dm thin metadata: fix __udivdi3 undefined on 32-bit
[linux/fpc-iii.git] / kernel / trace / trace_kprobe.c
blobf0ee722be520ecb33b1f0758ab28f367e5acb105
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 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 if (symbol) {
278 tk->symbol = kstrdup(symbol, GFP_KERNEL);
279 if (!tk->symbol)
280 goto error;
281 tk->rp.kp.symbol_name = tk->symbol;
282 tk->rp.kp.offset = offs;
283 } else
284 tk->rp.kp.addr = addr;
286 if (is_return)
287 tk->rp.handler = kretprobe_dispatcher;
288 else
289 tk->rp.kp.pre_handler = kprobe_dispatcher;
291 if (!event || !is_good_name(event)) {
292 ret = -EINVAL;
293 goto error;
296 tk->tp.call.class = &tk->tp.class;
297 tk->tp.call.name = kstrdup(event, GFP_KERNEL);
298 if (!tk->tp.call.name)
299 goto error;
301 if (!group || !is_good_name(group)) {
302 ret = -EINVAL;
303 goto error;
306 tk->tp.class.system = kstrdup(group, GFP_KERNEL);
307 if (!tk->tp.class.system)
308 goto error;
310 INIT_LIST_HEAD(&tk->list);
311 INIT_LIST_HEAD(&tk->tp.files);
312 return tk;
313 error:
314 kfree(tk->tp.call.name);
315 kfree(tk->symbol);
316 kfree(tk);
317 return ERR_PTR(ret);
320 static void free_trace_kprobe(struct trace_kprobe *tk)
322 int i;
324 for (i = 0; i < tk->tp.nr_args; i++)
325 traceprobe_free_probe_arg(&tk->tp.args[i]);
327 kfree(tk->tp.call.class->system);
328 kfree(tk->tp.call.name);
329 kfree(tk->symbol);
330 kfree(tk);
333 static struct trace_kprobe *find_trace_kprobe(const char *event,
334 const char *group)
336 struct trace_kprobe *tk;
338 list_for_each_entry(tk, &probe_list, list)
339 if (strcmp(trace_event_name(&tk->tp.call), event) == 0 &&
340 strcmp(tk->tp.call.class->system, group) == 0)
341 return tk;
342 return NULL;
346 * Enable trace_probe
347 * if the file is NULL, enable "perf" handler, or enable "trace" handler.
349 static int
350 enable_trace_kprobe(struct trace_kprobe *tk, struct trace_event_file *file)
352 struct event_file_link *link = NULL;
353 int ret = 0;
355 if (file) {
356 link = kmalloc(sizeof(*link), GFP_KERNEL);
357 if (!link) {
358 ret = -ENOMEM;
359 goto out;
362 link->file = file;
363 list_add_tail_rcu(&link->list, &tk->tp.files);
365 tk->tp.flags |= TP_FLAG_TRACE;
366 } else
367 tk->tp.flags |= TP_FLAG_PROFILE;
369 if (trace_probe_is_registered(&tk->tp) && !trace_kprobe_has_gone(tk)) {
370 if (trace_kprobe_is_return(tk))
371 ret = enable_kretprobe(&tk->rp);
372 else
373 ret = enable_kprobe(&tk->rp.kp);
376 if (ret) {
377 if (file) {
378 /* Notice the if is true on not WARN() */
379 if (!WARN_ON_ONCE(!link))
380 list_del_rcu(&link->list);
381 kfree(link);
382 tk->tp.flags &= ~TP_FLAG_TRACE;
383 } else {
384 tk->tp.flags &= ~TP_FLAG_PROFILE;
387 out:
388 return ret;
392 * Disable trace_probe
393 * if the file is NULL, disable "perf" handler, or disable "trace" handler.
395 static int
396 disable_trace_kprobe(struct trace_kprobe *tk, struct trace_event_file *file)
398 struct event_file_link *link = NULL;
399 int wait = 0;
400 int ret = 0;
402 if (file) {
403 link = find_event_file_link(&tk->tp, file);
404 if (!link) {
405 ret = -EINVAL;
406 goto out;
409 list_del_rcu(&link->list);
410 wait = 1;
411 if (!list_empty(&tk->tp.files))
412 goto out;
414 tk->tp.flags &= ~TP_FLAG_TRACE;
415 } else
416 tk->tp.flags &= ~TP_FLAG_PROFILE;
418 if (!trace_probe_is_enabled(&tk->tp) && trace_probe_is_registered(&tk->tp)) {
419 if (trace_kprobe_is_return(tk))
420 disable_kretprobe(&tk->rp);
421 else
422 disable_kprobe(&tk->rp.kp);
423 wait = 1;
425 out:
426 if (wait) {
428 * Synchronize with kprobe_trace_func/kretprobe_trace_func
429 * to ensure disabled (all running handlers are finished).
430 * This is not only for kfree(), but also the caller,
431 * trace_remove_event_call() supposes it for releasing
432 * event_call related objects, which will be accessed in
433 * the kprobe_trace_func/kretprobe_trace_func.
435 synchronize_sched();
436 kfree(link); /* Ignored if link == NULL */
439 return ret;
442 /* Internal register function - just handle k*probes and flags */
443 static int __register_trace_kprobe(struct trace_kprobe *tk)
445 int i, ret;
447 if (trace_probe_is_registered(&tk->tp))
448 return -EINVAL;
450 for (i = 0; i < tk->tp.nr_args; i++)
451 traceprobe_update_arg(&tk->tp.args[i]);
453 /* Set/clear disabled flag according to tp->flag */
454 if (trace_probe_is_enabled(&tk->tp))
455 tk->rp.kp.flags &= ~KPROBE_FLAG_DISABLED;
456 else
457 tk->rp.kp.flags |= KPROBE_FLAG_DISABLED;
459 if (trace_kprobe_is_return(tk))
460 ret = register_kretprobe(&tk->rp);
461 else
462 ret = register_kprobe(&tk->rp.kp);
464 if (ret == 0)
465 tk->tp.flags |= TP_FLAG_REGISTERED;
466 else {
467 pr_warning("Could not insert probe at %s+%lu: %d\n",
468 trace_kprobe_symbol(tk), trace_kprobe_offset(tk), ret);
469 if (ret == -ENOENT && trace_kprobe_is_on_module(tk)) {
470 pr_warning("This probe might be able to register after"
471 "target module is loaded. Continue.\n");
472 ret = 0;
473 } else if (ret == -EILSEQ) {
474 pr_warning("Probing address(0x%p) is not an "
475 "instruction boundary.\n",
476 tk->rp.kp.addr);
477 ret = -EINVAL;
481 return ret;
484 /* Internal unregister function - just handle k*probes and flags */
485 static void __unregister_trace_kprobe(struct trace_kprobe *tk)
487 if (trace_probe_is_registered(&tk->tp)) {
488 if (trace_kprobe_is_return(tk))
489 unregister_kretprobe(&tk->rp);
490 else
491 unregister_kprobe(&tk->rp.kp);
492 tk->tp.flags &= ~TP_FLAG_REGISTERED;
493 /* Cleanup kprobe for reuse */
494 if (tk->rp.kp.symbol_name)
495 tk->rp.kp.addr = NULL;
499 /* Unregister a trace_probe and probe_event: call with locking probe_lock */
500 static int unregister_trace_kprobe(struct trace_kprobe *tk)
502 /* Enabled event can not be unregistered */
503 if (trace_probe_is_enabled(&tk->tp))
504 return -EBUSY;
506 /* Will fail if probe is being used by ftrace or perf */
507 if (unregister_kprobe_event(tk))
508 return -EBUSY;
510 __unregister_trace_kprobe(tk);
511 list_del(&tk->list);
513 return 0;
516 /* Register a trace_probe and probe_event */
517 static int register_trace_kprobe(struct trace_kprobe *tk)
519 struct trace_kprobe *old_tk;
520 int ret;
522 mutex_lock(&probe_lock);
524 /* Delete old (same name) event if exist */
525 old_tk = find_trace_kprobe(trace_event_name(&tk->tp.call),
526 tk->tp.call.class->system);
527 if (old_tk) {
528 ret = unregister_trace_kprobe(old_tk);
529 if (ret < 0)
530 goto end;
531 free_trace_kprobe(old_tk);
534 /* Register new event */
535 ret = register_kprobe_event(tk);
536 if (ret) {
537 pr_warning("Failed to register probe event(%d)\n", ret);
538 goto end;
541 /* Register k*probe */
542 ret = __register_trace_kprobe(tk);
543 if (ret < 0)
544 unregister_kprobe_event(tk);
545 else
546 list_add_tail(&tk->list, &probe_list);
548 end:
549 mutex_unlock(&probe_lock);
550 return ret;
553 /* Module notifier call back, checking event on the module */
554 static int trace_kprobe_module_callback(struct notifier_block *nb,
555 unsigned long val, void *data)
557 struct module *mod = data;
558 struct trace_kprobe *tk;
559 int ret;
561 if (val != MODULE_STATE_COMING)
562 return NOTIFY_DONE;
564 /* Update probes on coming module */
565 mutex_lock(&probe_lock);
566 list_for_each_entry(tk, &probe_list, list) {
567 if (trace_kprobe_within_module(tk, mod)) {
568 /* Don't need to check busy - this should have gone. */
569 __unregister_trace_kprobe(tk);
570 ret = __register_trace_kprobe(tk);
571 if (ret)
572 pr_warning("Failed to re-register probe %s on"
573 "%s: %d\n",
574 trace_event_name(&tk->tp.call),
575 mod->name, ret);
578 mutex_unlock(&probe_lock);
580 return NOTIFY_DONE;
583 static struct notifier_block trace_kprobe_module_nb = {
584 .notifier_call = trace_kprobe_module_callback,
585 .priority = 1 /* Invoked after kprobe module callback */
588 static int create_trace_kprobe(int argc, char **argv)
591 * Argument syntax:
592 * - Add kprobe: p[:[GRP/]EVENT] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
593 * - Add kretprobe: r[:[GRP/]EVENT] [MOD:]KSYM[+0] [FETCHARGS]
594 * Fetch args:
595 * $retval : fetch return value
596 * $stack : fetch stack address
597 * $stackN : fetch Nth of stack (N:0-)
598 * @ADDR : fetch memory at ADDR (ADDR should be in kernel)
599 * @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
600 * %REG : fetch register REG
601 * Dereferencing memory fetch:
602 * +|-offs(ARG) : fetch memory at ARG +|- offs address.
603 * Alias name of args:
604 * NAME=FETCHARG : set NAME as alias of FETCHARG.
605 * Type of args:
606 * FETCHARG:TYPE : use TYPE instead of unsigned long.
608 struct trace_kprobe *tk;
609 int i, ret = 0;
610 bool is_return = false, is_delete = false;
611 char *symbol = NULL, *event = NULL, *group = NULL;
612 char *arg;
613 long offset = 0;
614 void *addr = NULL;
615 char buf[MAX_EVENT_NAME_LEN];
617 /* argc must be >= 1 */
618 if (argv[0][0] == 'p')
619 is_return = false;
620 else if (argv[0][0] == 'r')
621 is_return = true;
622 else if (argv[0][0] == '-')
623 is_delete = true;
624 else {
625 pr_info("Probe definition must be started with 'p', 'r' or"
626 " '-'.\n");
627 return -EINVAL;
630 if (argv[0][1] == ':') {
631 event = &argv[0][2];
632 if (strchr(event, '/')) {
633 group = event;
634 event = strchr(group, '/') + 1;
635 event[-1] = '\0';
636 if (strlen(group) == 0) {
637 pr_info("Group name is not specified\n");
638 return -EINVAL;
641 if (strlen(event) == 0) {
642 pr_info("Event name is not specified\n");
643 return -EINVAL;
646 if (!group)
647 group = KPROBE_EVENT_SYSTEM;
649 if (is_delete) {
650 if (!event) {
651 pr_info("Delete command needs an event name.\n");
652 return -EINVAL;
654 mutex_lock(&probe_lock);
655 tk = find_trace_kprobe(event, group);
656 if (!tk) {
657 mutex_unlock(&probe_lock);
658 pr_info("Event %s/%s doesn't exist.\n", group, event);
659 return -ENOENT;
661 /* delete an event */
662 ret = unregister_trace_kprobe(tk);
663 if (ret == 0)
664 free_trace_kprobe(tk);
665 mutex_unlock(&probe_lock);
666 return ret;
669 if (argc < 2) {
670 pr_info("Probe point is not specified.\n");
671 return -EINVAL;
674 /* try to parse an address. if that fails, try to read the
675 * input as a symbol. */
676 if (kstrtoul(argv[1], 0, (unsigned long *)&addr)) {
677 /* a symbol specified */
678 symbol = argv[1];
679 /* TODO: support .init module functions */
680 ret = traceprobe_split_symbol_offset(symbol, &offset);
681 if (ret || offset < 0 || offset > UINT_MAX) {
682 pr_info("Failed to parse either an address or a symbol.\n");
683 return ret;
685 if (offset && is_return) {
686 pr_info("Return probe must be used without offset.\n");
687 return -EINVAL;
689 } else if (is_return) {
690 pr_info("Return probe point must be a symbol.\n");
691 return -EINVAL;
693 argc -= 2; argv += 2;
695 /* setup a probe */
696 if (!event) {
697 /* Make a new event name */
698 if (symbol)
699 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
700 is_return ? 'r' : 'p', symbol, offset);
701 else
702 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
703 is_return ? 'r' : 'p', addr);
704 event = buf;
706 tk = alloc_trace_kprobe(group, event, addr, symbol, offset, argc,
707 is_return);
708 if (IS_ERR(tk)) {
709 pr_info("Failed to allocate trace_probe.(%d)\n",
710 (int)PTR_ERR(tk));
711 return PTR_ERR(tk);
714 /* parse arguments */
715 ret = 0;
716 for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
717 struct probe_arg *parg = &tk->tp.args[i];
719 /* Increment count for freeing args in error case */
720 tk->tp.nr_args++;
722 /* Parse argument name */
723 arg = strchr(argv[i], '=');
724 if (arg) {
725 *arg++ = '\0';
726 parg->name = kstrdup(argv[i], GFP_KERNEL);
727 } else {
728 arg = argv[i];
729 /* If argument name is omitted, set "argN" */
730 snprintf(buf, MAX_EVENT_NAME_LEN, "arg%d", i + 1);
731 parg->name = kstrdup(buf, GFP_KERNEL);
734 if (!parg->name) {
735 pr_info("Failed to allocate argument[%d] name.\n", i);
736 ret = -ENOMEM;
737 goto error;
740 if (!is_good_name(parg->name)) {
741 pr_info("Invalid argument[%d] name: %s\n",
742 i, parg->name);
743 ret = -EINVAL;
744 goto error;
747 if (traceprobe_conflict_field_name(parg->name,
748 tk->tp.args, i)) {
749 pr_info("Argument[%d] name '%s' conflicts with "
750 "another field.\n", i, argv[i]);
751 ret = -EINVAL;
752 goto error;
755 /* Parse fetch argument */
756 ret = traceprobe_parse_probe_arg(arg, &tk->tp.size, parg,
757 is_return, true,
758 kprobes_fetch_type_table);
759 if (ret) {
760 pr_info("Parse error at argument[%d]. (%d)\n", i, ret);
761 goto error;
765 ret = register_trace_kprobe(tk);
766 if (ret)
767 goto error;
768 return 0;
770 error:
771 free_trace_kprobe(tk);
772 return ret;
775 static int release_all_trace_kprobes(void)
777 struct trace_kprobe *tk;
778 int ret = 0;
780 mutex_lock(&probe_lock);
781 /* Ensure no probe is in use. */
782 list_for_each_entry(tk, &probe_list, list)
783 if (trace_probe_is_enabled(&tk->tp)) {
784 ret = -EBUSY;
785 goto end;
787 /* TODO: Use batch unregistration */
788 while (!list_empty(&probe_list)) {
789 tk = list_entry(probe_list.next, struct trace_kprobe, list);
790 ret = unregister_trace_kprobe(tk);
791 if (ret)
792 goto end;
793 free_trace_kprobe(tk);
796 end:
797 mutex_unlock(&probe_lock);
799 return ret;
802 /* Probes listing interfaces */
803 static void *probes_seq_start(struct seq_file *m, loff_t *pos)
805 mutex_lock(&probe_lock);
806 return seq_list_start(&probe_list, *pos);
809 static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
811 return seq_list_next(v, &probe_list, pos);
814 static void probes_seq_stop(struct seq_file *m, void *v)
816 mutex_unlock(&probe_lock);
819 static int probes_seq_show(struct seq_file *m, void *v)
821 struct trace_kprobe *tk = v;
822 int i;
824 seq_putc(m, trace_kprobe_is_return(tk) ? 'r' : 'p');
825 seq_printf(m, ":%s/%s", tk->tp.call.class->system,
826 trace_event_name(&tk->tp.call));
828 if (!tk->symbol)
829 seq_printf(m, " 0x%p", tk->rp.kp.addr);
830 else if (tk->rp.kp.offset)
831 seq_printf(m, " %s+%u", trace_kprobe_symbol(tk),
832 tk->rp.kp.offset);
833 else
834 seq_printf(m, " %s", trace_kprobe_symbol(tk));
836 for (i = 0; i < tk->tp.nr_args; i++)
837 seq_printf(m, " %s=%s", tk->tp.args[i].name, tk->tp.args[i].comm);
838 seq_putc(m, '\n');
840 return 0;
843 static const struct seq_operations probes_seq_op = {
844 .start = probes_seq_start,
845 .next = probes_seq_next,
846 .stop = probes_seq_stop,
847 .show = probes_seq_show
850 static int probes_open(struct inode *inode, struct file *file)
852 int ret;
854 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
855 ret = release_all_trace_kprobes();
856 if (ret < 0)
857 return ret;
860 return seq_open(file, &probes_seq_op);
863 static ssize_t probes_write(struct file *file, const char __user *buffer,
864 size_t count, loff_t *ppos)
866 return traceprobe_probes_write(file, buffer, count, ppos,
867 create_trace_kprobe);
870 static const struct file_operations kprobe_events_ops = {
871 .owner = THIS_MODULE,
872 .open = probes_open,
873 .read = seq_read,
874 .llseek = seq_lseek,
875 .release = seq_release,
876 .write = probes_write,
879 /* Probes profiling interfaces */
880 static int probes_profile_seq_show(struct seq_file *m, void *v)
882 struct trace_kprobe *tk = v;
884 seq_printf(m, " %-44s %15lu %15lu\n",
885 trace_event_name(&tk->tp.call), tk->nhit,
886 tk->rp.kp.nmissed);
888 return 0;
891 static const struct seq_operations profile_seq_op = {
892 .start = probes_seq_start,
893 .next = probes_seq_next,
894 .stop = probes_seq_stop,
895 .show = probes_profile_seq_show
898 static int profile_open(struct inode *inode, struct file *file)
900 return seq_open(file, &profile_seq_op);
903 static const struct file_operations kprobe_profile_ops = {
904 .owner = THIS_MODULE,
905 .open = profile_open,
906 .read = seq_read,
907 .llseek = seq_lseek,
908 .release = seq_release,
911 /* Kprobe handler */
912 static nokprobe_inline void
913 __kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs,
914 struct trace_event_file *trace_file)
916 struct kprobe_trace_entry_head *entry;
917 struct ring_buffer_event *event;
918 struct ring_buffer *buffer;
919 int size, dsize, pc;
920 unsigned long irq_flags;
921 struct trace_event_call *call = &tk->tp.call;
923 WARN_ON(call != trace_file->event_call);
925 if (trace_trigger_soft_disabled(trace_file))
926 return;
928 local_save_flags(irq_flags);
929 pc = preempt_count();
931 dsize = __get_data_size(&tk->tp, regs);
932 size = sizeof(*entry) + tk->tp.size + dsize;
934 event = trace_event_buffer_lock_reserve(&buffer, trace_file,
935 call->event.type,
936 size, irq_flags, pc);
937 if (!event)
938 return;
940 entry = ring_buffer_event_data(event);
941 entry->ip = (unsigned long)tk->rp.kp.addr;
942 store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
944 event_trigger_unlock_commit_regs(trace_file, buffer, event,
945 entry, irq_flags, pc, regs);
948 static void
949 kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs)
951 struct event_file_link *link;
953 list_for_each_entry_rcu(link, &tk->tp.files, list)
954 __kprobe_trace_func(tk, regs, link->file);
956 NOKPROBE_SYMBOL(kprobe_trace_func);
958 /* Kretprobe handler */
959 static nokprobe_inline void
960 __kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
961 struct pt_regs *regs,
962 struct trace_event_file *trace_file)
964 struct kretprobe_trace_entry_head *entry;
965 struct ring_buffer_event *event;
966 struct ring_buffer *buffer;
967 int size, pc, dsize;
968 unsigned long irq_flags;
969 struct trace_event_call *call = &tk->tp.call;
971 WARN_ON(call != trace_file->event_call);
973 if (trace_trigger_soft_disabled(trace_file))
974 return;
976 local_save_flags(irq_flags);
977 pc = preempt_count();
979 dsize = __get_data_size(&tk->tp, regs);
980 size = sizeof(*entry) + tk->tp.size + dsize;
982 event = trace_event_buffer_lock_reserve(&buffer, trace_file,
983 call->event.type,
984 size, irq_flags, pc);
985 if (!event)
986 return;
988 entry = ring_buffer_event_data(event);
989 entry->func = (unsigned long)tk->rp.kp.addr;
990 entry->ret_ip = (unsigned long)ri->ret_addr;
991 store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
993 event_trigger_unlock_commit_regs(trace_file, buffer, event,
994 entry, irq_flags, pc, regs);
997 static void
998 kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
999 struct pt_regs *regs)
1001 struct event_file_link *link;
1003 list_for_each_entry_rcu(link, &tk->tp.files, list)
1004 __kretprobe_trace_func(tk, ri, regs, link->file);
1006 NOKPROBE_SYMBOL(kretprobe_trace_func);
1008 /* Event entry printers */
1009 static enum print_line_t
1010 print_kprobe_event(struct trace_iterator *iter, int flags,
1011 struct trace_event *event)
1013 struct kprobe_trace_entry_head *field;
1014 struct trace_seq *s = &iter->seq;
1015 struct trace_probe *tp;
1016 u8 *data;
1017 int i;
1019 field = (struct kprobe_trace_entry_head *)iter->ent;
1020 tp = container_of(event, struct trace_probe, call.event);
1022 trace_seq_printf(s, "%s: (", trace_event_name(&tp->call));
1024 if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
1025 goto out;
1027 trace_seq_putc(s, ')');
1029 data = (u8 *)&field[1];
1030 for (i = 0; i < tp->nr_args; i++)
1031 if (!tp->args[i].type->print(s, tp->args[i].name,
1032 data + tp->args[i].offset, field))
1033 goto out;
1035 trace_seq_putc(s, '\n');
1036 out:
1037 return trace_handle_return(s);
1040 static enum print_line_t
1041 print_kretprobe_event(struct trace_iterator *iter, int flags,
1042 struct trace_event *event)
1044 struct kretprobe_trace_entry_head *field;
1045 struct trace_seq *s = &iter->seq;
1046 struct trace_probe *tp;
1047 u8 *data;
1048 int i;
1050 field = (struct kretprobe_trace_entry_head *)iter->ent;
1051 tp = container_of(event, struct trace_probe, call.event);
1053 trace_seq_printf(s, "%s: (", trace_event_name(&tp->call));
1055 if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1056 goto out;
1058 trace_seq_puts(s, " <- ");
1060 if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1061 goto out;
1063 trace_seq_putc(s, ')');
1065 data = (u8 *)&field[1];
1066 for (i = 0; i < tp->nr_args; i++)
1067 if (!tp->args[i].type->print(s, tp->args[i].name,
1068 data + tp->args[i].offset, field))
1069 goto out;
1071 trace_seq_putc(s, '\n');
1073 out:
1074 return trace_handle_return(s);
1078 static int kprobe_event_define_fields(struct trace_event_call *event_call)
1080 int ret, i;
1081 struct kprobe_trace_entry_head field;
1082 struct trace_kprobe *tk = (struct trace_kprobe *)event_call->data;
1084 DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1085 /* Set argument names as fields */
1086 for (i = 0; i < tk->tp.nr_args; i++) {
1087 struct probe_arg *parg = &tk->tp.args[i];
1089 ret = trace_define_field(event_call, parg->type->fmttype,
1090 parg->name,
1091 sizeof(field) + parg->offset,
1092 parg->type->size,
1093 parg->type->is_signed,
1094 FILTER_OTHER);
1095 if (ret)
1096 return ret;
1098 return 0;
1101 static int kretprobe_event_define_fields(struct trace_event_call *event_call)
1103 int ret, i;
1104 struct kretprobe_trace_entry_head field;
1105 struct trace_kprobe *tk = (struct trace_kprobe *)event_call->data;
1107 DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1108 DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1109 /* Set argument names as fields */
1110 for (i = 0; i < tk->tp.nr_args; i++) {
1111 struct probe_arg *parg = &tk->tp.args[i];
1113 ret = trace_define_field(event_call, parg->type->fmttype,
1114 parg->name,
1115 sizeof(field) + parg->offset,
1116 parg->type->size,
1117 parg->type->is_signed,
1118 FILTER_OTHER);
1119 if (ret)
1120 return ret;
1122 return 0;
1125 #ifdef CONFIG_PERF_EVENTS
1127 /* Kprobe profile handler */
1128 static void
1129 kprobe_perf_func(struct trace_kprobe *tk, struct pt_regs *regs)
1131 struct trace_event_call *call = &tk->tp.call;
1132 struct bpf_prog *prog = call->prog;
1133 struct kprobe_trace_entry_head *entry;
1134 struct hlist_head *head;
1135 int size, __size, dsize;
1136 int rctx;
1138 if (prog && !trace_call_bpf(prog, regs))
1139 return;
1141 head = this_cpu_ptr(call->perf_events);
1142 if (hlist_empty(head))
1143 return;
1145 dsize = __get_data_size(&tk->tp, regs);
1146 __size = sizeof(*entry) + tk->tp.size + dsize;
1147 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1148 size -= sizeof(u32);
1150 entry = perf_trace_buf_prepare(size, call->event.type, NULL, &rctx);
1151 if (!entry)
1152 return;
1154 entry->ip = (unsigned long)tk->rp.kp.addr;
1155 memset(&entry[1], 0, dsize);
1156 store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1157 perf_trace_buf_submit(entry, size, rctx, 0, 1, regs, head, NULL);
1159 NOKPROBE_SYMBOL(kprobe_perf_func);
1161 /* Kretprobe profile handler */
1162 static void
1163 kretprobe_perf_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1164 struct pt_regs *regs)
1166 struct trace_event_call *call = &tk->tp.call;
1167 struct bpf_prog *prog = call->prog;
1168 struct kretprobe_trace_entry_head *entry;
1169 struct hlist_head *head;
1170 int size, __size, dsize;
1171 int rctx;
1173 if (prog && !trace_call_bpf(prog, regs))
1174 return;
1176 head = this_cpu_ptr(call->perf_events);
1177 if (hlist_empty(head))
1178 return;
1180 dsize = __get_data_size(&tk->tp, regs);
1181 __size = sizeof(*entry) + tk->tp.size + dsize;
1182 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1183 size -= sizeof(u32);
1185 entry = perf_trace_buf_prepare(size, call->event.type, NULL, &rctx);
1186 if (!entry)
1187 return;
1189 entry->func = (unsigned long)tk->rp.kp.addr;
1190 entry->ret_ip = (unsigned long)ri->ret_addr;
1191 store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1192 perf_trace_buf_submit(entry, size, rctx, 0, 1, regs, head, NULL);
1194 NOKPROBE_SYMBOL(kretprobe_perf_func);
1195 #endif /* CONFIG_PERF_EVENTS */
1198 * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex.
1200 * kprobe_trace_self_tests_init() does enable_trace_probe/disable_trace_probe
1201 * lockless, but we can't race with this __init function.
1203 static int kprobe_register(struct trace_event_call *event,
1204 enum trace_reg type, void *data)
1206 struct trace_kprobe *tk = (struct trace_kprobe *)event->data;
1207 struct trace_event_file *file = data;
1209 switch (type) {
1210 case TRACE_REG_REGISTER:
1211 return enable_trace_kprobe(tk, file);
1212 case TRACE_REG_UNREGISTER:
1213 return disable_trace_kprobe(tk, file);
1215 #ifdef CONFIG_PERF_EVENTS
1216 case TRACE_REG_PERF_REGISTER:
1217 return enable_trace_kprobe(tk, NULL);
1218 case TRACE_REG_PERF_UNREGISTER:
1219 return disable_trace_kprobe(tk, NULL);
1220 case TRACE_REG_PERF_OPEN:
1221 case TRACE_REG_PERF_CLOSE:
1222 case TRACE_REG_PERF_ADD:
1223 case TRACE_REG_PERF_DEL:
1224 return 0;
1225 #endif
1227 return 0;
1230 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1232 struct trace_kprobe *tk = container_of(kp, struct trace_kprobe, rp.kp);
1234 tk->nhit++;
1236 if (tk->tp.flags & TP_FLAG_TRACE)
1237 kprobe_trace_func(tk, regs);
1238 #ifdef CONFIG_PERF_EVENTS
1239 if (tk->tp.flags & TP_FLAG_PROFILE)
1240 kprobe_perf_func(tk, regs);
1241 #endif
1242 return 0; /* We don't tweek kernel, so just return 0 */
1244 NOKPROBE_SYMBOL(kprobe_dispatcher);
1246 static int
1247 kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1249 struct trace_kprobe *tk = container_of(ri->rp, struct trace_kprobe, rp);
1251 tk->nhit++;
1253 if (tk->tp.flags & TP_FLAG_TRACE)
1254 kretprobe_trace_func(tk, ri, regs);
1255 #ifdef CONFIG_PERF_EVENTS
1256 if (tk->tp.flags & TP_FLAG_PROFILE)
1257 kretprobe_perf_func(tk, ri, regs);
1258 #endif
1259 return 0; /* We don't tweek kernel, so just return 0 */
1261 NOKPROBE_SYMBOL(kretprobe_dispatcher);
1263 static struct trace_event_functions kretprobe_funcs = {
1264 .trace = print_kretprobe_event
1267 static struct trace_event_functions kprobe_funcs = {
1268 .trace = print_kprobe_event
1271 static int register_kprobe_event(struct trace_kprobe *tk)
1273 struct trace_event_call *call = &tk->tp.call;
1274 int ret;
1276 /* Initialize trace_event_call */
1277 INIT_LIST_HEAD(&call->class->fields);
1278 if (trace_kprobe_is_return(tk)) {
1279 call->event.funcs = &kretprobe_funcs;
1280 call->class->define_fields = kretprobe_event_define_fields;
1281 } else {
1282 call->event.funcs = &kprobe_funcs;
1283 call->class->define_fields = kprobe_event_define_fields;
1285 if (set_print_fmt(&tk->tp, trace_kprobe_is_return(tk)) < 0)
1286 return -ENOMEM;
1287 ret = register_trace_event(&call->event);
1288 if (!ret) {
1289 kfree(call->print_fmt);
1290 return -ENODEV;
1292 call->flags = TRACE_EVENT_FL_KPROBE;
1293 call->class->reg = kprobe_register;
1294 call->data = tk;
1295 ret = trace_add_event_call(call);
1296 if (ret) {
1297 pr_info("Failed to register kprobe event: %s\n",
1298 trace_event_name(call));
1299 kfree(call->print_fmt);
1300 unregister_trace_event(&call->event);
1302 return ret;
1305 static int unregister_kprobe_event(struct trace_kprobe *tk)
1307 int ret;
1309 /* tp->event is unregistered in trace_remove_event_call() */
1310 ret = trace_remove_event_call(&tk->tp.call);
1311 if (!ret)
1312 kfree(tk->tp.call.print_fmt);
1313 return ret;
1316 /* Make a tracefs interface for controlling probe points */
1317 static __init int init_kprobe_trace(void)
1319 struct dentry *d_tracer;
1320 struct dentry *entry;
1322 if (register_module_notifier(&trace_kprobe_module_nb))
1323 return -EINVAL;
1325 d_tracer = tracing_init_dentry();
1326 if (IS_ERR(d_tracer))
1327 return 0;
1329 entry = tracefs_create_file("kprobe_events", 0644, d_tracer,
1330 NULL, &kprobe_events_ops);
1332 /* Event list interface */
1333 if (!entry)
1334 pr_warning("Could not create tracefs "
1335 "'kprobe_events' entry\n");
1337 /* Profile interface */
1338 entry = tracefs_create_file("kprobe_profile", 0444, d_tracer,
1339 NULL, &kprobe_profile_ops);
1341 if (!entry)
1342 pr_warning("Could not create tracefs "
1343 "'kprobe_profile' entry\n");
1344 return 0;
1346 fs_initcall(init_kprobe_trace);
1349 #ifdef CONFIG_FTRACE_STARTUP_TEST
1352 * The "__used" keeps gcc from removing the function symbol
1353 * from the kallsyms table.
1355 static __used int kprobe_trace_selftest_target(int a1, int a2, int a3,
1356 int a4, int a5, int a6)
1358 return a1 + a2 + a3 + a4 + a5 + a6;
1361 static struct trace_event_file *
1362 find_trace_probe_file(struct trace_kprobe *tk, struct trace_array *tr)
1364 struct trace_event_file *file;
1366 list_for_each_entry(file, &tr->events, list)
1367 if (file->event_call == &tk->tp.call)
1368 return file;
1370 return NULL;
1374 * Nobody but us can call enable_trace_kprobe/disable_trace_kprobe at this
1375 * stage, we can do this lockless.
1377 static __init int kprobe_trace_self_tests_init(void)
1379 int ret, warn = 0;
1380 int (*target)(int, int, int, int, int, int);
1381 struct trace_kprobe *tk;
1382 struct trace_event_file *file;
1384 if (tracing_is_disabled())
1385 return -ENODEV;
1387 target = kprobe_trace_selftest_target;
1389 pr_info("Testing kprobe tracing: ");
1391 ret = traceprobe_command("p:testprobe kprobe_trace_selftest_target "
1392 "$stack $stack0 +0($stack)",
1393 create_trace_kprobe);
1394 if (WARN_ON_ONCE(ret)) {
1395 pr_warn("error on probing function entry.\n");
1396 warn++;
1397 } else {
1398 /* Enable trace point */
1399 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1400 if (WARN_ON_ONCE(tk == NULL)) {
1401 pr_warn("error on getting new probe.\n");
1402 warn++;
1403 } else {
1404 file = find_trace_probe_file(tk, top_trace_array());
1405 if (WARN_ON_ONCE(file == NULL)) {
1406 pr_warn("error on getting probe file.\n");
1407 warn++;
1408 } else
1409 enable_trace_kprobe(tk, file);
1413 ret = traceprobe_command("r:testprobe2 kprobe_trace_selftest_target "
1414 "$retval", create_trace_kprobe);
1415 if (WARN_ON_ONCE(ret)) {
1416 pr_warn("error on probing function return.\n");
1417 warn++;
1418 } else {
1419 /* Enable trace point */
1420 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1421 if (WARN_ON_ONCE(tk == NULL)) {
1422 pr_warn("error on getting 2nd new probe.\n");
1423 warn++;
1424 } else {
1425 file = find_trace_probe_file(tk, top_trace_array());
1426 if (WARN_ON_ONCE(file == NULL)) {
1427 pr_warn("error on getting probe file.\n");
1428 warn++;
1429 } else
1430 enable_trace_kprobe(tk, file);
1434 if (warn)
1435 goto end;
1437 ret = target(1, 2, 3, 4, 5, 6);
1439 /* Disable trace points before removing it */
1440 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1441 if (WARN_ON_ONCE(tk == NULL)) {
1442 pr_warn("error on getting test probe.\n");
1443 warn++;
1444 } else {
1445 file = find_trace_probe_file(tk, top_trace_array());
1446 if (WARN_ON_ONCE(file == NULL)) {
1447 pr_warn("error on getting probe file.\n");
1448 warn++;
1449 } else
1450 disable_trace_kprobe(tk, file);
1453 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1454 if (WARN_ON_ONCE(tk == NULL)) {
1455 pr_warn("error on getting 2nd test probe.\n");
1456 warn++;
1457 } else {
1458 file = find_trace_probe_file(tk, top_trace_array());
1459 if (WARN_ON_ONCE(file == NULL)) {
1460 pr_warn("error on getting probe file.\n");
1461 warn++;
1462 } else
1463 disable_trace_kprobe(tk, file);
1466 ret = traceprobe_command("-:testprobe", create_trace_kprobe);
1467 if (WARN_ON_ONCE(ret)) {
1468 pr_warn("error on deleting a probe.\n");
1469 warn++;
1472 ret = traceprobe_command("-:testprobe2", create_trace_kprobe);
1473 if (WARN_ON_ONCE(ret)) {
1474 pr_warn("error on deleting a probe.\n");
1475 warn++;
1478 end:
1479 release_all_trace_kprobes();
1481 * Wait for the optimizer work to finish. Otherwise it might fiddle
1482 * with probes in already freed __init text.
1484 wait_for_kprobe_optimizer();
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