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)
403 struct event_file_link *link = NULL;
407 link = kmalloc(sizeof(*link), GFP_KERNEL);
414 list_add_tail_rcu(&link->list, &tk->tp.files);
416 tk->tp.flags |= TP_FLAG_TRACE;
418 tk->tp.flags |= TP_FLAG_PROFILE;
420 if (trace_probe_is_registered(&tk->tp) && !trace_kprobe_has_gone(tk)) {
421 if (trace_kprobe_is_return(tk))
422 ret = enable_kretprobe(&tk->rp);
424 ret = enable_kprobe(&tk->rp.kp);
429 /* Notice the if is true on not WARN() */
430 if (!WARN_ON_ONCE(!link))
431 list_del_rcu(&link->list);
433 tk->tp.flags &= ~TP_FLAG_TRACE;
435 tk->tp.flags &= ~TP_FLAG_PROFILE;
443 * Disable trace_probe
444 * if the file is NULL, disable "perf" handler, or disable "trace" handler.
447 disable_trace_kprobe(struct trace_kprobe *tk, struct trace_event_file *file)
449 struct event_file_link *link = NULL;
454 link = find_event_file_link(&tk->tp, file);
460 list_del_rcu(&link->list);
462 if (!list_empty(&tk->tp.files))
465 tk->tp.flags &= ~TP_FLAG_TRACE;
467 tk->tp.flags &= ~TP_FLAG_PROFILE;
469 if (!trace_probe_is_enabled(&tk->tp) && trace_probe_is_registered(&tk->tp)) {
470 if (trace_kprobe_is_return(tk))
471 disable_kretprobe(&tk->rp);
473 disable_kprobe(&tk->rp.kp);
478 * if tk is not added to any list, it must be a local trace_kprobe
479 * created with perf_event_open. We don't need to wait for these
482 if (list_empty(&tk->list))
487 * Synchronize with kprobe_trace_func/kretprobe_trace_func
488 * to ensure disabled (all running handlers are finished).
489 * This is not only for kfree(), but also the caller,
490 * trace_remove_event_call() supposes it for releasing
491 * event_call related objects, which will be accessed in
492 * the kprobe_trace_func/kretprobe_trace_func.
495 kfree(link); /* Ignored if link == NULL */
501 /* Internal register function - just handle k*probes and flags */
502 static int __register_trace_kprobe(struct trace_kprobe *tk)
506 if (trace_probe_is_registered(&tk->tp))
509 for (i = 0; i < tk->tp.nr_args; i++)
510 traceprobe_update_arg(&tk->tp.args[i]);
512 /* Set/clear disabled flag according to tp->flag */
513 if (trace_probe_is_enabled(&tk->tp))
514 tk->rp.kp.flags &= ~KPROBE_FLAG_DISABLED;
516 tk->rp.kp.flags |= KPROBE_FLAG_DISABLED;
518 if (trace_kprobe_is_return(tk))
519 ret = register_kretprobe(&tk->rp);
521 ret = register_kprobe(&tk->rp.kp);
524 tk->tp.flags |= TP_FLAG_REGISTERED;
526 if (ret == -ENOENT && trace_kprobe_is_on_module(tk)) {
527 pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
529 } else if (ret == -EILSEQ) {
530 pr_warn("Probing address(0x%p) is not an instruction boundary.\n",
539 /* Internal unregister function - just handle k*probes and flags */
540 static void __unregister_trace_kprobe(struct trace_kprobe *tk)
542 if (trace_probe_is_registered(&tk->tp)) {
543 if (trace_kprobe_is_return(tk))
544 unregister_kretprobe(&tk->rp);
546 unregister_kprobe(&tk->rp.kp);
547 tk->tp.flags &= ~TP_FLAG_REGISTERED;
548 /* Cleanup kprobe for reuse */
549 if (tk->rp.kp.symbol_name)
550 tk->rp.kp.addr = NULL;
554 /* Unregister a trace_probe and probe_event: call with locking probe_lock */
555 static int unregister_trace_kprobe(struct trace_kprobe *tk)
557 /* Enabled event can not be unregistered */
558 if (trace_probe_is_enabled(&tk->tp))
561 /* Will fail if probe is being used by ftrace or perf */
562 if (unregister_kprobe_event(tk))
565 __unregister_trace_kprobe(tk);
571 /* Register a trace_probe and probe_event */
572 static int register_trace_kprobe(struct trace_kprobe *tk)
574 struct trace_kprobe *old_tk;
577 mutex_lock(&probe_lock);
579 /* Delete old (same name) event if exist */
580 old_tk = find_trace_kprobe(trace_event_name(&tk->tp.call),
581 tk->tp.call.class->system);
583 ret = unregister_trace_kprobe(old_tk);
586 free_trace_kprobe(old_tk);
589 /* Register new event */
590 ret = register_kprobe_event(tk);
592 pr_warn("Failed to register probe event(%d)\n", ret);
596 /* Register k*probe */
597 ret = __register_trace_kprobe(tk);
599 unregister_kprobe_event(tk);
601 list_add_tail(&tk->list, &probe_list);
604 mutex_unlock(&probe_lock);
608 /* Module notifier call back, checking event on the module */
609 static int trace_kprobe_module_callback(struct notifier_block *nb,
610 unsigned long val, void *data)
612 struct module *mod = data;
613 struct trace_kprobe *tk;
616 if (val != MODULE_STATE_COMING)
619 /* Update probes on coming module */
620 mutex_lock(&probe_lock);
621 list_for_each_entry(tk, &probe_list, list) {
622 if (trace_kprobe_within_module(tk, mod)) {
623 /* Don't need to check busy - this should have gone. */
624 __unregister_trace_kprobe(tk);
625 ret = __register_trace_kprobe(tk);
627 pr_warn("Failed to re-register probe %s on %s: %d\n",
628 trace_event_name(&tk->tp.call),
632 mutex_unlock(&probe_lock);
637 static struct notifier_block trace_kprobe_module_nb = {
638 .notifier_call = trace_kprobe_module_callback,
639 .priority = 1 /* Invoked after kprobe module callback */
642 /* Convert certain expected symbols into '_' when generating event names */
643 static inline void sanitize_event_name(char *name)
645 while (*name++ != '\0')
646 if (*name == ':' || *name == '.')
650 static int create_trace_kprobe(int argc, char **argv)
655 * p[:[GRP/]EVENT] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
657 * r[MAXACTIVE][:[GRP/]EVENT] [MOD:]KSYM[+0] [FETCHARGS]
659 * $retval : fetch return value
660 * $stack : fetch stack address
661 * $stackN : fetch Nth of stack (N:0-)
662 * $comm : fetch current task comm
663 * @ADDR : fetch memory at ADDR (ADDR should be in kernel)
664 * @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
665 * %REG : fetch register REG
666 * Dereferencing memory fetch:
667 * +|-offs(ARG) : fetch memory at ARG +|- offs address.
668 * Alias name of args:
669 * NAME=FETCHARG : set NAME as alias of FETCHARG.
671 * FETCHARG:TYPE : use TYPE instead of unsigned long.
673 struct trace_kprobe *tk;
675 bool is_return = false, is_delete = false;
676 char *symbol = NULL, *event = NULL, *group = NULL;
681 char buf[MAX_EVENT_NAME_LEN];
683 /* argc must be >= 1 */
684 if (argv[0][0] == 'p')
686 else if (argv[0][0] == 'r')
688 else if (argv[0][0] == '-')
691 pr_info("Probe definition must be started with 'p', 'r' or"
696 event = strchr(&argv[0][1], ':');
701 if (is_return && isdigit(argv[0][1])) {
702 ret = kstrtouint(&argv[0][1], 0, &maxactive);
704 pr_info("Failed to parse maxactive.\n");
707 /* kretprobes instances are iterated over via a list. The
708 * maximum should stay reasonable.
710 if (maxactive > KRETPROBE_MAXACTIVE_MAX) {
711 pr_info("Maxactive is too big (%d > %d).\n",
712 maxactive, KRETPROBE_MAXACTIVE_MAX);
718 if (strchr(event, '/')) {
720 event = strchr(group, '/') + 1;
722 if (strlen(group) == 0) {
723 pr_info("Group name is not specified\n");
727 if (strlen(event) == 0) {
728 pr_info("Event name is not specified\n");
733 group = KPROBE_EVENT_SYSTEM;
737 pr_info("Delete command needs an event name.\n");
740 mutex_lock(&probe_lock);
741 tk = find_trace_kprobe(event, group);
743 mutex_unlock(&probe_lock);
744 pr_info("Event %s/%s doesn't exist.\n", group, event);
747 /* delete an event */
748 ret = unregister_trace_kprobe(tk);
750 free_trace_kprobe(tk);
751 mutex_unlock(&probe_lock);
756 pr_info("Probe point is not specified.\n");
760 /* try to parse an address. if that fails, try to read the
761 * input as a symbol. */
762 if (kstrtoul(argv[1], 0, (unsigned long *)&addr)) {
763 /* a symbol specified */
765 /* TODO: support .init module functions */
766 ret = traceprobe_split_symbol_offset(symbol, &offset);
767 if (ret || offset < 0 || offset > UINT_MAX) {
768 pr_info("Failed to parse either an address or a symbol.\n");
771 if (offset && is_return &&
772 !kprobe_on_func_entry(NULL, symbol, offset)) {
773 pr_info("Given offset is not valid for return probe.\n");
777 argc -= 2; argv += 2;
781 /* Make a new event name */
783 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
784 is_return ? 'r' : 'p', symbol, offset);
786 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
787 is_return ? 'r' : 'p', addr);
788 sanitize_event_name(buf);
791 tk = alloc_trace_kprobe(group, event, addr, symbol, offset, maxactive,
794 pr_info("Failed to allocate trace_probe.(%d)\n",
799 /* parse arguments */
801 for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
802 struct probe_arg *parg = &tk->tp.args[i];
804 /* Increment count for freeing args in error case */
807 /* Parse argument name */
808 arg = strchr(argv[i], '=');
811 parg->name = kstrdup(argv[i], GFP_KERNEL);
814 /* If argument name is omitted, set "argN" */
815 snprintf(buf, MAX_EVENT_NAME_LEN, "arg%d", i + 1);
816 parg->name = kstrdup(buf, GFP_KERNEL);
820 pr_info("Failed to allocate argument[%d] name.\n", i);
825 if (!is_good_name(parg->name)) {
826 pr_info("Invalid argument[%d] name: %s\n",
832 if (traceprobe_conflict_field_name(parg->name,
834 pr_info("Argument[%d] name '%s' conflicts with "
835 "another field.\n", i, argv[i]);
840 /* Parse fetch argument */
841 ret = traceprobe_parse_probe_arg(arg, &tk->tp.size, parg,
843 kprobes_fetch_type_table);
845 pr_info("Parse error at argument[%d]. (%d)\n", i, ret);
850 ret = register_trace_kprobe(tk);
856 free_trace_kprobe(tk);
860 static int release_all_trace_kprobes(void)
862 struct trace_kprobe *tk;
865 mutex_lock(&probe_lock);
866 /* Ensure no probe is in use. */
867 list_for_each_entry(tk, &probe_list, list)
868 if (trace_probe_is_enabled(&tk->tp)) {
872 /* TODO: Use batch unregistration */
873 while (!list_empty(&probe_list)) {
874 tk = list_entry(probe_list.next, struct trace_kprobe, list);
875 ret = unregister_trace_kprobe(tk);
878 free_trace_kprobe(tk);
882 mutex_unlock(&probe_lock);
887 /* Probes listing interfaces */
888 static void *probes_seq_start(struct seq_file *m, loff_t *pos)
890 mutex_lock(&probe_lock);
891 return seq_list_start(&probe_list, *pos);
894 static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
896 return seq_list_next(v, &probe_list, pos);
899 static void probes_seq_stop(struct seq_file *m, void *v)
901 mutex_unlock(&probe_lock);
904 static int probes_seq_show(struct seq_file *m, void *v)
906 struct trace_kprobe *tk = v;
909 seq_putc(m, trace_kprobe_is_return(tk) ? 'r' : 'p');
910 seq_printf(m, ":%s/%s", tk->tp.call.class->system,
911 trace_event_name(&tk->tp.call));
914 seq_printf(m, " 0x%p", tk->rp.kp.addr);
915 else if (tk->rp.kp.offset)
916 seq_printf(m, " %s+%u", trace_kprobe_symbol(tk),
919 seq_printf(m, " %s", trace_kprobe_symbol(tk));
921 for (i = 0; i < tk->tp.nr_args; i++)
922 seq_printf(m, " %s=%s", tk->tp.args[i].name, tk->tp.args[i].comm);
928 static const struct seq_operations probes_seq_op = {
929 .start = probes_seq_start,
930 .next = probes_seq_next,
931 .stop = probes_seq_stop,
932 .show = probes_seq_show
935 static int probes_open(struct inode *inode, struct file *file)
939 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
940 ret = release_all_trace_kprobes();
945 return seq_open(file, &probes_seq_op);
948 static ssize_t probes_write(struct file *file, const char __user *buffer,
949 size_t count, loff_t *ppos)
951 return trace_parse_run_command(file, buffer, count, ppos,
952 create_trace_kprobe);
955 static const struct file_operations kprobe_events_ops = {
956 .owner = THIS_MODULE,
960 .release = seq_release,
961 .write = probes_write,
964 /* Probes profiling interfaces */
965 static int probes_profile_seq_show(struct seq_file *m, void *v)
967 struct trace_kprobe *tk = v;
969 seq_printf(m, " %-44s %15lu %15lu\n",
970 trace_event_name(&tk->tp.call),
971 trace_kprobe_nhit(tk),
977 static const struct seq_operations profile_seq_op = {
978 .start = probes_seq_start,
979 .next = probes_seq_next,
980 .stop = probes_seq_stop,
981 .show = probes_profile_seq_show
984 static int profile_open(struct inode *inode, struct file *file)
986 return seq_open(file, &profile_seq_op);
989 static const struct file_operations kprobe_profile_ops = {
990 .owner = THIS_MODULE,
991 .open = profile_open,
994 .release = seq_release,
998 static nokprobe_inline void
999 __kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs,
1000 struct trace_event_file *trace_file)
1002 struct kprobe_trace_entry_head *entry;
1003 struct ring_buffer_event *event;
1004 struct ring_buffer *buffer;
1005 int size, dsize, pc;
1006 unsigned long irq_flags;
1007 struct trace_event_call *call = &tk->tp.call;
1009 WARN_ON(call != trace_file->event_call);
1011 if (trace_trigger_soft_disabled(trace_file))
1014 local_save_flags(irq_flags);
1015 pc = preempt_count();
1017 dsize = __get_data_size(&tk->tp, regs);
1018 size = sizeof(*entry) + tk->tp.size + dsize;
1020 event = trace_event_buffer_lock_reserve(&buffer, trace_file,
1022 size, irq_flags, pc);
1026 entry = ring_buffer_event_data(event);
1027 entry->ip = (unsigned long)tk->rp.kp.addr;
1028 store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1030 event_trigger_unlock_commit_regs(trace_file, buffer, event,
1031 entry, irq_flags, pc, regs);
1035 kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs)
1037 struct event_file_link *link;
1039 list_for_each_entry_rcu(link, &tk->tp.files, list)
1040 __kprobe_trace_func(tk, regs, link->file);
1042 NOKPROBE_SYMBOL(kprobe_trace_func);
1044 /* Kretprobe handler */
1045 static nokprobe_inline void
1046 __kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1047 struct pt_regs *regs,
1048 struct trace_event_file *trace_file)
1050 struct kretprobe_trace_entry_head *entry;
1051 struct ring_buffer_event *event;
1052 struct ring_buffer *buffer;
1053 int size, pc, dsize;
1054 unsigned long irq_flags;
1055 struct trace_event_call *call = &tk->tp.call;
1057 WARN_ON(call != trace_file->event_call);
1059 if (trace_trigger_soft_disabled(trace_file))
1062 local_save_flags(irq_flags);
1063 pc = preempt_count();
1065 dsize = __get_data_size(&tk->tp, regs);
1066 size = sizeof(*entry) + tk->tp.size + dsize;
1068 event = trace_event_buffer_lock_reserve(&buffer, trace_file,
1070 size, irq_flags, pc);
1074 entry = ring_buffer_event_data(event);
1075 entry->func = (unsigned long)tk->rp.kp.addr;
1076 entry->ret_ip = (unsigned long)ri->ret_addr;
1077 store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1079 event_trigger_unlock_commit_regs(trace_file, buffer, event,
1080 entry, irq_flags, pc, regs);
1084 kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1085 struct pt_regs *regs)
1087 struct event_file_link *link;
1089 list_for_each_entry_rcu(link, &tk->tp.files, list)
1090 __kretprobe_trace_func(tk, ri, regs, link->file);
1092 NOKPROBE_SYMBOL(kretprobe_trace_func);
1094 /* Event entry printers */
1095 static enum print_line_t
1096 print_kprobe_event(struct trace_iterator *iter, int flags,
1097 struct trace_event *event)
1099 struct kprobe_trace_entry_head *field;
1100 struct trace_seq *s = &iter->seq;
1101 struct trace_probe *tp;
1105 field = (struct kprobe_trace_entry_head *)iter->ent;
1106 tp = container_of(event, struct trace_probe, call.event);
1108 trace_seq_printf(s, "%s: (", trace_event_name(&tp->call));
1110 if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
1113 trace_seq_putc(s, ')');
1115 data = (u8 *)&field[1];
1116 for (i = 0; i < tp->nr_args; i++)
1117 if (!tp->args[i].type->print(s, tp->args[i].name,
1118 data + tp->args[i].offset, field))
1121 trace_seq_putc(s, '\n');
1123 return trace_handle_return(s);
1126 static enum print_line_t
1127 print_kretprobe_event(struct trace_iterator *iter, int flags,
1128 struct trace_event *event)
1130 struct kretprobe_trace_entry_head *field;
1131 struct trace_seq *s = &iter->seq;
1132 struct trace_probe *tp;
1136 field = (struct kretprobe_trace_entry_head *)iter->ent;
1137 tp = container_of(event, struct trace_probe, call.event);
1139 trace_seq_printf(s, "%s: (", trace_event_name(&tp->call));
1141 if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1144 trace_seq_puts(s, " <- ");
1146 if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1149 trace_seq_putc(s, ')');
1151 data = (u8 *)&field[1];
1152 for (i = 0; i < tp->nr_args; i++)
1153 if (!tp->args[i].type->print(s, tp->args[i].name,
1154 data + tp->args[i].offset, field))
1157 trace_seq_putc(s, '\n');
1160 return trace_handle_return(s);
1164 static int kprobe_event_define_fields(struct trace_event_call *event_call)
1167 struct kprobe_trace_entry_head field;
1168 struct trace_kprobe *tk = (struct trace_kprobe *)event_call->data;
1170 DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1171 /* Set argument names as fields */
1172 for (i = 0; i < tk->tp.nr_args; i++) {
1173 struct probe_arg *parg = &tk->tp.args[i];
1175 ret = trace_define_field(event_call, parg->type->fmttype,
1177 sizeof(field) + parg->offset,
1179 parg->type->is_signed,
1187 static int kretprobe_event_define_fields(struct trace_event_call *event_call)
1190 struct kretprobe_trace_entry_head field;
1191 struct trace_kprobe *tk = (struct trace_kprobe *)event_call->data;
1193 DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1194 DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1195 /* Set argument names as fields */
1196 for (i = 0; i < tk->tp.nr_args; i++) {
1197 struct probe_arg *parg = &tk->tp.args[i];
1199 ret = trace_define_field(event_call, parg->type->fmttype,
1201 sizeof(field) + parg->offset,
1203 parg->type->is_signed,
1211 #ifdef CONFIG_PERF_EVENTS
1213 /* Kprobe profile handler */
1215 kprobe_perf_func(struct trace_kprobe *tk, struct pt_regs *regs)
1217 struct trace_event_call *call = &tk->tp.call;
1218 struct kprobe_trace_entry_head *entry;
1219 struct hlist_head *head;
1220 int size, __size, dsize;
1223 if (bpf_prog_array_valid(call)) {
1224 unsigned long orig_ip = instruction_pointer(regs);
1227 ret = trace_call_bpf(call, regs);
1230 * We need to check and see if we modified the pc of the
1231 * pt_regs, and if so return 1 so that we don't do the
1234 if (orig_ip != instruction_pointer(regs))
1240 head = this_cpu_ptr(call->perf_events);
1241 if (hlist_empty(head))
1244 dsize = __get_data_size(&tk->tp, regs);
1245 __size = sizeof(*entry) + tk->tp.size + dsize;
1246 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1247 size -= sizeof(u32);
1249 entry = perf_trace_buf_alloc(size, NULL, &rctx);
1253 entry->ip = (unsigned long)tk->rp.kp.addr;
1254 memset(&entry[1], 0, dsize);
1255 store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1256 perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1260 NOKPROBE_SYMBOL(kprobe_perf_func);
1262 /* Kretprobe profile handler */
1264 kretprobe_perf_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1265 struct pt_regs *regs)
1267 struct trace_event_call *call = &tk->tp.call;
1268 struct kretprobe_trace_entry_head *entry;
1269 struct hlist_head *head;
1270 int size, __size, dsize;
1273 if (bpf_prog_array_valid(call) && !trace_call_bpf(call, regs))
1276 head = this_cpu_ptr(call->perf_events);
1277 if (hlist_empty(head))
1280 dsize = __get_data_size(&tk->tp, regs);
1281 __size = sizeof(*entry) + tk->tp.size + dsize;
1282 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1283 size -= sizeof(u32);
1285 entry = perf_trace_buf_alloc(size, NULL, &rctx);
1289 entry->func = (unsigned long)tk->rp.kp.addr;
1290 entry->ret_ip = (unsigned long)ri->ret_addr;
1291 store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1292 perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1295 NOKPROBE_SYMBOL(kretprobe_perf_func);
1297 int bpf_get_kprobe_info(const struct perf_event *event, u32 *fd_type,
1298 const char **symbol, u64 *probe_offset,
1299 u64 *probe_addr, bool perf_type_tracepoint)
1301 const char *pevent = trace_event_name(event->tp_event);
1302 const char *group = event->tp_event->class->system;
1303 struct trace_kprobe *tk;
1305 if (perf_type_tracepoint)
1306 tk = find_trace_kprobe(pevent, group);
1308 tk = event->tp_event->data;
1312 *fd_type = trace_kprobe_is_return(tk) ? BPF_FD_TYPE_KRETPROBE
1313 : BPF_FD_TYPE_KPROBE;
1315 *symbol = tk->symbol;
1316 *probe_offset = tk->rp.kp.offset;
1321 *probe_addr = (unsigned long)tk->rp.kp.addr;
1325 #endif /* CONFIG_PERF_EVENTS */
1328 * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex.
1330 * kprobe_trace_self_tests_init() does enable_trace_probe/disable_trace_probe
1331 * lockless, but we can't race with this __init function.
1333 static int kprobe_register(struct trace_event_call *event,
1334 enum trace_reg type, void *data)
1336 struct trace_kprobe *tk = (struct trace_kprobe *)event->data;
1337 struct trace_event_file *file = data;
1340 case TRACE_REG_REGISTER:
1341 return enable_trace_kprobe(tk, file);
1342 case TRACE_REG_UNREGISTER:
1343 return disable_trace_kprobe(tk, file);
1345 #ifdef CONFIG_PERF_EVENTS
1346 case TRACE_REG_PERF_REGISTER:
1347 return enable_trace_kprobe(tk, NULL);
1348 case TRACE_REG_PERF_UNREGISTER:
1349 return disable_trace_kprobe(tk, NULL);
1350 case TRACE_REG_PERF_OPEN:
1351 case TRACE_REG_PERF_CLOSE:
1352 case TRACE_REG_PERF_ADD:
1353 case TRACE_REG_PERF_DEL:
1360 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1362 struct trace_kprobe *tk = container_of(kp, struct trace_kprobe, rp.kp);
1365 raw_cpu_inc(*tk->nhit);
1367 if (tk->tp.flags & TP_FLAG_TRACE)
1368 kprobe_trace_func(tk, regs);
1369 #ifdef CONFIG_PERF_EVENTS
1370 if (tk->tp.flags & TP_FLAG_PROFILE)
1371 ret = kprobe_perf_func(tk, regs);
1375 NOKPROBE_SYMBOL(kprobe_dispatcher);
1378 kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1380 struct trace_kprobe *tk = container_of(ri->rp, struct trace_kprobe, rp);
1382 raw_cpu_inc(*tk->nhit);
1384 if (tk->tp.flags & TP_FLAG_TRACE)
1385 kretprobe_trace_func(tk, ri, regs);
1386 #ifdef CONFIG_PERF_EVENTS
1387 if (tk->tp.flags & TP_FLAG_PROFILE)
1388 kretprobe_perf_func(tk, ri, regs);
1390 return 0; /* We don't tweek kernel, so just return 0 */
1392 NOKPROBE_SYMBOL(kretprobe_dispatcher);
1394 static struct trace_event_functions kretprobe_funcs = {
1395 .trace = print_kretprobe_event
1398 static struct trace_event_functions kprobe_funcs = {
1399 .trace = print_kprobe_event
1402 static inline void init_trace_event_call(struct trace_kprobe *tk,
1403 struct trace_event_call *call)
1405 INIT_LIST_HEAD(&call->class->fields);
1406 if (trace_kprobe_is_return(tk)) {
1407 call->event.funcs = &kretprobe_funcs;
1408 call->class->define_fields = kretprobe_event_define_fields;
1410 call->event.funcs = &kprobe_funcs;
1411 call->class->define_fields = kprobe_event_define_fields;
1414 call->flags = TRACE_EVENT_FL_KPROBE;
1415 call->class->reg = kprobe_register;
1419 static int register_kprobe_event(struct trace_kprobe *tk)
1421 struct trace_event_call *call = &tk->tp.call;
1424 init_trace_event_call(tk, call);
1426 if (set_print_fmt(&tk->tp, trace_kprobe_is_return(tk)) < 0)
1428 ret = register_trace_event(&call->event);
1430 kfree(call->print_fmt);
1433 ret = trace_add_event_call(call);
1435 pr_info("Failed to register kprobe event: %s\n",
1436 trace_event_name(call));
1437 kfree(call->print_fmt);
1438 unregister_trace_event(&call->event);
1443 static int unregister_kprobe_event(struct trace_kprobe *tk)
1447 /* tp->event is unregistered in trace_remove_event_call() */
1448 ret = trace_remove_event_call(&tk->tp.call);
1450 kfree(tk->tp.call.print_fmt);
1454 #ifdef CONFIG_PERF_EVENTS
1455 /* create a trace_kprobe, but don't add it to global lists */
1456 struct trace_event_call *
1457 create_local_trace_kprobe(char *func, void *addr, unsigned long offs,
1460 struct trace_kprobe *tk;
1465 * local trace_kprobes are not added to probe_list, so they are never
1466 * searched in find_trace_kprobe(). Therefore, there is no concern of
1467 * duplicated name here.
1469 event = func ? func : "DUMMY_EVENT";
1471 tk = alloc_trace_kprobe(KPROBE_EVENT_SYSTEM, event, (void *)addr, func,
1472 offs, 0 /* maxactive */, 0 /* nargs */,
1476 pr_info("Failed to allocate trace_probe.(%d)\n",
1478 return ERR_CAST(tk);
1481 init_trace_event_call(tk, &tk->tp.call);
1483 if (set_print_fmt(&tk->tp, trace_kprobe_is_return(tk)) < 0) {
1488 ret = __register_trace_kprobe(tk);
1490 kfree(tk->tp.call.print_fmt);
1494 return &tk->tp.call;
1496 free_trace_kprobe(tk);
1497 return ERR_PTR(ret);
1500 void destroy_local_trace_kprobe(struct trace_event_call *event_call)
1502 struct trace_kprobe *tk;
1504 tk = container_of(event_call, struct trace_kprobe, tp.call);
1506 if (trace_probe_is_enabled(&tk->tp)) {
1511 __unregister_trace_kprobe(tk);
1513 kfree(tk->tp.call.print_fmt);
1514 free_trace_kprobe(tk);
1516 #endif /* CONFIG_PERF_EVENTS */
1518 /* Make a tracefs interface for controlling probe points */
1519 static __init int init_kprobe_trace(void)
1521 struct dentry *d_tracer;
1522 struct dentry *entry;
1524 if (register_module_notifier(&trace_kprobe_module_nb))
1527 d_tracer = tracing_init_dentry();
1528 if (IS_ERR(d_tracer))
1531 entry = tracefs_create_file("kprobe_events", 0644, d_tracer,
1532 NULL, &kprobe_events_ops);
1534 /* Event list interface */
1536 pr_warn("Could not create tracefs 'kprobe_events' entry\n");
1538 /* Profile interface */
1539 entry = tracefs_create_file("kprobe_profile", 0444, d_tracer,
1540 NULL, &kprobe_profile_ops);
1543 pr_warn("Could not create tracefs 'kprobe_profile' entry\n");
1546 fs_initcall(init_kprobe_trace);
1549 #ifdef CONFIG_FTRACE_STARTUP_TEST
1551 * The "__used" keeps gcc from removing the function symbol
1552 * from the kallsyms table. 'noinline' makes sure that there
1553 * isn't an inlined version used by the test method below
1555 static __used __init noinline int
1556 kprobe_trace_selftest_target(int a1, int a2, int a3, int a4, int a5, int a6)
1558 return a1 + a2 + a3 + a4 + a5 + a6;
1561 static __init struct trace_event_file *
1562 find_trace_probe_file(struct trace_kprobe *tk, struct trace_array *tr)
1564 struct trace_event_file *file;
1566 list_for_each_entry(file, &tr->events, list)
1567 if (file->event_call == &tk->tp.call)
1574 * Nobody but us can call enable_trace_kprobe/disable_trace_kprobe at this
1575 * stage, we can do this lockless.
1577 static __init int kprobe_trace_self_tests_init(void)
1580 int (*target)(int, int, int, int, int, int);
1581 struct trace_kprobe *tk;
1582 struct trace_event_file *file;
1584 if (tracing_is_disabled())
1587 target = kprobe_trace_selftest_target;
1589 pr_info("Testing kprobe tracing: ");
1591 ret = trace_run_command("p:testprobe kprobe_trace_selftest_target "
1592 "$stack $stack0 +0($stack)",
1593 create_trace_kprobe);
1594 if (WARN_ON_ONCE(ret)) {
1595 pr_warn("error on probing function entry.\n");
1598 /* Enable trace point */
1599 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1600 if (WARN_ON_ONCE(tk == NULL)) {
1601 pr_warn("error on getting new probe.\n");
1604 file = find_trace_probe_file(tk, top_trace_array());
1605 if (WARN_ON_ONCE(file == NULL)) {
1606 pr_warn("error on getting probe file.\n");
1609 enable_trace_kprobe(tk, file);
1613 ret = trace_run_command("r:testprobe2 kprobe_trace_selftest_target "
1614 "$retval", create_trace_kprobe);
1615 if (WARN_ON_ONCE(ret)) {
1616 pr_warn("error on probing function return.\n");
1619 /* Enable trace point */
1620 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1621 if (WARN_ON_ONCE(tk == NULL)) {
1622 pr_warn("error on getting 2nd new probe.\n");
1625 file = find_trace_probe_file(tk, top_trace_array());
1626 if (WARN_ON_ONCE(file == NULL)) {
1627 pr_warn("error on getting probe file.\n");
1630 enable_trace_kprobe(tk, file);
1637 ret = target(1, 2, 3, 4, 5, 6);
1640 * Not expecting an error here, the check is only to prevent the
1641 * optimizer from removing the call to target() as otherwise there
1642 * are no side-effects and the call is never performed.
1647 /* Disable trace points before removing it */
1648 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1649 if (WARN_ON_ONCE(tk == NULL)) {
1650 pr_warn("error on getting test probe.\n");
1653 if (trace_kprobe_nhit(tk) != 1) {
1654 pr_warn("incorrect number of testprobe hits\n");
1658 file = find_trace_probe_file(tk, top_trace_array());
1659 if (WARN_ON_ONCE(file == NULL)) {
1660 pr_warn("error on getting probe file.\n");
1663 disable_trace_kprobe(tk, file);
1666 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1667 if (WARN_ON_ONCE(tk == NULL)) {
1668 pr_warn("error on getting 2nd test probe.\n");
1671 if (trace_kprobe_nhit(tk) != 1) {
1672 pr_warn("incorrect number of testprobe2 hits\n");
1676 file = find_trace_probe_file(tk, top_trace_array());
1677 if (WARN_ON_ONCE(file == NULL)) {
1678 pr_warn("error on getting probe file.\n");
1681 disable_trace_kprobe(tk, file);
1684 ret = trace_run_command("-:testprobe", create_trace_kprobe);
1685 if (WARN_ON_ONCE(ret)) {
1686 pr_warn("error on deleting a probe.\n");
1690 ret = trace_run_command("-:testprobe2", create_trace_kprobe);
1691 if (WARN_ON_ONCE(ret)) {
1692 pr_warn("error on deleting a probe.\n");
1697 release_all_trace_kprobes();
1699 * Wait for the optimizer work to finish. Otherwise it might fiddle
1700 * with probes in already freed __init text.
1702 wait_for_kprobe_optimizer();
1704 pr_cont("NG: Some tests are failed. Please check them.\n");
1710 late_initcall(kprobe_trace_self_tests_init);