1 // SPDX-License-Identifier: GPL-2.0
3 * Kprobes-based tracing events
5 * Created by Masami Hiramatsu <mhiramat@redhat.com>
8 #define pr_fmt(fmt) "trace_kprobe: " fmt
10 #include <linux/bpf-cgroup.h>
11 #include <linux/security.h>
12 #include <linux/module.h>
13 #include <linux/uaccess.h>
14 #include <linux/rculist.h>
15 #include <linux/error-injection.h>
17 #include <asm/setup.h> /* for COMMAND_LINE_SIZE */
19 #include "trace_dynevent.h"
20 #include "trace_kprobe_selftest.h"
21 #include "trace_probe.h"
22 #include "trace_probe_tmpl.h"
23 #include "trace_probe_kernel.h"
25 #define KPROBE_EVENT_SYSTEM "kprobes"
26 #define KRETPROBE_MAXACTIVE_MAX 4096
28 /* Kprobe early definition from command line */
29 static char kprobe_boot_events_buf[COMMAND_LINE_SIZE] __initdata;
31 static int __init set_kprobe_boot_events(char *str)
33 strlcpy(kprobe_boot_events_buf, str, COMMAND_LINE_SIZE);
34 disable_tracing_selftest("running kprobe events");
38 __setup("kprobe_event=", set_kprobe_boot_events);
40 static int trace_kprobe_create(const char *raw_command);
41 static int trace_kprobe_show(struct seq_file *m, struct dyn_event *ev);
42 static int trace_kprobe_release(struct dyn_event *ev);
43 static bool trace_kprobe_is_busy(struct dyn_event *ev);
44 static bool trace_kprobe_match(const char *system, const char *event,
45 int argc, const char **argv, struct dyn_event *ev);
47 static struct dyn_event_operations trace_kprobe_ops = {
48 .create = trace_kprobe_create,
49 .show = trace_kprobe_show,
50 .is_busy = trace_kprobe_is_busy,
51 .free = trace_kprobe_release,
52 .match = trace_kprobe_match,
56 * Kprobe event core functions
59 struct dyn_event devent;
60 struct kretprobe rp; /* Use rp.kp for kprobe use */
61 unsigned long __percpu *nhit;
62 const char *symbol; /* symbol name */
63 struct trace_probe tp;
66 static bool is_trace_kprobe(struct dyn_event *ev)
68 return ev->ops == &trace_kprobe_ops;
71 static struct trace_kprobe *to_trace_kprobe(struct dyn_event *ev)
73 return container_of(ev, struct trace_kprobe, devent);
77 * for_each_trace_kprobe - iterate over the trace_kprobe list
78 * @pos: the struct trace_kprobe * for each entry
79 * @dpos: the struct dyn_event * to use as a loop cursor
81 #define for_each_trace_kprobe(pos, dpos) \
82 for_each_dyn_event(dpos) \
83 if (is_trace_kprobe(dpos) && (pos = to_trace_kprobe(dpos)))
85 static nokprobe_inline bool trace_kprobe_is_return(struct trace_kprobe *tk)
87 return tk->rp.handler != NULL;
90 static nokprobe_inline const char *trace_kprobe_symbol(struct trace_kprobe *tk)
92 return tk->symbol ? tk->symbol : "unknown";
95 static nokprobe_inline unsigned long trace_kprobe_offset(struct trace_kprobe *tk)
97 return tk->rp.kp.offset;
100 static nokprobe_inline bool trace_kprobe_has_gone(struct trace_kprobe *tk)
102 return kprobe_gone(&tk->rp.kp);
105 static nokprobe_inline bool trace_kprobe_within_module(struct trace_kprobe *tk,
108 int len = strlen(module_name(mod));
109 const char *name = trace_kprobe_symbol(tk);
111 return strncmp(module_name(mod), name, len) == 0 && name[len] == ':';
114 static nokprobe_inline bool trace_kprobe_module_exist(struct trace_kprobe *tk)
121 p = strchr(tk->symbol, ':');
125 rcu_read_lock_sched();
126 ret = !!find_module(tk->symbol);
127 rcu_read_unlock_sched();
133 static bool trace_kprobe_is_busy(struct dyn_event *ev)
135 struct trace_kprobe *tk = to_trace_kprobe(ev);
137 return trace_probe_is_enabled(&tk->tp);
140 static bool trace_kprobe_match_command_head(struct trace_kprobe *tk,
141 int argc, const char **argv)
143 char buf[MAX_ARGSTR_LEN + 1];
149 snprintf(buf, sizeof(buf), "0x%p", tk->rp.kp.addr);
150 else if (tk->rp.kp.offset)
151 snprintf(buf, sizeof(buf), "%s+%u",
152 trace_kprobe_symbol(tk), tk->rp.kp.offset);
154 snprintf(buf, sizeof(buf), "%s", trace_kprobe_symbol(tk));
155 if (strcmp(buf, argv[0]))
159 return trace_probe_match_command_args(&tk->tp, argc, argv);
162 static bool trace_kprobe_match(const char *system, const char *event,
163 int argc, const char **argv, struct dyn_event *ev)
165 struct trace_kprobe *tk = to_trace_kprobe(ev);
167 return (event[0] == '\0' ||
168 strcmp(trace_probe_name(&tk->tp), event) == 0) &&
169 (!system || strcmp(trace_probe_group_name(&tk->tp), system) == 0) &&
170 trace_kprobe_match_command_head(tk, argc, argv);
173 static nokprobe_inline unsigned long trace_kprobe_nhit(struct trace_kprobe *tk)
175 unsigned long nhit = 0;
178 for_each_possible_cpu(cpu)
179 nhit += *per_cpu_ptr(tk->nhit, cpu);
184 static nokprobe_inline bool trace_kprobe_is_registered(struct trace_kprobe *tk)
186 return !(list_empty(&tk->rp.kp.list) &&
187 hlist_unhashed(&tk->rp.kp.hlist));
190 /* Return 0 if it fails to find the symbol address */
191 static nokprobe_inline
192 unsigned long trace_kprobe_address(struct trace_kprobe *tk)
197 addr = (unsigned long)
198 kallsyms_lookup_name(trace_kprobe_symbol(tk));
200 addr += tk->rp.kp.offset;
202 addr = (unsigned long)tk->rp.kp.addr;
207 static nokprobe_inline struct trace_kprobe *
208 trace_kprobe_primary_from_call(struct trace_event_call *call)
210 struct trace_probe *tp;
212 tp = trace_probe_primary_from_call(call);
213 if (WARN_ON_ONCE(!tp))
216 return container_of(tp, struct trace_kprobe, tp);
219 bool trace_kprobe_on_func_entry(struct trace_event_call *call)
221 struct trace_kprobe *tk = trace_kprobe_primary_from_call(call);
223 return tk ? (kprobe_on_func_entry(tk->rp.kp.addr,
224 tk->rp.kp.addr ? NULL : tk->rp.kp.symbol_name,
225 tk->rp.kp.addr ? 0 : tk->rp.kp.offset) == 0) : false;
228 bool trace_kprobe_error_injectable(struct trace_event_call *call)
230 struct trace_kprobe *tk = trace_kprobe_primary_from_call(call);
232 return tk ? within_error_injection_list(trace_kprobe_address(tk)) :
236 static int register_kprobe_event(struct trace_kprobe *tk);
237 static int unregister_kprobe_event(struct trace_kprobe *tk);
239 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
240 static int kretprobe_dispatcher(struct kretprobe_instance *ri,
241 struct pt_regs *regs);
243 static void free_trace_kprobe(struct trace_kprobe *tk)
246 trace_probe_cleanup(&tk->tp);
248 free_percpu(tk->nhit);
254 * Allocate new trace_probe and initialize it (including kprobes).
256 static struct trace_kprobe *alloc_trace_kprobe(const char *group,
262 int nargs, bool is_return)
264 struct trace_kprobe *tk;
267 tk = kzalloc(struct_size(tk, tp.args, nargs), GFP_KERNEL);
271 tk->nhit = alloc_percpu(unsigned long);
276 tk->symbol = kstrdup(symbol, GFP_KERNEL);
279 tk->rp.kp.symbol_name = tk->symbol;
280 tk->rp.kp.offset = offs;
282 tk->rp.kp.addr = addr;
285 tk->rp.handler = kretprobe_dispatcher;
287 tk->rp.kp.pre_handler = kprobe_dispatcher;
289 tk->rp.maxactive = maxactive;
290 INIT_HLIST_NODE(&tk->rp.kp.hlist);
291 INIT_LIST_HEAD(&tk->rp.kp.list);
293 ret = trace_probe_init(&tk->tp, event, group, false);
297 dyn_event_init(&tk->devent, &trace_kprobe_ops);
300 free_trace_kprobe(tk);
304 static struct trace_kprobe *find_trace_kprobe(const char *event,
307 struct dyn_event *pos;
308 struct trace_kprobe *tk;
310 for_each_trace_kprobe(tk, pos)
311 if (strcmp(trace_probe_name(&tk->tp), event) == 0 &&
312 strcmp(trace_probe_group_name(&tk->tp), group) == 0)
317 static inline int __enable_trace_kprobe(struct trace_kprobe *tk)
321 if (trace_kprobe_is_registered(tk) && !trace_kprobe_has_gone(tk)) {
322 if (trace_kprobe_is_return(tk))
323 ret = enable_kretprobe(&tk->rp);
325 ret = enable_kprobe(&tk->rp.kp);
331 static void __disable_trace_kprobe(struct trace_probe *tp)
333 struct trace_kprobe *tk;
335 list_for_each_entry(tk, trace_probe_probe_list(tp), tp.list) {
336 if (!trace_kprobe_is_registered(tk))
338 if (trace_kprobe_is_return(tk))
339 disable_kretprobe(&tk->rp);
341 disable_kprobe(&tk->rp.kp);
347 * if the file is NULL, enable "perf" handler, or enable "trace" handler.
349 static int enable_trace_kprobe(struct trace_event_call *call,
350 struct trace_event_file *file)
352 struct trace_probe *tp;
353 struct trace_kprobe *tk;
357 tp = trace_probe_primary_from_call(call);
358 if (WARN_ON_ONCE(!tp))
360 enabled = trace_probe_is_enabled(tp);
362 /* This also changes "enabled" state */
364 ret = trace_probe_add_file(tp, file);
368 trace_probe_set_flag(tp, TP_FLAG_PROFILE);
373 list_for_each_entry(tk, trace_probe_probe_list(tp), tp.list) {
374 if (trace_kprobe_has_gone(tk))
376 ret = __enable_trace_kprobe(tk);
383 /* Failed to enable one of them. Roll back all */
385 __disable_trace_kprobe(tp);
387 trace_probe_remove_file(tp, file);
389 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
396 * Disable trace_probe
397 * if the file is NULL, disable "perf" handler, or disable "trace" handler.
399 static int disable_trace_kprobe(struct trace_event_call *call,
400 struct trace_event_file *file)
402 struct trace_probe *tp;
404 tp = trace_probe_primary_from_call(call);
405 if (WARN_ON_ONCE(!tp))
409 if (!trace_probe_get_file_link(tp, file))
411 if (!trace_probe_has_single_file(tp))
413 trace_probe_clear_flag(tp, TP_FLAG_TRACE);
415 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
417 if (!trace_probe_is_enabled(tp))
418 __disable_trace_kprobe(tp);
423 * Synchronization is done in below function. For perf event,
424 * file == NULL and perf_trace_event_unreg() calls
425 * tracepoint_synchronize_unregister() to ensure synchronize
426 * event. We don't need to care about it.
428 trace_probe_remove_file(tp, file);
433 #if defined(CONFIG_DYNAMIC_FTRACE) && \
434 !defined(CONFIG_KPROBE_EVENTS_ON_NOTRACE)
435 static bool __within_notrace_func(unsigned long addr)
437 unsigned long offset, size;
439 if (!addr || !kallsyms_lookup_size_offset(addr, &size, &offset))
442 /* Get the entry address of the target function */
446 * Since ftrace_location_range() does inclusive range check, we need
447 * to subtract 1 byte from the end address.
449 return !ftrace_location_range(addr, addr + size - 1);
452 static bool within_notrace_func(struct trace_kprobe *tk)
454 unsigned long addr = trace_kprobe_address(tk);
455 char symname[KSYM_NAME_LEN], *p;
457 if (!__within_notrace_func(addr))
460 /* Check if the address is on a suffixed-symbol */
461 if (!lookup_symbol_name(addr, symname)) {
462 p = strchr(symname, '.');
466 addr = (unsigned long)kprobe_lookup_name(symname, 0);
468 return __within_notrace_func(addr);
474 #define within_notrace_func(tk) (false)
477 /* Internal register function - just handle k*probes and flags */
478 static int __register_trace_kprobe(struct trace_kprobe *tk)
482 ret = security_locked_down(LOCKDOWN_KPROBES);
486 if (trace_kprobe_is_registered(tk))
489 if (within_notrace_func(tk)) {
490 pr_warn("Could not probe notrace function %s\n",
491 trace_kprobe_symbol(tk));
495 for (i = 0; i < tk->tp.nr_args; i++) {
496 ret = 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);
515 /* Internal unregister function - just handle k*probes and flags */
516 static void __unregister_trace_kprobe(struct trace_kprobe *tk)
518 if (trace_kprobe_is_registered(tk)) {
519 if (trace_kprobe_is_return(tk))
520 unregister_kretprobe(&tk->rp);
522 unregister_kprobe(&tk->rp.kp);
523 /* Cleanup kprobe for reuse and mark it unregistered */
524 INIT_HLIST_NODE(&tk->rp.kp.hlist);
525 INIT_LIST_HEAD(&tk->rp.kp.list);
526 if (tk->rp.kp.symbol_name)
527 tk->rp.kp.addr = NULL;
531 /* Unregister a trace_probe and probe_event */
532 static int unregister_trace_kprobe(struct trace_kprobe *tk)
534 /* If other probes are on the event, just unregister kprobe */
535 if (trace_probe_has_sibling(&tk->tp))
538 /* Enabled event can not be unregistered */
539 if (trace_probe_is_enabled(&tk->tp))
542 /* If there's a reference to the dynamic event */
543 if (trace_event_dyn_busy(trace_probe_event_call(&tk->tp)))
546 /* Will fail if probe is being used by ftrace or perf */
547 if (unregister_kprobe_event(tk))
551 __unregister_trace_kprobe(tk);
552 dyn_event_remove(&tk->devent);
553 trace_probe_unlink(&tk->tp);
558 static bool trace_kprobe_has_same_kprobe(struct trace_kprobe *orig,
559 struct trace_kprobe *comp)
561 struct trace_probe_event *tpe = orig->tp.event;
564 list_for_each_entry(orig, &tpe->probes, tp.list) {
565 if (strcmp(trace_kprobe_symbol(orig),
566 trace_kprobe_symbol(comp)) ||
567 trace_kprobe_offset(orig) != trace_kprobe_offset(comp))
571 * trace_probe_compare_arg_type() ensured that nr_args and
572 * each argument name and type are same. Let's compare comm.
574 for (i = 0; i < orig->tp.nr_args; i++) {
575 if (strcmp(orig->tp.args[i].comm,
576 comp->tp.args[i].comm))
580 if (i == orig->tp.nr_args)
587 static int append_trace_kprobe(struct trace_kprobe *tk, struct trace_kprobe *to)
591 ret = trace_probe_compare_arg_type(&tk->tp, &to->tp);
593 /* Note that argument starts index = 2 */
594 trace_probe_log_set_index(ret + 1);
595 trace_probe_log_err(0, DIFF_ARG_TYPE);
598 if (trace_kprobe_has_same_kprobe(to, tk)) {
599 trace_probe_log_set_index(0);
600 trace_probe_log_err(0, SAME_PROBE);
604 /* Append to existing event */
605 ret = trace_probe_append(&tk->tp, &to->tp);
609 /* Register k*probe */
610 ret = __register_trace_kprobe(tk);
611 if (ret == -ENOENT && !trace_kprobe_module_exist(tk)) {
612 pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
617 trace_probe_unlink(&tk->tp);
619 dyn_event_add(&tk->devent, trace_probe_event_call(&tk->tp));
624 /* Register a trace_probe and probe_event */
625 static int register_trace_kprobe(struct trace_kprobe *tk)
627 struct trace_kprobe *old_tk;
630 mutex_lock(&event_mutex);
632 old_tk = find_trace_kprobe(trace_probe_name(&tk->tp),
633 trace_probe_group_name(&tk->tp));
635 if (trace_kprobe_is_return(tk) != trace_kprobe_is_return(old_tk)) {
636 trace_probe_log_set_index(0);
637 trace_probe_log_err(0, DIFF_PROBE_TYPE);
640 ret = append_trace_kprobe(tk, old_tk);
645 /* Register new event */
646 ret = register_kprobe_event(tk);
648 if (ret == -EEXIST) {
649 trace_probe_log_set_index(0);
650 trace_probe_log_err(0, EVENT_EXIST);
652 pr_warn("Failed to register probe event(%d)\n", ret);
656 /* Register k*probe */
657 ret = __register_trace_kprobe(tk);
658 if (ret == -ENOENT && !trace_kprobe_module_exist(tk)) {
659 pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
664 unregister_kprobe_event(tk);
666 dyn_event_add(&tk->devent, trace_probe_event_call(&tk->tp));
669 mutex_unlock(&event_mutex);
673 /* Module notifier call back, checking event on the module */
674 static int trace_kprobe_module_callback(struct notifier_block *nb,
675 unsigned long val, void *data)
677 struct module *mod = data;
678 struct dyn_event *pos;
679 struct trace_kprobe *tk;
682 if (val != MODULE_STATE_COMING)
685 /* Update probes on coming module */
686 mutex_lock(&event_mutex);
687 for_each_trace_kprobe(tk, pos) {
688 if (trace_kprobe_within_module(tk, mod)) {
689 /* Don't need to check busy - this should have gone. */
690 __unregister_trace_kprobe(tk);
691 ret = __register_trace_kprobe(tk);
693 pr_warn("Failed to re-register probe %s on %s: %d\n",
694 trace_probe_name(&tk->tp),
695 module_name(mod), ret);
698 mutex_unlock(&event_mutex);
703 static struct notifier_block trace_kprobe_module_nb = {
704 .notifier_call = trace_kprobe_module_callback,
705 .priority = 1 /* Invoked after kprobe module callback */
708 static int count_symbols(void *data, unsigned long unused)
710 unsigned int *count = data;
717 static unsigned int number_of_same_symbols(char *func_name)
722 kallsyms_on_each_match_symbol(count_symbols, func_name, &count);
727 static int __trace_kprobe_create(int argc, const char *argv[])
732 * p[:[GRP/][EVENT]] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
734 * r[MAXACTIVE][:[GRP/][EVENT]] [MOD:]KSYM[+0] [FETCHARGS]
736 * p[:[GRP/][EVENT]] [MOD:]KSYM[+0]%return [FETCHARGS]
739 * $retval : fetch return value
740 * $stack : fetch stack address
741 * $stackN : fetch Nth of stack (N:0-)
742 * $comm : fetch current task comm
743 * @ADDR : fetch memory at ADDR (ADDR should be in kernel)
744 * @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
745 * %REG : fetch register REG
746 * Dereferencing memory fetch:
747 * +|-offs(ARG) : fetch memory at ARG +|- offs address.
748 * Alias name of args:
749 * NAME=FETCHARG : set NAME as alias of FETCHARG.
751 * FETCHARG:TYPE : use TYPE instead of unsigned long.
753 struct trace_kprobe *tk = NULL;
755 bool is_return = false;
756 char *symbol = NULL, *tmp = NULL;
757 const char *event = NULL, *group = KPROBE_EVENT_SYSTEM;
758 enum probe_print_type ptype;
762 char buf[MAX_EVENT_NAME_LEN];
763 char gbuf[MAX_EVENT_NAME_LEN];
764 unsigned int flags = TPARG_FL_KERNEL;
766 switch (argv[0][0]) {
778 trace_probe_log_init("trace_kprobe", argc, argv);
780 event = strchr(&argv[0][1], ':');
784 if (isdigit(argv[0][1])) {
786 trace_probe_log_err(1, MAXACT_NO_KPROBE);
790 len = event - &argv[0][1] - 1;
792 len = strlen(&argv[0][1]);
793 if (len > MAX_EVENT_NAME_LEN - 1) {
794 trace_probe_log_err(1, BAD_MAXACT);
797 memcpy(buf, &argv[0][1], len);
799 ret = kstrtouint(buf, 0, &maxactive);
800 if (ret || !maxactive) {
801 trace_probe_log_err(1, BAD_MAXACT);
804 /* kretprobes instances are iterated over via a list. The
805 * maximum should stay reasonable.
807 if (maxactive > KRETPROBE_MAXACTIVE_MAX) {
808 trace_probe_log_err(1, MAXACT_TOO_BIG);
813 /* try to parse an address. if that fails, try to read the
814 * input as a symbol. */
815 if (kstrtoul(argv[1], 0, (unsigned long *)&addr)) {
816 trace_probe_log_set_index(1);
817 /* Check whether uprobe event specified */
818 if (strchr(argv[1], '/') && strchr(argv[1], ':')) {
822 /* a symbol specified */
823 symbol = kstrdup(argv[1], GFP_KERNEL);
827 tmp = strchr(symbol, '%');
829 if (!strcmp(tmp, "%return")) {
833 trace_probe_log_err(tmp - symbol, BAD_ADDR_SUFFIX);
838 /* TODO: support .init module functions */
839 ret = traceprobe_split_symbol_offset(symbol, &offset);
840 if (ret || offset < 0 || offset > UINT_MAX) {
841 trace_probe_log_err(0, BAD_PROBE_ADDR);
845 flags |= TPARG_FL_RETURN;
846 ret = kprobe_on_func_entry(NULL, symbol, offset);
848 flags |= TPARG_FL_FENTRY;
849 /* Defer the ENOENT case until register kprobe */
850 if (ret == -EINVAL && is_return) {
851 trace_probe_log_err(0, BAD_RETPROBE);
856 if (symbol && !strchr(symbol, ':')) {
859 count = number_of_same_symbols(symbol);
862 * Users should use ADDR to remove the ambiguity of
865 trace_probe_log_err(0, NON_UNIQ_SYMBOL);
866 ret = -EADDRNOTAVAIL;
869 } else if (count == 0) {
871 * We can return ENOENT earlier than when register the
874 trace_probe_log_err(0, BAD_PROBE_ADDR);
881 trace_probe_log_set_index(0);
883 ret = traceprobe_parse_event_name(&event, &group, gbuf,
890 /* Make a new event name */
892 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
893 is_return ? 'r' : 'p', symbol, offset);
895 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
896 is_return ? 'r' : 'p', addr);
897 sanitize_event_name(buf);
902 tk = alloc_trace_kprobe(group, event, addr, symbol, offset, maxactive,
903 argc - 2, is_return);
906 /* This must return -ENOMEM, else there is a bug */
907 WARN_ON_ONCE(ret != -ENOMEM);
908 goto out; /* We know tk is not allocated */
910 argc -= 2; argv += 2;
912 /* parse arguments */
913 for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
914 trace_probe_log_set_index(i + 2);
915 ret = traceprobe_parse_probe_arg(&tk->tp, i, argv[i], flags);
917 goto error; /* This can be -ENOMEM */
920 ptype = is_return ? PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL;
921 ret = traceprobe_set_print_fmt(&tk->tp, ptype);
925 ret = register_trace_kprobe(tk);
927 trace_probe_log_set_index(1);
929 trace_probe_log_err(0, BAD_INSN_BNDRY);
930 else if (ret == -ENOENT)
931 trace_probe_log_err(0, BAD_PROBE_ADDR);
932 else if (ret != -ENOMEM && ret != -EEXIST)
933 trace_probe_log_err(0, FAIL_REG_PROBE);
938 trace_probe_log_clear();
945 free_trace_kprobe(tk);
949 static int trace_kprobe_create(const char *raw_command)
951 return trace_probe_create(raw_command, __trace_kprobe_create);
954 static int create_or_delete_trace_kprobe(const char *raw_command)
958 if (raw_command[0] == '-')
959 return dyn_event_release(raw_command, &trace_kprobe_ops);
961 ret = trace_kprobe_create(raw_command);
962 return ret == -ECANCELED ? -EINVAL : ret;
965 static int trace_kprobe_run_command(struct dynevent_cmd *cmd)
967 return create_or_delete_trace_kprobe(cmd->seq.buffer);
971 * kprobe_event_cmd_init - Initialize a kprobe event command object
972 * @cmd: A pointer to the dynevent_cmd struct representing the new event
973 * @buf: A pointer to the buffer used to build the command
974 * @maxlen: The length of the buffer passed in @buf
976 * Initialize a synthetic event command object. Use this before
977 * calling any of the other kprobe_event functions.
979 void kprobe_event_cmd_init(struct dynevent_cmd *cmd, char *buf, int maxlen)
981 dynevent_cmd_init(cmd, buf, maxlen, DYNEVENT_TYPE_KPROBE,
982 trace_kprobe_run_command);
984 EXPORT_SYMBOL_GPL(kprobe_event_cmd_init);
987 * __kprobe_event_gen_cmd_start - Generate a kprobe event command from arg list
988 * @cmd: A pointer to the dynevent_cmd struct representing the new event
989 * @name: The name of the kprobe event
990 * @loc: The location of the kprobe event
991 * @kretprobe: Is this a return probe?
992 * @...: Variable number of arg (pairs), one pair for each field
994 * NOTE: Users normally won't want to call this function directly, but
995 * rather use the kprobe_event_gen_cmd_start() wrapper, which automatically
996 * adds a NULL to the end of the arg list. If this function is used
997 * directly, make sure the last arg in the variable arg list is NULL.
999 * Generate a kprobe event command to be executed by
1000 * kprobe_event_gen_cmd_end(). This function can be used to generate the
1001 * complete command or only the first part of it; in the latter case,
1002 * kprobe_event_add_fields() can be used to add more fields following this.
1004 * Unlikely the synth_event_gen_cmd_start(), @loc must be specified. This
1005 * returns -EINVAL if @loc == NULL.
1007 * Return: 0 if successful, error otherwise.
1009 int __kprobe_event_gen_cmd_start(struct dynevent_cmd *cmd, bool kretprobe,
1010 const char *name, const char *loc, ...)
1012 char buf[MAX_EVENT_NAME_LEN];
1013 struct dynevent_arg arg;
1017 if (cmd->type != DYNEVENT_TYPE_KPROBE)
1024 snprintf(buf, MAX_EVENT_NAME_LEN, "r:kprobes/%s", name);
1026 snprintf(buf, MAX_EVENT_NAME_LEN, "p:kprobes/%s", name);
1028 ret = dynevent_str_add(cmd, buf);
1032 dynevent_arg_init(&arg, 0);
1034 ret = dynevent_arg_add(cmd, &arg, NULL);
1038 va_start(args, loc);
1042 field = va_arg(args, const char *);
1046 if (++cmd->n_fields > MAX_TRACE_ARGS) {
1052 ret = dynevent_arg_add(cmd, &arg, NULL);
1060 EXPORT_SYMBOL_GPL(__kprobe_event_gen_cmd_start);
1063 * __kprobe_event_add_fields - Add probe fields to a kprobe command from arg list
1064 * @cmd: A pointer to the dynevent_cmd struct representing the new event
1065 * @...: Variable number of arg (pairs), one pair for each field
1067 * NOTE: Users normally won't want to call this function directly, but
1068 * rather use the kprobe_event_add_fields() wrapper, which
1069 * automatically adds a NULL to the end of the arg list. If this
1070 * function is used directly, make sure the last arg in the variable
1073 * Add probe fields to an existing kprobe command using a variable
1074 * list of args. Fields are added in the same order they're listed.
1076 * Return: 0 if successful, error otherwise.
1078 int __kprobe_event_add_fields(struct dynevent_cmd *cmd, ...)
1080 struct dynevent_arg arg;
1084 if (cmd->type != DYNEVENT_TYPE_KPROBE)
1087 dynevent_arg_init(&arg, 0);
1089 va_start(args, cmd);
1093 field = va_arg(args, const char *);
1097 if (++cmd->n_fields > MAX_TRACE_ARGS) {
1103 ret = dynevent_arg_add(cmd, &arg, NULL);
1111 EXPORT_SYMBOL_GPL(__kprobe_event_add_fields);
1114 * kprobe_event_delete - Delete a kprobe event
1115 * @name: The name of the kprobe event to delete
1117 * Delete a kprobe event with the give @name from kernel code rather
1118 * than directly from the command line.
1120 * Return: 0 if successful, error otherwise.
1122 int kprobe_event_delete(const char *name)
1124 char buf[MAX_EVENT_NAME_LEN];
1126 snprintf(buf, MAX_EVENT_NAME_LEN, "-:%s", name);
1128 return create_or_delete_trace_kprobe(buf);
1130 EXPORT_SYMBOL_GPL(kprobe_event_delete);
1132 static int trace_kprobe_release(struct dyn_event *ev)
1134 struct trace_kprobe *tk = to_trace_kprobe(ev);
1135 int ret = unregister_trace_kprobe(tk);
1138 free_trace_kprobe(tk);
1142 static int trace_kprobe_show(struct seq_file *m, struct dyn_event *ev)
1144 struct trace_kprobe *tk = to_trace_kprobe(ev);
1147 seq_putc(m, trace_kprobe_is_return(tk) ? 'r' : 'p');
1148 if (trace_kprobe_is_return(tk) && tk->rp.maxactive)
1149 seq_printf(m, "%d", tk->rp.maxactive);
1150 seq_printf(m, ":%s/%s", trace_probe_group_name(&tk->tp),
1151 trace_probe_name(&tk->tp));
1154 seq_printf(m, " 0x%p", tk->rp.kp.addr);
1155 else if (tk->rp.kp.offset)
1156 seq_printf(m, " %s+%u", trace_kprobe_symbol(tk),
1159 seq_printf(m, " %s", trace_kprobe_symbol(tk));
1161 for (i = 0; i < tk->tp.nr_args; i++)
1162 seq_printf(m, " %s=%s", tk->tp.args[i].name, tk->tp.args[i].comm);
1168 static int probes_seq_show(struct seq_file *m, void *v)
1170 struct dyn_event *ev = v;
1172 if (!is_trace_kprobe(ev))
1175 return trace_kprobe_show(m, ev);
1178 static const struct seq_operations probes_seq_op = {
1179 .start = dyn_event_seq_start,
1180 .next = dyn_event_seq_next,
1181 .stop = dyn_event_seq_stop,
1182 .show = probes_seq_show
1185 static int probes_open(struct inode *inode, struct file *file)
1189 ret = security_locked_down(LOCKDOWN_TRACEFS);
1193 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
1194 ret = dyn_events_release_all(&trace_kprobe_ops);
1199 return seq_open(file, &probes_seq_op);
1202 static ssize_t probes_write(struct file *file, const char __user *buffer,
1203 size_t count, loff_t *ppos)
1205 return trace_parse_run_command(file, buffer, count, ppos,
1206 create_or_delete_trace_kprobe);
1209 static const struct file_operations kprobe_events_ops = {
1210 .owner = THIS_MODULE,
1211 .open = probes_open,
1213 .llseek = seq_lseek,
1214 .release = seq_release,
1215 .write = probes_write,
1218 /* Probes profiling interfaces */
1219 static int probes_profile_seq_show(struct seq_file *m, void *v)
1221 struct dyn_event *ev = v;
1222 struct trace_kprobe *tk;
1223 unsigned long nmissed;
1225 if (!is_trace_kprobe(ev))
1228 tk = to_trace_kprobe(ev);
1229 nmissed = trace_kprobe_is_return(tk) ?
1230 tk->rp.kp.nmissed + tk->rp.nmissed : tk->rp.kp.nmissed;
1231 seq_printf(m, " %-44s %15lu %15lu\n",
1232 trace_probe_name(&tk->tp),
1233 trace_kprobe_nhit(tk),
1239 static const struct seq_operations profile_seq_op = {
1240 .start = dyn_event_seq_start,
1241 .next = dyn_event_seq_next,
1242 .stop = dyn_event_seq_stop,
1243 .show = probes_profile_seq_show
1246 static int profile_open(struct inode *inode, struct file *file)
1250 ret = security_locked_down(LOCKDOWN_TRACEFS);
1254 return seq_open(file, &profile_seq_op);
1257 static const struct file_operations kprobe_profile_ops = {
1258 .owner = THIS_MODULE,
1259 .open = profile_open,
1261 .llseek = seq_lseek,
1262 .release = seq_release,
1265 /* Kprobe specific fetch functions */
1267 /* Return the length of string -- including null terminal byte */
1268 static nokprobe_inline int
1269 fetch_store_strlen_user(unsigned long addr)
1271 return kern_fetch_store_strlen_user(addr);
1274 /* Return the length of string -- including null terminal byte */
1275 static nokprobe_inline int
1276 fetch_store_strlen(unsigned long addr)
1278 return kern_fetch_store_strlen(addr);
1282 * Fetch a null-terminated string from user. Caller MUST set *(u32 *)buf
1283 * with max length and relative data location.
1285 static nokprobe_inline int
1286 fetch_store_string_user(unsigned long addr, void *dest, void *base)
1288 return kern_fetch_store_string_user(addr, dest, base);
1292 * Fetch a null-terminated string. Caller MUST set *(u32 *)buf with max
1293 * length and relative data location.
1295 static nokprobe_inline int
1296 fetch_store_string(unsigned long addr, void *dest, void *base)
1298 return kern_fetch_store_string(addr, dest, base);
1301 static nokprobe_inline int
1302 probe_mem_read_user(void *dest, void *src, size_t size)
1304 const void __user *uaddr = (__force const void __user *)src;
1306 return copy_from_user_nofault(dest, uaddr, size);
1309 static nokprobe_inline int
1310 probe_mem_read(void *dest, void *src, size_t size)
1312 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
1313 if ((unsigned long)src < TASK_SIZE)
1314 return probe_mem_read_user(dest, src, size);
1316 return copy_from_kernel_nofault(dest, src, size);
1319 /* Note that we don't verify it, since the code does not come from user space */
1321 process_fetch_insn(struct fetch_insn *code, void *rec, void *dest,
1324 struct pt_regs *regs = rec;
1328 /* 1st stage: get value from context */
1331 val = regs_get_register(regs, code->param);
1333 case FETCH_OP_STACK:
1334 val = regs_get_kernel_stack_nth(regs, code->param);
1336 case FETCH_OP_STACKP:
1337 val = kernel_stack_pointer(regs);
1339 case FETCH_OP_RETVAL:
1340 val = regs_return_value(regs);
1343 val = code->immediate;
1346 val = (unsigned long)current->comm;
1349 val = (unsigned long)code->data;
1351 #ifdef CONFIG_HAVE_FUNCTION_ARG_ACCESS_API
1353 val = regs_get_kernel_argument(regs, code->param);
1356 case FETCH_NOP_SYMBOL: /* Ignore a place holder */
1364 return process_fetch_insn_bottom(code, val, dest, base);
1366 NOKPROBE_SYMBOL(process_fetch_insn)
1368 /* Kprobe handler */
1369 static nokprobe_inline void
1370 __kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs,
1371 struct trace_event_file *trace_file)
1373 struct kprobe_trace_entry_head *entry;
1374 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1375 struct trace_event_buffer fbuffer;
1378 WARN_ON(call != trace_file->event_call);
1380 if (trace_trigger_soft_disabled(trace_file))
1383 dsize = __get_data_size(&tk->tp, regs);
1385 entry = trace_event_buffer_reserve(&fbuffer, trace_file,
1386 sizeof(*entry) + tk->tp.size + dsize);
1390 fbuffer.regs = regs;
1391 entry = fbuffer.entry = ring_buffer_event_data(fbuffer.event);
1392 entry->ip = (unsigned long)tk->rp.kp.addr;
1393 store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1395 trace_event_buffer_commit(&fbuffer);
1399 kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs)
1401 struct event_file_link *link;
1403 trace_probe_for_each_link_rcu(link, &tk->tp)
1404 __kprobe_trace_func(tk, regs, link->file);
1406 NOKPROBE_SYMBOL(kprobe_trace_func);
1408 /* Kretprobe handler */
1409 static nokprobe_inline void
1410 __kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1411 struct pt_regs *regs,
1412 struct trace_event_file *trace_file)
1414 struct kretprobe_trace_entry_head *entry;
1415 struct trace_event_buffer fbuffer;
1416 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1419 WARN_ON(call != trace_file->event_call);
1421 if (trace_trigger_soft_disabled(trace_file))
1424 dsize = __get_data_size(&tk->tp, regs);
1426 entry = trace_event_buffer_reserve(&fbuffer, trace_file,
1427 sizeof(*entry) + tk->tp.size + dsize);
1431 fbuffer.regs = regs;
1432 entry = fbuffer.entry = ring_buffer_event_data(fbuffer.event);
1433 entry->func = (unsigned long)tk->rp.kp.addr;
1434 entry->ret_ip = get_kretprobe_retaddr(ri);
1435 store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1437 trace_event_buffer_commit(&fbuffer);
1441 kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1442 struct pt_regs *regs)
1444 struct event_file_link *link;
1446 trace_probe_for_each_link_rcu(link, &tk->tp)
1447 __kretprobe_trace_func(tk, ri, regs, link->file);
1449 NOKPROBE_SYMBOL(kretprobe_trace_func);
1451 /* Event entry printers */
1452 static enum print_line_t
1453 print_kprobe_event(struct trace_iterator *iter, int flags,
1454 struct trace_event *event)
1456 struct kprobe_trace_entry_head *field;
1457 struct trace_seq *s = &iter->seq;
1458 struct trace_probe *tp;
1460 field = (struct kprobe_trace_entry_head *)iter->ent;
1461 tp = trace_probe_primary_from_call(
1462 container_of(event, struct trace_event_call, event));
1463 if (WARN_ON_ONCE(!tp))
1466 trace_seq_printf(s, "%s: (", trace_probe_name(tp));
1468 if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
1471 trace_seq_putc(s, ')');
1473 if (print_probe_args(s, tp->args, tp->nr_args,
1474 (u8 *)&field[1], field) < 0)
1477 trace_seq_putc(s, '\n');
1479 return trace_handle_return(s);
1482 static enum print_line_t
1483 print_kretprobe_event(struct trace_iterator *iter, int flags,
1484 struct trace_event *event)
1486 struct kretprobe_trace_entry_head *field;
1487 struct trace_seq *s = &iter->seq;
1488 struct trace_probe *tp;
1490 field = (struct kretprobe_trace_entry_head *)iter->ent;
1491 tp = trace_probe_primary_from_call(
1492 container_of(event, struct trace_event_call, event));
1493 if (WARN_ON_ONCE(!tp))
1496 trace_seq_printf(s, "%s: (", trace_probe_name(tp));
1498 if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1501 trace_seq_puts(s, " <- ");
1503 if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1506 trace_seq_putc(s, ')');
1508 if (print_probe_args(s, tp->args, tp->nr_args,
1509 (u8 *)&field[1], field) < 0)
1512 trace_seq_putc(s, '\n');
1515 return trace_handle_return(s);
1519 static int kprobe_event_define_fields(struct trace_event_call *event_call)
1522 struct kprobe_trace_entry_head field;
1523 struct trace_probe *tp;
1525 tp = trace_probe_primary_from_call(event_call);
1526 if (WARN_ON_ONCE(!tp))
1529 DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1531 return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
1534 static int kretprobe_event_define_fields(struct trace_event_call *event_call)
1537 struct kretprobe_trace_entry_head field;
1538 struct trace_probe *tp;
1540 tp = trace_probe_primary_from_call(event_call);
1541 if (WARN_ON_ONCE(!tp))
1544 DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1545 DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1547 return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
1550 #ifdef CONFIG_PERF_EVENTS
1552 /* Kprobe profile handler */
1554 kprobe_perf_func(struct trace_kprobe *tk, struct pt_regs *regs)
1556 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1557 struct kprobe_trace_entry_head *entry;
1558 struct hlist_head *head;
1559 int size, __size, dsize;
1562 if (bpf_prog_array_valid(call)) {
1563 unsigned long orig_ip = instruction_pointer(regs);
1566 ret = trace_call_bpf(call, regs);
1569 * We need to check and see if we modified the pc of the
1570 * pt_regs, and if so return 1 so that we don't do the
1573 if (orig_ip != instruction_pointer(regs))
1579 head = this_cpu_ptr(call->perf_events);
1580 if (hlist_empty(head))
1583 dsize = __get_data_size(&tk->tp, regs);
1584 __size = sizeof(*entry) + tk->tp.size + dsize;
1585 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1586 size -= sizeof(u32);
1588 entry = perf_trace_buf_alloc(size, NULL, &rctx);
1592 entry->ip = (unsigned long)tk->rp.kp.addr;
1593 memset(&entry[1], 0, dsize);
1594 store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1595 perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1599 NOKPROBE_SYMBOL(kprobe_perf_func);
1601 /* Kretprobe profile handler */
1603 kretprobe_perf_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1604 struct pt_regs *regs)
1606 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1607 struct kretprobe_trace_entry_head *entry;
1608 struct hlist_head *head;
1609 int size, __size, dsize;
1612 if (bpf_prog_array_valid(call) && !trace_call_bpf(call, regs))
1615 head = this_cpu_ptr(call->perf_events);
1616 if (hlist_empty(head))
1619 dsize = __get_data_size(&tk->tp, regs);
1620 __size = sizeof(*entry) + tk->tp.size + dsize;
1621 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1622 size -= sizeof(u32);
1624 entry = perf_trace_buf_alloc(size, NULL, &rctx);
1628 entry->func = (unsigned long)tk->rp.kp.addr;
1629 entry->ret_ip = get_kretprobe_retaddr(ri);
1630 store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1631 perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1634 NOKPROBE_SYMBOL(kretprobe_perf_func);
1636 int bpf_get_kprobe_info(const struct perf_event *event, u32 *fd_type,
1637 const char **symbol, u64 *probe_offset,
1638 u64 *probe_addr, bool perf_type_tracepoint)
1640 const char *pevent = trace_event_name(event->tp_event);
1641 const char *group = event->tp_event->class->system;
1642 struct trace_kprobe *tk;
1644 if (perf_type_tracepoint)
1645 tk = find_trace_kprobe(pevent, group);
1647 tk = trace_kprobe_primary_from_call(event->tp_event);
1651 *fd_type = trace_kprobe_is_return(tk) ? BPF_FD_TYPE_KRETPROBE
1652 : BPF_FD_TYPE_KPROBE;
1654 *symbol = tk->symbol;
1655 *probe_offset = tk->rp.kp.offset;
1660 *probe_addr = (unsigned long)tk->rp.kp.addr;
1664 #endif /* CONFIG_PERF_EVENTS */
1667 * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex.
1669 * kprobe_trace_self_tests_init() does enable_trace_probe/disable_trace_probe
1670 * lockless, but we can't race with this __init function.
1672 static int kprobe_register(struct trace_event_call *event,
1673 enum trace_reg type, void *data)
1675 struct trace_event_file *file = data;
1678 case TRACE_REG_REGISTER:
1679 return enable_trace_kprobe(event, file);
1680 case TRACE_REG_UNREGISTER:
1681 return disable_trace_kprobe(event, file);
1683 #ifdef CONFIG_PERF_EVENTS
1684 case TRACE_REG_PERF_REGISTER:
1685 return enable_trace_kprobe(event, NULL);
1686 case TRACE_REG_PERF_UNREGISTER:
1687 return disable_trace_kprobe(event, NULL);
1688 case TRACE_REG_PERF_OPEN:
1689 case TRACE_REG_PERF_CLOSE:
1690 case TRACE_REG_PERF_ADD:
1691 case TRACE_REG_PERF_DEL:
1698 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1700 struct trace_kprobe *tk = container_of(kp, struct trace_kprobe, rp.kp);
1703 raw_cpu_inc(*tk->nhit);
1705 if (trace_probe_test_flag(&tk->tp, TP_FLAG_TRACE))
1706 kprobe_trace_func(tk, regs);
1707 #ifdef CONFIG_PERF_EVENTS
1708 if (trace_probe_test_flag(&tk->tp, TP_FLAG_PROFILE))
1709 ret = kprobe_perf_func(tk, regs);
1713 NOKPROBE_SYMBOL(kprobe_dispatcher);
1716 kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1718 struct kretprobe *rp = get_kretprobe(ri);
1719 struct trace_kprobe *tk;
1722 * There is a small chance that get_kretprobe(ri) returns NULL when
1723 * the kretprobe is unregister on another CPU between kretprobe's
1724 * trampoline_handler and this function.
1729 tk = container_of(rp, struct trace_kprobe, rp);
1730 raw_cpu_inc(*tk->nhit);
1732 if (trace_probe_test_flag(&tk->tp, TP_FLAG_TRACE))
1733 kretprobe_trace_func(tk, ri, regs);
1734 #ifdef CONFIG_PERF_EVENTS
1735 if (trace_probe_test_flag(&tk->tp, TP_FLAG_PROFILE))
1736 kretprobe_perf_func(tk, ri, regs);
1738 return 0; /* We don't tweak kernel, so just return 0 */
1740 NOKPROBE_SYMBOL(kretprobe_dispatcher);
1742 static struct trace_event_functions kretprobe_funcs = {
1743 .trace = print_kretprobe_event
1746 static struct trace_event_functions kprobe_funcs = {
1747 .trace = print_kprobe_event
1750 static struct trace_event_fields kretprobe_fields_array[] = {
1751 { .type = TRACE_FUNCTION_TYPE,
1752 .define_fields = kretprobe_event_define_fields },
1756 static struct trace_event_fields kprobe_fields_array[] = {
1757 { .type = TRACE_FUNCTION_TYPE,
1758 .define_fields = kprobe_event_define_fields },
1762 static inline void init_trace_event_call(struct trace_kprobe *tk)
1764 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1766 if (trace_kprobe_is_return(tk)) {
1767 call->event.funcs = &kretprobe_funcs;
1768 call->class->fields_array = kretprobe_fields_array;
1770 call->event.funcs = &kprobe_funcs;
1771 call->class->fields_array = kprobe_fields_array;
1774 call->flags = TRACE_EVENT_FL_KPROBE;
1775 call->class->reg = kprobe_register;
1778 static int register_kprobe_event(struct trace_kprobe *tk)
1780 init_trace_event_call(tk);
1782 return trace_probe_register_event_call(&tk->tp);
1785 static int unregister_kprobe_event(struct trace_kprobe *tk)
1787 return trace_probe_unregister_event_call(&tk->tp);
1790 #ifdef CONFIG_PERF_EVENTS
1792 /* create a trace_kprobe, but don't add it to global lists */
1793 struct trace_event_call *
1794 create_local_trace_kprobe(char *func, void *addr, unsigned long offs,
1797 enum probe_print_type ptype;
1798 struct trace_kprobe *tk;
1805 count = number_of_same_symbols(func);
1808 * Users should use addr to remove the ambiguity of
1811 return ERR_PTR(-EADDRNOTAVAIL);
1812 else if (count == 0)
1814 * We can return ENOENT earlier than when register the
1817 return ERR_PTR(-ENOENT);
1821 * local trace_kprobes are not added to dyn_event, so they are never
1822 * searched in find_trace_kprobe(). Therefore, there is no concern of
1823 * duplicated name here.
1825 event = func ? func : "DUMMY_EVENT";
1827 tk = alloc_trace_kprobe(KPROBE_EVENT_SYSTEM, event, (void *)addr, func,
1828 offs, 0 /* maxactive */, 0 /* nargs */,
1832 pr_info("Failed to allocate trace_probe.(%d)\n",
1834 return ERR_CAST(tk);
1837 init_trace_event_call(tk);
1839 ptype = trace_kprobe_is_return(tk) ?
1840 PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL;
1841 if (traceprobe_set_print_fmt(&tk->tp, ptype) < 0) {
1846 ret = __register_trace_kprobe(tk);
1850 return trace_probe_event_call(&tk->tp);
1852 free_trace_kprobe(tk);
1853 return ERR_PTR(ret);
1856 void destroy_local_trace_kprobe(struct trace_event_call *event_call)
1858 struct trace_kprobe *tk;
1860 tk = trace_kprobe_primary_from_call(event_call);
1864 if (trace_probe_is_enabled(&tk->tp)) {
1869 __unregister_trace_kprobe(tk);
1871 free_trace_kprobe(tk);
1873 #endif /* CONFIG_PERF_EVENTS */
1875 static __init void enable_boot_kprobe_events(void)
1877 struct trace_array *tr = top_trace_array();
1878 struct trace_event_file *file;
1879 struct trace_kprobe *tk;
1880 struct dyn_event *pos;
1882 mutex_lock(&event_mutex);
1883 for_each_trace_kprobe(tk, pos) {
1884 list_for_each_entry(file, &tr->events, list)
1885 if (file->event_call == trace_probe_event_call(&tk->tp))
1886 trace_event_enable_disable(file, 1, 0);
1888 mutex_unlock(&event_mutex);
1891 static __init void setup_boot_kprobe_events(void)
1893 char *p, *cmd = kprobe_boot_events_buf;
1896 strreplace(kprobe_boot_events_buf, ',', ' ');
1898 while (cmd && *cmd != '\0') {
1899 p = strchr(cmd, ';');
1903 ret = create_or_delete_trace_kprobe(cmd);
1905 pr_warn("Failed to add event(%d): %s\n", ret, cmd);
1910 enable_boot_kprobe_events();
1914 * Register dynevent at core_initcall. This allows kernel to setup kprobe
1915 * events in postcore_initcall without tracefs.
1917 static __init int init_kprobe_trace_early(void)
1921 ret = dyn_event_register(&trace_kprobe_ops);
1925 if (register_module_notifier(&trace_kprobe_module_nb))
1930 core_initcall(init_kprobe_trace_early);
1932 /* Make a tracefs interface for controlling probe points */
1933 static __init int init_kprobe_trace(void)
1937 ret = tracing_init_dentry();
1941 /* Event list interface */
1942 trace_create_file("kprobe_events", TRACE_MODE_WRITE,
1943 NULL, NULL, &kprobe_events_ops);
1945 /* Profile interface */
1946 trace_create_file("kprobe_profile", TRACE_MODE_READ,
1947 NULL, NULL, &kprobe_profile_ops);
1949 setup_boot_kprobe_events();
1953 fs_initcall(init_kprobe_trace);
1956 #ifdef CONFIG_FTRACE_STARTUP_TEST
1957 static __init struct trace_event_file *
1958 find_trace_probe_file(struct trace_kprobe *tk, struct trace_array *tr)
1960 struct trace_event_file *file;
1962 list_for_each_entry(file, &tr->events, list)
1963 if (file->event_call == trace_probe_event_call(&tk->tp))
1970 * Nobody but us can call enable_trace_kprobe/disable_trace_kprobe at this
1971 * stage, we can do this lockless.
1973 static __init int kprobe_trace_self_tests_init(void)
1976 int (*target)(int, int, int, int, int, int);
1977 struct trace_kprobe *tk;
1978 struct trace_event_file *file;
1980 if (tracing_is_disabled())
1983 if (tracing_selftest_disabled)
1986 target = kprobe_trace_selftest_target;
1988 pr_info("Testing kprobe tracing: ");
1990 ret = create_or_delete_trace_kprobe("p:testprobe kprobe_trace_selftest_target $stack $stack0 +0($stack)");
1991 if (WARN_ON_ONCE(ret)) {
1992 pr_warn("error on probing function entry.\n");
1995 /* Enable trace point */
1996 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1997 if (WARN_ON_ONCE(tk == NULL)) {
1998 pr_warn("error on getting new probe.\n");
2001 file = find_trace_probe_file(tk, top_trace_array());
2002 if (WARN_ON_ONCE(file == NULL)) {
2003 pr_warn("error on getting probe file.\n");
2006 enable_trace_kprobe(
2007 trace_probe_event_call(&tk->tp), file);
2011 ret = create_or_delete_trace_kprobe("r:testprobe2 kprobe_trace_selftest_target $retval");
2012 if (WARN_ON_ONCE(ret)) {
2013 pr_warn("error on probing function return.\n");
2016 /* Enable trace point */
2017 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
2018 if (WARN_ON_ONCE(tk == NULL)) {
2019 pr_warn("error on getting 2nd new probe.\n");
2022 file = find_trace_probe_file(tk, top_trace_array());
2023 if (WARN_ON_ONCE(file == NULL)) {
2024 pr_warn("error on getting probe file.\n");
2027 enable_trace_kprobe(
2028 trace_probe_event_call(&tk->tp), file);
2035 ret = target(1, 2, 3, 4, 5, 6);
2038 * Not expecting an error here, the check is only to prevent the
2039 * optimizer from removing the call to target() as otherwise there
2040 * are no side-effects and the call is never performed.
2045 /* Disable trace points before removing it */
2046 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
2047 if (WARN_ON_ONCE(tk == NULL)) {
2048 pr_warn("error on getting test probe.\n");
2051 if (trace_kprobe_nhit(tk) != 1) {
2052 pr_warn("incorrect number of testprobe hits\n");
2056 file = find_trace_probe_file(tk, top_trace_array());
2057 if (WARN_ON_ONCE(file == NULL)) {
2058 pr_warn("error on getting probe file.\n");
2061 disable_trace_kprobe(
2062 trace_probe_event_call(&tk->tp), file);
2065 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
2066 if (WARN_ON_ONCE(tk == NULL)) {
2067 pr_warn("error on getting 2nd test probe.\n");
2070 if (trace_kprobe_nhit(tk) != 1) {
2071 pr_warn("incorrect number of testprobe2 hits\n");
2075 file = find_trace_probe_file(tk, top_trace_array());
2076 if (WARN_ON_ONCE(file == NULL)) {
2077 pr_warn("error on getting probe file.\n");
2080 disable_trace_kprobe(
2081 trace_probe_event_call(&tk->tp), file);
2084 ret = create_or_delete_trace_kprobe("-:testprobe");
2085 if (WARN_ON_ONCE(ret)) {
2086 pr_warn("error on deleting a probe.\n");
2090 ret = create_or_delete_trace_kprobe("-:testprobe2");
2091 if (WARN_ON_ONCE(ret)) {
2092 pr_warn("error on deleting a probe.\n");
2097 ret = dyn_events_release_all(&trace_kprobe_ops);
2098 if (WARN_ON_ONCE(ret)) {
2099 pr_warn("error on cleaning up probes.\n");
2103 * Wait for the optimizer work to finish. Otherwise it might fiddle
2104 * with probes in already freed __init text.
2106 wait_for_kprobe_optimizer();
2108 pr_cont("NG: Some tests are failed. Please check them.\n");
2114 late_initcall(kprobe_trace_self_tests_init);