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 #define SIZEOF_TRACE_KPROBE(n) \
84 (offsetof(struct trace_kprobe, tp.args) + \
85 (sizeof(struct probe_arg) * (n)))
87 static nokprobe_inline bool trace_kprobe_is_return(struct trace_kprobe *tk)
89 return tk->rp.handler != NULL;
92 static nokprobe_inline const char *trace_kprobe_symbol(struct trace_kprobe *tk)
94 return tk->symbol ? tk->symbol : "unknown";
97 static nokprobe_inline unsigned long trace_kprobe_offset(struct trace_kprobe *tk)
99 return tk->rp.kp.offset;
102 static nokprobe_inline bool trace_kprobe_has_gone(struct trace_kprobe *tk)
104 return !!(kprobe_gone(&tk->rp.kp));
107 static nokprobe_inline bool trace_kprobe_within_module(struct trace_kprobe *tk,
110 int len = strlen(module_name(mod));
111 const char *name = trace_kprobe_symbol(tk);
113 return strncmp(module_name(mod), name, len) == 0 && name[len] == ':';
116 static nokprobe_inline bool trace_kprobe_module_exist(struct trace_kprobe *tk)
123 p = strchr(tk->symbol, ':');
127 rcu_read_lock_sched();
128 ret = !!find_module(tk->symbol);
129 rcu_read_unlock_sched();
135 static bool trace_kprobe_is_busy(struct dyn_event *ev)
137 struct trace_kprobe *tk = to_trace_kprobe(ev);
139 return trace_probe_is_enabled(&tk->tp);
142 static bool trace_kprobe_match_command_head(struct trace_kprobe *tk,
143 int argc, const char **argv)
145 char buf[MAX_ARGSTR_LEN + 1];
151 snprintf(buf, sizeof(buf), "0x%p", tk->rp.kp.addr);
152 else if (tk->rp.kp.offset)
153 snprintf(buf, sizeof(buf), "%s+%u",
154 trace_kprobe_symbol(tk), tk->rp.kp.offset);
156 snprintf(buf, sizeof(buf), "%s", trace_kprobe_symbol(tk));
157 if (strcmp(buf, argv[0]))
161 return trace_probe_match_command_args(&tk->tp, argc, argv);
164 static bool trace_kprobe_match(const char *system, const char *event,
165 int argc, const char **argv, struct dyn_event *ev)
167 struct trace_kprobe *tk = to_trace_kprobe(ev);
169 return strcmp(trace_probe_name(&tk->tp), event) == 0 &&
170 (!system || strcmp(trace_probe_group_name(&tk->tp), system) == 0) &&
171 trace_kprobe_match_command_head(tk, argc, argv);
174 static nokprobe_inline unsigned long trace_kprobe_nhit(struct trace_kprobe *tk)
176 unsigned long nhit = 0;
179 for_each_possible_cpu(cpu)
180 nhit += *per_cpu_ptr(tk->nhit, cpu);
185 static nokprobe_inline bool trace_kprobe_is_registered(struct trace_kprobe *tk)
187 return !(list_empty(&tk->rp.kp.list) &&
188 hlist_unhashed(&tk->rp.kp.hlist));
191 /* Return 0 if it fails to find the symbol address */
192 static nokprobe_inline
193 unsigned long trace_kprobe_address(struct trace_kprobe *tk)
198 addr = (unsigned long)
199 kallsyms_lookup_name(trace_kprobe_symbol(tk));
201 addr += tk->rp.kp.offset;
203 addr = (unsigned long)tk->rp.kp.addr;
208 static nokprobe_inline struct trace_kprobe *
209 trace_kprobe_primary_from_call(struct trace_event_call *call)
211 struct trace_probe *tp;
213 tp = trace_probe_primary_from_call(call);
214 if (WARN_ON_ONCE(!tp))
217 return container_of(tp, struct trace_kprobe, tp);
220 bool trace_kprobe_on_func_entry(struct trace_event_call *call)
222 struct trace_kprobe *tk = trace_kprobe_primary_from_call(call);
224 return tk ? (kprobe_on_func_entry(tk->rp.kp.addr,
225 tk->rp.kp.addr ? NULL : tk->rp.kp.symbol_name,
226 tk->rp.kp.addr ? 0 : tk->rp.kp.offset) == 0) : false;
229 bool trace_kprobe_error_injectable(struct trace_event_call *call)
231 struct trace_kprobe *tk = trace_kprobe_primary_from_call(call);
233 return tk ? within_error_injection_list(trace_kprobe_address(tk)) :
237 static int register_kprobe_event(struct trace_kprobe *tk);
238 static int unregister_kprobe_event(struct trace_kprobe *tk);
240 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
241 static int kretprobe_dispatcher(struct kretprobe_instance *ri,
242 struct pt_regs *regs);
244 static void free_trace_kprobe(struct trace_kprobe *tk)
247 trace_probe_cleanup(&tk->tp);
249 free_percpu(tk->nhit);
255 * Allocate new trace_probe and initialize it (including kprobes).
257 static struct trace_kprobe *alloc_trace_kprobe(const char *group,
263 int nargs, bool is_return)
265 struct trace_kprobe *tk;
268 tk = kzalloc(SIZEOF_TRACE_KPROBE(nargs), GFP_KERNEL);
272 tk->nhit = alloc_percpu(unsigned long);
277 tk->symbol = kstrdup(symbol, GFP_KERNEL);
280 tk->rp.kp.symbol_name = tk->symbol;
281 tk->rp.kp.offset = offs;
283 tk->rp.kp.addr = addr;
286 tk->rp.handler = kretprobe_dispatcher;
288 tk->rp.kp.pre_handler = kprobe_dispatcher;
290 tk->rp.maxactive = maxactive;
291 INIT_HLIST_NODE(&tk->rp.kp.hlist);
292 INIT_LIST_HEAD(&tk->rp.kp.list);
294 ret = trace_probe_init(&tk->tp, event, group, false);
298 dyn_event_init(&tk->devent, &trace_kprobe_ops);
301 free_trace_kprobe(tk);
305 static struct trace_kprobe *find_trace_kprobe(const char *event,
308 struct dyn_event *pos;
309 struct trace_kprobe *tk;
311 for_each_trace_kprobe(tk, pos)
312 if (strcmp(trace_probe_name(&tk->tp), event) == 0 &&
313 strcmp(trace_probe_group_name(&tk->tp), group) == 0)
318 static inline int __enable_trace_kprobe(struct trace_kprobe *tk)
322 if (trace_kprobe_is_registered(tk) && !trace_kprobe_has_gone(tk)) {
323 if (trace_kprobe_is_return(tk))
324 ret = enable_kretprobe(&tk->rp);
326 ret = enable_kprobe(&tk->rp.kp);
332 static void __disable_trace_kprobe(struct trace_probe *tp)
334 struct trace_probe *pos;
335 struct trace_kprobe *tk;
337 list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
338 tk = container_of(pos, struct trace_kprobe, tp);
339 if (!trace_kprobe_is_registered(tk))
341 if (trace_kprobe_is_return(tk))
342 disable_kretprobe(&tk->rp);
344 disable_kprobe(&tk->rp.kp);
350 * if the file is NULL, enable "perf" handler, or enable "trace" handler.
352 static int enable_trace_kprobe(struct trace_event_call *call,
353 struct trace_event_file *file)
355 struct trace_probe *pos, *tp;
356 struct trace_kprobe *tk;
360 tp = trace_probe_primary_from_call(call);
361 if (WARN_ON_ONCE(!tp))
363 enabled = trace_probe_is_enabled(tp);
365 /* This also changes "enabled" state */
367 ret = trace_probe_add_file(tp, file);
371 trace_probe_set_flag(tp, TP_FLAG_PROFILE);
376 list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
377 tk = container_of(pos, struct trace_kprobe, tp);
378 if (trace_kprobe_has_gone(tk))
380 ret = __enable_trace_kprobe(tk);
387 /* Failed to enable one of them. Roll back all */
389 __disable_trace_kprobe(tp);
391 trace_probe_remove_file(tp, file);
393 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
400 * Disable trace_probe
401 * if the file is NULL, disable "perf" handler, or disable "trace" handler.
403 static int disable_trace_kprobe(struct trace_event_call *call,
404 struct trace_event_file *file)
406 struct trace_probe *tp;
408 tp = trace_probe_primary_from_call(call);
409 if (WARN_ON_ONCE(!tp))
413 if (!trace_probe_get_file_link(tp, file))
415 if (!trace_probe_has_single_file(tp))
417 trace_probe_clear_flag(tp, TP_FLAG_TRACE);
419 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
421 if (!trace_probe_is_enabled(tp))
422 __disable_trace_kprobe(tp);
427 * Synchronization is done in below function. For perf event,
428 * file == NULL and perf_trace_event_unreg() calls
429 * tracepoint_synchronize_unregister() to ensure synchronize
430 * event. We don't need to care about it.
432 trace_probe_remove_file(tp, file);
437 #if defined(CONFIG_DYNAMIC_FTRACE) && \
438 !defined(CONFIG_KPROBE_EVENTS_ON_NOTRACE)
439 static bool __within_notrace_func(unsigned long addr)
441 unsigned long offset, size;
443 if (!addr || !kallsyms_lookup_size_offset(addr, &size, &offset))
446 /* Get the entry address of the target function */
450 * Since ftrace_location_range() does inclusive range check, we need
451 * to subtract 1 byte from the end address.
453 return !ftrace_location_range(addr, addr + size - 1);
456 static bool within_notrace_func(struct trace_kprobe *tk)
458 unsigned long addr = trace_kprobe_address(tk);
459 char symname[KSYM_NAME_LEN], *p;
461 if (!__within_notrace_func(addr))
464 /* Check if the address is on a suffixed-symbol */
465 if (!lookup_symbol_name(addr, symname)) {
466 p = strchr(symname, '.');
470 addr = (unsigned long)kprobe_lookup_name(symname, 0);
472 return __within_notrace_func(addr);
478 #define within_notrace_func(tk) (false)
481 /* Internal register function - just handle k*probes and flags */
482 static int __register_trace_kprobe(struct trace_kprobe *tk)
486 ret = security_locked_down(LOCKDOWN_KPROBES);
490 if (trace_kprobe_is_registered(tk))
493 if (within_notrace_func(tk)) {
494 pr_warn("Could not probe notrace function %s\n",
495 trace_kprobe_symbol(tk));
499 for (i = 0; i < tk->tp.nr_args; i++) {
500 ret = traceprobe_update_arg(&tk->tp.args[i]);
505 /* Set/clear disabled flag according to tp->flag */
506 if (trace_probe_is_enabled(&tk->tp))
507 tk->rp.kp.flags &= ~KPROBE_FLAG_DISABLED;
509 tk->rp.kp.flags |= KPROBE_FLAG_DISABLED;
511 if (trace_kprobe_is_return(tk))
512 ret = register_kretprobe(&tk->rp);
514 ret = register_kprobe(&tk->rp.kp);
519 /* Internal unregister function - just handle k*probes and flags */
520 static void __unregister_trace_kprobe(struct trace_kprobe *tk)
522 if (trace_kprobe_is_registered(tk)) {
523 if (trace_kprobe_is_return(tk))
524 unregister_kretprobe(&tk->rp);
526 unregister_kprobe(&tk->rp.kp);
527 /* Cleanup kprobe for reuse and mark it unregistered */
528 INIT_HLIST_NODE(&tk->rp.kp.hlist);
529 INIT_LIST_HEAD(&tk->rp.kp.list);
530 if (tk->rp.kp.symbol_name)
531 tk->rp.kp.addr = NULL;
535 /* Unregister a trace_probe and probe_event */
536 static int unregister_trace_kprobe(struct trace_kprobe *tk)
538 /* If other probes are on the event, just unregister kprobe */
539 if (trace_probe_has_sibling(&tk->tp))
542 /* Enabled event can not be unregistered */
543 if (trace_probe_is_enabled(&tk->tp))
546 /* If there's a reference to the dynamic event */
547 if (trace_event_dyn_busy(trace_probe_event_call(&tk->tp)))
550 /* Will fail if probe is being used by ftrace or perf */
551 if (unregister_kprobe_event(tk))
555 __unregister_trace_kprobe(tk);
556 dyn_event_remove(&tk->devent);
557 trace_probe_unlink(&tk->tp);
562 static bool trace_kprobe_has_same_kprobe(struct trace_kprobe *orig,
563 struct trace_kprobe *comp)
565 struct trace_probe_event *tpe = orig->tp.event;
566 struct trace_probe *pos;
569 list_for_each_entry(pos, &tpe->probes, list) {
570 orig = container_of(pos, struct trace_kprobe, tp);
571 if (strcmp(trace_kprobe_symbol(orig),
572 trace_kprobe_symbol(comp)) ||
573 trace_kprobe_offset(orig) != trace_kprobe_offset(comp))
577 * trace_probe_compare_arg_type() ensured that nr_args and
578 * each argument name and type are same. Let's compare comm.
580 for (i = 0; i < orig->tp.nr_args; i++) {
581 if (strcmp(orig->tp.args[i].comm,
582 comp->tp.args[i].comm))
586 if (i == orig->tp.nr_args)
593 static int append_trace_kprobe(struct trace_kprobe *tk, struct trace_kprobe *to)
597 ret = trace_probe_compare_arg_type(&tk->tp, &to->tp);
599 /* Note that argument starts index = 2 */
600 trace_probe_log_set_index(ret + 1);
601 trace_probe_log_err(0, DIFF_ARG_TYPE);
604 if (trace_kprobe_has_same_kprobe(to, tk)) {
605 trace_probe_log_set_index(0);
606 trace_probe_log_err(0, SAME_PROBE);
610 /* Append to existing event */
611 ret = trace_probe_append(&tk->tp, &to->tp);
615 /* Register k*probe */
616 ret = __register_trace_kprobe(tk);
617 if (ret == -ENOENT && !trace_kprobe_module_exist(tk)) {
618 pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
623 trace_probe_unlink(&tk->tp);
625 dyn_event_add(&tk->devent, trace_probe_event_call(&tk->tp));
630 /* Register a trace_probe and probe_event */
631 static int register_trace_kprobe(struct trace_kprobe *tk)
633 struct trace_kprobe *old_tk;
636 mutex_lock(&event_mutex);
638 old_tk = find_trace_kprobe(trace_probe_name(&tk->tp),
639 trace_probe_group_name(&tk->tp));
641 if (trace_kprobe_is_return(tk) != trace_kprobe_is_return(old_tk)) {
642 trace_probe_log_set_index(0);
643 trace_probe_log_err(0, DIFF_PROBE_TYPE);
646 ret = append_trace_kprobe(tk, old_tk);
651 /* Register new event */
652 ret = register_kprobe_event(tk);
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 /* Convert certain expected symbols into '_' when generating event names */
711 static inline void sanitize_event_name(char *name)
713 while (*name++ != '\0')
714 if (*name == ':' || *name == '.')
718 static int __trace_kprobe_create(int argc, const char *argv[])
723 * p[:[GRP/]EVENT] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
725 * r[MAXACTIVE][:[GRP/]EVENT] [MOD:]KSYM[+0] [FETCHARGS]
727 * p:[GRP/]EVENT] [MOD:]KSYM[+0]%return [FETCHARGS]
730 * $retval : fetch return value
731 * $stack : fetch stack address
732 * $stackN : fetch Nth of stack (N:0-)
733 * $comm : fetch current task comm
734 * @ADDR : fetch memory at ADDR (ADDR should be in kernel)
735 * @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
736 * %REG : fetch register REG
737 * Dereferencing memory fetch:
738 * +|-offs(ARG) : fetch memory at ARG +|- offs address.
739 * Alias name of args:
740 * NAME=FETCHARG : set NAME as alias of FETCHARG.
742 * FETCHARG:TYPE : use TYPE instead of unsigned long.
744 struct trace_kprobe *tk = NULL;
746 bool is_return = false;
747 char *symbol = NULL, *tmp = NULL;
748 const char *event = NULL, *group = KPROBE_EVENT_SYSTEM;
752 char buf[MAX_EVENT_NAME_LEN];
753 unsigned int flags = TPARG_FL_KERNEL;
755 switch (argv[0][0]) {
767 trace_probe_log_init("trace_kprobe", argc, argv);
769 event = strchr(&argv[0][1], ':');
773 if (isdigit(argv[0][1])) {
775 trace_probe_log_err(1, MAXACT_NO_KPROBE);
779 len = event - &argv[0][1] - 1;
781 len = strlen(&argv[0][1]);
782 if (len > MAX_EVENT_NAME_LEN - 1) {
783 trace_probe_log_err(1, BAD_MAXACT);
786 memcpy(buf, &argv[0][1], len);
788 ret = kstrtouint(buf, 0, &maxactive);
789 if (ret || !maxactive) {
790 trace_probe_log_err(1, BAD_MAXACT);
793 /* kretprobes instances are iterated over via a list. The
794 * maximum should stay reasonable.
796 if (maxactive > KRETPROBE_MAXACTIVE_MAX) {
797 trace_probe_log_err(1, MAXACT_TOO_BIG);
802 /* try to parse an address. if that fails, try to read the
803 * input as a symbol. */
804 if (kstrtoul(argv[1], 0, (unsigned long *)&addr)) {
805 trace_probe_log_set_index(1);
806 /* Check whether uprobe event specified */
807 if (strchr(argv[1], '/') && strchr(argv[1], ':')) {
811 /* a symbol specified */
812 symbol = kstrdup(argv[1], GFP_KERNEL);
816 tmp = strchr(symbol, '%');
818 if (!strcmp(tmp, "%return")) {
822 trace_probe_log_err(tmp - symbol, BAD_ADDR_SUFFIX);
827 /* TODO: support .init module functions */
828 ret = traceprobe_split_symbol_offset(symbol, &offset);
829 if (ret || offset < 0 || offset > UINT_MAX) {
830 trace_probe_log_err(0, BAD_PROBE_ADDR);
834 flags |= TPARG_FL_RETURN;
835 ret = kprobe_on_func_entry(NULL, symbol, offset);
837 flags |= TPARG_FL_FENTRY;
838 /* Defer the ENOENT case until register kprobe */
839 if (ret == -EINVAL && is_return) {
840 trace_probe_log_err(0, BAD_RETPROBE);
845 trace_probe_log_set_index(0);
847 ret = traceprobe_parse_event_name(&event, &group, buf,
852 /* Make a new event name */
854 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
855 is_return ? 'r' : 'p', symbol, offset);
857 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
858 is_return ? 'r' : 'p', addr);
859 sanitize_event_name(buf);
864 tk = alloc_trace_kprobe(group, event, addr, symbol, offset, maxactive,
865 argc - 2, is_return);
868 /* This must return -ENOMEM, else there is a bug */
869 WARN_ON_ONCE(ret != -ENOMEM);
870 goto out; /* We know tk is not allocated */
872 argc -= 2; argv += 2;
874 /* parse arguments */
875 for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
876 tmp = kstrdup(argv[i], GFP_KERNEL);
882 trace_probe_log_set_index(i + 2);
883 ret = traceprobe_parse_probe_arg(&tk->tp, i, tmp, flags);
886 goto error; /* This can be -ENOMEM */
889 ret = traceprobe_set_print_fmt(&tk->tp, is_return);
893 ret = register_trace_kprobe(tk);
895 trace_probe_log_set_index(1);
897 trace_probe_log_err(0, BAD_INSN_BNDRY);
898 else if (ret == -ENOENT)
899 trace_probe_log_err(0, BAD_PROBE_ADDR);
900 else if (ret != -ENOMEM && ret != -EEXIST)
901 trace_probe_log_err(0, FAIL_REG_PROBE);
906 trace_probe_log_clear();
913 free_trace_kprobe(tk);
917 static int trace_kprobe_create(const char *raw_command)
919 return trace_probe_create(raw_command, __trace_kprobe_create);
922 static int create_or_delete_trace_kprobe(const char *raw_command)
926 if (raw_command[0] == '-')
927 return dyn_event_release(raw_command, &trace_kprobe_ops);
929 ret = trace_kprobe_create(raw_command);
930 return ret == -ECANCELED ? -EINVAL : ret;
933 static int trace_kprobe_run_command(struct dynevent_cmd *cmd)
935 return create_or_delete_trace_kprobe(cmd->seq.buffer);
939 * kprobe_event_cmd_init - Initialize a kprobe event command object
940 * @cmd: A pointer to the dynevent_cmd struct representing the new event
941 * @buf: A pointer to the buffer used to build the command
942 * @maxlen: The length of the buffer passed in @buf
944 * Initialize a synthetic event command object. Use this before
945 * calling any of the other kprobe_event functions.
947 void kprobe_event_cmd_init(struct dynevent_cmd *cmd, char *buf, int maxlen)
949 dynevent_cmd_init(cmd, buf, maxlen, DYNEVENT_TYPE_KPROBE,
950 trace_kprobe_run_command);
952 EXPORT_SYMBOL_GPL(kprobe_event_cmd_init);
955 * __kprobe_event_gen_cmd_start - Generate a kprobe event command from arg list
956 * @cmd: A pointer to the dynevent_cmd struct representing the new event
957 * @name: The name of the kprobe event
958 * @loc: The location of the kprobe event
959 * @kretprobe: Is this a return probe?
960 * @args: Variable number of arg (pairs), one pair for each field
962 * NOTE: Users normally won't want to call this function directly, but
963 * rather use the kprobe_event_gen_cmd_start() wrapper, which automatically
964 * adds a NULL to the end of the arg list. If this function is used
965 * directly, make sure the last arg in the variable arg list is NULL.
967 * Generate a kprobe event command to be executed by
968 * kprobe_event_gen_cmd_end(). This function can be used to generate the
969 * complete command or only the first part of it; in the latter case,
970 * kprobe_event_add_fields() can be used to add more fields following this.
972 * Unlikely the synth_event_gen_cmd_start(), @loc must be specified. This
973 * returns -EINVAL if @loc == NULL.
975 * Return: 0 if successful, error otherwise.
977 int __kprobe_event_gen_cmd_start(struct dynevent_cmd *cmd, bool kretprobe,
978 const char *name, const char *loc, ...)
980 char buf[MAX_EVENT_NAME_LEN];
981 struct dynevent_arg arg;
985 if (cmd->type != DYNEVENT_TYPE_KPROBE)
992 snprintf(buf, MAX_EVENT_NAME_LEN, "r:kprobes/%s", name);
994 snprintf(buf, MAX_EVENT_NAME_LEN, "p:kprobes/%s", name);
996 ret = dynevent_str_add(cmd, buf);
1000 dynevent_arg_init(&arg, 0);
1002 ret = dynevent_arg_add(cmd, &arg, NULL);
1006 va_start(args, loc);
1010 field = va_arg(args, const char *);
1014 if (++cmd->n_fields > MAX_TRACE_ARGS) {
1020 ret = dynevent_arg_add(cmd, &arg, NULL);
1028 EXPORT_SYMBOL_GPL(__kprobe_event_gen_cmd_start);
1031 * __kprobe_event_add_fields - Add probe fields to a kprobe command from arg list
1032 * @cmd: A pointer to the dynevent_cmd struct representing the new event
1033 * @args: Variable number of arg (pairs), one pair for each field
1035 * NOTE: Users normally won't want to call this function directly, but
1036 * rather use the kprobe_event_add_fields() wrapper, which
1037 * automatically adds a NULL to the end of the arg list. If this
1038 * function is used directly, make sure the last arg in the variable
1041 * Add probe fields to an existing kprobe command using a variable
1042 * list of args. Fields are added in the same order they're listed.
1044 * Return: 0 if successful, error otherwise.
1046 int __kprobe_event_add_fields(struct dynevent_cmd *cmd, ...)
1048 struct dynevent_arg arg;
1052 if (cmd->type != DYNEVENT_TYPE_KPROBE)
1055 dynevent_arg_init(&arg, 0);
1057 va_start(args, cmd);
1061 field = va_arg(args, const char *);
1065 if (++cmd->n_fields > MAX_TRACE_ARGS) {
1071 ret = dynevent_arg_add(cmd, &arg, NULL);
1079 EXPORT_SYMBOL_GPL(__kprobe_event_add_fields);
1082 * kprobe_event_delete - Delete a kprobe event
1083 * @name: The name of the kprobe event to delete
1085 * Delete a kprobe event with the give @name from kernel code rather
1086 * than directly from the command line.
1088 * Return: 0 if successful, error otherwise.
1090 int kprobe_event_delete(const char *name)
1092 char buf[MAX_EVENT_NAME_LEN];
1094 snprintf(buf, MAX_EVENT_NAME_LEN, "-:%s", name);
1096 return create_or_delete_trace_kprobe(buf);
1098 EXPORT_SYMBOL_GPL(kprobe_event_delete);
1100 static int trace_kprobe_release(struct dyn_event *ev)
1102 struct trace_kprobe *tk = to_trace_kprobe(ev);
1103 int ret = unregister_trace_kprobe(tk);
1106 free_trace_kprobe(tk);
1110 static int trace_kprobe_show(struct seq_file *m, struct dyn_event *ev)
1112 struct trace_kprobe *tk = to_trace_kprobe(ev);
1115 seq_putc(m, trace_kprobe_is_return(tk) ? 'r' : 'p');
1116 if (trace_kprobe_is_return(tk) && tk->rp.maxactive)
1117 seq_printf(m, "%d", tk->rp.maxactive);
1118 seq_printf(m, ":%s/%s", trace_probe_group_name(&tk->tp),
1119 trace_probe_name(&tk->tp));
1122 seq_printf(m, " 0x%p", tk->rp.kp.addr);
1123 else if (tk->rp.kp.offset)
1124 seq_printf(m, " %s+%u", trace_kprobe_symbol(tk),
1127 seq_printf(m, " %s", trace_kprobe_symbol(tk));
1129 for (i = 0; i < tk->tp.nr_args; i++)
1130 seq_printf(m, " %s=%s", tk->tp.args[i].name, tk->tp.args[i].comm);
1136 static int probes_seq_show(struct seq_file *m, void *v)
1138 struct dyn_event *ev = v;
1140 if (!is_trace_kprobe(ev))
1143 return trace_kprobe_show(m, ev);
1146 static const struct seq_operations probes_seq_op = {
1147 .start = dyn_event_seq_start,
1148 .next = dyn_event_seq_next,
1149 .stop = dyn_event_seq_stop,
1150 .show = probes_seq_show
1153 static int probes_open(struct inode *inode, struct file *file)
1157 ret = security_locked_down(LOCKDOWN_TRACEFS);
1161 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
1162 ret = dyn_events_release_all(&trace_kprobe_ops);
1167 return seq_open(file, &probes_seq_op);
1170 static ssize_t probes_write(struct file *file, const char __user *buffer,
1171 size_t count, loff_t *ppos)
1173 return trace_parse_run_command(file, buffer, count, ppos,
1174 create_or_delete_trace_kprobe);
1177 static const struct file_operations kprobe_events_ops = {
1178 .owner = THIS_MODULE,
1179 .open = probes_open,
1181 .llseek = seq_lseek,
1182 .release = seq_release,
1183 .write = probes_write,
1186 /* Probes profiling interfaces */
1187 static int probes_profile_seq_show(struct seq_file *m, void *v)
1189 struct dyn_event *ev = v;
1190 struct trace_kprobe *tk;
1192 if (!is_trace_kprobe(ev))
1195 tk = to_trace_kprobe(ev);
1196 seq_printf(m, " %-44s %15lu %15lu\n",
1197 trace_probe_name(&tk->tp),
1198 trace_kprobe_nhit(tk),
1204 static const struct seq_operations profile_seq_op = {
1205 .start = dyn_event_seq_start,
1206 .next = dyn_event_seq_next,
1207 .stop = dyn_event_seq_stop,
1208 .show = probes_profile_seq_show
1211 static int profile_open(struct inode *inode, struct file *file)
1215 ret = security_locked_down(LOCKDOWN_TRACEFS);
1219 return seq_open(file, &profile_seq_op);
1222 static const struct file_operations kprobe_profile_ops = {
1223 .owner = THIS_MODULE,
1224 .open = profile_open,
1226 .llseek = seq_lseek,
1227 .release = seq_release,
1230 /* Kprobe specific fetch functions */
1232 /* Return the length of string -- including null terminal byte */
1233 static nokprobe_inline int
1234 fetch_store_strlen_user(unsigned long addr)
1236 const void __user *uaddr = (__force const void __user *)addr;
1238 return strnlen_user_nofault(uaddr, MAX_STRING_SIZE);
1241 /* Return the length of string -- including null terminal byte */
1242 static nokprobe_inline int
1243 fetch_store_strlen(unsigned long addr)
1248 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
1249 if (addr < TASK_SIZE)
1250 return fetch_store_strlen_user(addr);
1254 ret = copy_from_kernel_nofault(&c, (u8 *)addr + len, 1);
1256 } while (c && ret == 0 && len < MAX_STRING_SIZE);
1258 return (ret < 0) ? ret : len;
1262 * Fetch a null-terminated string from user. Caller MUST set *(u32 *)buf
1263 * with max length and relative data location.
1265 static nokprobe_inline int
1266 fetch_store_string_user(unsigned long addr, void *dest, void *base)
1268 const void __user *uaddr = (__force const void __user *)addr;
1269 int maxlen = get_loc_len(*(u32 *)dest);
1273 if (unlikely(!maxlen))
1276 __dest = get_loc_data(dest, base);
1278 ret = strncpy_from_user_nofault(__dest, uaddr, maxlen);
1280 *(u32 *)dest = make_data_loc(ret, __dest - base);
1286 * Fetch a null-terminated string. Caller MUST set *(u32 *)buf with max
1287 * length and relative data location.
1289 static nokprobe_inline int
1290 fetch_store_string(unsigned long addr, void *dest, void *base)
1292 int maxlen = get_loc_len(*(u32 *)dest);
1296 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
1297 if ((unsigned long)addr < TASK_SIZE)
1298 return fetch_store_string_user(addr, dest, base);
1301 if (unlikely(!maxlen))
1304 __dest = get_loc_data(dest, base);
1307 * Try to get string again, since the string can be changed while
1310 ret = strncpy_from_kernel_nofault(__dest, (void *)addr, maxlen);
1312 *(u32 *)dest = make_data_loc(ret, __dest - base);
1317 static nokprobe_inline int
1318 probe_mem_read_user(void *dest, void *src, size_t size)
1320 const void __user *uaddr = (__force const void __user *)src;
1322 return copy_from_user_nofault(dest, uaddr, size);
1325 static nokprobe_inline int
1326 probe_mem_read(void *dest, void *src, size_t size)
1328 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
1329 if ((unsigned long)src < TASK_SIZE)
1330 return probe_mem_read_user(dest, src, size);
1332 return copy_from_kernel_nofault(dest, src, size);
1335 /* Note that we don't verify it, since the code does not come from user space */
1337 process_fetch_insn(struct fetch_insn *code, struct pt_regs *regs, void *dest,
1343 /* 1st stage: get value from context */
1346 val = regs_get_register(regs, code->param);
1348 case FETCH_OP_STACK:
1349 val = regs_get_kernel_stack_nth(regs, code->param);
1351 case FETCH_OP_STACKP:
1352 val = kernel_stack_pointer(regs);
1354 case FETCH_OP_RETVAL:
1355 val = regs_return_value(regs);
1358 val = code->immediate;
1361 val = (unsigned long)current->comm;
1364 val = (unsigned long)code->data;
1366 #ifdef CONFIG_HAVE_FUNCTION_ARG_ACCESS_API
1368 val = regs_get_kernel_argument(regs, code->param);
1371 case FETCH_NOP_SYMBOL: /* Ignore a place holder */
1379 return process_fetch_insn_bottom(code, val, dest, base);
1381 NOKPROBE_SYMBOL(process_fetch_insn)
1383 /* Kprobe handler */
1384 static nokprobe_inline void
1385 __kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs,
1386 struct trace_event_file *trace_file)
1388 struct kprobe_trace_entry_head *entry;
1389 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1390 struct trace_event_buffer fbuffer;
1393 WARN_ON(call != trace_file->event_call);
1395 if (trace_trigger_soft_disabled(trace_file))
1398 fbuffer.trace_ctx = tracing_gen_ctx();
1399 fbuffer.trace_file = trace_file;
1401 dsize = __get_data_size(&tk->tp, regs);
1404 trace_event_buffer_lock_reserve(&fbuffer.buffer, trace_file,
1406 sizeof(*entry) + tk->tp.size + dsize,
1411 fbuffer.regs = regs;
1412 entry = fbuffer.entry = ring_buffer_event_data(fbuffer.event);
1413 entry->ip = (unsigned long)tk->rp.kp.addr;
1414 store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1416 trace_event_buffer_commit(&fbuffer);
1420 kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs)
1422 struct event_file_link *link;
1424 trace_probe_for_each_link_rcu(link, &tk->tp)
1425 __kprobe_trace_func(tk, regs, link->file);
1427 NOKPROBE_SYMBOL(kprobe_trace_func);
1429 /* Kretprobe handler */
1430 static nokprobe_inline void
1431 __kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1432 struct pt_regs *regs,
1433 struct trace_event_file *trace_file)
1435 struct kretprobe_trace_entry_head *entry;
1436 struct trace_event_buffer fbuffer;
1437 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1440 WARN_ON(call != trace_file->event_call);
1442 if (trace_trigger_soft_disabled(trace_file))
1445 fbuffer.trace_ctx = tracing_gen_ctx();
1446 fbuffer.trace_file = trace_file;
1448 dsize = __get_data_size(&tk->tp, regs);
1450 trace_event_buffer_lock_reserve(&fbuffer.buffer, trace_file,
1452 sizeof(*entry) + tk->tp.size + dsize,
1457 fbuffer.regs = regs;
1458 entry = fbuffer.entry = ring_buffer_event_data(fbuffer.event);
1459 entry->func = (unsigned long)tk->rp.kp.addr;
1460 entry->ret_ip = (unsigned long)ri->ret_addr;
1461 store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1463 trace_event_buffer_commit(&fbuffer);
1467 kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1468 struct pt_regs *regs)
1470 struct event_file_link *link;
1472 trace_probe_for_each_link_rcu(link, &tk->tp)
1473 __kretprobe_trace_func(tk, ri, regs, link->file);
1475 NOKPROBE_SYMBOL(kretprobe_trace_func);
1477 /* Event entry printers */
1478 static enum print_line_t
1479 print_kprobe_event(struct trace_iterator *iter, int flags,
1480 struct trace_event *event)
1482 struct kprobe_trace_entry_head *field;
1483 struct trace_seq *s = &iter->seq;
1484 struct trace_probe *tp;
1486 field = (struct kprobe_trace_entry_head *)iter->ent;
1487 tp = trace_probe_primary_from_call(
1488 container_of(event, struct trace_event_call, event));
1489 if (WARN_ON_ONCE(!tp))
1492 trace_seq_printf(s, "%s: (", trace_probe_name(tp));
1494 if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
1497 trace_seq_putc(s, ')');
1499 if (print_probe_args(s, tp->args, tp->nr_args,
1500 (u8 *)&field[1], field) < 0)
1503 trace_seq_putc(s, '\n');
1505 return trace_handle_return(s);
1508 static enum print_line_t
1509 print_kretprobe_event(struct trace_iterator *iter, int flags,
1510 struct trace_event *event)
1512 struct kretprobe_trace_entry_head *field;
1513 struct trace_seq *s = &iter->seq;
1514 struct trace_probe *tp;
1516 field = (struct kretprobe_trace_entry_head *)iter->ent;
1517 tp = trace_probe_primary_from_call(
1518 container_of(event, struct trace_event_call, event));
1519 if (WARN_ON_ONCE(!tp))
1522 trace_seq_printf(s, "%s: (", trace_probe_name(tp));
1524 if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1527 trace_seq_puts(s, " <- ");
1529 if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1532 trace_seq_putc(s, ')');
1534 if (print_probe_args(s, tp->args, tp->nr_args,
1535 (u8 *)&field[1], field) < 0)
1538 trace_seq_putc(s, '\n');
1541 return trace_handle_return(s);
1545 static int kprobe_event_define_fields(struct trace_event_call *event_call)
1548 struct kprobe_trace_entry_head field;
1549 struct trace_probe *tp;
1551 tp = trace_probe_primary_from_call(event_call);
1552 if (WARN_ON_ONCE(!tp))
1555 DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1557 return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
1560 static int kretprobe_event_define_fields(struct trace_event_call *event_call)
1563 struct kretprobe_trace_entry_head field;
1564 struct trace_probe *tp;
1566 tp = trace_probe_primary_from_call(event_call);
1567 if (WARN_ON_ONCE(!tp))
1570 DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1571 DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1573 return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
1576 #ifdef CONFIG_PERF_EVENTS
1578 /* Kprobe profile handler */
1580 kprobe_perf_func(struct trace_kprobe *tk, struct pt_regs *regs)
1582 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1583 struct kprobe_trace_entry_head *entry;
1584 struct hlist_head *head;
1585 int size, __size, dsize;
1588 if (bpf_prog_array_valid(call)) {
1589 unsigned long orig_ip = instruction_pointer(regs);
1592 ret = trace_call_bpf(call, regs);
1595 * We need to check and see if we modified the pc of the
1596 * pt_regs, and if so return 1 so that we don't do the
1599 if (orig_ip != instruction_pointer(regs))
1605 head = this_cpu_ptr(call->perf_events);
1606 if (hlist_empty(head))
1609 dsize = __get_data_size(&tk->tp, regs);
1610 __size = sizeof(*entry) + tk->tp.size + dsize;
1611 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1612 size -= sizeof(u32);
1614 entry = perf_trace_buf_alloc(size, NULL, &rctx);
1618 entry->ip = (unsigned long)tk->rp.kp.addr;
1619 memset(&entry[1], 0, dsize);
1620 store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1621 perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1625 NOKPROBE_SYMBOL(kprobe_perf_func);
1627 /* Kretprobe profile handler */
1629 kretprobe_perf_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1630 struct pt_regs *regs)
1632 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1633 struct kretprobe_trace_entry_head *entry;
1634 struct hlist_head *head;
1635 int size, __size, dsize;
1638 if (bpf_prog_array_valid(call) && !trace_call_bpf(call, regs))
1641 head = this_cpu_ptr(call->perf_events);
1642 if (hlist_empty(head))
1645 dsize = __get_data_size(&tk->tp, regs);
1646 __size = sizeof(*entry) + tk->tp.size + dsize;
1647 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1648 size -= sizeof(u32);
1650 entry = perf_trace_buf_alloc(size, NULL, &rctx);
1654 entry->func = (unsigned long)tk->rp.kp.addr;
1655 entry->ret_ip = (unsigned long)ri->ret_addr;
1656 store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1657 perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1660 NOKPROBE_SYMBOL(kretprobe_perf_func);
1662 int bpf_get_kprobe_info(const struct perf_event *event, u32 *fd_type,
1663 const char **symbol, u64 *probe_offset,
1664 u64 *probe_addr, bool perf_type_tracepoint)
1666 const char *pevent = trace_event_name(event->tp_event);
1667 const char *group = event->tp_event->class->system;
1668 struct trace_kprobe *tk;
1670 if (perf_type_tracepoint)
1671 tk = find_trace_kprobe(pevent, group);
1673 tk = trace_kprobe_primary_from_call(event->tp_event);
1677 *fd_type = trace_kprobe_is_return(tk) ? BPF_FD_TYPE_KRETPROBE
1678 : BPF_FD_TYPE_KPROBE;
1680 *symbol = tk->symbol;
1681 *probe_offset = tk->rp.kp.offset;
1686 *probe_addr = (unsigned long)tk->rp.kp.addr;
1690 #endif /* CONFIG_PERF_EVENTS */
1693 * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex.
1695 * kprobe_trace_self_tests_init() does enable_trace_probe/disable_trace_probe
1696 * lockless, but we can't race with this __init function.
1698 static int kprobe_register(struct trace_event_call *event,
1699 enum trace_reg type, void *data)
1701 struct trace_event_file *file = data;
1704 case TRACE_REG_REGISTER:
1705 return enable_trace_kprobe(event, file);
1706 case TRACE_REG_UNREGISTER:
1707 return disable_trace_kprobe(event, file);
1709 #ifdef CONFIG_PERF_EVENTS
1710 case TRACE_REG_PERF_REGISTER:
1711 return enable_trace_kprobe(event, NULL);
1712 case TRACE_REG_PERF_UNREGISTER:
1713 return disable_trace_kprobe(event, NULL);
1714 case TRACE_REG_PERF_OPEN:
1715 case TRACE_REG_PERF_CLOSE:
1716 case TRACE_REG_PERF_ADD:
1717 case TRACE_REG_PERF_DEL:
1724 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1726 struct trace_kprobe *tk = container_of(kp, struct trace_kprobe, rp.kp);
1729 raw_cpu_inc(*tk->nhit);
1731 if (trace_probe_test_flag(&tk->tp, TP_FLAG_TRACE))
1732 kprobe_trace_func(tk, regs);
1733 #ifdef CONFIG_PERF_EVENTS
1734 if (trace_probe_test_flag(&tk->tp, TP_FLAG_PROFILE))
1735 ret = kprobe_perf_func(tk, regs);
1739 NOKPROBE_SYMBOL(kprobe_dispatcher);
1742 kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1744 struct kretprobe *rp = get_kretprobe(ri);
1745 struct trace_kprobe *tk = container_of(rp, struct trace_kprobe, rp);
1747 raw_cpu_inc(*tk->nhit);
1749 if (trace_probe_test_flag(&tk->tp, TP_FLAG_TRACE))
1750 kretprobe_trace_func(tk, ri, regs);
1751 #ifdef CONFIG_PERF_EVENTS
1752 if (trace_probe_test_flag(&tk->tp, TP_FLAG_PROFILE))
1753 kretprobe_perf_func(tk, ri, regs);
1755 return 0; /* We don't tweak kernel, so just return 0 */
1757 NOKPROBE_SYMBOL(kretprobe_dispatcher);
1759 static struct trace_event_functions kretprobe_funcs = {
1760 .trace = print_kretprobe_event
1763 static struct trace_event_functions kprobe_funcs = {
1764 .trace = print_kprobe_event
1767 static struct trace_event_fields kretprobe_fields_array[] = {
1768 { .type = TRACE_FUNCTION_TYPE,
1769 .define_fields = kretprobe_event_define_fields },
1773 static struct trace_event_fields kprobe_fields_array[] = {
1774 { .type = TRACE_FUNCTION_TYPE,
1775 .define_fields = kprobe_event_define_fields },
1779 static inline void init_trace_event_call(struct trace_kprobe *tk)
1781 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1783 if (trace_kprobe_is_return(tk)) {
1784 call->event.funcs = &kretprobe_funcs;
1785 call->class->fields_array = kretprobe_fields_array;
1787 call->event.funcs = &kprobe_funcs;
1788 call->class->fields_array = kprobe_fields_array;
1791 call->flags = TRACE_EVENT_FL_KPROBE;
1792 call->class->reg = kprobe_register;
1795 static int register_kprobe_event(struct trace_kprobe *tk)
1797 init_trace_event_call(tk);
1799 return trace_probe_register_event_call(&tk->tp);
1802 static int unregister_kprobe_event(struct trace_kprobe *tk)
1804 return trace_probe_unregister_event_call(&tk->tp);
1807 #ifdef CONFIG_PERF_EVENTS
1808 /* create a trace_kprobe, but don't add it to global lists */
1809 struct trace_event_call *
1810 create_local_trace_kprobe(char *func, void *addr, unsigned long offs,
1813 struct trace_kprobe *tk;
1818 * local trace_kprobes are not added to dyn_event, so they are never
1819 * searched in find_trace_kprobe(). Therefore, there is no concern of
1820 * duplicated name here.
1822 event = func ? func : "DUMMY_EVENT";
1824 tk = alloc_trace_kprobe(KPROBE_EVENT_SYSTEM, event, (void *)addr, func,
1825 offs, 0 /* maxactive */, 0 /* nargs */,
1829 pr_info("Failed to allocate trace_probe.(%d)\n",
1831 return ERR_CAST(tk);
1834 init_trace_event_call(tk);
1836 if (traceprobe_set_print_fmt(&tk->tp, trace_kprobe_is_return(tk)) < 0) {
1841 ret = __register_trace_kprobe(tk);
1845 return trace_probe_event_call(&tk->tp);
1847 free_trace_kprobe(tk);
1848 return ERR_PTR(ret);
1851 void destroy_local_trace_kprobe(struct trace_event_call *event_call)
1853 struct trace_kprobe *tk;
1855 tk = trace_kprobe_primary_from_call(event_call);
1859 if (trace_probe_is_enabled(&tk->tp)) {
1864 __unregister_trace_kprobe(tk);
1866 free_trace_kprobe(tk);
1868 #endif /* CONFIG_PERF_EVENTS */
1870 static __init void enable_boot_kprobe_events(void)
1872 struct trace_array *tr = top_trace_array();
1873 struct trace_event_file *file;
1874 struct trace_kprobe *tk;
1875 struct dyn_event *pos;
1877 mutex_lock(&event_mutex);
1878 for_each_trace_kprobe(tk, pos) {
1879 list_for_each_entry(file, &tr->events, list)
1880 if (file->event_call == trace_probe_event_call(&tk->tp))
1881 trace_event_enable_disable(file, 1, 0);
1883 mutex_unlock(&event_mutex);
1886 static __init void setup_boot_kprobe_events(void)
1888 char *p, *cmd = kprobe_boot_events_buf;
1891 strreplace(kprobe_boot_events_buf, ',', ' ');
1893 while (cmd && *cmd != '\0') {
1894 p = strchr(cmd, ';');
1898 ret = create_or_delete_trace_kprobe(cmd);
1900 pr_warn("Failed to add event(%d): %s\n", ret, cmd);
1905 enable_boot_kprobe_events();
1909 * Register dynevent at core_initcall. This allows kernel to setup kprobe
1910 * events in postcore_initcall without tracefs.
1912 static __init int init_kprobe_trace_early(void)
1916 ret = dyn_event_register(&trace_kprobe_ops);
1920 if (register_module_notifier(&trace_kprobe_module_nb))
1925 core_initcall(init_kprobe_trace_early);
1927 /* Make a tracefs interface for controlling probe points */
1928 static __init int init_kprobe_trace(void)
1931 struct dentry *entry;
1933 ret = tracing_init_dentry();
1937 entry = tracefs_create_file("kprobe_events", 0644, NULL,
1938 NULL, &kprobe_events_ops);
1940 /* Event list interface */
1942 pr_warn("Could not create tracefs 'kprobe_events' entry\n");
1944 /* Profile interface */
1945 entry = tracefs_create_file("kprobe_profile", 0444, NULL,
1946 NULL, &kprobe_profile_ops);
1949 pr_warn("Could not create tracefs 'kprobe_profile' entry\n");
1951 setup_boot_kprobe_events();
1955 fs_initcall(init_kprobe_trace);
1958 #ifdef CONFIG_FTRACE_STARTUP_TEST
1959 static __init struct trace_event_file *
1960 find_trace_probe_file(struct trace_kprobe *tk, struct trace_array *tr)
1962 struct trace_event_file *file;
1964 list_for_each_entry(file, &tr->events, list)
1965 if (file->event_call == trace_probe_event_call(&tk->tp))
1972 * Nobody but us can call enable_trace_kprobe/disable_trace_kprobe at this
1973 * stage, we can do this lockless.
1975 static __init int kprobe_trace_self_tests_init(void)
1978 int (*target)(int, int, int, int, int, int);
1979 struct trace_kprobe *tk;
1980 struct trace_event_file *file;
1982 if (tracing_is_disabled())
1985 if (tracing_selftest_disabled)
1988 target = kprobe_trace_selftest_target;
1990 pr_info("Testing kprobe tracing: ");
1992 ret = create_or_delete_trace_kprobe("p:testprobe kprobe_trace_selftest_target $stack $stack0 +0($stack)");
1993 if (WARN_ON_ONCE(ret)) {
1994 pr_warn("error on probing function entry.\n");
1997 /* Enable trace point */
1998 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1999 if (WARN_ON_ONCE(tk == NULL)) {
2000 pr_warn("error on getting new probe.\n");
2003 file = find_trace_probe_file(tk, top_trace_array());
2004 if (WARN_ON_ONCE(file == NULL)) {
2005 pr_warn("error on getting probe file.\n");
2008 enable_trace_kprobe(
2009 trace_probe_event_call(&tk->tp), file);
2013 ret = create_or_delete_trace_kprobe("r:testprobe2 kprobe_trace_selftest_target $retval");
2014 if (WARN_ON_ONCE(ret)) {
2015 pr_warn("error on probing function return.\n");
2018 /* Enable trace point */
2019 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
2020 if (WARN_ON_ONCE(tk == NULL)) {
2021 pr_warn("error on getting 2nd new probe.\n");
2024 file = find_trace_probe_file(tk, top_trace_array());
2025 if (WARN_ON_ONCE(file == NULL)) {
2026 pr_warn("error on getting probe file.\n");
2029 enable_trace_kprobe(
2030 trace_probe_event_call(&tk->tp), file);
2037 ret = target(1, 2, 3, 4, 5, 6);
2040 * Not expecting an error here, the check is only to prevent the
2041 * optimizer from removing the call to target() as otherwise there
2042 * are no side-effects and the call is never performed.
2047 /* Disable trace points before removing it */
2048 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
2049 if (WARN_ON_ONCE(tk == NULL)) {
2050 pr_warn("error on getting test probe.\n");
2053 if (trace_kprobe_nhit(tk) != 1) {
2054 pr_warn("incorrect number of testprobe hits\n");
2058 file = find_trace_probe_file(tk, top_trace_array());
2059 if (WARN_ON_ONCE(file == NULL)) {
2060 pr_warn("error on getting probe file.\n");
2063 disable_trace_kprobe(
2064 trace_probe_event_call(&tk->tp), file);
2067 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
2068 if (WARN_ON_ONCE(tk == NULL)) {
2069 pr_warn("error on getting 2nd test probe.\n");
2072 if (trace_kprobe_nhit(tk) != 1) {
2073 pr_warn("incorrect number of testprobe2 hits\n");
2077 file = find_trace_probe_file(tk, top_trace_array());
2078 if (WARN_ON_ONCE(file == NULL)) {
2079 pr_warn("error on getting probe file.\n");
2082 disable_trace_kprobe(
2083 trace_probe_event_call(&tk->tp), file);
2086 ret = create_or_delete_trace_kprobe("-:testprobe");
2087 if (WARN_ON_ONCE(ret)) {
2088 pr_warn("error on deleting a probe.\n");
2092 ret = create_or_delete_trace_kprobe("-:testprobe2");
2093 if (WARN_ON_ONCE(ret)) {
2094 pr_warn("error on deleting a probe.\n");
2099 ret = dyn_events_release_all(&trace_kprobe_ops);
2100 if (WARN_ON_ONCE(ret)) {
2101 pr_warn("error on cleaning up probes.\n");
2105 * Wait for the optimizer work to finish. Otherwise it might fiddle
2106 * with probes in already freed __init text.
2108 wait_for_kprobe_optimizer();
2110 pr_cont("NG: Some tests are failed. Please check them.\n");
2116 late_initcall(kprobe_trace_self_tests_init);