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"
23 #define KPROBE_EVENT_SYSTEM "kprobes"
24 #define KRETPROBE_MAXACTIVE_MAX 4096
26 /* Kprobe early definition from command line */
27 static char kprobe_boot_events_buf[COMMAND_LINE_SIZE] __initdata;
29 static int __init set_kprobe_boot_events(char *str)
31 strlcpy(kprobe_boot_events_buf, str, COMMAND_LINE_SIZE);
32 disable_tracing_selftest("running kprobe events");
36 __setup("kprobe_event=", set_kprobe_boot_events);
38 static int trace_kprobe_create(const char *raw_command);
39 static int trace_kprobe_show(struct seq_file *m, struct dyn_event *ev);
40 static int trace_kprobe_release(struct dyn_event *ev);
41 static bool trace_kprobe_is_busy(struct dyn_event *ev);
42 static bool trace_kprobe_match(const char *system, const char *event,
43 int argc, const char **argv, struct dyn_event *ev);
45 static struct dyn_event_operations trace_kprobe_ops = {
46 .create = trace_kprobe_create,
47 .show = trace_kprobe_show,
48 .is_busy = trace_kprobe_is_busy,
49 .free = trace_kprobe_release,
50 .match = trace_kprobe_match,
54 * Kprobe event core functions
57 struct dyn_event devent;
58 struct kretprobe rp; /* Use rp.kp for kprobe use */
59 unsigned long __percpu *nhit;
60 const char *symbol; /* symbol name */
61 struct trace_probe tp;
64 static bool is_trace_kprobe(struct dyn_event *ev)
66 return ev->ops == &trace_kprobe_ops;
69 static struct trace_kprobe *to_trace_kprobe(struct dyn_event *ev)
71 return container_of(ev, struct trace_kprobe, devent);
75 * for_each_trace_kprobe - iterate over the trace_kprobe list
76 * @pos: the struct trace_kprobe * for each entry
77 * @dpos: the struct dyn_event * to use as a loop cursor
79 #define for_each_trace_kprobe(pos, dpos) \
80 for_each_dyn_event(dpos) \
81 if (is_trace_kprobe(dpos) && (pos = to_trace_kprobe(dpos)))
83 static nokprobe_inline bool trace_kprobe_is_return(struct trace_kprobe *tk)
85 return tk->rp.handler != NULL;
88 static nokprobe_inline const char *trace_kprobe_symbol(struct trace_kprobe *tk)
90 return tk->symbol ? tk->symbol : "unknown";
93 static nokprobe_inline unsigned long trace_kprobe_offset(struct trace_kprobe *tk)
95 return tk->rp.kp.offset;
98 static nokprobe_inline bool trace_kprobe_has_gone(struct trace_kprobe *tk)
100 return !!(kprobe_gone(&tk->rp.kp));
103 static nokprobe_inline bool trace_kprobe_within_module(struct trace_kprobe *tk,
106 int len = strlen(module_name(mod));
107 const char *name = trace_kprobe_symbol(tk);
109 return strncmp(module_name(mod), name, len) == 0 && name[len] == ':';
112 static nokprobe_inline bool trace_kprobe_module_exist(struct trace_kprobe *tk)
119 p = strchr(tk->symbol, ':');
123 rcu_read_lock_sched();
124 ret = !!find_module(tk->symbol);
125 rcu_read_unlock_sched();
131 static bool trace_kprobe_is_busy(struct dyn_event *ev)
133 struct trace_kprobe *tk = to_trace_kprobe(ev);
135 return trace_probe_is_enabled(&tk->tp);
138 static bool trace_kprobe_match_command_head(struct trace_kprobe *tk,
139 int argc, const char **argv)
141 char buf[MAX_ARGSTR_LEN + 1];
147 snprintf(buf, sizeof(buf), "0x%p", tk->rp.kp.addr);
148 else if (tk->rp.kp.offset)
149 snprintf(buf, sizeof(buf), "%s+%u",
150 trace_kprobe_symbol(tk), tk->rp.kp.offset);
152 snprintf(buf, sizeof(buf), "%s", trace_kprobe_symbol(tk));
153 if (strcmp(buf, argv[0]))
157 return trace_probe_match_command_args(&tk->tp, argc, argv);
160 static bool trace_kprobe_match(const char *system, const char *event,
161 int argc, const char **argv, struct dyn_event *ev)
163 struct trace_kprobe *tk = to_trace_kprobe(ev);
165 return strcmp(trace_probe_name(&tk->tp), event) == 0 &&
166 (!system || strcmp(trace_probe_group_name(&tk->tp), system) == 0) &&
167 trace_kprobe_match_command_head(tk, argc, argv);
170 static nokprobe_inline unsigned long trace_kprobe_nhit(struct trace_kprobe *tk)
172 unsigned long nhit = 0;
175 for_each_possible_cpu(cpu)
176 nhit += *per_cpu_ptr(tk->nhit, cpu);
181 static nokprobe_inline bool trace_kprobe_is_registered(struct trace_kprobe *tk)
183 return !(list_empty(&tk->rp.kp.list) &&
184 hlist_unhashed(&tk->rp.kp.hlist));
187 /* Return 0 if it fails to find the symbol address */
188 static nokprobe_inline
189 unsigned long trace_kprobe_address(struct trace_kprobe *tk)
194 addr = (unsigned long)
195 kallsyms_lookup_name(trace_kprobe_symbol(tk));
197 addr += tk->rp.kp.offset;
199 addr = (unsigned long)tk->rp.kp.addr;
204 static nokprobe_inline struct trace_kprobe *
205 trace_kprobe_primary_from_call(struct trace_event_call *call)
207 struct trace_probe *tp;
209 tp = trace_probe_primary_from_call(call);
210 if (WARN_ON_ONCE(!tp))
213 return container_of(tp, struct trace_kprobe, tp);
216 bool trace_kprobe_on_func_entry(struct trace_event_call *call)
218 struct trace_kprobe *tk = trace_kprobe_primary_from_call(call);
220 return tk ? (kprobe_on_func_entry(tk->rp.kp.addr,
221 tk->rp.kp.addr ? NULL : tk->rp.kp.symbol_name,
222 tk->rp.kp.addr ? 0 : tk->rp.kp.offset) == 0) : false;
225 bool trace_kprobe_error_injectable(struct trace_event_call *call)
227 struct trace_kprobe *tk = trace_kprobe_primary_from_call(call);
229 return tk ? within_error_injection_list(trace_kprobe_address(tk)) :
233 static int register_kprobe_event(struct trace_kprobe *tk);
234 static int unregister_kprobe_event(struct trace_kprobe *tk);
236 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
237 static int kretprobe_dispatcher(struct kretprobe_instance *ri,
238 struct pt_regs *regs);
240 static void free_trace_kprobe(struct trace_kprobe *tk)
243 trace_probe_cleanup(&tk->tp);
245 free_percpu(tk->nhit);
251 * Allocate new trace_probe and initialize it (including kprobes).
253 static struct trace_kprobe *alloc_trace_kprobe(const char *group,
259 int nargs, bool is_return)
261 struct trace_kprobe *tk;
264 tk = kzalloc(struct_size(tk, tp.args, nargs), GFP_KERNEL);
268 tk->nhit = alloc_percpu(unsigned long);
273 tk->symbol = kstrdup(symbol, GFP_KERNEL);
276 tk->rp.kp.symbol_name = tk->symbol;
277 tk->rp.kp.offset = offs;
279 tk->rp.kp.addr = addr;
282 tk->rp.handler = kretprobe_dispatcher;
284 tk->rp.kp.pre_handler = kprobe_dispatcher;
286 tk->rp.maxactive = maxactive;
287 INIT_HLIST_NODE(&tk->rp.kp.hlist);
288 INIT_LIST_HEAD(&tk->rp.kp.list);
290 ret = trace_probe_init(&tk->tp, event, group, false);
294 dyn_event_init(&tk->devent, &trace_kprobe_ops);
297 free_trace_kprobe(tk);
301 static struct trace_kprobe *find_trace_kprobe(const char *event,
304 struct dyn_event *pos;
305 struct trace_kprobe *tk;
307 for_each_trace_kprobe(tk, pos)
308 if (strcmp(trace_probe_name(&tk->tp), event) == 0 &&
309 strcmp(trace_probe_group_name(&tk->tp), group) == 0)
314 static inline int __enable_trace_kprobe(struct trace_kprobe *tk)
318 if (trace_kprobe_is_registered(tk) && !trace_kprobe_has_gone(tk)) {
319 if (trace_kprobe_is_return(tk))
320 ret = enable_kretprobe(&tk->rp);
322 ret = enable_kprobe(&tk->rp.kp);
328 static void __disable_trace_kprobe(struct trace_probe *tp)
330 struct trace_probe *pos;
331 struct trace_kprobe *tk;
333 list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
334 tk = container_of(pos, struct trace_kprobe, tp);
335 if (!trace_kprobe_is_registered(tk))
337 if (trace_kprobe_is_return(tk))
338 disable_kretprobe(&tk->rp);
340 disable_kprobe(&tk->rp.kp);
346 * if the file is NULL, enable "perf" handler, or enable "trace" handler.
348 static int enable_trace_kprobe(struct trace_event_call *call,
349 struct trace_event_file *file)
351 struct trace_probe *pos, *tp;
352 struct trace_kprobe *tk;
356 tp = trace_probe_primary_from_call(call);
357 if (WARN_ON_ONCE(!tp))
359 enabled = trace_probe_is_enabled(tp);
361 /* This also changes "enabled" state */
363 ret = trace_probe_add_file(tp, file);
367 trace_probe_set_flag(tp, TP_FLAG_PROFILE);
372 list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
373 tk = container_of(pos, struct trace_kprobe, tp);
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;
562 struct trace_probe *pos;
565 list_for_each_entry(pos, &tpe->probes, list) {
566 orig = container_of(pos, struct trace_kprobe, tp);
567 if (strcmp(trace_kprobe_symbol(orig),
568 trace_kprobe_symbol(comp)) ||
569 trace_kprobe_offset(orig) != trace_kprobe_offset(comp))
573 * trace_probe_compare_arg_type() ensured that nr_args and
574 * each argument name and type are same. Let's compare comm.
576 for (i = 0; i < orig->tp.nr_args; i++) {
577 if (strcmp(orig->tp.args[i].comm,
578 comp->tp.args[i].comm))
582 if (i == orig->tp.nr_args)
589 static int append_trace_kprobe(struct trace_kprobe *tk, struct trace_kprobe *to)
593 ret = trace_probe_compare_arg_type(&tk->tp, &to->tp);
595 /* Note that argument starts index = 2 */
596 trace_probe_log_set_index(ret + 1);
597 trace_probe_log_err(0, DIFF_ARG_TYPE);
600 if (trace_kprobe_has_same_kprobe(to, tk)) {
601 trace_probe_log_set_index(0);
602 trace_probe_log_err(0, SAME_PROBE);
606 /* Append to existing event */
607 ret = trace_probe_append(&tk->tp, &to->tp);
611 /* Register k*probe */
612 ret = __register_trace_kprobe(tk);
613 if (ret == -ENOENT && !trace_kprobe_module_exist(tk)) {
614 pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
619 trace_probe_unlink(&tk->tp);
621 dyn_event_add(&tk->devent, trace_probe_event_call(&tk->tp));
626 /* Register a trace_probe and probe_event */
627 static int register_trace_kprobe(struct trace_kprobe *tk)
629 struct trace_kprobe *old_tk;
632 mutex_lock(&event_mutex);
634 old_tk = find_trace_kprobe(trace_probe_name(&tk->tp),
635 trace_probe_group_name(&tk->tp));
637 if (trace_kprobe_is_return(tk) != trace_kprobe_is_return(old_tk)) {
638 trace_probe_log_set_index(0);
639 trace_probe_log_err(0, DIFF_PROBE_TYPE);
642 ret = append_trace_kprobe(tk, old_tk);
647 /* Register new event */
648 ret = register_kprobe_event(tk);
650 if (ret == -EEXIST) {
651 trace_probe_log_set_index(0);
652 trace_probe_log_err(0, EVENT_EXIST);
654 pr_warn("Failed to register probe event(%d)\n", ret);
658 /* Register k*probe */
659 ret = __register_trace_kprobe(tk);
660 if (ret == -ENOENT && !trace_kprobe_module_exist(tk)) {
661 pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
666 unregister_kprobe_event(tk);
668 dyn_event_add(&tk->devent, trace_probe_event_call(&tk->tp));
671 mutex_unlock(&event_mutex);
675 /* Module notifier call back, checking event on the module */
676 static int trace_kprobe_module_callback(struct notifier_block *nb,
677 unsigned long val, void *data)
679 struct module *mod = data;
680 struct dyn_event *pos;
681 struct trace_kprobe *tk;
684 if (val != MODULE_STATE_COMING)
687 /* Update probes on coming module */
688 mutex_lock(&event_mutex);
689 for_each_trace_kprobe(tk, pos) {
690 if (trace_kprobe_within_module(tk, mod)) {
691 /* Don't need to check busy - this should have gone. */
692 __unregister_trace_kprobe(tk);
693 ret = __register_trace_kprobe(tk);
695 pr_warn("Failed to re-register probe %s on %s: %d\n",
696 trace_probe_name(&tk->tp),
697 module_name(mod), ret);
700 mutex_unlock(&event_mutex);
705 static struct notifier_block trace_kprobe_module_nb = {
706 .notifier_call = trace_kprobe_module_callback,
707 .priority = 1 /* Invoked after kprobe module callback */
710 static int __trace_kprobe_create(int argc, const char *argv[])
715 * p[:[GRP/]EVENT] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
717 * r[MAXACTIVE][:[GRP/]EVENT] [MOD:]KSYM[+0] [FETCHARGS]
719 * p:[GRP/]EVENT] [MOD:]KSYM[+0]%return [FETCHARGS]
722 * $retval : fetch return value
723 * $stack : fetch stack address
724 * $stackN : fetch Nth of stack (N:0-)
725 * $comm : fetch current task comm
726 * @ADDR : fetch memory at ADDR (ADDR should be in kernel)
727 * @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
728 * %REG : fetch register REG
729 * Dereferencing memory fetch:
730 * +|-offs(ARG) : fetch memory at ARG +|- offs address.
731 * Alias name of args:
732 * NAME=FETCHARG : set NAME as alias of FETCHARG.
734 * FETCHARG:TYPE : use TYPE instead of unsigned long.
736 struct trace_kprobe *tk = NULL;
738 bool is_return = false;
739 char *symbol = NULL, *tmp = NULL;
740 const char *event = NULL, *group = KPROBE_EVENT_SYSTEM;
741 enum probe_print_type ptype;
745 char buf[MAX_EVENT_NAME_LEN];
746 unsigned int flags = TPARG_FL_KERNEL;
748 switch (argv[0][0]) {
760 trace_probe_log_init("trace_kprobe", argc, argv);
762 event = strchr(&argv[0][1], ':');
766 if (isdigit(argv[0][1])) {
768 trace_probe_log_err(1, MAXACT_NO_KPROBE);
772 len = event - &argv[0][1] - 1;
774 len = strlen(&argv[0][1]);
775 if (len > MAX_EVENT_NAME_LEN - 1) {
776 trace_probe_log_err(1, BAD_MAXACT);
779 memcpy(buf, &argv[0][1], len);
781 ret = kstrtouint(buf, 0, &maxactive);
782 if (ret || !maxactive) {
783 trace_probe_log_err(1, BAD_MAXACT);
786 /* kretprobes instances are iterated over via a list. The
787 * maximum should stay reasonable.
789 if (maxactive > KRETPROBE_MAXACTIVE_MAX) {
790 trace_probe_log_err(1, MAXACT_TOO_BIG);
795 /* try to parse an address. if that fails, try to read the
796 * input as a symbol. */
797 if (kstrtoul(argv[1], 0, (unsigned long *)&addr)) {
798 trace_probe_log_set_index(1);
799 /* Check whether uprobe event specified */
800 if (strchr(argv[1], '/') && strchr(argv[1], ':')) {
804 /* a symbol specified */
805 symbol = kstrdup(argv[1], GFP_KERNEL);
809 tmp = strchr(symbol, '%');
811 if (!strcmp(tmp, "%return")) {
815 trace_probe_log_err(tmp - symbol, BAD_ADDR_SUFFIX);
820 /* TODO: support .init module functions */
821 ret = traceprobe_split_symbol_offset(symbol, &offset);
822 if (ret || offset < 0 || offset > UINT_MAX) {
823 trace_probe_log_err(0, BAD_PROBE_ADDR);
827 flags |= TPARG_FL_RETURN;
828 ret = kprobe_on_func_entry(NULL, symbol, offset);
830 flags |= TPARG_FL_FENTRY;
831 /* Defer the ENOENT case until register kprobe */
832 if (ret == -EINVAL && is_return) {
833 trace_probe_log_err(0, BAD_RETPROBE);
838 trace_probe_log_set_index(0);
840 ret = traceprobe_parse_event_name(&event, &group, buf,
845 /* Make a new event name */
847 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
848 is_return ? 'r' : 'p', symbol, offset);
850 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
851 is_return ? 'r' : 'p', addr);
852 sanitize_event_name(buf);
857 tk = alloc_trace_kprobe(group, event, addr, symbol, offset, maxactive,
858 argc - 2, is_return);
861 /* This must return -ENOMEM, else there is a bug */
862 WARN_ON_ONCE(ret != -ENOMEM);
863 goto out; /* We know tk is not allocated */
865 argc -= 2; argv += 2;
867 /* parse arguments */
868 for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
869 trace_probe_log_set_index(i + 2);
870 ret = traceprobe_parse_probe_arg(&tk->tp, i, argv[i], flags);
872 goto error; /* This can be -ENOMEM */
875 ptype = is_return ? PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL;
876 ret = traceprobe_set_print_fmt(&tk->tp, ptype);
880 ret = register_trace_kprobe(tk);
882 trace_probe_log_set_index(1);
884 trace_probe_log_err(0, BAD_INSN_BNDRY);
885 else if (ret == -ENOENT)
886 trace_probe_log_err(0, BAD_PROBE_ADDR);
887 else if (ret != -ENOMEM && ret != -EEXIST)
888 trace_probe_log_err(0, FAIL_REG_PROBE);
893 trace_probe_log_clear();
900 free_trace_kprobe(tk);
904 static int trace_kprobe_create(const char *raw_command)
906 return trace_probe_create(raw_command, __trace_kprobe_create);
909 static int create_or_delete_trace_kprobe(const char *raw_command)
913 if (raw_command[0] == '-')
914 return dyn_event_release(raw_command, &trace_kprobe_ops);
916 ret = trace_kprobe_create(raw_command);
917 return ret == -ECANCELED ? -EINVAL : ret;
920 static int trace_kprobe_run_command(struct dynevent_cmd *cmd)
922 return create_or_delete_trace_kprobe(cmd->seq.buffer);
926 * kprobe_event_cmd_init - Initialize a kprobe event command object
927 * @cmd: A pointer to the dynevent_cmd struct representing the new event
928 * @buf: A pointer to the buffer used to build the command
929 * @maxlen: The length of the buffer passed in @buf
931 * Initialize a synthetic event command object. Use this before
932 * calling any of the other kprobe_event functions.
934 void kprobe_event_cmd_init(struct dynevent_cmd *cmd, char *buf, int maxlen)
936 dynevent_cmd_init(cmd, buf, maxlen, DYNEVENT_TYPE_KPROBE,
937 trace_kprobe_run_command);
939 EXPORT_SYMBOL_GPL(kprobe_event_cmd_init);
942 * __kprobe_event_gen_cmd_start - Generate a kprobe event command from arg list
943 * @cmd: A pointer to the dynevent_cmd struct representing the new event
944 * @name: The name of the kprobe event
945 * @loc: The location of the kprobe event
946 * @kretprobe: Is this a return probe?
947 * @args: Variable number of arg (pairs), one pair for each field
949 * NOTE: Users normally won't want to call this function directly, but
950 * rather use the kprobe_event_gen_cmd_start() wrapper, which automatically
951 * adds a NULL to the end of the arg list. If this function is used
952 * directly, make sure the last arg in the variable arg list is NULL.
954 * Generate a kprobe event command to be executed by
955 * kprobe_event_gen_cmd_end(). This function can be used to generate the
956 * complete command or only the first part of it; in the latter case,
957 * kprobe_event_add_fields() can be used to add more fields following this.
959 * Unlikely the synth_event_gen_cmd_start(), @loc must be specified. This
960 * returns -EINVAL if @loc == NULL.
962 * Return: 0 if successful, error otherwise.
964 int __kprobe_event_gen_cmd_start(struct dynevent_cmd *cmd, bool kretprobe,
965 const char *name, const char *loc, ...)
967 char buf[MAX_EVENT_NAME_LEN];
968 struct dynevent_arg arg;
972 if (cmd->type != DYNEVENT_TYPE_KPROBE)
979 snprintf(buf, MAX_EVENT_NAME_LEN, "r:kprobes/%s", name);
981 snprintf(buf, MAX_EVENT_NAME_LEN, "p:kprobes/%s", name);
983 ret = dynevent_str_add(cmd, buf);
987 dynevent_arg_init(&arg, 0);
989 ret = dynevent_arg_add(cmd, &arg, NULL);
997 field = va_arg(args, const char *);
1001 if (++cmd->n_fields > MAX_TRACE_ARGS) {
1007 ret = dynevent_arg_add(cmd, &arg, NULL);
1015 EXPORT_SYMBOL_GPL(__kprobe_event_gen_cmd_start);
1018 * __kprobe_event_add_fields - Add probe fields to a kprobe command from arg list
1019 * @cmd: A pointer to the dynevent_cmd struct representing the new event
1020 * @args: Variable number of arg (pairs), one pair for each field
1022 * NOTE: Users normally won't want to call this function directly, but
1023 * rather use the kprobe_event_add_fields() wrapper, which
1024 * automatically adds a NULL to the end of the arg list. If this
1025 * function is used directly, make sure the last arg in the variable
1028 * Add probe fields to an existing kprobe command using a variable
1029 * list of args. Fields are added in the same order they're listed.
1031 * Return: 0 if successful, error otherwise.
1033 int __kprobe_event_add_fields(struct dynevent_cmd *cmd, ...)
1035 struct dynevent_arg arg;
1039 if (cmd->type != DYNEVENT_TYPE_KPROBE)
1042 dynevent_arg_init(&arg, 0);
1044 va_start(args, cmd);
1048 field = va_arg(args, const char *);
1052 if (++cmd->n_fields > MAX_TRACE_ARGS) {
1058 ret = dynevent_arg_add(cmd, &arg, NULL);
1066 EXPORT_SYMBOL_GPL(__kprobe_event_add_fields);
1069 * kprobe_event_delete - Delete a kprobe event
1070 * @name: The name of the kprobe event to delete
1072 * Delete a kprobe event with the give @name from kernel code rather
1073 * than directly from the command line.
1075 * Return: 0 if successful, error otherwise.
1077 int kprobe_event_delete(const char *name)
1079 char buf[MAX_EVENT_NAME_LEN];
1081 snprintf(buf, MAX_EVENT_NAME_LEN, "-:%s", name);
1083 return create_or_delete_trace_kprobe(buf);
1085 EXPORT_SYMBOL_GPL(kprobe_event_delete);
1087 static int trace_kprobe_release(struct dyn_event *ev)
1089 struct trace_kprobe *tk = to_trace_kprobe(ev);
1090 int ret = unregister_trace_kprobe(tk);
1093 free_trace_kprobe(tk);
1097 static int trace_kprobe_show(struct seq_file *m, struct dyn_event *ev)
1099 struct trace_kprobe *tk = to_trace_kprobe(ev);
1102 seq_putc(m, trace_kprobe_is_return(tk) ? 'r' : 'p');
1103 if (trace_kprobe_is_return(tk) && tk->rp.maxactive)
1104 seq_printf(m, "%d", tk->rp.maxactive);
1105 seq_printf(m, ":%s/%s", trace_probe_group_name(&tk->tp),
1106 trace_probe_name(&tk->tp));
1109 seq_printf(m, " 0x%p", tk->rp.kp.addr);
1110 else if (tk->rp.kp.offset)
1111 seq_printf(m, " %s+%u", trace_kprobe_symbol(tk),
1114 seq_printf(m, " %s", trace_kprobe_symbol(tk));
1116 for (i = 0; i < tk->tp.nr_args; i++)
1117 seq_printf(m, " %s=%s", tk->tp.args[i].name, tk->tp.args[i].comm);
1123 static int probes_seq_show(struct seq_file *m, void *v)
1125 struct dyn_event *ev = v;
1127 if (!is_trace_kprobe(ev))
1130 return trace_kprobe_show(m, ev);
1133 static const struct seq_operations probes_seq_op = {
1134 .start = dyn_event_seq_start,
1135 .next = dyn_event_seq_next,
1136 .stop = dyn_event_seq_stop,
1137 .show = probes_seq_show
1140 static int probes_open(struct inode *inode, struct file *file)
1144 ret = security_locked_down(LOCKDOWN_TRACEFS);
1148 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
1149 ret = dyn_events_release_all(&trace_kprobe_ops);
1154 return seq_open(file, &probes_seq_op);
1157 static ssize_t probes_write(struct file *file, const char __user *buffer,
1158 size_t count, loff_t *ppos)
1160 return trace_parse_run_command(file, buffer, count, ppos,
1161 create_or_delete_trace_kprobe);
1164 static const struct file_operations kprobe_events_ops = {
1165 .owner = THIS_MODULE,
1166 .open = probes_open,
1168 .llseek = seq_lseek,
1169 .release = seq_release,
1170 .write = probes_write,
1173 /* Probes profiling interfaces */
1174 static int probes_profile_seq_show(struct seq_file *m, void *v)
1176 struct dyn_event *ev = v;
1177 struct trace_kprobe *tk;
1179 if (!is_trace_kprobe(ev))
1182 tk = to_trace_kprobe(ev);
1183 seq_printf(m, " %-44s %15lu %15lu\n",
1184 trace_probe_name(&tk->tp),
1185 trace_kprobe_nhit(tk),
1191 static const struct seq_operations profile_seq_op = {
1192 .start = dyn_event_seq_start,
1193 .next = dyn_event_seq_next,
1194 .stop = dyn_event_seq_stop,
1195 .show = probes_profile_seq_show
1198 static int profile_open(struct inode *inode, struct file *file)
1202 ret = security_locked_down(LOCKDOWN_TRACEFS);
1206 return seq_open(file, &profile_seq_op);
1209 static const struct file_operations kprobe_profile_ops = {
1210 .owner = THIS_MODULE,
1211 .open = profile_open,
1213 .llseek = seq_lseek,
1214 .release = seq_release,
1217 /* Kprobe specific fetch functions */
1219 /* Return the length of string -- including null terminal byte */
1220 static nokprobe_inline int
1221 fetch_store_strlen_user(unsigned long addr)
1223 const void __user *uaddr = (__force const void __user *)addr;
1225 return strnlen_user_nofault(uaddr, MAX_STRING_SIZE);
1228 /* Return the length of string -- including null terminal byte */
1229 static nokprobe_inline int
1230 fetch_store_strlen(unsigned long addr)
1235 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
1236 if (addr < TASK_SIZE)
1237 return fetch_store_strlen_user(addr);
1241 ret = copy_from_kernel_nofault(&c, (u8 *)addr + len, 1);
1243 } while (c && ret == 0 && len < MAX_STRING_SIZE);
1245 return (ret < 0) ? ret : len;
1249 * Fetch a null-terminated string from user. Caller MUST set *(u32 *)buf
1250 * with max length and relative data location.
1252 static nokprobe_inline int
1253 fetch_store_string_user(unsigned long addr, void *dest, void *base)
1255 const void __user *uaddr = (__force const void __user *)addr;
1256 int maxlen = get_loc_len(*(u32 *)dest);
1260 if (unlikely(!maxlen))
1263 __dest = get_loc_data(dest, base);
1265 ret = strncpy_from_user_nofault(__dest, uaddr, maxlen);
1267 *(u32 *)dest = make_data_loc(ret, __dest - base);
1273 * Fetch a null-terminated string. Caller MUST set *(u32 *)buf with max
1274 * length and relative data location.
1276 static nokprobe_inline int
1277 fetch_store_string(unsigned long addr, void *dest, void *base)
1279 int maxlen = get_loc_len(*(u32 *)dest);
1283 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
1284 if ((unsigned long)addr < TASK_SIZE)
1285 return fetch_store_string_user(addr, dest, base);
1288 if (unlikely(!maxlen))
1291 __dest = get_loc_data(dest, base);
1294 * Try to get string again, since the string can be changed while
1297 ret = strncpy_from_kernel_nofault(__dest, (void *)addr, maxlen);
1299 *(u32 *)dest = make_data_loc(ret, __dest - base);
1304 static nokprobe_inline int
1305 probe_mem_read_user(void *dest, void *src, size_t size)
1307 const void __user *uaddr = (__force const void __user *)src;
1309 return copy_from_user_nofault(dest, uaddr, size);
1312 static nokprobe_inline int
1313 probe_mem_read(void *dest, void *src, size_t size)
1315 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
1316 if ((unsigned long)src < TASK_SIZE)
1317 return probe_mem_read_user(dest, src, size);
1319 return copy_from_kernel_nofault(dest, src, size);
1322 /* Note that we don't verify it, since the code does not come from user space */
1324 process_fetch_insn(struct fetch_insn *code, void *rec, void *dest,
1327 struct pt_regs *regs = rec;
1331 /* 1st stage: get value from context */
1334 val = regs_get_register(regs, code->param);
1336 case FETCH_OP_STACK:
1337 val = regs_get_kernel_stack_nth(regs, code->param);
1339 case FETCH_OP_STACKP:
1340 val = kernel_stack_pointer(regs);
1342 case FETCH_OP_RETVAL:
1343 val = regs_return_value(regs);
1346 val = code->immediate;
1349 val = (unsigned long)current->comm;
1352 val = (unsigned long)code->data;
1354 #ifdef CONFIG_HAVE_FUNCTION_ARG_ACCESS_API
1356 val = regs_get_kernel_argument(regs, code->param);
1359 case FETCH_NOP_SYMBOL: /* Ignore a place holder */
1367 return process_fetch_insn_bottom(code, val, dest, base);
1369 NOKPROBE_SYMBOL(process_fetch_insn)
1371 /* Kprobe handler */
1372 static nokprobe_inline void
1373 __kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs,
1374 struct trace_event_file *trace_file)
1376 struct kprobe_trace_entry_head *entry;
1377 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1378 struct trace_event_buffer fbuffer;
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);
1392 trace_event_buffer_lock_reserve(&fbuffer.buffer, trace_file,
1394 sizeof(*entry) + tk->tp.size + dsize,
1399 fbuffer.regs = regs;
1400 entry = fbuffer.entry = ring_buffer_event_data(fbuffer.event);
1401 entry->ip = (unsigned long)tk->rp.kp.addr;
1402 store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1404 trace_event_buffer_commit(&fbuffer);
1408 kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs)
1410 struct event_file_link *link;
1412 trace_probe_for_each_link_rcu(link, &tk->tp)
1413 __kprobe_trace_func(tk, regs, link->file);
1415 NOKPROBE_SYMBOL(kprobe_trace_func);
1417 /* Kretprobe handler */
1418 static nokprobe_inline void
1419 __kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1420 struct pt_regs *regs,
1421 struct trace_event_file *trace_file)
1423 struct kretprobe_trace_entry_head *entry;
1424 struct trace_event_buffer fbuffer;
1425 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1428 WARN_ON(call != trace_file->event_call);
1430 if (trace_trigger_soft_disabled(trace_file))
1433 fbuffer.trace_ctx = tracing_gen_ctx();
1434 fbuffer.trace_file = trace_file;
1436 dsize = __get_data_size(&tk->tp, regs);
1438 trace_event_buffer_lock_reserve(&fbuffer.buffer, trace_file,
1440 sizeof(*entry) + tk->tp.size + dsize,
1445 fbuffer.regs = regs;
1446 entry = fbuffer.entry = ring_buffer_event_data(fbuffer.event);
1447 entry->func = (unsigned long)tk->rp.kp.addr;
1448 entry->ret_ip = (unsigned long)ri->ret_addr;
1449 store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1451 trace_event_buffer_commit(&fbuffer);
1455 kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1456 struct pt_regs *regs)
1458 struct event_file_link *link;
1460 trace_probe_for_each_link_rcu(link, &tk->tp)
1461 __kretprobe_trace_func(tk, ri, regs, link->file);
1463 NOKPROBE_SYMBOL(kretprobe_trace_func);
1465 /* Event entry printers */
1466 static enum print_line_t
1467 print_kprobe_event(struct trace_iterator *iter, int flags,
1468 struct trace_event *event)
1470 struct kprobe_trace_entry_head *field;
1471 struct trace_seq *s = &iter->seq;
1472 struct trace_probe *tp;
1474 field = (struct kprobe_trace_entry_head *)iter->ent;
1475 tp = trace_probe_primary_from_call(
1476 container_of(event, struct trace_event_call, event));
1477 if (WARN_ON_ONCE(!tp))
1480 trace_seq_printf(s, "%s: (", trace_probe_name(tp));
1482 if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
1485 trace_seq_putc(s, ')');
1487 if (print_probe_args(s, tp->args, tp->nr_args,
1488 (u8 *)&field[1], field) < 0)
1491 trace_seq_putc(s, '\n');
1493 return trace_handle_return(s);
1496 static enum print_line_t
1497 print_kretprobe_event(struct trace_iterator *iter, int flags,
1498 struct trace_event *event)
1500 struct kretprobe_trace_entry_head *field;
1501 struct trace_seq *s = &iter->seq;
1502 struct trace_probe *tp;
1504 field = (struct kretprobe_trace_entry_head *)iter->ent;
1505 tp = trace_probe_primary_from_call(
1506 container_of(event, struct trace_event_call, event));
1507 if (WARN_ON_ONCE(!tp))
1510 trace_seq_printf(s, "%s: (", trace_probe_name(tp));
1512 if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1515 trace_seq_puts(s, " <- ");
1517 if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1520 trace_seq_putc(s, ')');
1522 if (print_probe_args(s, tp->args, tp->nr_args,
1523 (u8 *)&field[1], field) < 0)
1526 trace_seq_putc(s, '\n');
1529 return trace_handle_return(s);
1533 static int kprobe_event_define_fields(struct trace_event_call *event_call)
1536 struct kprobe_trace_entry_head field;
1537 struct trace_probe *tp;
1539 tp = trace_probe_primary_from_call(event_call);
1540 if (WARN_ON_ONCE(!tp))
1543 DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1545 return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
1548 static int kretprobe_event_define_fields(struct trace_event_call *event_call)
1551 struct kretprobe_trace_entry_head field;
1552 struct trace_probe *tp;
1554 tp = trace_probe_primary_from_call(event_call);
1555 if (WARN_ON_ONCE(!tp))
1558 DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1559 DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1561 return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
1564 #ifdef CONFIG_PERF_EVENTS
1566 /* Kprobe profile handler */
1568 kprobe_perf_func(struct trace_kprobe *tk, struct pt_regs *regs)
1570 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1571 struct kprobe_trace_entry_head *entry;
1572 struct hlist_head *head;
1573 int size, __size, dsize;
1576 if (bpf_prog_array_valid(call)) {
1577 unsigned long orig_ip = instruction_pointer(regs);
1580 ret = trace_call_bpf(call, regs);
1583 * We need to check and see if we modified the pc of the
1584 * pt_regs, and if so return 1 so that we don't do the
1587 if (orig_ip != instruction_pointer(regs))
1593 head = this_cpu_ptr(call->perf_events);
1594 if (hlist_empty(head))
1597 dsize = __get_data_size(&tk->tp, regs);
1598 __size = sizeof(*entry) + tk->tp.size + dsize;
1599 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1600 size -= sizeof(u32);
1602 entry = perf_trace_buf_alloc(size, NULL, &rctx);
1606 entry->ip = (unsigned long)tk->rp.kp.addr;
1607 memset(&entry[1], 0, dsize);
1608 store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1609 perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1613 NOKPROBE_SYMBOL(kprobe_perf_func);
1615 /* Kretprobe profile handler */
1617 kretprobe_perf_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1618 struct pt_regs *regs)
1620 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1621 struct kretprobe_trace_entry_head *entry;
1622 struct hlist_head *head;
1623 int size, __size, dsize;
1626 if (bpf_prog_array_valid(call) && !trace_call_bpf(call, regs))
1629 head = this_cpu_ptr(call->perf_events);
1630 if (hlist_empty(head))
1633 dsize = __get_data_size(&tk->tp, regs);
1634 __size = sizeof(*entry) + tk->tp.size + dsize;
1635 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1636 size -= sizeof(u32);
1638 entry = perf_trace_buf_alloc(size, NULL, &rctx);
1642 entry->func = (unsigned long)tk->rp.kp.addr;
1643 entry->ret_ip = (unsigned long)ri->ret_addr;
1644 store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1645 perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1648 NOKPROBE_SYMBOL(kretprobe_perf_func);
1650 int bpf_get_kprobe_info(const struct perf_event *event, u32 *fd_type,
1651 const char **symbol, u64 *probe_offset,
1652 u64 *probe_addr, bool perf_type_tracepoint)
1654 const char *pevent = trace_event_name(event->tp_event);
1655 const char *group = event->tp_event->class->system;
1656 struct trace_kprobe *tk;
1658 if (perf_type_tracepoint)
1659 tk = find_trace_kprobe(pevent, group);
1661 tk = trace_kprobe_primary_from_call(event->tp_event);
1665 *fd_type = trace_kprobe_is_return(tk) ? BPF_FD_TYPE_KRETPROBE
1666 : BPF_FD_TYPE_KPROBE;
1668 *symbol = tk->symbol;
1669 *probe_offset = tk->rp.kp.offset;
1674 *probe_addr = (unsigned long)tk->rp.kp.addr;
1678 #endif /* CONFIG_PERF_EVENTS */
1681 * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex.
1683 * kprobe_trace_self_tests_init() does enable_trace_probe/disable_trace_probe
1684 * lockless, but we can't race with this __init function.
1686 static int kprobe_register(struct trace_event_call *event,
1687 enum trace_reg type, void *data)
1689 struct trace_event_file *file = data;
1692 case TRACE_REG_REGISTER:
1693 return enable_trace_kprobe(event, file);
1694 case TRACE_REG_UNREGISTER:
1695 return disable_trace_kprobe(event, file);
1697 #ifdef CONFIG_PERF_EVENTS
1698 case TRACE_REG_PERF_REGISTER:
1699 return enable_trace_kprobe(event, NULL);
1700 case TRACE_REG_PERF_UNREGISTER:
1701 return disable_trace_kprobe(event, NULL);
1702 case TRACE_REG_PERF_OPEN:
1703 case TRACE_REG_PERF_CLOSE:
1704 case TRACE_REG_PERF_ADD:
1705 case TRACE_REG_PERF_DEL:
1712 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1714 struct trace_kprobe *tk = container_of(kp, struct trace_kprobe, rp.kp);
1717 raw_cpu_inc(*tk->nhit);
1719 if (trace_probe_test_flag(&tk->tp, TP_FLAG_TRACE))
1720 kprobe_trace_func(tk, regs);
1721 #ifdef CONFIG_PERF_EVENTS
1722 if (trace_probe_test_flag(&tk->tp, TP_FLAG_PROFILE))
1723 ret = kprobe_perf_func(tk, regs);
1727 NOKPROBE_SYMBOL(kprobe_dispatcher);
1730 kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1732 struct kretprobe *rp = get_kretprobe(ri);
1733 struct trace_kprobe *tk = container_of(rp, struct trace_kprobe, rp);
1735 raw_cpu_inc(*tk->nhit);
1737 if (trace_probe_test_flag(&tk->tp, TP_FLAG_TRACE))
1738 kretprobe_trace_func(tk, ri, regs);
1739 #ifdef CONFIG_PERF_EVENTS
1740 if (trace_probe_test_flag(&tk->tp, TP_FLAG_PROFILE))
1741 kretprobe_perf_func(tk, ri, regs);
1743 return 0; /* We don't tweak kernel, so just return 0 */
1745 NOKPROBE_SYMBOL(kretprobe_dispatcher);
1747 static struct trace_event_functions kretprobe_funcs = {
1748 .trace = print_kretprobe_event
1751 static struct trace_event_functions kprobe_funcs = {
1752 .trace = print_kprobe_event
1755 static struct trace_event_fields kretprobe_fields_array[] = {
1756 { .type = TRACE_FUNCTION_TYPE,
1757 .define_fields = kretprobe_event_define_fields },
1761 static struct trace_event_fields kprobe_fields_array[] = {
1762 { .type = TRACE_FUNCTION_TYPE,
1763 .define_fields = kprobe_event_define_fields },
1767 static inline void init_trace_event_call(struct trace_kprobe *tk)
1769 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1771 if (trace_kprobe_is_return(tk)) {
1772 call->event.funcs = &kretprobe_funcs;
1773 call->class->fields_array = kretprobe_fields_array;
1775 call->event.funcs = &kprobe_funcs;
1776 call->class->fields_array = kprobe_fields_array;
1779 call->flags = TRACE_EVENT_FL_KPROBE;
1780 call->class->reg = kprobe_register;
1783 static int register_kprobe_event(struct trace_kprobe *tk)
1785 init_trace_event_call(tk);
1787 return trace_probe_register_event_call(&tk->tp);
1790 static int unregister_kprobe_event(struct trace_kprobe *tk)
1792 return trace_probe_unregister_event_call(&tk->tp);
1795 #ifdef CONFIG_PERF_EVENTS
1796 /* create a trace_kprobe, but don't add it to global lists */
1797 struct trace_event_call *
1798 create_local_trace_kprobe(char *func, void *addr, unsigned long offs,
1801 enum probe_print_type ptype;
1802 struct trace_kprobe *tk;
1807 * local trace_kprobes are not added to dyn_event, so they are never
1808 * searched in find_trace_kprobe(). Therefore, there is no concern of
1809 * duplicated name here.
1811 event = func ? func : "DUMMY_EVENT";
1813 tk = alloc_trace_kprobe(KPROBE_EVENT_SYSTEM, event, (void *)addr, func,
1814 offs, 0 /* maxactive */, 0 /* nargs */,
1818 pr_info("Failed to allocate trace_probe.(%d)\n",
1820 return ERR_CAST(tk);
1823 init_trace_event_call(tk);
1825 ptype = trace_kprobe_is_return(tk) ?
1826 PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL;
1827 if (traceprobe_set_print_fmt(&tk->tp, ptype) < 0) {
1832 ret = __register_trace_kprobe(tk);
1836 return trace_probe_event_call(&tk->tp);
1838 free_trace_kprobe(tk);
1839 return ERR_PTR(ret);
1842 void destroy_local_trace_kprobe(struct trace_event_call *event_call)
1844 struct trace_kprobe *tk;
1846 tk = trace_kprobe_primary_from_call(event_call);
1850 if (trace_probe_is_enabled(&tk->tp)) {
1855 __unregister_trace_kprobe(tk);
1857 free_trace_kprobe(tk);
1859 #endif /* CONFIG_PERF_EVENTS */
1861 static __init void enable_boot_kprobe_events(void)
1863 struct trace_array *tr = top_trace_array();
1864 struct trace_event_file *file;
1865 struct trace_kprobe *tk;
1866 struct dyn_event *pos;
1868 mutex_lock(&event_mutex);
1869 for_each_trace_kprobe(tk, pos) {
1870 list_for_each_entry(file, &tr->events, list)
1871 if (file->event_call == trace_probe_event_call(&tk->tp))
1872 trace_event_enable_disable(file, 1, 0);
1874 mutex_unlock(&event_mutex);
1877 static __init void setup_boot_kprobe_events(void)
1879 char *p, *cmd = kprobe_boot_events_buf;
1882 strreplace(kprobe_boot_events_buf, ',', ' ');
1884 while (cmd && *cmd != '\0') {
1885 p = strchr(cmd, ';');
1889 ret = create_or_delete_trace_kprobe(cmd);
1891 pr_warn("Failed to add event(%d): %s\n", ret, cmd);
1896 enable_boot_kprobe_events();
1900 * Register dynevent at core_initcall. This allows kernel to setup kprobe
1901 * events in postcore_initcall without tracefs.
1903 static __init int init_kprobe_trace_early(void)
1907 ret = dyn_event_register(&trace_kprobe_ops);
1911 if (register_module_notifier(&trace_kprobe_module_nb))
1916 core_initcall(init_kprobe_trace_early);
1918 /* Make a tracefs interface for controlling probe points */
1919 static __init int init_kprobe_trace(void)
1922 struct dentry *entry;
1924 ret = tracing_init_dentry();
1928 entry = tracefs_create_file("kprobe_events", 0644, NULL,
1929 NULL, &kprobe_events_ops);
1931 /* Event list interface */
1933 pr_warn("Could not create tracefs 'kprobe_events' entry\n");
1935 /* Profile interface */
1936 entry = tracefs_create_file("kprobe_profile", 0444, NULL,
1937 NULL, &kprobe_profile_ops);
1940 pr_warn("Could not create tracefs 'kprobe_profile' entry\n");
1942 setup_boot_kprobe_events();
1946 fs_initcall(init_kprobe_trace);
1949 #ifdef CONFIG_FTRACE_STARTUP_TEST
1950 static __init struct trace_event_file *
1951 find_trace_probe_file(struct trace_kprobe *tk, struct trace_array *tr)
1953 struct trace_event_file *file;
1955 list_for_each_entry(file, &tr->events, list)
1956 if (file->event_call == trace_probe_event_call(&tk->tp))
1963 * Nobody but us can call enable_trace_kprobe/disable_trace_kprobe at this
1964 * stage, we can do this lockless.
1966 static __init int kprobe_trace_self_tests_init(void)
1969 int (*target)(int, int, int, int, int, int);
1970 struct trace_kprobe *tk;
1971 struct trace_event_file *file;
1973 if (tracing_is_disabled())
1976 if (tracing_selftest_disabled)
1979 target = kprobe_trace_selftest_target;
1981 pr_info("Testing kprobe tracing: ");
1983 ret = create_or_delete_trace_kprobe("p:testprobe kprobe_trace_selftest_target $stack $stack0 +0($stack)");
1984 if (WARN_ON_ONCE(ret)) {
1985 pr_warn("error on probing function entry.\n");
1988 /* Enable trace point */
1989 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1990 if (WARN_ON_ONCE(tk == NULL)) {
1991 pr_warn("error on getting new probe.\n");
1994 file = find_trace_probe_file(tk, top_trace_array());
1995 if (WARN_ON_ONCE(file == NULL)) {
1996 pr_warn("error on getting probe file.\n");
1999 enable_trace_kprobe(
2000 trace_probe_event_call(&tk->tp), file);
2004 ret = create_or_delete_trace_kprobe("r:testprobe2 kprobe_trace_selftest_target $retval");
2005 if (WARN_ON_ONCE(ret)) {
2006 pr_warn("error on probing function return.\n");
2009 /* Enable trace point */
2010 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
2011 if (WARN_ON_ONCE(tk == NULL)) {
2012 pr_warn("error on getting 2nd new probe.\n");
2015 file = find_trace_probe_file(tk, top_trace_array());
2016 if (WARN_ON_ONCE(file == NULL)) {
2017 pr_warn("error on getting probe file.\n");
2020 enable_trace_kprobe(
2021 trace_probe_event_call(&tk->tp), file);
2028 ret = target(1, 2, 3, 4, 5, 6);
2031 * Not expecting an error here, the check is only to prevent the
2032 * optimizer from removing the call to target() as otherwise there
2033 * are no side-effects and the call is never performed.
2038 /* Disable trace points before removing it */
2039 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
2040 if (WARN_ON_ONCE(tk == NULL)) {
2041 pr_warn("error on getting test probe.\n");
2044 if (trace_kprobe_nhit(tk) != 1) {
2045 pr_warn("incorrect number of testprobe hits\n");
2049 file = find_trace_probe_file(tk, top_trace_array());
2050 if (WARN_ON_ONCE(file == NULL)) {
2051 pr_warn("error on getting probe file.\n");
2054 disable_trace_kprobe(
2055 trace_probe_event_call(&tk->tp), file);
2058 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
2059 if (WARN_ON_ONCE(tk == NULL)) {
2060 pr_warn("error on getting 2nd test probe.\n");
2063 if (trace_kprobe_nhit(tk) != 1) {
2064 pr_warn("incorrect number of testprobe2 hits\n");
2068 file = find_trace_probe_file(tk, top_trace_array());
2069 if (WARN_ON_ONCE(file == NULL)) {
2070 pr_warn("error on getting probe file.\n");
2073 disable_trace_kprobe(
2074 trace_probe_event_call(&tk->tp), file);
2077 ret = create_or_delete_trace_kprobe("-:testprobe");
2078 if (WARN_ON_ONCE(ret)) {
2079 pr_warn("error on deleting a probe.\n");
2083 ret = create_or_delete_trace_kprobe("-:testprobe2");
2084 if (WARN_ON_ONCE(ret)) {
2085 pr_warn("error on deleting a probe.\n");
2090 ret = dyn_events_release_all(&trace_kprobe_ops);
2091 if (WARN_ON_ONCE(ret)) {
2092 pr_warn("error on cleaning up probes.\n");
2096 * Wait for the optimizer work to finish. Otherwise it might fiddle
2097 * with probes in already freed __init text.
2099 wait_for_kprobe_optimizer();
2101 pr_cont("NG: Some tests are failed. Please check them.\n");
2107 late_initcall(kprobe_trace_self_tests_init);