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"
28 * Kprobe event core functions
31 struct list_head list;
32 struct kretprobe rp; /* Use rp.kp for kprobe use */
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 list_head files;
39 ssize_t size; /* trace entry size */
41 struct probe_arg args[];
44 struct event_file_link {
45 struct ftrace_event_file *file;
46 struct list_head list;
49 #define SIZEOF_TRACE_PROBE(n) \
50 (offsetof(struct trace_probe, args) + \
51 (sizeof(struct probe_arg) * (n)))
54 static __kprobes bool trace_probe_is_return(struct trace_probe *tp)
56 return tp->rp.handler != NULL;
59 static __kprobes const char *trace_probe_symbol(struct trace_probe *tp)
61 return tp->symbol ? tp->symbol : "unknown";
64 static __kprobes unsigned long trace_probe_offset(struct trace_probe *tp)
66 return tp->rp.kp.offset;
69 static __kprobes bool trace_probe_is_enabled(struct trace_probe *tp)
71 return !!(tp->flags & (TP_FLAG_TRACE | TP_FLAG_PROFILE));
74 static __kprobes bool trace_probe_is_registered(struct trace_probe *tp)
76 return !!(tp->flags & TP_FLAG_REGISTERED);
79 static __kprobes bool trace_probe_has_gone(struct trace_probe *tp)
81 return !!(kprobe_gone(&tp->rp.kp));
84 static __kprobes bool trace_probe_within_module(struct trace_probe *tp,
87 int len = strlen(mod->name);
88 const char *name = trace_probe_symbol(tp);
89 return strncmp(mod->name, name, len) == 0 && name[len] == ':';
92 static __kprobes bool trace_probe_is_on_module(struct trace_probe *tp)
94 return !!strchr(trace_probe_symbol(tp), ':');
97 static int register_probe_event(struct trace_probe *tp);
98 static int unregister_probe_event(struct trace_probe *tp);
100 static DEFINE_MUTEX(probe_lock);
101 static LIST_HEAD(probe_list);
103 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
104 static int kretprobe_dispatcher(struct kretprobe_instance *ri,
105 struct pt_regs *regs);
108 * Allocate new trace_probe and initialize it (including kprobes).
110 static struct trace_probe *alloc_trace_probe(const char *group,
115 int nargs, bool is_return)
117 struct trace_probe *tp;
120 tp = kzalloc(SIZEOF_TRACE_PROBE(nargs), GFP_KERNEL);
125 tp->symbol = kstrdup(symbol, GFP_KERNEL);
128 tp->rp.kp.symbol_name = tp->symbol;
129 tp->rp.kp.offset = offs;
131 tp->rp.kp.addr = addr;
134 tp->rp.handler = kretprobe_dispatcher;
136 tp->rp.kp.pre_handler = kprobe_dispatcher;
138 if (!event || !is_good_name(event)) {
143 tp->call.class = &tp->class;
144 tp->call.name = kstrdup(event, GFP_KERNEL);
148 if (!group || !is_good_name(group)) {
153 tp->class.system = kstrdup(group, GFP_KERNEL);
154 if (!tp->class.system)
157 INIT_LIST_HEAD(&tp->list);
158 INIT_LIST_HEAD(&tp->files);
161 kfree(tp->call.name);
167 static void free_trace_probe(struct trace_probe *tp)
171 for (i = 0; i < tp->nr_args; i++)
172 traceprobe_free_probe_arg(&tp->args[i]);
174 kfree(tp->call.class->system);
175 kfree(tp->call.name);
180 static struct trace_probe *find_trace_probe(const char *event,
183 struct trace_probe *tp;
185 list_for_each_entry(tp, &probe_list, list)
186 if (strcmp(tp->call.name, event) == 0 &&
187 strcmp(tp->call.class->system, group) == 0)
194 * if the file is NULL, enable "perf" handler, or enable "trace" handler.
197 enable_trace_probe(struct trace_probe *tp, struct ftrace_event_file *file)
202 struct event_file_link *link;
204 link = kmalloc(sizeof(*link), GFP_KERNEL);
211 list_add_tail_rcu(&link->list, &tp->files);
213 tp->flags |= TP_FLAG_TRACE;
215 tp->flags |= TP_FLAG_PROFILE;
217 if (trace_probe_is_registered(tp) && !trace_probe_has_gone(tp)) {
218 if (trace_probe_is_return(tp))
219 ret = enable_kretprobe(&tp->rp);
221 ret = enable_kprobe(&tp->rp.kp);
227 static struct event_file_link *
228 find_event_file_link(struct trace_probe *tp, struct ftrace_event_file *file)
230 struct event_file_link *link;
232 list_for_each_entry(link, &tp->files, list)
233 if (link->file == file)
240 * Disable trace_probe
241 * if the file is NULL, disable "perf" handler, or disable "trace" handler.
244 disable_trace_probe(struct trace_probe *tp, struct ftrace_event_file *file)
246 struct event_file_link *link = NULL;
251 link = find_event_file_link(tp, file);
257 list_del_rcu(&link->list);
259 if (!list_empty(&tp->files))
262 tp->flags &= ~TP_FLAG_TRACE;
264 tp->flags &= ~TP_FLAG_PROFILE;
266 if (!trace_probe_is_enabled(tp) && trace_probe_is_registered(tp)) {
267 if (trace_probe_is_return(tp))
268 disable_kretprobe(&tp->rp);
270 disable_kprobe(&tp->rp.kp);
276 * Synchronize with kprobe_trace_func/kretprobe_trace_func
277 * to ensure disabled (all running handlers are finished).
278 * This is not only for kfree(), but also the caller,
279 * trace_remove_event_call() supposes it for releasing
280 * event_call related objects, which will be accessed in
281 * the kprobe_trace_func/kretprobe_trace_func.
284 kfree(link); /* Ignored if link == NULL */
290 /* Internal register function - just handle k*probes and flags */
291 static int __register_trace_probe(struct trace_probe *tp)
295 if (trace_probe_is_registered(tp))
298 for (i = 0; i < tp->nr_args; i++)
299 traceprobe_update_arg(&tp->args[i]);
301 /* Set/clear disabled flag according to tp->flag */
302 if (trace_probe_is_enabled(tp))
303 tp->rp.kp.flags &= ~KPROBE_FLAG_DISABLED;
305 tp->rp.kp.flags |= KPROBE_FLAG_DISABLED;
307 if (trace_probe_is_return(tp))
308 ret = register_kretprobe(&tp->rp);
310 ret = register_kprobe(&tp->rp.kp);
313 tp->flags |= TP_FLAG_REGISTERED;
315 pr_warning("Could not insert probe at %s+%lu: %d\n",
316 trace_probe_symbol(tp), trace_probe_offset(tp), ret);
317 if (ret == -ENOENT && trace_probe_is_on_module(tp)) {
318 pr_warning("This probe might be able to register after"
319 "target module is loaded. Continue.\n");
321 } else if (ret == -EILSEQ) {
322 pr_warning("Probing address(0x%p) is not an "
323 "instruction boundary.\n",
332 /* Internal unregister function - just handle k*probes and flags */
333 static void __unregister_trace_probe(struct trace_probe *tp)
335 if (trace_probe_is_registered(tp)) {
336 if (trace_probe_is_return(tp))
337 unregister_kretprobe(&tp->rp);
339 unregister_kprobe(&tp->rp.kp);
340 tp->flags &= ~TP_FLAG_REGISTERED;
341 /* Cleanup kprobe for reuse */
342 if (tp->rp.kp.symbol_name)
343 tp->rp.kp.addr = NULL;
347 /* Unregister a trace_probe and probe_event: call with locking probe_lock */
348 static int unregister_trace_probe(struct trace_probe *tp)
350 /* Enabled event can not be unregistered */
351 if (trace_probe_is_enabled(tp))
354 /* Will fail if probe is being used by ftrace or perf */
355 if (unregister_probe_event(tp))
358 __unregister_trace_probe(tp);
364 /* Register a trace_probe and probe_event */
365 static int register_trace_probe(struct trace_probe *tp)
367 struct trace_probe *old_tp;
370 mutex_lock(&probe_lock);
372 /* Delete old (same name) event if exist */
373 old_tp = find_trace_probe(tp->call.name, tp->call.class->system);
375 ret = unregister_trace_probe(old_tp);
378 free_trace_probe(old_tp);
381 /* Register new event */
382 ret = register_probe_event(tp);
384 pr_warning("Failed to register probe event(%d)\n", ret);
388 /* Register k*probe */
389 ret = __register_trace_probe(tp);
391 unregister_probe_event(tp);
393 list_add_tail(&tp->list, &probe_list);
396 mutex_unlock(&probe_lock);
400 /* Module notifier call back, checking event on the module */
401 static int trace_probe_module_callback(struct notifier_block *nb,
402 unsigned long val, void *data)
404 struct module *mod = data;
405 struct trace_probe *tp;
408 if (val != MODULE_STATE_COMING)
411 /* Update probes on coming module */
412 mutex_lock(&probe_lock);
413 list_for_each_entry(tp, &probe_list, list) {
414 if (trace_probe_within_module(tp, mod)) {
415 /* Don't need to check busy - this should have gone. */
416 __unregister_trace_probe(tp);
417 ret = __register_trace_probe(tp);
419 pr_warning("Failed to re-register probe %s on"
421 tp->call.name, mod->name, ret);
424 mutex_unlock(&probe_lock);
429 static struct notifier_block trace_probe_module_nb = {
430 .notifier_call = trace_probe_module_callback,
431 .priority = 1 /* Invoked after kprobe module callback */
434 static int create_trace_probe(int argc, char **argv)
438 * - Add kprobe: p[:[GRP/]EVENT] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
439 * - Add kretprobe: r[:[GRP/]EVENT] [MOD:]KSYM[+0] [FETCHARGS]
441 * $retval : fetch return value
442 * $stack : fetch stack address
443 * $stackN : fetch Nth of stack (N:0-)
444 * @ADDR : fetch memory at ADDR (ADDR should be in kernel)
445 * @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
446 * %REG : fetch register REG
447 * Dereferencing memory fetch:
448 * +|-offs(ARG) : fetch memory at ARG +|- offs address.
449 * Alias name of args:
450 * NAME=FETCHARG : set NAME as alias of FETCHARG.
452 * FETCHARG:TYPE : use TYPE instead of unsigned long.
454 struct trace_probe *tp;
456 bool is_return = false, is_delete = false;
457 char *symbol = NULL, *event = NULL, *group = NULL;
459 unsigned long offset = 0;
461 char buf[MAX_EVENT_NAME_LEN];
463 /* argc must be >= 1 */
464 if (argv[0][0] == 'p')
466 else if (argv[0][0] == 'r')
468 else if (argv[0][0] == '-')
471 pr_info("Probe definition must be started with 'p', 'r' or"
476 if (argv[0][1] == ':') {
478 if (strchr(event, '/')) {
480 event = strchr(group, '/') + 1;
482 if (strlen(group) == 0) {
483 pr_info("Group name is not specified\n");
487 if (strlen(event) == 0) {
488 pr_info("Event name is not specified\n");
493 group = KPROBE_EVENT_SYSTEM;
497 pr_info("Delete command needs an event name.\n");
500 mutex_lock(&probe_lock);
501 tp = find_trace_probe(event, group);
503 mutex_unlock(&probe_lock);
504 pr_info("Event %s/%s doesn't exist.\n", group, event);
507 /* delete an event */
508 ret = unregister_trace_probe(tp);
510 free_trace_probe(tp);
511 mutex_unlock(&probe_lock);
516 pr_info("Probe point is not specified.\n");
519 if (isdigit(argv[1][0])) {
521 pr_info("Return probe point must be a symbol.\n");
524 /* an address specified */
525 ret = kstrtoul(&argv[1][0], 0, (unsigned long *)&addr);
527 pr_info("Failed to parse address.\n");
531 /* a symbol specified */
533 /* TODO: support .init module functions */
534 ret = traceprobe_split_symbol_offset(symbol, &offset);
536 pr_info("Failed to parse symbol.\n");
539 if (offset && is_return) {
540 pr_info("Return probe must be used without offset.\n");
544 argc -= 2; argv += 2;
548 /* Make a new event name */
550 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
551 is_return ? 'r' : 'p', symbol, offset);
553 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
554 is_return ? 'r' : 'p', addr);
557 tp = alloc_trace_probe(group, event, addr, symbol, offset, argc,
560 pr_info("Failed to allocate trace_probe.(%d)\n",
565 /* parse arguments */
567 for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
568 /* Increment count for freeing args in error case */
571 /* Parse argument name */
572 arg = strchr(argv[i], '=');
575 tp->args[i].name = kstrdup(argv[i], GFP_KERNEL);
578 /* If argument name is omitted, set "argN" */
579 snprintf(buf, MAX_EVENT_NAME_LEN, "arg%d", i + 1);
580 tp->args[i].name = kstrdup(buf, GFP_KERNEL);
583 if (!tp->args[i].name) {
584 pr_info("Failed to allocate argument[%d] name.\n", i);
589 if (!is_good_name(tp->args[i].name)) {
590 pr_info("Invalid argument[%d] name: %s\n",
591 i, tp->args[i].name);
596 if (traceprobe_conflict_field_name(tp->args[i].name,
598 pr_info("Argument[%d] name '%s' conflicts with "
599 "another field.\n", i, argv[i]);
604 /* Parse fetch argument */
605 ret = traceprobe_parse_probe_arg(arg, &tp->size, &tp->args[i],
608 pr_info("Parse error at argument[%d]. (%d)\n", i, ret);
613 ret = register_trace_probe(tp);
619 free_trace_probe(tp);
623 static int release_all_trace_probes(void)
625 struct trace_probe *tp;
628 mutex_lock(&probe_lock);
629 /* Ensure no probe is in use. */
630 list_for_each_entry(tp, &probe_list, list)
631 if (trace_probe_is_enabled(tp)) {
635 /* TODO: Use batch unregistration */
636 while (!list_empty(&probe_list)) {
637 tp = list_entry(probe_list.next, struct trace_probe, list);
638 ret = unregister_trace_probe(tp);
641 free_trace_probe(tp);
645 mutex_unlock(&probe_lock);
650 /* Probes listing interfaces */
651 static void *probes_seq_start(struct seq_file *m, loff_t *pos)
653 mutex_lock(&probe_lock);
654 return seq_list_start(&probe_list, *pos);
657 static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
659 return seq_list_next(v, &probe_list, pos);
662 static void probes_seq_stop(struct seq_file *m, void *v)
664 mutex_unlock(&probe_lock);
667 static int probes_seq_show(struct seq_file *m, void *v)
669 struct trace_probe *tp = v;
672 seq_printf(m, "%c", trace_probe_is_return(tp) ? 'r' : 'p');
673 seq_printf(m, ":%s/%s", tp->call.class->system, tp->call.name);
676 seq_printf(m, " 0x%p", tp->rp.kp.addr);
677 else if (tp->rp.kp.offset)
678 seq_printf(m, " %s+%u", trace_probe_symbol(tp),
681 seq_printf(m, " %s", trace_probe_symbol(tp));
683 for (i = 0; i < tp->nr_args; i++)
684 seq_printf(m, " %s=%s", tp->args[i].name, tp->args[i].comm);
690 static const struct seq_operations probes_seq_op = {
691 .start = probes_seq_start,
692 .next = probes_seq_next,
693 .stop = probes_seq_stop,
694 .show = probes_seq_show
697 static int probes_open(struct inode *inode, struct file *file)
701 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
702 ret = release_all_trace_probes();
707 return seq_open(file, &probes_seq_op);
710 static ssize_t probes_write(struct file *file, const char __user *buffer,
711 size_t count, loff_t *ppos)
713 return traceprobe_probes_write(file, buffer, count, ppos,
717 static const struct file_operations kprobe_events_ops = {
718 .owner = THIS_MODULE,
722 .release = seq_release,
723 .write = probes_write,
726 /* Probes profiling interfaces */
727 static int probes_profile_seq_show(struct seq_file *m, void *v)
729 struct trace_probe *tp = v;
731 seq_printf(m, " %-44s %15lu %15lu\n", tp->call.name, tp->nhit,
737 static const struct seq_operations profile_seq_op = {
738 .start = probes_seq_start,
739 .next = probes_seq_next,
740 .stop = probes_seq_stop,
741 .show = probes_profile_seq_show
744 static int profile_open(struct inode *inode, struct file *file)
746 return seq_open(file, &profile_seq_op);
749 static const struct file_operations kprobe_profile_ops = {
750 .owner = THIS_MODULE,
751 .open = profile_open,
754 .release = seq_release,
757 /* Sum up total data length for dynamic arraies (strings) */
758 static __kprobes int __get_data_size(struct trace_probe *tp,
759 struct pt_regs *regs)
764 for (i = 0; i < tp->nr_args; i++)
765 if (unlikely(tp->args[i].fetch_size.fn)) {
766 call_fetch(&tp->args[i].fetch_size, regs, &len);
773 /* Store the value of each argument */
774 static __kprobes void store_trace_args(int ent_size, struct trace_probe *tp,
775 struct pt_regs *regs,
776 u8 *data, int maxlen)
780 u32 *dl; /* Data (relative) location */
782 for (i = 0; i < tp->nr_args; i++) {
783 if (unlikely(tp->args[i].fetch_size.fn)) {
785 * First, we set the relative location and
786 * maximum data length to *dl
788 dl = (u32 *)(data + tp->args[i].offset);
789 *dl = make_data_rloc(maxlen, end - tp->args[i].offset);
790 /* Then try to fetch string or dynamic array data */
791 call_fetch(&tp->args[i].fetch, regs, dl);
792 /* Reduce maximum length */
793 end += get_rloc_len(*dl);
794 maxlen -= get_rloc_len(*dl);
795 /* Trick here, convert data_rloc to data_loc */
796 *dl = convert_rloc_to_loc(*dl,
797 ent_size + tp->args[i].offset);
799 /* Just fetching data normally */
800 call_fetch(&tp->args[i].fetch, regs,
801 data + tp->args[i].offset);
806 static __kprobes void
807 __kprobe_trace_func(struct trace_probe *tp, struct pt_regs *regs,
808 struct ftrace_event_file *ftrace_file)
810 struct kprobe_trace_entry_head *entry;
811 struct ring_buffer_event *event;
812 struct ring_buffer *buffer;
814 unsigned long irq_flags;
815 struct ftrace_event_call *call = &tp->call;
817 WARN_ON(call != ftrace_file->event_call);
819 if (test_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT, &ftrace_file->flags))
822 local_save_flags(irq_flags);
823 pc = preempt_count();
825 dsize = __get_data_size(tp, regs);
826 size = sizeof(*entry) + tp->size + dsize;
828 event = trace_event_buffer_lock_reserve(&buffer, ftrace_file,
830 size, irq_flags, pc);
834 entry = ring_buffer_event_data(event);
835 entry->ip = (unsigned long)tp->rp.kp.addr;
836 store_trace_args(sizeof(*entry), tp, regs, (u8 *)&entry[1], dsize);
838 if (!filter_check_discard(ftrace_file, entry, buffer, event))
839 trace_buffer_unlock_commit_regs(buffer, event,
840 irq_flags, pc, regs);
843 static __kprobes void
844 kprobe_trace_func(struct trace_probe *tp, struct pt_regs *regs)
846 struct event_file_link *link;
848 list_for_each_entry_rcu(link, &tp->files, list)
849 __kprobe_trace_func(tp, regs, link->file);
852 /* Kretprobe handler */
853 static __kprobes void
854 __kretprobe_trace_func(struct trace_probe *tp, struct kretprobe_instance *ri,
855 struct pt_regs *regs,
856 struct ftrace_event_file *ftrace_file)
858 struct kretprobe_trace_entry_head *entry;
859 struct ring_buffer_event *event;
860 struct ring_buffer *buffer;
862 unsigned long irq_flags;
863 struct ftrace_event_call *call = &tp->call;
865 WARN_ON(call != ftrace_file->event_call);
867 if (test_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT, &ftrace_file->flags))
870 local_save_flags(irq_flags);
871 pc = preempt_count();
873 dsize = __get_data_size(tp, regs);
874 size = sizeof(*entry) + tp->size + dsize;
876 event = trace_event_buffer_lock_reserve(&buffer, ftrace_file,
878 size, irq_flags, pc);
882 entry = ring_buffer_event_data(event);
883 entry->func = (unsigned long)tp->rp.kp.addr;
884 entry->ret_ip = (unsigned long)ri->ret_addr;
885 store_trace_args(sizeof(*entry), tp, regs, (u8 *)&entry[1], dsize);
887 if (!filter_check_discard(ftrace_file, entry, buffer, event))
888 trace_buffer_unlock_commit_regs(buffer, event,
889 irq_flags, pc, regs);
892 static __kprobes void
893 kretprobe_trace_func(struct trace_probe *tp, struct kretprobe_instance *ri,
894 struct pt_regs *regs)
896 struct event_file_link *link;
898 list_for_each_entry_rcu(link, &tp->files, list)
899 __kretprobe_trace_func(tp, ri, regs, link->file);
902 /* Event entry printers */
903 static enum print_line_t
904 print_kprobe_event(struct trace_iterator *iter, int flags,
905 struct trace_event *event)
907 struct kprobe_trace_entry_head *field;
908 struct trace_seq *s = &iter->seq;
909 struct trace_probe *tp;
913 field = (struct kprobe_trace_entry_head *)iter->ent;
914 tp = container_of(event, struct trace_probe, call.event);
916 if (!trace_seq_printf(s, "%s: (", tp->call.name))
919 if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
922 if (!trace_seq_puts(s, ")"))
925 data = (u8 *)&field[1];
926 for (i = 0; i < tp->nr_args; i++)
927 if (!tp->args[i].type->print(s, tp->args[i].name,
928 data + tp->args[i].offset, field))
931 if (!trace_seq_puts(s, "\n"))
934 return TRACE_TYPE_HANDLED;
936 return TRACE_TYPE_PARTIAL_LINE;
939 static enum print_line_t
940 print_kretprobe_event(struct trace_iterator *iter, int flags,
941 struct trace_event *event)
943 struct kretprobe_trace_entry_head *field;
944 struct trace_seq *s = &iter->seq;
945 struct trace_probe *tp;
949 field = (struct kretprobe_trace_entry_head *)iter->ent;
950 tp = container_of(event, struct trace_probe, call.event);
952 if (!trace_seq_printf(s, "%s: (", tp->call.name))
955 if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
958 if (!trace_seq_puts(s, " <- "))
961 if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
964 if (!trace_seq_puts(s, ")"))
967 data = (u8 *)&field[1];
968 for (i = 0; i < tp->nr_args; i++)
969 if (!tp->args[i].type->print(s, tp->args[i].name,
970 data + tp->args[i].offset, field))
973 if (!trace_seq_puts(s, "\n"))
976 return TRACE_TYPE_HANDLED;
978 return TRACE_TYPE_PARTIAL_LINE;
982 static int kprobe_event_define_fields(struct ftrace_event_call *event_call)
985 struct kprobe_trace_entry_head field;
986 struct trace_probe *tp = (struct trace_probe *)event_call->data;
988 DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
989 /* Set argument names as fields */
990 for (i = 0; i < tp->nr_args; i++) {
991 ret = trace_define_field(event_call, tp->args[i].type->fmttype,
993 sizeof(field) + tp->args[i].offset,
994 tp->args[i].type->size,
995 tp->args[i].type->is_signed,
1003 static int kretprobe_event_define_fields(struct ftrace_event_call *event_call)
1006 struct kretprobe_trace_entry_head field;
1007 struct trace_probe *tp = (struct trace_probe *)event_call->data;
1009 DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1010 DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1011 /* Set argument names as fields */
1012 for (i = 0; i < tp->nr_args; i++) {
1013 ret = trace_define_field(event_call, tp->args[i].type->fmttype,
1015 sizeof(field) + tp->args[i].offset,
1016 tp->args[i].type->size,
1017 tp->args[i].type->is_signed,
1025 static int __set_print_fmt(struct trace_probe *tp, char *buf, int len)
1030 const char *fmt, *arg;
1032 if (!trace_probe_is_return(tp)) {
1034 arg = "REC->" FIELD_STRING_IP;
1036 fmt = "(%lx <- %lx)";
1037 arg = "REC->" FIELD_STRING_FUNC ", REC->" FIELD_STRING_RETIP;
1040 /* When len=0, we just calculate the needed length */
1041 #define LEN_OR_ZERO (len ? len - pos : 0)
1043 pos += snprintf(buf + pos, LEN_OR_ZERO, "\"%s", fmt);
1045 for (i = 0; i < tp->nr_args; i++) {
1046 pos += snprintf(buf + pos, LEN_OR_ZERO, " %s=%s",
1047 tp->args[i].name, tp->args[i].type->fmt);
1050 pos += snprintf(buf + pos, LEN_OR_ZERO, "\", %s", arg);
1052 for (i = 0; i < tp->nr_args; i++) {
1053 if (strcmp(tp->args[i].type->name, "string") == 0)
1054 pos += snprintf(buf + pos, LEN_OR_ZERO,
1058 pos += snprintf(buf + pos, LEN_OR_ZERO, ", REC->%s",
1064 /* return the length of print_fmt */
1068 static int set_print_fmt(struct trace_probe *tp)
1073 /* First: called with 0 length to calculate the needed length */
1074 len = __set_print_fmt(tp, NULL, 0);
1075 print_fmt = kmalloc(len + 1, GFP_KERNEL);
1079 /* Second: actually write the @print_fmt */
1080 __set_print_fmt(tp, print_fmt, len + 1);
1081 tp->call.print_fmt = print_fmt;
1086 #ifdef CONFIG_PERF_EVENTS
1088 /* Kprobe profile handler */
1089 static __kprobes void
1090 kprobe_perf_func(struct trace_probe *tp, struct pt_regs *regs)
1092 struct ftrace_event_call *call = &tp->call;
1093 struct kprobe_trace_entry_head *entry;
1094 struct hlist_head *head;
1095 int size, __size, dsize;
1098 head = this_cpu_ptr(call->perf_events);
1099 if (hlist_empty(head))
1102 dsize = __get_data_size(tp, regs);
1103 __size = sizeof(*entry) + tp->size + dsize;
1104 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1105 size -= sizeof(u32);
1107 entry = perf_trace_buf_prepare(size, call->event.type, regs, &rctx);
1111 entry->ip = (unsigned long)tp->rp.kp.addr;
1112 memset(&entry[1], 0, dsize);
1113 store_trace_args(sizeof(*entry), tp, regs, (u8 *)&entry[1], dsize);
1114 perf_trace_buf_submit(entry, size, rctx, 0, 1, regs, head, NULL);
1117 /* Kretprobe profile handler */
1118 static __kprobes void
1119 kretprobe_perf_func(struct trace_probe *tp, struct kretprobe_instance *ri,
1120 struct pt_regs *regs)
1122 struct ftrace_event_call *call = &tp->call;
1123 struct kretprobe_trace_entry_head *entry;
1124 struct hlist_head *head;
1125 int size, __size, dsize;
1128 head = this_cpu_ptr(call->perf_events);
1129 if (hlist_empty(head))
1132 dsize = __get_data_size(tp, regs);
1133 __size = sizeof(*entry) + tp->size + dsize;
1134 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1135 size -= sizeof(u32);
1137 entry = perf_trace_buf_prepare(size, call->event.type, regs, &rctx);
1141 entry->func = (unsigned long)tp->rp.kp.addr;
1142 entry->ret_ip = (unsigned long)ri->ret_addr;
1143 store_trace_args(sizeof(*entry), tp, regs, (u8 *)&entry[1], dsize);
1144 perf_trace_buf_submit(entry, size, rctx, 0, 1, regs, head, NULL);
1146 #endif /* CONFIG_PERF_EVENTS */
1149 * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex.
1151 * kprobe_trace_self_tests_init() does enable_trace_probe/disable_trace_probe
1152 * lockless, but we can't race with this __init function.
1155 int kprobe_register(struct ftrace_event_call *event,
1156 enum trace_reg type, void *data)
1158 struct trace_probe *tp = (struct trace_probe *)event->data;
1159 struct ftrace_event_file *file = data;
1162 case TRACE_REG_REGISTER:
1163 return enable_trace_probe(tp, file);
1164 case TRACE_REG_UNREGISTER:
1165 return disable_trace_probe(tp, file);
1167 #ifdef CONFIG_PERF_EVENTS
1168 case TRACE_REG_PERF_REGISTER:
1169 return enable_trace_probe(tp, NULL);
1170 case TRACE_REG_PERF_UNREGISTER:
1171 return disable_trace_probe(tp, NULL);
1172 case TRACE_REG_PERF_OPEN:
1173 case TRACE_REG_PERF_CLOSE:
1174 case TRACE_REG_PERF_ADD:
1175 case TRACE_REG_PERF_DEL:
1183 int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1185 struct trace_probe *tp = container_of(kp, struct trace_probe, rp.kp);
1189 if (tp->flags & TP_FLAG_TRACE)
1190 kprobe_trace_func(tp, regs);
1191 #ifdef CONFIG_PERF_EVENTS
1192 if (tp->flags & TP_FLAG_PROFILE)
1193 kprobe_perf_func(tp, regs);
1195 return 0; /* We don't tweek kernel, so just return 0 */
1199 int kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1201 struct trace_probe *tp = container_of(ri->rp, struct trace_probe, rp);
1205 if (tp->flags & TP_FLAG_TRACE)
1206 kretprobe_trace_func(tp, ri, regs);
1207 #ifdef CONFIG_PERF_EVENTS
1208 if (tp->flags & TP_FLAG_PROFILE)
1209 kretprobe_perf_func(tp, ri, regs);
1211 return 0; /* We don't tweek kernel, so just return 0 */
1214 static struct trace_event_functions kretprobe_funcs = {
1215 .trace = print_kretprobe_event
1218 static struct trace_event_functions kprobe_funcs = {
1219 .trace = print_kprobe_event
1222 static int register_probe_event(struct trace_probe *tp)
1224 struct ftrace_event_call *call = &tp->call;
1227 /* Initialize ftrace_event_call */
1228 INIT_LIST_HEAD(&call->class->fields);
1229 if (trace_probe_is_return(tp)) {
1230 call->event.funcs = &kretprobe_funcs;
1231 call->class->define_fields = kretprobe_event_define_fields;
1233 call->event.funcs = &kprobe_funcs;
1234 call->class->define_fields = kprobe_event_define_fields;
1236 if (set_print_fmt(tp) < 0)
1238 ret = register_ftrace_event(&call->event);
1240 kfree(call->print_fmt);
1244 call->class->reg = kprobe_register;
1246 ret = trace_add_event_call(call);
1248 pr_info("Failed to register kprobe event: %s\n", call->name);
1249 kfree(call->print_fmt);
1250 unregister_ftrace_event(&call->event);
1255 static int unregister_probe_event(struct trace_probe *tp)
1259 /* tp->event is unregistered in trace_remove_event_call() */
1260 ret = trace_remove_event_call(&tp->call);
1262 kfree(tp->call.print_fmt);
1266 /* Make a debugfs interface for controlling probe points */
1267 static __init int init_kprobe_trace(void)
1269 struct dentry *d_tracer;
1270 struct dentry *entry;
1272 if (register_module_notifier(&trace_probe_module_nb))
1275 d_tracer = tracing_init_dentry();
1279 entry = debugfs_create_file("kprobe_events", 0644, d_tracer,
1280 NULL, &kprobe_events_ops);
1282 /* Event list interface */
1284 pr_warning("Could not create debugfs "
1285 "'kprobe_events' entry\n");
1287 /* Profile interface */
1288 entry = debugfs_create_file("kprobe_profile", 0444, d_tracer,
1289 NULL, &kprobe_profile_ops);
1292 pr_warning("Could not create debugfs "
1293 "'kprobe_profile' entry\n");
1296 fs_initcall(init_kprobe_trace);
1299 #ifdef CONFIG_FTRACE_STARTUP_TEST
1302 * The "__used" keeps gcc from removing the function symbol
1303 * from the kallsyms table.
1305 static __used int kprobe_trace_selftest_target(int a1, int a2, int a3,
1306 int a4, int a5, int a6)
1308 return a1 + a2 + a3 + a4 + a5 + a6;
1311 static struct ftrace_event_file *
1312 find_trace_probe_file(struct trace_probe *tp, struct trace_array *tr)
1314 struct ftrace_event_file *file;
1316 list_for_each_entry(file, &tr->events, list)
1317 if (file->event_call == &tp->call)
1324 * Nobody but us can call enable_trace_probe/disable_trace_probe at this
1325 * stage, we can do this lockless.
1327 static __init int kprobe_trace_self_tests_init(void)
1330 int (*target)(int, int, int, int, int, int);
1331 struct trace_probe *tp;
1332 struct ftrace_event_file *file;
1334 target = kprobe_trace_selftest_target;
1336 pr_info("Testing kprobe tracing: ");
1338 ret = traceprobe_command("p:testprobe kprobe_trace_selftest_target "
1339 "$stack $stack0 +0($stack)",
1340 create_trace_probe);
1341 if (WARN_ON_ONCE(ret)) {
1342 pr_warn("error on probing function entry.\n");
1345 /* Enable trace point */
1346 tp = find_trace_probe("testprobe", KPROBE_EVENT_SYSTEM);
1347 if (WARN_ON_ONCE(tp == NULL)) {
1348 pr_warn("error on getting new probe.\n");
1351 file = find_trace_probe_file(tp, top_trace_array());
1352 if (WARN_ON_ONCE(file == NULL)) {
1353 pr_warn("error on getting probe file.\n");
1356 enable_trace_probe(tp, file);
1360 ret = traceprobe_command("r:testprobe2 kprobe_trace_selftest_target "
1361 "$retval", create_trace_probe);
1362 if (WARN_ON_ONCE(ret)) {
1363 pr_warn("error on probing function return.\n");
1366 /* Enable trace point */
1367 tp = find_trace_probe("testprobe2", KPROBE_EVENT_SYSTEM);
1368 if (WARN_ON_ONCE(tp == NULL)) {
1369 pr_warn("error on getting 2nd new probe.\n");
1372 file = find_trace_probe_file(tp, top_trace_array());
1373 if (WARN_ON_ONCE(file == NULL)) {
1374 pr_warn("error on getting probe file.\n");
1377 enable_trace_probe(tp, file);
1384 ret = target(1, 2, 3, 4, 5, 6);
1386 /* Disable trace points before removing it */
1387 tp = find_trace_probe("testprobe", KPROBE_EVENT_SYSTEM);
1388 if (WARN_ON_ONCE(tp == NULL)) {
1389 pr_warn("error on getting test probe.\n");
1392 file = find_trace_probe_file(tp, top_trace_array());
1393 if (WARN_ON_ONCE(file == NULL)) {
1394 pr_warn("error on getting probe file.\n");
1397 disable_trace_probe(tp, file);
1400 tp = find_trace_probe("testprobe2", KPROBE_EVENT_SYSTEM);
1401 if (WARN_ON_ONCE(tp == NULL)) {
1402 pr_warn("error on getting 2nd test probe.\n");
1405 file = find_trace_probe_file(tp, top_trace_array());
1406 if (WARN_ON_ONCE(file == NULL)) {
1407 pr_warn("error on getting probe file.\n");
1410 disable_trace_probe(tp, file);
1413 ret = traceprobe_command("-:testprobe", create_trace_probe);
1414 if (WARN_ON_ONCE(ret)) {
1415 pr_warn("error on deleting a probe.\n");
1419 ret = traceprobe_command("-:testprobe2", create_trace_probe);
1420 if (WARN_ON_ONCE(ret)) {
1421 pr_warn("error on deleting a probe.\n");
1426 release_all_trace_probes();
1428 pr_cont("NG: Some tests are failed. Please check them.\n");
1434 late_initcall(kprobe_trace_self_tests_init);