[i].name,
1081 data + tp->args[i].offset, field))
1082 goto partial;
1084 if (!trace_seq_puts(s, "\n"))
1085 goto partial;
1087 return TRACE_TYPE_HANDLED;
1088 partial:
1089 return TRACE_TYPE_PARTIAL_LINE;
1093 static int kprobe_event_define_fields(struct ftrace_event_call *event_call)
1095 int ret, i;
1096 struct kprobe_trace_entry_head field;
1097 struct trace_kprobe *tk = (struct trace_kprobe *)event_call->data;
1099 DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1100 /* Set argument names as fields */
1101 for (i = 0; i < tk->tp.nr_args; i++) {
1102 struct probe_arg *parg = &tk->tp.args[i];
1104 ret = trace_define_field(event_call, parg->type->fmttype,
1105 parg->name,
1106 sizeof(field) + parg->offset,
1107 parg->type->size,
1108 parg->type->is_signed,
1109 FILTER_OTHER);
1110 if (ret)
1111 return ret;
1113 return 0;
1116 static int kretprobe_event_define_fields(struct ftrace_event_call *event_call)
1118 int ret, i;
1119 struct kretprobe_trace_entry_head field;
1120 struct trace_kprobe *tk = (struct trace_kprobe *)event_call->data;
1122 DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1123 DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1124 /* Set argument names as fields */
1125 for (i = 0; i < tk->tp.nr_args; i++) {
1126 struct probe_arg *parg = &tk->tp.args[i];
1128 ret = trace_define_field(event_call, parg->type->fmttype,
1129 parg->name,
1130 sizeof(field) + parg->offset,
1131 parg->type->size,
1132 parg->type->is_signed,
1133 FILTER_OTHER);
1134 if (ret)
1135 return ret;
1137 return 0;
1140 #ifdef CONFIG_PERF_EVENTS
1142 /* Kprobe profile handler */
1143 static void
1144 kprobe_perf_func(struct trace_kprobe *tk, struct pt_regs *regs)
1146 struct ftrace_event_call *call = &tk->tp.call;
1147 struct kprobe_trace_entry_head *entry;
1148 struct hlist_head *head;
1149 int size, __size, dsize;
1150 int rctx;
1152 head = this_cpu_ptr(call->perf_events);
1153 if (hlist_empty(head))
1154 return;
1156 dsize = __get_data_size(&tk->tp, regs);
1157 __size = sizeof(*entry) + tk->tp.size + dsize;
1158 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1159 size -= sizeof(u32);
1161 entry = perf_trace_buf_prepare(size, call->event.type, regs, &rctx);
1162 if (!entry)
1163 return;
1165 entry->ip = (unsigned long)tk->rp.kp.addr;
1166 memset(&entry[1], 0, dsize);
1167 store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1168 perf_trace_buf_submit(entry, size, rctx, 0, 1, regs, head, NULL);
1170 NOKPROBE_SYMBOL(kprobe_perf_func);
1172 /* Kretprobe profile handler */
1173 static void
1174 kretprobe_perf_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1175 struct pt_regs *regs)
1177 struct ftrace_event_call *call = &tk->tp.call;
1178 struct kretprobe_trace_entry_head *entry;
1179 struct hlist_head *head;
1180 int size, __size, dsize;
1181 int rctx;
1183 head = this_cpu_ptr(call->perf_events);
1184 if (hlist_empty(head))
1185 return;
1187 dsize = __get_data_size(&tk->tp, regs);
1188 __size = sizeof(*entry) + tk->tp.size + dsize;
1189 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1190 size -= sizeof(u32);
1192 entry = perf_trace_buf_prepare(size, call->event.type, regs, &rctx);
1193 if (!entry)
1194 return;
1196 entry->func = (unsigned long)tk->rp.kp.addr;
1197 entry->ret_ip = (unsigned long)ri->ret_addr;
1198 store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1199 perf_trace_buf_submit(entry, size, rctx, 0, 1, regs, head, NULL);
1201 NOKPROBE_SYMBOL(kretprobe_perf_func);
1202 #endif /* CONFIG_PERF_EVENTS */
1205 * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex.
1207 * kprobe_trace_self_tests_init() does enable_trace_probe/disable_trace_probe
1208 * lockless, but we can't race with this __init function.
1210 static int kprobe_register(struct ftrace_event_call *event,
1211 enum trace_reg type, void *data)
1213 struct trace_kprobe *tk = (struct trace_kprobe *)event->data;
1214 struct ftrace_event_file *file = data;
1216 switch (type) {
1217 case TRACE_REG_REGISTER:
1218 return enable_trace_kprobe(tk, file);
1219 case TRACE_REG_UNREGISTER:
1220 return disable_trace_kprobe(tk, file);
1222 #ifdef CONFIG_PERF_EVENTS
1223 case TRACE_REG_PERF_REGISTER:
1224 return enable_trace_kprobe(tk, NULL);
1225 case TRACE_REG_PERF_UNREGISTER:
1226 return disable_trace_kprobe(tk, NULL);
1227 case TRACE_REG_PERF_OPEN:
1228 case TRACE_REG_PERF_CLOSE:
1229 case TRACE_REG_PERF_ADD:
1230 case TRACE_REG_PERF_DEL:
1231 return 0;
1232 #endif
1234 return 0;
1237 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1239 struct trace_kprobe *tk = container_of(kp, struct trace_kprobe, rp.kp);
1241 tk->nhit++;
1243 if (tk->tp.flags & TP_FLAG_TRACE)
1244 kprobe_trace_func(tk, regs);
1245 #ifdef CONFIG_PERF_EVENTS
1246 if (tk->tp.flags & TP_FLAG_PROFILE)
1247 kprobe_perf_func(tk, regs);
1248 #endif
1249 return 0; /* We don't tweek kernel, so just return 0 */
1251 NOKPROBE_SYMBOL(kprobe_dispatcher);
1253 static int
1254 kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1256 struct trace_kprobe *tk = container_of(ri->rp, struct trace_kprobe, rp);
1258 tk->nhit++;
1260 if (tk->tp.flags & TP_FLAG_TRACE)
1261 kretprobe_trace_func(tk, ri, regs);
1262 #ifdef CONFIG_PERF_EVENTS
1263 if (tk->tp.flags & TP_FLAG_PROFILE)
1264 kretprobe_perf_func(tk, ri, regs);
1265 #endif
1266 return 0; /* We don't tweek kernel, so just return 0 */
1268 NOKPROBE_SYMBOL(kretprobe_dispatcher);
1270 static struct trace_event_functions kretprobe_funcs = {
1271 .trace = print_kretprobe_event
1274 static struct trace_event_functions kprobe_funcs = {
1275 .trace = print_kprobe_event
1278 static int register_kprobe_event(struct trace_kprobe *tk)
1280 struct ftrace_event_call *call = &tk->tp.call;
1281 int ret;
1283 /* Initialize ftrace_event_call */
1284 INIT_LIST_HEAD(&call->class->fields);
1285 if (trace_kprobe_is_return(tk)) {
1286 call->event.funcs = &kretprobe_funcs;
1287 call->class->define_fields = kretprobe_event_define_fields;
1288 } else {
1289 call->event.funcs = &kprobe_funcs;
1290 call->class->define_fields = kprobe_event_define_fields;
1292 if (set_print_fmt(&tk->tp, trace_kprobe_is_return(tk)) < 0)
1293 return -ENOMEM;
1294 ret = register_ftrace_event(&call->event);
1295 if (!ret) {
1296 kfree(call->print_fmt);
1297 return -ENODEV;
1299 call->flags = 0;
1300 call->class->reg = kprobe_register;
1301 call->data = tk;
1302 ret = trace_add_event_call(call);
1303 if (ret) {
1304 pr_info("Failed to register kprobe event: %s\n",
1305 ftrace_event_name(call));
1306 kfree(call->print_fmt);
1307 unregister_ftrace_event(&call->event);
1309 return ret;
1312 static int unregister_kprobe_event(struct trace_kprobe *tk)
1314 int ret;
1316 /* tp->event is unregistered in trace_remove_event_call() */
1317 ret = trace_remove_event_call(&tk->tp.call);
1318 if (!ret)
1319 kfree(tk->tp.call.print_fmt);
1320 return ret;
1323 /* Make a debugfs interface for controlling probe points */
1324 static __init int init_kprobe_trace(void)
1326 struct dentry *d_tracer;
1327 struct dentry *entry;
1329 if (register_module_notifier(&trace_kprobe_module_nb))
1330 return -EINVAL;
1332 d_tracer = tracing_init_dentry();
1333 if (!d_tracer)
1334 return 0;
1336 entry = debugfs_create_file("kprobe_events", 0644, d_tracer,
1337 NULL, &kprobe_events_ops);
1339 /* Event list interface */
1340 if (!entry)
1341 pr_warning("Could not create debugfs "
1342 "'kprobe_events' entry\n");
1344 /* Profile interface */
1345 entry = debugfs_create_file("kprobe_profile", 0444, d_tracer,
1346 NULL, &kprobe_profile_ops);
1348 if (!entry)
1349 pr_warning("Could not create debugfs "
1350 "'kprobe_profile' entry\n");
1351 return 0;
1353 fs_initcall(init_kprobe_trace);
1356 #ifdef CONFIG_FTRACE_STARTUP_TEST
1359 * The "__used" keeps gcc from removing the function symbol
1360 * from the kallsyms table.
1362 static __used int kprobe_trace_selftest_target(int a1, int a2, int a3,
1363 int a4, int a5, int a6)
1365 return a1 + a2 + a3 + a4 + a5 + a6;
1368 static struct ftrace_event_file *
1369 find_trace_probe_file(struct trace_kprobe *tk, struct trace_array *tr)
1371 struct ftrace_event_file *file;
1373 list_for_each_entry(file, &tr->events, list)
1374 if (file->event_call == &tk->tp.call)
1375 return file;
1377 return NULL;
1381 * Nobody but us can call enable_trace_kprobe/disable_trace_kprobe at this
1382 * stage, we can do this lockless.
1384 static __init int kprobe_trace_self_tests_init(void)
1386 int ret, warn = 0;
1387 int (*target)(int, int, int, int, int, int);
1388 struct trace_kprobe *tk;
1389 struct ftrace_event_file *file;
1391 if (tracing_is_disabled())
1392 return -ENODEV;
1394 target = kprobe_trace_selftest_target;
1396 pr_info("Testing kprobe tracing: ");
1398 ret = traceprobe_command("p:testprobe kprobe_trace_selftest_target "
1399 "$stack $stack0 +0($stack)",
1400 create_trace_kprobe);
1401 if (WARN_ON_ONCE(ret)) {
1402 pr_warn("error on probing function entry.\n");
1403 warn++;
1404 } else {
1405 /* Enable trace point */
1406 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1407 if (WARN_ON_ONCE(tk == NULL)) {
1408 pr_warn("error on getting new probe.\n");
1409 warn++;
1410 } else {
1411 file = find_trace_probe_file(tk, top_trace_array());
1412 if (WARN_ON_ONCE(file == NULL)) {
1413 pr_warn("error on getting probe file.\n");
1414 warn++;
1415 } else
1416 enable_trace_kprobe(tk, file);
1420 ret = traceprobe_command("r:testprobe2 kprobe_trace_selftest_target "
1421 "$retval", create_trace_kprobe);
1422 if (WARN_ON_ONCE(ret)) {
1423 pr_warn("error on probing function return.\n");
1424 warn++;
1425 } else {
1426 /* Enable trace point */
1427 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1428 if (WARN_ON_ONCE(tk == NULL)) {
1429 pr_warn("error on getting 2nd new probe.\n");
1430 warn++;
1431 } else {
1432 file = find_trace_probe_file(tk, top_trace_array());
1433 if (WARN_ON_ONCE(file == NULL)) {
1434 pr_warn("error on getting probe file.\n");
1435 warn++;
1436 } else
1437 enable_trace_kprobe(tk, file);
1441 if (warn)
1442 goto end;
1444 ret = target(1, 2, 3, 4, 5, 6);
1446 /* Disable trace points before removing it */
1447 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1448 if (WARN_ON_ONCE(tk == NULL)) {
1449 pr_warn("error on getting test probe.\n");
1450 warn++;
1451 } else {
1452 file = find_trace_probe_file(tk, top_trace_array());
1453 if (WARN_ON_ONCE(file == NULL)) {
1454 pr_warn("error on getting probe file.\n");
1455 warn++;
1456 } else
1457 disable_trace_kprobe(tk, file);
1460 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1461 if (WARN_ON_ONCE(tk == NULL)) {
1462 pr_warn("error on getting 2nd test probe.\n");
1463 warn++;
1464 } else {
1465 file = find_trace_probe_file(tk, top_trace_array());
1466 if (WARN_ON_ONCE(file == NULL)) {
1467 pr_warn("error on getting probe file.\n");
1468 warn++;
1469 } else
1470 disable_trace_kprobe(tk, file);
1473 ret = traceprobe_command("-:testprobe", create_trace_kprobe);
1474 if (WARN_ON_ONCE(ret)) {
1475 pr_warn("error on deleting a probe.\n");
1476 warn++;
1479 ret = traceprobe_command("-:testprobe2", create_trace_kprobe);
1480 if (WARN_ON_ONCE(ret)) {
1481 pr_warn("error on deleting a probe.\n");
1482 warn++;
1485 end:
1486 release_all_trace_kprobes();
1487 if (warn)
1488 pr_cont("NG: Some tests are failed. Please check them.\n");
1489 else
1490 pr_cont("OK\n");
1491 return 0;
1494 late_initcall(kprobe_trace_self_tests_init);
1496 #endif