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>
22 #include <linux/kprobes.h>
23 #include <linux/seq_file.h>
24 #include <linux/slab.h>
25 #include <linux/smp.h>
26 #include <linux/debugfs.h>
27 #include <linux/types.h>
28 #include <linux/string.h>
29 #include <linux/ctype.h>
30 #include <linux/ptrace.h>
31 #include <linux/perf_event.h>
34 #include "trace_output.h"
36 #define MAX_TRACE_ARGS 128
37 #define MAX_ARGSTR_LEN 63
38 #define MAX_EVENT_NAME_LEN 64
39 #define KPROBE_EVENT_SYSTEM "kprobes"
41 /* Reserved field names */
42 #define FIELD_STRING_IP "__probe_ip"
43 #define FIELD_STRING_NARGS "__probe_nargs"
44 #define FIELD_STRING_RETIP "__probe_ret_ip"
45 #define FIELD_STRING_FUNC "__probe_func"
47 const char *reserved_field_names[] = {
50 "common_preempt_count",
61 unsigned long (*func)(struct pt_regs *, void *);
65 static __kprobes unsigned long call_fetch(struct fetch_func *f,
68 return f->func(regs, f->data);
72 static __kprobes unsigned long fetch_register(struct pt_regs *regs,
75 return regs_get_register(regs, (unsigned int)((unsigned long)offset));
78 static __kprobes unsigned long fetch_stack(struct pt_regs *regs,
81 return regs_get_kernel_stack_nth(regs,
82 (unsigned int)((unsigned long)num));
85 static __kprobes unsigned long fetch_memory(struct pt_regs *regs, void *addr)
89 if (probe_kernel_address(addr, retval))
94 static __kprobes unsigned long fetch_retvalue(struct pt_regs *regs,
97 return regs_return_value(regs);
100 static __kprobes unsigned long fetch_stack_address(struct pt_regs *regs,
103 return kernel_stack_pointer(regs);
106 /* Memory fetching by symbol */
107 struct symbol_cache {
113 static unsigned long update_symbol_cache(struct symbol_cache *sc)
115 sc->addr = (unsigned long)kallsyms_lookup_name(sc->symbol);
117 sc->addr += sc->offset;
121 static void free_symbol_cache(struct symbol_cache *sc)
127 static struct symbol_cache *alloc_symbol_cache(const char *sym, long offset)
129 struct symbol_cache *sc;
131 if (!sym || strlen(sym) == 0)
133 sc = kzalloc(sizeof(struct symbol_cache), GFP_KERNEL);
137 sc->symbol = kstrdup(sym, GFP_KERNEL);
144 update_symbol_cache(sc);
148 static __kprobes unsigned long fetch_symbol(struct pt_regs *regs, void *data)
150 struct symbol_cache *sc = data;
153 return fetch_memory(regs, (void *)sc->addr);
158 /* Special indirect memory access interface */
159 struct indirect_fetch_data {
160 struct fetch_func orig;
164 static __kprobes unsigned long fetch_indirect(struct pt_regs *regs, void *data)
166 struct indirect_fetch_data *ind = data;
169 addr = call_fetch(&ind->orig, regs);
172 return fetch_memory(regs, (void *)addr);
177 static __kprobes void free_indirect_fetch_data(struct indirect_fetch_data *data)
179 if (data->orig.func == fetch_indirect)
180 free_indirect_fetch_data(data->orig.data);
181 else if (data->orig.func == fetch_symbol)
182 free_symbol_cache(data->orig.data);
187 * Kprobe event core functions
191 struct fetch_func fetch;
195 /* Flags for trace_probe */
196 #define TP_FLAG_TRACE 1
197 #define TP_FLAG_PROFILE 2
200 struct list_head list;
201 struct kretprobe rp; /* Use rp.kp for kprobe use */
203 unsigned int flags; /* For TP_FLAG_* */
204 const char *symbol; /* symbol name */
205 struct ftrace_event_call call;
206 struct trace_event event;
207 unsigned int nr_args;
208 struct probe_arg args[];
211 #define SIZEOF_TRACE_PROBE(n) \
212 (offsetof(struct trace_probe, args) + \
213 (sizeof(struct probe_arg) * (n)))
215 static __kprobes int probe_is_return(struct trace_probe *tp)
217 return tp->rp.handler != NULL;
220 static __kprobes const char *probe_symbol(struct trace_probe *tp)
222 return tp->symbol ? tp->symbol : "unknown";
225 static int probe_arg_string(char *buf, size_t n, struct fetch_func *ff)
229 if (ff->func == fetch_register) {
231 name = regs_query_register_name((unsigned int)((long)ff->data));
232 ret = snprintf(buf, n, "%%%s", name);
233 } else if (ff->func == fetch_stack)
234 ret = snprintf(buf, n, "$stack%lu", (unsigned long)ff->data);
235 else if (ff->func == fetch_memory)
236 ret = snprintf(buf, n, "@0x%p", ff->data);
237 else if (ff->func == fetch_symbol) {
238 struct symbol_cache *sc = ff->data;
240 ret = snprintf(buf, n, "@%s%+ld", sc->symbol,
243 ret = snprintf(buf, n, "@%s", sc->symbol);
244 } else if (ff->func == fetch_retvalue)
245 ret = snprintf(buf, n, "$retval");
246 else if (ff->func == fetch_stack_address)
247 ret = snprintf(buf, n, "$stack");
248 else if (ff->func == fetch_indirect) {
249 struct indirect_fetch_data *id = ff->data;
251 ret = snprintf(buf, n, "%+ld(", id->offset);
255 ret = probe_arg_string(buf + l, n - l, &id->orig);
259 ret = snprintf(buf + l, n - l, ")");
268 static int register_probe_event(struct trace_probe *tp);
269 static void unregister_probe_event(struct trace_probe *tp);
271 static DEFINE_MUTEX(probe_lock);
272 static LIST_HEAD(probe_list);
274 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
275 static int kretprobe_dispatcher(struct kretprobe_instance *ri,
276 struct pt_regs *regs);
278 /* Check the name is good for event/group */
279 static int check_event_name(const char *name)
281 if (!isalpha(*name) && *name != '_')
283 while (*++name != '\0') {
284 if (!isalpha(*name) && !isdigit(*name) && *name != '_')
291 * Allocate new trace_probe and initialize it (including kprobes).
293 static struct trace_probe *alloc_trace_probe(const char *group,
298 int nargs, int is_return)
300 struct trace_probe *tp;
303 tp = kzalloc(SIZEOF_TRACE_PROBE(nargs), GFP_KERNEL);
308 tp->symbol = kstrdup(symbol, GFP_KERNEL);
311 tp->rp.kp.symbol_name = tp->symbol;
312 tp->rp.kp.offset = offs;
314 tp->rp.kp.addr = addr;
317 tp->rp.handler = kretprobe_dispatcher;
319 tp->rp.kp.pre_handler = kprobe_dispatcher;
321 if (!event || !check_event_name(event)) {
326 tp->call.name = kstrdup(event, GFP_KERNEL);
330 if (!group || !check_event_name(group)) {
335 tp->call.system = kstrdup(group, GFP_KERNEL);
336 if (!tp->call.system)
339 INIT_LIST_HEAD(&tp->list);
342 kfree(tp->call.name);
348 static void free_probe_arg(struct probe_arg *arg)
350 if (arg->fetch.func == fetch_symbol)
351 free_symbol_cache(arg->fetch.data);
352 else if (arg->fetch.func == fetch_indirect)
353 free_indirect_fetch_data(arg->fetch.data);
357 static void free_trace_probe(struct trace_probe *tp)
361 for (i = 0; i < tp->nr_args; i++)
362 free_probe_arg(&tp->args[i]);
364 kfree(tp->call.system);
365 kfree(tp->call.name);
370 static struct trace_probe *find_probe_event(const char *event,
373 struct trace_probe *tp;
375 list_for_each_entry(tp, &probe_list, list)
376 if (strcmp(tp->call.name, event) == 0 &&
377 strcmp(tp->call.system, group) == 0)
382 /* Unregister a trace_probe and probe_event: call with locking probe_lock */
383 static void unregister_trace_probe(struct trace_probe *tp)
385 if (probe_is_return(tp))
386 unregister_kretprobe(&tp->rp);
388 unregister_kprobe(&tp->rp.kp);
390 unregister_probe_event(tp);
393 /* Register a trace_probe and probe_event */
394 static int register_trace_probe(struct trace_probe *tp)
396 struct trace_probe *old_tp;
399 mutex_lock(&probe_lock);
401 /* register as an event */
402 old_tp = find_probe_event(tp->call.name, tp->call.system);
404 /* delete old event */
405 unregister_trace_probe(old_tp);
406 free_trace_probe(old_tp);
408 ret = register_probe_event(tp);
410 pr_warning("Faild to register probe event(%d)\n", ret);
414 tp->rp.kp.flags |= KPROBE_FLAG_DISABLED;
415 if (probe_is_return(tp))
416 ret = register_kretprobe(&tp->rp);
418 ret = register_kprobe(&tp->rp.kp);
421 pr_warning("Could not insert probe(%d)\n", ret);
422 if (ret == -EILSEQ) {
423 pr_warning("Probing address(0x%p) is not an "
424 "instruction boundary.\n",
428 unregister_probe_event(tp);
430 list_add_tail(&tp->list, &probe_list);
432 mutex_unlock(&probe_lock);
436 /* Split symbol and offset. */
437 static int split_symbol_offset(char *symbol, unsigned long *offset)
445 tmp = strchr(symbol, '+');
447 /* skip sign because strict_strtol doesn't accept '+' */
448 ret = strict_strtoul(tmp + 1, 0, offset);
457 #define PARAM_MAX_ARGS 16
458 #define PARAM_MAX_STACK (THREAD_SIZE / sizeof(unsigned long))
460 static int parse_probe_vars(char *arg, struct fetch_func *ff, int is_return)
465 if (strcmp(arg, "retval") == 0) {
467 ff->func = fetch_retvalue;
471 } else if (strncmp(arg, "stack", 5) == 0) {
472 if (arg[5] == '\0') {
473 ff->func = fetch_stack_address;
475 } else if (isdigit(arg[5])) {
476 ret = strict_strtoul(arg + 5, 10, ¶m);
477 if (ret || param > PARAM_MAX_STACK)
480 ff->func = fetch_stack;
481 ff->data = (void *)param;
490 /* Recursive argument parser */
491 static int __parse_probe_arg(char *arg, struct fetch_func *ff, int is_return)
500 ret = parse_probe_vars(arg + 1, ff, is_return);
502 case '%': /* named register */
503 ret = regs_query_register_offset(arg + 1);
505 ff->func = fetch_register;
506 ff->data = (void *)(unsigned long)ret;
510 case '@': /* memory or symbol */
511 if (isdigit(arg[1])) {
512 ret = strict_strtoul(arg + 1, 0, ¶m);
515 ff->func = fetch_memory;
516 ff->data = (void *)param;
518 ret = split_symbol_offset(arg + 1, &offset);
521 ff->data = alloc_symbol_cache(arg + 1, offset);
523 ff->func = fetch_symbol;
528 case '+': /* indirect memory */
530 tmp = strchr(arg, '(');
536 ret = strict_strtol(arg + 1, 0, &offset);
542 tmp = strrchr(arg, ')');
544 struct indirect_fetch_data *id;
546 id = kzalloc(sizeof(struct indirect_fetch_data),
551 ret = __parse_probe_arg(arg, &id->orig, is_return);
555 ff->func = fetch_indirect;
556 ff->data = (void *)id;
562 /* TODO: support custom handler */
568 /* String length checking wrapper */
569 static int parse_probe_arg(char *arg, struct fetch_func *ff, int is_return)
571 if (strlen(arg) > MAX_ARGSTR_LEN) {
572 pr_info("Argument is too long.: %s\n", arg);
575 return __parse_probe_arg(arg, ff, is_return);
578 /* Return 1 if name is reserved or already used by another argument */
579 static int conflict_field_name(const char *name,
580 struct probe_arg *args, int narg)
583 for (i = 0; i < ARRAY_SIZE(reserved_field_names); i++)
584 if (strcmp(reserved_field_names[i], name) == 0)
586 for (i = 0; i < narg; i++)
587 if (strcmp(args[i].name, name) == 0)
592 static int create_trace_probe(int argc, char **argv)
596 * - Add kprobe: p[:[GRP/]EVENT] KSYM[+OFFS]|KADDR [FETCHARGS]
597 * - Add kretprobe: r[:[GRP/]EVENT] KSYM[+0] [FETCHARGS]
599 * $retval : fetch return value
600 * $stack : fetch stack address
601 * $stackN : fetch Nth of stack (N:0-)
602 * @ADDR : fetch memory at ADDR (ADDR should be in kernel)
603 * @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
604 * %REG : fetch register REG
605 * Indirect memory fetch:
606 * +|-offs(ARG) : fetch memory at ARG +|- offs address.
607 * Alias name of args:
608 * NAME=FETCHARG : set NAME as alias of FETCHARG.
610 struct trace_probe *tp;
612 int is_return = 0, is_delete = 0;
613 char *symbol = NULL, *event = NULL, *arg = NULL, *group = NULL;
614 unsigned long offset = 0;
616 char buf[MAX_EVENT_NAME_LEN];
618 /* argc must be >= 1 */
619 if (argv[0][0] == 'p')
621 else if (argv[0][0] == 'r')
623 else if (argv[0][0] == '-')
626 pr_info("Probe definition must be started with 'p', 'r' or"
631 if (argv[0][1] == ':') {
633 if (strchr(event, '/')) {
635 event = strchr(group, '/') + 1;
637 if (strlen(group) == 0) {
638 pr_info("Group name is not specified\n");
642 if (strlen(event) == 0) {
643 pr_info("Event name is not specified\n");
648 group = KPROBE_EVENT_SYSTEM;
652 pr_info("Delete command needs an event name.\n");
655 tp = find_probe_event(event, group);
657 pr_info("Event %s/%s doesn't exist.\n", group, event);
660 /* delete an event */
661 unregister_trace_probe(tp);
662 free_trace_probe(tp);
667 pr_info("Probe point is not specified.\n");
670 if (isdigit(argv[1][0])) {
672 pr_info("Return probe point must be a symbol.\n");
675 /* an address specified */
676 ret = strict_strtoul(&argv[1][0], 0, (unsigned long *)&addr);
678 pr_info("Failed to parse address.\n");
682 /* a symbol specified */
684 /* TODO: support .init module functions */
685 ret = split_symbol_offset(symbol, &offset);
687 pr_info("Failed to parse symbol.\n");
690 if (offset && is_return) {
691 pr_info("Return probe must be used without offset.\n");
695 argc -= 2; argv += 2;
699 /* Make a new event name */
701 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
702 is_return ? 'r' : 'p', symbol, offset);
704 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
705 is_return ? 'r' : 'p', addr);
708 tp = alloc_trace_probe(group, event, addr, symbol, offset, argc,
711 pr_info("Failed to allocate trace_probe.(%d)\n",
716 /* parse arguments */
718 for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
719 /* Parse argument name */
720 arg = strchr(argv[i], '=');
726 if (conflict_field_name(argv[i], tp->args, i)) {
727 pr_info("Argument%d name '%s' conflicts with "
728 "another field.\n", i, argv[i]);
733 tp->args[i].name = kstrdup(argv[i], GFP_KERNEL);
734 if (!tp->args[i].name) {
735 pr_info("Failed to allocate argument%d name '%s'.\n",
741 /* Parse fetch argument */
742 ret = parse_probe_arg(arg, &tp->args[i].fetch, is_return);
744 pr_info("Parse error at argument%d. (%d)\n", i, ret);
745 kfree(tp->args[i].name);
752 ret = register_trace_probe(tp);
758 free_trace_probe(tp);
762 static void cleanup_all_probes(void)
764 struct trace_probe *tp;
766 mutex_lock(&probe_lock);
767 /* TODO: Use batch unregistration */
768 while (!list_empty(&probe_list)) {
769 tp = list_entry(probe_list.next, struct trace_probe, list);
770 unregister_trace_probe(tp);
771 free_trace_probe(tp);
773 mutex_unlock(&probe_lock);
777 /* Probes listing interfaces */
778 static void *probes_seq_start(struct seq_file *m, loff_t *pos)
780 mutex_lock(&probe_lock);
781 return seq_list_start(&probe_list, *pos);
784 static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
786 return seq_list_next(v, &probe_list, pos);
789 static void probes_seq_stop(struct seq_file *m, void *v)
791 mutex_unlock(&probe_lock);
794 static int probes_seq_show(struct seq_file *m, void *v)
796 struct trace_probe *tp = v;
798 char buf[MAX_ARGSTR_LEN + 1];
800 seq_printf(m, "%c", probe_is_return(tp) ? 'r' : 'p');
801 seq_printf(m, ":%s/%s", tp->call.system, tp->call.name);
804 seq_printf(m, " 0x%p", tp->rp.kp.addr);
805 else if (tp->rp.kp.offset)
806 seq_printf(m, " %s+%u", probe_symbol(tp), tp->rp.kp.offset);
808 seq_printf(m, " %s", probe_symbol(tp));
810 for (i = 0; i < tp->nr_args; i++) {
811 ret = probe_arg_string(buf, MAX_ARGSTR_LEN, &tp->args[i].fetch);
813 pr_warning("Argument%d decoding error(%d).\n", i, ret);
816 seq_printf(m, " %s=%s", tp->args[i].name, buf);
822 static const struct seq_operations probes_seq_op = {
823 .start = probes_seq_start,
824 .next = probes_seq_next,
825 .stop = probes_seq_stop,
826 .show = probes_seq_show
829 static int probes_open(struct inode *inode, struct file *file)
831 if ((file->f_mode & FMODE_WRITE) &&
832 (file->f_flags & O_TRUNC))
833 cleanup_all_probes();
835 return seq_open(file, &probes_seq_op);
838 static int command_trace_probe(const char *buf)
841 int argc = 0, ret = 0;
843 argv = argv_split(GFP_KERNEL, buf, &argc);
848 ret = create_trace_probe(argc, argv);
854 #define WRITE_BUFSIZE 128
856 static ssize_t probes_write(struct file *file, const char __user *buffer,
857 size_t count, loff_t *ppos)
864 kbuf = kmalloc(WRITE_BUFSIZE, GFP_KERNEL);
869 while (done < count) {
871 if (size >= WRITE_BUFSIZE)
872 size = WRITE_BUFSIZE - 1;
873 if (copy_from_user(kbuf, buffer + done, size)) {
878 tmp = strchr(kbuf, '\n');
881 size = tmp - kbuf + 1;
882 } else if (done + size < count) {
883 pr_warning("Line length is too long: "
884 "Should be less than %d.", WRITE_BUFSIZE);
889 /* Remove comments */
890 tmp = strchr(kbuf, '#');
894 ret = command_trace_probe(kbuf);
904 static const struct file_operations kprobe_events_ops = {
905 .owner = THIS_MODULE,
909 .release = seq_release,
910 .write = probes_write,
913 /* Probes profiling interfaces */
914 static int probes_profile_seq_show(struct seq_file *m, void *v)
916 struct trace_probe *tp = v;
918 seq_printf(m, " %-44s %15lu %15lu\n", tp->call.name, tp->nhit,
924 static const struct seq_operations profile_seq_op = {
925 .start = probes_seq_start,
926 .next = probes_seq_next,
927 .stop = probes_seq_stop,
928 .show = probes_profile_seq_show
931 static int profile_open(struct inode *inode, struct file *file)
933 return seq_open(file, &profile_seq_op);
936 static const struct file_operations kprobe_profile_ops = {
937 .owner = THIS_MODULE,
938 .open = profile_open,
941 .release = seq_release,
945 static __kprobes void kprobe_trace_func(struct kprobe *kp, struct pt_regs *regs)
947 struct trace_probe *tp = container_of(kp, struct trace_probe, rp.kp);
948 struct kprobe_trace_entry *entry;
949 struct ring_buffer_event *event;
950 struct ring_buffer *buffer;
952 unsigned long irq_flags;
953 struct ftrace_event_call *call = &tp->call;
957 local_save_flags(irq_flags);
958 pc = preempt_count();
960 size = SIZEOF_KPROBE_TRACE_ENTRY(tp->nr_args);
962 event = trace_current_buffer_lock_reserve(&buffer, call->id, size,
967 entry = ring_buffer_event_data(event);
968 entry->nargs = tp->nr_args;
969 entry->ip = (unsigned long)kp->addr;
970 for (i = 0; i < tp->nr_args; i++)
971 entry->args[i] = call_fetch(&tp->args[i].fetch, regs);
973 if (!filter_current_check_discard(buffer, call, entry, event))
974 trace_nowake_buffer_unlock_commit(buffer, event, irq_flags, pc);
977 /* Kretprobe handler */
978 static __kprobes void kretprobe_trace_func(struct kretprobe_instance *ri,
979 struct pt_regs *regs)
981 struct trace_probe *tp = container_of(ri->rp, struct trace_probe, rp);
982 struct kretprobe_trace_entry *entry;
983 struct ring_buffer_event *event;
984 struct ring_buffer *buffer;
986 unsigned long irq_flags;
987 struct ftrace_event_call *call = &tp->call;
989 local_save_flags(irq_flags);
990 pc = preempt_count();
992 size = SIZEOF_KRETPROBE_TRACE_ENTRY(tp->nr_args);
994 event = trace_current_buffer_lock_reserve(&buffer, call->id, size,
999 entry = ring_buffer_event_data(event);
1000 entry->nargs = tp->nr_args;
1001 entry->func = (unsigned long)tp->rp.kp.addr;
1002 entry->ret_ip = (unsigned long)ri->ret_addr;
1003 for (i = 0; i < tp->nr_args; i++)
1004 entry->args[i] = call_fetch(&tp->args[i].fetch, regs);
1006 if (!filter_current_check_discard(buffer, call, entry, event))
1007 trace_nowake_buffer_unlock_commit(buffer, event, irq_flags, pc);
1010 /* Event entry printers */
1012 print_kprobe_event(struct trace_iterator *iter, int flags)
1014 struct kprobe_trace_entry *field;
1015 struct trace_seq *s = &iter->seq;
1016 struct trace_event *event;
1017 struct trace_probe *tp;
1020 field = (struct kprobe_trace_entry *)iter->ent;
1021 event = ftrace_find_event(field->ent.type);
1022 tp = container_of(event, struct trace_probe, event);
1024 if (!trace_seq_printf(s, "%s: (", tp->call.name))
1027 if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
1030 if (!trace_seq_puts(s, ")"))
1033 for (i = 0; i < field->nargs; i++)
1034 if (!trace_seq_printf(s, " %s=%lx",
1035 tp->args[i].name, field->args[i]))
1038 if (!trace_seq_puts(s, "\n"))
1041 return TRACE_TYPE_HANDLED;
1043 return TRACE_TYPE_PARTIAL_LINE;
1047 print_kretprobe_event(struct trace_iterator *iter, int flags)
1049 struct kretprobe_trace_entry *field;
1050 struct trace_seq *s = &iter->seq;
1051 struct trace_event *event;
1052 struct trace_probe *tp;
1055 field = (struct kretprobe_trace_entry *)iter->ent;
1056 event = ftrace_find_event(field->ent.type);
1057 tp = container_of(event, struct trace_probe, event);
1059 if (!trace_seq_printf(s, "%s: (", tp->call.name))
1062 if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1065 if (!trace_seq_puts(s, " <- "))
1068 if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1071 if (!trace_seq_puts(s, ")"))
1074 for (i = 0; i < field->nargs; i++)
1075 if (!trace_seq_printf(s, " %s=%lx",
1076 tp->args[i].name, field->args[i]))
1079 if (!trace_seq_puts(s, "\n"))
1082 return TRACE_TYPE_HANDLED;
1084 return TRACE_TYPE_PARTIAL_LINE;
1087 static int probe_event_enable(struct ftrace_event_call *call)
1089 struct trace_probe *tp = (struct trace_probe *)call->data;
1091 tp->flags |= TP_FLAG_TRACE;
1092 if (probe_is_return(tp))
1093 return enable_kretprobe(&tp->rp);
1095 return enable_kprobe(&tp->rp.kp);
1098 static void probe_event_disable(struct ftrace_event_call *call)
1100 struct trace_probe *tp = (struct trace_probe *)call->data;
1102 tp->flags &= ~TP_FLAG_TRACE;
1103 if (!(tp->flags & (TP_FLAG_TRACE | TP_FLAG_PROFILE))) {
1104 if (probe_is_return(tp))
1105 disable_kretprobe(&tp->rp);
1107 disable_kprobe(&tp->rp.kp);
1111 static int probe_event_raw_init(struct ftrace_event_call *event_call)
1113 INIT_LIST_HEAD(&event_call->fields);
1119 #define DEFINE_FIELD(type, item, name, is_signed) \
1121 ret = trace_define_field(event_call, #type, name, \
1122 offsetof(typeof(field), item), \
1123 sizeof(field.item), is_signed, \
1129 static int kprobe_event_define_fields(struct ftrace_event_call *event_call)
1132 struct kprobe_trace_entry field;
1133 struct trace_probe *tp = (struct trace_probe *)event_call->data;
1135 DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1136 DEFINE_FIELD(int, nargs, FIELD_STRING_NARGS, 1);
1137 /* Set argument names as fields */
1138 for (i = 0; i < tp->nr_args; i++)
1139 DEFINE_FIELD(unsigned long, args[i], tp->args[i].name, 0);
1143 static int kretprobe_event_define_fields(struct ftrace_event_call *event_call)
1146 struct kretprobe_trace_entry field;
1147 struct trace_probe *tp = (struct trace_probe *)event_call->data;
1149 DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1150 DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1151 DEFINE_FIELD(int, nargs, FIELD_STRING_NARGS, 1);
1152 /* Set argument names as fields */
1153 for (i = 0; i < tp->nr_args; i++)
1154 DEFINE_FIELD(unsigned long, args[i], tp->args[i].name, 0);
1158 static int __set_print_fmt(struct trace_probe *tp, char *buf, int len)
1163 const char *fmt, *arg;
1165 if (!probe_is_return(tp)) {
1167 arg = "REC->" FIELD_STRING_IP;
1169 fmt = "(%lx <- %lx)";
1170 arg = "REC->" FIELD_STRING_FUNC ", REC->" FIELD_STRING_RETIP;
1173 /* When len=0, we just calculate the needed length */
1174 #define LEN_OR_ZERO (len ? len - pos : 0)
1176 pos += snprintf(buf + pos, LEN_OR_ZERO, "\"%s", fmt);
1178 for (i = 0; i < tp->nr_args; i++) {
1179 pos += snprintf(buf + pos, LEN_OR_ZERO, " %s=%%lx",
1183 pos += snprintf(buf + pos, LEN_OR_ZERO, "\", %s", arg);
1185 for (i = 0; i < tp->nr_args; i++) {
1186 pos += snprintf(buf + pos, LEN_OR_ZERO, ", REC->%s",
1192 /* return the length of print_fmt */
1196 static int set_print_fmt(struct trace_probe *tp)
1201 /* First: called with 0 length to calculate the needed length */
1202 len = __set_print_fmt(tp, NULL, 0);
1203 print_fmt = kmalloc(len + 1, GFP_KERNEL);
1207 /* Second: actually write the @print_fmt */
1208 __set_print_fmt(tp, print_fmt, len + 1);
1209 tp->call.print_fmt = print_fmt;
1214 #ifdef CONFIG_PERF_EVENTS
1216 /* Kprobe profile handler */
1217 static __kprobes void kprobe_profile_func(struct kprobe *kp,
1218 struct pt_regs *regs)
1220 struct trace_probe *tp = container_of(kp, struct trace_probe, rp.kp);
1221 struct ftrace_event_call *call = &tp->call;
1222 struct kprobe_trace_entry *entry;
1223 int size, __size, i;
1224 unsigned long irq_flags;
1227 __size = SIZEOF_KPROBE_TRACE_ENTRY(tp->nr_args);
1228 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1229 size -= sizeof(u32);
1230 if (WARN_ONCE(size > FTRACE_MAX_PROFILE_SIZE,
1231 "profile buffer not large enough"))
1234 entry = ftrace_perf_buf_prepare(size, call->id, &rctx, &irq_flags);
1238 entry->nargs = tp->nr_args;
1239 entry->ip = (unsigned long)kp->addr;
1240 for (i = 0; i < tp->nr_args; i++)
1241 entry->args[i] = call_fetch(&tp->args[i].fetch, regs);
1243 ftrace_perf_buf_submit(entry, size, rctx, entry->ip, 1, irq_flags);
1246 /* Kretprobe profile handler */
1247 static __kprobes void kretprobe_profile_func(struct kretprobe_instance *ri,
1248 struct pt_regs *regs)
1250 struct trace_probe *tp = container_of(ri->rp, struct trace_probe, rp);
1251 struct ftrace_event_call *call = &tp->call;
1252 struct kretprobe_trace_entry *entry;
1253 int size, __size, i;
1254 unsigned long irq_flags;
1257 __size = SIZEOF_KRETPROBE_TRACE_ENTRY(tp->nr_args);
1258 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1259 size -= sizeof(u32);
1260 if (WARN_ONCE(size > FTRACE_MAX_PROFILE_SIZE,
1261 "profile buffer not large enough"))
1264 entry = ftrace_perf_buf_prepare(size, call->id, &rctx, &irq_flags);
1268 entry->nargs = tp->nr_args;
1269 entry->func = (unsigned long)tp->rp.kp.addr;
1270 entry->ret_ip = (unsigned long)ri->ret_addr;
1271 for (i = 0; i < tp->nr_args; i++)
1272 entry->args[i] = call_fetch(&tp->args[i].fetch, regs);
1274 ftrace_perf_buf_submit(entry, size, rctx, entry->ret_ip, 1, irq_flags);
1277 static int probe_profile_enable(struct ftrace_event_call *call)
1279 struct trace_probe *tp = (struct trace_probe *)call->data;
1281 tp->flags |= TP_FLAG_PROFILE;
1283 if (probe_is_return(tp))
1284 return enable_kretprobe(&tp->rp);
1286 return enable_kprobe(&tp->rp.kp);
1289 static void probe_profile_disable(struct ftrace_event_call *call)
1291 struct trace_probe *tp = (struct trace_probe *)call->data;
1293 tp->flags &= ~TP_FLAG_PROFILE;
1295 if (!(tp->flags & TP_FLAG_TRACE)) {
1296 if (probe_is_return(tp))
1297 disable_kretprobe(&tp->rp);
1299 disable_kprobe(&tp->rp.kp);
1302 #endif /* CONFIG_PERF_EVENTS */
1306 int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1308 struct trace_probe *tp = container_of(kp, struct trace_probe, rp.kp);
1310 if (tp->flags & TP_FLAG_TRACE)
1311 kprobe_trace_func(kp, regs);
1312 #ifdef CONFIG_PERF_EVENTS
1313 if (tp->flags & TP_FLAG_PROFILE)
1314 kprobe_profile_func(kp, regs);
1316 return 0; /* We don't tweek kernel, so just return 0 */
1320 int kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1322 struct trace_probe *tp = container_of(ri->rp, struct trace_probe, rp);
1324 if (tp->flags & TP_FLAG_TRACE)
1325 kretprobe_trace_func(ri, regs);
1326 #ifdef CONFIG_PERF_EVENTS
1327 if (tp->flags & TP_FLAG_PROFILE)
1328 kretprobe_profile_func(ri, regs);
1330 return 0; /* We don't tweek kernel, so just return 0 */
1333 static int register_probe_event(struct trace_probe *tp)
1335 struct ftrace_event_call *call = &tp->call;
1338 /* Initialize ftrace_event_call */
1339 if (probe_is_return(tp)) {
1340 tp->event.trace = print_kretprobe_event;
1341 call->raw_init = probe_event_raw_init;
1342 call->define_fields = kretprobe_event_define_fields;
1344 tp->event.trace = print_kprobe_event;
1345 call->raw_init = probe_event_raw_init;
1346 call->define_fields = kprobe_event_define_fields;
1348 if (set_print_fmt(tp) < 0)
1350 call->event = &tp->event;
1351 call->id = register_ftrace_event(&tp->event);
1353 kfree(call->print_fmt);
1357 call->regfunc = probe_event_enable;
1358 call->unregfunc = probe_event_disable;
1360 #ifdef CONFIG_PERF_EVENTS
1361 call->profile_enable = probe_profile_enable;
1362 call->profile_disable = probe_profile_disable;
1365 ret = trace_add_event_call(call);
1367 pr_info("Failed to register kprobe event: %s\n", call->name);
1368 kfree(call->print_fmt);
1369 unregister_ftrace_event(&tp->event);
1374 static void unregister_probe_event(struct trace_probe *tp)
1376 /* tp->event is unregistered in trace_remove_event_call() */
1377 trace_remove_event_call(&tp->call);
1378 kfree(tp->call.print_fmt);
1381 /* Make a debugfs interface for controling probe points */
1382 static __init int init_kprobe_trace(void)
1384 struct dentry *d_tracer;
1385 struct dentry *entry;
1387 d_tracer = tracing_init_dentry();
1391 entry = debugfs_create_file("kprobe_events", 0644, d_tracer,
1392 NULL, &kprobe_events_ops);
1394 /* Event list interface */
1396 pr_warning("Could not create debugfs "
1397 "'kprobe_events' entry\n");
1399 /* Profile interface */
1400 entry = debugfs_create_file("kprobe_profile", 0444, d_tracer,
1401 NULL, &kprobe_profile_ops);
1404 pr_warning("Could not create debugfs "
1405 "'kprobe_profile' entry\n");
1408 fs_initcall(init_kprobe_trace);
1411 #ifdef CONFIG_FTRACE_STARTUP_TEST
1413 static int kprobe_trace_selftest_target(int a1, int a2, int a3,
1414 int a4, int a5, int a6)
1416 return a1 + a2 + a3 + a4 + a5 + a6;
1419 static __init int kprobe_trace_self_tests_init(void)
1422 int (*target)(int, int, int, int, int, int);
1423 struct trace_probe *tp;
1425 target = kprobe_trace_selftest_target;
1427 pr_info("Testing kprobe tracing: ");
1429 ret = command_trace_probe("p:testprobe kprobe_trace_selftest_target "
1430 "$stack $stack0 +0($stack)");
1431 if (WARN_ON_ONCE(ret)) {
1432 pr_warning("error on probing function entry.\n");
1435 /* Enable trace point */
1436 tp = find_probe_event("testprobe", KPROBE_EVENT_SYSTEM);
1437 if (WARN_ON_ONCE(tp == NULL)) {
1438 pr_warning("error on getting new probe.\n");
1441 probe_event_enable(&tp->call);
1444 ret = command_trace_probe("r:testprobe2 kprobe_trace_selftest_target "
1446 if (WARN_ON_ONCE(ret)) {
1447 pr_warning("error on probing function return.\n");
1450 /* Enable trace point */
1451 tp = find_probe_event("testprobe2", KPROBE_EVENT_SYSTEM);
1452 if (WARN_ON_ONCE(tp == NULL)) {
1453 pr_warning("error on getting new probe.\n");
1456 probe_event_enable(&tp->call);
1462 ret = target(1, 2, 3, 4, 5, 6);
1464 ret = command_trace_probe("-:testprobe");
1465 if (WARN_ON_ONCE(ret)) {
1466 pr_warning("error on deleting a probe.\n");
1470 ret = command_trace_probe("-:testprobe2");
1471 if (WARN_ON_ONCE(ret)) {
1472 pr_warning("error on deleting a probe.\n");
1477 cleanup_all_probes();
1479 pr_cont("NG: Some tests are failed. Please check them.\n");
1485 late_initcall(kprobe_trace_self_tests_init);