perf hists: Move sort__has_sym into struct perf_hpp_list
[linux/fpc-iii.git] / kernel / trace / trace_kprobe.c
blob919e0ddd8fcc386de7e6e168f5d2a9911118c331
1 /*
2 * Kprobes-based tracing events
4 * Created by Masami Hiramatsu <mhiramat@redhat.com>
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 as
8 * published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write to the Free Software
17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
20 #include <linux/module.h>
21 #include <linux/uaccess.h>
23 #include "trace_probe.h"
25 #define KPROBE_EVENT_SYSTEM "kprobes"
27 /**
28 * Kprobe event core functions
30 struct trace_kprobe {
31 struct list_head list;
32 struct kretprobe rp; /* Use rp.kp for kprobe use */
33 unsigned long __percpu *nhit;
34 const char *symbol; /* symbol name */
35 struct trace_probe tp;
38 #define SIZEOF_TRACE_KPROBE(n) \
39 (offsetof(struct trace_kprobe, tp.args) + \
40 (sizeof(struct probe_arg) * (n)))
43 static nokprobe_inline bool trace_kprobe_is_return(struct trace_kprobe *tk)
45 return tk->rp.handler != NULL;
48 static nokprobe_inline const char *trace_kprobe_symbol(struct trace_kprobe *tk)
50 return tk->symbol ? tk->symbol : "unknown";
53 static nokprobe_inline unsigned long trace_kprobe_offset(struct trace_kprobe *tk)
55 return tk->rp.kp.offset;
58 static nokprobe_inline bool trace_kprobe_has_gone(struct trace_kprobe *tk)
60 return !!(kprobe_gone(&tk->rp.kp));
63 static nokprobe_inline bool trace_kprobe_within_module(struct trace_kprobe *tk,
64 struct module *mod)
66 int len = strlen(mod->name);
67 const char *name = trace_kprobe_symbol(tk);
68 return strncmp(mod->name, name, len) == 0 && name[len] == ':';
71 static nokprobe_inline bool trace_kprobe_is_on_module(struct trace_kprobe *tk)
73 return !!strchr(trace_kprobe_symbol(tk), ':');
76 static int register_kprobe_event(struct trace_kprobe *tk);
77 static int unregister_kprobe_event(struct trace_kprobe *tk);
79 static DEFINE_MUTEX(probe_lock);
80 static LIST_HEAD(probe_list);
82 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
83 static int kretprobe_dispatcher(struct kretprobe_instance *ri,
84 struct pt_regs *regs);
86 /* Memory fetching by symbol */
87 struct symbol_cache {
88 char *symbol;
89 long offset;
90 unsigned long addr;
93 unsigned long update_symbol_cache(struct symbol_cache *sc)
95 sc->addr = (unsigned long)kallsyms_lookup_name(sc->symbol);
97 if (sc->addr)
98 sc->addr += sc->offset;
100 return sc->addr;
103 void free_symbol_cache(struct symbol_cache *sc)
105 kfree(sc->symbol);
106 kfree(sc);
109 struct symbol_cache *alloc_symbol_cache(const char *sym, long offset)
111 struct symbol_cache *sc;
113 if (!sym || strlen(sym) == 0)
114 return NULL;
116 sc = kzalloc(sizeof(struct symbol_cache), GFP_KERNEL);
117 if (!sc)
118 return NULL;
120 sc->symbol = kstrdup(sym, GFP_KERNEL);
121 if (!sc->symbol) {
122 kfree(sc);
123 return NULL;
125 sc->offset = offset;
126 update_symbol_cache(sc);
128 return sc;
132 * Kprobes-specific fetch functions
134 #define DEFINE_FETCH_stack(type) \
135 static void FETCH_FUNC_NAME(stack, type)(struct pt_regs *regs, \
136 void *offset, void *dest) \
138 *(type *)dest = (type)regs_get_kernel_stack_nth(regs, \
139 (unsigned int)((unsigned long)offset)); \
141 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(stack, type));
143 DEFINE_BASIC_FETCH_FUNCS(stack)
144 /* No string on the stack entry */
145 #define fetch_stack_string NULL
146 #define fetch_stack_string_size NULL
148 #define DEFINE_FETCH_memory(type) \
149 static void FETCH_FUNC_NAME(memory, type)(struct pt_regs *regs, \
150 void *addr, void *dest) \
152 type retval; \
153 if (probe_kernel_address(addr, retval)) \
154 *(type *)dest = 0; \
155 else \
156 *(type *)dest = retval; \
158 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(memory, type));
160 DEFINE_BASIC_FETCH_FUNCS(memory)
162 * Fetch a null-terminated string. Caller MUST set *(u32 *)dest with max
163 * length and relative data location.
165 static void FETCH_FUNC_NAME(memory, string)(struct pt_regs *regs,
166 void *addr, void *dest)
168 int maxlen = get_rloc_len(*(u32 *)dest);
169 u8 *dst = get_rloc_data(dest);
170 long ret;
172 if (!maxlen)
173 return;
176 * Try to get string again, since the string can be changed while
177 * probing.
179 ret = strncpy_from_unsafe(dst, addr, maxlen);
181 if (ret < 0) { /* Failed to fetch string */
182 dst[0] = '\0';
183 *(u32 *)dest = make_data_rloc(0, get_rloc_offs(*(u32 *)dest));
184 } else {
185 *(u32 *)dest = make_data_rloc(ret, get_rloc_offs(*(u32 *)dest));
188 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(memory, string));
190 /* Return the length of string -- including null terminal byte */
191 static void FETCH_FUNC_NAME(memory, string_size)(struct pt_regs *regs,
192 void *addr, void *dest)
194 mm_segment_t old_fs;
195 int ret, len = 0;
196 u8 c;
198 old_fs = get_fs();
199 set_fs(KERNEL_DS);
200 pagefault_disable();
202 do {
203 ret = __copy_from_user_inatomic(&c, (u8 *)addr + len, 1);
204 len++;
205 } while (c && ret == 0 && len < MAX_STRING_SIZE);
207 pagefault_enable();
208 set_fs(old_fs);
210 if (ret < 0) /* Failed to check the length */
211 *(u32 *)dest = 0;
212 else
213 *(u32 *)dest = len;
215 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(memory, string_size));
217 #define DEFINE_FETCH_symbol(type) \
218 void FETCH_FUNC_NAME(symbol, type)(struct pt_regs *regs, void *data, void *dest)\
220 struct symbol_cache *sc = data; \
221 if (sc->addr) \
222 fetch_memory_##type(regs, (void *)sc->addr, dest); \
223 else \
224 *(type *)dest = 0; \
226 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(symbol, type));
228 DEFINE_BASIC_FETCH_FUNCS(symbol)
229 DEFINE_FETCH_symbol(string)
230 DEFINE_FETCH_symbol(string_size)
232 /* kprobes don't support file_offset fetch methods */
233 #define fetch_file_offset_u8 NULL
234 #define fetch_file_offset_u16 NULL
235 #define fetch_file_offset_u32 NULL
236 #define fetch_file_offset_u64 NULL
237 #define fetch_file_offset_string NULL
238 #define fetch_file_offset_string_size NULL
240 /* Fetch type information table */
241 static const struct fetch_type kprobes_fetch_type_table[] = {
242 /* Special types */
243 [FETCH_TYPE_STRING] = __ASSIGN_FETCH_TYPE("string", string, string,
244 sizeof(u32), 1, "__data_loc char[]"),
245 [FETCH_TYPE_STRSIZE] = __ASSIGN_FETCH_TYPE("string_size", u32,
246 string_size, sizeof(u32), 0, "u32"),
247 /* Basic types */
248 ASSIGN_FETCH_TYPE(u8, u8, 0),
249 ASSIGN_FETCH_TYPE(u16, u16, 0),
250 ASSIGN_FETCH_TYPE(u32, u32, 0),
251 ASSIGN_FETCH_TYPE(u64, u64, 0),
252 ASSIGN_FETCH_TYPE(s8, u8, 1),
253 ASSIGN_FETCH_TYPE(s16, u16, 1),
254 ASSIGN_FETCH_TYPE(s32, u32, 1),
255 ASSIGN_FETCH_TYPE(s64, u64, 1),
257 ASSIGN_FETCH_TYPE_END
261 * Allocate new trace_probe and initialize it (including kprobes).
263 static struct trace_kprobe *alloc_trace_kprobe(const char *group,
264 const char *event,
265 void *addr,
266 const char *symbol,
267 unsigned long offs,
268 int nargs, bool is_return)
270 struct trace_kprobe *tk;
271 int ret = -ENOMEM;
273 tk = kzalloc(SIZEOF_TRACE_KPROBE(nargs), GFP_KERNEL);
274 if (!tk)
275 return ERR_PTR(ret);
277 tk->nhit = alloc_percpu(unsigned long);
278 if (!tk->nhit)
279 goto error;
281 if (symbol) {
282 tk->symbol = kstrdup(symbol, GFP_KERNEL);
283 if (!tk->symbol)
284 goto error;
285 tk->rp.kp.symbol_name = tk->symbol;
286 tk->rp.kp.offset = offs;
287 } else
288 tk->rp.kp.addr = addr;
290 if (is_return)
291 tk->rp.handler = kretprobe_dispatcher;
292 else
293 tk->rp.kp.pre_handler = kprobe_dispatcher;
295 if (!event || !is_good_name(event)) {
296 ret = -EINVAL;
297 goto error;
300 tk->tp.call.class = &tk->tp.class;
301 tk->tp.call.name = kstrdup(event, GFP_KERNEL);
302 if (!tk->tp.call.name)
303 goto error;
305 if (!group || !is_good_name(group)) {
306 ret = -EINVAL;
307 goto error;
310 tk->tp.class.system = kstrdup(group, GFP_KERNEL);
311 if (!tk->tp.class.system)
312 goto error;
314 INIT_LIST_HEAD(&tk->list);
315 INIT_LIST_HEAD(&tk->tp.files);
316 return tk;
317 error:
318 kfree(tk->tp.call.name);
319 kfree(tk->symbol);
320 free_percpu(tk->nhit);
321 kfree(tk);
322 return ERR_PTR(ret);
325 static void free_trace_kprobe(struct trace_kprobe *tk)
327 int i;
329 for (i = 0; i < tk->tp.nr_args; i++)
330 traceprobe_free_probe_arg(&tk->tp.args[i]);
332 kfree(tk->tp.call.class->system);
333 kfree(tk->tp.call.name);
334 kfree(tk->symbol);
335 free_percpu(tk->nhit);
336 kfree(tk);
339 static struct trace_kprobe *find_trace_kprobe(const char *event,
340 const char *group)
342 struct trace_kprobe *tk;
344 list_for_each_entry(tk, &probe_list, list)
345 if (strcmp(trace_event_name(&tk->tp.call), event) == 0 &&
346 strcmp(tk->tp.call.class->system, group) == 0)
347 return tk;
348 return NULL;
352 * Enable trace_probe
353 * if the file is NULL, enable "perf" handler, or enable "trace" handler.
355 static int
356 enable_trace_kprobe(struct trace_kprobe *tk, struct trace_event_file *file)
358 int ret = 0;
360 if (file) {
361 struct event_file_link *link;
363 link = kmalloc(sizeof(*link), GFP_KERNEL);
364 if (!link) {
365 ret = -ENOMEM;
366 goto out;
369 link->file = file;
370 list_add_tail_rcu(&link->list, &tk->tp.files);
372 tk->tp.flags |= TP_FLAG_TRACE;
373 } else
374 tk->tp.flags |= TP_FLAG_PROFILE;
376 if (trace_probe_is_registered(&tk->tp) && !trace_kprobe_has_gone(tk)) {
377 if (trace_kprobe_is_return(tk))
378 ret = enable_kretprobe(&tk->rp);
379 else
380 ret = enable_kprobe(&tk->rp.kp);
382 out:
383 return ret;
387 * Disable trace_probe
388 * if the file is NULL, disable "perf" handler, or disable "trace" handler.
390 static int
391 disable_trace_kprobe(struct trace_kprobe *tk, struct trace_event_file *file)
393 struct event_file_link *link = NULL;
394 int wait = 0;
395 int ret = 0;
397 if (file) {
398 link = find_event_file_link(&tk->tp, file);
399 if (!link) {
400 ret = -EINVAL;
401 goto out;
404 list_del_rcu(&link->list);
405 wait = 1;
406 if (!list_empty(&tk->tp.files))
407 goto out;
409 tk->tp.flags &= ~TP_FLAG_TRACE;
410 } else
411 tk->tp.flags &= ~TP_FLAG_PROFILE;
413 if (!trace_probe_is_enabled(&tk->tp) && trace_probe_is_registered(&tk->tp)) {
414 if (trace_kprobe_is_return(tk))
415 disable_kretprobe(&tk->rp);
416 else
417 disable_kprobe(&tk->rp.kp);
418 wait = 1;
420 out:
421 if (wait) {
423 * Synchronize with kprobe_trace_func/kretprobe_trace_func
424 * to ensure disabled (all running handlers are finished).
425 * This is not only for kfree(), but also the caller,
426 * trace_remove_event_call() supposes it for releasing
427 * event_call related objects, which will be accessed in
428 * the kprobe_trace_func/kretprobe_trace_func.
430 synchronize_sched();
431 kfree(link); /* Ignored if link == NULL */
434 return ret;
437 /* Internal register function - just handle k*probes and flags */
438 static int __register_trace_kprobe(struct trace_kprobe *tk)
440 int i, ret;
442 if (trace_probe_is_registered(&tk->tp))
443 return -EINVAL;
445 for (i = 0; i < tk->tp.nr_args; i++)
446 traceprobe_update_arg(&tk->tp.args[i]);
448 /* Set/clear disabled flag according to tp->flag */
449 if (trace_probe_is_enabled(&tk->tp))
450 tk->rp.kp.flags &= ~KPROBE_FLAG_DISABLED;
451 else
452 tk->rp.kp.flags |= KPROBE_FLAG_DISABLED;
454 if (trace_kprobe_is_return(tk))
455 ret = register_kretprobe(&tk->rp);
456 else
457 ret = register_kprobe(&tk->rp.kp);
459 if (ret == 0)
460 tk->tp.flags |= TP_FLAG_REGISTERED;
461 else {
462 pr_warn("Could not insert probe at %s+%lu: %d\n",
463 trace_kprobe_symbol(tk), trace_kprobe_offset(tk), ret);
464 if (ret == -ENOENT && trace_kprobe_is_on_module(tk)) {
465 pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
466 ret = 0;
467 } else if (ret == -EILSEQ) {
468 pr_warn("Probing address(0x%p) is not an instruction boundary.\n",
469 tk->rp.kp.addr);
470 ret = -EINVAL;
474 return ret;
477 /* Internal unregister function - just handle k*probes and flags */
478 static void __unregister_trace_kprobe(struct trace_kprobe *tk)
480 if (trace_probe_is_registered(&tk->tp)) {
481 if (trace_kprobe_is_return(tk))
482 unregister_kretprobe(&tk->rp);
483 else
484 unregister_kprobe(&tk->rp.kp);
485 tk->tp.flags &= ~TP_FLAG_REGISTERED;
486 /* Cleanup kprobe for reuse */
487 if (tk->rp.kp.symbol_name)
488 tk->rp.kp.addr = NULL;
492 /* Unregister a trace_probe and probe_event: call with locking probe_lock */
493 static int unregister_trace_kprobe(struct trace_kprobe *tk)
495 /* Enabled event can not be unregistered */
496 if (trace_probe_is_enabled(&tk->tp))
497 return -EBUSY;
499 /* Will fail if probe is being used by ftrace or perf */
500 if (unregister_kprobe_event(tk))
501 return -EBUSY;
503 __unregister_trace_kprobe(tk);
504 list_del(&tk->list);
506 return 0;
509 /* Register a trace_probe and probe_event */
510 static int register_trace_kprobe(struct trace_kprobe *tk)
512 struct trace_kprobe *old_tk;
513 int ret;
515 mutex_lock(&probe_lock);
517 /* Delete old (same name) event if exist */
518 old_tk = find_trace_kprobe(trace_event_name(&tk->tp.call),
519 tk->tp.call.class->system);
520 if (old_tk) {
521 ret = unregister_trace_kprobe(old_tk);
522 if (ret < 0)
523 goto end;
524 free_trace_kprobe(old_tk);
527 /* Register new event */
528 ret = register_kprobe_event(tk);
529 if (ret) {
530 pr_warn("Failed to register probe event(%d)\n", ret);
531 goto end;
534 /* Register k*probe */
535 ret = __register_trace_kprobe(tk);
536 if (ret < 0)
537 unregister_kprobe_event(tk);
538 else
539 list_add_tail(&tk->list, &probe_list);
541 end:
542 mutex_unlock(&probe_lock);
543 return ret;
546 /* Module notifier call back, checking event on the module */
547 static int trace_kprobe_module_callback(struct notifier_block *nb,
548 unsigned long val, void *data)
550 struct module *mod = data;
551 struct trace_kprobe *tk;
552 int ret;
554 if (val != MODULE_STATE_COMING)
555 return NOTIFY_DONE;
557 /* Update probes on coming module */
558 mutex_lock(&probe_lock);
559 list_for_each_entry(tk, &probe_list, list) {
560 if (trace_kprobe_within_module(tk, mod)) {
561 /* Don't need to check busy - this should have gone. */
562 __unregister_trace_kprobe(tk);
563 ret = __register_trace_kprobe(tk);
564 if (ret)
565 pr_warn("Failed to re-register probe %s on %s: %d\n",
566 trace_event_name(&tk->tp.call),
567 mod->name, ret);
570 mutex_unlock(&probe_lock);
572 return NOTIFY_DONE;
575 static struct notifier_block trace_kprobe_module_nb = {
576 .notifier_call = trace_kprobe_module_callback,
577 .priority = 1 /* Invoked after kprobe module callback */
580 static int create_trace_kprobe(int argc, char **argv)
583 * Argument syntax:
584 * - Add kprobe: p[:[GRP/]EVENT] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
585 * - Add kretprobe: r[:[GRP/]EVENT] [MOD:]KSYM[+0] [FETCHARGS]
586 * Fetch args:
587 * $retval : fetch return value
588 * $stack : fetch stack address
589 * $stackN : fetch Nth of stack (N:0-)
590 * @ADDR : fetch memory at ADDR (ADDR should be in kernel)
591 * @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
592 * %REG : fetch register REG
593 * Dereferencing memory fetch:
594 * +|-offs(ARG) : fetch memory at ARG +|- offs address.
595 * Alias name of args:
596 * NAME=FETCHARG : set NAME as alias of FETCHARG.
597 * Type of args:
598 * FETCHARG:TYPE : use TYPE instead of unsigned long.
600 struct trace_kprobe *tk;
601 int i, ret = 0;
602 bool is_return = false, is_delete = false;
603 char *symbol = NULL, *event = NULL, *group = NULL;
604 char *arg;
605 unsigned long offset = 0;
606 void *addr = NULL;
607 char buf[MAX_EVENT_NAME_LEN];
609 /* argc must be >= 1 */
610 if (argv[0][0] == 'p')
611 is_return = false;
612 else if (argv[0][0] == 'r')
613 is_return = true;
614 else if (argv[0][0] == '-')
615 is_delete = true;
616 else {
617 pr_info("Probe definition must be started with 'p', 'r' or"
618 " '-'.\n");
619 return -EINVAL;
622 if (argv[0][1] == ':') {
623 event = &argv[0][2];
624 if (strchr(event, '/')) {
625 group = event;
626 event = strchr(group, '/') + 1;
627 event[-1] = '\0';
628 if (strlen(group) == 0) {
629 pr_info("Group name is not specified\n");
630 return -EINVAL;
633 if (strlen(event) == 0) {
634 pr_info("Event name is not specified\n");
635 return -EINVAL;
638 if (!group)
639 group = KPROBE_EVENT_SYSTEM;
641 if (is_delete) {
642 if (!event) {
643 pr_info("Delete command needs an event name.\n");
644 return -EINVAL;
646 mutex_lock(&probe_lock);
647 tk = find_trace_kprobe(event, group);
648 if (!tk) {
649 mutex_unlock(&probe_lock);
650 pr_info("Event %s/%s doesn't exist.\n", group, event);
651 return -ENOENT;
653 /* delete an event */
654 ret = unregister_trace_kprobe(tk);
655 if (ret == 0)
656 free_trace_kprobe(tk);
657 mutex_unlock(&probe_lock);
658 return ret;
661 if (argc < 2) {
662 pr_info("Probe point is not specified.\n");
663 return -EINVAL;
665 if (isdigit(argv[1][0])) {
666 if (is_return) {
667 pr_info("Return probe point must be a symbol.\n");
668 return -EINVAL;
670 /* an address specified */
671 ret = kstrtoul(&argv[1][0], 0, (unsigned long *)&addr);
672 if (ret) {
673 pr_info("Failed to parse address.\n");
674 return ret;
676 } else {
677 /* a symbol specified */
678 symbol = argv[1];
679 /* TODO: support .init module functions */
680 ret = traceprobe_split_symbol_offset(symbol, &offset);
681 if (ret) {
682 pr_info("Failed to parse symbol.\n");
683 return ret;
685 if (offset && is_return) {
686 pr_info("Return probe must be used without offset.\n");
687 return -EINVAL;
690 argc -= 2; argv += 2;
692 /* setup a probe */
693 if (!event) {
694 /* Make a new event name */
695 if (symbol)
696 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
697 is_return ? 'r' : 'p', symbol, offset);
698 else
699 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
700 is_return ? 'r' : 'p', addr);
701 event = buf;
703 tk = alloc_trace_kprobe(group, event, addr, symbol, offset, argc,
704 is_return);
705 if (IS_ERR(tk)) {
706 pr_info("Failed to allocate trace_probe.(%d)\n",
707 (int)PTR_ERR(tk));
708 return PTR_ERR(tk);
711 /* parse arguments */
712 ret = 0;
713 for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
714 struct probe_arg *parg = &tk->tp.args[i];
716 /* Increment count for freeing args in error case */
717 tk->tp.nr_args++;
719 /* Parse argument name */
720 arg = strchr(argv[i], '=');
721 if (arg) {
722 *arg++ = '\0';
723 parg->name = kstrdup(argv[i], GFP_KERNEL);
724 } else {
725 arg = argv[i];
726 /* If argument name is omitted, set "argN" */
727 snprintf(buf, MAX_EVENT_NAME_LEN, "arg%d", i + 1);
728 parg->name = kstrdup(buf, GFP_KERNEL);
731 if (!parg->name) {
732 pr_info("Failed to allocate argument[%d] name.\n", i);
733 ret = -ENOMEM;
734 goto error;
737 if (!is_good_name(parg->name)) {
738 pr_info("Invalid argument[%d] name: %s\n",
739 i, parg->name);
740 ret = -EINVAL;
741 goto error;
744 if (traceprobe_conflict_field_name(parg->name,
745 tk->tp.args, i)) {
746 pr_info("Argument[%d] name '%s' conflicts with "
747 "another field.\n", i, argv[i]);
748 ret = -EINVAL;
749 goto error;
752 /* Parse fetch argument */
753 ret = traceprobe_parse_probe_arg(arg, &tk->tp.size, parg,
754 is_return, true,
755 kprobes_fetch_type_table);
756 if (ret) {
757 pr_info("Parse error at argument[%d]. (%d)\n", i, ret);
758 goto error;
762 ret = register_trace_kprobe(tk);
763 if (ret)
764 goto error;
765 return 0;
767 error:
768 free_trace_kprobe(tk);
769 return ret;
772 static int release_all_trace_kprobes(void)
774 struct trace_kprobe *tk;
775 int ret = 0;
777 mutex_lock(&probe_lock);
778 /* Ensure no probe is in use. */
779 list_for_each_entry(tk, &probe_list, list)
780 if (trace_probe_is_enabled(&tk->tp)) {
781 ret = -EBUSY;
782 goto end;
784 /* TODO: Use batch unregistration */
785 while (!list_empty(&probe_list)) {
786 tk = list_entry(probe_list.next, struct trace_kprobe, list);
787 ret = unregister_trace_kprobe(tk);
788 if (ret)
789 goto end;
790 free_trace_kprobe(tk);
793 end:
794 mutex_unlock(&probe_lock);
796 return ret;
799 /* Probes listing interfaces */
800 static void *probes_seq_start(struct seq_file *m, loff_t *pos)
802 mutex_lock(&probe_lock);
803 return seq_list_start(&probe_list, *pos);
806 static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
808 return seq_list_next(v, &probe_list, pos);
811 static void probes_seq_stop(struct seq_file *m, void *v)
813 mutex_unlock(&probe_lock);
816 static int probes_seq_show(struct seq_file *m, void *v)
818 struct trace_kprobe *tk = v;
819 int i;
821 seq_putc(m, trace_kprobe_is_return(tk) ? 'r' : 'p');
822 seq_printf(m, ":%s/%s", tk->tp.call.class->system,
823 trace_event_name(&tk->tp.call));
825 if (!tk->symbol)
826 seq_printf(m, " 0x%p", tk->rp.kp.addr);
827 else if (tk->rp.kp.offset)
828 seq_printf(m, " %s+%u", trace_kprobe_symbol(tk),
829 tk->rp.kp.offset);
830 else
831 seq_printf(m, " %s", trace_kprobe_symbol(tk));
833 for (i = 0; i < tk->tp.nr_args; i++)
834 seq_printf(m, " %s=%s", tk->tp.args[i].name, tk->tp.args[i].comm);
835 seq_putc(m, '\n');
837 return 0;
840 static const struct seq_operations probes_seq_op = {
841 .start = probes_seq_start,
842 .next = probes_seq_next,
843 .stop = probes_seq_stop,
844 .show = probes_seq_show
847 static int probes_open(struct inode *inode, struct file *file)
849 int ret;
851 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
852 ret = release_all_trace_kprobes();
853 if (ret < 0)
854 return ret;
857 return seq_open(file, &probes_seq_op);
860 static ssize_t probes_write(struct file *file, const char __user *buffer,
861 size_t count, loff_t *ppos)
863 return traceprobe_probes_write(file, buffer, count, ppos,
864 create_trace_kprobe);
867 static const struct file_operations kprobe_events_ops = {
868 .owner = THIS_MODULE,
869 .open = probes_open,
870 .read = seq_read,
871 .llseek = seq_lseek,
872 .release = seq_release,
873 .write = probes_write,
876 /* Probes profiling interfaces */
877 static int probes_profile_seq_show(struct seq_file *m, void *v)
879 struct trace_kprobe *tk = v;
880 unsigned long nhit = 0;
881 int cpu;
883 for_each_possible_cpu(cpu)
884 nhit += *per_cpu_ptr(tk->nhit, cpu);
886 seq_printf(m, " %-44s %15lu %15lu\n",
887 trace_event_name(&tk->tp.call), nhit,
888 tk->rp.kp.nmissed);
890 return 0;
893 static const struct seq_operations profile_seq_op = {
894 .start = probes_seq_start,
895 .next = probes_seq_next,
896 .stop = probes_seq_stop,
897 .show = probes_profile_seq_show
900 static int profile_open(struct inode *inode, struct file *file)
902 return seq_open(file, &profile_seq_op);
905 static const struct file_operations kprobe_profile_ops = {
906 .owner = THIS_MODULE,
907 .open = profile_open,
908 .read = seq_read,
909 .llseek = seq_lseek,
910 .release = seq_release,
913 /* Kprobe handler */
914 static nokprobe_inline void
915 __kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs,
916 struct trace_event_file *trace_file)
918 struct kprobe_trace_entry_head *entry;
919 struct ring_buffer_event *event;
920 struct ring_buffer *buffer;
921 int size, dsize, pc;
922 unsigned long irq_flags;
923 struct trace_event_call *call = &tk->tp.call;
925 WARN_ON(call != trace_file->event_call);
927 if (trace_trigger_soft_disabled(trace_file))
928 return;
930 local_save_flags(irq_flags);
931 pc = preempt_count();
933 dsize = __get_data_size(&tk->tp, regs);
934 size = sizeof(*entry) + tk->tp.size + dsize;
936 event = trace_event_buffer_lock_reserve(&buffer, trace_file,
937 call->event.type,
938 size, irq_flags, pc);
939 if (!event)
940 return;
942 entry = ring_buffer_event_data(event);
943 entry->ip = (unsigned long)tk->rp.kp.addr;
944 store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
946 event_trigger_unlock_commit_regs(trace_file, buffer, event,
947 entry, irq_flags, pc, regs);
950 static void
951 kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs)
953 struct event_file_link *link;
955 list_for_each_entry_rcu(link, &tk->tp.files, list)
956 __kprobe_trace_func(tk, regs, link->file);
958 NOKPROBE_SYMBOL(kprobe_trace_func);
960 /* Kretprobe handler */
961 static nokprobe_inline void
962 __kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
963 struct pt_regs *regs,
964 struct trace_event_file *trace_file)
966 struct kretprobe_trace_entry_head *entry;
967 struct ring_buffer_event *event;
968 struct ring_buffer *buffer;
969 int size, pc, dsize;
970 unsigned long irq_flags;
971 struct trace_event_call *call = &tk->tp.call;
973 WARN_ON(call != trace_file->event_call);
975 if (trace_trigger_soft_disabled(trace_file))
976 return;
978 local_save_flags(irq_flags);
979 pc = preempt_count();
981 dsize = __get_data_size(&tk->tp, regs);
982 size = sizeof(*entry) + tk->tp.size + dsize;
984 event = trace_event_buffer_lock_reserve(&buffer, trace_file,
985 call->event.type,
986 size, irq_flags, pc);
987 if (!event)
988 return;
990 entry = ring_buffer_event_data(event);
991 entry->func = (unsigned long)tk->rp.kp.addr;
992 entry->ret_ip = (unsigned long)ri->ret_addr;
993 store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
995 event_trigger_unlock_commit_regs(trace_file, buffer, event,
996 entry, irq_flags, pc, regs);
999 static void
1000 kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1001 struct pt_regs *regs)
1003 struct event_file_link *link;
1005 list_for_each_entry_rcu(link, &tk->tp.files, list)
1006 __kretprobe_trace_func(tk, ri, regs, link->file);
1008 NOKPROBE_SYMBOL(kretprobe_trace_func);
1010 /* Event entry printers */
1011 static enum print_line_t
1012 print_kprobe_event(struct trace_iterator *iter, int flags,
1013 struct trace_event *event)
1015 struct kprobe_trace_entry_head *field;
1016 struct trace_seq *s = &iter->seq;
1017 struct trace_probe *tp;
1018 u8 *data;
1019 int i;
1021 field = (struct kprobe_trace_entry_head *)iter->ent;
1022 tp = container_of(event, struct trace_probe, call.event);
1024 trace_seq_printf(s, "%s: (", trace_event_name(&tp->call));
1026 if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
1027 goto out;
1029 trace_seq_putc(s, ')');
1031 data = (u8 *)&field[1];
1032 for (i = 0; i < tp->nr_args; i++)
1033 if (!tp->args[i].type->print(s, tp->args[i].name,
1034 data + tp->args[i].offset, field))
1035 goto out;
1037 trace_seq_putc(s, '\n');
1038 out:
1039 return trace_handle_return(s);
1042 static enum print_line_t
1043 print_kretprobe_event(struct trace_iterator *iter, int flags,
1044 struct trace_event *event)
1046 struct kretprobe_trace_entry_head *field;
1047 struct trace_seq *s = &iter->seq;
1048 struct trace_probe *tp;
1049 u8 *data;
1050 int i;
1052 field = (struct kretprobe_trace_entry_head *)iter->ent;
1053 tp = container_of(event, struct trace_probe, call.event);
1055 trace_seq_printf(s, "%s: (", trace_event_name(&tp->call));
1057 if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1058 goto out;
1060 trace_seq_puts(s, " <- ");
1062 if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1063 goto out;
1065 trace_seq_putc(s, ')');
1067 data = (u8 *)&field[1];
1068 for (i = 0; i < tp->nr_args; i++)
1069 if (!tp->args[i].type->print(s, tp->args[i].name,
1070 data + tp->args[i].offset, field))
1071 goto out;
1073 trace_seq_putc(s, '\n');
1075 out:
1076 return trace_handle_return(s);
1080 static int kprobe_event_define_fields(struct trace_event_call *event_call)
1082 int ret, i;
1083 struct kprobe_trace_entry_head field;
1084 struct trace_kprobe *tk = (struct trace_kprobe *)event_call->data;
1086 DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1087 /* Set argument names as fields */
1088 for (i = 0; i < tk->tp.nr_args; i++) {
1089 struct probe_arg *parg = &tk->tp.args[i];
1091 ret = trace_define_field(event_call, parg->type->fmttype,
1092 parg->name,
1093 sizeof(field) + parg->offset,
1094 parg->type->size,
1095 parg->type->is_signed,
1096 FILTER_OTHER);
1097 if (ret)
1098 return ret;
1100 return 0;
1103 static int kretprobe_event_define_fields(struct trace_event_call *event_call)
1105 int ret, i;
1106 struct kretprobe_trace_entry_head field;
1107 struct trace_kprobe *tk = (struct trace_kprobe *)event_call->data;
1109 DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1110 DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1111 /* Set argument names as fields */
1112 for (i = 0; i < tk->tp.nr_args; i++) {
1113 struct probe_arg *parg = &tk->tp.args[i];
1115 ret = trace_define_field(event_call, parg->type->fmttype,
1116 parg->name,
1117 sizeof(field) + parg->offset,
1118 parg->type->size,
1119 parg->type->is_signed,
1120 FILTER_OTHER);
1121 if (ret)
1122 return ret;
1124 return 0;
1127 #ifdef CONFIG_PERF_EVENTS
1129 /* Kprobe profile handler */
1130 static void
1131 kprobe_perf_func(struct trace_kprobe *tk, struct pt_regs *regs)
1133 struct trace_event_call *call = &tk->tp.call;
1134 struct bpf_prog *prog = call->prog;
1135 struct kprobe_trace_entry_head *entry;
1136 struct hlist_head *head;
1137 int size, __size, dsize;
1138 int rctx;
1140 if (prog && !trace_call_bpf(prog, regs))
1141 return;
1143 head = this_cpu_ptr(call->perf_events);
1144 if (hlist_empty(head))
1145 return;
1147 dsize = __get_data_size(&tk->tp, regs);
1148 __size = sizeof(*entry) + tk->tp.size + dsize;
1149 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1150 size -= sizeof(u32);
1152 entry = perf_trace_buf_prepare(size, call->event.type, NULL, &rctx);
1153 if (!entry)
1154 return;
1156 entry->ip = (unsigned long)tk->rp.kp.addr;
1157 memset(&entry[1], 0, dsize);
1158 store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1159 perf_trace_buf_submit(entry, size, rctx, 0, 1, regs, head, NULL);
1161 NOKPROBE_SYMBOL(kprobe_perf_func);
1163 /* Kretprobe profile handler */
1164 static void
1165 kretprobe_perf_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1166 struct pt_regs *regs)
1168 struct trace_event_call *call = &tk->tp.call;
1169 struct bpf_prog *prog = call->prog;
1170 struct kretprobe_trace_entry_head *entry;
1171 struct hlist_head *head;
1172 int size, __size, dsize;
1173 int rctx;
1175 if (prog && !trace_call_bpf(prog, regs))
1176 return;
1178 head = this_cpu_ptr(call->perf_events);
1179 if (hlist_empty(head))
1180 return;
1182 dsize = __get_data_size(&tk->tp, regs);
1183 __size = sizeof(*entry) + tk->tp.size + dsize;
1184 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1185 size -= sizeof(u32);
1187 entry = perf_trace_buf_prepare(size, call->event.type, NULL, &rctx);
1188 if (!entry)
1189 return;
1191 entry->func = (unsigned long)tk->rp.kp.addr;
1192 entry->ret_ip = (unsigned long)ri->ret_addr;
1193 store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1194 perf_trace_buf_submit(entry, size, rctx, 0, 1, regs, head, NULL);
1196 NOKPROBE_SYMBOL(kretprobe_perf_func);
1197 #endif /* CONFIG_PERF_EVENTS */
1200 * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex.
1202 * kprobe_trace_self_tests_init() does enable_trace_probe/disable_trace_probe
1203 * lockless, but we can't race with this __init function.
1205 static int kprobe_register(struct trace_event_call *event,
1206 enum trace_reg type, void *data)
1208 struct trace_kprobe *tk = (struct trace_kprobe *)event->data;
1209 struct trace_event_file *file = data;
1211 switch (type) {
1212 case TRACE_REG_REGISTER:
1213 return enable_trace_kprobe(tk, file);
1214 case TRACE_REG_UNREGISTER:
1215 return disable_trace_kprobe(tk, file);
1217 #ifdef CONFIG_PERF_EVENTS
1218 case TRACE_REG_PERF_REGISTER:
1219 return enable_trace_kprobe(tk, NULL);
1220 case TRACE_REG_PERF_UNREGISTER:
1221 return disable_trace_kprobe(tk, NULL);
1222 case TRACE_REG_PERF_OPEN:
1223 case TRACE_REG_PERF_CLOSE:
1224 case TRACE_REG_PERF_ADD:
1225 case TRACE_REG_PERF_DEL:
1226 return 0;
1227 #endif
1229 return 0;
1232 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1234 struct trace_kprobe *tk = container_of(kp, struct trace_kprobe, rp.kp);
1236 raw_cpu_inc(*tk->nhit);
1238 if (tk->tp.flags & TP_FLAG_TRACE)
1239 kprobe_trace_func(tk, regs);
1240 #ifdef CONFIG_PERF_EVENTS
1241 if (tk->tp.flags & TP_FLAG_PROFILE)
1242 kprobe_perf_func(tk, regs);
1243 #endif
1244 return 0; /* We don't tweek kernel, so just return 0 */
1246 NOKPROBE_SYMBOL(kprobe_dispatcher);
1248 static int
1249 kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1251 struct trace_kprobe *tk = container_of(ri->rp, struct trace_kprobe, rp);
1253 raw_cpu_inc(*tk->nhit);
1255 if (tk->tp.flags & TP_FLAG_TRACE)
1256 kretprobe_trace_func(tk, ri, regs);
1257 #ifdef CONFIG_PERF_EVENTS
1258 if (tk->tp.flags & TP_FLAG_PROFILE)
1259 kretprobe_perf_func(tk, ri, regs);
1260 #endif
1261 return 0; /* We don't tweek kernel, so just return 0 */
1263 NOKPROBE_SYMBOL(kretprobe_dispatcher);
1265 static struct trace_event_functions kretprobe_funcs = {
1266 .trace = print_kretprobe_event
1269 static struct trace_event_functions kprobe_funcs = {
1270 .trace = print_kprobe_event
1273 static int register_kprobe_event(struct trace_kprobe *tk)
1275 struct trace_event_call *call = &tk->tp.call;
1276 int ret;
1278 /* Initialize trace_event_call */
1279 INIT_LIST_HEAD(&call->class->fields);
1280 if (trace_kprobe_is_return(tk)) {
1281 call->event.funcs = &kretprobe_funcs;
1282 call->class->define_fields = kretprobe_event_define_fields;
1283 } else {
1284 call->event.funcs = &kprobe_funcs;
1285 call->class->define_fields = kprobe_event_define_fields;
1287 if (set_print_fmt(&tk->tp, trace_kprobe_is_return(tk)) < 0)
1288 return -ENOMEM;
1289 ret = register_trace_event(&call->event);
1290 if (!ret) {
1291 kfree(call->print_fmt);
1292 return -ENODEV;
1294 call->flags = TRACE_EVENT_FL_KPROBE;
1295 call->class->reg = kprobe_register;
1296 call->data = tk;
1297 ret = trace_add_event_call(call);
1298 if (ret) {
1299 pr_info("Failed to register kprobe event: %s\n",
1300 trace_event_name(call));
1301 kfree(call->print_fmt);
1302 unregister_trace_event(&call->event);
1304 return ret;
1307 static int unregister_kprobe_event(struct trace_kprobe *tk)
1309 int ret;
1311 /* tp->event is unregistered in trace_remove_event_call() */
1312 ret = trace_remove_event_call(&tk->tp.call);
1313 if (!ret)
1314 kfree(tk->tp.call.print_fmt);
1315 return ret;
1318 /* Make a tracefs interface for controlling probe points */
1319 static __init int init_kprobe_trace(void)
1321 struct dentry *d_tracer;
1322 struct dentry *entry;
1324 if (register_module_notifier(&trace_kprobe_module_nb))
1325 return -EINVAL;
1327 d_tracer = tracing_init_dentry();
1328 if (IS_ERR(d_tracer))
1329 return 0;
1331 entry = tracefs_create_file("kprobe_events", 0644, d_tracer,
1332 NULL, &kprobe_events_ops);
1334 /* Event list interface */
1335 if (!entry)
1336 pr_warn("Could not create tracefs 'kprobe_events' entry\n");
1338 /* Profile interface */
1339 entry = tracefs_create_file("kprobe_profile", 0444, d_tracer,
1340 NULL, &kprobe_profile_ops);
1342 if (!entry)
1343 pr_warn("Could not create tracefs '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();
1480 if (warn)
1481 pr_cont("NG: Some tests are failed. Please check them.\n");
1482 else
1483 pr_cont("OK\n");
1484 return 0;
1487 late_initcall(kprobe_trace_self_tests_init);
1489 #endif