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/security.h>
11 #include <linux/module.h>
12 #include <linux/uaccess.h>
13 #include <linux/rculist.h>
14 #include <linux/error-injection.h>
16 #include <asm/setup.h> /* for COMMAND_LINE_SIZE */
18 #include "trace_dynevent.h"
19 #include "trace_kprobe_selftest.h"
20 #include "trace_probe.h"
21 #include "trace_probe_tmpl.h"
22 #include "trace_probe_kernel.h"
24 #define KPROBE_EVENT_SYSTEM "kprobes"
25 #define KRETPROBE_MAXACTIVE_MAX 4096
27 /* Kprobe early definition from command line */
28 static char kprobe_boot_events_buf[COMMAND_LINE_SIZE] __initdata;
30 static int __init set_kprobe_boot_events(char *str)
32 strlcpy(kprobe_boot_events_buf, str, COMMAND_LINE_SIZE);
33 disable_tracing_selftest("running kprobe events");
37 __setup("kprobe_event=", set_kprobe_boot_events);
39 static int trace_kprobe_create(const char *raw_command);
40 static int trace_kprobe_show(struct seq_file *m, struct dyn_event *ev);
41 static int trace_kprobe_release(struct dyn_event *ev);
42 static bool trace_kprobe_is_busy(struct dyn_event *ev);
43 static bool trace_kprobe_match(const char *system, const char *event,
44 int argc, const char **argv, struct dyn_event *ev);
46 static struct dyn_event_operations trace_kprobe_ops = {
47 .create = trace_kprobe_create,
48 .show = trace_kprobe_show,
49 .is_busy = trace_kprobe_is_busy,
50 .free = trace_kprobe_release,
51 .match = trace_kprobe_match,
55 * Kprobe event core functions
58 struct dyn_event devent;
59 struct kretprobe rp; /* Use rp.kp for kprobe use */
60 unsigned long __percpu *nhit;
61 const char *symbol; /* symbol name */
62 struct trace_probe tp;
65 static bool is_trace_kprobe(struct dyn_event *ev)
67 return ev->ops == &trace_kprobe_ops;
70 static struct trace_kprobe *to_trace_kprobe(struct dyn_event *ev)
72 return container_of(ev, struct trace_kprobe, devent);
76 * for_each_trace_kprobe - iterate over the trace_kprobe list
77 * @pos: the struct trace_kprobe * for each entry
78 * @dpos: the struct dyn_event * to use as a loop cursor
80 #define for_each_trace_kprobe(pos, dpos) \
81 for_each_dyn_event(dpos) \
82 if (is_trace_kprobe(dpos) && (pos = to_trace_kprobe(dpos)))
84 static nokprobe_inline bool trace_kprobe_is_return(struct trace_kprobe *tk)
86 return tk->rp.handler != NULL;
89 static nokprobe_inline const char *trace_kprobe_symbol(struct trace_kprobe *tk)
91 return tk->symbol ? tk->symbol : "unknown";
94 static nokprobe_inline unsigned long trace_kprobe_offset(struct trace_kprobe *tk)
96 return tk->rp.kp.offset;
99 static nokprobe_inline bool trace_kprobe_has_gone(struct trace_kprobe *tk)
101 return !!(kprobe_gone(&tk->rp.kp));
104 static nokprobe_inline bool trace_kprobe_within_module(struct trace_kprobe *tk,
107 int len = strlen(module_name(mod));
108 const char *name = trace_kprobe_symbol(tk);
110 return strncmp(module_name(mod), name, len) == 0 && name[len] == ':';
113 static nokprobe_inline bool trace_kprobe_module_exist(struct trace_kprobe *tk)
120 p = strchr(tk->symbol, ':');
124 rcu_read_lock_sched();
125 ret = !!find_module(tk->symbol);
126 rcu_read_unlock_sched();
132 static bool trace_kprobe_is_busy(struct dyn_event *ev)
134 struct trace_kprobe *tk = to_trace_kprobe(ev);
136 return trace_probe_is_enabled(&tk->tp);
139 static bool trace_kprobe_match_command_head(struct trace_kprobe *tk,
140 int argc, const char **argv)
142 char buf[MAX_ARGSTR_LEN + 1];
148 snprintf(buf, sizeof(buf), "0x%p", tk->rp.kp.addr);
149 else if (tk->rp.kp.offset)
150 snprintf(buf, sizeof(buf), "%s+%u",
151 trace_kprobe_symbol(tk), tk->rp.kp.offset);
153 snprintf(buf, sizeof(buf), "%s", trace_kprobe_symbol(tk));
154 if (strcmp(buf, argv[0]))
158 return trace_probe_match_command_args(&tk->tp, argc, argv);
161 static bool trace_kprobe_match(const char *system, const char *event,
162 int argc, const char **argv, struct dyn_event *ev)
164 struct trace_kprobe *tk = to_trace_kprobe(ev);
166 return strcmp(trace_probe_name(&tk->tp), event) == 0 &&
167 (!system || strcmp(trace_probe_group_name(&tk->tp), system) == 0) &&
168 trace_kprobe_match_command_head(tk, argc, argv);
171 static nokprobe_inline unsigned long trace_kprobe_nhit(struct trace_kprobe *tk)
173 unsigned long nhit = 0;
176 for_each_possible_cpu(cpu)
177 nhit += *per_cpu_ptr(tk->nhit, cpu);
182 static nokprobe_inline bool trace_kprobe_is_registered(struct trace_kprobe *tk)
184 return !(list_empty(&tk->rp.kp.list) &&
185 hlist_unhashed(&tk->rp.kp.hlist));
188 /* Return 0 if it fails to find the symbol address */
189 static nokprobe_inline
190 unsigned long trace_kprobe_address(struct trace_kprobe *tk)
195 addr = (unsigned long)
196 kallsyms_lookup_name(trace_kprobe_symbol(tk));
198 addr += tk->rp.kp.offset;
200 addr = (unsigned long)tk->rp.kp.addr;
205 static nokprobe_inline struct trace_kprobe *
206 trace_kprobe_primary_from_call(struct trace_event_call *call)
208 struct trace_probe *tp;
210 tp = trace_probe_primary_from_call(call);
211 if (WARN_ON_ONCE(!tp))
214 return container_of(tp, struct trace_kprobe, tp);
217 bool trace_kprobe_on_func_entry(struct trace_event_call *call)
219 struct trace_kprobe *tk = trace_kprobe_primary_from_call(call);
221 return tk ? (kprobe_on_func_entry(tk->rp.kp.addr,
222 tk->rp.kp.addr ? NULL : tk->rp.kp.symbol_name,
223 tk->rp.kp.addr ? 0 : tk->rp.kp.offset) == 0) : false;
226 bool trace_kprobe_error_injectable(struct trace_event_call *call)
228 struct trace_kprobe *tk = trace_kprobe_primary_from_call(call);
230 return tk ? within_error_injection_list(trace_kprobe_address(tk)) :
234 static int register_kprobe_event(struct trace_kprobe *tk);
235 static int unregister_kprobe_event(struct trace_kprobe *tk);
237 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
238 static int kretprobe_dispatcher(struct kretprobe_instance *ri,
239 struct pt_regs *regs);
241 static void free_trace_kprobe(struct trace_kprobe *tk)
244 trace_probe_cleanup(&tk->tp);
246 free_percpu(tk->nhit);
252 * Allocate new trace_probe and initialize it (including kprobes).
254 static struct trace_kprobe *alloc_trace_kprobe(const char *group,
260 int nargs, bool is_return)
262 struct trace_kprobe *tk;
265 tk = kzalloc(struct_size(tk, tp.args, nargs), GFP_KERNEL);
269 tk->nhit = alloc_percpu(unsigned long);
274 tk->symbol = kstrdup(symbol, GFP_KERNEL);
277 tk->rp.kp.symbol_name = tk->symbol;
278 tk->rp.kp.offset = offs;
280 tk->rp.kp.addr = addr;
283 tk->rp.handler = kretprobe_dispatcher;
285 tk->rp.kp.pre_handler = kprobe_dispatcher;
287 tk->rp.maxactive = maxactive;
288 INIT_HLIST_NODE(&tk->rp.kp.hlist);
289 INIT_LIST_HEAD(&tk->rp.kp.list);
291 ret = trace_probe_init(&tk->tp, event, group, false);
295 dyn_event_init(&tk->devent, &trace_kprobe_ops);
298 free_trace_kprobe(tk);
302 static struct trace_kprobe *find_trace_kprobe(const char *event,
305 struct dyn_event *pos;
306 struct trace_kprobe *tk;
308 for_each_trace_kprobe(tk, pos)
309 if (strcmp(trace_probe_name(&tk->tp), event) == 0 &&
310 strcmp(trace_probe_group_name(&tk->tp), group) == 0)
315 static inline int __enable_trace_kprobe(struct trace_kprobe *tk)
319 if (trace_kprobe_is_registered(tk) && !trace_kprobe_has_gone(tk)) {
320 if (trace_kprobe_is_return(tk))
321 ret = enable_kretprobe(&tk->rp);
323 ret = enable_kprobe(&tk->rp.kp);
329 static void __disable_trace_kprobe(struct trace_probe *tp)
331 struct trace_probe *pos;
332 struct trace_kprobe *tk;
334 list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
335 tk = container_of(pos, struct trace_kprobe, tp);
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 *pos, *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(pos, trace_probe_probe_list(tp), list) {
374 tk = container_of(pos, struct trace_kprobe, tp);
375 if (trace_kprobe_has_gone(tk))
377 ret = __enable_trace_kprobe(tk);
384 /* Failed to enable one of them. Roll back all */
386 __disable_trace_kprobe(tp);
388 trace_probe_remove_file(tp, file);
390 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
397 * Disable trace_probe
398 * if the file is NULL, disable "perf" handler, or disable "trace" handler.
400 static int disable_trace_kprobe(struct trace_event_call *call,
401 struct trace_event_file *file)
403 struct trace_probe *tp;
405 tp = trace_probe_primary_from_call(call);
406 if (WARN_ON_ONCE(!tp))
410 if (!trace_probe_get_file_link(tp, file))
412 if (!trace_probe_has_single_file(tp))
414 trace_probe_clear_flag(tp, TP_FLAG_TRACE);
416 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
418 if (!trace_probe_is_enabled(tp))
419 __disable_trace_kprobe(tp);
424 * Synchronization is done in below function. For perf event,
425 * file == NULL and perf_trace_event_unreg() calls
426 * tracepoint_synchronize_unregister() to ensure synchronize
427 * event. We don't need to care about it.
429 trace_probe_remove_file(tp, file);
434 #if defined(CONFIG_DYNAMIC_FTRACE) && \
435 !defined(CONFIG_KPROBE_EVENTS_ON_NOTRACE)
436 static bool __within_notrace_func(unsigned long addr)
438 unsigned long offset, size;
440 if (!addr || !kallsyms_lookup_size_offset(addr, &size, &offset))
443 /* Get the entry address of the target function */
447 * Since ftrace_location_range() does inclusive range check, we need
448 * to subtract 1 byte from the end address.
450 return !ftrace_location_range(addr, addr + size - 1);
453 static bool within_notrace_func(struct trace_kprobe *tk)
455 unsigned long addr = trace_kprobe_address(tk);
456 char symname[KSYM_NAME_LEN], *p;
458 if (!__within_notrace_func(addr))
461 /* Check if the address is on a suffixed-symbol */
462 if (!lookup_symbol_name(addr, symname)) {
463 p = strchr(symname, '.');
467 addr = (unsigned long)kprobe_lookup_name(symname, 0);
469 return __within_notrace_func(addr);
475 #define within_notrace_func(tk) (false)
478 /* Internal register function - just handle k*probes and flags */
479 static int __register_trace_kprobe(struct trace_kprobe *tk)
483 ret = security_locked_down(LOCKDOWN_KPROBES);
487 if (trace_kprobe_is_registered(tk))
490 if (within_notrace_func(tk)) {
491 pr_warn("Could not probe notrace function %s\n",
492 trace_kprobe_symbol(tk));
496 for (i = 0; i < tk->tp.nr_args; i++) {
497 ret = traceprobe_update_arg(&tk->tp.args[i]);
502 /* Set/clear disabled flag according to tp->flag */
503 if (trace_probe_is_enabled(&tk->tp))
504 tk->rp.kp.flags &= ~KPROBE_FLAG_DISABLED;
506 tk->rp.kp.flags |= KPROBE_FLAG_DISABLED;
508 if (trace_kprobe_is_return(tk))
509 ret = register_kretprobe(&tk->rp);
511 ret = register_kprobe(&tk->rp.kp);
516 /* Internal unregister function - just handle k*probes and flags */
517 static void __unregister_trace_kprobe(struct trace_kprobe *tk)
519 if (trace_kprobe_is_registered(tk)) {
520 if (trace_kprobe_is_return(tk))
521 unregister_kretprobe(&tk->rp);
523 unregister_kprobe(&tk->rp.kp);
524 /* Cleanup kprobe for reuse and mark it unregistered */
525 INIT_HLIST_NODE(&tk->rp.kp.hlist);
526 INIT_LIST_HEAD(&tk->rp.kp.list);
527 if (tk->rp.kp.symbol_name)
528 tk->rp.kp.addr = NULL;
532 /* Unregister a trace_probe and probe_event */
533 static int unregister_trace_kprobe(struct trace_kprobe *tk)
535 /* If other probes are on the event, just unregister kprobe */
536 if (trace_probe_has_sibling(&tk->tp))
539 /* Enabled event can not be unregistered */
540 if (trace_probe_is_enabled(&tk->tp))
543 /* If there's a reference to the dynamic event */
544 if (trace_event_dyn_busy(trace_probe_event_call(&tk->tp)))
547 /* Will fail if probe is being used by ftrace or perf */
548 if (unregister_kprobe_event(tk))
552 __unregister_trace_kprobe(tk);
553 dyn_event_remove(&tk->devent);
554 trace_probe_unlink(&tk->tp);
559 static bool trace_kprobe_has_same_kprobe(struct trace_kprobe *orig,
560 struct trace_kprobe *comp)
562 struct trace_probe_event *tpe = orig->tp.event;
563 struct trace_probe *pos;
566 list_for_each_entry(pos, &tpe->probes, list) {
567 orig = container_of(pos, struct trace_kprobe, tp);
568 if (strcmp(trace_kprobe_symbol(orig),
569 trace_kprobe_symbol(comp)) ||
570 trace_kprobe_offset(orig) != trace_kprobe_offset(comp))
574 * trace_probe_compare_arg_type() ensured that nr_args and
575 * each argument name and type are same. Let's compare comm.
577 for (i = 0; i < orig->tp.nr_args; i++) {
578 if (strcmp(orig->tp.args[i].comm,
579 comp->tp.args[i].comm))
583 if (i == orig->tp.nr_args)
590 static int append_trace_kprobe(struct trace_kprobe *tk, struct trace_kprobe *to)
594 ret = trace_probe_compare_arg_type(&tk->tp, &to->tp);
596 /* Note that argument starts index = 2 */
597 trace_probe_log_set_index(ret + 1);
598 trace_probe_log_err(0, DIFF_ARG_TYPE);
601 if (trace_kprobe_has_same_kprobe(to, tk)) {
602 trace_probe_log_set_index(0);
603 trace_probe_log_err(0, SAME_PROBE);
607 /* Append to existing event */
608 ret = trace_probe_append(&tk->tp, &to->tp);
612 /* Register k*probe */
613 ret = __register_trace_kprobe(tk);
614 if (ret == -ENOENT && !trace_kprobe_module_exist(tk)) {
615 pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
620 trace_probe_unlink(&tk->tp);
622 dyn_event_add(&tk->devent, trace_probe_event_call(&tk->tp));
627 /* Register a trace_probe and probe_event */
628 static int register_trace_kprobe(struct trace_kprobe *tk)
630 struct trace_kprobe *old_tk;
633 mutex_lock(&event_mutex);
635 old_tk = find_trace_kprobe(trace_probe_name(&tk->tp),
636 trace_probe_group_name(&tk->tp));
638 if (trace_kprobe_is_return(tk) != trace_kprobe_is_return(old_tk)) {
639 trace_probe_log_set_index(0);
640 trace_probe_log_err(0, DIFF_PROBE_TYPE);
643 ret = append_trace_kprobe(tk, old_tk);
648 /* Register new event */
649 ret = register_kprobe_event(tk);
651 if (ret == -EEXIST) {
652 trace_probe_log_set_index(0);
653 trace_probe_log_err(0, EVENT_EXIST);
655 pr_warn("Failed to register probe event(%d)\n", ret);
659 /* Register k*probe */
660 ret = __register_trace_kprobe(tk);
661 if (ret == -ENOENT && !trace_kprobe_module_exist(tk)) {
662 pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
667 unregister_kprobe_event(tk);
669 dyn_event_add(&tk->devent, trace_probe_event_call(&tk->tp));
672 mutex_unlock(&event_mutex);
676 /* Module notifier call back, checking event on the module */
677 static int trace_kprobe_module_callback(struct notifier_block *nb,
678 unsigned long val, void *data)
680 struct module *mod = data;
681 struct dyn_event *pos;
682 struct trace_kprobe *tk;
685 if (val != MODULE_STATE_COMING)
688 /* Update probes on coming module */
689 mutex_lock(&event_mutex);
690 for_each_trace_kprobe(tk, pos) {
691 if (trace_kprobe_within_module(tk, mod)) {
692 /* Don't need to check busy - this should have gone. */
693 __unregister_trace_kprobe(tk);
694 ret = __register_trace_kprobe(tk);
696 pr_warn("Failed to re-register probe %s on %s: %d\n",
697 trace_probe_name(&tk->tp),
698 module_name(mod), ret);
701 mutex_unlock(&event_mutex);
706 static struct notifier_block trace_kprobe_module_nb = {
707 .notifier_call = trace_kprobe_module_callback,
708 .priority = 1 /* Invoked after kprobe module callback */
711 static int __trace_kprobe_create(int argc, const char *argv[])
716 * p[:[GRP/]EVENT] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
718 * r[MAXACTIVE][:[GRP/]EVENT] [MOD:]KSYM[+0] [FETCHARGS]
720 * p:[GRP/]EVENT] [MOD:]KSYM[+0]%return [FETCHARGS]
723 * $retval : fetch return value
724 * $stack : fetch stack address
725 * $stackN : fetch Nth of stack (N:0-)
726 * $comm : fetch current task comm
727 * @ADDR : fetch memory at ADDR (ADDR should be in kernel)
728 * @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
729 * %REG : fetch register REG
730 * Dereferencing memory fetch:
731 * +|-offs(ARG) : fetch memory at ARG +|- offs address.
732 * Alias name of args:
733 * NAME=FETCHARG : set NAME as alias of FETCHARG.
735 * FETCHARG:TYPE : use TYPE instead of unsigned long.
737 struct trace_kprobe *tk = NULL;
739 bool is_return = false;
740 char *symbol = NULL, *tmp = NULL;
741 const char *event = NULL, *group = KPROBE_EVENT_SYSTEM;
742 enum probe_print_type ptype;
746 char buf[MAX_EVENT_NAME_LEN];
747 unsigned int flags = TPARG_FL_KERNEL;
749 switch (argv[0][0]) {
761 trace_probe_log_init("trace_kprobe", argc, argv);
763 event = strchr(&argv[0][1], ':');
767 if (isdigit(argv[0][1])) {
769 trace_probe_log_err(1, MAXACT_NO_KPROBE);
773 len = event - &argv[0][1] - 1;
775 len = strlen(&argv[0][1]);
776 if (len > MAX_EVENT_NAME_LEN - 1) {
777 trace_probe_log_err(1, BAD_MAXACT);
780 memcpy(buf, &argv[0][1], len);
782 ret = kstrtouint(buf, 0, &maxactive);
783 if (ret || !maxactive) {
784 trace_probe_log_err(1, BAD_MAXACT);
787 /* kretprobes instances are iterated over via a list. The
788 * maximum should stay reasonable.
790 if (maxactive > KRETPROBE_MAXACTIVE_MAX) {
791 trace_probe_log_err(1, MAXACT_TOO_BIG);
796 /* try to parse an address. if that fails, try to read the
797 * input as a symbol. */
798 if (kstrtoul(argv[1], 0, (unsigned long *)&addr)) {
799 trace_probe_log_set_index(1);
800 /* Check whether uprobe event specified */
801 if (strchr(argv[1], '/') && strchr(argv[1], ':')) {
805 /* a symbol specified */
806 symbol = kstrdup(argv[1], GFP_KERNEL);
810 tmp = strchr(symbol, '%');
812 if (!strcmp(tmp, "%return")) {
816 trace_probe_log_err(tmp - symbol, BAD_ADDR_SUFFIX);
821 /* TODO: support .init module functions */
822 ret = traceprobe_split_symbol_offset(symbol, &offset);
823 if (ret || offset < 0 || offset > UINT_MAX) {
824 trace_probe_log_err(0, BAD_PROBE_ADDR);
828 flags |= TPARG_FL_RETURN;
829 ret = kprobe_on_func_entry(NULL, symbol, offset);
831 flags |= TPARG_FL_FENTRY;
832 /* Defer the ENOENT case until register kprobe */
833 if (ret == -EINVAL && is_return) {
834 trace_probe_log_err(0, BAD_RETPROBE);
839 trace_probe_log_set_index(0);
841 ret = traceprobe_parse_event_name(&event, &group, buf,
846 /* Make a new event name */
848 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
849 is_return ? 'r' : 'p', symbol, offset);
851 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
852 is_return ? 'r' : 'p', addr);
853 sanitize_event_name(buf);
858 tk = alloc_trace_kprobe(group, event, addr, symbol, offset, maxactive,
859 argc - 2, is_return);
862 /* This must return -ENOMEM, else there is a bug */
863 WARN_ON_ONCE(ret != -ENOMEM);
864 goto out; /* We know tk is not allocated */
866 argc -= 2; argv += 2;
868 /* parse arguments */
869 for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
870 trace_probe_log_set_index(i + 2);
871 ret = traceprobe_parse_probe_arg(&tk->tp, i, argv[i], flags);
873 goto error; /* This can be -ENOMEM */
876 ptype = is_return ? PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL;
877 ret = traceprobe_set_print_fmt(&tk->tp, ptype);
881 ret = register_trace_kprobe(tk);
883 trace_probe_log_set_index(1);
885 trace_probe_log_err(0, BAD_INSN_BNDRY);
886 else if (ret == -ENOENT)
887 trace_probe_log_err(0, BAD_PROBE_ADDR);
888 else if (ret != -ENOMEM && ret != -EEXIST)
889 trace_probe_log_err(0, FAIL_REG_PROBE);
894 trace_probe_log_clear();
901 free_trace_kprobe(tk);
905 static int trace_kprobe_create(const char *raw_command)
907 return trace_probe_create(raw_command, __trace_kprobe_create);
910 static int create_or_delete_trace_kprobe(const char *raw_command)
914 if (raw_command[0] == '-')
915 return dyn_event_release(raw_command, &trace_kprobe_ops);
917 ret = trace_kprobe_create(raw_command);
918 return ret == -ECANCELED ? -EINVAL : ret;
921 static int trace_kprobe_run_command(struct dynevent_cmd *cmd)
923 return create_or_delete_trace_kprobe(cmd->seq.buffer);
927 * kprobe_event_cmd_init - Initialize a kprobe event command object
928 * @cmd: A pointer to the dynevent_cmd struct representing the new event
929 * @buf: A pointer to the buffer used to build the command
930 * @maxlen: The length of the buffer passed in @buf
932 * Initialize a synthetic event command object. Use this before
933 * calling any of the other kprobe_event functions.
935 void kprobe_event_cmd_init(struct dynevent_cmd *cmd, char *buf, int maxlen)
937 dynevent_cmd_init(cmd, buf, maxlen, DYNEVENT_TYPE_KPROBE,
938 trace_kprobe_run_command);
940 EXPORT_SYMBOL_GPL(kprobe_event_cmd_init);
943 * __kprobe_event_gen_cmd_start - Generate a kprobe event command from arg list
944 * @cmd: A pointer to the dynevent_cmd struct representing the new event
945 * @name: The name of the kprobe event
946 * @loc: The location of the kprobe event
947 * @kretprobe: Is this a return probe?
948 * @args: Variable number of arg (pairs), one pair for each field
950 * NOTE: Users normally won't want to call this function directly, but
951 * rather use the kprobe_event_gen_cmd_start() wrapper, which automatically
952 * adds a NULL to the end of the arg list. If this function is used
953 * directly, make sure the last arg in the variable arg list is NULL.
955 * Generate a kprobe event command to be executed by
956 * kprobe_event_gen_cmd_end(). This function can be used to generate the
957 * complete command or only the first part of it; in the latter case,
958 * kprobe_event_add_fields() can be used to add more fields following this.
960 * Unlikely the synth_event_gen_cmd_start(), @loc must be specified. This
961 * returns -EINVAL if @loc == NULL.
963 * Return: 0 if successful, error otherwise.
965 int __kprobe_event_gen_cmd_start(struct dynevent_cmd *cmd, bool kretprobe,
966 const char *name, const char *loc, ...)
968 char buf[MAX_EVENT_NAME_LEN];
969 struct dynevent_arg arg;
973 if (cmd->type != DYNEVENT_TYPE_KPROBE)
980 snprintf(buf, MAX_EVENT_NAME_LEN, "r:kprobes/%s", name);
982 snprintf(buf, MAX_EVENT_NAME_LEN, "p:kprobes/%s", name);
984 ret = dynevent_str_add(cmd, buf);
988 dynevent_arg_init(&arg, 0);
990 ret = dynevent_arg_add(cmd, &arg, NULL);
998 field = va_arg(args, const char *);
1002 if (++cmd->n_fields > MAX_TRACE_ARGS) {
1008 ret = dynevent_arg_add(cmd, &arg, NULL);
1016 EXPORT_SYMBOL_GPL(__kprobe_event_gen_cmd_start);
1019 * __kprobe_event_add_fields - Add probe fields to a kprobe command from arg list
1020 * @cmd: A pointer to the dynevent_cmd struct representing the new event
1021 * @args: Variable number of arg (pairs), one pair for each field
1023 * NOTE: Users normally won't want to call this function directly, but
1024 * rather use the kprobe_event_add_fields() wrapper, which
1025 * automatically adds a NULL to the end of the arg list. If this
1026 * function is used directly, make sure the last arg in the variable
1029 * Add probe fields to an existing kprobe command using a variable
1030 * list of args. Fields are added in the same order they're listed.
1032 * Return: 0 if successful, error otherwise.
1034 int __kprobe_event_add_fields(struct dynevent_cmd *cmd, ...)
1036 struct dynevent_arg arg;
1040 if (cmd->type != DYNEVENT_TYPE_KPROBE)
1043 dynevent_arg_init(&arg, 0);
1045 va_start(args, cmd);
1049 field = va_arg(args, const char *);
1053 if (++cmd->n_fields > MAX_TRACE_ARGS) {
1059 ret = dynevent_arg_add(cmd, &arg, NULL);
1067 EXPORT_SYMBOL_GPL(__kprobe_event_add_fields);
1070 * kprobe_event_delete - Delete a kprobe event
1071 * @name: The name of the kprobe event to delete
1073 * Delete a kprobe event with the give @name from kernel code rather
1074 * than directly from the command line.
1076 * Return: 0 if successful, error otherwise.
1078 int kprobe_event_delete(const char *name)
1080 char buf[MAX_EVENT_NAME_LEN];
1082 snprintf(buf, MAX_EVENT_NAME_LEN, "-:%s", name);
1084 return create_or_delete_trace_kprobe(buf);
1086 EXPORT_SYMBOL_GPL(kprobe_event_delete);
1088 static int trace_kprobe_release(struct dyn_event *ev)
1090 struct trace_kprobe *tk = to_trace_kprobe(ev);
1091 int ret = unregister_trace_kprobe(tk);
1094 free_trace_kprobe(tk);
1098 static int trace_kprobe_show(struct seq_file *m, struct dyn_event *ev)
1100 struct trace_kprobe *tk = to_trace_kprobe(ev);
1103 seq_putc(m, trace_kprobe_is_return(tk) ? 'r' : 'p');
1104 if (trace_kprobe_is_return(tk) && tk->rp.maxactive)
1105 seq_printf(m, "%d", tk->rp.maxactive);
1106 seq_printf(m, ":%s/%s", trace_probe_group_name(&tk->tp),
1107 trace_probe_name(&tk->tp));
1110 seq_printf(m, " 0x%p", tk->rp.kp.addr);
1111 else if (tk->rp.kp.offset)
1112 seq_printf(m, " %s+%u", trace_kprobe_symbol(tk),
1115 seq_printf(m, " %s", trace_kprobe_symbol(tk));
1117 for (i = 0; i < tk->tp.nr_args; i++)
1118 seq_printf(m, " %s=%s", tk->tp.args[i].name, tk->tp.args[i].comm);
1124 static int probes_seq_show(struct seq_file *m, void *v)
1126 struct dyn_event *ev = v;
1128 if (!is_trace_kprobe(ev))
1131 return trace_kprobe_show(m, ev);
1134 static const struct seq_operations probes_seq_op = {
1135 .start = dyn_event_seq_start,
1136 .next = dyn_event_seq_next,
1137 .stop = dyn_event_seq_stop,
1138 .show = probes_seq_show
1141 static int probes_open(struct inode *inode, struct file *file)
1145 ret = security_locked_down(LOCKDOWN_TRACEFS);
1149 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
1150 ret = dyn_events_release_all(&trace_kprobe_ops);
1155 return seq_open(file, &probes_seq_op);
1158 static ssize_t probes_write(struct file *file, const char __user *buffer,
1159 size_t count, loff_t *ppos)
1161 return trace_parse_run_command(file, buffer, count, ppos,
1162 create_or_delete_trace_kprobe);
1165 static const struct file_operations kprobe_events_ops = {
1166 .owner = THIS_MODULE,
1167 .open = probes_open,
1169 .llseek = seq_lseek,
1170 .release = seq_release,
1171 .write = probes_write,
1174 /* Probes profiling interfaces */
1175 static int probes_profile_seq_show(struct seq_file *m, void *v)
1177 struct dyn_event *ev = v;
1178 struct trace_kprobe *tk;
1179 unsigned long nmissed;
1181 if (!is_trace_kprobe(ev))
1184 tk = to_trace_kprobe(ev);
1185 nmissed = trace_kprobe_is_return(tk) ?
1186 tk->rp.kp.nmissed + tk->rp.nmissed : tk->rp.kp.nmissed;
1187 seq_printf(m, " %-44s %15lu %15lu\n",
1188 trace_probe_name(&tk->tp),
1189 trace_kprobe_nhit(tk),
1195 static const struct seq_operations profile_seq_op = {
1196 .start = dyn_event_seq_start,
1197 .next = dyn_event_seq_next,
1198 .stop = dyn_event_seq_stop,
1199 .show = probes_profile_seq_show
1202 static int profile_open(struct inode *inode, struct file *file)
1206 ret = security_locked_down(LOCKDOWN_TRACEFS);
1210 return seq_open(file, &profile_seq_op);
1213 static const struct file_operations kprobe_profile_ops = {
1214 .owner = THIS_MODULE,
1215 .open = profile_open,
1217 .llseek = seq_lseek,
1218 .release = seq_release,
1221 /* Kprobe specific fetch functions */
1223 /* Return the length of string -- including null terminal byte */
1224 static nokprobe_inline int
1225 fetch_store_strlen_user(unsigned long addr)
1227 return kern_fetch_store_strlen_user(addr);
1230 /* Return the length of string -- including null terminal byte */
1231 static nokprobe_inline int
1232 fetch_store_strlen(unsigned long addr)
1234 return kern_fetch_store_strlen(addr);
1238 * Fetch a null-terminated string from user. Caller MUST set *(u32 *)buf
1239 * with max length and relative data location.
1241 static nokprobe_inline int
1242 fetch_store_string_user(unsigned long addr, void *dest, void *base)
1244 return kern_fetch_store_string_user(addr, dest, base);
1248 * Fetch a null-terminated string. Caller MUST set *(u32 *)buf with max
1249 * length and relative data location.
1251 static nokprobe_inline int
1252 fetch_store_string(unsigned long addr, void *dest, void *base)
1254 return kern_fetch_store_string(addr, dest, base);
1257 static nokprobe_inline int
1258 probe_mem_read_user(void *dest, void *src, size_t size)
1260 const void __user *uaddr = (__force const void __user *)src;
1262 return copy_from_user_nofault(dest, uaddr, size);
1265 static nokprobe_inline int
1266 probe_mem_read(void *dest, void *src, size_t size)
1268 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
1269 if ((unsigned long)src < TASK_SIZE)
1270 return probe_mem_read_user(dest, src, size);
1272 return copy_from_kernel_nofault(dest, src, size);
1275 /* Note that we don't verify it, since the code does not come from user space */
1277 process_fetch_insn(struct fetch_insn *code, void *rec, void *dest,
1280 struct pt_regs *regs = rec;
1284 /* 1st stage: get value from context */
1287 val = regs_get_register(regs, code->param);
1289 case FETCH_OP_STACK:
1290 val = regs_get_kernel_stack_nth(regs, code->param);
1292 case FETCH_OP_STACKP:
1293 val = kernel_stack_pointer(regs);
1295 case FETCH_OP_RETVAL:
1296 val = regs_return_value(regs);
1299 val = code->immediate;
1302 val = (unsigned long)current->comm;
1305 val = (unsigned long)code->data;
1307 #ifdef CONFIG_HAVE_FUNCTION_ARG_ACCESS_API
1309 val = regs_get_kernel_argument(regs, code->param);
1312 case FETCH_NOP_SYMBOL: /* Ignore a place holder */
1320 return process_fetch_insn_bottom(code, val, dest, base);
1322 NOKPROBE_SYMBOL(process_fetch_insn)
1324 /* Kprobe handler */
1325 static nokprobe_inline void
1326 __kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs,
1327 struct trace_event_file *trace_file)
1329 struct kprobe_trace_entry_head *entry;
1330 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1331 struct trace_event_buffer fbuffer;
1334 WARN_ON(call != trace_file->event_call);
1336 if (trace_trigger_soft_disabled(trace_file))
1339 fbuffer.trace_ctx = tracing_gen_ctx();
1340 fbuffer.trace_file = trace_file;
1342 dsize = __get_data_size(&tk->tp, regs);
1345 trace_event_buffer_lock_reserve(&fbuffer.buffer, trace_file,
1347 sizeof(*entry) + tk->tp.size + dsize,
1352 fbuffer.regs = regs;
1353 entry = fbuffer.entry = ring_buffer_event_data(fbuffer.event);
1354 entry->ip = (unsigned long)tk->rp.kp.addr;
1355 store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1357 trace_event_buffer_commit(&fbuffer);
1361 kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs)
1363 struct event_file_link *link;
1365 trace_probe_for_each_link_rcu(link, &tk->tp)
1366 __kprobe_trace_func(tk, regs, link->file);
1368 NOKPROBE_SYMBOL(kprobe_trace_func);
1370 /* Kretprobe handler */
1371 static nokprobe_inline void
1372 __kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1373 struct pt_regs *regs,
1374 struct trace_event_file *trace_file)
1376 struct kretprobe_trace_entry_head *entry;
1377 struct trace_event_buffer fbuffer;
1378 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1381 WARN_ON(call != trace_file->event_call);
1383 if (trace_trigger_soft_disabled(trace_file))
1386 fbuffer.trace_ctx = tracing_gen_ctx();
1387 fbuffer.trace_file = trace_file;
1389 dsize = __get_data_size(&tk->tp, regs);
1391 trace_event_buffer_lock_reserve(&fbuffer.buffer, trace_file,
1393 sizeof(*entry) + tk->tp.size + dsize,
1398 fbuffer.regs = regs;
1399 entry = fbuffer.entry = ring_buffer_event_data(fbuffer.event);
1400 entry->func = (unsigned long)tk->rp.kp.addr;
1401 entry->ret_ip = (unsigned long)ri->ret_addr;
1402 store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1404 trace_event_buffer_commit(&fbuffer);
1408 kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1409 struct pt_regs *regs)
1411 struct event_file_link *link;
1413 trace_probe_for_each_link_rcu(link, &tk->tp)
1414 __kretprobe_trace_func(tk, ri, regs, link->file);
1416 NOKPROBE_SYMBOL(kretprobe_trace_func);
1418 /* Event entry printers */
1419 static enum print_line_t
1420 print_kprobe_event(struct trace_iterator *iter, int flags,
1421 struct trace_event *event)
1423 struct kprobe_trace_entry_head *field;
1424 struct trace_seq *s = &iter->seq;
1425 struct trace_probe *tp;
1427 field = (struct kprobe_trace_entry_head *)iter->ent;
1428 tp = trace_probe_primary_from_call(
1429 container_of(event, struct trace_event_call, event));
1430 if (WARN_ON_ONCE(!tp))
1433 trace_seq_printf(s, "%s: (", trace_probe_name(tp));
1435 if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
1438 trace_seq_putc(s, ')');
1440 if (print_probe_args(s, tp->args, tp->nr_args,
1441 (u8 *)&field[1], field) < 0)
1444 trace_seq_putc(s, '\n');
1446 return trace_handle_return(s);
1449 static enum print_line_t
1450 print_kretprobe_event(struct trace_iterator *iter, int flags,
1451 struct trace_event *event)
1453 struct kretprobe_trace_entry_head *field;
1454 struct trace_seq *s = &iter->seq;
1455 struct trace_probe *tp;
1457 field = (struct kretprobe_trace_entry_head *)iter->ent;
1458 tp = trace_probe_primary_from_call(
1459 container_of(event, struct trace_event_call, event));
1460 if (WARN_ON_ONCE(!tp))
1463 trace_seq_printf(s, "%s: (", trace_probe_name(tp));
1465 if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1468 trace_seq_puts(s, " <- ");
1470 if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1473 trace_seq_putc(s, ')');
1475 if (print_probe_args(s, tp->args, tp->nr_args,
1476 (u8 *)&field[1], field) < 0)
1479 trace_seq_putc(s, '\n');
1482 return trace_handle_return(s);
1486 static int kprobe_event_define_fields(struct trace_event_call *event_call)
1489 struct kprobe_trace_entry_head field;
1490 struct trace_probe *tp;
1492 tp = trace_probe_primary_from_call(event_call);
1493 if (WARN_ON_ONCE(!tp))
1496 DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1498 return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
1501 static int kretprobe_event_define_fields(struct trace_event_call *event_call)
1504 struct kretprobe_trace_entry_head field;
1505 struct trace_probe *tp;
1507 tp = trace_probe_primary_from_call(event_call);
1508 if (WARN_ON_ONCE(!tp))
1511 DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1512 DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1514 return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
1517 #ifdef CONFIG_PERF_EVENTS
1519 /* Kprobe profile handler */
1521 kprobe_perf_func(struct trace_kprobe *tk, struct pt_regs *regs)
1523 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1524 struct kprobe_trace_entry_head *entry;
1525 struct hlist_head *head;
1526 int size, __size, dsize;
1529 if (bpf_prog_array_valid(call)) {
1530 unsigned long orig_ip = instruction_pointer(regs);
1533 ret = trace_call_bpf(call, regs);
1536 * We need to check and see if we modified the pc of the
1537 * pt_regs, and if so return 1 so that we don't do the
1540 if (orig_ip != instruction_pointer(regs))
1546 head = this_cpu_ptr(call->perf_events);
1547 if (hlist_empty(head))
1550 dsize = __get_data_size(&tk->tp, regs);
1551 __size = sizeof(*entry) + tk->tp.size + dsize;
1552 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1553 size -= sizeof(u32);
1555 entry = perf_trace_buf_alloc(size, NULL, &rctx);
1559 entry->ip = (unsigned long)tk->rp.kp.addr;
1560 memset(&entry[1], 0, dsize);
1561 store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1562 perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1566 NOKPROBE_SYMBOL(kprobe_perf_func);
1568 /* Kretprobe profile handler */
1570 kretprobe_perf_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1571 struct pt_regs *regs)
1573 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1574 struct kretprobe_trace_entry_head *entry;
1575 struct hlist_head *head;
1576 int size, __size, dsize;
1579 if (bpf_prog_array_valid(call) && !trace_call_bpf(call, regs))
1582 head = this_cpu_ptr(call->perf_events);
1583 if (hlist_empty(head))
1586 dsize = __get_data_size(&tk->tp, regs);
1587 __size = sizeof(*entry) + tk->tp.size + dsize;
1588 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1589 size -= sizeof(u32);
1591 entry = perf_trace_buf_alloc(size, NULL, &rctx);
1595 entry->func = (unsigned long)tk->rp.kp.addr;
1596 entry->ret_ip = (unsigned long)ri->ret_addr;
1597 store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1598 perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1601 NOKPROBE_SYMBOL(kretprobe_perf_func);
1603 int bpf_get_kprobe_info(const struct perf_event *event, u32 *fd_type,
1604 const char **symbol, u64 *probe_offset,
1605 u64 *probe_addr, bool perf_type_tracepoint)
1607 const char *pevent = trace_event_name(event->tp_event);
1608 const char *group = event->tp_event->class->system;
1609 struct trace_kprobe *tk;
1611 if (perf_type_tracepoint)
1612 tk = find_trace_kprobe(pevent, group);
1614 tk = trace_kprobe_primary_from_call(event->tp_event);
1618 *fd_type = trace_kprobe_is_return(tk) ? BPF_FD_TYPE_KRETPROBE
1619 : BPF_FD_TYPE_KPROBE;
1621 *symbol = tk->symbol;
1622 *probe_offset = tk->rp.kp.offset;
1627 *probe_addr = (unsigned long)tk->rp.kp.addr;
1631 #endif /* CONFIG_PERF_EVENTS */
1634 * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex.
1636 * kprobe_trace_self_tests_init() does enable_trace_probe/disable_trace_probe
1637 * lockless, but we can't race with this __init function.
1639 static int kprobe_register(struct trace_event_call *event,
1640 enum trace_reg type, void *data)
1642 struct trace_event_file *file = data;
1645 case TRACE_REG_REGISTER:
1646 return enable_trace_kprobe(event, file);
1647 case TRACE_REG_UNREGISTER:
1648 return disable_trace_kprobe(event, file);
1650 #ifdef CONFIG_PERF_EVENTS
1651 case TRACE_REG_PERF_REGISTER:
1652 return enable_trace_kprobe(event, NULL);
1653 case TRACE_REG_PERF_UNREGISTER:
1654 return disable_trace_kprobe(event, NULL);
1655 case TRACE_REG_PERF_OPEN:
1656 case TRACE_REG_PERF_CLOSE:
1657 case TRACE_REG_PERF_ADD:
1658 case TRACE_REG_PERF_DEL:
1665 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1667 struct trace_kprobe *tk = container_of(kp, struct trace_kprobe, rp.kp);
1670 raw_cpu_inc(*tk->nhit);
1672 if (trace_probe_test_flag(&tk->tp, TP_FLAG_TRACE))
1673 kprobe_trace_func(tk, regs);
1674 #ifdef CONFIG_PERF_EVENTS
1675 if (trace_probe_test_flag(&tk->tp, TP_FLAG_PROFILE))
1676 ret = kprobe_perf_func(tk, regs);
1680 NOKPROBE_SYMBOL(kprobe_dispatcher);
1683 kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1685 struct kretprobe *rp = get_kretprobe(ri);
1686 struct trace_kprobe *tk;
1689 * There is a small chance that get_kretprobe(ri) returns NULL when
1690 * the kretprobe is unregister on another CPU between kretprobe's
1691 * trampoline_handler and this function.
1696 tk = container_of(rp, struct trace_kprobe, rp);
1697 raw_cpu_inc(*tk->nhit);
1699 if (trace_probe_test_flag(&tk->tp, TP_FLAG_TRACE))
1700 kretprobe_trace_func(tk, ri, regs);
1701 #ifdef CONFIG_PERF_EVENTS
1702 if (trace_probe_test_flag(&tk->tp, TP_FLAG_PROFILE))
1703 kretprobe_perf_func(tk, ri, regs);
1705 return 0; /* We don't tweak kernel, so just return 0 */
1707 NOKPROBE_SYMBOL(kretprobe_dispatcher);
1709 static struct trace_event_functions kretprobe_funcs = {
1710 .trace = print_kretprobe_event
1713 static struct trace_event_functions kprobe_funcs = {
1714 .trace = print_kprobe_event
1717 static struct trace_event_fields kretprobe_fields_array[] = {
1718 { .type = TRACE_FUNCTION_TYPE,
1719 .define_fields = kretprobe_event_define_fields },
1723 static struct trace_event_fields kprobe_fields_array[] = {
1724 { .type = TRACE_FUNCTION_TYPE,
1725 .define_fields = kprobe_event_define_fields },
1729 static inline void init_trace_event_call(struct trace_kprobe *tk)
1731 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1733 if (trace_kprobe_is_return(tk)) {
1734 call->event.funcs = &kretprobe_funcs;
1735 call->class->fields_array = kretprobe_fields_array;
1737 call->event.funcs = &kprobe_funcs;
1738 call->class->fields_array = kprobe_fields_array;
1741 call->flags = TRACE_EVENT_FL_KPROBE;
1742 call->class->reg = kprobe_register;
1745 static int register_kprobe_event(struct trace_kprobe *tk)
1747 init_trace_event_call(tk);
1749 return trace_probe_register_event_call(&tk->tp);
1752 static int unregister_kprobe_event(struct trace_kprobe *tk)
1754 return trace_probe_unregister_event_call(&tk->tp);
1757 #ifdef CONFIG_PERF_EVENTS
1758 /* create a trace_kprobe, but don't add it to global lists */
1759 struct trace_event_call *
1760 create_local_trace_kprobe(char *func, void *addr, unsigned long offs,
1763 enum probe_print_type ptype;
1764 struct trace_kprobe *tk;
1769 * local trace_kprobes are not added to dyn_event, so they are never
1770 * searched in find_trace_kprobe(). Therefore, there is no concern of
1771 * duplicated name here.
1773 event = func ? func : "DUMMY_EVENT";
1775 tk = alloc_trace_kprobe(KPROBE_EVENT_SYSTEM, event, (void *)addr, func,
1776 offs, 0 /* maxactive */, 0 /* nargs */,
1780 pr_info("Failed to allocate trace_probe.(%d)\n",
1782 return ERR_CAST(tk);
1785 init_trace_event_call(tk);
1787 ptype = trace_kprobe_is_return(tk) ?
1788 PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL;
1789 if (traceprobe_set_print_fmt(&tk->tp, ptype) < 0) {
1794 ret = __register_trace_kprobe(tk);
1798 return trace_probe_event_call(&tk->tp);
1800 free_trace_kprobe(tk);
1801 return ERR_PTR(ret);
1804 void destroy_local_trace_kprobe(struct trace_event_call *event_call)
1806 struct trace_kprobe *tk;
1808 tk = trace_kprobe_primary_from_call(event_call);
1812 if (trace_probe_is_enabled(&tk->tp)) {
1817 __unregister_trace_kprobe(tk);
1819 free_trace_kprobe(tk);
1821 #endif /* CONFIG_PERF_EVENTS */
1823 static __init void enable_boot_kprobe_events(void)
1825 struct trace_array *tr = top_trace_array();
1826 struct trace_event_file *file;
1827 struct trace_kprobe *tk;
1828 struct dyn_event *pos;
1830 mutex_lock(&event_mutex);
1831 for_each_trace_kprobe(tk, pos) {
1832 list_for_each_entry(file, &tr->events, list)
1833 if (file->event_call == trace_probe_event_call(&tk->tp))
1834 trace_event_enable_disable(file, 1, 0);
1836 mutex_unlock(&event_mutex);
1839 static __init void setup_boot_kprobe_events(void)
1841 char *p, *cmd = kprobe_boot_events_buf;
1844 strreplace(kprobe_boot_events_buf, ',', ' ');
1846 while (cmd && *cmd != '\0') {
1847 p = strchr(cmd, ';');
1851 ret = create_or_delete_trace_kprobe(cmd);
1853 pr_warn("Failed to add event(%d): %s\n", ret, cmd);
1858 enable_boot_kprobe_events();
1862 * Register dynevent at core_initcall. This allows kernel to setup kprobe
1863 * events in postcore_initcall without tracefs.
1865 static __init int init_kprobe_trace_early(void)
1869 ret = dyn_event_register(&trace_kprobe_ops);
1873 if (register_module_notifier(&trace_kprobe_module_nb))
1878 core_initcall(init_kprobe_trace_early);
1880 /* Make a tracefs interface for controlling probe points */
1881 static __init int init_kprobe_trace(void)
1884 struct dentry *entry;
1886 ret = tracing_init_dentry();
1890 entry = tracefs_create_file("kprobe_events", TRACE_MODE_WRITE,
1891 NULL, NULL, &kprobe_events_ops);
1893 /* Event list interface */
1895 pr_warn("Could not create tracefs 'kprobe_events' entry\n");
1897 /* Profile interface */
1898 entry = tracefs_create_file("kprobe_profile", TRACE_MODE_READ,
1899 NULL, NULL, &kprobe_profile_ops);
1902 pr_warn("Could not create tracefs 'kprobe_profile' entry\n");
1904 setup_boot_kprobe_events();
1908 fs_initcall(init_kprobe_trace);
1911 #ifdef CONFIG_FTRACE_STARTUP_TEST
1912 static __init struct trace_event_file *
1913 find_trace_probe_file(struct trace_kprobe *tk, struct trace_array *tr)
1915 struct trace_event_file *file;
1917 list_for_each_entry(file, &tr->events, list)
1918 if (file->event_call == trace_probe_event_call(&tk->tp))
1925 * Nobody but us can call enable_trace_kprobe/disable_trace_kprobe at this
1926 * stage, we can do this lockless.
1928 static __init int kprobe_trace_self_tests_init(void)
1931 int (*target)(int, int, int, int, int, int);
1932 struct trace_kprobe *tk;
1933 struct trace_event_file *file;
1935 if (tracing_is_disabled())
1938 if (tracing_selftest_disabled)
1941 target = kprobe_trace_selftest_target;
1943 pr_info("Testing kprobe tracing: ");
1945 ret = create_or_delete_trace_kprobe("p:testprobe kprobe_trace_selftest_target $stack $stack0 +0($stack)");
1946 if (WARN_ON_ONCE(ret)) {
1947 pr_warn("error on probing function entry.\n");
1950 /* Enable trace point */
1951 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1952 if (WARN_ON_ONCE(tk == NULL)) {
1953 pr_warn("error on getting new probe.\n");
1956 file = find_trace_probe_file(tk, top_trace_array());
1957 if (WARN_ON_ONCE(file == NULL)) {
1958 pr_warn("error on getting probe file.\n");
1961 enable_trace_kprobe(
1962 trace_probe_event_call(&tk->tp), file);
1966 ret = create_or_delete_trace_kprobe("r:testprobe2 kprobe_trace_selftest_target $retval");
1967 if (WARN_ON_ONCE(ret)) {
1968 pr_warn("error on probing function return.\n");
1971 /* Enable trace point */
1972 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1973 if (WARN_ON_ONCE(tk == NULL)) {
1974 pr_warn("error on getting 2nd new probe.\n");
1977 file = find_trace_probe_file(tk, top_trace_array());
1978 if (WARN_ON_ONCE(file == NULL)) {
1979 pr_warn("error on getting probe file.\n");
1982 enable_trace_kprobe(
1983 trace_probe_event_call(&tk->tp), file);
1990 ret = target(1, 2, 3, 4, 5, 6);
1993 * Not expecting an error here, the check is only to prevent the
1994 * optimizer from removing the call to target() as otherwise there
1995 * are no side-effects and the call is never performed.
2000 /* Disable trace points before removing it */
2001 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
2002 if (WARN_ON_ONCE(tk == NULL)) {
2003 pr_warn("error on getting test probe.\n");
2006 if (trace_kprobe_nhit(tk) != 1) {
2007 pr_warn("incorrect number of testprobe hits\n");
2011 file = find_trace_probe_file(tk, top_trace_array());
2012 if (WARN_ON_ONCE(file == NULL)) {
2013 pr_warn("error on getting probe file.\n");
2016 disable_trace_kprobe(
2017 trace_probe_event_call(&tk->tp), file);
2020 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
2021 if (WARN_ON_ONCE(tk == NULL)) {
2022 pr_warn("error on getting 2nd test probe.\n");
2025 if (trace_kprobe_nhit(tk) != 1) {
2026 pr_warn("incorrect number of testprobe2 hits\n");
2030 file = find_trace_probe_file(tk, top_trace_array());
2031 if (WARN_ON_ONCE(file == NULL)) {
2032 pr_warn("error on getting probe file.\n");
2035 disable_trace_kprobe(
2036 trace_probe_event_call(&tk->tp), file);
2039 ret = create_or_delete_trace_kprobe("-:testprobe");
2040 if (WARN_ON_ONCE(ret)) {
2041 pr_warn("error on deleting a probe.\n");
2045 ret = create_or_delete_trace_kprobe("-:testprobe2");
2046 if (WARN_ON_ONCE(ret)) {
2047 pr_warn("error on deleting a probe.\n");
2052 ret = dyn_events_release_all(&trace_kprobe_ops);
2053 if (WARN_ON_ONCE(ret)) {
2054 pr_warn("error on cleaning up probes.\n");
2058 * Wait for the optimizer work to finish. Otherwise it might fiddle
2059 * with probes in already freed __init text.
2061 wait_for_kprobe_optimizer();
2063 pr_cont("NG: Some tests are failed. Please check them.\n");
2069 late_initcall(kprobe_trace_self_tests_init);