drbd: merge_bvec_fn: properly remap bvm->bi_bdev
[linux/fpc-iii.git] / kernel / trace / trace_kprobe.c
blobbdbae450c13e48e77871317cb19b5667d63a0d44
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 struct event_file_link {
39 struct ftrace_event_file *file;
40 struct list_head list;
43 #define SIZEOF_TRACE_KPROBE(n) \
44 (offsetof(struct trace_kprobe, tp.args) + \
45 (sizeof(struct probe_arg) * (n)))
48 static __kprobes bool trace_kprobe_is_return(struct trace_kprobe *tk)
50 return tk->rp.handler != NULL;
53 static __kprobes const char *trace_kprobe_symbol(struct trace_kprobe *tk)
55 return tk->symbol ? tk->symbol : "unknown";
58 static __kprobes unsigned long trace_kprobe_offset(struct trace_kprobe *tk)
60 return tk->rp.kp.offset;
63 static __kprobes bool trace_kprobe_has_gone(struct trace_kprobe *tk)
65 return !!(kprobe_gone(&tk->rp.kp));
68 static __kprobes bool trace_kprobe_within_module(struct trace_kprobe *tk,
69 struct module *mod)
71 int len = strlen(mod->name);
72 const char *name = trace_kprobe_symbol(tk);
73 return strncmp(mod->name, name, len) == 0 && name[len] == ':';
76 static __kprobes bool trace_kprobe_is_on_module(struct trace_kprobe *tk)
78 return !!strchr(trace_kprobe_symbol(tk), ':');
81 static int register_kprobe_event(struct trace_kprobe *tk);
82 static int unregister_kprobe_event(struct trace_kprobe *tk);
84 static DEFINE_MUTEX(probe_lock);
85 static LIST_HEAD(probe_list);
87 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
88 static int kretprobe_dispatcher(struct kretprobe_instance *ri,
89 struct pt_regs *regs);
91 /* Memory fetching by symbol */
92 struct symbol_cache {
93 char *symbol;
94 long offset;
95 unsigned long addr;
98 unsigned long update_symbol_cache(struct symbol_cache *sc)
100 sc->addr = (unsigned long)kallsyms_lookup_name(sc->symbol);
102 if (sc->addr)
103 sc->addr += sc->offset;
105 return sc->addr;
108 void free_symbol_cache(struct symbol_cache *sc)
110 kfree(sc->symbol);
111 kfree(sc);
114 struct symbol_cache *alloc_symbol_cache(const char *sym, long offset)
116 struct symbol_cache *sc;
118 if (!sym || strlen(sym) == 0)
119 return NULL;
121 sc = kzalloc(sizeof(struct symbol_cache), GFP_KERNEL);
122 if (!sc)
123 return NULL;
125 sc->symbol = kstrdup(sym, GFP_KERNEL);
126 if (!sc->symbol) {
127 kfree(sc);
128 return NULL;
130 sc->offset = offset;
131 update_symbol_cache(sc);
133 return sc;
137 * Kprobes-specific fetch functions
139 #define DEFINE_FETCH_stack(type) \
140 static __kprobes void FETCH_FUNC_NAME(stack, type)(struct pt_regs *regs,\
141 void *offset, void *dest) \
143 *(type *)dest = (type)regs_get_kernel_stack_nth(regs, \
144 (unsigned int)((unsigned long)offset)); \
146 DEFINE_BASIC_FETCH_FUNCS(stack)
147 /* No string on the stack entry */
148 #define fetch_stack_string NULL
149 #define fetch_stack_string_size NULL
151 #define DEFINE_FETCH_memory(type) \
152 static __kprobes void FETCH_FUNC_NAME(memory, type)(struct pt_regs *regs,\
153 void *addr, void *dest) \
155 type retval; \
156 if (probe_kernel_address(addr, retval)) \
157 *(type *)dest = 0; \
158 else \
159 *(type *)dest = retval; \
161 DEFINE_BASIC_FETCH_FUNCS(memory)
163 * Fetch a null-terminated string. Caller MUST set *(u32 *)dest with max
164 * length and relative data location.
166 static __kprobes void FETCH_FUNC_NAME(memory, string)(struct pt_regs *regs,
167 void *addr, void *dest)
169 long ret;
170 int maxlen = get_rloc_len(*(u32 *)dest);
171 u8 *dst = get_rloc_data(dest);
172 u8 *src = addr;
173 mm_segment_t old_fs = get_fs();
175 if (!maxlen)
176 return;
179 * Try to get string again, since the string can be changed while
180 * probing.
182 set_fs(KERNEL_DS);
183 pagefault_disable();
186 ret = __copy_from_user_inatomic(dst++, src++, 1);
187 while (dst[-1] && ret == 0 && src - (u8 *)addr < maxlen);
189 dst[-1] = '\0';
190 pagefault_enable();
191 set_fs(old_fs);
193 if (ret < 0) { /* Failed to fetch string */
194 ((u8 *)get_rloc_data(dest))[0] = '\0';
195 *(u32 *)dest = make_data_rloc(0, get_rloc_offs(*(u32 *)dest));
196 } else {
197 *(u32 *)dest = make_data_rloc(src - (u8 *)addr,
198 get_rloc_offs(*(u32 *)dest));
202 /* Return the length of string -- including null terminal byte */
203 static __kprobes 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;
228 #define DEFINE_FETCH_symbol(type) \
229 __kprobes void FETCH_FUNC_NAME(symbol, type)(struct pt_regs *regs, \
230 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 DEFINE_BASIC_FETCH_FUNCS(symbol)
239 DEFINE_FETCH_symbol(string)
240 DEFINE_FETCH_symbol(string_size)
242 /* kprobes don't support file_offset fetch methods */
243 #define fetch_file_offset_u8 NULL
244 #define fetch_file_offset_u16 NULL
245 #define fetch_file_offset_u32 NULL
246 #define fetch_file_offset_u64 NULL
247 #define fetch_file_offset_string NULL
248 #define fetch_file_offset_string_size NULL
250 /* Fetch type information table */
251 const struct fetch_type kprobes_fetch_type_table[] = {
252 /* Special types */
253 [FETCH_TYPE_STRING] = __ASSIGN_FETCH_TYPE("string", string, string,
254 sizeof(u32), 1, "__data_loc char[]"),
255 [FETCH_TYPE_STRSIZE] = __ASSIGN_FETCH_TYPE("string_size", u32,
256 string_size, sizeof(u32), 0, "u32"),
257 /* Basic types */
258 ASSIGN_FETCH_TYPE(u8, u8, 0),
259 ASSIGN_FETCH_TYPE(u16, u16, 0),
260 ASSIGN_FETCH_TYPE(u32, u32, 0),
261 ASSIGN_FETCH_TYPE(u64, u64, 0),
262 ASSIGN_FETCH_TYPE(s8, u8, 1),
263 ASSIGN_FETCH_TYPE(s16, u16, 1),
264 ASSIGN_FETCH_TYPE(s32, u32, 1),
265 ASSIGN_FETCH_TYPE(s64, u64, 1),
267 ASSIGN_FETCH_TYPE_END
271 * Allocate new trace_probe and initialize it (including kprobes).
273 static struct trace_kprobe *alloc_trace_kprobe(const char *group,
274 const char *event,
275 void *addr,
276 const char *symbol,
277 unsigned long offs,
278 int nargs, bool is_return)
280 struct trace_kprobe *tk;
281 int ret = -ENOMEM;
283 tk = kzalloc(SIZEOF_TRACE_KPROBE(nargs), GFP_KERNEL);
284 if (!tk)
285 return ERR_PTR(ret);
287 if (symbol) {
288 tk->symbol = kstrdup(symbol, GFP_KERNEL);
289 if (!tk->symbol)
290 goto error;
291 tk->rp.kp.symbol_name = tk->symbol;
292 tk->rp.kp.offset = offs;
293 } else
294 tk->rp.kp.addr = addr;
296 if (is_return)
297 tk->rp.handler = kretprobe_dispatcher;
298 else
299 tk->rp.kp.pre_handler = kprobe_dispatcher;
301 if (!event || !is_good_name(event)) {
302 ret = -EINVAL;
303 goto error;
306 tk->tp.call.class = &tk->tp.class;
307 tk->tp.call.name = kstrdup(event, GFP_KERNEL);
308 if (!tk->tp.call.name)
309 goto error;
311 if (!group || !is_good_name(group)) {
312 ret = -EINVAL;
313 goto error;
316 tk->tp.class.system = kstrdup(group, GFP_KERNEL);
317 if (!tk->tp.class.system)
318 goto error;
320 INIT_LIST_HEAD(&tk->list);
321 INIT_LIST_HEAD(&tk->tp.files);
322 return tk;
323 error:
324 kfree(tk->tp.call.name);
325 kfree(tk->symbol);
326 kfree(tk);
327 return ERR_PTR(ret);
330 static void free_trace_kprobe(struct trace_kprobe *tk)
332 int i;
334 for (i = 0; i < tk->tp.nr_args; i++)
335 traceprobe_free_probe_arg(&tk->tp.args[i]);
337 kfree(tk->tp.call.class->system);
338 kfree(tk->tp.call.name);
339 kfree(tk->symbol);
340 kfree(tk);
343 static struct trace_kprobe *find_trace_kprobe(const char *event,
344 const char *group)
346 struct trace_kprobe *tk;
348 list_for_each_entry(tk, &probe_list, list)
349 if (strcmp(tk->tp.call.name, event) == 0 &&
350 strcmp(tk->tp.call.class->system, group) == 0)
351 return tk;
352 return NULL;
356 * Enable trace_probe
357 * if the file is NULL, enable "perf" handler, or enable "trace" handler.
359 static int
360 enable_trace_kprobe(struct trace_kprobe *tk, struct ftrace_event_file *file)
362 int ret = 0;
364 if (file) {
365 struct event_file_link *link;
367 link = kmalloc(sizeof(*link), GFP_KERNEL);
368 if (!link) {
369 ret = -ENOMEM;
370 goto out;
373 link->file = file;
374 list_add_tail_rcu(&link->list, &tk->tp.files);
376 tk->tp.flags |= TP_FLAG_TRACE;
377 } else
378 tk->tp.flags |= TP_FLAG_PROFILE;
380 if (trace_probe_is_registered(&tk->tp) && !trace_kprobe_has_gone(tk)) {
381 if (trace_kprobe_is_return(tk))
382 ret = enable_kretprobe(&tk->rp);
383 else
384 ret = enable_kprobe(&tk->rp.kp);
386 out:
387 return ret;
390 static struct event_file_link *
391 find_event_file_link(struct trace_probe *tp, struct ftrace_event_file *file)
393 struct event_file_link *link;
395 list_for_each_entry(link, &tp->files, list)
396 if (link->file == file)
397 return link;
399 return NULL;
403 * Disable trace_probe
404 * if the file is NULL, disable "perf" handler, or disable "trace" handler.
406 static int
407 disable_trace_kprobe(struct trace_kprobe *tk, struct ftrace_event_file *file)
409 struct event_file_link *link = NULL;
410 int wait = 0;
411 int ret = 0;
413 if (file) {
414 link = find_event_file_link(&tk->tp, file);
415 if (!link) {
416 ret = -EINVAL;
417 goto out;
420 list_del_rcu(&link->list);
421 wait = 1;
422 if (!list_empty(&tk->tp.files))
423 goto out;
425 tk->tp.flags &= ~TP_FLAG_TRACE;
426 } else
427 tk->tp.flags &= ~TP_FLAG_PROFILE;
429 if (!trace_probe_is_enabled(&tk->tp) && trace_probe_is_registered(&tk->tp)) {
430 if (trace_kprobe_is_return(tk))
431 disable_kretprobe(&tk->rp);
432 else
433 disable_kprobe(&tk->rp.kp);
434 wait = 1;
436 out:
437 if (wait) {
439 * Synchronize with kprobe_trace_func/kretprobe_trace_func
440 * to ensure disabled (all running handlers are finished).
441 * This is not only for kfree(), but also the caller,
442 * trace_remove_event_call() supposes it for releasing
443 * event_call related objects, which will be accessed in
444 * the kprobe_trace_func/kretprobe_trace_func.
446 synchronize_sched();
447 kfree(link); /* Ignored if link == NULL */
450 return ret;
453 /* Internal register function - just handle k*probes and flags */
454 static int __register_trace_kprobe(struct trace_kprobe *tk)
456 int i, ret;
458 if (trace_probe_is_registered(&tk->tp))
459 return -EINVAL;
461 for (i = 0; i < tk->tp.nr_args; i++)
462 traceprobe_update_arg(&tk->tp.args[i]);
464 /* Set/clear disabled flag according to tp->flag */
465 if (trace_probe_is_enabled(&tk->tp))
466 tk->rp.kp.flags &= ~KPROBE_FLAG_DISABLED;
467 else
468 tk->rp.kp.flags |= KPROBE_FLAG_DISABLED;
470 if (trace_kprobe_is_return(tk))
471 ret = register_kretprobe(&tk->rp);
472 else
473 ret = register_kprobe(&tk->rp.kp);
475 if (ret == 0)
476 tk->tp.flags |= TP_FLAG_REGISTERED;
477 else {
478 pr_warning("Could not insert probe at %s+%lu: %d\n",
479 trace_kprobe_symbol(tk), trace_kprobe_offset(tk), ret);
480 if (ret == -ENOENT && trace_kprobe_is_on_module(tk)) {
481 pr_warning("This probe might be able to register after"
482 "target module is loaded. Continue.\n");
483 ret = 0;
484 } else if (ret == -EILSEQ) {
485 pr_warning("Probing address(0x%p) is not an "
486 "instruction boundary.\n",
487 tk->rp.kp.addr);
488 ret = -EINVAL;
492 return ret;
495 /* Internal unregister function - just handle k*probes and flags */
496 static void __unregister_trace_kprobe(struct trace_kprobe *tk)
498 if (trace_probe_is_registered(&tk->tp)) {
499 if (trace_kprobe_is_return(tk))
500 unregister_kretprobe(&tk->rp);
501 else
502 unregister_kprobe(&tk->rp.kp);
503 tk->tp.flags &= ~TP_FLAG_REGISTERED;
504 /* Cleanup kprobe for reuse */
505 if (tk->rp.kp.symbol_name)
506 tk->rp.kp.addr = NULL;
510 /* Unregister a trace_probe and probe_event: call with locking probe_lock */
511 static int unregister_trace_kprobe(struct trace_kprobe *tk)
513 /* Enabled event can not be unregistered */
514 if (trace_probe_is_enabled(&tk->tp))
515 return -EBUSY;
517 /* Will fail if probe is being used by ftrace or perf */
518 if (unregister_kprobe_event(tk))
519 return -EBUSY;
521 __unregister_trace_kprobe(tk);
522 list_del(&tk->list);
524 return 0;
527 /* Register a trace_probe and probe_event */
528 static int register_trace_kprobe(struct trace_kprobe *tk)
530 struct trace_kprobe *old_tk;
531 int ret;
533 mutex_lock(&probe_lock);
535 /* Delete old (same name) event if exist */
536 old_tk = find_trace_kprobe(tk->tp.call.name, tk->tp.call.class->system);
537 if (old_tk) {
538 ret = unregister_trace_kprobe(old_tk);
539 if (ret < 0)
540 goto end;
541 free_trace_kprobe(old_tk);
544 /* Register new event */
545 ret = register_kprobe_event(tk);
546 if (ret) {
547 pr_warning("Failed to register probe event(%d)\n", ret);
548 goto end;
551 /* Register k*probe */
552 ret = __register_trace_kprobe(tk);
553 if (ret < 0)
554 unregister_kprobe_event(tk);
555 else
556 list_add_tail(&tk->list, &probe_list);
558 end:
559 mutex_unlock(&probe_lock);
560 return ret;
563 /* Module notifier call back, checking event on the module */
564 static int trace_kprobe_module_callback(struct notifier_block *nb,
565 unsigned long val, void *data)
567 struct module *mod = data;
568 struct trace_kprobe *tk;
569 int ret;
571 if (val != MODULE_STATE_COMING)
572 return NOTIFY_DONE;
574 /* Update probes on coming module */
575 mutex_lock(&probe_lock);
576 list_for_each_entry(tk, &probe_list, list) {
577 if (trace_kprobe_within_module(tk, mod)) {
578 /* Don't need to check busy - this should have gone. */
579 __unregister_trace_kprobe(tk);
580 ret = __register_trace_kprobe(tk);
581 if (ret)
582 pr_warning("Failed to re-register probe %s on"
583 "%s: %d\n",
584 tk->tp.call.name, mod->name, ret);
587 mutex_unlock(&probe_lock);
589 return NOTIFY_DONE;
592 static struct notifier_block trace_kprobe_module_nb = {
593 .notifier_call = trace_kprobe_module_callback,
594 .priority = 1 /* Invoked after kprobe module callback */
597 static int create_trace_kprobe(int argc, char **argv)
600 * Argument syntax:
601 * - Add kprobe: p[:[GRP/]EVENT] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
602 * - Add kretprobe: r[:[GRP/]EVENT] [MOD:]KSYM[+0] [FETCHARGS]
603 * Fetch args:
604 * $retval : fetch return value
605 * $stack : fetch stack address
606 * $stackN : fetch Nth of stack (N:0-)
607 * @ADDR : fetch memory at ADDR (ADDR should be in kernel)
608 * @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
609 * %REG : fetch register REG
610 * Dereferencing memory fetch:
611 * +|-offs(ARG) : fetch memory at ARG +|- offs address.
612 * Alias name of args:
613 * NAME=FETCHARG : set NAME as alias of FETCHARG.
614 * Type of args:
615 * FETCHARG:TYPE : use TYPE instead of unsigned long.
617 struct trace_kprobe *tk;
618 int i, ret = 0;
619 bool is_return = false, is_delete = false;
620 char *symbol = NULL, *event = NULL, *group = NULL;
621 char *arg;
622 unsigned long offset = 0;
623 void *addr = NULL;
624 char buf[MAX_EVENT_NAME_LEN];
626 /* argc must be >= 1 */
627 if (argv[0][0] == 'p')
628 is_return = false;
629 else if (argv[0][0] == 'r')
630 is_return = true;
631 else if (argv[0][0] == '-')
632 is_delete = true;
633 else {
634 pr_info("Probe definition must be started with 'p', 'r' or"
635 " '-'.\n");
636 return -EINVAL;
639 if (argv[0][1] == ':') {
640 event = &argv[0][2];
641 if (strchr(event, '/')) {
642 group = event;
643 event = strchr(group, '/') + 1;
644 event[-1] = '\0';
645 if (strlen(group) == 0) {
646 pr_info("Group name is not specified\n");
647 return -EINVAL;
650 if (strlen(event) == 0) {
651 pr_info("Event name is not specified\n");
652 return -EINVAL;
655 if (!group)
656 group = KPROBE_EVENT_SYSTEM;
658 if (is_delete) {
659 if (!event) {
660 pr_info("Delete command needs an event name.\n");
661 return -EINVAL;
663 mutex_lock(&probe_lock);
664 tk = find_trace_kprobe(event, group);
665 if (!tk) {
666 mutex_unlock(&probe_lock);
667 pr_info("Event %s/%s doesn't exist.\n", group, event);
668 return -ENOENT;
670 /* delete an event */
671 ret = unregister_trace_kprobe(tk);
672 if (ret == 0)
673 free_trace_kprobe(tk);
674 mutex_unlock(&probe_lock);
675 return ret;
678 if (argc < 2) {
679 pr_info("Probe point is not specified.\n");
680 return -EINVAL;
682 if (isdigit(argv[1][0])) {
683 if (is_return) {
684 pr_info("Return probe point must be a symbol.\n");
685 return -EINVAL;
687 /* an address specified */
688 ret = kstrtoul(&argv[1][0], 0, (unsigned long *)&addr);
689 if (ret) {
690 pr_info("Failed to parse address.\n");
691 return ret;
693 } else {
694 /* a symbol specified */
695 symbol = argv[1];
696 /* TODO: support .init module functions */
697 ret = traceprobe_split_symbol_offset(symbol, &offset);
698 if (ret) {
699 pr_info("Failed to parse symbol.\n");
700 return ret;
702 if (offset && is_return) {
703 pr_info("Return probe must be used without offset.\n");
704 return -EINVAL;
707 argc -= 2; argv += 2;
709 /* setup a probe */
710 if (!event) {
711 /* Make a new event name */
712 if (symbol)
713 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
714 is_return ? 'r' : 'p', symbol, offset);
715 else
716 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
717 is_return ? 'r' : 'p', addr);
718 event = buf;
720 tk = alloc_trace_kprobe(group, event, addr, symbol, offset, argc,
721 is_return);
722 if (IS_ERR(tk)) {
723 pr_info("Failed to allocate trace_probe.(%d)\n",
724 (int)PTR_ERR(tk));
725 return PTR_ERR(tk);
728 /* parse arguments */
729 ret = 0;
730 for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
731 struct probe_arg *parg = &tk->tp.args[i];
733 /* Increment count for freeing args in error case */
734 tk->tp.nr_args++;
736 /* Parse argument name */
737 arg = strchr(argv[i], '=');
738 if (arg) {
739 *arg++ = '\0';
740 parg->name = kstrdup(argv[i], GFP_KERNEL);
741 } else {
742 arg = argv[i];
743 /* If argument name is omitted, set "argN" */
744 snprintf(buf, MAX_EVENT_NAME_LEN, "arg%d", i + 1);
745 parg->name = kstrdup(buf, GFP_KERNEL);
748 if (!parg->name) {
749 pr_info("Failed to allocate argument[%d] name.\n", i);
750 ret = -ENOMEM;
751 goto error;
754 if (!is_good_name(parg->name)) {
755 pr_info("Invalid argument[%d] name: %s\n",
756 i, parg->name);
757 ret = -EINVAL;
758 goto error;
761 if (traceprobe_conflict_field_name(parg->name,
762 tk->tp.args, i)) {
763 pr_info("Argument[%d] name '%s' conflicts with "
764 "another field.\n", i, argv[i]);
765 ret = -EINVAL;
766 goto error;
769 /* Parse fetch argument */
770 ret = traceprobe_parse_probe_arg(arg, &tk->tp.size, parg,
771 is_return, true);
772 if (ret) {
773 pr_info("Parse error at argument[%d]. (%d)\n", i, ret);
774 goto error;
778 ret = register_trace_kprobe(tk);
779 if (ret)
780 goto error;
781 return 0;
783 error:
784 free_trace_kprobe(tk);
785 return ret;
788 static int release_all_trace_kprobes(void)
790 struct trace_kprobe *tk;
791 int ret = 0;
793 mutex_lock(&probe_lock);
794 /* Ensure no probe is in use. */
795 list_for_each_entry(tk, &probe_list, list)
796 if (trace_probe_is_enabled(&tk->tp)) {
797 ret = -EBUSY;
798 goto end;
800 /* TODO: Use batch unregistration */
801 while (!list_empty(&probe_list)) {
802 tk = list_entry(probe_list.next, struct trace_kprobe, list);
803 ret = unregister_trace_kprobe(tk);
804 if (ret)
805 goto end;
806 free_trace_kprobe(tk);
809 end:
810 mutex_unlock(&probe_lock);
812 return ret;
815 /* Probes listing interfaces */
816 static void *probes_seq_start(struct seq_file *m, loff_t *pos)
818 mutex_lock(&probe_lock);
819 return seq_list_start(&probe_list, *pos);
822 static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
824 return seq_list_next(v, &probe_list, pos);
827 static void probes_seq_stop(struct seq_file *m, void *v)
829 mutex_unlock(&probe_lock);
832 static int probes_seq_show(struct seq_file *m, void *v)
834 struct trace_kprobe *tk = v;
835 int i;
837 seq_printf(m, "%c", trace_kprobe_is_return(tk) ? 'r' : 'p');
838 seq_printf(m, ":%s/%s", tk->tp.call.class->system, tk->tp.call.name);
840 if (!tk->symbol)
841 seq_printf(m, " 0x%p", tk->rp.kp.addr);
842 else if (tk->rp.kp.offset)
843 seq_printf(m, " %s+%u", trace_kprobe_symbol(tk),
844 tk->rp.kp.offset);
845 else
846 seq_printf(m, " %s", trace_kprobe_symbol(tk));
848 for (i = 0; i < tk->tp.nr_args; i++)
849 seq_printf(m, " %s=%s", tk->tp.args[i].name, tk->tp.args[i].comm);
850 seq_printf(m, "\n");
852 return 0;
855 static const struct seq_operations probes_seq_op = {
856 .start = probes_seq_start,
857 .next = probes_seq_next,
858 .stop = probes_seq_stop,
859 .show = probes_seq_show
862 static int probes_open(struct inode *inode, struct file *file)
864 int ret;
866 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
867 ret = release_all_trace_kprobes();
868 if (ret < 0)
869 return ret;
872 return seq_open(file, &probes_seq_op);
875 static ssize_t probes_write(struct file *file, const char __user *buffer,
876 size_t count, loff_t *ppos)
878 return traceprobe_probes_write(file, buffer, count, ppos,
879 create_trace_kprobe);
882 static const struct file_operations kprobe_events_ops = {
883 .owner = THIS_MODULE,
884 .open = probes_open,
885 .read = seq_read,
886 .llseek = seq_lseek,
887 .release = seq_release,
888 .write = probes_write,
891 /* Probes profiling interfaces */
892 static int probes_profile_seq_show(struct seq_file *m, void *v)
894 struct trace_kprobe *tk = v;
896 seq_printf(m, " %-44s %15lu %15lu\n", tk->tp.call.name, tk->nhit,
897 tk->rp.kp.nmissed);
899 return 0;
902 static const struct seq_operations profile_seq_op = {
903 .start = probes_seq_start,
904 .next = probes_seq_next,
905 .stop = probes_seq_stop,
906 .show = probes_profile_seq_show
909 static int profile_open(struct inode *inode, struct file *file)
911 return seq_open(file, &profile_seq_op);
914 static const struct file_operations kprobe_profile_ops = {
915 .owner = THIS_MODULE,
916 .open = profile_open,
917 .read = seq_read,
918 .llseek = seq_lseek,
919 .release = seq_release,
922 /* Kprobe handler */
923 static __kprobes void
924 __kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs,
925 struct ftrace_event_file *ftrace_file)
927 struct kprobe_trace_entry_head *entry;
928 struct ring_buffer_event *event;
929 struct ring_buffer *buffer;
930 int size, dsize, pc;
931 unsigned long irq_flags;
932 struct ftrace_event_call *call = &tk->tp.call;
934 WARN_ON(call != ftrace_file->event_call);
936 if (ftrace_trigger_soft_disabled(ftrace_file))
937 return;
939 local_save_flags(irq_flags);
940 pc = preempt_count();
942 dsize = __get_data_size(&tk->tp, regs);
943 size = sizeof(*entry) + tk->tp.size + dsize;
945 event = trace_event_buffer_lock_reserve(&buffer, ftrace_file,
946 call->event.type,
947 size, irq_flags, pc);
948 if (!event)
949 return;
951 entry = ring_buffer_event_data(event);
952 entry->ip = (unsigned long)tk->rp.kp.addr;
953 store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
955 event_trigger_unlock_commit_regs(ftrace_file, buffer, event,
956 entry, irq_flags, pc, regs);
959 static __kprobes void
960 kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs)
962 struct event_file_link *link;
964 list_for_each_entry_rcu(link, &tk->tp.files, list)
965 __kprobe_trace_func(tk, regs, link->file);
968 /* Kretprobe handler */
969 static __kprobes void
970 __kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
971 struct pt_regs *regs,
972 struct ftrace_event_file *ftrace_file)
974 struct kretprobe_trace_entry_head *entry;
975 struct ring_buffer_event *event;
976 struct ring_buffer *buffer;
977 int size, pc, dsize;
978 unsigned long irq_flags;
979 struct ftrace_event_call *call = &tk->tp.call;
981 WARN_ON(call != ftrace_file->event_call);
983 if (ftrace_trigger_soft_disabled(ftrace_file))
984 return;
986 local_save_flags(irq_flags);
987 pc = preempt_count();
989 dsize = __get_data_size(&tk->tp, regs);
990 size = sizeof(*entry) + tk->tp.size + dsize;
992 event = trace_event_buffer_lock_reserve(&buffer, ftrace_file,
993 call->event.type,
994 size, irq_flags, pc);
995 if (!event)
996 return;
998 entry = ring_buffer_event_data(event);
999 entry->func = (unsigned long)tk->rp.kp.addr;
1000 entry->ret_ip = (unsigned long)ri->ret_addr;
1001 store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1003 event_trigger_unlock_commit_regs(ftrace_file, buffer, event,
1004 entry, irq_flags, pc, regs);
1007 static __kprobes void
1008 kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1009 struct pt_regs *regs)
1011 struct event_file_link *link;
1013 list_for_each_entry_rcu(link, &tk->tp.files, list)
1014 __kretprobe_trace_func(tk, ri, regs, link->file);
1017 /* Event entry printers */
1018 static enum print_line_t
1019 print_kprobe_event(struct trace_iterator *iter, int flags,
1020 struct trace_event *event)
1022 struct kprobe_trace_entry_head *field;
1023 struct trace_seq *s = &iter->seq;
1024 struct trace_probe *tp;
1025 u8 *data;
1026 int i;
1028 field = (struct kprobe_trace_entry_head *)iter->ent;
1029 tp = container_of(event, struct trace_probe, call.event);
1031 if (!trace_seq_printf(s, "%s: (", tp->call.name))
1032 goto partial;
1034 if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
1035 goto partial;
1037 if (!trace_seq_puts(s, ")"))
1038 goto partial;
1040 data = (u8 *)&field[1];
1041 for (i = 0; i < tp->nr_args; i++)
1042 if (!tp->args[i].type->print(s, tp->args[i].name,
1043 data + tp->args[i].offset, field))
1044 goto partial;
1046 if (!trace_seq_puts(s, "\n"))
1047 goto partial;
1049 return TRACE_TYPE_HANDLED;
1050 partial:
1051 return TRACE_TYPE_PARTIAL_LINE;
1054 static enum print_line_t
1055 print_kretprobe_event(struct trace_iterator *iter, int flags,
1056 struct trace_event *event)
1058 struct kretprobe_trace_entry_head *field;
1059 struct trace_seq *s = &iter->seq;
1060 struct trace_probe *tp;
1061 u8 *data;
1062 int i;
1064 field = (struct kretprobe_trace_entry_head *)iter->ent;
1065 tp = container_of(event, struct trace_probe, call.event);
1067 if (!trace_seq_printf(s, "%s: (", tp->call.name))
1068 goto partial;
1070 if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1071 goto partial;
1073 if (!trace_seq_puts(s, " <- "))
1074 goto partial;
1076 if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1077 goto partial;
1079 if (!trace_seq_puts(s, ")"))
1080 goto partial;
1082 data = (u8 *)&field[1];
1083 for (i = 0; i < tp->nr_args; i++)
1084 if (!tp->args[i].type->print(s, tp->args[i].name,
1085 data + tp->args[i].offset, field))
1086 goto partial;
1088 if (!trace_seq_puts(s, "\n"))
1089 goto partial;
1091 return TRACE_TYPE_HANDLED;
1092 partial:
1093 return TRACE_TYPE_PARTIAL_LINE;
1097 static int kprobe_event_define_fields(struct ftrace_event_call *event_call)
1099 int ret, i;
1100 struct kprobe_trace_entry_head field;
1101 struct trace_kprobe *tk = (struct trace_kprobe *)event_call->data;
1103 DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1104 /* Set argument names as fields */
1105 for (i = 0; i < tk->tp.nr_args; i++) {
1106 struct probe_arg *parg = &tk->tp.args[i];
1108 ret = trace_define_field(event_call, parg->type->fmttype,
1109 parg->name,
1110 sizeof(field) + parg->offset,
1111 parg->type->size,
1112 parg->type->is_signed,
1113 FILTER_OTHER);
1114 if (ret)
1115 return ret;
1117 return 0;
1120 static int kretprobe_event_define_fields(struct ftrace_event_call *event_call)
1122 int ret, i;
1123 struct kretprobe_trace_entry_head field;
1124 struct trace_kprobe *tk = (struct trace_kprobe *)event_call->data;
1126 DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1127 DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1128 /* Set argument names as fields */
1129 for (i = 0; i < tk->tp.nr_args; i++) {
1130 struct probe_arg *parg = &tk->tp.args[i];
1132 ret = trace_define_field(event_call, parg->type->fmttype,
1133 parg->name,
1134 sizeof(field) + parg->offset,
1135 parg->type->size,
1136 parg->type->is_signed,
1137 FILTER_OTHER);
1138 if (ret)
1139 return ret;
1141 return 0;
1144 #ifdef CONFIG_PERF_EVENTS
1146 /* Kprobe profile handler */
1147 static __kprobes void
1148 kprobe_perf_func(struct trace_kprobe *tk, struct pt_regs *regs)
1150 struct ftrace_event_call *call = &tk->tp.call;
1151 struct kprobe_trace_entry_head *entry;
1152 struct hlist_head *head;
1153 int size, __size, dsize;
1154 int rctx;
1156 head = this_cpu_ptr(call->perf_events);
1157 if (hlist_empty(head))
1158 return;
1160 dsize = __get_data_size(&tk->tp, regs);
1161 __size = sizeof(*entry) + tk->tp.size + dsize;
1162 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1163 size -= sizeof(u32);
1165 entry = perf_trace_buf_prepare(size, call->event.type, regs, &rctx);
1166 if (!entry)
1167 return;
1169 entry->ip = (unsigned long)tk->rp.kp.addr;
1170 memset(&entry[1], 0, dsize);
1171 store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1172 perf_trace_buf_submit(entry, size, rctx, 0, 1, regs, head, NULL);
1175 /* Kretprobe profile handler */
1176 static __kprobes void
1177 kretprobe_perf_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1178 struct pt_regs *regs)
1180 struct ftrace_event_call *call = &tk->tp.call;
1181 struct kretprobe_trace_entry_head *entry;
1182 struct hlist_head *head;
1183 int size, __size, dsize;
1184 int rctx;
1186 head = this_cpu_ptr(call->perf_events);
1187 if (hlist_empty(head))
1188 return;
1190 dsize = __get_data_size(&tk->tp, regs);
1191 __size = sizeof(*entry) + tk->tp.size + dsize;
1192 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1193 size -= sizeof(u32);
1195 entry = perf_trace_buf_prepare(size, call->event.type, regs, &rctx);
1196 if (!entry)
1197 return;
1199 entry->func = (unsigned long)tk->rp.kp.addr;
1200 entry->ret_ip = (unsigned long)ri->ret_addr;
1201 store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1202 perf_trace_buf_submit(entry, size, rctx, 0, 1, regs, head, NULL);
1204 #endif /* CONFIG_PERF_EVENTS */
1207 * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex.
1209 * kprobe_trace_self_tests_init() does enable_trace_probe/disable_trace_probe
1210 * lockless, but we can't race with this __init function.
1212 static __kprobes
1213 int kprobe_register(struct ftrace_event_call *event,
1214 enum trace_reg type, void *data)
1216 struct trace_kprobe *tk = (struct trace_kprobe *)event->data;
1217 struct ftrace_event_file *file = data;
1219 switch (type) {
1220 case TRACE_REG_REGISTER:
1221 return enable_trace_kprobe(tk, file);
1222 case TRACE_REG_UNREGISTER:
1223 return disable_trace_kprobe(tk, file);
1225 #ifdef CONFIG_PERF_EVENTS
1226 case TRACE_REG_PERF_REGISTER:
1227 return enable_trace_kprobe(tk, NULL);
1228 case TRACE_REG_PERF_UNREGISTER:
1229 return disable_trace_kprobe(tk, NULL);
1230 case TRACE_REG_PERF_OPEN:
1231 case TRACE_REG_PERF_CLOSE:
1232 case TRACE_REG_PERF_ADD:
1233 case TRACE_REG_PERF_DEL:
1234 return 0;
1235 #endif
1237 return 0;
1240 static __kprobes
1241 int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1243 struct trace_kprobe *tk = container_of(kp, struct trace_kprobe, rp.kp);
1245 tk->nhit++;
1247 if (tk->tp.flags & TP_FLAG_TRACE)
1248 kprobe_trace_func(tk, regs);
1249 #ifdef CONFIG_PERF_EVENTS
1250 if (tk->tp.flags & TP_FLAG_PROFILE)
1251 kprobe_perf_func(tk, regs);
1252 #endif
1253 return 0; /* We don't tweek kernel, so just return 0 */
1256 static __kprobes
1257 int kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1259 struct trace_kprobe *tk = container_of(ri->rp, struct trace_kprobe, rp);
1261 tk->nhit++;
1263 if (tk->tp.flags & TP_FLAG_TRACE)
1264 kretprobe_trace_func(tk, ri, regs);
1265 #ifdef CONFIG_PERF_EVENTS
1266 if (tk->tp.flags & TP_FLAG_PROFILE)
1267 kretprobe_perf_func(tk, ri, regs);
1268 #endif
1269 return 0; /* We don't tweek kernel, so just return 0 */
1272 static struct trace_event_functions kretprobe_funcs = {
1273 .trace = print_kretprobe_event
1276 static struct trace_event_functions kprobe_funcs = {
1277 .trace = print_kprobe_event
1280 static int register_kprobe_event(struct trace_kprobe *tk)
1282 struct ftrace_event_call *call = &tk->tp.call;
1283 int ret;
1285 /* Initialize ftrace_event_call */
1286 INIT_LIST_HEAD(&call->class->fields);
1287 if (trace_kprobe_is_return(tk)) {
1288 call->event.funcs = &kretprobe_funcs;
1289 call->class->define_fields = kretprobe_event_define_fields;
1290 } else {
1291 call->event.funcs = &kprobe_funcs;
1292 call->class->define_fields = kprobe_event_define_fields;
1294 if (set_print_fmt(&tk->tp, trace_kprobe_is_return(tk)) < 0)
1295 return -ENOMEM;
1296 ret = register_ftrace_event(&call->event);
1297 if (!ret) {
1298 kfree(call->print_fmt);
1299 return -ENODEV;
1301 call->flags = 0;
1302 call->class->reg = kprobe_register;
1303 call->data = tk;
1304 ret = trace_add_event_call(call);
1305 if (ret) {
1306 pr_info("Failed to register kprobe event: %s\n", call->name);
1307 kfree(call->print_fmt);
1308 unregister_ftrace_event(&call->event);
1310 return ret;
1313 static int unregister_kprobe_event(struct trace_kprobe *tk)
1315 int ret;
1317 /* tp->event is unregistered in trace_remove_event_call() */
1318 ret = trace_remove_event_call(&tk->tp.call);
1319 if (!ret)
1320 kfree(tk->tp.call.print_fmt);
1321 return ret;
1324 /* Make a debugfs interface for controlling probe points */
1325 static __init int init_kprobe_trace(void)
1327 struct dentry *d_tracer;
1328 struct dentry *entry;
1330 if (register_module_notifier(&trace_kprobe_module_nb))
1331 return -EINVAL;
1333 d_tracer = tracing_init_dentry();
1334 if (!d_tracer)
1335 return 0;
1337 entry = debugfs_create_file("kprobe_events", 0644, d_tracer,
1338 NULL, &kprobe_events_ops);
1340 /* Event list interface */
1341 if (!entry)
1342 pr_warning("Could not create debugfs "
1343 "'kprobe_events' entry\n");
1345 /* Profile interface */
1346 entry = debugfs_create_file("kprobe_profile", 0444, d_tracer,
1347 NULL, &kprobe_profile_ops);
1349 if (!entry)
1350 pr_warning("Could not create debugfs "
1351 "'kprobe_profile' entry\n");
1352 return 0;
1354 fs_initcall(init_kprobe_trace);
1357 #ifdef CONFIG_FTRACE_STARTUP_TEST
1360 * The "__used" keeps gcc from removing the function symbol
1361 * from the kallsyms table.
1363 static __used int kprobe_trace_selftest_target(int a1, int a2, int a3,
1364 int a4, int a5, int a6)
1366 return a1 + a2 + a3 + a4 + a5 + a6;
1369 static struct ftrace_event_file *
1370 find_trace_probe_file(struct trace_kprobe *tk, struct trace_array *tr)
1372 struct ftrace_event_file *file;
1374 list_for_each_entry(file, &tr->events, list)
1375 if (file->event_call == &tk->tp.call)
1376 return file;
1378 return NULL;
1382 * Nobody but us can call enable_trace_kprobe/disable_trace_kprobe at this
1383 * stage, we can do this lockless.
1385 static __init int kprobe_trace_self_tests_init(void)
1387 int ret, warn = 0;
1388 int (*target)(int, int, int, int, int, int);
1389 struct trace_kprobe *tk;
1390 struct ftrace_event_file *file;
1392 target = kprobe_trace_selftest_target;
1394 pr_info("Testing kprobe tracing: ");
1396 ret = traceprobe_command("p:testprobe kprobe_trace_selftest_target "
1397 "$stack $stack0 +0($stack)",
1398 create_trace_kprobe);
1399 if (WARN_ON_ONCE(ret)) {
1400 pr_warn("error on probing function entry.\n");
1401 warn++;
1402 } else {
1403 /* Enable trace point */
1404 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1405 if (WARN_ON_ONCE(tk == NULL)) {
1406 pr_warn("error on getting new probe.\n");
1407 warn++;
1408 } else {
1409 file = find_trace_probe_file(tk, top_trace_array());
1410 if (WARN_ON_ONCE(file == NULL)) {
1411 pr_warn("error on getting probe file.\n");
1412 warn++;
1413 } else
1414 enable_trace_kprobe(tk, file);
1418 ret = traceprobe_command("r:testprobe2 kprobe_trace_selftest_target "
1419 "$retval", create_trace_kprobe);
1420 if (WARN_ON_ONCE(ret)) {
1421 pr_warn("error on probing function return.\n");
1422 warn++;
1423 } else {
1424 /* Enable trace point */
1425 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1426 if (WARN_ON_ONCE(tk == NULL)) {
1427 pr_warn("error on getting 2nd new probe.\n");
1428 warn++;
1429 } else {
1430 file = find_trace_probe_file(tk, top_trace_array());
1431 if (WARN_ON_ONCE(file == NULL)) {
1432 pr_warn("error on getting probe file.\n");
1433 warn++;
1434 } else
1435 enable_trace_kprobe(tk, file);
1439 if (warn)
1440 goto end;
1442 ret = target(1, 2, 3, 4, 5, 6);
1444 /* Disable trace points before removing it */
1445 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1446 if (WARN_ON_ONCE(tk == NULL)) {
1447 pr_warn("error on getting test probe.\n");
1448 warn++;
1449 } else {
1450 file = find_trace_probe_file(tk, top_trace_array());
1451 if (WARN_ON_ONCE(file == NULL)) {
1452 pr_warn("error on getting probe file.\n");
1453 warn++;
1454 } else
1455 disable_trace_kprobe(tk, file);
1458 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1459 if (WARN_ON_ONCE(tk == NULL)) {
1460 pr_warn("error on getting 2nd test probe.\n");
1461 warn++;
1462 } else {
1463 file = find_trace_probe_file(tk, top_trace_array());
1464 if (WARN_ON_ONCE(file == NULL)) {
1465 pr_warn("error on getting probe file.\n");
1466 warn++;
1467 } else
1468 disable_trace_kprobe(tk, file);
1471 ret = traceprobe_command("-:testprobe", create_trace_kprobe);
1472 if (WARN_ON_ONCE(ret)) {
1473 pr_warn("error on deleting a probe.\n");
1474 warn++;
1477 ret = traceprobe_command("-:testprobe2", create_trace_kprobe);
1478 if (WARN_ON_ONCE(ret)) {
1479 pr_warn("error on deleting a probe.\n");
1480 warn++;
1483 end:
1484 release_all_trace_kprobes();
1485 if (warn)
1486 pr_cont("NG: Some tests are failed. Please check them.\n");
1487 else
1488 pr_cont("OK\n");
1489 return 0;
1492 late_initcall(kprobe_trace_self_tests_init);
1494 #endif