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>
25 #include "trace_probe.h"
27 #define KPROBE_EVENT_SYSTEM "kprobes"
30 * Kprobe event core functions
33 struct list_head list;
34 struct kretprobe rp; /* Use rp.kp for kprobe use */
35 unsigned long __percpu *nhit;
36 const char *symbol; /* symbol name */
37 struct trace_probe tp;
40 #define SIZEOF_TRACE_KPROBE(n) \
41 (offsetof(struct trace_kprobe, tp.args) + \
42 (sizeof(struct probe_arg) * (n)))
45 static nokprobe_inline bool trace_kprobe_is_return(struct trace_kprobe *tk)
47 return tk->rp.handler != NULL;
50 static nokprobe_inline const char *trace_kprobe_symbol(struct trace_kprobe *tk)
52 return tk->symbol ? tk->symbol : "unknown";
55 static nokprobe_inline unsigned long trace_kprobe_offset(struct trace_kprobe *tk)
57 return tk->rp.kp.offset;
60 static nokprobe_inline bool trace_kprobe_has_gone(struct trace_kprobe *tk)
62 return !!(kprobe_gone(&tk->rp.kp));
65 static nokprobe_inline bool trace_kprobe_within_module(struct trace_kprobe *tk,
68 int len = strlen(mod->name);
69 const char *name = trace_kprobe_symbol(tk);
70 return strncmp(mod->name, name, len) == 0 && name[len] == ':';
73 static nokprobe_inline bool trace_kprobe_is_on_module(struct trace_kprobe *tk)
75 return !!strchr(trace_kprobe_symbol(tk), ':');
78 static nokprobe_inline unsigned long trace_kprobe_nhit(struct trace_kprobe *tk)
80 unsigned long nhit = 0;
83 for_each_possible_cpu(cpu)
84 nhit += *per_cpu_ptr(tk->nhit, cpu);
89 static int register_kprobe_event(struct trace_kprobe *tk);
90 static int unregister_kprobe_event(struct trace_kprobe *tk);
92 static DEFINE_MUTEX(probe_lock);
93 static LIST_HEAD(probe_list);
95 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
96 static int kretprobe_dispatcher(struct kretprobe_instance *ri,
97 struct pt_regs *regs);
99 /* Memory fetching by symbol */
100 struct symbol_cache {
106 unsigned long update_symbol_cache(struct symbol_cache *sc)
108 sc->addr = (unsigned long)kallsyms_lookup_name(sc->symbol);
111 sc->addr += sc->offset;
116 void free_symbol_cache(struct symbol_cache *sc)
122 struct symbol_cache *alloc_symbol_cache(const char *sym, long offset)
124 struct symbol_cache *sc;
126 if (!sym || strlen(sym) == 0)
129 sc = kzalloc(sizeof(struct symbol_cache), GFP_KERNEL);
133 sc->symbol = kstrdup(sym, GFP_KERNEL);
139 update_symbol_cache(sc);
145 * Kprobes-specific fetch functions
147 #define DEFINE_FETCH_stack(type) \
148 static void FETCH_FUNC_NAME(stack, type)(struct pt_regs *regs, \
149 void *offset, void *dest) \
151 *(type *)dest = (type)regs_get_kernel_stack_nth(regs, \
152 (unsigned int)((unsigned long)offset)); \
154 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(stack, type));
156 DEFINE_BASIC_FETCH_FUNCS(stack)
157 /* No string on the stack entry */
158 #define fetch_stack_string NULL
159 #define fetch_stack_string_size NULL
161 #define DEFINE_FETCH_memory(type) \
162 static void FETCH_FUNC_NAME(memory, type)(struct pt_regs *regs, \
163 void *addr, void *dest) \
166 if (probe_kernel_address(addr, retval)) \
169 *(type *)dest = retval; \
171 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(memory, type));
173 DEFINE_BASIC_FETCH_FUNCS(memory)
175 * Fetch a null-terminated string. Caller MUST set *(u32 *)dest with max
176 * length and relative data location.
178 static void FETCH_FUNC_NAME(memory, string)(struct pt_regs *regs,
179 void *addr, void *dest)
181 int maxlen = get_rloc_len(*(u32 *)dest);
182 u8 *dst = get_rloc_data(dest);
189 * Try to get string again, since the string can be changed while
192 ret = strncpy_from_unsafe(dst, addr, maxlen);
194 if (ret < 0) { /* Failed to fetch string */
196 *(u32 *)dest = make_data_rloc(0, get_rloc_offs(*(u32 *)dest));
198 *(u32 *)dest = make_data_rloc(ret, get_rloc_offs(*(u32 *)dest));
201 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(memory, string));
203 /* Return the length of string -- including null terminal byte */
204 static void FETCH_FUNC_NAME(memory, string_size)(struct pt_regs *regs,
205 void *addr, void *dest)
216 ret = __copy_from_user_inatomic(&c, (u8 *)addr + len, 1);
218 } while (c && ret == 0 && len < MAX_STRING_SIZE);
223 if (ret < 0) /* Failed to check the length */
228 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(memory, string_size));
230 #define DEFINE_FETCH_symbol(type) \
231 void FETCH_FUNC_NAME(symbol, type)(struct pt_regs *regs, void *data, void *dest)\
233 struct symbol_cache *sc = data; \
235 fetch_memory_##type(regs, (void *)sc->addr, dest); \
239 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(symbol, type));
241 DEFINE_BASIC_FETCH_FUNCS(symbol)
242 DEFINE_FETCH_symbol(string)
243 DEFINE_FETCH_symbol(string_size)
245 /* kprobes don't support file_offset fetch methods */
246 #define fetch_file_offset_u8 NULL
247 #define fetch_file_offset_u16 NULL
248 #define fetch_file_offset_u32 NULL
249 #define fetch_file_offset_u64 NULL
250 #define fetch_file_offset_string NULL
251 #define fetch_file_offset_string_size NULL
253 /* Fetch type information table */
254 static const struct fetch_type kprobes_fetch_type_table[] = {
256 [FETCH_TYPE_STRING] = __ASSIGN_FETCH_TYPE("string", string, string,
257 sizeof(u32), 1, "__data_loc char[]"),
258 [FETCH_TYPE_STRSIZE] = __ASSIGN_FETCH_TYPE("string_size", u32,
259 string_size, sizeof(u32), 0, "u32"),
261 ASSIGN_FETCH_TYPE(u8, u8, 0),
262 ASSIGN_FETCH_TYPE(u16, u16, 0),
263 ASSIGN_FETCH_TYPE(u32, u32, 0),
264 ASSIGN_FETCH_TYPE(u64, u64, 0),
265 ASSIGN_FETCH_TYPE(s8, u8, 1),
266 ASSIGN_FETCH_TYPE(s16, u16, 1),
267 ASSIGN_FETCH_TYPE(s32, u32, 1),
268 ASSIGN_FETCH_TYPE(s64, u64, 1),
269 ASSIGN_FETCH_TYPE_ALIAS(x8, u8, u8, 0),
270 ASSIGN_FETCH_TYPE_ALIAS(x16, u16, u16, 0),
271 ASSIGN_FETCH_TYPE_ALIAS(x32, u32, u32, 0),
272 ASSIGN_FETCH_TYPE_ALIAS(x64, u64, u64, 0),
274 ASSIGN_FETCH_TYPE_END
278 * Allocate new trace_probe and initialize it (including kprobes).
280 static struct trace_kprobe *alloc_trace_kprobe(const char *group,
285 int nargs, bool is_return)
287 struct trace_kprobe *tk;
290 tk = kzalloc(SIZEOF_TRACE_KPROBE(nargs), GFP_KERNEL);
294 tk->nhit = alloc_percpu(unsigned long);
299 tk->symbol = kstrdup(symbol, GFP_KERNEL);
302 tk->rp.kp.symbol_name = tk->symbol;
303 tk->rp.kp.offset = offs;
305 tk->rp.kp.addr = addr;
308 tk->rp.handler = kretprobe_dispatcher;
310 tk->rp.kp.pre_handler = kprobe_dispatcher;
312 if (!event || !is_good_name(event)) {
317 tk->tp.call.class = &tk->tp.class;
318 tk->tp.call.name = kstrdup(event, GFP_KERNEL);
319 if (!tk->tp.call.name)
322 if (!group || !is_good_name(group)) {
327 tk->tp.class.system = kstrdup(group, GFP_KERNEL);
328 if (!tk->tp.class.system)
331 INIT_LIST_HEAD(&tk->list);
332 INIT_LIST_HEAD(&tk->tp.files);
335 kfree(tk->tp.call.name);
337 free_percpu(tk->nhit);
342 static void free_trace_kprobe(struct trace_kprobe *tk)
346 for (i = 0; i < tk->tp.nr_args; i++)
347 traceprobe_free_probe_arg(&tk->tp.args[i]);
349 kfree(tk->tp.call.class->system);
350 kfree(tk->tp.call.name);
352 free_percpu(tk->nhit);
356 static struct trace_kprobe *find_trace_kprobe(const char *event,
359 struct trace_kprobe *tk;
361 list_for_each_entry(tk, &probe_list, list)
362 if (strcmp(trace_event_name(&tk->tp.call), event) == 0 &&
363 strcmp(tk->tp.call.class->system, group) == 0)
370 * if the file is NULL, enable "perf" handler, or enable "trace" handler.
373 enable_trace_kprobe(struct trace_kprobe *tk, struct trace_event_file *file)
378 struct event_file_link *link;
380 link = kmalloc(sizeof(*link), GFP_KERNEL);
387 list_add_tail_rcu(&link->list, &tk->tp.files);
389 tk->tp.flags |= TP_FLAG_TRACE;
391 tk->tp.flags |= TP_FLAG_PROFILE;
393 if (trace_probe_is_registered(&tk->tp) && !trace_kprobe_has_gone(tk)) {
394 if (trace_kprobe_is_return(tk))
395 ret = enable_kretprobe(&tk->rp);
397 ret = enable_kprobe(&tk->rp.kp);
404 * Disable trace_probe
405 * if the file is NULL, disable "perf" handler, or disable "trace" handler.
408 disable_trace_kprobe(struct trace_kprobe *tk, struct trace_event_file *file)
410 struct event_file_link *link = NULL;
415 link = find_event_file_link(&tk->tp, file);
421 list_del_rcu(&link->list);
423 if (!list_empty(&tk->tp.files))
426 tk->tp.flags &= ~TP_FLAG_TRACE;
428 tk->tp.flags &= ~TP_FLAG_PROFILE;
430 if (!trace_probe_is_enabled(&tk->tp) && trace_probe_is_registered(&tk->tp)) {
431 if (trace_kprobe_is_return(tk))
432 disable_kretprobe(&tk->rp);
434 disable_kprobe(&tk->rp.kp);
440 * Synchronize with kprobe_trace_func/kretprobe_trace_func
441 * to ensure disabled (all running handlers are finished).
442 * This is not only for kfree(), but also the caller,
443 * trace_remove_event_call() supposes it for releasing
444 * event_call related objects, which will be accessed in
445 * the kprobe_trace_func/kretprobe_trace_func.
448 kfree(link); /* Ignored if link == NULL */
454 /* Internal register function - just handle k*probes and flags */
455 static int __register_trace_kprobe(struct trace_kprobe *tk)
459 if (trace_probe_is_registered(&tk->tp))
462 for (i = 0; i < tk->tp.nr_args; i++)
463 traceprobe_update_arg(&tk->tp.args[i]);
465 /* Set/clear disabled flag according to tp->flag */
466 if (trace_probe_is_enabled(&tk->tp))
467 tk->rp.kp.flags &= ~KPROBE_FLAG_DISABLED;
469 tk->rp.kp.flags |= KPROBE_FLAG_DISABLED;
471 if (trace_kprobe_is_return(tk))
472 ret = register_kretprobe(&tk->rp);
474 ret = register_kprobe(&tk->rp.kp);
477 tk->tp.flags |= TP_FLAG_REGISTERED;
479 pr_warn("Could not insert probe at %s+%lu: %d\n",
480 trace_kprobe_symbol(tk), trace_kprobe_offset(tk), ret);
481 if (ret == -ENOENT && trace_kprobe_is_on_module(tk)) {
482 pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
484 } else if (ret == -EILSEQ) {
485 pr_warn("Probing address(0x%p) is not an instruction boundary.\n",
494 /* Internal unregister function - just handle k*probes and flags */
495 static void __unregister_trace_kprobe(struct trace_kprobe *tk)
497 if (trace_probe_is_registered(&tk->tp)) {
498 if (trace_kprobe_is_return(tk))
499 unregister_kretprobe(&tk->rp);
501 unregister_kprobe(&tk->rp.kp);
502 tk->tp.flags &= ~TP_FLAG_REGISTERED;
503 /* Cleanup kprobe for reuse */
504 if (tk->rp.kp.symbol_name)
505 tk->rp.kp.addr = NULL;
509 /* Unregister a trace_probe and probe_event: call with locking probe_lock */
510 static int unregister_trace_kprobe(struct trace_kprobe *tk)
512 /* Enabled event can not be unregistered */
513 if (trace_probe_is_enabled(&tk->tp))
516 /* Will fail if probe is being used by ftrace or perf */
517 if (unregister_kprobe_event(tk))
520 __unregister_trace_kprobe(tk);
526 /* Register a trace_probe and probe_event */
527 static int register_trace_kprobe(struct trace_kprobe *tk)
529 struct trace_kprobe *old_tk;
532 mutex_lock(&probe_lock);
534 /* Delete old (same name) event if exist */
535 old_tk = find_trace_kprobe(trace_event_name(&tk->tp.call),
536 tk->tp.call.class->system);
538 ret = unregister_trace_kprobe(old_tk);
541 free_trace_kprobe(old_tk);
544 /* Register new event */
545 ret = register_kprobe_event(tk);
547 pr_warn("Failed to register probe event(%d)\n", ret);
551 /* Register k*probe */
552 ret = __register_trace_kprobe(tk);
554 unregister_kprobe_event(tk);
556 list_add_tail(&tk->list, &probe_list);
559 mutex_unlock(&probe_lock);
563 /* Module notifier call back, checking event on the module */
564 static int trace_kprobe_module_callback(struct notifier_block *nb,
565 unsigned long val, void *data)
567 struct module *mod = data;
568 struct trace_kprobe *tk;
571 if (val != MODULE_STATE_COMING)
574 /* Update probes on coming module */
575 mutex_lock(&probe_lock);
576 list_for_each_entry(tk, &probe_list, list) {
577 if (trace_kprobe_within_module(tk, mod)) {
578 /* Don't need to check busy - this should have gone. */
579 __unregister_trace_kprobe(tk);
580 ret = __register_trace_kprobe(tk);
582 pr_warn("Failed to re-register probe %s on %s: %d\n",
583 trace_event_name(&tk->tp.call),
587 mutex_unlock(&probe_lock);
592 static struct notifier_block trace_kprobe_module_nb = {
593 .notifier_call = trace_kprobe_module_callback,
594 .priority = 1 /* Invoked after kprobe module callback */
597 static int create_trace_kprobe(int argc, char **argv)
601 * - Add kprobe: p[:[GRP/]EVENT] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
602 * - Add kretprobe: r[:[GRP/]EVENT] [MOD:]KSYM[+0] [FETCHARGS]
604 * $retval : fetch return value
605 * $stack : fetch stack address
606 * $stackN : fetch Nth of stack (N:0-)
607 * $comm : fetch current task comm
608 * @ADDR : fetch memory at ADDR (ADDR should be in kernel)
609 * @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
610 * %REG : fetch register REG
611 * Dereferencing memory fetch:
612 * +|-offs(ARG) : fetch memory at ARG +|- offs address.
613 * Alias name of args:
614 * NAME=FETCHARG : set NAME as alias of FETCHARG.
616 * FETCHARG:TYPE : use TYPE instead of unsigned long.
618 struct trace_kprobe *tk;
620 bool is_return = false, is_delete = false;
621 char *symbol = NULL, *event = NULL, *group = NULL;
623 unsigned long offset = 0;
625 char buf[MAX_EVENT_NAME_LEN];
627 /* argc must be >= 1 */
628 if (argv[0][0] == 'p')
630 else if (argv[0][0] == 'r')
632 else if (argv[0][0] == '-')
635 pr_info("Probe definition must be started with 'p', 'r' or"
640 if (argv[0][1] == ':') {
642 if (strchr(event, '/')) {
644 event = strchr(group, '/') + 1;
646 if (strlen(group) == 0) {
647 pr_info("Group name is not specified\n");
651 if (strlen(event) == 0) {
652 pr_info("Event name is not specified\n");
657 group = KPROBE_EVENT_SYSTEM;
661 pr_info("Delete command needs an event name.\n");
664 mutex_lock(&probe_lock);
665 tk = find_trace_kprobe(event, group);
667 mutex_unlock(&probe_lock);
668 pr_info("Event %s/%s doesn't exist.\n", group, event);
671 /* delete an event */
672 ret = unregister_trace_kprobe(tk);
674 free_trace_kprobe(tk);
675 mutex_unlock(&probe_lock);
680 pr_info("Probe point is not specified.\n");
683 if (isdigit(argv[1][0])) {
685 pr_info("Return probe point must be a symbol.\n");
688 /* an address specified */
689 ret = kstrtoul(&argv[1][0], 0, (unsigned long *)&addr);
691 pr_info("Failed to parse address.\n");
695 /* a symbol specified */
697 /* TODO: support .init module functions */
698 ret = traceprobe_split_symbol_offset(symbol, &offset);
700 pr_info("Failed to parse symbol.\n");
703 if (offset && is_return) {
704 pr_info("Return probe must be used without offset.\n");
708 argc -= 2; argv += 2;
712 /* Make a new event name */
714 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
715 is_return ? 'r' : 'p', symbol, offset);
717 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
718 is_return ? 'r' : 'p', addr);
721 tk = alloc_trace_kprobe(group, event, addr, symbol, offset, argc,
724 pr_info("Failed to allocate trace_probe.(%d)\n",
729 /* parse arguments */
731 for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
732 struct probe_arg *parg = &tk->tp.args[i];
734 /* Increment count for freeing args in error case */
737 /* Parse argument name */
738 arg = strchr(argv[i], '=');
741 parg->name = kstrdup(argv[i], GFP_KERNEL);
744 /* If argument name is omitted, set "argN" */
745 snprintf(buf, MAX_EVENT_NAME_LEN, "arg%d", i + 1);
746 parg->name = kstrdup(buf, GFP_KERNEL);
750 pr_info("Failed to allocate argument[%d] name.\n", i);
755 if (!is_good_name(parg->name)) {
756 pr_info("Invalid argument[%d] name: %s\n",
762 if (traceprobe_conflict_field_name(parg->name,
764 pr_info("Argument[%d] name '%s' conflicts with "
765 "another field.\n", i, argv[i]);
770 /* Parse fetch argument */
771 ret = traceprobe_parse_probe_arg(arg, &tk->tp.size, parg,
773 kprobes_fetch_type_table);
775 pr_info("Parse error at argument[%d]. (%d)\n", i, ret);
780 ret = register_trace_kprobe(tk);
786 free_trace_kprobe(tk);
790 static int release_all_trace_kprobes(void)
792 struct trace_kprobe *tk;
795 mutex_lock(&probe_lock);
796 /* Ensure no probe is in use. */
797 list_for_each_entry(tk, &probe_list, list)
798 if (trace_probe_is_enabled(&tk->tp)) {
802 /* TODO: Use batch unregistration */
803 while (!list_empty(&probe_list)) {
804 tk = list_entry(probe_list.next, struct trace_kprobe, list);
805 ret = unregister_trace_kprobe(tk);
808 free_trace_kprobe(tk);
812 mutex_unlock(&probe_lock);
817 /* Probes listing interfaces */
818 static void *probes_seq_start(struct seq_file *m, loff_t *pos)
820 mutex_lock(&probe_lock);
821 return seq_list_start(&probe_list, *pos);
824 static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
826 return seq_list_next(v, &probe_list, pos);
829 static void probes_seq_stop(struct seq_file *m, void *v)
831 mutex_unlock(&probe_lock);
834 static int probes_seq_show(struct seq_file *m, void *v)
836 struct trace_kprobe *tk = v;
839 seq_putc(m, trace_kprobe_is_return(tk) ? 'r' : 'p');
840 seq_printf(m, ":%s/%s", tk->tp.call.class->system,
841 trace_event_name(&tk->tp.call));
844 seq_printf(m, " 0x%p", tk->rp.kp.addr);
845 else if (tk->rp.kp.offset)
846 seq_printf(m, " %s+%u", trace_kprobe_symbol(tk),
849 seq_printf(m, " %s", trace_kprobe_symbol(tk));
851 for (i = 0; i < tk->tp.nr_args; i++)
852 seq_printf(m, " %s=%s", tk->tp.args[i].name, tk->tp.args[i].comm);
858 static const struct seq_operations probes_seq_op = {
859 .start = probes_seq_start,
860 .next = probes_seq_next,
861 .stop = probes_seq_stop,
862 .show = probes_seq_show
865 static int probes_open(struct inode *inode, struct file *file)
869 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
870 ret = release_all_trace_kprobes();
875 return seq_open(file, &probes_seq_op);
878 static ssize_t probes_write(struct file *file, const char __user *buffer,
879 size_t count, loff_t *ppos)
881 return traceprobe_probes_write(file, buffer, count, ppos,
882 create_trace_kprobe);
885 static const struct file_operations kprobe_events_ops = {
886 .owner = THIS_MODULE,
890 .release = seq_release,
891 .write = probes_write,
894 /* Probes profiling interfaces */
895 static int probes_profile_seq_show(struct seq_file *m, void *v)
897 struct trace_kprobe *tk = v;
899 seq_printf(m, " %-44s %15lu %15lu\n",
900 trace_event_name(&tk->tp.call),
901 trace_kprobe_nhit(tk),
907 static const struct seq_operations profile_seq_op = {
908 .start = probes_seq_start,
909 .next = probes_seq_next,
910 .stop = probes_seq_stop,
911 .show = probes_profile_seq_show
914 static int profile_open(struct inode *inode, struct file *file)
916 return seq_open(file, &profile_seq_op);
919 static const struct file_operations kprobe_profile_ops = {
920 .owner = THIS_MODULE,
921 .open = profile_open,
924 .release = seq_release,
928 static nokprobe_inline void
929 __kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs,
930 struct trace_event_file *trace_file)
932 struct kprobe_trace_entry_head *entry;
933 struct ring_buffer_event *event;
934 struct ring_buffer *buffer;
936 unsigned long irq_flags;
937 struct trace_event_call *call = &tk->tp.call;
939 WARN_ON(call != trace_file->event_call);
941 if (trace_trigger_soft_disabled(trace_file))
944 local_save_flags(irq_flags);
945 pc = preempt_count();
947 dsize = __get_data_size(&tk->tp, regs);
948 size = sizeof(*entry) + tk->tp.size + dsize;
950 event = trace_event_buffer_lock_reserve(&buffer, trace_file,
952 size, irq_flags, pc);
956 entry = ring_buffer_event_data(event);
957 entry->ip = (unsigned long)tk->rp.kp.addr;
958 store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
960 event_trigger_unlock_commit_regs(trace_file, buffer, event,
961 entry, irq_flags, pc, regs);
965 kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs)
967 struct event_file_link *link;
969 list_for_each_entry_rcu(link, &tk->tp.files, list)
970 __kprobe_trace_func(tk, regs, link->file);
972 NOKPROBE_SYMBOL(kprobe_trace_func);
974 /* Kretprobe handler */
975 static nokprobe_inline void
976 __kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
977 struct pt_regs *regs,
978 struct trace_event_file *trace_file)
980 struct kretprobe_trace_entry_head *entry;
981 struct ring_buffer_event *event;
982 struct ring_buffer *buffer;
984 unsigned long irq_flags;
985 struct trace_event_call *call = &tk->tp.call;
987 WARN_ON(call != trace_file->event_call);
989 if (trace_trigger_soft_disabled(trace_file))
992 local_save_flags(irq_flags);
993 pc = preempt_count();
995 dsize = __get_data_size(&tk->tp, regs);
996 size = sizeof(*entry) + tk->tp.size + dsize;
998 event = trace_event_buffer_lock_reserve(&buffer, trace_file,
1000 size, irq_flags, pc);
1004 entry = ring_buffer_event_data(event);
1005 entry->func = (unsigned long)tk->rp.kp.addr;
1006 entry->ret_ip = (unsigned long)ri->ret_addr;
1007 store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1009 event_trigger_unlock_commit_regs(trace_file, buffer, event,
1010 entry, irq_flags, pc, regs);
1014 kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1015 struct pt_regs *regs)
1017 struct event_file_link *link;
1019 list_for_each_entry_rcu(link, &tk->tp.files, list)
1020 __kretprobe_trace_func(tk, ri, regs, link->file);
1022 NOKPROBE_SYMBOL(kretprobe_trace_func);
1024 /* Event entry printers */
1025 static enum print_line_t
1026 print_kprobe_event(struct trace_iterator *iter, int flags,
1027 struct trace_event *event)
1029 struct kprobe_trace_entry_head *field;
1030 struct trace_seq *s = &iter->seq;
1031 struct trace_probe *tp;
1035 field = (struct kprobe_trace_entry_head *)iter->ent;
1036 tp = container_of(event, struct trace_probe, call.event);
1038 trace_seq_printf(s, "%s: (", trace_event_name(&tp->call));
1040 if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
1043 trace_seq_putc(s, ')');
1045 data = (u8 *)&field[1];
1046 for (i = 0; i < tp->nr_args; i++)
1047 if (!tp->args[i].type->print(s, tp->args[i].name,
1048 data + tp->args[i].offset, field))
1051 trace_seq_putc(s, '\n');
1053 return trace_handle_return(s);
1056 static enum print_line_t
1057 print_kretprobe_event(struct trace_iterator *iter, int flags,
1058 struct trace_event *event)
1060 struct kretprobe_trace_entry_head *field;
1061 struct trace_seq *s = &iter->seq;
1062 struct trace_probe *tp;
1066 field = (struct kretprobe_trace_entry_head *)iter->ent;
1067 tp = container_of(event, struct trace_probe, call.event);
1069 trace_seq_printf(s, "%s: (", trace_event_name(&tp->call));
1071 if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1074 trace_seq_puts(s, " <- ");
1076 if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1079 trace_seq_putc(s, ')');
1081 data = (u8 *)&field[1];
1082 for (i = 0; i < tp->nr_args; i++)
1083 if (!tp->args[i].type->print(s, tp->args[i].name,
1084 data + tp->args[i].offset, field))
1087 trace_seq_putc(s, '\n');
1090 return trace_handle_return(s);
1094 static int kprobe_event_define_fields(struct trace_event_call *event_call)
1097 struct kprobe_trace_entry_head field;
1098 struct trace_kprobe *tk = (struct trace_kprobe *)event_call->data;
1100 DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1101 /* Set argument names as fields */
1102 for (i = 0; i < tk->tp.nr_args; i++) {
1103 struct probe_arg *parg = &tk->tp.args[i];
1105 ret = trace_define_field(event_call, parg->type->fmttype,
1107 sizeof(field) + parg->offset,
1109 parg->type->is_signed,
1117 static int kretprobe_event_define_fields(struct trace_event_call *event_call)
1120 struct kretprobe_trace_entry_head field;
1121 struct trace_kprobe *tk = (struct trace_kprobe *)event_call->data;
1123 DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1124 DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1125 /* Set argument names as fields */
1126 for (i = 0; i < tk->tp.nr_args; i++) {
1127 struct probe_arg *parg = &tk->tp.args[i];
1129 ret = trace_define_field(event_call, parg->type->fmttype,
1131 sizeof(field) + parg->offset,
1133 parg->type->is_signed,
1141 #ifdef CONFIG_PERF_EVENTS
1143 /* Kprobe profile handler */
1145 kprobe_perf_func(struct trace_kprobe *tk, struct pt_regs *regs)
1147 struct trace_event_call *call = &tk->tp.call;
1148 struct bpf_prog *prog = call->prog;
1149 struct kprobe_trace_entry_head *entry;
1150 struct hlist_head *head;
1151 int size, __size, dsize;
1154 if (prog && !trace_call_bpf(prog, regs))
1157 head = this_cpu_ptr(call->perf_events);
1158 if (hlist_empty(head))
1161 dsize = __get_data_size(&tk->tp, regs);
1162 __size = sizeof(*entry) + tk->tp.size + dsize;
1163 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1164 size -= sizeof(u32);
1166 entry = perf_trace_buf_alloc(size, NULL, &rctx);
1170 entry->ip = (unsigned long)tk->rp.kp.addr;
1171 memset(&entry[1], 0, dsize);
1172 store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1173 perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1176 NOKPROBE_SYMBOL(kprobe_perf_func);
1178 /* Kretprobe profile handler */
1180 kretprobe_perf_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1181 struct pt_regs *regs)
1183 struct trace_event_call *call = &tk->tp.call;
1184 struct bpf_prog *prog = call->prog;
1185 struct kretprobe_trace_entry_head *entry;
1186 struct hlist_head *head;
1187 int size, __size, dsize;
1190 if (prog && !trace_call_bpf(prog, regs))
1193 head = this_cpu_ptr(call->perf_events);
1194 if (hlist_empty(head))
1197 dsize = __get_data_size(&tk->tp, regs);
1198 __size = sizeof(*entry) + tk->tp.size + dsize;
1199 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1200 size -= sizeof(u32);
1202 entry = perf_trace_buf_alloc(size, NULL, &rctx);
1206 entry->func = (unsigned long)tk->rp.kp.addr;
1207 entry->ret_ip = (unsigned long)ri->ret_addr;
1208 store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1209 perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1212 NOKPROBE_SYMBOL(kretprobe_perf_func);
1213 #endif /* CONFIG_PERF_EVENTS */
1216 * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex.
1218 * kprobe_trace_self_tests_init() does enable_trace_probe/disable_trace_probe
1219 * lockless, but we can't race with this __init function.
1221 static int kprobe_register(struct trace_event_call *event,
1222 enum trace_reg type, void *data)
1224 struct trace_kprobe *tk = (struct trace_kprobe *)event->data;
1225 struct trace_event_file *file = data;
1228 case TRACE_REG_REGISTER:
1229 return enable_trace_kprobe(tk, file);
1230 case TRACE_REG_UNREGISTER:
1231 return disable_trace_kprobe(tk, file);
1233 #ifdef CONFIG_PERF_EVENTS
1234 case TRACE_REG_PERF_REGISTER:
1235 return enable_trace_kprobe(tk, NULL);
1236 case TRACE_REG_PERF_UNREGISTER:
1237 return disable_trace_kprobe(tk, NULL);
1238 case TRACE_REG_PERF_OPEN:
1239 case TRACE_REG_PERF_CLOSE:
1240 case TRACE_REG_PERF_ADD:
1241 case TRACE_REG_PERF_DEL:
1248 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1250 struct trace_kprobe *tk = container_of(kp, struct trace_kprobe, rp.kp);
1252 raw_cpu_inc(*tk->nhit);
1254 if (tk->tp.flags & TP_FLAG_TRACE)
1255 kprobe_trace_func(tk, regs);
1256 #ifdef CONFIG_PERF_EVENTS
1257 if (tk->tp.flags & TP_FLAG_PROFILE)
1258 kprobe_perf_func(tk, regs);
1260 return 0; /* We don't tweek kernel, so just return 0 */
1262 NOKPROBE_SYMBOL(kprobe_dispatcher);
1265 kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1267 struct trace_kprobe *tk = container_of(ri->rp, struct trace_kprobe, rp);
1269 raw_cpu_inc(*tk->nhit);
1271 if (tk->tp.flags & TP_FLAG_TRACE)
1272 kretprobe_trace_func(tk, ri, regs);
1273 #ifdef CONFIG_PERF_EVENTS
1274 if (tk->tp.flags & TP_FLAG_PROFILE)
1275 kretprobe_perf_func(tk, ri, regs);
1277 return 0; /* We don't tweek kernel, so just return 0 */
1279 NOKPROBE_SYMBOL(kretprobe_dispatcher);
1281 static struct trace_event_functions kretprobe_funcs = {
1282 .trace = print_kretprobe_event
1285 static struct trace_event_functions kprobe_funcs = {
1286 .trace = print_kprobe_event
1289 static int register_kprobe_event(struct trace_kprobe *tk)
1291 struct trace_event_call *call = &tk->tp.call;
1294 /* Initialize trace_event_call */
1295 INIT_LIST_HEAD(&call->class->fields);
1296 if (trace_kprobe_is_return(tk)) {
1297 call->event.funcs = &kretprobe_funcs;
1298 call->class->define_fields = kretprobe_event_define_fields;
1300 call->event.funcs = &kprobe_funcs;
1301 call->class->define_fields = kprobe_event_define_fields;
1303 if (set_print_fmt(&tk->tp, trace_kprobe_is_return(tk)) < 0)
1305 ret = register_trace_event(&call->event);
1307 kfree(call->print_fmt);
1310 call->flags = TRACE_EVENT_FL_KPROBE;
1311 call->class->reg = kprobe_register;
1313 ret = trace_add_event_call(call);
1315 pr_info("Failed to register kprobe event: %s\n",
1316 trace_event_name(call));
1317 kfree(call->print_fmt);
1318 unregister_trace_event(&call->event);
1323 static int unregister_kprobe_event(struct trace_kprobe *tk)
1327 /* tp->event is unregistered in trace_remove_event_call() */
1328 ret = trace_remove_event_call(&tk->tp.call);
1330 kfree(tk->tp.call.print_fmt);
1334 /* Make a tracefs interface for controlling probe points */
1335 static __init int init_kprobe_trace(void)
1337 struct dentry *d_tracer;
1338 struct dentry *entry;
1340 if (register_module_notifier(&trace_kprobe_module_nb))
1343 d_tracer = tracing_init_dentry();
1344 if (IS_ERR(d_tracer))
1347 entry = tracefs_create_file("kprobe_events", 0644, d_tracer,
1348 NULL, &kprobe_events_ops);
1350 /* Event list interface */
1352 pr_warn("Could not create tracefs 'kprobe_events' entry\n");
1354 /* Profile interface */
1355 entry = tracefs_create_file("kprobe_profile", 0444, d_tracer,
1356 NULL, &kprobe_profile_ops);
1359 pr_warn("Could not create tracefs 'kprobe_profile' entry\n");
1362 fs_initcall(init_kprobe_trace);
1365 #ifdef CONFIG_FTRACE_STARTUP_TEST
1367 * The "__used" keeps gcc from removing the function symbol
1368 * from the kallsyms table. 'noinline' makes sure that there
1369 * isn't an inlined version used by the test method below
1371 static __used __init noinline int
1372 kprobe_trace_selftest_target(int a1, int a2, int a3, int a4, int a5, int a6)
1374 return a1 + a2 + a3 + a4 + a5 + a6;
1377 static __init struct trace_event_file *
1378 find_trace_probe_file(struct trace_kprobe *tk, struct trace_array *tr)
1380 struct trace_event_file *file;
1382 list_for_each_entry(file, &tr->events, list)
1383 if (file->event_call == &tk->tp.call)
1390 * Nobody but us can call enable_trace_kprobe/disable_trace_kprobe at this
1391 * stage, we can do this lockless.
1393 static __init int kprobe_trace_self_tests_init(void)
1396 int (*target)(int, int, int, int, int, int);
1397 struct trace_kprobe *tk;
1398 struct trace_event_file *file;
1400 if (tracing_is_disabled())
1403 target = kprobe_trace_selftest_target;
1405 pr_info("Testing kprobe tracing: ");
1407 ret = traceprobe_command("p:testprobe kprobe_trace_selftest_target "
1408 "$stack $stack0 +0($stack)",
1409 create_trace_kprobe);
1410 if (WARN_ON_ONCE(ret)) {
1411 pr_warn("error on probing function entry.\n");
1414 /* Enable trace point */
1415 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1416 if (WARN_ON_ONCE(tk == NULL)) {
1417 pr_warn("error on getting new probe.\n");
1420 file = find_trace_probe_file(tk, top_trace_array());
1421 if (WARN_ON_ONCE(file == NULL)) {
1422 pr_warn("error on getting probe file.\n");
1425 enable_trace_kprobe(tk, file);
1429 ret = traceprobe_command("r:testprobe2 kprobe_trace_selftest_target "
1430 "$retval", create_trace_kprobe);
1431 if (WARN_ON_ONCE(ret)) {
1432 pr_warn("error on probing function return.\n");
1435 /* Enable trace point */
1436 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1437 if (WARN_ON_ONCE(tk == NULL)) {
1438 pr_warn("error on getting 2nd new probe.\n");
1441 file = find_trace_probe_file(tk, top_trace_array());
1442 if (WARN_ON_ONCE(file == NULL)) {
1443 pr_warn("error on getting probe file.\n");
1446 enable_trace_kprobe(tk, file);
1453 ret = target(1, 2, 3, 4, 5, 6);
1456 * Not expecting an error here, the check is only to prevent the
1457 * optimizer from removing the call to target() as otherwise there
1458 * are no side-effects and the call is never performed.
1463 /* Disable trace points before removing it */
1464 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1465 if (WARN_ON_ONCE(tk == NULL)) {
1466 pr_warn("error on getting test probe.\n");
1469 if (trace_kprobe_nhit(tk) != 1) {
1470 pr_warn("incorrect number of testprobe hits\n");
1474 file = find_trace_probe_file(tk, top_trace_array());
1475 if (WARN_ON_ONCE(file == NULL)) {
1476 pr_warn("error on getting probe file.\n");
1479 disable_trace_kprobe(tk, file);
1482 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1483 if (WARN_ON_ONCE(tk == NULL)) {
1484 pr_warn("error on getting 2nd test probe.\n");
1487 if (trace_kprobe_nhit(tk) != 1) {
1488 pr_warn("incorrect number of testprobe2 hits\n");
1492 file = find_trace_probe_file(tk, top_trace_array());
1493 if (WARN_ON_ONCE(file == NULL)) {
1494 pr_warn("error on getting probe file.\n");
1497 disable_trace_kprobe(tk, file);
1500 ret = traceprobe_command("-:testprobe", create_trace_kprobe);
1501 if (WARN_ON_ONCE(ret)) {
1502 pr_warn("error on deleting a probe.\n");
1506 ret = traceprobe_command("-:testprobe2", create_trace_kprobe);
1507 if (WARN_ON_ONCE(ret)) {
1508 pr_warn("error on deleting a probe.\n");
1513 release_all_trace_kprobes();
1515 pr_cont("NG: Some tests are failed. Please check them.\n");
1521 late_initcall(kprobe_trace_self_tests_init);