1 // SPDX-License-Identifier: GPL-2.0
3 * Kprobes-based tracing events
5 * Created by Masami Hiramatsu <mhiramat@redhat.com>
8 #define pr_fmt(fmt) "trace_kprobe: " fmt
10 #include <linux/bpf-cgroup.h>
11 #include <linux/security.h>
12 #include <linux/module.h>
13 #include <linux/uaccess.h>
14 #include <linux/rculist.h>
15 #include <linux/error-injection.h>
17 #include <asm/setup.h> /* for COMMAND_LINE_SIZE */
19 #include "trace_dynevent.h"
20 #include "trace_kprobe_selftest.h"
21 #include "trace_probe.h"
22 #include "trace_probe_tmpl.h"
23 #include "trace_probe_kernel.h"
25 #define KPROBE_EVENT_SYSTEM "kprobes"
26 #define KRETPROBE_MAXACTIVE_MAX 4096
28 /* Kprobe early definition from command line */
29 static char kprobe_boot_events_buf[COMMAND_LINE_SIZE] __initdata;
31 static int __init set_kprobe_boot_events(char *str)
33 strscpy(kprobe_boot_events_buf, str, COMMAND_LINE_SIZE);
34 disable_tracing_selftest("running kprobe events");
38 __setup("kprobe_event=", set_kprobe_boot_events);
40 static int trace_kprobe_create(const char *raw_command);
41 static int trace_kprobe_show(struct seq_file *m, struct dyn_event *ev);
42 static int trace_kprobe_release(struct dyn_event *ev);
43 static bool trace_kprobe_is_busy(struct dyn_event *ev);
44 static bool trace_kprobe_match(const char *system, const char *event,
45 int argc, const char **argv, struct dyn_event *ev);
47 static struct dyn_event_operations trace_kprobe_ops = {
48 .create = trace_kprobe_create,
49 .show = trace_kprobe_show,
50 .is_busy = trace_kprobe_is_busy,
51 .free = trace_kprobe_release,
52 .match = trace_kprobe_match,
56 * Kprobe event core functions
59 struct dyn_event devent;
60 struct kretprobe rp; /* Use rp.kp for kprobe use */
61 unsigned long __percpu *nhit;
62 const char *symbol; /* symbol name */
63 struct trace_probe tp;
66 static bool is_trace_kprobe(struct dyn_event *ev)
68 return ev->ops == &trace_kprobe_ops;
71 static struct trace_kprobe *to_trace_kprobe(struct dyn_event *ev)
73 return container_of(ev, struct trace_kprobe, devent);
77 * for_each_trace_kprobe - iterate over the trace_kprobe list
78 * @pos: the struct trace_kprobe * for each entry
79 * @dpos: the struct dyn_event * to use as a loop cursor
81 #define for_each_trace_kprobe(pos, dpos) \
82 for_each_dyn_event(dpos) \
83 if (is_trace_kprobe(dpos) && (pos = to_trace_kprobe(dpos)))
85 static nokprobe_inline bool trace_kprobe_is_return(struct trace_kprobe *tk)
87 return tk->rp.handler != NULL;
90 static nokprobe_inline const char *trace_kprobe_symbol(struct trace_kprobe *tk)
92 return tk->symbol ? tk->symbol : "unknown";
95 static nokprobe_inline unsigned long trace_kprobe_offset(struct trace_kprobe *tk)
97 return tk->rp.kp.offset;
100 static nokprobe_inline bool trace_kprobe_has_gone(struct trace_kprobe *tk)
102 return kprobe_gone(&tk->rp.kp);
105 static nokprobe_inline bool trace_kprobe_within_module(struct trace_kprobe *tk,
108 int len = strlen(module_name(mod));
109 const char *name = trace_kprobe_symbol(tk);
111 return strncmp(module_name(mod), name, len) == 0 && name[len] == ':';
114 static nokprobe_inline bool trace_kprobe_module_exist(struct trace_kprobe *tk)
121 p = strchr(tk->symbol, ':');
125 rcu_read_lock_sched();
126 ret = !!find_module(tk->symbol);
127 rcu_read_unlock_sched();
133 static bool trace_kprobe_is_busy(struct dyn_event *ev)
135 struct trace_kprobe *tk = to_trace_kprobe(ev);
137 return trace_probe_is_enabled(&tk->tp);
140 static bool trace_kprobe_match_command_head(struct trace_kprobe *tk,
141 int argc, const char **argv)
143 char buf[MAX_ARGSTR_LEN + 1];
149 snprintf(buf, sizeof(buf), "0x%p", tk->rp.kp.addr);
150 else if (tk->rp.kp.offset)
151 snprintf(buf, sizeof(buf), "%s+%u",
152 trace_kprobe_symbol(tk), tk->rp.kp.offset);
154 snprintf(buf, sizeof(buf), "%s", trace_kprobe_symbol(tk));
155 if (strcmp(buf, argv[0]))
159 return trace_probe_match_command_args(&tk->tp, argc, argv);
162 static bool trace_kprobe_match(const char *system, const char *event,
163 int argc, const char **argv, struct dyn_event *ev)
165 struct trace_kprobe *tk = to_trace_kprobe(ev);
167 return (event[0] == '\0' ||
168 strcmp(trace_probe_name(&tk->tp), event) == 0) &&
169 (!system || strcmp(trace_probe_group_name(&tk->tp), system) == 0) &&
170 trace_kprobe_match_command_head(tk, argc, argv);
173 static nokprobe_inline unsigned long trace_kprobe_nhit(struct trace_kprobe *tk)
175 unsigned long nhit = 0;
178 for_each_possible_cpu(cpu)
179 nhit += *per_cpu_ptr(tk->nhit, cpu);
184 static nokprobe_inline bool trace_kprobe_is_registered(struct trace_kprobe *tk)
186 return !(list_empty(&tk->rp.kp.list) &&
187 hlist_unhashed(&tk->rp.kp.hlist));
190 /* Return 0 if it fails to find the symbol address */
191 static nokprobe_inline
192 unsigned long trace_kprobe_address(struct trace_kprobe *tk)
197 addr = (unsigned long)
198 kallsyms_lookup_name(trace_kprobe_symbol(tk));
200 addr += tk->rp.kp.offset;
202 addr = (unsigned long)tk->rp.kp.addr;
207 static nokprobe_inline struct trace_kprobe *
208 trace_kprobe_primary_from_call(struct trace_event_call *call)
210 struct trace_probe *tp;
212 tp = trace_probe_primary_from_call(call);
213 if (WARN_ON_ONCE(!tp))
216 return container_of(tp, struct trace_kprobe, tp);
219 bool trace_kprobe_on_func_entry(struct trace_event_call *call)
221 struct trace_kprobe *tk = trace_kprobe_primary_from_call(call);
223 return tk ? (kprobe_on_func_entry(tk->rp.kp.addr,
224 tk->rp.kp.addr ? NULL : tk->rp.kp.symbol_name,
225 tk->rp.kp.addr ? 0 : tk->rp.kp.offset) == 0) : false;
228 bool trace_kprobe_error_injectable(struct trace_event_call *call)
230 struct trace_kprobe *tk = trace_kprobe_primary_from_call(call);
232 return tk ? within_error_injection_list(trace_kprobe_address(tk)) :
236 static int register_kprobe_event(struct trace_kprobe *tk);
237 static int unregister_kprobe_event(struct trace_kprobe *tk);
239 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
240 static int kretprobe_dispatcher(struct kretprobe_instance *ri,
241 struct pt_regs *regs);
243 static void free_trace_kprobe(struct trace_kprobe *tk)
246 trace_probe_cleanup(&tk->tp);
248 free_percpu(tk->nhit);
254 * Allocate new trace_probe and initialize it (including kprobes).
256 static struct trace_kprobe *alloc_trace_kprobe(const char *group,
262 int nargs, bool is_return)
264 struct trace_kprobe *tk;
267 tk = kzalloc(struct_size(tk, tp.args, nargs), GFP_KERNEL);
271 tk->nhit = alloc_percpu(unsigned long);
276 tk->symbol = kstrdup(symbol, GFP_KERNEL);
279 tk->rp.kp.symbol_name = tk->symbol;
280 tk->rp.kp.offset = offs;
282 tk->rp.kp.addr = addr;
285 tk->rp.handler = kretprobe_dispatcher;
287 tk->rp.kp.pre_handler = kprobe_dispatcher;
289 tk->rp.maxactive = maxactive;
290 INIT_HLIST_NODE(&tk->rp.kp.hlist);
291 INIT_LIST_HEAD(&tk->rp.kp.list);
293 ret = trace_probe_init(&tk->tp, event, group, false);
297 dyn_event_init(&tk->devent, &trace_kprobe_ops);
300 free_trace_kprobe(tk);
304 static struct trace_kprobe *find_trace_kprobe(const char *event,
307 struct dyn_event *pos;
308 struct trace_kprobe *tk;
310 for_each_trace_kprobe(tk, pos)
311 if (strcmp(trace_probe_name(&tk->tp), event) == 0 &&
312 strcmp(trace_probe_group_name(&tk->tp), group) == 0)
317 static inline int __enable_trace_kprobe(struct trace_kprobe *tk)
321 if (trace_kprobe_is_registered(tk) && !trace_kprobe_has_gone(tk)) {
322 if (trace_kprobe_is_return(tk))
323 ret = enable_kretprobe(&tk->rp);
325 ret = enable_kprobe(&tk->rp.kp);
331 static void __disable_trace_kprobe(struct trace_probe *tp)
333 struct trace_kprobe *tk;
335 list_for_each_entry(tk, trace_probe_probe_list(tp), tp.list) {
336 if (!trace_kprobe_is_registered(tk))
338 if (trace_kprobe_is_return(tk))
339 disable_kretprobe(&tk->rp);
341 disable_kprobe(&tk->rp.kp);
347 * if the file is NULL, enable "perf" handler, or enable "trace" handler.
349 static int enable_trace_kprobe(struct trace_event_call *call,
350 struct trace_event_file *file)
352 struct trace_probe *tp;
353 struct trace_kprobe *tk;
357 tp = trace_probe_primary_from_call(call);
358 if (WARN_ON_ONCE(!tp))
360 enabled = trace_probe_is_enabled(tp);
362 /* This also changes "enabled" state */
364 ret = trace_probe_add_file(tp, file);
368 trace_probe_set_flag(tp, TP_FLAG_PROFILE);
373 list_for_each_entry(tk, trace_probe_probe_list(tp), tp.list) {
374 if (trace_kprobe_has_gone(tk))
376 ret = __enable_trace_kprobe(tk);
383 /* Failed to enable one of them. Roll back all */
385 __disable_trace_kprobe(tp);
387 trace_probe_remove_file(tp, file);
389 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
396 * Disable trace_probe
397 * if the file is NULL, disable "perf" handler, or disable "trace" handler.
399 static int disable_trace_kprobe(struct trace_event_call *call,
400 struct trace_event_file *file)
402 struct trace_probe *tp;
404 tp = trace_probe_primary_from_call(call);
405 if (WARN_ON_ONCE(!tp))
409 if (!trace_probe_get_file_link(tp, file))
411 if (!trace_probe_has_single_file(tp))
413 trace_probe_clear_flag(tp, TP_FLAG_TRACE);
415 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
417 if (!trace_probe_is_enabled(tp))
418 __disable_trace_kprobe(tp);
423 * Synchronization is done in below function. For perf event,
424 * file == NULL and perf_trace_event_unreg() calls
425 * tracepoint_synchronize_unregister() to ensure synchronize
426 * event. We don't need to care about it.
428 trace_probe_remove_file(tp, file);
433 #if defined(CONFIG_DYNAMIC_FTRACE) && \
434 !defined(CONFIG_KPROBE_EVENTS_ON_NOTRACE)
435 static bool __within_notrace_func(unsigned long addr)
437 unsigned long offset, size;
439 if (!addr || !kallsyms_lookup_size_offset(addr, &size, &offset))
442 /* Get the entry address of the target function */
446 * Since ftrace_location_range() does inclusive range check, we need
447 * to subtract 1 byte from the end address.
449 return !ftrace_location_range(addr, addr + size - 1);
452 static bool within_notrace_func(struct trace_kprobe *tk)
454 unsigned long addr = trace_kprobe_address(tk);
455 char symname[KSYM_NAME_LEN], *p;
457 if (!__within_notrace_func(addr))
460 /* Check if the address is on a suffixed-symbol */
461 if (!lookup_symbol_name(addr, symname)) {
462 p = strchr(symname, '.');
466 addr = (unsigned long)kprobe_lookup_name(symname, 0);
468 return __within_notrace_func(addr);
474 #define within_notrace_func(tk) (false)
477 /* Internal register function - just handle k*probes and flags */
478 static int __register_trace_kprobe(struct trace_kprobe *tk)
482 ret = security_locked_down(LOCKDOWN_KPROBES);
486 if (trace_kprobe_is_registered(tk))
489 if (within_notrace_func(tk)) {
490 pr_warn("Could not probe notrace function %s\n",
491 trace_kprobe_symbol(tk));
495 for (i = 0; i < tk->tp.nr_args; i++) {
496 ret = traceprobe_update_arg(&tk->tp.args[i]);
501 /* Set/clear disabled flag according to tp->flag */
502 if (trace_probe_is_enabled(&tk->tp))
503 tk->rp.kp.flags &= ~KPROBE_FLAG_DISABLED;
505 tk->rp.kp.flags |= KPROBE_FLAG_DISABLED;
507 if (trace_kprobe_is_return(tk))
508 ret = register_kretprobe(&tk->rp);
510 ret = register_kprobe(&tk->rp.kp);
515 /* Internal unregister function - just handle k*probes and flags */
516 static void __unregister_trace_kprobe(struct trace_kprobe *tk)
518 if (trace_kprobe_is_registered(tk)) {
519 if (trace_kprobe_is_return(tk))
520 unregister_kretprobe(&tk->rp);
522 unregister_kprobe(&tk->rp.kp);
523 /* Cleanup kprobe for reuse and mark it unregistered */
524 INIT_HLIST_NODE(&tk->rp.kp.hlist);
525 INIT_LIST_HEAD(&tk->rp.kp.list);
526 if (tk->rp.kp.symbol_name)
527 tk->rp.kp.addr = NULL;
531 /* Unregister a trace_probe and probe_event */
532 static int unregister_trace_kprobe(struct trace_kprobe *tk)
534 /* If other probes are on the event, just unregister kprobe */
535 if (trace_probe_has_sibling(&tk->tp))
538 /* Enabled event can not be unregistered */
539 if (trace_probe_is_enabled(&tk->tp))
542 /* If there's a reference to the dynamic event */
543 if (trace_event_dyn_busy(trace_probe_event_call(&tk->tp)))
546 /* Will fail if probe is being used by ftrace or perf */
547 if (unregister_kprobe_event(tk))
551 __unregister_trace_kprobe(tk);
552 dyn_event_remove(&tk->devent);
553 trace_probe_unlink(&tk->tp);
558 static bool trace_kprobe_has_same_kprobe(struct trace_kprobe *orig,
559 struct trace_kprobe *comp)
561 struct trace_probe_event *tpe = orig->tp.event;
564 list_for_each_entry(orig, &tpe->probes, tp.list) {
565 if (strcmp(trace_kprobe_symbol(orig),
566 trace_kprobe_symbol(comp)) ||
567 trace_kprobe_offset(orig) != trace_kprobe_offset(comp))
571 * trace_probe_compare_arg_type() ensured that nr_args and
572 * each argument name and type are same. Let's compare comm.
574 for (i = 0; i < orig->tp.nr_args; i++) {
575 if (strcmp(orig->tp.args[i].comm,
576 comp->tp.args[i].comm))
580 if (i == orig->tp.nr_args)
587 static int append_trace_kprobe(struct trace_kprobe *tk, struct trace_kprobe *to)
591 ret = trace_probe_compare_arg_type(&tk->tp, &to->tp);
593 /* Note that argument starts index = 2 */
594 trace_probe_log_set_index(ret + 1);
595 trace_probe_log_err(0, DIFF_ARG_TYPE);
598 if (trace_kprobe_has_same_kprobe(to, tk)) {
599 trace_probe_log_set_index(0);
600 trace_probe_log_err(0, SAME_PROBE);
604 /* Append to existing event */
605 ret = trace_probe_append(&tk->tp, &to->tp);
609 /* Register k*probe */
610 ret = __register_trace_kprobe(tk);
611 if (ret == -ENOENT && !trace_kprobe_module_exist(tk)) {
612 pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
617 trace_probe_unlink(&tk->tp);
619 dyn_event_add(&tk->devent, trace_probe_event_call(&tk->tp));
624 /* Register a trace_probe and probe_event */
625 static int register_trace_kprobe(struct trace_kprobe *tk)
627 struct trace_kprobe *old_tk;
630 mutex_lock(&event_mutex);
632 old_tk = find_trace_kprobe(trace_probe_name(&tk->tp),
633 trace_probe_group_name(&tk->tp));
635 if (trace_kprobe_is_return(tk) != trace_kprobe_is_return(old_tk)) {
636 trace_probe_log_set_index(0);
637 trace_probe_log_err(0, DIFF_PROBE_TYPE);
640 ret = append_trace_kprobe(tk, old_tk);
645 /* Register new event */
646 ret = register_kprobe_event(tk);
648 if (ret == -EEXIST) {
649 trace_probe_log_set_index(0);
650 trace_probe_log_err(0, EVENT_EXIST);
652 pr_warn("Failed to register probe event(%d)\n", ret);
656 /* Register k*probe */
657 ret = __register_trace_kprobe(tk);
658 if (ret == -ENOENT && !trace_kprobe_module_exist(tk)) {
659 pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
664 unregister_kprobe_event(tk);
666 dyn_event_add(&tk->devent, trace_probe_event_call(&tk->tp));
669 mutex_unlock(&event_mutex);
673 /* Module notifier call back, checking event on the module */
674 static int trace_kprobe_module_callback(struct notifier_block *nb,
675 unsigned long val, void *data)
677 struct module *mod = data;
678 struct dyn_event *pos;
679 struct trace_kprobe *tk;
682 if (val != MODULE_STATE_COMING)
685 /* Update probes on coming module */
686 mutex_lock(&event_mutex);
687 for_each_trace_kprobe(tk, pos) {
688 if (trace_kprobe_within_module(tk, mod)) {
689 /* Don't need to check busy - this should have gone. */
690 __unregister_trace_kprobe(tk);
691 ret = __register_trace_kprobe(tk);
693 pr_warn("Failed to re-register probe %s on %s: %d\n",
694 trace_probe_name(&tk->tp),
695 module_name(mod), ret);
698 mutex_unlock(&event_mutex);
703 static struct notifier_block trace_kprobe_module_nb = {
704 .notifier_call = trace_kprobe_module_callback,
705 .priority = 1 /* Invoked after kprobe module callback */
708 static int count_symbols(void *data, unsigned long unused)
710 unsigned int *count = data;
717 struct sym_count_ctx {
722 static int count_mod_symbols(void *data, const char *name, unsigned long unused)
724 struct sym_count_ctx *ctx = data;
726 if (strcmp(name, ctx->name) == 0)
732 static unsigned int number_of_same_symbols(char *func_name)
734 struct sym_count_ctx ctx = { .count = 0, .name = func_name };
736 kallsyms_on_each_match_symbol(count_symbols, func_name, &ctx.count);
738 module_kallsyms_on_each_symbol(NULL, count_mod_symbols, &ctx);
743 static int __trace_kprobe_create(int argc, const char *argv[])
748 * p[:[GRP/][EVENT]] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
750 * r[MAXACTIVE][:[GRP/][EVENT]] [MOD:]KSYM[+0] [FETCHARGS]
752 * p[:[GRP/][EVENT]] [MOD:]KSYM[+0]%return [FETCHARGS]
755 * $retval : fetch return value
756 * $stack : fetch stack address
757 * $stackN : fetch Nth of stack (N:0-)
758 * $comm : fetch current task comm
759 * @ADDR : fetch memory at ADDR (ADDR should be in kernel)
760 * @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
761 * %REG : fetch register REG
762 * Dereferencing memory fetch:
763 * +|-offs(ARG) : fetch memory at ARG +|- offs address.
764 * Alias name of args:
765 * NAME=FETCHARG : set NAME as alias of FETCHARG.
767 * FETCHARG:TYPE : use TYPE instead of unsigned long.
769 struct trace_kprobe *tk = NULL;
770 int i, len, new_argc = 0, ret = 0;
771 bool is_return = false;
772 char *symbol = NULL, *tmp = NULL;
773 const char **new_argv = NULL;
774 const char *event = NULL, *group = KPROBE_EVENT_SYSTEM;
775 enum probe_print_type ptype;
779 char buf[MAX_EVENT_NAME_LEN];
780 char gbuf[MAX_EVENT_NAME_LEN];
781 char abuf[MAX_BTF_ARGS_LEN];
782 struct traceprobe_parse_context ctx = { .flags = TPARG_FL_KERNEL };
784 switch (argv[0][0]) {
796 trace_probe_log_init("trace_kprobe", argc, argv);
798 event = strchr(&argv[0][1], ':');
802 if (isdigit(argv[0][1])) {
804 trace_probe_log_err(1, BAD_MAXACT_TYPE);
808 len = event - &argv[0][1] - 1;
810 len = strlen(&argv[0][1]);
811 if (len > MAX_EVENT_NAME_LEN - 1) {
812 trace_probe_log_err(1, BAD_MAXACT);
815 memcpy(buf, &argv[0][1], len);
817 ret = kstrtouint(buf, 0, &maxactive);
818 if (ret || !maxactive) {
819 trace_probe_log_err(1, BAD_MAXACT);
822 /* kretprobes instances are iterated over via a list. The
823 * maximum should stay reasonable.
825 if (maxactive > KRETPROBE_MAXACTIVE_MAX) {
826 trace_probe_log_err(1, MAXACT_TOO_BIG);
831 /* try to parse an address. if that fails, try to read the
832 * input as a symbol. */
833 if (kstrtoul(argv[1], 0, (unsigned long *)&addr)) {
834 trace_probe_log_set_index(1);
835 /* Check whether uprobe event specified */
836 if (strchr(argv[1], '/') && strchr(argv[1], ':')) {
840 /* a symbol specified */
841 symbol = kstrdup(argv[1], GFP_KERNEL);
845 tmp = strchr(symbol, '%');
847 if (!strcmp(tmp, "%return")) {
851 trace_probe_log_err(tmp - symbol, BAD_ADDR_SUFFIX);
856 /* TODO: support .init module functions */
857 ret = traceprobe_split_symbol_offset(symbol, &offset);
858 if (ret || offset < 0 || offset > UINT_MAX) {
859 trace_probe_log_err(0, BAD_PROBE_ADDR);
863 ctx.flags |= TPARG_FL_RETURN;
864 ret = kprobe_on_func_entry(NULL, symbol, offset);
865 if (ret == 0 && !is_return)
866 ctx.flags |= TPARG_FL_FENTRY;
867 /* Defer the ENOENT case until register kprobe */
868 if (ret == -EINVAL && is_return) {
869 trace_probe_log_err(0, BAD_RETPROBE);
874 if (symbol && !strchr(symbol, ':')) {
877 count = number_of_same_symbols(symbol);
880 * Users should use ADDR to remove the ambiguity of
883 trace_probe_log_err(0, NON_UNIQ_SYMBOL);
884 ret = -EADDRNOTAVAIL;
887 } else if (count == 0) {
889 * We can return ENOENT earlier than when register the
892 trace_probe_log_err(0, BAD_PROBE_ADDR);
899 trace_probe_log_set_index(0);
901 ret = traceprobe_parse_event_name(&event, &group, gbuf,
908 /* Make a new event name */
910 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
911 is_return ? 'r' : 'p', symbol, offset);
913 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
914 is_return ? 'r' : 'p', addr);
915 sanitize_event_name(buf);
919 argc -= 2; argv += 2;
920 ctx.funcname = symbol;
921 new_argv = traceprobe_expand_meta_args(argc, argv, &new_argc,
922 abuf, MAX_BTF_ARGS_LEN, &ctx);
923 if (IS_ERR(new_argv)) {
924 ret = PTR_ERR(new_argv);
934 tk = alloc_trace_kprobe(group, event, addr, symbol, offset, maxactive,
938 /* This must return -ENOMEM, else there is a bug */
939 WARN_ON_ONCE(ret != -ENOMEM);
940 goto out; /* We know tk is not allocated */
943 /* parse arguments */
944 for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
945 trace_probe_log_set_index(i + 2);
947 ret = traceprobe_parse_probe_arg(&tk->tp, i, argv[i], &ctx);
949 goto error; /* This can be -ENOMEM */
952 ptype = is_return ? PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL;
953 ret = traceprobe_set_print_fmt(&tk->tp, ptype);
957 ret = register_trace_kprobe(tk);
959 trace_probe_log_set_index(1);
961 trace_probe_log_err(0, BAD_INSN_BNDRY);
962 else if (ret == -ENOENT)
963 trace_probe_log_err(0, BAD_PROBE_ADDR);
964 else if (ret != -ENOMEM && ret != -EEXIST)
965 trace_probe_log_err(0, FAIL_REG_PROBE);
970 traceprobe_finish_parse(&ctx);
971 trace_probe_log_clear();
979 free_trace_kprobe(tk);
983 static int trace_kprobe_create(const char *raw_command)
985 return trace_probe_create(raw_command, __trace_kprobe_create);
988 static int create_or_delete_trace_kprobe(const char *raw_command)
992 if (raw_command[0] == '-')
993 return dyn_event_release(raw_command, &trace_kprobe_ops);
995 ret = trace_kprobe_create(raw_command);
996 return ret == -ECANCELED ? -EINVAL : ret;
999 static int trace_kprobe_run_command(struct dynevent_cmd *cmd)
1001 return create_or_delete_trace_kprobe(cmd->seq.buffer);
1005 * kprobe_event_cmd_init - Initialize a kprobe event command object
1006 * @cmd: A pointer to the dynevent_cmd struct representing the new event
1007 * @buf: A pointer to the buffer used to build the command
1008 * @maxlen: The length of the buffer passed in @buf
1010 * Initialize a synthetic event command object. Use this before
1011 * calling any of the other kprobe_event functions.
1013 void kprobe_event_cmd_init(struct dynevent_cmd *cmd, char *buf, int maxlen)
1015 dynevent_cmd_init(cmd, buf, maxlen, DYNEVENT_TYPE_KPROBE,
1016 trace_kprobe_run_command);
1018 EXPORT_SYMBOL_GPL(kprobe_event_cmd_init);
1021 * __kprobe_event_gen_cmd_start - Generate a kprobe event command from arg list
1022 * @cmd: A pointer to the dynevent_cmd struct representing the new event
1023 * @kretprobe: Is this a return probe?
1024 * @name: The name of the kprobe event
1025 * @loc: The location of the kprobe event
1026 * @...: Variable number of arg (pairs), one pair for each field
1028 * NOTE: Users normally won't want to call this function directly, but
1029 * rather use the kprobe_event_gen_cmd_start() wrapper, which automatically
1030 * adds a NULL to the end of the arg list. If this function is used
1031 * directly, make sure the last arg in the variable arg list is NULL.
1033 * Generate a kprobe event command to be executed by
1034 * kprobe_event_gen_cmd_end(). This function can be used to generate the
1035 * complete command or only the first part of it; in the latter case,
1036 * kprobe_event_add_fields() can be used to add more fields following this.
1038 * Unlikely the synth_event_gen_cmd_start(), @loc must be specified. This
1039 * returns -EINVAL if @loc == NULL.
1041 * Return: 0 if successful, error otherwise.
1043 int __kprobe_event_gen_cmd_start(struct dynevent_cmd *cmd, bool kretprobe,
1044 const char *name, const char *loc, ...)
1046 char buf[MAX_EVENT_NAME_LEN];
1047 struct dynevent_arg arg;
1051 if (cmd->type != DYNEVENT_TYPE_KPROBE)
1058 snprintf(buf, MAX_EVENT_NAME_LEN, "r:kprobes/%s", name);
1060 snprintf(buf, MAX_EVENT_NAME_LEN, "p:kprobes/%s", name);
1062 ret = dynevent_str_add(cmd, buf);
1066 dynevent_arg_init(&arg, 0);
1068 ret = dynevent_arg_add(cmd, &arg, NULL);
1072 va_start(args, loc);
1076 field = va_arg(args, const char *);
1080 if (++cmd->n_fields > MAX_TRACE_ARGS) {
1086 ret = dynevent_arg_add(cmd, &arg, NULL);
1094 EXPORT_SYMBOL_GPL(__kprobe_event_gen_cmd_start);
1097 * __kprobe_event_add_fields - Add probe fields to a kprobe command from arg list
1098 * @cmd: A pointer to the dynevent_cmd struct representing the new event
1099 * @...: Variable number of arg (pairs), one pair for each field
1101 * NOTE: Users normally won't want to call this function directly, but
1102 * rather use the kprobe_event_add_fields() wrapper, which
1103 * automatically adds a NULL to the end of the arg list. If this
1104 * function is used directly, make sure the last arg in the variable
1107 * Add probe fields to an existing kprobe command using a variable
1108 * list of args. Fields are added in the same order they're listed.
1110 * Return: 0 if successful, error otherwise.
1112 int __kprobe_event_add_fields(struct dynevent_cmd *cmd, ...)
1114 struct dynevent_arg arg;
1118 if (cmd->type != DYNEVENT_TYPE_KPROBE)
1121 dynevent_arg_init(&arg, 0);
1123 va_start(args, cmd);
1127 field = va_arg(args, const char *);
1131 if (++cmd->n_fields > MAX_TRACE_ARGS) {
1137 ret = dynevent_arg_add(cmd, &arg, NULL);
1145 EXPORT_SYMBOL_GPL(__kprobe_event_add_fields);
1148 * kprobe_event_delete - Delete a kprobe event
1149 * @name: The name of the kprobe event to delete
1151 * Delete a kprobe event with the give @name from kernel code rather
1152 * than directly from the command line.
1154 * Return: 0 if successful, error otherwise.
1156 int kprobe_event_delete(const char *name)
1158 char buf[MAX_EVENT_NAME_LEN];
1160 snprintf(buf, MAX_EVENT_NAME_LEN, "-:%s", name);
1162 return create_or_delete_trace_kprobe(buf);
1164 EXPORT_SYMBOL_GPL(kprobe_event_delete);
1166 static int trace_kprobe_release(struct dyn_event *ev)
1168 struct trace_kprobe *tk = to_trace_kprobe(ev);
1169 int ret = unregister_trace_kprobe(tk);
1172 free_trace_kprobe(tk);
1176 static int trace_kprobe_show(struct seq_file *m, struct dyn_event *ev)
1178 struct trace_kprobe *tk = to_trace_kprobe(ev);
1181 seq_putc(m, trace_kprobe_is_return(tk) ? 'r' : 'p');
1182 if (trace_kprobe_is_return(tk) && tk->rp.maxactive)
1183 seq_printf(m, "%d", tk->rp.maxactive);
1184 seq_printf(m, ":%s/%s", trace_probe_group_name(&tk->tp),
1185 trace_probe_name(&tk->tp));
1188 seq_printf(m, " 0x%p", tk->rp.kp.addr);
1189 else if (tk->rp.kp.offset)
1190 seq_printf(m, " %s+%u", trace_kprobe_symbol(tk),
1193 seq_printf(m, " %s", trace_kprobe_symbol(tk));
1195 for (i = 0; i < tk->tp.nr_args; i++)
1196 seq_printf(m, " %s=%s", tk->tp.args[i].name, tk->tp.args[i].comm);
1202 static int probes_seq_show(struct seq_file *m, void *v)
1204 struct dyn_event *ev = v;
1206 if (!is_trace_kprobe(ev))
1209 return trace_kprobe_show(m, ev);
1212 static const struct seq_operations probes_seq_op = {
1213 .start = dyn_event_seq_start,
1214 .next = dyn_event_seq_next,
1215 .stop = dyn_event_seq_stop,
1216 .show = probes_seq_show
1219 static int probes_open(struct inode *inode, struct file *file)
1223 ret = security_locked_down(LOCKDOWN_TRACEFS);
1227 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
1228 ret = dyn_events_release_all(&trace_kprobe_ops);
1233 return seq_open(file, &probes_seq_op);
1236 static ssize_t probes_write(struct file *file, const char __user *buffer,
1237 size_t count, loff_t *ppos)
1239 return trace_parse_run_command(file, buffer, count, ppos,
1240 create_or_delete_trace_kprobe);
1243 static const struct file_operations kprobe_events_ops = {
1244 .owner = THIS_MODULE,
1245 .open = probes_open,
1247 .llseek = seq_lseek,
1248 .release = seq_release,
1249 .write = probes_write,
1252 /* Probes profiling interfaces */
1253 static int probes_profile_seq_show(struct seq_file *m, void *v)
1255 struct dyn_event *ev = v;
1256 struct trace_kprobe *tk;
1257 unsigned long nmissed;
1259 if (!is_trace_kprobe(ev))
1262 tk = to_trace_kprobe(ev);
1263 nmissed = trace_kprobe_is_return(tk) ?
1264 tk->rp.kp.nmissed + tk->rp.nmissed : tk->rp.kp.nmissed;
1265 seq_printf(m, " %-44s %15lu %15lu\n",
1266 trace_probe_name(&tk->tp),
1267 trace_kprobe_nhit(tk),
1273 static const struct seq_operations profile_seq_op = {
1274 .start = dyn_event_seq_start,
1275 .next = dyn_event_seq_next,
1276 .stop = dyn_event_seq_stop,
1277 .show = probes_profile_seq_show
1280 static int profile_open(struct inode *inode, struct file *file)
1284 ret = security_locked_down(LOCKDOWN_TRACEFS);
1288 return seq_open(file, &profile_seq_op);
1291 static const struct file_operations kprobe_profile_ops = {
1292 .owner = THIS_MODULE,
1293 .open = profile_open,
1295 .llseek = seq_lseek,
1296 .release = seq_release,
1299 /* Note that we don't verify it, since the code does not come from user space */
1301 process_fetch_insn(struct fetch_insn *code, void *rec, void *dest,
1304 struct pt_regs *regs = rec;
1309 /* 1st stage: get value from context */
1312 val = regs_get_register(regs, code->param);
1314 case FETCH_OP_STACK:
1315 val = regs_get_kernel_stack_nth(regs, code->param);
1317 case FETCH_OP_STACKP:
1318 val = kernel_stack_pointer(regs);
1320 case FETCH_OP_RETVAL:
1321 val = regs_return_value(regs);
1323 #ifdef CONFIG_HAVE_FUNCTION_ARG_ACCESS_API
1325 val = regs_get_kernel_argument(regs, code->param);
1328 case FETCH_NOP_SYMBOL: /* Ignore a place holder */
1332 ret = process_common_fetch_insn(code, &val);
1338 return process_fetch_insn_bottom(code, val, dest, base);
1340 NOKPROBE_SYMBOL(process_fetch_insn)
1342 /* Kprobe handler */
1343 static nokprobe_inline void
1344 __kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs,
1345 struct trace_event_file *trace_file)
1347 struct kprobe_trace_entry_head *entry;
1348 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1349 struct trace_event_buffer fbuffer;
1352 WARN_ON(call != trace_file->event_call);
1354 if (trace_trigger_soft_disabled(trace_file))
1357 dsize = __get_data_size(&tk->tp, regs);
1359 entry = trace_event_buffer_reserve(&fbuffer, trace_file,
1360 sizeof(*entry) + tk->tp.size + dsize);
1364 fbuffer.regs = regs;
1365 entry->ip = (unsigned long)tk->rp.kp.addr;
1366 store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1368 trace_event_buffer_commit(&fbuffer);
1372 kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs)
1374 struct event_file_link *link;
1376 trace_probe_for_each_link_rcu(link, &tk->tp)
1377 __kprobe_trace_func(tk, regs, link->file);
1379 NOKPROBE_SYMBOL(kprobe_trace_func);
1381 /* Kretprobe handler */
1382 static nokprobe_inline void
1383 __kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1384 struct pt_regs *regs,
1385 struct trace_event_file *trace_file)
1387 struct kretprobe_trace_entry_head *entry;
1388 struct trace_event_buffer fbuffer;
1389 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1392 WARN_ON(call != trace_file->event_call);
1394 if (trace_trigger_soft_disabled(trace_file))
1397 dsize = __get_data_size(&tk->tp, regs);
1399 entry = trace_event_buffer_reserve(&fbuffer, trace_file,
1400 sizeof(*entry) + tk->tp.size + dsize);
1404 fbuffer.regs = regs;
1405 entry->func = (unsigned long)tk->rp.kp.addr;
1406 entry->ret_ip = get_kretprobe_retaddr(ri);
1407 store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1409 trace_event_buffer_commit(&fbuffer);
1413 kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1414 struct pt_regs *regs)
1416 struct event_file_link *link;
1418 trace_probe_for_each_link_rcu(link, &tk->tp)
1419 __kretprobe_trace_func(tk, ri, regs, link->file);
1421 NOKPROBE_SYMBOL(kretprobe_trace_func);
1423 /* Event entry printers */
1424 static enum print_line_t
1425 print_kprobe_event(struct trace_iterator *iter, int flags,
1426 struct trace_event *event)
1428 struct kprobe_trace_entry_head *field;
1429 struct trace_seq *s = &iter->seq;
1430 struct trace_probe *tp;
1432 field = (struct kprobe_trace_entry_head *)iter->ent;
1433 tp = trace_probe_primary_from_call(
1434 container_of(event, struct trace_event_call, event));
1435 if (WARN_ON_ONCE(!tp))
1438 trace_seq_printf(s, "%s: (", trace_probe_name(tp));
1440 if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
1443 trace_seq_putc(s, ')');
1445 if (trace_probe_print_args(s, tp->args, tp->nr_args,
1446 (u8 *)&field[1], field) < 0)
1449 trace_seq_putc(s, '\n');
1451 return trace_handle_return(s);
1454 static enum print_line_t
1455 print_kretprobe_event(struct trace_iterator *iter, int flags,
1456 struct trace_event *event)
1458 struct kretprobe_trace_entry_head *field;
1459 struct trace_seq *s = &iter->seq;
1460 struct trace_probe *tp;
1462 field = (struct kretprobe_trace_entry_head *)iter->ent;
1463 tp = trace_probe_primary_from_call(
1464 container_of(event, struct trace_event_call, event));
1465 if (WARN_ON_ONCE(!tp))
1468 trace_seq_printf(s, "%s: (", trace_probe_name(tp));
1470 if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1473 trace_seq_puts(s, " <- ");
1475 if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1478 trace_seq_putc(s, ')');
1480 if (trace_probe_print_args(s, tp->args, tp->nr_args,
1481 (u8 *)&field[1], field) < 0)
1484 trace_seq_putc(s, '\n');
1487 return trace_handle_return(s);
1491 static int kprobe_event_define_fields(struct trace_event_call *event_call)
1494 struct kprobe_trace_entry_head field;
1495 struct trace_probe *tp;
1497 tp = trace_probe_primary_from_call(event_call);
1498 if (WARN_ON_ONCE(!tp))
1501 DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1503 return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
1506 static int kretprobe_event_define_fields(struct trace_event_call *event_call)
1509 struct kretprobe_trace_entry_head field;
1510 struct trace_probe *tp;
1512 tp = trace_probe_primary_from_call(event_call);
1513 if (WARN_ON_ONCE(!tp))
1516 DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1517 DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1519 return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
1522 #ifdef CONFIG_PERF_EVENTS
1524 /* Kprobe profile handler */
1526 kprobe_perf_func(struct trace_kprobe *tk, struct pt_regs *regs)
1528 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1529 struct kprobe_trace_entry_head *entry;
1530 struct hlist_head *head;
1531 int size, __size, dsize;
1534 if (bpf_prog_array_valid(call)) {
1535 unsigned long orig_ip = instruction_pointer(regs);
1538 ret = trace_call_bpf(call, regs);
1541 * We need to check and see if we modified the pc of the
1542 * pt_regs, and if so return 1 so that we don't do the
1545 if (orig_ip != instruction_pointer(regs))
1551 head = this_cpu_ptr(call->perf_events);
1552 if (hlist_empty(head))
1555 dsize = __get_data_size(&tk->tp, regs);
1556 __size = sizeof(*entry) + tk->tp.size + dsize;
1557 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1558 size -= sizeof(u32);
1560 entry = perf_trace_buf_alloc(size, NULL, &rctx);
1564 entry->ip = (unsigned long)tk->rp.kp.addr;
1565 memset(&entry[1], 0, dsize);
1566 store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1567 perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1571 NOKPROBE_SYMBOL(kprobe_perf_func);
1573 /* Kretprobe profile handler */
1575 kretprobe_perf_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1576 struct pt_regs *regs)
1578 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1579 struct kretprobe_trace_entry_head *entry;
1580 struct hlist_head *head;
1581 int size, __size, dsize;
1584 if (bpf_prog_array_valid(call) && !trace_call_bpf(call, regs))
1587 head = this_cpu_ptr(call->perf_events);
1588 if (hlist_empty(head))
1591 dsize = __get_data_size(&tk->tp, regs);
1592 __size = sizeof(*entry) + tk->tp.size + dsize;
1593 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1594 size -= sizeof(u32);
1596 entry = perf_trace_buf_alloc(size, NULL, &rctx);
1600 entry->func = (unsigned long)tk->rp.kp.addr;
1601 entry->ret_ip = get_kretprobe_retaddr(ri);
1602 store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1603 perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1606 NOKPROBE_SYMBOL(kretprobe_perf_func);
1608 int bpf_get_kprobe_info(const struct perf_event *event, u32 *fd_type,
1609 const char **symbol, u64 *probe_offset,
1610 u64 *probe_addr, bool perf_type_tracepoint)
1612 const char *pevent = trace_event_name(event->tp_event);
1613 const char *group = event->tp_event->class->system;
1614 struct trace_kprobe *tk;
1616 if (perf_type_tracepoint)
1617 tk = find_trace_kprobe(pevent, group);
1619 tk = trace_kprobe_primary_from_call(event->tp_event);
1623 *fd_type = trace_kprobe_is_return(tk) ? BPF_FD_TYPE_KRETPROBE
1624 : BPF_FD_TYPE_KPROBE;
1625 *probe_offset = tk->rp.kp.offset;
1626 *probe_addr = kallsyms_show_value(current_cred()) ?
1627 (unsigned long)tk->rp.kp.addr : 0;
1628 *symbol = tk->symbol;
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
1759 /* create a trace_kprobe, but don't add it to global lists */
1760 struct trace_event_call *
1761 create_local_trace_kprobe(char *func, void *addr, unsigned long offs,
1764 enum probe_print_type ptype;
1765 struct trace_kprobe *tk;
1772 count = number_of_same_symbols(func);
1775 * Users should use addr to remove the ambiguity of
1778 return ERR_PTR(-EADDRNOTAVAIL);
1779 else if (count == 0)
1781 * We can return ENOENT earlier than when register the
1784 return ERR_PTR(-ENOENT);
1788 * local trace_kprobes are not added to dyn_event, so they are never
1789 * searched in find_trace_kprobe(). Therefore, there is no concern of
1790 * duplicated name here.
1792 event = func ? func : "DUMMY_EVENT";
1794 tk = alloc_trace_kprobe(KPROBE_EVENT_SYSTEM, event, (void *)addr, func,
1795 offs, 0 /* maxactive */, 0 /* nargs */,
1799 pr_info("Failed to allocate trace_probe.(%d)\n",
1801 return ERR_CAST(tk);
1804 init_trace_event_call(tk);
1806 ptype = trace_kprobe_is_return(tk) ?
1807 PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL;
1808 if (traceprobe_set_print_fmt(&tk->tp, ptype) < 0) {
1813 ret = __register_trace_kprobe(tk);
1817 return trace_probe_event_call(&tk->tp);
1819 free_trace_kprobe(tk);
1820 return ERR_PTR(ret);
1823 void destroy_local_trace_kprobe(struct trace_event_call *event_call)
1825 struct trace_kprobe *tk;
1827 tk = trace_kprobe_primary_from_call(event_call);
1831 if (trace_probe_is_enabled(&tk->tp)) {
1836 __unregister_trace_kprobe(tk);
1838 free_trace_kprobe(tk);
1840 #endif /* CONFIG_PERF_EVENTS */
1842 static __init void enable_boot_kprobe_events(void)
1844 struct trace_array *tr = top_trace_array();
1845 struct trace_event_file *file;
1846 struct trace_kprobe *tk;
1847 struct dyn_event *pos;
1849 mutex_lock(&event_mutex);
1850 for_each_trace_kprobe(tk, pos) {
1851 list_for_each_entry(file, &tr->events, list)
1852 if (file->event_call == trace_probe_event_call(&tk->tp))
1853 trace_event_enable_disable(file, 1, 0);
1855 mutex_unlock(&event_mutex);
1858 static __init void setup_boot_kprobe_events(void)
1860 char *p, *cmd = kprobe_boot_events_buf;
1863 strreplace(kprobe_boot_events_buf, ',', ' ');
1865 while (cmd && *cmd != '\0') {
1866 p = strchr(cmd, ';');
1870 ret = create_or_delete_trace_kprobe(cmd);
1872 pr_warn("Failed to add event(%d): %s\n", ret, cmd);
1877 enable_boot_kprobe_events();
1881 * Register dynevent at core_initcall. This allows kernel to setup kprobe
1882 * events in postcore_initcall without tracefs.
1884 static __init int init_kprobe_trace_early(void)
1888 ret = dyn_event_register(&trace_kprobe_ops);
1892 if (register_module_notifier(&trace_kprobe_module_nb))
1897 core_initcall(init_kprobe_trace_early);
1899 /* Make a tracefs interface for controlling probe points */
1900 static __init int init_kprobe_trace(void)
1904 ret = tracing_init_dentry();
1908 /* Event list interface */
1909 trace_create_file("kprobe_events", TRACE_MODE_WRITE,
1910 NULL, NULL, &kprobe_events_ops);
1912 /* Profile interface */
1913 trace_create_file("kprobe_profile", TRACE_MODE_READ,
1914 NULL, NULL, &kprobe_profile_ops);
1916 setup_boot_kprobe_events();
1920 fs_initcall(init_kprobe_trace);
1923 #ifdef CONFIG_FTRACE_STARTUP_TEST
1924 static __init struct trace_event_file *
1925 find_trace_probe_file(struct trace_kprobe *tk, struct trace_array *tr)
1927 struct trace_event_file *file;
1929 list_for_each_entry(file, &tr->events, list)
1930 if (file->event_call == trace_probe_event_call(&tk->tp))
1937 * Nobody but us can call enable_trace_kprobe/disable_trace_kprobe at this
1938 * stage, we can do this lockless.
1940 static __init int kprobe_trace_self_tests_init(void)
1943 int (*target)(int, int, int, int, int, int);
1944 struct trace_kprobe *tk;
1945 struct trace_event_file *file;
1947 if (tracing_is_disabled())
1950 if (tracing_selftest_disabled)
1953 target = kprobe_trace_selftest_target;
1955 pr_info("Testing kprobe tracing: ");
1957 ret = create_or_delete_trace_kprobe("p:testprobe kprobe_trace_selftest_target $stack $stack0 +0($stack)");
1958 if (WARN_ON_ONCE(ret)) {
1959 pr_warn("error on probing function entry.\n");
1962 /* Enable trace point */
1963 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1964 if (WARN_ON_ONCE(tk == NULL)) {
1965 pr_warn("error on getting new probe.\n");
1968 file = find_trace_probe_file(tk, top_trace_array());
1969 if (WARN_ON_ONCE(file == NULL)) {
1970 pr_warn("error on getting probe file.\n");
1973 enable_trace_kprobe(
1974 trace_probe_event_call(&tk->tp), file);
1978 ret = create_or_delete_trace_kprobe("r:testprobe2 kprobe_trace_selftest_target $retval");
1979 if (WARN_ON_ONCE(ret)) {
1980 pr_warn("error on probing function return.\n");
1983 /* Enable trace point */
1984 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1985 if (WARN_ON_ONCE(tk == NULL)) {
1986 pr_warn("error on getting 2nd new probe.\n");
1989 file = find_trace_probe_file(tk, top_trace_array());
1990 if (WARN_ON_ONCE(file == NULL)) {
1991 pr_warn("error on getting probe file.\n");
1994 enable_trace_kprobe(
1995 trace_probe_event_call(&tk->tp), file);
2002 ret = target(1, 2, 3, 4, 5, 6);
2005 * Not expecting an error here, the check is only to prevent the
2006 * optimizer from removing the call to target() as otherwise there
2007 * are no side-effects and the call is never performed.
2012 /* Disable trace points before removing it */
2013 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
2014 if (WARN_ON_ONCE(tk == NULL)) {
2015 pr_warn("error on getting test probe.\n");
2018 if (trace_kprobe_nhit(tk) != 1) {
2019 pr_warn("incorrect number of testprobe hits\n");
2023 file = find_trace_probe_file(tk, top_trace_array());
2024 if (WARN_ON_ONCE(file == NULL)) {
2025 pr_warn("error on getting probe file.\n");
2028 disable_trace_kprobe(
2029 trace_probe_event_call(&tk->tp), file);
2032 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
2033 if (WARN_ON_ONCE(tk == NULL)) {
2034 pr_warn("error on getting 2nd test probe.\n");
2037 if (trace_kprobe_nhit(tk) != 1) {
2038 pr_warn("incorrect number of testprobe2 hits\n");
2042 file = find_trace_probe_file(tk, top_trace_array());
2043 if (WARN_ON_ONCE(file == NULL)) {
2044 pr_warn("error on getting probe file.\n");
2047 disable_trace_kprobe(
2048 trace_probe_event_call(&tk->tp), file);
2051 ret = create_or_delete_trace_kprobe("-:testprobe");
2052 if (WARN_ON_ONCE(ret)) {
2053 pr_warn("error on deleting a probe.\n");
2057 ret = create_or_delete_trace_kprobe("-:testprobe2");
2058 if (WARN_ON_ONCE(ret)) {
2059 pr_warn("error on deleting a probe.\n");
2064 ret = dyn_events_release_all(&trace_kprobe_ops);
2065 if (WARN_ON_ONCE(ret)) {
2066 pr_warn("error on cleaning up probes.\n");
2070 * Wait for the optimizer work to finish. Otherwise it might fiddle
2071 * with probes in already freed __init text.
2073 wait_for_kprobe_optimizer();
2075 pr_cont("NG: Some tests are failed. Please check them.\n");
2081 late_initcall(kprobe_trace_self_tests_init);