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
19 #define pr_fmt(fmt) "trace_kprobe: " fmt
21 #include <linux/module.h>
22 #include <linux/uaccess.h>
23 #include <linux/rculist.h>
24 #include <linux/error-injection.h>
26 #include "trace_probe.h"
28 #define KPROBE_EVENT_SYSTEM "kprobes"
29 #define KRETPROBE_MAXACTIVE_MAX 4096
32 * Kprobe event core functions
35 struct list_head list;
36 struct kretprobe rp; /* Use rp.kp for kprobe use */
37 unsigned long __percpu *nhit;
38 const char *symbol; /* symbol name */
39 struct trace_probe tp;
42 #define SIZEOF_TRACE_KPROBE(n) \
43 (offsetof(struct trace_kprobe, tp.args) + \
44 (sizeof(struct probe_arg) * (n)))
46 static nokprobe_inline bool trace_kprobe_is_return(struct trace_kprobe *tk)
48 return tk->rp.handler != NULL;
51 static nokprobe_inline const char *trace_kprobe_symbol(struct trace_kprobe *tk)
53 return tk->symbol ? tk->symbol : "unknown";
56 static nokprobe_inline unsigned long trace_kprobe_offset(struct trace_kprobe *tk)
58 return tk->rp.kp.offset;
61 static nokprobe_inline bool trace_kprobe_has_gone(struct trace_kprobe *tk)
63 return !!(kprobe_gone(&tk->rp.kp));
66 static nokprobe_inline bool trace_kprobe_within_module(struct trace_kprobe *tk,
69 int len = strlen(mod->name);
70 const char *name = trace_kprobe_symbol(tk);
71 return strncmp(mod->name, name, len) == 0 && name[len] == ':';
74 static nokprobe_inline bool trace_kprobe_is_on_module(struct trace_kprobe *tk)
76 return !!strchr(trace_kprobe_symbol(tk), ':');
79 static nokprobe_inline unsigned long trace_kprobe_nhit(struct trace_kprobe *tk)
81 unsigned long nhit = 0;
84 for_each_possible_cpu(cpu)
85 nhit += *per_cpu_ptr(tk->nhit, cpu);
90 bool trace_kprobe_on_func_entry(struct trace_event_call *call)
92 struct trace_kprobe *tk = (struct trace_kprobe *)call->data;
94 return kprobe_on_func_entry(tk->rp.kp.addr,
95 tk->rp.kp.addr ? NULL : tk->rp.kp.symbol_name,
96 tk->rp.kp.addr ? 0 : tk->rp.kp.offset);
99 bool trace_kprobe_error_injectable(struct trace_event_call *call)
101 struct trace_kprobe *tk = (struct trace_kprobe *)call->data;
105 addr = (unsigned long)
106 kallsyms_lookup_name(trace_kprobe_symbol(tk));
107 addr += tk->rp.kp.offset;
109 addr = (unsigned long)tk->rp.kp.addr;
111 return within_error_injection_list(addr);
114 static int register_kprobe_event(struct trace_kprobe *tk);
115 static int unregister_kprobe_event(struct trace_kprobe *tk);
117 static DEFINE_MUTEX(probe_lock);
118 static LIST_HEAD(probe_list);
120 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
121 static int kretprobe_dispatcher(struct kretprobe_instance *ri,
122 struct pt_regs *regs);
124 /* Memory fetching by symbol */
125 struct symbol_cache {
131 unsigned long update_symbol_cache(struct symbol_cache *sc)
133 sc->addr = (unsigned long)kallsyms_lookup_name(sc->symbol);
136 sc->addr += sc->offset;
141 void free_symbol_cache(struct symbol_cache *sc)
147 struct symbol_cache *alloc_symbol_cache(const char *sym, long offset)
149 struct symbol_cache *sc;
151 if (!sym || strlen(sym) == 0)
154 sc = kzalloc(sizeof(struct symbol_cache), GFP_KERNEL);
158 sc->symbol = kstrdup(sym, GFP_KERNEL);
164 update_symbol_cache(sc);
170 * Kprobes-specific fetch functions
172 #define DEFINE_FETCH_stack(type) \
173 static void FETCH_FUNC_NAME(stack, type)(struct pt_regs *regs, \
174 void *offset, void *dest) \
176 *(type *)dest = (type)regs_get_kernel_stack_nth(regs, \
177 (unsigned int)((unsigned long)offset)); \
179 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(stack, type));
181 DEFINE_BASIC_FETCH_FUNCS(stack)
182 /* No string on the stack entry */
183 #define fetch_stack_string NULL
184 #define fetch_stack_string_size NULL
186 #define DEFINE_FETCH_memory(type) \
187 static void FETCH_FUNC_NAME(memory, type)(struct pt_regs *regs, \
188 void *addr, void *dest) \
191 if (probe_kernel_address(addr, retval)) \
194 *(type *)dest = retval; \
196 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(memory, type));
198 DEFINE_BASIC_FETCH_FUNCS(memory)
200 * Fetch a null-terminated string. Caller MUST set *(u32 *)dest with max
201 * length and relative data location.
203 static void FETCH_FUNC_NAME(memory, string)(struct pt_regs *regs,
204 void *addr, void *dest)
206 int maxlen = get_rloc_len(*(u32 *)dest);
207 u8 *dst = get_rloc_data(dest);
214 * Try to get string again, since the string can be changed while
217 ret = strncpy_from_unsafe(dst, addr, maxlen);
219 if (ret < 0) { /* Failed to fetch string */
221 *(u32 *)dest = make_data_rloc(0, get_rloc_offs(*(u32 *)dest));
223 *(u32 *)dest = make_data_rloc(ret, get_rloc_offs(*(u32 *)dest));
226 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(memory, string));
228 /* Return the length of string -- including null terminal byte */
229 static void FETCH_FUNC_NAME(memory, string_size)(struct pt_regs *regs,
230 void *addr, void *dest)
241 ret = __copy_from_user_inatomic(&c, (u8 *)addr + len, 1);
243 } while (c && ret == 0 && len < MAX_STRING_SIZE);
248 if (ret < 0) /* Failed to check the length */
253 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(memory, string_size));
255 #define DEFINE_FETCH_symbol(type) \
256 void FETCH_FUNC_NAME(symbol, type)(struct pt_regs *regs, void *data, void *dest)\
258 struct symbol_cache *sc = data; \
260 fetch_memory_##type(regs, (void *)sc->addr, dest); \
264 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(symbol, type));
266 DEFINE_BASIC_FETCH_FUNCS(symbol)
267 DEFINE_FETCH_symbol(string)
268 DEFINE_FETCH_symbol(string_size)
270 /* kprobes don't support file_offset fetch methods */
271 #define fetch_file_offset_u8 NULL
272 #define fetch_file_offset_u16 NULL
273 #define fetch_file_offset_u32 NULL
274 #define fetch_file_offset_u64 NULL
275 #define fetch_file_offset_string NULL
276 #define fetch_file_offset_string_size NULL
278 /* Fetch type information table */
279 static const struct fetch_type kprobes_fetch_type_table[] = {
281 [FETCH_TYPE_STRING] = __ASSIGN_FETCH_TYPE("string", string, string,
282 sizeof(u32), 1, "__data_loc char[]"),
283 [FETCH_TYPE_STRSIZE] = __ASSIGN_FETCH_TYPE("string_size", u32,
284 string_size, sizeof(u32), 0, "u32"),
286 ASSIGN_FETCH_TYPE(u8, u8, 0),
287 ASSIGN_FETCH_TYPE(u16, u16, 0),
288 ASSIGN_FETCH_TYPE(u32, u32, 0),
289 ASSIGN_FETCH_TYPE(u64, u64, 0),
290 ASSIGN_FETCH_TYPE(s8, u8, 1),
291 ASSIGN_FETCH_TYPE(s16, u16, 1),
292 ASSIGN_FETCH_TYPE(s32, u32, 1),
293 ASSIGN_FETCH_TYPE(s64, u64, 1),
294 ASSIGN_FETCH_TYPE_ALIAS(x8, u8, u8, 0),
295 ASSIGN_FETCH_TYPE_ALIAS(x16, u16, u16, 0),
296 ASSIGN_FETCH_TYPE_ALIAS(x32, u32, u32, 0),
297 ASSIGN_FETCH_TYPE_ALIAS(x64, u64, u64, 0),
299 ASSIGN_FETCH_TYPE_END
303 * Allocate new trace_probe and initialize it (including kprobes).
305 static struct trace_kprobe *alloc_trace_kprobe(const char *group,
311 int nargs, bool is_return)
313 struct trace_kprobe *tk;
316 tk = kzalloc(SIZEOF_TRACE_KPROBE(nargs), GFP_KERNEL);
320 tk->nhit = alloc_percpu(unsigned long);
325 tk->symbol = kstrdup(symbol, GFP_KERNEL);
328 tk->rp.kp.symbol_name = tk->symbol;
329 tk->rp.kp.offset = offs;
331 tk->rp.kp.addr = addr;
334 tk->rp.handler = kretprobe_dispatcher;
336 tk->rp.kp.pre_handler = kprobe_dispatcher;
338 tk->rp.maxactive = maxactive;
340 if (!event || !is_good_name(event)) {
345 tk->tp.call.class = &tk->tp.class;
346 tk->tp.call.name = kstrdup(event, GFP_KERNEL);
347 if (!tk->tp.call.name)
350 if (!group || !is_good_name(group)) {
355 tk->tp.class.system = kstrdup(group, GFP_KERNEL);
356 if (!tk->tp.class.system)
359 INIT_LIST_HEAD(&tk->list);
360 INIT_LIST_HEAD(&tk->tp.files);
363 kfree(tk->tp.call.name);
365 free_percpu(tk->nhit);
370 static void free_trace_kprobe(struct trace_kprobe *tk)
374 for (i = 0; i < tk->tp.nr_args; i++)
375 traceprobe_free_probe_arg(&tk->tp.args[i]);
377 kfree(tk->tp.call.class->system);
378 kfree(tk->tp.call.name);
380 free_percpu(tk->nhit);
384 static struct trace_kprobe *find_trace_kprobe(const char *event,
387 struct trace_kprobe *tk;
389 list_for_each_entry(tk, &probe_list, list)
390 if (strcmp(trace_event_name(&tk->tp.call), event) == 0 &&
391 strcmp(tk->tp.call.class->system, group) == 0)
398 * if the file is NULL, enable "perf" handler, or enable "trace" handler.
401 enable_trace_kprobe(struct trace_kprobe *tk, struct trace_event_file *file)
406 struct event_file_link *link;
408 link = kmalloc(sizeof(*link), GFP_KERNEL);
415 list_add_tail_rcu(&link->list, &tk->tp.files);
417 tk->tp.flags |= TP_FLAG_TRACE;
419 tk->tp.flags |= TP_FLAG_PROFILE;
421 if (trace_probe_is_registered(&tk->tp) && !trace_kprobe_has_gone(tk)) {
422 if (trace_kprobe_is_return(tk))
423 ret = enable_kretprobe(&tk->rp);
425 ret = enable_kprobe(&tk->rp.kp);
432 * Disable trace_probe
433 * if the file is NULL, disable "perf" handler, or disable "trace" handler.
436 disable_trace_kprobe(struct trace_kprobe *tk, struct trace_event_file *file)
438 struct event_file_link *link = NULL;
443 link = find_event_file_link(&tk->tp, file);
449 list_del_rcu(&link->list);
451 if (!list_empty(&tk->tp.files))
454 tk->tp.flags &= ~TP_FLAG_TRACE;
456 tk->tp.flags &= ~TP_FLAG_PROFILE;
458 if (!trace_probe_is_enabled(&tk->tp) && trace_probe_is_registered(&tk->tp)) {
459 if (trace_kprobe_is_return(tk))
460 disable_kretprobe(&tk->rp);
462 disable_kprobe(&tk->rp.kp);
467 * if tk is not added to any list, it must be a local trace_kprobe
468 * created with perf_event_open. We don't need to wait for these
471 if (list_empty(&tk->list))
476 * Synchronize with kprobe_trace_func/kretprobe_trace_func
477 * to ensure disabled (all running handlers are finished).
478 * This is not only for kfree(), but also the caller,
479 * trace_remove_event_call() supposes it for releasing
480 * event_call related objects, which will be accessed in
481 * the kprobe_trace_func/kretprobe_trace_func.
484 kfree(link); /* Ignored if link == NULL */
490 /* Internal register function - just handle k*probes and flags */
491 static int __register_trace_kprobe(struct trace_kprobe *tk)
495 if (trace_probe_is_registered(&tk->tp))
498 for (i = 0; i < tk->tp.nr_args; i++)
499 traceprobe_update_arg(&tk->tp.args[i]);
501 /* Set/clear disabled flag according to tp->flag */
502 if (trace_probe_is_enabled(&tk->tp))
503 tk->rp.kp.flags &= ~KPROBE_FLAG_DISABLED;
505 tk->rp.kp.flags |= KPROBE_FLAG_DISABLED;
507 if (trace_kprobe_is_return(tk))
508 ret = register_kretprobe(&tk->rp);
510 ret = register_kprobe(&tk->rp.kp);
513 tk->tp.flags |= TP_FLAG_REGISTERED;
515 pr_warn("Could not insert probe at %s+%lu: %d\n",
516 trace_kprobe_symbol(tk), trace_kprobe_offset(tk), ret);
517 if (ret == -ENOENT && trace_kprobe_is_on_module(tk)) {
518 pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
520 } else if (ret == -EILSEQ) {
521 pr_warn("Probing address(0x%p) is not an instruction boundary.\n",
530 /* Internal unregister function - just handle k*probes and flags */
531 static void __unregister_trace_kprobe(struct trace_kprobe *tk)
533 if (trace_probe_is_registered(&tk->tp)) {
534 if (trace_kprobe_is_return(tk))
535 unregister_kretprobe(&tk->rp);
537 unregister_kprobe(&tk->rp.kp);
538 tk->tp.flags &= ~TP_FLAG_REGISTERED;
539 /* Cleanup kprobe for reuse */
540 if (tk->rp.kp.symbol_name)
541 tk->rp.kp.addr = NULL;
545 /* Unregister a trace_probe and probe_event: call with locking probe_lock */
546 static int unregister_trace_kprobe(struct trace_kprobe *tk)
548 /* Enabled event can not be unregistered */
549 if (trace_probe_is_enabled(&tk->tp))
552 /* Will fail if probe is being used by ftrace or perf */
553 if (unregister_kprobe_event(tk))
556 __unregister_trace_kprobe(tk);
562 /* Register a trace_probe and probe_event */
563 static int register_trace_kprobe(struct trace_kprobe *tk)
565 struct trace_kprobe *old_tk;
568 mutex_lock(&probe_lock);
570 /* Delete old (same name) event if exist */
571 old_tk = find_trace_kprobe(trace_event_name(&tk->tp.call),
572 tk->tp.call.class->system);
574 ret = unregister_trace_kprobe(old_tk);
577 free_trace_kprobe(old_tk);
580 /* Register new event */
581 ret = register_kprobe_event(tk);
583 pr_warn("Failed to register probe event(%d)\n", ret);
587 /* Register k*probe */
588 ret = __register_trace_kprobe(tk);
590 unregister_kprobe_event(tk);
592 list_add_tail(&tk->list, &probe_list);
595 mutex_unlock(&probe_lock);
599 /* Module notifier call back, checking event on the module */
600 static int trace_kprobe_module_callback(struct notifier_block *nb,
601 unsigned long val, void *data)
603 struct module *mod = data;
604 struct trace_kprobe *tk;
607 if (val != MODULE_STATE_COMING)
610 /* Update probes on coming module */
611 mutex_lock(&probe_lock);
612 list_for_each_entry(tk, &probe_list, list) {
613 if (trace_kprobe_within_module(tk, mod)) {
614 /* Don't need to check busy - this should have gone. */
615 __unregister_trace_kprobe(tk);
616 ret = __register_trace_kprobe(tk);
618 pr_warn("Failed to re-register probe %s on %s: %d\n",
619 trace_event_name(&tk->tp.call),
623 mutex_unlock(&probe_lock);
628 static struct notifier_block trace_kprobe_module_nb = {
629 .notifier_call = trace_kprobe_module_callback,
630 .priority = 1 /* Invoked after kprobe module callback */
633 /* Convert certain expected symbols into '_' when generating event names */
634 static inline void sanitize_event_name(char *name)
636 while (*name++ != '\0')
637 if (*name == ':' || *name == '.')
641 static int create_trace_kprobe(int argc, char **argv)
646 * p[:[GRP/]EVENT] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
648 * r[MAXACTIVE][:[GRP/]EVENT] [MOD:]KSYM[+0] [FETCHARGS]
650 * $retval : fetch return value
651 * $stack : fetch stack address
652 * $stackN : fetch Nth of stack (N:0-)
653 * $comm : fetch current task comm
654 * @ADDR : fetch memory at ADDR (ADDR should be in kernel)
655 * @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
656 * %REG : fetch register REG
657 * Dereferencing memory fetch:
658 * +|-offs(ARG) : fetch memory at ARG +|- offs address.
659 * Alias name of args:
660 * NAME=FETCHARG : set NAME as alias of FETCHARG.
662 * FETCHARG:TYPE : use TYPE instead of unsigned long.
664 struct trace_kprobe *tk;
666 bool is_return = false, is_delete = false;
667 char *symbol = NULL, *event = NULL, *group = NULL;
670 unsigned long offset = 0;
672 char buf[MAX_EVENT_NAME_LEN];
674 /* argc must be >= 1 */
675 if (argv[0][0] == 'p')
677 else if (argv[0][0] == 'r')
679 else if (argv[0][0] == '-')
682 pr_info("Probe definition must be started with 'p', 'r' or"
687 event = strchr(&argv[0][1], ':');
692 if (is_return && isdigit(argv[0][1])) {
693 ret = kstrtouint(&argv[0][1], 0, &maxactive);
695 pr_info("Failed to parse maxactive.\n");
698 /* kretprobes instances are iterated over via a list. The
699 * maximum should stay reasonable.
701 if (maxactive > KRETPROBE_MAXACTIVE_MAX) {
702 pr_info("Maxactive is too big (%d > %d).\n",
703 maxactive, KRETPROBE_MAXACTIVE_MAX);
709 if (strchr(event, '/')) {
711 event = strchr(group, '/') + 1;
713 if (strlen(group) == 0) {
714 pr_info("Group name is not specified\n");
718 if (strlen(event) == 0) {
719 pr_info("Event name is not specified\n");
724 group = KPROBE_EVENT_SYSTEM;
728 pr_info("Delete command needs an event name.\n");
731 mutex_lock(&probe_lock);
732 tk = find_trace_kprobe(event, group);
734 mutex_unlock(&probe_lock);
735 pr_info("Event %s/%s doesn't exist.\n", group, event);
738 /* delete an event */
739 ret = unregister_trace_kprobe(tk);
741 free_trace_kprobe(tk);
742 mutex_unlock(&probe_lock);
747 pr_info("Probe point is not specified.\n");
751 /* try to parse an address. if that fails, try to read the
752 * input as a symbol. */
753 if (kstrtoul(argv[1], 0, (unsigned long *)&addr)) {
754 /* a symbol specified */
756 /* TODO: support .init module functions */
757 ret = traceprobe_split_symbol_offset(symbol, &offset);
759 pr_info("Failed to parse either an address or a symbol.\n");
762 if (offset && is_return &&
763 !kprobe_on_func_entry(NULL, symbol, offset)) {
764 pr_info("Given offset is not valid for return probe.\n");
768 argc -= 2; argv += 2;
772 /* Make a new event name */
774 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
775 is_return ? 'r' : 'p', symbol, offset);
777 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
778 is_return ? 'r' : 'p', addr);
779 sanitize_event_name(buf);
782 tk = alloc_trace_kprobe(group, event, addr, symbol, offset, maxactive,
785 pr_info("Failed to allocate trace_probe.(%d)\n",
790 /* parse arguments */
792 for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
793 struct probe_arg *parg = &tk->tp.args[i];
795 /* Increment count for freeing args in error case */
798 /* Parse argument name */
799 arg = strchr(argv[i], '=');
802 parg->name = kstrdup(argv[i], GFP_KERNEL);
805 /* If argument name is omitted, set "argN" */
806 snprintf(buf, MAX_EVENT_NAME_LEN, "arg%d", i + 1);
807 parg->name = kstrdup(buf, GFP_KERNEL);
811 pr_info("Failed to allocate argument[%d] name.\n", i);
816 if (!is_good_name(parg->name)) {
817 pr_info("Invalid argument[%d] name: %s\n",
823 if (traceprobe_conflict_field_name(parg->name,
825 pr_info("Argument[%d] name '%s' conflicts with "
826 "another field.\n", i, argv[i]);
831 /* Parse fetch argument */
832 ret = traceprobe_parse_probe_arg(arg, &tk->tp.size, parg,
834 kprobes_fetch_type_table);
836 pr_info("Parse error at argument[%d]. (%d)\n", i, ret);
841 ret = register_trace_kprobe(tk);
847 free_trace_kprobe(tk);
851 static int release_all_trace_kprobes(void)
853 struct trace_kprobe *tk;
856 mutex_lock(&probe_lock);
857 /* Ensure no probe is in use. */
858 list_for_each_entry(tk, &probe_list, list)
859 if (trace_probe_is_enabled(&tk->tp)) {
863 /* TODO: Use batch unregistration */
864 while (!list_empty(&probe_list)) {
865 tk = list_entry(probe_list.next, struct trace_kprobe, list);
866 ret = unregister_trace_kprobe(tk);
869 free_trace_kprobe(tk);
873 mutex_unlock(&probe_lock);
878 /* Probes listing interfaces */
879 static void *probes_seq_start(struct seq_file *m, loff_t *pos)
881 mutex_lock(&probe_lock);
882 return seq_list_start(&probe_list, *pos);
885 static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
887 return seq_list_next(v, &probe_list, pos);
890 static void probes_seq_stop(struct seq_file *m, void *v)
892 mutex_unlock(&probe_lock);
895 static int probes_seq_show(struct seq_file *m, void *v)
897 struct trace_kprobe *tk = v;
900 seq_putc(m, trace_kprobe_is_return(tk) ? 'r' : 'p');
901 seq_printf(m, ":%s/%s", tk->tp.call.class->system,
902 trace_event_name(&tk->tp.call));
905 seq_printf(m, " 0x%p", tk->rp.kp.addr);
906 else if (tk->rp.kp.offset)
907 seq_printf(m, " %s+%u", trace_kprobe_symbol(tk),
910 seq_printf(m, " %s", trace_kprobe_symbol(tk));
912 for (i = 0; i < tk->tp.nr_args; i++)
913 seq_printf(m, " %s=%s", tk->tp.args[i].name, tk->tp.args[i].comm);
919 static const struct seq_operations probes_seq_op = {
920 .start = probes_seq_start,
921 .next = probes_seq_next,
922 .stop = probes_seq_stop,
923 .show = probes_seq_show
926 static int probes_open(struct inode *inode, struct file *file)
930 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
931 ret = release_all_trace_kprobes();
936 return seq_open(file, &probes_seq_op);
939 static ssize_t probes_write(struct file *file, const char __user *buffer,
940 size_t count, loff_t *ppos)
942 return trace_parse_run_command(file, buffer, count, ppos,
943 create_trace_kprobe);
946 static const struct file_operations kprobe_events_ops = {
947 .owner = THIS_MODULE,
951 .release = seq_release,
952 .write = probes_write,
955 /* Probes profiling interfaces */
956 static int probes_profile_seq_show(struct seq_file *m, void *v)
958 struct trace_kprobe *tk = v;
960 seq_printf(m, " %-44s %15lu %15lu\n",
961 trace_event_name(&tk->tp.call),
962 trace_kprobe_nhit(tk),
968 static const struct seq_operations profile_seq_op = {
969 .start = probes_seq_start,
970 .next = probes_seq_next,
971 .stop = probes_seq_stop,
972 .show = probes_profile_seq_show
975 static int profile_open(struct inode *inode, struct file *file)
977 return seq_open(file, &profile_seq_op);
980 static const struct file_operations kprobe_profile_ops = {
981 .owner = THIS_MODULE,
982 .open = profile_open,
985 .release = seq_release,
989 static nokprobe_inline void
990 __kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs,
991 struct trace_event_file *trace_file)
993 struct kprobe_trace_entry_head *entry;
994 struct ring_buffer_event *event;
995 struct ring_buffer *buffer;
997 unsigned long irq_flags;
998 struct trace_event_call *call = &tk->tp.call;
1000 WARN_ON(call != trace_file->event_call);
1002 if (trace_trigger_soft_disabled(trace_file))
1005 local_save_flags(irq_flags);
1006 pc = preempt_count();
1008 dsize = __get_data_size(&tk->tp, regs);
1009 size = sizeof(*entry) + tk->tp.size + dsize;
1011 event = trace_event_buffer_lock_reserve(&buffer, trace_file,
1013 size, irq_flags, pc);
1017 entry = ring_buffer_event_data(event);
1018 entry->ip = (unsigned long)tk->rp.kp.addr;
1019 store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1021 event_trigger_unlock_commit_regs(trace_file, buffer, event,
1022 entry, irq_flags, pc, regs);
1026 kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs)
1028 struct event_file_link *link;
1030 list_for_each_entry_rcu(link, &tk->tp.files, list)
1031 __kprobe_trace_func(tk, regs, link->file);
1033 NOKPROBE_SYMBOL(kprobe_trace_func);
1035 /* Kretprobe handler */
1036 static nokprobe_inline void
1037 __kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1038 struct pt_regs *regs,
1039 struct trace_event_file *trace_file)
1041 struct kretprobe_trace_entry_head *entry;
1042 struct ring_buffer_event *event;
1043 struct ring_buffer *buffer;
1044 int size, pc, dsize;
1045 unsigned long irq_flags;
1046 struct trace_event_call *call = &tk->tp.call;
1048 WARN_ON(call != trace_file->event_call);
1050 if (trace_trigger_soft_disabled(trace_file))
1053 local_save_flags(irq_flags);
1054 pc = preempt_count();
1056 dsize = __get_data_size(&tk->tp, regs);
1057 size = sizeof(*entry) + tk->tp.size + dsize;
1059 event = trace_event_buffer_lock_reserve(&buffer, trace_file,
1061 size, irq_flags, pc);
1065 entry = ring_buffer_event_data(event);
1066 entry->func = (unsigned long)tk->rp.kp.addr;
1067 entry->ret_ip = (unsigned long)ri->ret_addr;
1068 store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1070 event_trigger_unlock_commit_regs(trace_file, buffer, event,
1071 entry, irq_flags, pc, regs);
1075 kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1076 struct pt_regs *regs)
1078 struct event_file_link *link;
1080 list_for_each_entry_rcu(link, &tk->tp.files, list)
1081 __kretprobe_trace_func(tk, ri, regs, link->file);
1083 NOKPROBE_SYMBOL(kretprobe_trace_func);
1085 /* Event entry printers */
1086 static enum print_line_t
1087 print_kprobe_event(struct trace_iterator *iter, int flags,
1088 struct trace_event *event)
1090 struct kprobe_trace_entry_head *field;
1091 struct trace_seq *s = &iter->seq;
1092 struct trace_probe *tp;
1096 field = (struct kprobe_trace_entry_head *)iter->ent;
1097 tp = container_of(event, struct trace_probe, call.event);
1099 trace_seq_printf(s, "%s: (", trace_event_name(&tp->call));
1101 if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
1104 trace_seq_putc(s, ')');
1106 data = (u8 *)&field[1];
1107 for (i = 0; i < tp->nr_args; i++)
1108 if (!tp->args[i].type->print(s, tp->args[i].name,
1109 data + tp->args[i].offset, field))
1112 trace_seq_putc(s, '\n');
1114 return trace_handle_return(s);
1117 static enum print_line_t
1118 print_kretprobe_event(struct trace_iterator *iter, int flags,
1119 struct trace_event *event)
1121 struct kretprobe_trace_entry_head *field;
1122 struct trace_seq *s = &iter->seq;
1123 struct trace_probe *tp;
1127 field = (struct kretprobe_trace_entry_head *)iter->ent;
1128 tp = container_of(event, struct trace_probe, call.event);
1130 trace_seq_printf(s, "%s: (", trace_event_name(&tp->call));
1132 if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1135 trace_seq_puts(s, " <- ");
1137 if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1140 trace_seq_putc(s, ')');
1142 data = (u8 *)&field[1];
1143 for (i = 0; i < tp->nr_args; i++)
1144 if (!tp->args[i].type->print(s, tp->args[i].name,
1145 data + tp->args[i].offset, field))
1148 trace_seq_putc(s, '\n');
1151 return trace_handle_return(s);
1155 static int kprobe_event_define_fields(struct trace_event_call *event_call)
1158 struct kprobe_trace_entry_head field;
1159 struct trace_kprobe *tk = (struct trace_kprobe *)event_call->data;
1161 DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1162 /* Set argument names as fields */
1163 for (i = 0; i < tk->tp.nr_args; i++) {
1164 struct probe_arg *parg = &tk->tp.args[i];
1166 ret = trace_define_field(event_call, parg->type->fmttype,
1168 sizeof(field) + parg->offset,
1170 parg->type->is_signed,
1178 static int kretprobe_event_define_fields(struct trace_event_call *event_call)
1181 struct kretprobe_trace_entry_head field;
1182 struct trace_kprobe *tk = (struct trace_kprobe *)event_call->data;
1184 DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1185 DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1186 /* Set argument names as fields */
1187 for (i = 0; i < tk->tp.nr_args; i++) {
1188 struct probe_arg *parg = &tk->tp.args[i];
1190 ret = trace_define_field(event_call, parg->type->fmttype,
1192 sizeof(field) + parg->offset,
1194 parg->type->is_signed,
1202 #ifdef CONFIG_PERF_EVENTS
1204 /* Kprobe profile handler */
1206 kprobe_perf_func(struct trace_kprobe *tk, struct pt_regs *regs)
1208 struct trace_event_call *call = &tk->tp.call;
1209 struct kprobe_trace_entry_head *entry;
1210 struct hlist_head *head;
1211 int size, __size, dsize;
1214 if (bpf_prog_array_valid(call)) {
1215 unsigned long orig_ip = instruction_pointer(regs);
1218 ret = trace_call_bpf(call, regs);
1221 * We need to check and see if we modified the pc of the
1222 * pt_regs, and if so clear the kprobe and return 1 so that we
1223 * don't do the single stepping.
1224 * The ftrace kprobe handler leaves it up to us to re-enable
1225 * preemption here before returning if we've modified the ip.
1227 if (orig_ip != instruction_pointer(regs)) {
1228 reset_current_kprobe();
1229 preempt_enable_no_resched();
1236 head = this_cpu_ptr(call->perf_events);
1237 if (hlist_empty(head))
1240 dsize = __get_data_size(&tk->tp, regs);
1241 __size = sizeof(*entry) + tk->tp.size + dsize;
1242 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1243 size -= sizeof(u32);
1245 entry = perf_trace_buf_alloc(size, NULL, &rctx);
1249 entry->ip = (unsigned long)tk->rp.kp.addr;
1250 memset(&entry[1], 0, dsize);
1251 store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1252 perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1256 NOKPROBE_SYMBOL(kprobe_perf_func);
1258 /* Kretprobe profile handler */
1260 kretprobe_perf_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1261 struct pt_regs *regs)
1263 struct trace_event_call *call = &tk->tp.call;
1264 struct kretprobe_trace_entry_head *entry;
1265 struct hlist_head *head;
1266 int size, __size, dsize;
1269 if (bpf_prog_array_valid(call) && !trace_call_bpf(call, regs))
1272 head = this_cpu_ptr(call->perf_events);
1273 if (hlist_empty(head))
1276 dsize = __get_data_size(&tk->tp, regs);
1277 __size = sizeof(*entry) + tk->tp.size + dsize;
1278 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1279 size -= sizeof(u32);
1281 entry = perf_trace_buf_alloc(size, NULL, &rctx);
1285 entry->func = (unsigned long)tk->rp.kp.addr;
1286 entry->ret_ip = (unsigned long)ri->ret_addr;
1287 store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1288 perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1291 NOKPROBE_SYMBOL(kretprobe_perf_func);
1292 #endif /* CONFIG_PERF_EVENTS */
1295 * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex.
1297 * kprobe_trace_self_tests_init() does enable_trace_probe/disable_trace_probe
1298 * lockless, but we can't race with this __init function.
1300 static int kprobe_register(struct trace_event_call *event,
1301 enum trace_reg type, void *data)
1303 struct trace_kprobe *tk = (struct trace_kprobe *)event->data;
1304 struct trace_event_file *file = data;
1307 case TRACE_REG_REGISTER:
1308 return enable_trace_kprobe(tk, file);
1309 case TRACE_REG_UNREGISTER:
1310 return disable_trace_kprobe(tk, file);
1312 #ifdef CONFIG_PERF_EVENTS
1313 case TRACE_REG_PERF_REGISTER:
1314 return enable_trace_kprobe(tk, NULL);
1315 case TRACE_REG_PERF_UNREGISTER:
1316 return disable_trace_kprobe(tk, NULL);
1317 case TRACE_REG_PERF_OPEN:
1318 case TRACE_REG_PERF_CLOSE:
1319 case TRACE_REG_PERF_ADD:
1320 case TRACE_REG_PERF_DEL:
1327 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1329 struct trace_kprobe *tk = container_of(kp, struct trace_kprobe, rp.kp);
1332 raw_cpu_inc(*tk->nhit);
1334 if (tk->tp.flags & TP_FLAG_TRACE)
1335 kprobe_trace_func(tk, regs);
1336 #ifdef CONFIG_PERF_EVENTS
1337 if (tk->tp.flags & TP_FLAG_PROFILE)
1338 ret = kprobe_perf_func(tk, regs);
1342 NOKPROBE_SYMBOL(kprobe_dispatcher);
1345 kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1347 struct trace_kprobe *tk = container_of(ri->rp, struct trace_kprobe, rp);
1349 raw_cpu_inc(*tk->nhit);
1351 if (tk->tp.flags & TP_FLAG_TRACE)
1352 kretprobe_trace_func(tk, ri, regs);
1353 #ifdef CONFIG_PERF_EVENTS
1354 if (tk->tp.flags & TP_FLAG_PROFILE)
1355 kretprobe_perf_func(tk, ri, regs);
1357 return 0; /* We don't tweek kernel, so just return 0 */
1359 NOKPROBE_SYMBOL(kretprobe_dispatcher);
1361 static struct trace_event_functions kretprobe_funcs = {
1362 .trace = print_kretprobe_event
1365 static struct trace_event_functions kprobe_funcs = {
1366 .trace = print_kprobe_event
1369 static inline void init_trace_event_call(struct trace_kprobe *tk,
1370 struct trace_event_call *call)
1372 INIT_LIST_HEAD(&call->class->fields);
1373 if (trace_kprobe_is_return(tk)) {
1374 call->event.funcs = &kretprobe_funcs;
1375 call->class->define_fields = kretprobe_event_define_fields;
1377 call->event.funcs = &kprobe_funcs;
1378 call->class->define_fields = kprobe_event_define_fields;
1381 call->flags = TRACE_EVENT_FL_KPROBE;
1382 call->class->reg = kprobe_register;
1386 static int register_kprobe_event(struct trace_kprobe *tk)
1388 struct trace_event_call *call = &tk->tp.call;
1391 init_trace_event_call(tk, call);
1393 if (set_print_fmt(&tk->tp, trace_kprobe_is_return(tk)) < 0)
1395 ret = register_trace_event(&call->event);
1397 kfree(call->print_fmt);
1400 ret = trace_add_event_call(call);
1402 pr_info("Failed to register kprobe event: %s\n",
1403 trace_event_name(call));
1404 kfree(call->print_fmt);
1405 unregister_trace_event(&call->event);
1410 static int unregister_kprobe_event(struct trace_kprobe *tk)
1414 /* tp->event is unregistered in trace_remove_event_call() */
1415 ret = trace_remove_event_call(&tk->tp.call);
1417 kfree(tk->tp.call.print_fmt);
1421 #ifdef CONFIG_PERF_EVENTS
1422 /* create a trace_kprobe, but don't add it to global lists */
1423 struct trace_event_call *
1424 create_local_trace_kprobe(char *func, void *addr, unsigned long offs,
1427 struct trace_kprobe *tk;
1432 * local trace_kprobes are not added to probe_list, so they are never
1433 * searched in find_trace_kprobe(). Therefore, there is no concern of
1434 * duplicated name here.
1436 event = func ? func : "DUMMY_EVENT";
1438 tk = alloc_trace_kprobe(KPROBE_EVENT_SYSTEM, event, (void *)addr, func,
1439 offs, 0 /* maxactive */, 0 /* nargs */,
1443 pr_info("Failed to allocate trace_probe.(%d)\n",
1445 return ERR_CAST(tk);
1448 init_trace_event_call(tk, &tk->tp.call);
1450 if (set_print_fmt(&tk->tp, trace_kprobe_is_return(tk)) < 0) {
1455 ret = __register_trace_kprobe(tk);
1459 return &tk->tp.call;
1461 free_trace_kprobe(tk);
1462 return ERR_PTR(ret);
1465 void destroy_local_trace_kprobe(struct trace_event_call *event_call)
1467 struct trace_kprobe *tk;
1469 tk = container_of(event_call, struct trace_kprobe, tp.call);
1471 if (trace_probe_is_enabled(&tk->tp)) {
1476 __unregister_trace_kprobe(tk);
1477 free_trace_kprobe(tk);
1479 #endif /* CONFIG_PERF_EVENTS */
1481 /* Make a tracefs interface for controlling probe points */
1482 static __init int init_kprobe_trace(void)
1484 struct dentry *d_tracer;
1485 struct dentry *entry;
1487 if (register_module_notifier(&trace_kprobe_module_nb))
1490 d_tracer = tracing_init_dentry();
1491 if (IS_ERR(d_tracer))
1494 entry = tracefs_create_file("kprobe_events", 0644, d_tracer,
1495 NULL, &kprobe_events_ops);
1497 /* Event list interface */
1499 pr_warn("Could not create tracefs 'kprobe_events' entry\n");
1501 /* Profile interface */
1502 entry = tracefs_create_file("kprobe_profile", 0444, d_tracer,
1503 NULL, &kprobe_profile_ops);
1506 pr_warn("Could not create tracefs 'kprobe_profile' entry\n");
1509 fs_initcall(init_kprobe_trace);
1512 #ifdef CONFIG_FTRACE_STARTUP_TEST
1514 * The "__used" keeps gcc from removing the function symbol
1515 * from the kallsyms table. 'noinline' makes sure that there
1516 * isn't an inlined version used by the test method below
1518 static __used __init noinline int
1519 kprobe_trace_selftest_target(int a1, int a2, int a3, int a4, int a5, int a6)
1521 return a1 + a2 + a3 + a4 + a5 + a6;
1524 static __init struct trace_event_file *
1525 find_trace_probe_file(struct trace_kprobe *tk, struct trace_array *tr)
1527 struct trace_event_file *file;
1529 list_for_each_entry(file, &tr->events, list)
1530 if (file->event_call == &tk->tp.call)
1537 * Nobody but us can call enable_trace_kprobe/disable_trace_kprobe at this
1538 * stage, we can do this lockless.
1540 static __init int kprobe_trace_self_tests_init(void)
1543 int (*target)(int, int, int, int, int, int);
1544 struct trace_kprobe *tk;
1545 struct trace_event_file *file;
1547 if (tracing_is_disabled())
1550 target = kprobe_trace_selftest_target;
1552 pr_info("Testing kprobe tracing: ");
1554 ret = trace_run_command("p:testprobe kprobe_trace_selftest_target "
1555 "$stack $stack0 +0($stack)",
1556 create_trace_kprobe);
1557 if (WARN_ON_ONCE(ret)) {
1558 pr_warn("error on probing function entry.\n");
1561 /* Enable trace point */
1562 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1563 if (WARN_ON_ONCE(tk == NULL)) {
1564 pr_warn("error on getting new probe.\n");
1567 file = find_trace_probe_file(tk, top_trace_array());
1568 if (WARN_ON_ONCE(file == NULL)) {
1569 pr_warn("error on getting probe file.\n");
1572 enable_trace_kprobe(tk, file);
1576 ret = trace_run_command("r:testprobe2 kprobe_trace_selftest_target "
1577 "$retval", create_trace_kprobe);
1578 if (WARN_ON_ONCE(ret)) {
1579 pr_warn("error on probing function return.\n");
1582 /* Enable trace point */
1583 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1584 if (WARN_ON_ONCE(tk == NULL)) {
1585 pr_warn("error on getting 2nd new probe.\n");
1588 file = find_trace_probe_file(tk, top_trace_array());
1589 if (WARN_ON_ONCE(file == NULL)) {
1590 pr_warn("error on getting probe file.\n");
1593 enable_trace_kprobe(tk, file);
1600 ret = target(1, 2, 3, 4, 5, 6);
1603 * Not expecting an error here, the check is only to prevent the
1604 * optimizer from removing the call to target() as otherwise there
1605 * are no side-effects and the call is never performed.
1610 /* Disable trace points before removing it */
1611 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1612 if (WARN_ON_ONCE(tk == NULL)) {
1613 pr_warn("error on getting test probe.\n");
1616 if (trace_kprobe_nhit(tk) != 1) {
1617 pr_warn("incorrect number of testprobe hits\n");
1621 file = find_trace_probe_file(tk, top_trace_array());
1622 if (WARN_ON_ONCE(file == NULL)) {
1623 pr_warn("error on getting probe file.\n");
1626 disable_trace_kprobe(tk, file);
1629 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1630 if (WARN_ON_ONCE(tk == NULL)) {
1631 pr_warn("error on getting 2nd test probe.\n");
1634 if (trace_kprobe_nhit(tk) != 1) {
1635 pr_warn("incorrect number of testprobe2 hits\n");
1639 file = find_trace_probe_file(tk, top_trace_array());
1640 if (WARN_ON_ONCE(file == NULL)) {
1641 pr_warn("error on getting probe file.\n");
1644 disable_trace_kprobe(tk, file);
1647 ret = trace_run_command("-:testprobe", create_trace_kprobe);
1648 if (WARN_ON_ONCE(ret)) {
1649 pr_warn("error on deleting a probe.\n");
1653 ret = trace_run_command("-:testprobe2", create_trace_kprobe);
1654 if (WARN_ON_ONCE(ret)) {
1655 pr_warn("error on deleting a probe.\n");
1660 release_all_trace_kprobes();
1662 * Wait for the optimizer work to finish. Otherwise it might fiddle
1663 * with probes in already freed __init text.
1665 wait_for_kprobe_optimizer();
1667 pr_cont("NG: Some tests are failed. Please check them.\n");
1673 late_initcall(kprobe_trace_self_tests_init);