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;
28 static bool kprobe_boot_events_enabled __initdata;
30 static int __init set_kprobe_boot_events(char *str)
32 strlcpy(kprobe_boot_events_buf, str, COMMAND_LINE_SIZE);
35 __setup("kprobe_event=", set_kprobe_boot_events);
37 static int trace_kprobe_create(int argc, const char **argv);
38 static int trace_kprobe_show(struct seq_file *m, struct dyn_event *ev);
39 static int trace_kprobe_release(struct dyn_event *ev);
40 static bool trace_kprobe_is_busy(struct dyn_event *ev);
41 static bool trace_kprobe_match(const char *system, const char *event,
42 int argc, const char **argv, struct dyn_event *ev);
44 static struct dyn_event_operations trace_kprobe_ops = {
45 .create = trace_kprobe_create,
46 .show = trace_kprobe_show,
47 .is_busy = trace_kprobe_is_busy,
48 .free = trace_kprobe_release,
49 .match = trace_kprobe_match,
53 * Kprobe event core functions
56 struct dyn_event devent;
57 struct kretprobe rp; /* Use rp.kp for kprobe use */
58 unsigned long __percpu *nhit;
59 const char *symbol; /* symbol name */
60 struct trace_probe tp;
63 static bool is_trace_kprobe(struct dyn_event *ev)
65 return ev->ops == &trace_kprobe_ops;
68 static struct trace_kprobe *to_trace_kprobe(struct dyn_event *ev)
70 return container_of(ev, struct trace_kprobe, devent);
74 * for_each_trace_kprobe - iterate over the trace_kprobe list
75 * @pos: the struct trace_kprobe * for each entry
76 * @dpos: the struct dyn_event * to use as a loop cursor
78 #define for_each_trace_kprobe(pos, dpos) \
79 for_each_dyn_event(dpos) \
80 if (is_trace_kprobe(dpos) && (pos = to_trace_kprobe(dpos)))
82 #define SIZEOF_TRACE_KPROBE(n) \
83 (offsetof(struct trace_kprobe, tp.args) + \
84 (sizeof(struct probe_arg) * (n)))
86 static nokprobe_inline bool trace_kprobe_is_return(struct trace_kprobe *tk)
88 return tk->rp.handler != NULL;
91 static nokprobe_inline const char *trace_kprobe_symbol(struct trace_kprobe *tk)
93 return tk->symbol ? tk->symbol : "unknown";
96 static nokprobe_inline unsigned long trace_kprobe_offset(struct trace_kprobe *tk)
98 return tk->rp.kp.offset;
101 static nokprobe_inline bool trace_kprobe_has_gone(struct trace_kprobe *tk)
103 return !!(kprobe_gone(&tk->rp.kp));
106 static nokprobe_inline bool trace_kprobe_within_module(struct trace_kprobe *tk,
109 int len = strlen(module_name(mod));
110 const char *name = trace_kprobe_symbol(tk);
112 return strncmp(module_name(mod), name, len) == 0 && name[len] == ':';
115 static nokprobe_inline bool trace_kprobe_module_exist(struct trace_kprobe *tk)
122 p = strchr(tk->symbol, ':');
126 mutex_lock(&module_mutex);
127 ret = !!find_module(tk->symbol);
128 mutex_unlock(&module_mutex);
134 static bool trace_kprobe_is_busy(struct dyn_event *ev)
136 struct trace_kprobe *tk = to_trace_kprobe(ev);
138 return trace_probe_is_enabled(&tk->tp);
141 static bool trace_kprobe_match_command_head(struct trace_kprobe *tk,
142 int argc, const char **argv)
144 char buf[MAX_ARGSTR_LEN + 1];
150 snprintf(buf, sizeof(buf), "0x%p", tk->rp.kp.addr);
151 else if (tk->rp.kp.offset)
152 snprintf(buf, sizeof(buf), "%s+%u",
153 trace_kprobe_symbol(tk), tk->rp.kp.offset);
155 snprintf(buf, sizeof(buf), "%s", trace_kprobe_symbol(tk));
156 if (strcmp(buf, argv[0]))
160 return trace_probe_match_command_args(&tk->tp, argc, argv);
163 static bool trace_kprobe_match(const char *system, const char *event,
164 int argc, const char **argv, struct dyn_event *ev)
166 struct trace_kprobe *tk = to_trace_kprobe(ev);
168 return 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) : 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(SIZEOF_TRACE_KPROBE(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_probe *pos;
334 struct trace_kprobe *tk;
336 list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
337 tk = container_of(pos, struct trace_kprobe, tp);
338 if (!trace_kprobe_is_registered(tk))
340 if (trace_kprobe_is_return(tk))
341 disable_kretprobe(&tk->rp);
343 disable_kprobe(&tk->rp.kp);
349 * if the file is NULL, enable "perf" handler, or enable "trace" handler.
351 static int enable_trace_kprobe(struct trace_event_call *call,
352 struct trace_event_file *file)
354 struct trace_probe *pos, *tp;
355 struct trace_kprobe *tk;
359 tp = trace_probe_primary_from_call(call);
360 if (WARN_ON_ONCE(!tp))
362 enabled = trace_probe_is_enabled(tp);
364 /* This also changes "enabled" state */
366 ret = trace_probe_add_file(tp, file);
370 trace_probe_set_flag(tp, TP_FLAG_PROFILE);
375 list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
376 tk = container_of(pos, struct trace_kprobe, tp);
377 if (trace_kprobe_has_gone(tk))
379 ret = __enable_trace_kprobe(tk);
386 /* Failed to enable one of them. Roll back all */
388 __disable_trace_kprobe(tp);
390 trace_probe_remove_file(tp, file);
392 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
399 * Disable trace_probe
400 * if the file is NULL, disable "perf" handler, or disable "trace" handler.
402 static int disable_trace_kprobe(struct trace_event_call *call,
403 struct trace_event_file *file)
405 struct trace_probe *tp;
407 tp = trace_probe_primary_from_call(call);
408 if (WARN_ON_ONCE(!tp))
412 if (!trace_probe_get_file_link(tp, file))
414 if (!trace_probe_has_single_file(tp))
416 trace_probe_clear_flag(tp, TP_FLAG_TRACE);
418 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
420 if (!trace_probe_is_enabled(tp))
421 __disable_trace_kprobe(tp);
426 * Synchronization is done in below function. For perf event,
427 * file == NULL and perf_trace_event_unreg() calls
428 * tracepoint_synchronize_unregister() to ensure synchronize
429 * event. We don't need to care about it.
431 trace_probe_remove_file(tp, file);
436 #if defined(CONFIG_KPROBES_ON_FTRACE) && \
437 !defined(CONFIG_KPROBE_EVENTS_ON_NOTRACE)
438 static bool __within_notrace_func(unsigned long addr)
440 unsigned long offset, size;
442 if (!addr || !kallsyms_lookup_size_offset(addr, &size, &offset))
445 /* Get the entry address of the target function */
449 * Since ftrace_location_range() does inclusive range check, we need
450 * to subtract 1 byte from the end address.
452 return !ftrace_location_range(addr, addr + size - 1);
455 static bool within_notrace_func(struct trace_kprobe *tk)
457 unsigned long addr = trace_kprobe_address(tk);
458 char symname[KSYM_NAME_LEN], *p;
460 if (!__within_notrace_func(addr))
463 /* Check if the address is on a suffixed-symbol */
464 if (!lookup_symbol_name(addr, symname)) {
465 p = strchr(symname, '.');
469 addr = (unsigned long)kprobe_lookup_name(symname, 0);
471 return __within_notrace_func(addr);
477 #define within_notrace_func(tk) (false)
480 /* Internal register function - just handle k*probes and flags */
481 static int __register_trace_kprobe(struct trace_kprobe *tk)
485 ret = security_locked_down(LOCKDOWN_KPROBES);
489 if (trace_kprobe_is_registered(tk))
492 if (within_notrace_func(tk)) {
493 pr_warn("Could not probe notrace function %s\n",
494 trace_kprobe_symbol(tk));
498 for (i = 0; i < tk->tp.nr_args; i++) {
499 ret = traceprobe_update_arg(&tk->tp.args[i]);
504 /* Set/clear disabled flag according to tp->flag */
505 if (trace_probe_is_enabled(&tk->tp))
506 tk->rp.kp.flags &= ~KPROBE_FLAG_DISABLED;
508 tk->rp.kp.flags |= KPROBE_FLAG_DISABLED;
510 if (trace_kprobe_is_return(tk))
511 ret = register_kretprobe(&tk->rp);
513 ret = register_kprobe(&tk->rp.kp);
518 /* Internal unregister function - just handle k*probes and flags */
519 static void __unregister_trace_kprobe(struct trace_kprobe *tk)
521 if (trace_kprobe_is_registered(tk)) {
522 if (trace_kprobe_is_return(tk))
523 unregister_kretprobe(&tk->rp);
525 unregister_kprobe(&tk->rp.kp);
526 /* Cleanup kprobe for reuse and mark it unregistered */
527 INIT_HLIST_NODE(&tk->rp.kp.hlist);
528 INIT_LIST_HEAD(&tk->rp.kp.list);
529 if (tk->rp.kp.symbol_name)
530 tk->rp.kp.addr = NULL;
534 /* Unregister a trace_probe and probe_event */
535 static int unregister_trace_kprobe(struct trace_kprobe *tk)
537 /* If other probes are on the event, just unregister kprobe */
538 if (trace_probe_has_sibling(&tk->tp))
541 /* Enabled event can not be unregistered */
542 if (trace_probe_is_enabled(&tk->tp))
545 /* Will fail if probe is being used by ftrace or perf */
546 if (unregister_kprobe_event(tk))
550 __unregister_trace_kprobe(tk);
551 dyn_event_remove(&tk->devent);
552 trace_probe_unlink(&tk->tp);
557 static bool trace_kprobe_has_same_kprobe(struct trace_kprobe *orig,
558 struct trace_kprobe *comp)
560 struct trace_probe_event *tpe = orig->tp.event;
561 struct trace_probe *pos;
564 list_for_each_entry(pos, &tpe->probes, list) {
565 orig = container_of(pos, struct trace_kprobe, tp);
566 if (strcmp(trace_kprobe_symbol(orig),
567 trace_kprobe_symbol(comp)) ||
568 trace_kprobe_offset(orig) != trace_kprobe_offset(comp))
572 * trace_probe_compare_arg_type() ensured that nr_args and
573 * each argument name and type are same. Let's compare comm.
575 for (i = 0; i < orig->tp.nr_args; i++) {
576 if (strcmp(orig->tp.args[i].comm,
577 comp->tp.args[i].comm))
581 if (i == orig->tp.nr_args)
588 static int append_trace_kprobe(struct trace_kprobe *tk, struct trace_kprobe *to)
592 ret = trace_probe_compare_arg_type(&tk->tp, &to->tp);
594 /* Note that argument starts index = 2 */
595 trace_probe_log_set_index(ret + 1);
596 trace_probe_log_err(0, DIFF_ARG_TYPE);
599 if (trace_kprobe_has_same_kprobe(to, tk)) {
600 trace_probe_log_set_index(0);
601 trace_probe_log_err(0, SAME_PROBE);
605 /* Append to existing event */
606 ret = trace_probe_append(&tk->tp, &to->tp);
610 /* Register k*probe */
611 ret = __register_trace_kprobe(tk);
612 if (ret == -ENOENT && !trace_kprobe_module_exist(tk)) {
613 pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
618 trace_probe_unlink(&tk->tp);
620 dyn_event_add(&tk->devent);
625 /* Register a trace_probe and probe_event */
626 static int register_trace_kprobe(struct trace_kprobe *tk)
628 struct trace_kprobe *old_tk;
631 mutex_lock(&event_mutex);
633 old_tk = find_trace_kprobe(trace_probe_name(&tk->tp),
634 trace_probe_group_name(&tk->tp));
636 if (trace_kprobe_is_return(tk) != trace_kprobe_is_return(old_tk)) {
637 trace_probe_log_set_index(0);
638 trace_probe_log_err(0, DIFF_PROBE_TYPE);
641 ret = append_trace_kprobe(tk, old_tk);
646 /* Register new event */
647 ret = register_kprobe_event(tk);
649 pr_warn("Failed to register probe event(%d)\n", ret);
653 /* Register k*probe */
654 ret = __register_trace_kprobe(tk);
655 if (ret == -ENOENT && !trace_kprobe_module_exist(tk)) {
656 pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
661 unregister_kprobe_event(tk);
663 dyn_event_add(&tk->devent);
666 mutex_unlock(&event_mutex);
670 /* Module notifier call back, checking event on the module */
671 static int trace_kprobe_module_callback(struct notifier_block *nb,
672 unsigned long val, void *data)
674 struct module *mod = data;
675 struct dyn_event *pos;
676 struct trace_kprobe *tk;
679 if (val != MODULE_STATE_COMING)
682 /* Update probes on coming module */
683 mutex_lock(&event_mutex);
684 for_each_trace_kprobe(tk, pos) {
685 if (trace_kprobe_within_module(tk, mod)) {
686 /* Don't need to check busy - this should have gone. */
687 __unregister_trace_kprobe(tk);
688 ret = __register_trace_kprobe(tk);
690 pr_warn("Failed to re-register probe %s on %s: %d\n",
691 trace_probe_name(&tk->tp),
692 module_name(mod), ret);
695 mutex_unlock(&event_mutex);
700 static struct notifier_block trace_kprobe_module_nb = {
701 .notifier_call = trace_kprobe_module_callback,
702 .priority = 1 /* Invoked after kprobe module callback */
705 /* Convert certain expected symbols into '_' when generating event names */
706 static inline void sanitize_event_name(char *name)
708 while (*name++ != '\0')
709 if (*name == ':' || *name == '.')
713 static int trace_kprobe_create(int argc, const char *argv[])
718 * p[:[GRP/]EVENT] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
720 * r[MAXACTIVE][:[GRP/]EVENT] [MOD:]KSYM[+0] [FETCHARGS]
722 * p:[GRP/]EVENT] [MOD:]KSYM[+0]%return [FETCHARGS]
725 * $retval : fetch return value
726 * $stack : fetch stack address
727 * $stackN : fetch Nth of stack (N:0-)
728 * $comm : fetch current task comm
729 * @ADDR : fetch memory at ADDR (ADDR should be in kernel)
730 * @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
731 * %REG : fetch register REG
732 * Dereferencing memory fetch:
733 * +|-offs(ARG) : fetch memory at ARG +|- offs address.
734 * Alias name of args:
735 * NAME=FETCHARG : set NAME as alias of FETCHARG.
737 * FETCHARG:TYPE : use TYPE instead of unsigned long.
739 struct trace_kprobe *tk = NULL;
741 bool is_return = false;
742 char *symbol = NULL, *tmp = NULL;
743 const char *event = NULL, *group = KPROBE_EVENT_SYSTEM;
747 char buf[MAX_EVENT_NAME_LEN];
748 unsigned int flags = TPARG_FL_KERNEL;
750 switch (argv[0][0]) {
762 trace_probe_log_init("trace_kprobe", argc, argv);
764 event = strchr(&argv[0][1], ':');
768 if (isdigit(argv[0][1])) {
770 trace_probe_log_err(1, MAXACT_NO_KPROBE);
774 len = event - &argv[0][1] - 1;
776 len = strlen(&argv[0][1]);
777 if (len > MAX_EVENT_NAME_LEN - 1) {
778 trace_probe_log_err(1, BAD_MAXACT);
781 memcpy(buf, &argv[0][1], len);
783 ret = kstrtouint(buf, 0, &maxactive);
784 if (ret || !maxactive) {
785 trace_probe_log_err(1, BAD_MAXACT);
788 /* kretprobes instances are iterated over via a list. The
789 * maximum should stay reasonable.
791 if (maxactive > KRETPROBE_MAXACTIVE_MAX) {
792 trace_probe_log_err(1, MAXACT_TOO_BIG);
797 /* try to parse an address. if that fails, try to read the
798 * input as a symbol. */
799 if (kstrtoul(argv[1], 0, (unsigned long *)&addr)) {
800 trace_probe_log_set_index(1);
801 /* Check whether uprobe event specified */
802 if (strchr(argv[1], '/') && strchr(argv[1], ':')) {
806 /* a symbol specified */
807 symbol = kstrdup(argv[1], GFP_KERNEL);
811 tmp = strchr(symbol, '%');
813 if (!strcmp(tmp, "%return")) {
817 trace_probe_log_err(tmp - symbol, BAD_ADDR_SUFFIX);
822 /* TODO: support .init module functions */
823 ret = traceprobe_split_symbol_offset(symbol, &offset);
824 if (ret || offset < 0 || offset > UINT_MAX) {
825 trace_probe_log_err(0, BAD_PROBE_ADDR);
829 flags |= TPARG_FL_RETURN;
830 if (kprobe_on_func_entry(NULL, symbol, offset))
831 flags |= TPARG_FL_FENTRY;
832 if (offset && is_return && !(flags & TPARG_FL_FENTRY)) {
833 trace_probe_log_err(0, BAD_RETPROBE);
838 trace_probe_log_set_index(0);
840 ret = traceprobe_parse_event_name(&event, &group, buf,
845 /* Make a new event name */
847 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
848 is_return ? 'r' : 'p', symbol, offset);
850 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
851 is_return ? 'r' : 'p', addr);
852 sanitize_event_name(buf);
857 tk = alloc_trace_kprobe(group, event, addr, symbol, offset, maxactive,
858 argc - 2, is_return);
861 /* This must return -ENOMEM, else there is a bug */
862 WARN_ON_ONCE(ret != -ENOMEM);
863 goto out; /* We know tk is not allocated */
865 argc -= 2; argv += 2;
867 /* parse arguments */
868 for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
869 tmp = kstrdup(argv[i], GFP_KERNEL);
875 trace_probe_log_set_index(i + 2);
876 ret = traceprobe_parse_probe_arg(&tk->tp, i, tmp, flags);
879 goto error; /* This can be -ENOMEM */
882 ret = traceprobe_set_print_fmt(&tk->tp, is_return);
886 ret = register_trace_kprobe(tk);
888 trace_probe_log_set_index(1);
890 trace_probe_log_err(0, BAD_INSN_BNDRY);
891 else if (ret == -ENOENT)
892 trace_probe_log_err(0, BAD_PROBE_ADDR);
893 else if (ret != -ENOMEM && ret != -EEXIST)
894 trace_probe_log_err(0, FAIL_REG_PROBE);
899 trace_probe_log_clear();
906 free_trace_kprobe(tk);
910 static int create_or_delete_trace_kprobe(int argc, char **argv)
914 if (argv[0][0] == '-')
915 return dyn_event_release(argc, argv, &trace_kprobe_ops);
917 ret = trace_kprobe_create(argc, (const char **)argv);
918 return ret == -ECANCELED ? -EINVAL : ret;
921 static int trace_kprobe_run_command(struct dynevent_cmd *cmd)
923 return trace_run_command(cmd->seq.buffer, create_or_delete_trace_kprobe);
927 * kprobe_event_cmd_init - Initialize a kprobe event command object
928 * @cmd: A pointer to the dynevent_cmd struct representing the new event
929 * @buf: A pointer to the buffer used to build the command
930 * @maxlen: The length of the buffer passed in @buf
932 * Initialize a synthetic event command object. Use this before
933 * calling any of the other kprobe_event functions.
935 void kprobe_event_cmd_init(struct dynevent_cmd *cmd, char *buf, int maxlen)
937 dynevent_cmd_init(cmd, buf, maxlen, DYNEVENT_TYPE_KPROBE,
938 trace_kprobe_run_command);
940 EXPORT_SYMBOL_GPL(kprobe_event_cmd_init);
943 * __kprobe_event_gen_cmd_start - Generate a kprobe event command from arg list
944 * @cmd: A pointer to the dynevent_cmd struct representing the new event
945 * @name: The name of the kprobe event
946 * @loc: The location of the kprobe event
947 * @kretprobe: Is this a return probe?
948 * @args: Variable number of arg (pairs), one pair for each field
950 * NOTE: Users normally won't want to call this function directly, but
951 * rather use the kprobe_event_gen_cmd_start() wrapper, which automatically
952 * adds a NULL to the end of the arg list. If this function is used
953 * directly, make sure the last arg in the variable arg list is NULL.
955 * Generate a kprobe event command to be executed by
956 * kprobe_event_gen_cmd_end(). This function can be used to generate the
957 * complete command or only the first part of it; in the latter case,
958 * kprobe_event_add_fields() can be used to add more fields following this.
960 * Unlikely the synth_event_gen_cmd_start(), @loc must be specified. This
961 * returns -EINVAL if @loc == NULL.
963 * Return: 0 if successful, error otherwise.
965 int __kprobe_event_gen_cmd_start(struct dynevent_cmd *cmd, bool kretprobe,
966 const char *name, const char *loc, ...)
968 char buf[MAX_EVENT_NAME_LEN];
969 struct dynevent_arg arg;
973 if (cmd->type != DYNEVENT_TYPE_KPROBE)
980 snprintf(buf, MAX_EVENT_NAME_LEN, "r:kprobes/%s", name);
982 snprintf(buf, MAX_EVENT_NAME_LEN, "p:kprobes/%s", name);
984 ret = dynevent_str_add(cmd, buf);
988 dynevent_arg_init(&arg, 0);
990 ret = dynevent_arg_add(cmd, &arg, NULL);
998 field = va_arg(args, const char *);
1002 if (++cmd->n_fields > MAX_TRACE_ARGS) {
1008 ret = dynevent_arg_add(cmd, &arg, NULL);
1016 EXPORT_SYMBOL_GPL(__kprobe_event_gen_cmd_start);
1019 * __kprobe_event_add_fields - Add probe fields to a kprobe command from arg list
1020 * @cmd: A pointer to the dynevent_cmd struct representing the new event
1021 * @args: Variable number of arg (pairs), one pair for each field
1023 * NOTE: Users normally won't want to call this function directly, but
1024 * rather use the kprobe_event_add_fields() wrapper, which
1025 * automatically adds a NULL to the end of the arg list. If this
1026 * function is used directly, make sure the last arg in the variable
1029 * Add probe fields to an existing kprobe command using a variable
1030 * list of args. Fields are added in the same order they're listed.
1032 * Return: 0 if successful, error otherwise.
1034 int __kprobe_event_add_fields(struct dynevent_cmd *cmd, ...)
1036 struct dynevent_arg arg;
1040 if (cmd->type != DYNEVENT_TYPE_KPROBE)
1043 dynevent_arg_init(&arg, 0);
1045 va_start(args, cmd);
1049 field = va_arg(args, const char *);
1053 if (++cmd->n_fields > MAX_TRACE_ARGS) {
1059 ret = dynevent_arg_add(cmd, &arg, NULL);
1067 EXPORT_SYMBOL_GPL(__kprobe_event_add_fields);
1070 * kprobe_event_delete - Delete a kprobe event
1071 * @name: The name of the kprobe event to delete
1073 * Delete a kprobe event with the give @name from kernel code rather
1074 * than directly from the command line.
1076 * Return: 0 if successful, error otherwise.
1078 int kprobe_event_delete(const char *name)
1080 char buf[MAX_EVENT_NAME_LEN];
1082 snprintf(buf, MAX_EVENT_NAME_LEN, "-:%s", name);
1084 return trace_run_command(buf, create_or_delete_trace_kprobe);
1086 EXPORT_SYMBOL_GPL(kprobe_event_delete);
1088 static int trace_kprobe_release(struct dyn_event *ev)
1090 struct trace_kprobe *tk = to_trace_kprobe(ev);
1091 int ret = unregister_trace_kprobe(tk);
1094 free_trace_kprobe(tk);
1098 static int trace_kprobe_show(struct seq_file *m, struct dyn_event *ev)
1100 struct trace_kprobe *tk = to_trace_kprobe(ev);
1103 seq_putc(m, trace_kprobe_is_return(tk) ? 'r' : 'p');
1104 if (trace_kprobe_is_return(tk) && tk->rp.maxactive)
1105 seq_printf(m, "%d", tk->rp.maxactive);
1106 seq_printf(m, ":%s/%s", trace_probe_group_name(&tk->tp),
1107 trace_probe_name(&tk->tp));
1110 seq_printf(m, " 0x%p", tk->rp.kp.addr);
1111 else if (tk->rp.kp.offset)
1112 seq_printf(m, " %s+%u", trace_kprobe_symbol(tk),
1115 seq_printf(m, " %s", trace_kprobe_symbol(tk));
1117 for (i = 0; i < tk->tp.nr_args; i++)
1118 seq_printf(m, " %s=%s", tk->tp.args[i].name, tk->tp.args[i].comm);
1124 static int probes_seq_show(struct seq_file *m, void *v)
1126 struct dyn_event *ev = v;
1128 if (!is_trace_kprobe(ev))
1131 return trace_kprobe_show(m, ev);
1134 static const struct seq_operations probes_seq_op = {
1135 .start = dyn_event_seq_start,
1136 .next = dyn_event_seq_next,
1137 .stop = dyn_event_seq_stop,
1138 .show = probes_seq_show
1141 static int probes_open(struct inode *inode, struct file *file)
1145 ret = security_locked_down(LOCKDOWN_TRACEFS);
1149 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
1150 ret = dyn_events_release_all(&trace_kprobe_ops);
1155 return seq_open(file, &probes_seq_op);
1158 static ssize_t probes_write(struct file *file, const char __user *buffer,
1159 size_t count, loff_t *ppos)
1161 return trace_parse_run_command(file, buffer, count, ppos,
1162 create_or_delete_trace_kprobe);
1165 static const struct file_operations kprobe_events_ops = {
1166 .owner = THIS_MODULE,
1167 .open = probes_open,
1169 .llseek = seq_lseek,
1170 .release = seq_release,
1171 .write = probes_write,
1174 /* Probes profiling interfaces */
1175 static int probes_profile_seq_show(struct seq_file *m, void *v)
1177 struct dyn_event *ev = v;
1178 struct trace_kprobe *tk;
1180 if (!is_trace_kprobe(ev))
1183 tk = to_trace_kprobe(ev);
1184 seq_printf(m, " %-44s %15lu %15lu\n",
1185 trace_probe_name(&tk->tp),
1186 trace_kprobe_nhit(tk),
1192 static const struct seq_operations profile_seq_op = {
1193 .start = dyn_event_seq_start,
1194 .next = dyn_event_seq_next,
1195 .stop = dyn_event_seq_stop,
1196 .show = probes_profile_seq_show
1199 static int profile_open(struct inode *inode, struct file *file)
1203 ret = security_locked_down(LOCKDOWN_TRACEFS);
1207 return seq_open(file, &profile_seq_op);
1210 static const struct file_operations kprobe_profile_ops = {
1211 .owner = THIS_MODULE,
1212 .open = profile_open,
1214 .llseek = seq_lseek,
1215 .release = seq_release,
1218 /* Kprobe specific fetch functions */
1220 /* Return the length of string -- including null terminal byte */
1221 static nokprobe_inline int
1222 fetch_store_strlen_user(unsigned long addr)
1224 const void __user *uaddr = (__force const void __user *)addr;
1226 return strnlen_user_nofault(uaddr, MAX_STRING_SIZE);
1229 /* Return the length of string -- including null terminal byte */
1230 static nokprobe_inline int
1231 fetch_store_strlen(unsigned long addr)
1236 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
1237 if (addr < TASK_SIZE)
1238 return fetch_store_strlen_user(addr);
1242 ret = copy_from_kernel_nofault(&c, (u8 *)addr + len, 1);
1244 } while (c && ret == 0 && len < MAX_STRING_SIZE);
1246 return (ret < 0) ? ret : len;
1250 * Fetch a null-terminated string from user. Caller MUST set *(u32 *)buf
1251 * with max length and relative data location.
1253 static nokprobe_inline int
1254 fetch_store_string_user(unsigned long addr, void *dest, void *base)
1256 const void __user *uaddr = (__force const void __user *)addr;
1257 int maxlen = get_loc_len(*(u32 *)dest);
1261 if (unlikely(!maxlen))
1264 __dest = get_loc_data(dest, base);
1266 ret = strncpy_from_user_nofault(__dest, uaddr, maxlen);
1268 *(u32 *)dest = make_data_loc(ret, __dest - base);
1274 * Fetch a null-terminated string. Caller MUST set *(u32 *)buf with max
1275 * length and relative data location.
1277 static nokprobe_inline int
1278 fetch_store_string(unsigned long addr, void *dest, void *base)
1280 int maxlen = get_loc_len(*(u32 *)dest);
1284 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
1285 if ((unsigned long)addr < TASK_SIZE)
1286 return fetch_store_string_user(addr, dest, base);
1289 if (unlikely(!maxlen))
1292 __dest = get_loc_data(dest, base);
1295 * Try to get string again, since the string can be changed while
1298 ret = strncpy_from_kernel_nofault(__dest, (void *)addr, maxlen);
1300 *(u32 *)dest = make_data_loc(ret, __dest - base);
1305 static nokprobe_inline int
1306 probe_mem_read_user(void *dest, void *src, size_t size)
1308 const void __user *uaddr = (__force const void __user *)src;
1310 return copy_from_user_nofault(dest, uaddr, size);
1313 static nokprobe_inline int
1314 probe_mem_read(void *dest, void *src, size_t size)
1316 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
1317 if ((unsigned long)src < TASK_SIZE)
1318 return probe_mem_read_user(dest, src, size);
1320 return copy_from_kernel_nofault(dest, src, size);
1323 /* Note that we don't verify it, since the code does not come from user space */
1325 process_fetch_insn(struct fetch_insn *code, struct pt_regs *regs, void *dest,
1331 /* 1st stage: get value from context */
1334 val = regs_get_register(regs, code->param);
1336 case FETCH_OP_STACK:
1337 val = regs_get_kernel_stack_nth(regs, code->param);
1339 case FETCH_OP_STACKP:
1340 val = kernel_stack_pointer(regs);
1342 case FETCH_OP_RETVAL:
1343 val = regs_return_value(regs);
1346 val = code->immediate;
1349 val = (unsigned long)current->comm;
1352 val = (unsigned long)code->data;
1354 #ifdef CONFIG_HAVE_FUNCTION_ARG_ACCESS_API
1356 val = regs_get_kernel_argument(regs, code->param);
1359 case FETCH_NOP_SYMBOL: /* Ignore a place holder */
1367 return process_fetch_insn_bottom(code, val, dest, base);
1369 NOKPROBE_SYMBOL(process_fetch_insn)
1371 /* Kprobe handler */
1372 static nokprobe_inline void
1373 __kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs,
1374 struct trace_event_file *trace_file)
1376 struct kprobe_trace_entry_head *entry;
1377 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1378 struct trace_event_buffer fbuffer;
1381 WARN_ON(call != trace_file->event_call);
1383 if (trace_trigger_soft_disabled(trace_file))
1386 local_save_flags(fbuffer.flags);
1387 fbuffer.pc = preempt_count();
1388 fbuffer.trace_file = trace_file;
1390 dsize = __get_data_size(&tk->tp, regs);
1393 trace_event_buffer_lock_reserve(&fbuffer.buffer, trace_file,
1395 sizeof(*entry) + tk->tp.size + dsize,
1396 fbuffer.flags, fbuffer.pc);
1400 fbuffer.regs = regs;
1401 entry = fbuffer.entry = ring_buffer_event_data(fbuffer.event);
1402 entry->ip = (unsigned long)tk->rp.kp.addr;
1403 store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1405 trace_event_buffer_commit(&fbuffer);
1409 kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs)
1411 struct event_file_link *link;
1413 trace_probe_for_each_link_rcu(link, &tk->tp)
1414 __kprobe_trace_func(tk, regs, link->file);
1416 NOKPROBE_SYMBOL(kprobe_trace_func);
1418 /* Kretprobe handler */
1419 static nokprobe_inline void
1420 __kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1421 struct pt_regs *regs,
1422 struct trace_event_file *trace_file)
1424 struct kretprobe_trace_entry_head *entry;
1425 struct trace_event_buffer fbuffer;
1426 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1429 WARN_ON(call != trace_file->event_call);
1431 if (trace_trigger_soft_disabled(trace_file))
1434 local_save_flags(fbuffer.flags);
1435 fbuffer.pc = preempt_count();
1436 fbuffer.trace_file = trace_file;
1438 dsize = __get_data_size(&tk->tp, regs);
1440 trace_event_buffer_lock_reserve(&fbuffer.buffer, trace_file,
1442 sizeof(*entry) + tk->tp.size + dsize,
1443 fbuffer.flags, fbuffer.pc);
1447 fbuffer.regs = regs;
1448 entry = fbuffer.entry = ring_buffer_event_data(fbuffer.event);
1449 entry->func = (unsigned long)tk->rp.kp.addr;
1450 entry->ret_ip = (unsigned long)ri->ret_addr;
1451 store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1453 trace_event_buffer_commit(&fbuffer);
1457 kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1458 struct pt_regs *regs)
1460 struct event_file_link *link;
1462 trace_probe_for_each_link_rcu(link, &tk->tp)
1463 __kretprobe_trace_func(tk, ri, regs, link->file);
1465 NOKPROBE_SYMBOL(kretprobe_trace_func);
1467 /* Event entry printers */
1468 static enum print_line_t
1469 print_kprobe_event(struct trace_iterator *iter, int flags,
1470 struct trace_event *event)
1472 struct kprobe_trace_entry_head *field;
1473 struct trace_seq *s = &iter->seq;
1474 struct trace_probe *tp;
1476 field = (struct kprobe_trace_entry_head *)iter->ent;
1477 tp = trace_probe_primary_from_call(
1478 container_of(event, struct trace_event_call, event));
1479 if (WARN_ON_ONCE(!tp))
1482 trace_seq_printf(s, "%s: (", trace_probe_name(tp));
1484 if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
1487 trace_seq_putc(s, ')');
1489 if (print_probe_args(s, tp->args, tp->nr_args,
1490 (u8 *)&field[1], field) < 0)
1493 trace_seq_putc(s, '\n');
1495 return trace_handle_return(s);
1498 static enum print_line_t
1499 print_kretprobe_event(struct trace_iterator *iter, int flags,
1500 struct trace_event *event)
1502 struct kretprobe_trace_entry_head *field;
1503 struct trace_seq *s = &iter->seq;
1504 struct trace_probe *tp;
1506 field = (struct kretprobe_trace_entry_head *)iter->ent;
1507 tp = trace_probe_primary_from_call(
1508 container_of(event, struct trace_event_call, event));
1509 if (WARN_ON_ONCE(!tp))
1512 trace_seq_printf(s, "%s: (", trace_probe_name(tp));
1514 if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1517 trace_seq_puts(s, " <- ");
1519 if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1522 trace_seq_putc(s, ')');
1524 if (print_probe_args(s, tp->args, tp->nr_args,
1525 (u8 *)&field[1], field) < 0)
1528 trace_seq_putc(s, '\n');
1531 return trace_handle_return(s);
1535 static int kprobe_event_define_fields(struct trace_event_call *event_call)
1538 struct kprobe_trace_entry_head field;
1539 struct trace_probe *tp;
1541 tp = trace_probe_primary_from_call(event_call);
1542 if (WARN_ON_ONCE(!tp))
1545 DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1547 return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
1550 static int kretprobe_event_define_fields(struct trace_event_call *event_call)
1553 struct kretprobe_trace_entry_head field;
1554 struct trace_probe *tp;
1556 tp = trace_probe_primary_from_call(event_call);
1557 if (WARN_ON_ONCE(!tp))
1560 DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1561 DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1563 return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
1566 #ifdef CONFIG_PERF_EVENTS
1568 /* Kprobe profile handler */
1570 kprobe_perf_func(struct trace_kprobe *tk, struct pt_regs *regs)
1572 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1573 struct kprobe_trace_entry_head *entry;
1574 struct hlist_head *head;
1575 int size, __size, dsize;
1578 if (bpf_prog_array_valid(call)) {
1579 unsigned long orig_ip = instruction_pointer(regs);
1582 ret = trace_call_bpf(call, regs);
1585 * We need to check and see if we modified the pc of the
1586 * pt_regs, and if so return 1 so that we don't do the
1589 if (orig_ip != instruction_pointer(regs))
1595 head = this_cpu_ptr(call->perf_events);
1596 if (hlist_empty(head))
1599 dsize = __get_data_size(&tk->tp, regs);
1600 __size = sizeof(*entry) + tk->tp.size + dsize;
1601 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1602 size -= sizeof(u32);
1604 entry = perf_trace_buf_alloc(size, NULL, &rctx);
1608 entry->ip = (unsigned long)tk->rp.kp.addr;
1609 memset(&entry[1], 0, dsize);
1610 store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1611 perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1615 NOKPROBE_SYMBOL(kprobe_perf_func);
1617 /* Kretprobe profile handler */
1619 kretprobe_perf_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1620 struct pt_regs *regs)
1622 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1623 struct kretprobe_trace_entry_head *entry;
1624 struct hlist_head *head;
1625 int size, __size, dsize;
1628 if (bpf_prog_array_valid(call) && !trace_call_bpf(call, regs))
1631 head = this_cpu_ptr(call->perf_events);
1632 if (hlist_empty(head))
1635 dsize = __get_data_size(&tk->tp, regs);
1636 __size = sizeof(*entry) + tk->tp.size + dsize;
1637 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1638 size -= sizeof(u32);
1640 entry = perf_trace_buf_alloc(size, NULL, &rctx);
1644 entry->func = (unsigned long)tk->rp.kp.addr;
1645 entry->ret_ip = (unsigned long)ri->ret_addr;
1646 store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1647 perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1650 NOKPROBE_SYMBOL(kretprobe_perf_func);
1652 int bpf_get_kprobe_info(const struct perf_event *event, u32 *fd_type,
1653 const char **symbol, u64 *probe_offset,
1654 u64 *probe_addr, bool perf_type_tracepoint)
1656 const char *pevent = trace_event_name(event->tp_event);
1657 const char *group = event->tp_event->class->system;
1658 struct trace_kprobe *tk;
1660 if (perf_type_tracepoint)
1661 tk = find_trace_kprobe(pevent, group);
1663 tk = trace_kprobe_primary_from_call(event->tp_event);
1667 *fd_type = trace_kprobe_is_return(tk) ? BPF_FD_TYPE_KRETPROBE
1668 : BPF_FD_TYPE_KPROBE;
1670 *symbol = tk->symbol;
1671 *probe_offset = tk->rp.kp.offset;
1676 *probe_addr = (unsigned long)tk->rp.kp.addr;
1680 #endif /* CONFIG_PERF_EVENTS */
1683 * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex.
1685 * kprobe_trace_self_tests_init() does enable_trace_probe/disable_trace_probe
1686 * lockless, but we can't race with this __init function.
1688 static int kprobe_register(struct trace_event_call *event,
1689 enum trace_reg type, void *data)
1691 struct trace_event_file *file = data;
1694 case TRACE_REG_REGISTER:
1695 return enable_trace_kprobe(event, file);
1696 case TRACE_REG_UNREGISTER:
1697 return disable_trace_kprobe(event, file);
1699 #ifdef CONFIG_PERF_EVENTS
1700 case TRACE_REG_PERF_REGISTER:
1701 return enable_trace_kprobe(event, NULL);
1702 case TRACE_REG_PERF_UNREGISTER:
1703 return disable_trace_kprobe(event, NULL);
1704 case TRACE_REG_PERF_OPEN:
1705 case TRACE_REG_PERF_CLOSE:
1706 case TRACE_REG_PERF_ADD:
1707 case TRACE_REG_PERF_DEL:
1714 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1716 struct trace_kprobe *tk = container_of(kp, struct trace_kprobe, rp.kp);
1719 raw_cpu_inc(*tk->nhit);
1721 if (trace_probe_test_flag(&tk->tp, TP_FLAG_TRACE))
1722 kprobe_trace_func(tk, regs);
1723 #ifdef CONFIG_PERF_EVENTS
1724 if (trace_probe_test_flag(&tk->tp, TP_FLAG_PROFILE))
1725 ret = kprobe_perf_func(tk, regs);
1729 NOKPROBE_SYMBOL(kprobe_dispatcher);
1732 kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1734 struct kretprobe *rp = get_kretprobe(ri);
1735 struct trace_kprobe *tk = container_of(rp, struct trace_kprobe, rp);
1737 raw_cpu_inc(*tk->nhit);
1739 if (trace_probe_test_flag(&tk->tp, TP_FLAG_TRACE))
1740 kretprobe_trace_func(tk, ri, regs);
1741 #ifdef CONFIG_PERF_EVENTS
1742 if (trace_probe_test_flag(&tk->tp, TP_FLAG_PROFILE))
1743 kretprobe_perf_func(tk, ri, regs);
1745 return 0; /* We don't tweek kernel, so just return 0 */
1747 NOKPROBE_SYMBOL(kretprobe_dispatcher);
1749 static struct trace_event_functions kretprobe_funcs = {
1750 .trace = print_kretprobe_event
1753 static struct trace_event_functions kprobe_funcs = {
1754 .trace = print_kprobe_event
1757 static struct trace_event_fields kretprobe_fields_array[] = {
1758 { .type = TRACE_FUNCTION_TYPE,
1759 .define_fields = kretprobe_event_define_fields },
1763 static struct trace_event_fields kprobe_fields_array[] = {
1764 { .type = TRACE_FUNCTION_TYPE,
1765 .define_fields = kprobe_event_define_fields },
1769 static inline void init_trace_event_call(struct trace_kprobe *tk)
1771 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1773 if (trace_kprobe_is_return(tk)) {
1774 call->event.funcs = &kretprobe_funcs;
1775 call->class->fields_array = kretprobe_fields_array;
1777 call->event.funcs = &kprobe_funcs;
1778 call->class->fields_array = kprobe_fields_array;
1781 call->flags = TRACE_EVENT_FL_KPROBE;
1782 call->class->reg = kprobe_register;
1785 static int register_kprobe_event(struct trace_kprobe *tk)
1787 init_trace_event_call(tk);
1789 return trace_probe_register_event_call(&tk->tp);
1792 static int unregister_kprobe_event(struct trace_kprobe *tk)
1794 return trace_probe_unregister_event_call(&tk->tp);
1797 #ifdef CONFIG_PERF_EVENTS
1798 /* create a trace_kprobe, but don't add it to global lists */
1799 struct trace_event_call *
1800 create_local_trace_kprobe(char *func, void *addr, unsigned long offs,
1803 struct trace_kprobe *tk;
1808 * local trace_kprobes are not added to dyn_event, so they are never
1809 * searched in find_trace_kprobe(). Therefore, there is no concern of
1810 * duplicated name here.
1812 event = func ? func : "DUMMY_EVENT";
1814 tk = alloc_trace_kprobe(KPROBE_EVENT_SYSTEM, event, (void *)addr, func,
1815 offs, 0 /* maxactive */, 0 /* nargs */,
1819 pr_info("Failed to allocate trace_probe.(%d)\n",
1821 return ERR_CAST(tk);
1824 init_trace_event_call(tk);
1826 if (traceprobe_set_print_fmt(&tk->tp, trace_kprobe_is_return(tk)) < 0) {
1831 ret = __register_trace_kprobe(tk);
1835 return trace_probe_event_call(&tk->tp);
1837 free_trace_kprobe(tk);
1838 return ERR_PTR(ret);
1841 void destroy_local_trace_kprobe(struct trace_event_call *event_call)
1843 struct trace_kprobe *tk;
1845 tk = trace_kprobe_primary_from_call(event_call);
1849 if (trace_probe_is_enabled(&tk->tp)) {
1854 __unregister_trace_kprobe(tk);
1856 free_trace_kprobe(tk);
1858 #endif /* CONFIG_PERF_EVENTS */
1860 static __init void enable_boot_kprobe_events(void)
1862 struct trace_array *tr = top_trace_array();
1863 struct trace_event_file *file;
1864 struct trace_kprobe *tk;
1865 struct dyn_event *pos;
1867 mutex_lock(&event_mutex);
1868 for_each_trace_kprobe(tk, pos) {
1869 list_for_each_entry(file, &tr->events, list)
1870 if (file->event_call == trace_probe_event_call(&tk->tp))
1871 trace_event_enable_disable(file, 1, 0);
1873 mutex_unlock(&event_mutex);
1876 static __init void setup_boot_kprobe_events(void)
1878 char *p, *cmd = kprobe_boot_events_buf;
1881 strreplace(kprobe_boot_events_buf, ',', ' ');
1883 while (cmd && *cmd != '\0') {
1884 p = strchr(cmd, ';');
1888 ret = trace_run_command(cmd, create_or_delete_trace_kprobe);
1890 pr_warn("Failed to add event(%d): %s\n", ret, cmd);
1892 kprobe_boot_events_enabled = true;
1897 enable_boot_kprobe_events();
1901 * Register dynevent at core_initcall. This allows kernel to setup kprobe
1902 * events in postcore_initcall without tracefs.
1904 static __init int init_kprobe_trace_early(void)
1908 ret = dyn_event_register(&trace_kprobe_ops);
1912 if (register_module_notifier(&trace_kprobe_module_nb))
1917 core_initcall(init_kprobe_trace_early);
1919 /* Make a tracefs interface for controlling probe points */
1920 static __init int init_kprobe_trace(void)
1923 struct dentry *entry;
1925 ret = tracing_init_dentry();
1929 entry = tracefs_create_file("kprobe_events", 0644, NULL,
1930 NULL, &kprobe_events_ops);
1932 /* Event list interface */
1934 pr_warn("Could not create tracefs 'kprobe_events' entry\n");
1936 /* Profile interface */
1937 entry = tracefs_create_file("kprobe_profile", 0444, NULL,
1938 NULL, &kprobe_profile_ops);
1941 pr_warn("Could not create tracefs 'kprobe_profile' entry\n");
1943 setup_boot_kprobe_events();
1947 fs_initcall(init_kprobe_trace);
1950 #ifdef CONFIG_FTRACE_STARTUP_TEST
1951 static __init struct trace_event_file *
1952 find_trace_probe_file(struct trace_kprobe *tk, struct trace_array *tr)
1954 struct trace_event_file *file;
1956 list_for_each_entry(file, &tr->events, list)
1957 if (file->event_call == trace_probe_event_call(&tk->tp))
1964 * Nobody but us can call enable_trace_kprobe/disable_trace_kprobe at this
1965 * stage, we can do this lockless.
1967 static __init int kprobe_trace_self_tests_init(void)
1970 int (*target)(int, int, int, int, int, int);
1971 struct trace_kprobe *tk;
1972 struct trace_event_file *file;
1974 if (tracing_is_disabled())
1977 if (kprobe_boot_events_enabled) {
1978 pr_info("Skipping kprobe tests due to kprobe_event on cmdline\n");
1982 target = kprobe_trace_selftest_target;
1984 pr_info("Testing kprobe tracing: ");
1986 ret = trace_run_command("p:testprobe kprobe_trace_selftest_target $stack $stack0 +0($stack)",
1987 create_or_delete_trace_kprobe);
1988 if (WARN_ON_ONCE(ret)) {
1989 pr_warn("error on probing function entry.\n");
1992 /* Enable trace point */
1993 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1994 if (WARN_ON_ONCE(tk == NULL)) {
1995 pr_warn("error on getting new probe.\n");
1998 file = find_trace_probe_file(tk, top_trace_array());
1999 if (WARN_ON_ONCE(file == NULL)) {
2000 pr_warn("error on getting probe file.\n");
2003 enable_trace_kprobe(
2004 trace_probe_event_call(&tk->tp), file);
2008 ret = trace_run_command("r:testprobe2 kprobe_trace_selftest_target $retval",
2009 create_or_delete_trace_kprobe);
2010 if (WARN_ON_ONCE(ret)) {
2011 pr_warn("error on probing function return.\n");
2014 /* Enable trace point */
2015 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
2016 if (WARN_ON_ONCE(tk == NULL)) {
2017 pr_warn("error on getting 2nd new probe.\n");
2020 file = find_trace_probe_file(tk, top_trace_array());
2021 if (WARN_ON_ONCE(file == NULL)) {
2022 pr_warn("error on getting probe file.\n");
2025 enable_trace_kprobe(
2026 trace_probe_event_call(&tk->tp), file);
2033 ret = target(1, 2, 3, 4, 5, 6);
2036 * Not expecting an error here, the check is only to prevent the
2037 * optimizer from removing the call to target() as otherwise there
2038 * are no side-effects and the call is never performed.
2043 /* Disable trace points before removing it */
2044 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
2045 if (WARN_ON_ONCE(tk == NULL)) {
2046 pr_warn("error on getting test probe.\n");
2049 if (trace_kprobe_nhit(tk) != 1) {
2050 pr_warn("incorrect number of testprobe hits\n");
2054 file = find_trace_probe_file(tk, top_trace_array());
2055 if (WARN_ON_ONCE(file == NULL)) {
2056 pr_warn("error on getting probe file.\n");
2059 disable_trace_kprobe(
2060 trace_probe_event_call(&tk->tp), file);
2063 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
2064 if (WARN_ON_ONCE(tk == NULL)) {
2065 pr_warn("error on getting 2nd test probe.\n");
2068 if (trace_kprobe_nhit(tk) != 1) {
2069 pr_warn("incorrect number of testprobe2 hits\n");
2073 file = find_trace_probe_file(tk, top_trace_array());
2074 if (WARN_ON_ONCE(file == NULL)) {
2075 pr_warn("error on getting probe file.\n");
2078 disable_trace_kprobe(
2079 trace_probe_event_call(&tk->tp), file);
2082 ret = trace_run_command("-:testprobe", create_or_delete_trace_kprobe);
2083 if (WARN_ON_ONCE(ret)) {
2084 pr_warn("error on deleting a probe.\n");
2088 ret = trace_run_command("-:testprobe2", create_or_delete_trace_kprobe);
2089 if (WARN_ON_ONCE(ret)) {
2090 pr_warn("error on deleting a probe.\n");
2095 ret = dyn_events_release_all(&trace_kprobe_ops);
2096 if (WARN_ON_ONCE(ret)) {
2097 pr_warn("error on cleaning up probes.\n");
2101 * Wait for the optimizer work to finish. Otherwise it might fiddle
2102 * with probes in already freed __init text.
2104 wait_for_kprobe_optimizer();
2106 pr_cont("NG: Some tests are failed. Please check them.\n");
2112 late_initcall(kprobe_trace_self_tests_init);