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 void 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 __unregister_trace_probe(tp);
356 unregister_probe_event(tp);
361 /* Register a trace_probe and probe_event */
362 static int register_trace_probe(struct trace_probe *tp)
364 struct trace_probe *old_tp;
367 mutex_lock(&probe_lock);
369 /* Delete old (same name) event if exist */
370 old_tp = find_trace_probe(tp->call.name, tp->call.class->system);
372 ret = unregister_trace_probe(old_tp);
375 free_trace_probe(old_tp);
378 /* Register new event */
379 ret = register_probe_event(tp);
381 pr_warning("Failed to register probe event(%d)\n", ret);
385 /* Register k*probe */
386 ret = __register_trace_probe(tp);
388 unregister_probe_event(tp);
390 list_add_tail(&tp->list, &probe_list);
393 mutex_unlock(&probe_lock);
397 /* Module notifier call back, checking event on the module */
398 static int trace_probe_module_callback(struct notifier_block *nb,
399 unsigned long val, void *data)
401 struct module *mod = data;
402 struct trace_probe *tp;
405 if (val != MODULE_STATE_COMING)
408 /* Update probes on coming module */
409 mutex_lock(&probe_lock);
410 list_for_each_entry(tp, &probe_list, list) {
411 if (trace_probe_within_module(tp, mod)) {
412 /* Don't need to check busy - this should have gone. */
413 __unregister_trace_probe(tp);
414 ret = __register_trace_probe(tp);
416 pr_warning("Failed to re-register probe %s on"
418 tp->call.name, mod->name, ret);
421 mutex_unlock(&probe_lock);
426 static struct notifier_block trace_probe_module_nb = {
427 .notifier_call = trace_probe_module_callback,
428 .priority = 1 /* Invoked after kprobe module callback */
431 static int create_trace_probe(int argc, char **argv)
435 * - Add kprobe: p[:[GRP/]EVENT] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
436 * - Add kretprobe: r[:[GRP/]EVENT] [MOD:]KSYM[+0] [FETCHARGS]
438 * $retval : fetch return value
439 * $stack : fetch stack address
440 * $stackN : fetch Nth of stack (N:0-)
441 * @ADDR : fetch memory at ADDR (ADDR should be in kernel)
442 * @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
443 * %REG : fetch register REG
444 * Dereferencing memory fetch:
445 * +|-offs(ARG) : fetch memory at ARG +|- offs address.
446 * Alias name of args:
447 * NAME=FETCHARG : set NAME as alias of FETCHARG.
449 * FETCHARG:TYPE : use TYPE instead of unsigned long.
451 struct trace_probe *tp;
453 bool is_return = false, is_delete = false;
454 char *symbol = NULL, *event = NULL, *group = NULL;
456 unsigned long offset = 0;
458 char buf[MAX_EVENT_NAME_LEN];
460 /* argc must be >= 1 */
461 if (argv[0][0] == 'p')
463 else if (argv[0][0] == 'r')
465 else if (argv[0][0] == '-')
468 pr_info("Probe definition must be started with 'p', 'r' or"
473 if (argv[0][1] == ':') {
475 if (strchr(event, '/')) {
477 event = strchr(group, '/') + 1;
479 if (strlen(group) == 0) {
480 pr_info("Group name is not specified\n");
484 if (strlen(event) == 0) {
485 pr_info("Event name is not specified\n");
490 group = KPROBE_EVENT_SYSTEM;
494 pr_info("Delete command needs an event name.\n");
497 mutex_lock(&probe_lock);
498 tp = find_trace_probe(event, group);
500 mutex_unlock(&probe_lock);
501 pr_info("Event %s/%s doesn't exist.\n", group, event);
504 /* delete an event */
505 ret = unregister_trace_probe(tp);
507 free_trace_probe(tp);
508 mutex_unlock(&probe_lock);
513 pr_info("Probe point is not specified.\n");
516 if (isdigit(argv[1][0])) {
518 pr_info("Return probe point must be a symbol.\n");
521 /* an address specified */
522 ret = kstrtoul(&argv[1][0], 0, (unsigned long *)&addr);
524 pr_info("Failed to parse address.\n");
528 /* a symbol specified */
530 /* TODO: support .init module functions */
531 ret = traceprobe_split_symbol_offset(symbol, &offset);
533 pr_info("Failed to parse symbol.\n");
536 if (offset && is_return) {
537 pr_info("Return probe must be used without offset.\n");
541 argc -= 2; argv += 2;
545 /* Make a new event name */
547 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
548 is_return ? 'r' : 'p', symbol, offset);
550 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
551 is_return ? 'r' : 'p', addr);
554 tp = alloc_trace_probe(group, event, addr, symbol, offset, argc,
557 pr_info("Failed to allocate trace_probe.(%d)\n",
562 /* parse arguments */
564 for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
565 /* Increment count for freeing args in error case */
568 /* Parse argument name */
569 arg = strchr(argv[i], '=');
572 tp->args[i].name = kstrdup(argv[i], GFP_KERNEL);
575 /* If argument name is omitted, set "argN" */
576 snprintf(buf, MAX_EVENT_NAME_LEN, "arg%d", i + 1);
577 tp->args[i].name = kstrdup(buf, GFP_KERNEL);
580 if (!tp->args[i].name) {
581 pr_info("Failed to allocate argument[%d] name.\n", i);
586 if (!is_good_name(tp->args[i].name)) {
587 pr_info("Invalid argument[%d] name: %s\n",
588 i, tp->args[i].name);
593 if (traceprobe_conflict_field_name(tp->args[i].name,
595 pr_info("Argument[%d] name '%s' conflicts with "
596 "another field.\n", i, argv[i]);
601 /* Parse fetch argument */
602 ret = traceprobe_parse_probe_arg(arg, &tp->size, &tp->args[i],
605 pr_info("Parse error at argument[%d]. (%d)\n", i, ret);
610 ret = register_trace_probe(tp);
616 free_trace_probe(tp);
620 static int release_all_trace_probes(void)
622 struct trace_probe *tp;
625 mutex_lock(&probe_lock);
626 /* Ensure no probe is in use. */
627 list_for_each_entry(tp, &probe_list, list)
628 if (trace_probe_is_enabled(tp)) {
632 /* TODO: Use batch unregistration */
633 while (!list_empty(&probe_list)) {
634 tp = list_entry(probe_list.next, struct trace_probe, list);
635 unregister_trace_probe(tp);
636 free_trace_probe(tp);
640 mutex_unlock(&probe_lock);
645 /* Probes listing interfaces */
646 static void *probes_seq_start(struct seq_file *m, loff_t *pos)
648 mutex_lock(&probe_lock);
649 return seq_list_start(&probe_list, *pos);
652 static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
654 return seq_list_next(v, &probe_list, pos);
657 static void probes_seq_stop(struct seq_file *m, void *v)
659 mutex_unlock(&probe_lock);
662 static int probes_seq_show(struct seq_file *m, void *v)
664 struct trace_probe *tp = v;
667 seq_printf(m, "%c", trace_probe_is_return(tp) ? 'r' : 'p');
668 seq_printf(m, ":%s/%s", tp->call.class->system, tp->call.name);
671 seq_printf(m, " 0x%p", tp->rp.kp.addr);
672 else if (tp->rp.kp.offset)
673 seq_printf(m, " %s+%u", trace_probe_symbol(tp),
676 seq_printf(m, " %s", trace_probe_symbol(tp));
678 for (i = 0; i < tp->nr_args; i++)
679 seq_printf(m, " %s=%s", tp->args[i].name, tp->args[i].comm);
685 static const struct seq_operations probes_seq_op = {
686 .start = probes_seq_start,
687 .next = probes_seq_next,
688 .stop = probes_seq_stop,
689 .show = probes_seq_show
692 static int probes_open(struct inode *inode, struct file *file)
696 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
697 ret = release_all_trace_probes();
702 return seq_open(file, &probes_seq_op);
705 static ssize_t probes_write(struct file *file, const char __user *buffer,
706 size_t count, loff_t *ppos)
708 return traceprobe_probes_write(file, buffer, count, ppos,
712 static const struct file_operations kprobe_events_ops = {
713 .owner = THIS_MODULE,
717 .release = seq_release,
718 .write = probes_write,
721 /* Probes profiling interfaces */
722 static int probes_profile_seq_show(struct seq_file *m, void *v)
724 struct trace_probe *tp = v;
726 seq_printf(m, " %-44s %15lu %15lu\n", tp->call.name, tp->nhit,
732 static const struct seq_operations profile_seq_op = {
733 .start = probes_seq_start,
734 .next = probes_seq_next,
735 .stop = probes_seq_stop,
736 .show = probes_profile_seq_show
739 static int profile_open(struct inode *inode, struct file *file)
741 return seq_open(file, &profile_seq_op);
744 static const struct file_operations kprobe_profile_ops = {
745 .owner = THIS_MODULE,
746 .open = profile_open,
749 .release = seq_release,
752 /* Sum up total data length for dynamic arraies (strings) */
753 static __kprobes int __get_data_size(struct trace_probe *tp,
754 struct pt_regs *regs)
759 for (i = 0; i < tp->nr_args; i++)
760 if (unlikely(tp->args[i].fetch_size.fn)) {
761 call_fetch(&tp->args[i].fetch_size, regs, &len);
768 /* Store the value of each argument */
769 static __kprobes void store_trace_args(int ent_size, struct trace_probe *tp,
770 struct pt_regs *regs,
771 u8 *data, int maxlen)
775 u32 *dl; /* Data (relative) location */
777 for (i = 0; i < tp->nr_args; i++) {
778 if (unlikely(tp->args[i].fetch_size.fn)) {
780 * First, we set the relative location and
781 * maximum data length to *dl
783 dl = (u32 *)(data + tp->args[i].offset);
784 *dl = make_data_rloc(maxlen, end - tp->args[i].offset);
785 /* Then try to fetch string or dynamic array data */
786 call_fetch(&tp->args[i].fetch, regs, dl);
787 /* Reduce maximum length */
788 end += get_rloc_len(*dl);
789 maxlen -= get_rloc_len(*dl);
790 /* Trick here, convert data_rloc to data_loc */
791 *dl = convert_rloc_to_loc(*dl,
792 ent_size + tp->args[i].offset);
794 /* Just fetching data normally */
795 call_fetch(&tp->args[i].fetch, regs,
796 data + tp->args[i].offset);
801 static __kprobes void
802 __kprobe_trace_func(struct trace_probe *tp, struct pt_regs *regs,
803 struct ftrace_event_file *ftrace_file)
805 struct kprobe_trace_entry_head *entry;
806 struct ring_buffer_event *event;
807 struct ring_buffer *buffer;
809 unsigned long irq_flags;
810 struct ftrace_event_call *call = &tp->call;
812 WARN_ON(call != ftrace_file->event_call);
814 if (test_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT, &ftrace_file->flags))
817 local_save_flags(irq_flags);
818 pc = preempt_count();
820 dsize = __get_data_size(tp, regs);
821 size = sizeof(*entry) + tp->size + dsize;
823 event = trace_event_buffer_lock_reserve(&buffer, ftrace_file,
825 size, irq_flags, pc);
829 entry = ring_buffer_event_data(event);
830 entry->ip = (unsigned long)tp->rp.kp.addr;
831 store_trace_args(sizeof(*entry), tp, regs, (u8 *)&entry[1], dsize);
833 if (!filter_current_check_discard(buffer, call, entry, event))
834 trace_buffer_unlock_commit_regs(buffer, event,
835 irq_flags, pc, regs);
838 static __kprobes void
839 kprobe_trace_func(struct trace_probe *tp, struct pt_regs *regs)
841 struct event_file_link *link;
843 list_for_each_entry_rcu(link, &tp->files, list)
844 __kprobe_trace_func(tp, regs, link->file);
847 /* Kretprobe handler */
848 static __kprobes void
849 __kretprobe_trace_func(struct trace_probe *tp, struct kretprobe_instance *ri,
850 struct pt_regs *regs,
851 struct ftrace_event_file *ftrace_file)
853 struct kretprobe_trace_entry_head *entry;
854 struct ring_buffer_event *event;
855 struct ring_buffer *buffer;
857 unsigned long irq_flags;
858 struct ftrace_event_call *call = &tp->call;
860 WARN_ON(call != ftrace_file->event_call);
862 if (test_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT, &ftrace_file->flags))
865 local_save_flags(irq_flags);
866 pc = preempt_count();
868 dsize = __get_data_size(tp, regs);
869 size = sizeof(*entry) + tp->size + dsize;
871 event = trace_event_buffer_lock_reserve(&buffer, ftrace_file,
873 size, irq_flags, pc);
877 entry = ring_buffer_event_data(event);
878 entry->func = (unsigned long)tp->rp.kp.addr;
879 entry->ret_ip = (unsigned long)ri->ret_addr;
880 store_trace_args(sizeof(*entry), tp, regs, (u8 *)&entry[1], dsize);
882 if (!filter_current_check_discard(buffer, call, entry, event))
883 trace_buffer_unlock_commit_regs(buffer, event,
884 irq_flags, pc, regs);
887 static __kprobes void
888 kretprobe_trace_func(struct trace_probe *tp, struct kretprobe_instance *ri,
889 struct pt_regs *regs)
891 struct event_file_link *link;
893 list_for_each_entry_rcu(link, &tp->files, list)
894 __kretprobe_trace_func(tp, ri, regs, link->file);
897 /* Event entry printers */
898 static enum print_line_t
899 print_kprobe_event(struct trace_iterator *iter, int flags,
900 struct trace_event *event)
902 struct kprobe_trace_entry_head *field;
903 struct trace_seq *s = &iter->seq;
904 struct trace_probe *tp;
908 field = (struct kprobe_trace_entry_head *)iter->ent;
909 tp = container_of(event, struct trace_probe, call.event);
911 if (!trace_seq_printf(s, "%s: (", tp->call.name))
914 if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
917 if (!trace_seq_puts(s, ")"))
920 data = (u8 *)&field[1];
921 for (i = 0; i < tp->nr_args; i++)
922 if (!tp->args[i].type->print(s, tp->args[i].name,
923 data + tp->args[i].offset, field))
926 if (!trace_seq_puts(s, "\n"))
929 return TRACE_TYPE_HANDLED;
931 return TRACE_TYPE_PARTIAL_LINE;
934 static enum print_line_t
935 print_kretprobe_event(struct trace_iterator *iter, int flags,
936 struct trace_event *event)
938 struct kretprobe_trace_entry_head *field;
939 struct trace_seq *s = &iter->seq;
940 struct trace_probe *tp;
944 field = (struct kretprobe_trace_entry_head *)iter->ent;
945 tp = container_of(event, struct trace_probe, call.event);
947 if (!trace_seq_printf(s, "%s: (", tp->call.name))
950 if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
953 if (!trace_seq_puts(s, " <- "))
956 if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
959 if (!trace_seq_puts(s, ")"))
962 data = (u8 *)&field[1];
963 for (i = 0; i < tp->nr_args; i++)
964 if (!tp->args[i].type->print(s, tp->args[i].name,
965 data + tp->args[i].offset, field))
968 if (!trace_seq_puts(s, "\n"))
971 return TRACE_TYPE_HANDLED;
973 return TRACE_TYPE_PARTIAL_LINE;
977 static int kprobe_event_define_fields(struct ftrace_event_call *event_call)
980 struct kprobe_trace_entry_head field;
981 struct trace_probe *tp = (struct trace_probe *)event_call->data;
983 DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
984 /* Set argument names as fields */
985 for (i = 0; i < tp->nr_args; i++) {
986 ret = trace_define_field(event_call, tp->args[i].type->fmttype,
988 sizeof(field) + tp->args[i].offset,
989 tp->args[i].type->size,
990 tp->args[i].type->is_signed,
998 static int kretprobe_event_define_fields(struct ftrace_event_call *event_call)
1001 struct kretprobe_trace_entry_head field;
1002 struct trace_probe *tp = (struct trace_probe *)event_call->data;
1004 DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1005 DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1006 /* Set argument names as fields */
1007 for (i = 0; i < tp->nr_args; i++) {
1008 ret = trace_define_field(event_call, tp->args[i].type->fmttype,
1010 sizeof(field) + tp->args[i].offset,
1011 tp->args[i].type->size,
1012 tp->args[i].type->is_signed,
1020 static int __set_print_fmt(struct trace_probe *tp, char *buf, int len)
1025 const char *fmt, *arg;
1027 if (!trace_probe_is_return(tp)) {
1029 arg = "REC->" FIELD_STRING_IP;
1031 fmt = "(%lx <- %lx)";
1032 arg = "REC->" FIELD_STRING_FUNC ", REC->" FIELD_STRING_RETIP;
1035 /* When len=0, we just calculate the needed length */
1036 #define LEN_OR_ZERO (len ? len - pos : 0)
1038 pos += snprintf(buf + pos, LEN_OR_ZERO, "\"%s", fmt);
1040 for (i = 0; i < tp->nr_args; i++) {
1041 pos += snprintf(buf + pos, LEN_OR_ZERO, " %s=%s",
1042 tp->args[i].name, tp->args[i].type->fmt);
1045 pos += snprintf(buf + pos, LEN_OR_ZERO, "\", %s", arg);
1047 for (i = 0; i < tp->nr_args; i++) {
1048 if (strcmp(tp->args[i].type->name, "string") == 0)
1049 pos += snprintf(buf + pos, LEN_OR_ZERO,
1053 pos += snprintf(buf + pos, LEN_OR_ZERO, ", REC->%s",
1059 /* return the length of print_fmt */
1063 static int set_print_fmt(struct trace_probe *tp)
1068 /* First: called with 0 length to calculate the needed length */
1069 len = __set_print_fmt(tp, NULL, 0);
1070 print_fmt = kmalloc(len + 1, GFP_KERNEL);
1074 /* Second: actually write the @print_fmt */
1075 __set_print_fmt(tp, print_fmt, len + 1);
1076 tp->call.print_fmt = print_fmt;
1081 #ifdef CONFIG_PERF_EVENTS
1083 /* Kprobe profile handler */
1084 static __kprobes void
1085 kprobe_perf_func(struct trace_probe *tp, struct pt_regs *regs)
1087 struct ftrace_event_call *call = &tp->call;
1088 struct kprobe_trace_entry_head *entry;
1089 struct hlist_head *head;
1090 int size, __size, dsize;
1093 head = this_cpu_ptr(call->perf_events);
1094 if (hlist_empty(head))
1097 dsize = __get_data_size(tp, regs);
1098 __size = sizeof(*entry) + tp->size + dsize;
1099 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1100 size -= sizeof(u32);
1102 entry = perf_trace_buf_prepare(size, call->event.type, regs, &rctx);
1106 entry->ip = (unsigned long)tp->rp.kp.addr;
1107 memset(&entry[1], 0, dsize);
1108 store_trace_args(sizeof(*entry), tp, regs, (u8 *)&entry[1], dsize);
1109 perf_trace_buf_submit(entry, size, rctx, 0, 1, regs, head, NULL);
1112 /* Kretprobe profile handler */
1113 static __kprobes void
1114 kretprobe_perf_func(struct trace_probe *tp, struct kretprobe_instance *ri,
1115 struct pt_regs *regs)
1117 struct ftrace_event_call *call = &tp->call;
1118 struct kretprobe_trace_entry_head *entry;
1119 struct hlist_head *head;
1120 int size, __size, dsize;
1123 head = this_cpu_ptr(call->perf_events);
1124 if (hlist_empty(head))
1127 dsize = __get_data_size(tp, regs);
1128 __size = sizeof(*entry) + tp->size + dsize;
1129 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1130 size -= sizeof(u32);
1132 entry = perf_trace_buf_prepare(size, call->event.type, regs, &rctx);
1136 entry->func = (unsigned long)tp->rp.kp.addr;
1137 entry->ret_ip = (unsigned long)ri->ret_addr;
1138 store_trace_args(sizeof(*entry), tp, regs, (u8 *)&entry[1], dsize);
1139 perf_trace_buf_submit(entry, size, rctx, 0, 1, regs, head, NULL);
1141 #endif /* CONFIG_PERF_EVENTS */
1144 * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex.
1146 * kprobe_trace_self_tests_init() does enable_trace_probe/disable_trace_probe
1147 * lockless, but we can't race with this __init function.
1150 int kprobe_register(struct ftrace_event_call *event,
1151 enum trace_reg type, void *data)
1153 struct trace_probe *tp = (struct trace_probe *)event->data;
1154 struct ftrace_event_file *file = data;
1157 case TRACE_REG_REGISTER:
1158 return enable_trace_probe(tp, file);
1159 case TRACE_REG_UNREGISTER:
1160 return disable_trace_probe(tp, file);
1162 #ifdef CONFIG_PERF_EVENTS
1163 case TRACE_REG_PERF_REGISTER:
1164 return enable_trace_probe(tp, NULL);
1165 case TRACE_REG_PERF_UNREGISTER:
1166 return disable_trace_probe(tp, NULL);
1167 case TRACE_REG_PERF_OPEN:
1168 case TRACE_REG_PERF_CLOSE:
1169 case TRACE_REG_PERF_ADD:
1170 case TRACE_REG_PERF_DEL:
1178 int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1180 struct trace_probe *tp = container_of(kp, struct trace_probe, rp.kp);
1184 if (tp->flags & TP_FLAG_TRACE)
1185 kprobe_trace_func(tp, regs);
1186 #ifdef CONFIG_PERF_EVENTS
1187 if (tp->flags & TP_FLAG_PROFILE)
1188 kprobe_perf_func(tp, regs);
1190 return 0; /* We don't tweek kernel, so just return 0 */
1194 int kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1196 struct trace_probe *tp = container_of(ri->rp, struct trace_probe, rp);
1200 if (tp->flags & TP_FLAG_TRACE)
1201 kretprobe_trace_func(tp, ri, regs);
1202 #ifdef CONFIG_PERF_EVENTS
1203 if (tp->flags & TP_FLAG_PROFILE)
1204 kretprobe_perf_func(tp, ri, regs);
1206 return 0; /* We don't tweek kernel, so just return 0 */
1209 static struct trace_event_functions kretprobe_funcs = {
1210 .trace = print_kretprobe_event
1213 static struct trace_event_functions kprobe_funcs = {
1214 .trace = print_kprobe_event
1217 static int register_probe_event(struct trace_probe *tp)
1219 struct ftrace_event_call *call = &tp->call;
1222 /* Initialize ftrace_event_call */
1223 INIT_LIST_HEAD(&call->class->fields);
1224 if (trace_probe_is_return(tp)) {
1225 call->event.funcs = &kretprobe_funcs;
1226 call->class->define_fields = kretprobe_event_define_fields;
1228 call->event.funcs = &kprobe_funcs;
1229 call->class->define_fields = kprobe_event_define_fields;
1231 if (set_print_fmt(tp) < 0)
1233 ret = register_ftrace_event(&call->event);
1235 kfree(call->print_fmt);
1239 call->class->reg = kprobe_register;
1241 ret = trace_add_event_call(call);
1243 pr_info("Failed to register kprobe event: %s\n", call->name);
1244 kfree(call->print_fmt);
1245 unregister_ftrace_event(&call->event);
1250 static void unregister_probe_event(struct trace_probe *tp)
1252 /* tp->event is unregistered in trace_remove_event_call() */
1253 trace_remove_event_call(&tp->call);
1254 kfree(tp->call.print_fmt);
1257 /* Make a debugfs interface for controlling probe points */
1258 static __init int init_kprobe_trace(void)
1260 struct dentry *d_tracer;
1261 struct dentry *entry;
1263 if (register_module_notifier(&trace_probe_module_nb))
1266 d_tracer = tracing_init_dentry();
1270 entry = debugfs_create_file("kprobe_events", 0644, d_tracer,
1271 NULL, &kprobe_events_ops);
1273 /* Event list interface */
1275 pr_warning("Could not create debugfs "
1276 "'kprobe_events' entry\n");
1278 /* Profile interface */
1279 entry = debugfs_create_file("kprobe_profile", 0444, d_tracer,
1280 NULL, &kprobe_profile_ops);
1283 pr_warning("Could not create debugfs "
1284 "'kprobe_profile' entry\n");
1287 fs_initcall(init_kprobe_trace);
1290 #ifdef CONFIG_FTRACE_STARTUP_TEST
1293 * The "__used" keeps gcc from removing the function symbol
1294 * from the kallsyms table.
1296 static __used int kprobe_trace_selftest_target(int a1, int a2, int a3,
1297 int a4, int a5, int a6)
1299 return a1 + a2 + a3 + a4 + a5 + a6;
1302 static struct ftrace_event_file *
1303 find_trace_probe_file(struct trace_probe *tp, struct trace_array *tr)
1305 struct ftrace_event_file *file;
1307 list_for_each_entry(file, &tr->events, list)
1308 if (file->event_call == &tp->call)
1315 * Nobody but us can call enable_trace_probe/disable_trace_probe at this
1316 * stage, we can do this lockless.
1318 static __init int kprobe_trace_self_tests_init(void)
1321 int (*target)(int, int, int, int, int, int);
1322 struct trace_probe *tp;
1323 struct ftrace_event_file *file;
1325 target = kprobe_trace_selftest_target;
1327 pr_info("Testing kprobe tracing: ");
1329 ret = traceprobe_command("p:testprobe kprobe_trace_selftest_target "
1330 "$stack $stack0 +0($stack)",
1331 create_trace_probe);
1332 if (WARN_ON_ONCE(ret)) {
1333 pr_warn("error on probing function entry.\n");
1336 /* Enable trace point */
1337 tp = find_trace_probe("testprobe", KPROBE_EVENT_SYSTEM);
1338 if (WARN_ON_ONCE(tp == NULL)) {
1339 pr_warn("error on getting new probe.\n");
1342 file = find_trace_probe_file(tp, top_trace_array());
1343 if (WARN_ON_ONCE(file == NULL)) {
1344 pr_warn("error on getting probe file.\n");
1347 enable_trace_probe(tp, file);
1351 ret = traceprobe_command("r:testprobe2 kprobe_trace_selftest_target "
1352 "$retval", create_trace_probe);
1353 if (WARN_ON_ONCE(ret)) {
1354 pr_warn("error on probing function return.\n");
1357 /* Enable trace point */
1358 tp = find_trace_probe("testprobe2", KPROBE_EVENT_SYSTEM);
1359 if (WARN_ON_ONCE(tp == NULL)) {
1360 pr_warn("error on getting 2nd new probe.\n");
1363 file = find_trace_probe_file(tp, top_trace_array());
1364 if (WARN_ON_ONCE(file == NULL)) {
1365 pr_warn("error on getting probe file.\n");
1368 enable_trace_probe(tp, file);
1375 ret = target(1, 2, 3, 4, 5, 6);
1377 /* Disable trace points before removing it */
1378 tp = find_trace_probe("testprobe", KPROBE_EVENT_SYSTEM);
1379 if (WARN_ON_ONCE(tp == NULL)) {
1380 pr_warn("error on getting test probe.\n");
1383 file = find_trace_probe_file(tp, top_trace_array());
1384 if (WARN_ON_ONCE(file == NULL)) {
1385 pr_warn("error on getting probe file.\n");
1388 disable_trace_probe(tp, file);
1391 tp = find_trace_probe("testprobe2", KPROBE_EVENT_SYSTEM);
1392 if (WARN_ON_ONCE(tp == NULL)) {
1393 pr_warn("error on getting 2nd test probe.\n");
1396 file = find_trace_probe_file(tp, top_trace_array());
1397 if (WARN_ON_ONCE(file == NULL)) {
1398 pr_warn("error on getting probe file.\n");
1401 disable_trace_probe(tp, file);
1404 ret = traceprobe_command("-:testprobe", create_trace_probe);
1405 if (WARN_ON_ONCE(ret)) {
1406 pr_warn("error on deleting a probe.\n");
1410 ret = traceprobe_command("-:testprobe2", create_trace_probe);
1411 if (WARN_ON_ONCE(ret)) {
1412 pr_warn("error on deleting a probe.\n");
1417 release_all_trace_probes();
1419 pr_cont("NG: Some tests are failed. Please check them.\n");
1425 late_initcall(kprobe_trace_self_tests_init);