mm/writeback: commit reason of WB_REASON_FORKER_THREAD mismatch name
[linux-2.6.git] / kernel / trace / trace_kprobe.c
blob9f46e98ba8f22a712962ccc68e17b34f166439ba
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_probe {
31 struct list_head list;
32 struct kretprobe rp; /* Use rp.kp for kprobe use */
33 unsigned long nhit;
34 unsigned int flags; /* For TP_FLAG_* */
35 const char *symbol; /* symbol name */
36 struct ftrace_event_class class;
37 struct ftrace_event_call call;
38 struct ftrace_event_file * __rcu *files;
39 ssize_t size; /* trace entry size */
40 unsigned int nr_args;
41 struct probe_arg args[];
44 #define SIZEOF_TRACE_PROBE(n) \
45 (offsetof(struct trace_probe, args) + \
46 (sizeof(struct probe_arg) * (n)))
49 static __kprobes bool trace_probe_is_return(struct trace_probe *tp)
51 return tp->rp.handler != NULL;
54 static __kprobes const char *trace_probe_symbol(struct trace_probe *tp)
56 return tp->symbol ? tp->symbol : "unknown";
59 static __kprobes unsigned long trace_probe_offset(struct trace_probe *tp)
61 return tp->rp.kp.offset;
64 static __kprobes bool trace_probe_is_enabled(struct trace_probe *tp)
66 return !!(tp->flags & (TP_FLAG_TRACE | TP_FLAG_PROFILE));
69 static __kprobes bool trace_probe_is_registered(struct trace_probe *tp)
71 return !!(tp->flags & TP_FLAG_REGISTERED);
74 static __kprobes bool trace_probe_has_gone(struct trace_probe *tp)
76 return !!(kprobe_gone(&tp->rp.kp));
79 static __kprobes bool trace_probe_within_module(struct trace_probe *tp,
80 struct module *mod)
82 int len = strlen(mod->name);
83 const char *name = trace_probe_symbol(tp);
84 return strncmp(mod->name, name, len) == 0 && name[len] == ':';
87 static __kprobes bool trace_probe_is_on_module(struct trace_probe *tp)
89 return !!strchr(trace_probe_symbol(tp), ':');
92 static int register_probe_event(struct trace_probe *tp);
93 static void unregister_probe_event(struct trace_probe *tp);
95 static DEFINE_MUTEX(probe_lock);
96 static LIST_HEAD(probe_list);
98 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
99 static int kretprobe_dispatcher(struct kretprobe_instance *ri,
100 struct pt_regs *regs);
103 * Allocate new trace_probe and initialize it (including kprobes).
105 static struct trace_probe *alloc_trace_probe(const char *group,
106 const char *event,
107 void *addr,
108 const char *symbol,
109 unsigned long offs,
110 int nargs, bool is_return)
112 struct trace_probe *tp;
113 int ret = -ENOMEM;
115 tp = kzalloc(SIZEOF_TRACE_PROBE(nargs), GFP_KERNEL);
116 if (!tp)
117 return ERR_PTR(ret);
119 if (symbol) {
120 tp->symbol = kstrdup(symbol, GFP_KERNEL);
121 if (!tp->symbol)
122 goto error;
123 tp->rp.kp.symbol_name = tp->symbol;
124 tp->rp.kp.offset = offs;
125 } else
126 tp->rp.kp.addr = addr;
128 if (is_return)
129 tp->rp.handler = kretprobe_dispatcher;
130 else
131 tp->rp.kp.pre_handler = kprobe_dispatcher;
133 if (!event || !is_good_name(event)) {
134 ret = -EINVAL;
135 goto error;
138 tp->call.class = &tp->class;
139 tp->call.name = kstrdup(event, GFP_KERNEL);
140 if (!tp->call.name)
141 goto error;
143 if (!group || !is_good_name(group)) {
144 ret = -EINVAL;
145 goto error;
148 tp->class.system = kstrdup(group, GFP_KERNEL);
149 if (!tp->class.system)
150 goto error;
152 INIT_LIST_HEAD(&tp->list);
153 return tp;
154 error:
155 kfree(tp->call.name);
156 kfree(tp->symbol);
157 kfree(tp);
158 return ERR_PTR(ret);
161 static void free_trace_probe(struct trace_probe *tp)
163 int i;
165 for (i = 0; i < tp->nr_args; i++)
166 traceprobe_free_probe_arg(&tp->args[i]);
168 kfree(tp->call.class->system);
169 kfree(tp->call.name);
170 kfree(tp->symbol);
171 kfree(tp);
174 static struct trace_probe *find_trace_probe(const char *event,
175 const char *group)
177 struct trace_probe *tp;
179 list_for_each_entry(tp, &probe_list, list)
180 if (strcmp(tp->call.name, event) == 0 &&
181 strcmp(tp->call.class->system, group) == 0)
182 return tp;
183 return NULL;
186 static int trace_probe_nr_files(struct trace_probe *tp)
188 struct ftrace_event_file **file;
189 int ret = 0;
192 * Since all tp->files updater is protected by probe_enable_lock,
193 * we don't need to lock an rcu_read_lock.
195 file = rcu_dereference_raw(tp->files);
196 if (file)
197 while (*(file++))
198 ret++;
200 return ret;
203 static DEFINE_MUTEX(probe_enable_lock);
206 * Enable trace_probe
207 * if the file is NULL, enable "perf" handler, or enable "trace" handler.
209 static int
210 enable_trace_probe(struct trace_probe *tp, struct ftrace_event_file *file)
212 int ret = 0;
214 mutex_lock(&probe_enable_lock);
216 if (file) {
217 struct ftrace_event_file **new, **old;
218 int n = trace_probe_nr_files(tp);
220 old = rcu_dereference_raw(tp->files);
221 /* 1 is for new one and 1 is for stopper */
222 new = kzalloc((n + 2) * sizeof(struct ftrace_event_file *),
223 GFP_KERNEL);
224 if (!new) {
225 ret = -ENOMEM;
226 goto out_unlock;
228 memcpy(new, old, n * sizeof(struct ftrace_event_file *));
229 new[n] = file;
230 /* The last one keeps a NULL */
232 rcu_assign_pointer(tp->files, new);
233 tp->flags |= TP_FLAG_TRACE;
235 if (old) {
236 /* Make sure the probe is done with old files */
237 synchronize_sched();
238 kfree(old);
240 } else
241 tp->flags |= TP_FLAG_PROFILE;
243 if (trace_probe_is_enabled(tp) && trace_probe_is_registered(tp) &&
244 !trace_probe_has_gone(tp)) {
245 if (trace_probe_is_return(tp))
246 ret = enable_kretprobe(&tp->rp);
247 else
248 ret = enable_kprobe(&tp->rp.kp);
251 out_unlock:
252 mutex_unlock(&probe_enable_lock);
254 return ret;
257 static int
258 trace_probe_file_index(struct trace_probe *tp, struct ftrace_event_file *file)
260 struct ftrace_event_file **files;
261 int i;
264 * Since all tp->files updater is protected by probe_enable_lock,
265 * we don't need to lock an rcu_read_lock.
267 files = rcu_dereference_raw(tp->files);
268 if (files) {
269 for (i = 0; files[i]; i++)
270 if (files[i] == file)
271 return i;
274 return -1;
278 * Disable trace_probe
279 * if the file is NULL, disable "perf" handler, or disable "trace" handler.
281 static int
282 disable_trace_probe(struct trace_probe *tp, struct ftrace_event_file *file)
284 int ret = 0;
286 mutex_lock(&probe_enable_lock);
288 if (file) {
289 struct ftrace_event_file **new, **old;
290 int n = trace_probe_nr_files(tp);
291 int i, j;
293 old = rcu_dereference_raw(tp->files);
294 if (n == 0 || trace_probe_file_index(tp, file) < 0) {
295 ret = -EINVAL;
296 goto out_unlock;
299 if (n == 1) { /* Remove the last file */
300 tp->flags &= ~TP_FLAG_TRACE;
301 new = NULL;
302 } else {
303 new = kzalloc(n * sizeof(struct ftrace_event_file *),
304 GFP_KERNEL);
305 if (!new) {
306 ret = -ENOMEM;
307 goto out_unlock;
310 /* This copy & check loop copies the NULL stopper too */
311 for (i = 0, j = 0; j < n && i < n + 1; i++)
312 if (old[i] != file)
313 new[j++] = old[i];
316 rcu_assign_pointer(tp->files, new);
318 /* Make sure the probe is done with old files */
319 synchronize_sched();
320 kfree(old);
321 } else
322 tp->flags &= ~TP_FLAG_PROFILE;
324 if (!trace_probe_is_enabled(tp) && trace_probe_is_registered(tp)) {
325 if (trace_probe_is_return(tp))
326 disable_kretprobe(&tp->rp);
327 else
328 disable_kprobe(&tp->rp.kp);
331 out_unlock:
332 mutex_unlock(&probe_enable_lock);
334 return ret;
337 /* Internal register function - just handle k*probes and flags */
338 static int __register_trace_probe(struct trace_probe *tp)
340 int i, ret;
342 if (trace_probe_is_registered(tp))
343 return -EINVAL;
345 for (i = 0; i < tp->nr_args; i++)
346 traceprobe_update_arg(&tp->args[i]);
348 /* Set/clear disabled flag according to tp->flag */
349 if (trace_probe_is_enabled(tp))
350 tp->rp.kp.flags &= ~KPROBE_FLAG_DISABLED;
351 else
352 tp->rp.kp.flags |= KPROBE_FLAG_DISABLED;
354 if (trace_probe_is_return(tp))
355 ret = register_kretprobe(&tp->rp);
356 else
357 ret = register_kprobe(&tp->rp.kp);
359 if (ret == 0)
360 tp->flags |= TP_FLAG_REGISTERED;
361 else {
362 pr_warning("Could not insert probe at %s+%lu: %d\n",
363 trace_probe_symbol(tp), trace_probe_offset(tp), ret);
364 if (ret == -ENOENT && trace_probe_is_on_module(tp)) {
365 pr_warning("This probe might be able to register after"
366 "target module is loaded. Continue.\n");
367 ret = 0;
368 } else if (ret == -EILSEQ) {
369 pr_warning("Probing address(0x%p) is not an "
370 "instruction boundary.\n",
371 tp->rp.kp.addr);
372 ret = -EINVAL;
376 return ret;
379 /* Internal unregister function - just handle k*probes and flags */
380 static void __unregister_trace_probe(struct trace_probe *tp)
382 if (trace_probe_is_registered(tp)) {
383 if (trace_probe_is_return(tp))
384 unregister_kretprobe(&tp->rp);
385 else
386 unregister_kprobe(&tp->rp.kp);
387 tp->flags &= ~TP_FLAG_REGISTERED;
388 /* Cleanup kprobe for reuse */
389 if (tp->rp.kp.symbol_name)
390 tp->rp.kp.addr = NULL;
394 /* Unregister a trace_probe and probe_event: call with locking probe_lock */
395 static int unregister_trace_probe(struct trace_probe *tp)
397 /* Enabled event can not be unregistered */
398 if (trace_probe_is_enabled(tp))
399 return -EBUSY;
401 __unregister_trace_probe(tp);
402 list_del(&tp->list);
403 unregister_probe_event(tp);
405 return 0;
408 /* Register a trace_probe and probe_event */
409 static int register_trace_probe(struct trace_probe *tp)
411 struct trace_probe *old_tp;
412 int ret;
414 mutex_lock(&probe_lock);
416 /* Delete old (same name) event if exist */
417 old_tp = find_trace_probe(tp->call.name, tp->call.class->system);
418 if (old_tp) {
419 ret = unregister_trace_probe(old_tp);
420 if (ret < 0)
421 goto end;
422 free_trace_probe(old_tp);
425 /* Register new event */
426 ret = register_probe_event(tp);
427 if (ret) {
428 pr_warning("Failed to register probe event(%d)\n", ret);
429 goto end;
432 /* Register k*probe */
433 ret = __register_trace_probe(tp);
434 if (ret < 0)
435 unregister_probe_event(tp);
436 else
437 list_add_tail(&tp->list, &probe_list);
439 end:
440 mutex_unlock(&probe_lock);
441 return ret;
444 /* Module notifier call back, checking event on the module */
445 static int trace_probe_module_callback(struct notifier_block *nb,
446 unsigned long val, void *data)
448 struct module *mod = data;
449 struct trace_probe *tp;
450 int ret;
452 if (val != MODULE_STATE_COMING)
453 return NOTIFY_DONE;
455 /* Update probes on coming module */
456 mutex_lock(&probe_lock);
457 list_for_each_entry(tp, &probe_list, list) {
458 if (trace_probe_within_module(tp, mod)) {
459 /* Don't need to check busy - this should have gone. */
460 __unregister_trace_probe(tp);
461 ret = __register_trace_probe(tp);
462 if (ret)
463 pr_warning("Failed to re-register probe %s on"
464 "%s: %d\n",
465 tp->call.name, mod->name, ret);
468 mutex_unlock(&probe_lock);
470 return NOTIFY_DONE;
473 static struct notifier_block trace_probe_module_nb = {
474 .notifier_call = trace_probe_module_callback,
475 .priority = 1 /* Invoked after kprobe module callback */
478 static int create_trace_probe(int argc, char **argv)
481 * Argument syntax:
482 * - Add kprobe: p[:[GRP/]EVENT] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
483 * - Add kretprobe: r[:[GRP/]EVENT] [MOD:]KSYM[+0] [FETCHARGS]
484 * Fetch args:
485 * $retval : fetch return value
486 * $stack : fetch stack address
487 * $stackN : fetch Nth of stack (N:0-)
488 * @ADDR : fetch memory at ADDR (ADDR should be in kernel)
489 * @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
490 * %REG : fetch register REG
491 * Dereferencing memory fetch:
492 * +|-offs(ARG) : fetch memory at ARG +|- offs address.
493 * Alias name of args:
494 * NAME=FETCHARG : set NAME as alias of FETCHARG.
495 * Type of args:
496 * FETCHARG:TYPE : use TYPE instead of unsigned long.
498 struct trace_probe *tp;
499 int i, ret = 0;
500 bool is_return = false, is_delete = false;
501 char *symbol = NULL, *event = NULL, *group = NULL;
502 char *arg;
503 unsigned long offset = 0;
504 void *addr = NULL;
505 char buf[MAX_EVENT_NAME_LEN];
507 /* argc must be >= 1 */
508 if (argv[0][0] == 'p')
509 is_return = false;
510 else if (argv[0][0] == 'r')
511 is_return = true;
512 else if (argv[0][0] == '-')
513 is_delete = true;
514 else {
515 pr_info("Probe definition must be started with 'p', 'r' or"
516 " '-'.\n");
517 return -EINVAL;
520 if (argv[0][1] == ':') {
521 event = &argv[0][2];
522 if (strchr(event, '/')) {
523 group = event;
524 event = strchr(group, '/') + 1;
525 event[-1] = '\0';
526 if (strlen(group) == 0) {
527 pr_info("Group name is not specified\n");
528 return -EINVAL;
531 if (strlen(event) == 0) {
532 pr_info("Event name is not specified\n");
533 return -EINVAL;
536 if (!group)
537 group = KPROBE_EVENT_SYSTEM;
539 if (is_delete) {
540 if (!event) {
541 pr_info("Delete command needs an event name.\n");
542 return -EINVAL;
544 mutex_lock(&probe_lock);
545 tp = find_trace_probe(event, group);
546 if (!tp) {
547 mutex_unlock(&probe_lock);
548 pr_info("Event %s/%s doesn't exist.\n", group, event);
549 return -ENOENT;
551 /* delete an event */
552 ret = unregister_trace_probe(tp);
553 if (ret == 0)
554 free_trace_probe(tp);
555 mutex_unlock(&probe_lock);
556 return ret;
559 if (argc < 2) {
560 pr_info("Probe point is not specified.\n");
561 return -EINVAL;
563 if (isdigit(argv[1][0])) {
564 if (is_return) {
565 pr_info("Return probe point must be a symbol.\n");
566 return -EINVAL;
568 /* an address specified */
569 ret = kstrtoul(&argv[1][0], 0, (unsigned long *)&addr);
570 if (ret) {
571 pr_info("Failed to parse address.\n");
572 return ret;
574 } else {
575 /* a symbol specified */
576 symbol = argv[1];
577 /* TODO: support .init module functions */
578 ret = traceprobe_split_symbol_offset(symbol, &offset);
579 if (ret) {
580 pr_info("Failed to parse symbol.\n");
581 return ret;
583 if (offset && is_return) {
584 pr_info("Return probe must be used without offset.\n");
585 return -EINVAL;
588 argc -= 2; argv += 2;
590 /* setup a probe */
591 if (!event) {
592 /* Make a new event name */
593 if (symbol)
594 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
595 is_return ? 'r' : 'p', symbol, offset);
596 else
597 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
598 is_return ? 'r' : 'p', addr);
599 event = buf;
601 tp = alloc_trace_probe(group, event, addr, symbol, offset, argc,
602 is_return);
603 if (IS_ERR(tp)) {
604 pr_info("Failed to allocate trace_probe.(%d)\n",
605 (int)PTR_ERR(tp));
606 return PTR_ERR(tp);
609 /* parse arguments */
610 ret = 0;
611 for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
612 /* Increment count for freeing args in error case */
613 tp->nr_args++;
615 /* Parse argument name */
616 arg = strchr(argv[i], '=');
617 if (arg) {
618 *arg++ = '\0';
619 tp->args[i].name = kstrdup(argv[i], GFP_KERNEL);
620 } else {
621 arg = argv[i];
622 /* If argument name is omitted, set "argN" */
623 snprintf(buf, MAX_EVENT_NAME_LEN, "arg%d", i + 1);
624 tp->args[i].name = kstrdup(buf, GFP_KERNEL);
627 if (!tp->args[i].name) {
628 pr_info("Failed to allocate argument[%d] name.\n", i);
629 ret = -ENOMEM;
630 goto error;
633 if (!is_good_name(tp->args[i].name)) {
634 pr_info("Invalid argument[%d] name: %s\n",
635 i, tp->args[i].name);
636 ret = -EINVAL;
637 goto error;
640 if (traceprobe_conflict_field_name(tp->args[i].name,
641 tp->args, i)) {
642 pr_info("Argument[%d] name '%s' conflicts with "
643 "another field.\n", i, argv[i]);
644 ret = -EINVAL;
645 goto error;
648 /* Parse fetch argument */
649 ret = traceprobe_parse_probe_arg(arg, &tp->size, &tp->args[i],
650 is_return, true);
651 if (ret) {
652 pr_info("Parse error at argument[%d]. (%d)\n", i, ret);
653 goto error;
657 ret = register_trace_probe(tp);
658 if (ret)
659 goto error;
660 return 0;
662 error:
663 free_trace_probe(tp);
664 return ret;
667 static int release_all_trace_probes(void)
669 struct trace_probe *tp;
670 int ret = 0;
672 mutex_lock(&probe_lock);
673 /* Ensure no probe is in use. */
674 list_for_each_entry(tp, &probe_list, list)
675 if (trace_probe_is_enabled(tp)) {
676 ret = -EBUSY;
677 goto end;
679 /* TODO: Use batch unregistration */
680 while (!list_empty(&probe_list)) {
681 tp = list_entry(probe_list.next, struct trace_probe, list);
682 unregister_trace_probe(tp);
683 free_trace_probe(tp);
686 end:
687 mutex_unlock(&probe_lock);
689 return ret;
692 /* Probes listing interfaces */
693 static void *probes_seq_start(struct seq_file *m, loff_t *pos)
695 mutex_lock(&probe_lock);
696 return seq_list_start(&probe_list, *pos);
699 static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
701 return seq_list_next(v, &probe_list, pos);
704 static void probes_seq_stop(struct seq_file *m, void *v)
706 mutex_unlock(&probe_lock);
709 static int probes_seq_show(struct seq_file *m, void *v)
711 struct trace_probe *tp = v;
712 int i;
714 seq_printf(m, "%c", trace_probe_is_return(tp) ? 'r' : 'p');
715 seq_printf(m, ":%s/%s", tp->call.class->system, tp->call.name);
717 if (!tp->symbol)
718 seq_printf(m, " 0x%p", tp->rp.kp.addr);
719 else if (tp->rp.kp.offset)
720 seq_printf(m, " %s+%u", trace_probe_symbol(tp),
721 tp->rp.kp.offset);
722 else
723 seq_printf(m, " %s", trace_probe_symbol(tp));
725 for (i = 0; i < tp->nr_args; i++)
726 seq_printf(m, " %s=%s", tp->args[i].name, tp->args[i].comm);
727 seq_printf(m, "\n");
729 return 0;
732 static const struct seq_operations probes_seq_op = {
733 .start = probes_seq_start,
734 .next = probes_seq_next,
735 .stop = probes_seq_stop,
736 .show = probes_seq_show
739 static int probes_open(struct inode *inode, struct file *file)
741 int ret;
743 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
744 ret = release_all_trace_probes();
745 if (ret < 0)
746 return ret;
749 return seq_open(file, &probes_seq_op);
752 static ssize_t probes_write(struct file *file, const char __user *buffer,
753 size_t count, loff_t *ppos)
755 return traceprobe_probes_write(file, buffer, count, ppos,
756 create_trace_probe);
759 static const struct file_operations kprobe_events_ops = {
760 .owner = THIS_MODULE,
761 .open = probes_open,
762 .read = seq_read,
763 .llseek = seq_lseek,
764 .release = seq_release,
765 .write = probes_write,
768 /* Probes profiling interfaces */
769 static int probes_profile_seq_show(struct seq_file *m, void *v)
771 struct trace_probe *tp = v;
773 seq_printf(m, " %-44s %15lu %15lu\n", tp->call.name, tp->nhit,
774 tp->rp.kp.nmissed);
776 return 0;
779 static const struct seq_operations profile_seq_op = {
780 .start = probes_seq_start,
781 .next = probes_seq_next,
782 .stop = probes_seq_stop,
783 .show = probes_profile_seq_show
786 static int profile_open(struct inode *inode, struct file *file)
788 return seq_open(file, &profile_seq_op);
791 static const struct file_operations kprobe_profile_ops = {
792 .owner = THIS_MODULE,
793 .open = profile_open,
794 .read = seq_read,
795 .llseek = seq_lseek,
796 .release = seq_release,
799 /* Sum up total data length for dynamic arraies (strings) */
800 static __kprobes int __get_data_size(struct trace_probe *tp,
801 struct pt_regs *regs)
803 int i, ret = 0;
804 u32 len;
806 for (i = 0; i < tp->nr_args; i++)
807 if (unlikely(tp->args[i].fetch_size.fn)) {
808 call_fetch(&tp->args[i].fetch_size, regs, &len);
809 ret += len;
812 return ret;
815 /* Store the value of each argument */
816 static __kprobes void store_trace_args(int ent_size, struct trace_probe *tp,
817 struct pt_regs *regs,
818 u8 *data, int maxlen)
820 int i;
821 u32 end = tp->size;
822 u32 *dl; /* Data (relative) location */
824 for (i = 0; i < tp->nr_args; i++) {
825 if (unlikely(tp->args[i].fetch_size.fn)) {
827 * First, we set the relative location and
828 * maximum data length to *dl
830 dl = (u32 *)(data + tp->args[i].offset);
831 *dl = make_data_rloc(maxlen, end - tp->args[i].offset);
832 /* Then try to fetch string or dynamic array data */
833 call_fetch(&tp->args[i].fetch, regs, dl);
834 /* Reduce maximum length */
835 end += get_rloc_len(*dl);
836 maxlen -= get_rloc_len(*dl);
837 /* Trick here, convert data_rloc to data_loc */
838 *dl = convert_rloc_to_loc(*dl,
839 ent_size + tp->args[i].offset);
840 } else
841 /* Just fetching data normally */
842 call_fetch(&tp->args[i].fetch, regs,
843 data + tp->args[i].offset);
847 /* Kprobe handler */
848 static __kprobes void
849 __kprobe_trace_func(struct trace_probe *tp, struct pt_regs *regs,
850 struct ftrace_event_file *ftrace_file)
852 struct kprobe_trace_entry_head *entry;
853 struct ring_buffer_event *event;
854 struct ring_buffer *buffer;
855 int size, dsize, pc;
856 unsigned long irq_flags;
857 struct ftrace_event_call *call = &tp->call;
859 WARN_ON(call != ftrace_file->event_call);
861 if (test_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT, &ftrace_file->flags))
862 return;
864 local_save_flags(irq_flags);
865 pc = preempt_count();
867 dsize = __get_data_size(tp, regs);
868 size = sizeof(*entry) + tp->size + dsize;
870 event = trace_event_buffer_lock_reserve(&buffer, ftrace_file,
871 call->event.type,
872 size, irq_flags, pc);
873 if (!event)
874 return;
876 entry = ring_buffer_event_data(event);
877 entry->ip = (unsigned long)tp->rp.kp.addr;
878 store_trace_args(sizeof(*entry), tp, regs, (u8 *)&entry[1], dsize);
880 if (!filter_current_check_discard(buffer, call, entry, event))
881 trace_buffer_unlock_commit_regs(buffer, event,
882 irq_flags, pc, regs);
885 static __kprobes void
886 kprobe_trace_func(struct trace_probe *tp, struct pt_regs *regs)
889 * Note: preempt is already disabled around the kprobe handler.
890 * However, we still need an smp_read_barrier_depends() corresponding
891 * to smp_wmb() in rcu_assign_pointer() to access the pointer.
893 struct ftrace_event_file **file = rcu_dereference_raw(tp->files);
895 if (unlikely(!file))
896 return;
898 while (*file) {
899 __kprobe_trace_func(tp, regs, *file);
900 file++;
904 /* Kretprobe handler */
905 static __kprobes void
906 __kretprobe_trace_func(struct trace_probe *tp, struct kretprobe_instance *ri,
907 struct pt_regs *regs,
908 struct ftrace_event_file *ftrace_file)
910 struct kretprobe_trace_entry_head *entry;
911 struct ring_buffer_event *event;
912 struct ring_buffer *buffer;
913 int size, pc, dsize;
914 unsigned long irq_flags;
915 struct ftrace_event_call *call = &tp->call;
917 WARN_ON(call != ftrace_file->event_call);
919 if (test_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT, &ftrace_file->flags))
920 return;
922 local_save_flags(irq_flags);
923 pc = preempt_count();
925 dsize = __get_data_size(tp, regs);
926 size = sizeof(*entry) + tp->size + dsize;
928 event = trace_event_buffer_lock_reserve(&buffer, ftrace_file,
929 call->event.type,
930 size, irq_flags, pc);
931 if (!event)
932 return;
934 entry = ring_buffer_event_data(event);
935 entry->func = (unsigned long)tp->rp.kp.addr;
936 entry->ret_ip = (unsigned long)ri->ret_addr;
937 store_trace_args(sizeof(*entry), tp, regs, (u8 *)&entry[1], dsize);
939 if (!filter_current_check_discard(buffer, call, entry, event))
940 trace_buffer_unlock_commit_regs(buffer, event,
941 irq_flags, pc, regs);
944 static __kprobes void
945 kretprobe_trace_func(struct trace_probe *tp, struct kretprobe_instance *ri,
946 struct pt_regs *regs)
949 * Note: preempt is already disabled around the kprobe handler.
950 * However, we still need an smp_read_barrier_depends() corresponding
951 * to smp_wmb() in rcu_assign_pointer() to access the pointer.
953 struct ftrace_event_file **file = rcu_dereference_raw(tp->files);
955 if (unlikely(!file))
956 return;
958 while (*file) {
959 __kretprobe_trace_func(tp, ri, regs, *file);
960 file++;
964 /* Event entry printers */
965 static enum print_line_t
966 print_kprobe_event(struct trace_iterator *iter, int flags,
967 struct trace_event *event)
969 struct kprobe_trace_entry_head *field;
970 struct trace_seq *s = &iter->seq;
971 struct trace_probe *tp;
972 u8 *data;
973 int i;
975 field = (struct kprobe_trace_entry_head *)iter->ent;
976 tp = container_of(event, struct trace_probe, call.event);
978 if (!trace_seq_printf(s, "%s: (", tp->call.name))
979 goto partial;
981 if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
982 goto partial;
984 if (!trace_seq_puts(s, ")"))
985 goto partial;
987 data = (u8 *)&field[1];
988 for (i = 0; i < tp->nr_args; i++)
989 if (!tp->args[i].type->print(s, tp->args[i].name,
990 data + tp->args[i].offset, field))
991 goto partial;
993 if (!trace_seq_puts(s, "\n"))
994 goto partial;
996 return TRACE_TYPE_HANDLED;
997 partial:
998 return TRACE_TYPE_PARTIAL_LINE;
1001 static enum print_line_t
1002 print_kretprobe_event(struct trace_iterator *iter, int flags,
1003 struct trace_event *event)
1005 struct kretprobe_trace_entry_head *field;
1006 struct trace_seq *s = &iter->seq;
1007 struct trace_probe *tp;
1008 u8 *data;
1009 int i;
1011 field = (struct kretprobe_trace_entry_head *)iter->ent;
1012 tp = container_of(event, struct trace_probe, call.event);
1014 if (!trace_seq_printf(s, "%s: (", tp->call.name))
1015 goto partial;
1017 if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1018 goto partial;
1020 if (!trace_seq_puts(s, " <- "))
1021 goto partial;
1023 if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1024 goto partial;
1026 if (!trace_seq_puts(s, ")"))
1027 goto partial;
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 partial;
1035 if (!trace_seq_puts(s, "\n"))
1036 goto partial;
1038 return TRACE_TYPE_HANDLED;
1039 partial:
1040 return TRACE_TYPE_PARTIAL_LINE;
1044 static int kprobe_event_define_fields(struct ftrace_event_call *event_call)
1046 int ret, i;
1047 struct kprobe_trace_entry_head field;
1048 struct trace_probe *tp = (struct trace_probe *)event_call->data;
1050 DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1051 /* Set argument names as fields */
1052 for (i = 0; i < tp->nr_args; i++) {
1053 ret = trace_define_field(event_call, tp->args[i].type->fmttype,
1054 tp->args[i].name,
1055 sizeof(field) + tp->args[i].offset,
1056 tp->args[i].type->size,
1057 tp->args[i].type->is_signed,
1058 FILTER_OTHER);
1059 if (ret)
1060 return ret;
1062 return 0;
1065 static int kretprobe_event_define_fields(struct ftrace_event_call *event_call)
1067 int ret, i;
1068 struct kretprobe_trace_entry_head field;
1069 struct trace_probe *tp = (struct trace_probe *)event_call->data;
1071 DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1072 DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1073 /* Set argument names as fields */
1074 for (i = 0; i < tp->nr_args; i++) {
1075 ret = trace_define_field(event_call, tp->args[i].type->fmttype,
1076 tp->args[i].name,
1077 sizeof(field) + tp->args[i].offset,
1078 tp->args[i].type->size,
1079 tp->args[i].type->is_signed,
1080 FILTER_OTHER);
1081 if (ret)
1082 return ret;
1084 return 0;
1087 static int __set_print_fmt(struct trace_probe *tp, char *buf, int len)
1089 int i;
1090 int pos = 0;
1092 const char *fmt, *arg;
1094 if (!trace_probe_is_return(tp)) {
1095 fmt = "(%lx)";
1096 arg = "REC->" FIELD_STRING_IP;
1097 } else {
1098 fmt = "(%lx <- %lx)";
1099 arg = "REC->" FIELD_STRING_FUNC ", REC->" FIELD_STRING_RETIP;
1102 /* When len=0, we just calculate the needed length */
1103 #define LEN_OR_ZERO (len ? len - pos : 0)
1105 pos += snprintf(buf + pos, LEN_OR_ZERO, "\"%s", fmt);
1107 for (i = 0; i < tp->nr_args; i++) {
1108 pos += snprintf(buf + pos, LEN_OR_ZERO, " %s=%s",
1109 tp->args[i].name, tp->args[i].type->fmt);
1112 pos += snprintf(buf + pos, LEN_OR_ZERO, "\", %s", arg);
1114 for (i = 0; i < tp->nr_args; i++) {
1115 if (strcmp(tp->args[i].type->name, "string") == 0)
1116 pos += snprintf(buf + pos, LEN_OR_ZERO,
1117 ", __get_str(%s)",
1118 tp->args[i].name);
1119 else
1120 pos += snprintf(buf + pos, LEN_OR_ZERO, ", REC->%s",
1121 tp->args[i].name);
1124 #undef LEN_OR_ZERO
1126 /* return the length of print_fmt */
1127 return pos;
1130 static int set_print_fmt(struct trace_probe *tp)
1132 int len;
1133 char *print_fmt;
1135 /* First: called with 0 length to calculate the needed length */
1136 len = __set_print_fmt(tp, NULL, 0);
1137 print_fmt = kmalloc(len + 1, GFP_KERNEL);
1138 if (!print_fmt)
1139 return -ENOMEM;
1141 /* Second: actually write the @print_fmt */
1142 __set_print_fmt(tp, print_fmt, len + 1);
1143 tp->call.print_fmt = print_fmt;
1145 return 0;
1148 #ifdef CONFIG_PERF_EVENTS
1150 /* Kprobe profile handler */
1151 static __kprobes void
1152 kprobe_perf_func(struct trace_probe *tp, struct pt_regs *regs)
1154 struct ftrace_event_call *call = &tp->call;
1155 struct kprobe_trace_entry_head *entry;
1156 struct hlist_head *head;
1157 int size, __size, dsize;
1158 int rctx;
1160 dsize = __get_data_size(tp, regs);
1161 __size = sizeof(*entry) + tp->size + dsize;
1162 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1163 size -= sizeof(u32);
1164 if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE,
1165 "profile buffer not large enough"))
1166 return;
1168 entry = perf_trace_buf_prepare(size, call->event.type, regs, &rctx);
1169 if (!entry)
1170 return;
1172 entry->ip = (unsigned long)tp->rp.kp.addr;
1173 memset(&entry[1], 0, dsize);
1174 store_trace_args(sizeof(*entry), tp, regs, (u8 *)&entry[1], dsize);
1176 head = this_cpu_ptr(call->perf_events);
1177 perf_trace_buf_submit(entry, size, rctx,
1178 entry->ip, 1, regs, head, NULL);
1181 /* Kretprobe profile handler */
1182 static __kprobes void
1183 kretprobe_perf_func(struct trace_probe *tp, struct kretprobe_instance *ri,
1184 struct pt_regs *regs)
1186 struct ftrace_event_call *call = &tp->call;
1187 struct kretprobe_trace_entry_head *entry;
1188 struct hlist_head *head;
1189 int size, __size, dsize;
1190 int rctx;
1192 dsize = __get_data_size(tp, regs);
1193 __size = sizeof(*entry) + tp->size + dsize;
1194 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1195 size -= sizeof(u32);
1196 if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE,
1197 "profile buffer not large enough"))
1198 return;
1200 entry = perf_trace_buf_prepare(size, call->event.type, regs, &rctx);
1201 if (!entry)
1202 return;
1204 entry->func = (unsigned long)tp->rp.kp.addr;
1205 entry->ret_ip = (unsigned long)ri->ret_addr;
1206 store_trace_args(sizeof(*entry), tp, regs, (u8 *)&entry[1], dsize);
1208 head = this_cpu_ptr(call->perf_events);
1209 perf_trace_buf_submit(entry, size, rctx,
1210 entry->ret_ip, 1, regs, head, NULL);
1212 #endif /* CONFIG_PERF_EVENTS */
1214 static __kprobes
1215 int kprobe_register(struct ftrace_event_call *event,
1216 enum trace_reg type, void *data)
1218 struct trace_probe *tp = (struct trace_probe *)event->data;
1219 struct ftrace_event_file *file = data;
1221 switch (type) {
1222 case TRACE_REG_REGISTER:
1223 return enable_trace_probe(tp, file);
1224 case TRACE_REG_UNREGISTER:
1225 return disable_trace_probe(tp, file);
1227 #ifdef CONFIG_PERF_EVENTS
1228 case TRACE_REG_PERF_REGISTER:
1229 return enable_trace_probe(tp, NULL);
1230 case TRACE_REG_PERF_UNREGISTER:
1231 return disable_trace_probe(tp, NULL);
1232 case TRACE_REG_PERF_OPEN:
1233 case TRACE_REG_PERF_CLOSE:
1234 case TRACE_REG_PERF_ADD:
1235 case TRACE_REG_PERF_DEL:
1236 return 0;
1237 #endif
1239 return 0;
1242 static __kprobes
1243 int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1245 struct trace_probe *tp = container_of(kp, struct trace_probe, rp.kp);
1247 tp->nhit++;
1249 if (tp->flags & TP_FLAG_TRACE)
1250 kprobe_trace_func(tp, regs);
1251 #ifdef CONFIG_PERF_EVENTS
1252 if (tp->flags & TP_FLAG_PROFILE)
1253 kprobe_perf_func(tp, regs);
1254 #endif
1255 return 0; /* We don't tweek kernel, so just return 0 */
1258 static __kprobes
1259 int kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1261 struct trace_probe *tp = container_of(ri->rp, struct trace_probe, rp);
1263 tp->nhit++;
1265 if (tp->flags & TP_FLAG_TRACE)
1266 kretprobe_trace_func(tp, ri, regs);
1267 #ifdef CONFIG_PERF_EVENTS
1268 if (tp->flags & TP_FLAG_PROFILE)
1269 kretprobe_perf_func(tp, ri, regs);
1270 #endif
1271 return 0; /* We don't tweek kernel, so just return 0 */
1274 static struct trace_event_functions kretprobe_funcs = {
1275 .trace = print_kretprobe_event
1278 static struct trace_event_functions kprobe_funcs = {
1279 .trace = print_kprobe_event
1282 static int register_probe_event(struct trace_probe *tp)
1284 struct ftrace_event_call *call = &tp->call;
1285 int ret;
1287 /* Initialize ftrace_event_call */
1288 INIT_LIST_HEAD(&call->class->fields);
1289 if (trace_probe_is_return(tp)) {
1290 call->event.funcs = &kretprobe_funcs;
1291 call->class->define_fields = kretprobe_event_define_fields;
1292 } else {
1293 call->event.funcs = &kprobe_funcs;
1294 call->class->define_fields = kprobe_event_define_fields;
1296 if (set_print_fmt(tp) < 0)
1297 return -ENOMEM;
1298 ret = register_ftrace_event(&call->event);
1299 if (!ret) {
1300 kfree(call->print_fmt);
1301 return -ENODEV;
1303 call->flags = 0;
1304 call->class->reg = kprobe_register;
1305 call->data = tp;
1306 ret = trace_add_event_call(call);
1307 if (ret) {
1308 pr_info("Failed to register kprobe event: %s\n", call->name);
1309 kfree(call->print_fmt);
1310 unregister_ftrace_event(&call->event);
1312 return ret;
1315 static void unregister_probe_event(struct trace_probe *tp)
1317 /* tp->event is unregistered in trace_remove_event_call() */
1318 trace_remove_event_call(&tp->call);
1319 kfree(tp->call.print_fmt);
1322 /* Make a debugfs interface for controlling probe points */
1323 static __init int init_kprobe_trace(void)
1325 struct dentry *d_tracer;
1326 struct dentry *entry;
1328 if (register_module_notifier(&trace_probe_module_nb))
1329 return -EINVAL;
1331 d_tracer = tracing_init_dentry();
1332 if (!d_tracer)
1333 return 0;
1335 entry = debugfs_create_file("kprobe_events", 0644, d_tracer,
1336 NULL, &kprobe_events_ops);
1338 /* Event list interface */
1339 if (!entry)
1340 pr_warning("Could not create debugfs "
1341 "'kprobe_events' entry\n");
1343 /* Profile interface */
1344 entry = debugfs_create_file("kprobe_profile", 0444, d_tracer,
1345 NULL, &kprobe_profile_ops);
1347 if (!entry)
1348 pr_warning("Could not create debugfs "
1349 "'kprobe_profile' entry\n");
1350 return 0;
1352 fs_initcall(init_kprobe_trace);
1355 #ifdef CONFIG_FTRACE_STARTUP_TEST
1358 * The "__used" keeps gcc from removing the function symbol
1359 * from the kallsyms table.
1361 static __used int kprobe_trace_selftest_target(int a1, int a2, int a3,
1362 int a4, int a5, int a6)
1364 return a1 + a2 + a3 + a4 + a5 + a6;
1367 static struct ftrace_event_file *
1368 find_trace_probe_file(struct trace_probe *tp, struct trace_array *tr)
1370 struct ftrace_event_file *file;
1372 list_for_each_entry(file, &tr->events, list)
1373 if (file->event_call == &tp->call)
1374 return file;
1376 return NULL;
1379 static __init int kprobe_trace_self_tests_init(void)
1381 int ret, warn = 0;
1382 int (*target)(int, int, int, int, int, int);
1383 struct trace_probe *tp;
1384 struct ftrace_event_file *file;
1386 target = kprobe_trace_selftest_target;
1388 pr_info("Testing kprobe tracing: ");
1390 ret = traceprobe_command("p:testprobe kprobe_trace_selftest_target "
1391 "$stack $stack0 +0($stack)",
1392 create_trace_probe);
1393 if (WARN_ON_ONCE(ret)) {
1394 pr_warn("error on probing function entry.\n");
1395 warn++;
1396 } else {
1397 /* Enable trace point */
1398 tp = find_trace_probe("testprobe", KPROBE_EVENT_SYSTEM);
1399 if (WARN_ON_ONCE(tp == NULL)) {
1400 pr_warn("error on getting new probe.\n");
1401 warn++;
1402 } else {
1403 file = find_trace_probe_file(tp, top_trace_array());
1404 if (WARN_ON_ONCE(file == NULL)) {
1405 pr_warn("error on getting probe file.\n");
1406 warn++;
1407 } else
1408 enable_trace_probe(tp, file);
1412 ret = traceprobe_command("r:testprobe2 kprobe_trace_selftest_target "
1413 "$retval", create_trace_probe);
1414 if (WARN_ON_ONCE(ret)) {
1415 pr_warn("error on probing function return.\n");
1416 warn++;
1417 } else {
1418 /* Enable trace point */
1419 tp = find_trace_probe("testprobe2", KPROBE_EVENT_SYSTEM);
1420 if (WARN_ON_ONCE(tp == NULL)) {
1421 pr_warn("error on getting 2nd new probe.\n");
1422 warn++;
1423 } else {
1424 file = find_trace_probe_file(tp, top_trace_array());
1425 if (WARN_ON_ONCE(file == NULL)) {
1426 pr_warn("error on getting probe file.\n");
1427 warn++;
1428 } else
1429 enable_trace_probe(tp, file);
1433 if (warn)
1434 goto end;
1436 ret = target(1, 2, 3, 4, 5, 6);
1438 /* Disable trace points before removing it */
1439 tp = find_trace_probe("testprobe", KPROBE_EVENT_SYSTEM);
1440 if (WARN_ON_ONCE(tp == NULL)) {
1441 pr_warn("error on getting test probe.\n");
1442 warn++;
1443 } else {
1444 file = find_trace_probe_file(tp, top_trace_array());
1445 if (WARN_ON_ONCE(file == NULL)) {
1446 pr_warn("error on getting probe file.\n");
1447 warn++;
1448 } else
1449 disable_trace_probe(tp, file);
1452 tp = find_trace_probe("testprobe2", KPROBE_EVENT_SYSTEM);
1453 if (WARN_ON_ONCE(tp == NULL)) {
1454 pr_warn("error on getting 2nd test probe.\n");
1455 warn++;
1456 } else {
1457 file = find_trace_probe_file(tp, top_trace_array());
1458 if (WARN_ON_ONCE(file == NULL)) {
1459 pr_warn("error on getting probe file.\n");
1460 warn++;
1461 } else
1462 disable_trace_probe(tp, file);
1465 ret = traceprobe_command("-:testprobe", create_trace_probe);
1466 if (WARN_ON_ONCE(ret)) {
1467 pr_warn("error on deleting a probe.\n");
1468 warn++;
1471 ret = traceprobe_command("-:testprobe2", create_trace_probe);
1472 if (WARN_ON_ONCE(ret)) {
1473 pr_warn("error on deleting a probe.\n");
1474 warn++;
1477 end:
1478 release_all_trace_probes();
1479 if (warn)
1480 pr_cont("NG: Some tests are failed. Please check them.\n");
1481 else
1482 pr_cont("OK\n");
1483 return 0;
1486 late_initcall(kprobe_trace_self_tests_init);
1488 #endif