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(mod->name);
110 const char *name = trace_kprobe_symbol(tk);
111 return strncmp(mod->name, name, len) == 0 && name[len] == ':';
114 static nokprobe_inline bool trace_kprobe_module_exist(struct trace_kprobe *tk)
121 p = strchr(tk->symbol, ':');
125 mutex_lock(&module_mutex);
126 ret = !!find_module(tk->symbol);
127 mutex_unlock(&module_mutex);
133 static bool trace_kprobe_is_busy(struct dyn_event *ev)
135 struct trace_kprobe *tk = to_trace_kprobe(ev);
137 return trace_probe_is_enabled(&tk->tp);
140 static bool trace_kprobe_match_command_head(struct trace_kprobe *tk,
141 int argc, const char **argv)
143 char buf[MAX_ARGSTR_LEN + 1];
149 snprintf(buf, sizeof(buf), "0x%p", tk->rp.kp.addr);
150 else if (tk->rp.kp.offset)
151 snprintf(buf, sizeof(buf), "%s+%u",
152 trace_kprobe_symbol(tk), tk->rp.kp.offset);
154 snprintf(buf, sizeof(buf), "%s", trace_kprobe_symbol(tk));
155 if (strcmp(buf, argv[0]))
159 return trace_probe_match_command_args(&tk->tp, argc, argv);
162 static bool trace_kprobe_match(const char *system, const char *event,
163 int argc, const char **argv, struct dyn_event *ev)
165 struct trace_kprobe *tk = to_trace_kprobe(ev);
167 return strcmp(trace_probe_name(&tk->tp), event) == 0 &&
168 (!system || strcmp(trace_probe_group_name(&tk->tp), system) == 0) &&
169 trace_kprobe_match_command_head(tk, argc, argv);
172 static nokprobe_inline unsigned long trace_kprobe_nhit(struct trace_kprobe *tk)
174 unsigned long nhit = 0;
177 for_each_possible_cpu(cpu)
178 nhit += *per_cpu_ptr(tk->nhit, cpu);
183 static nokprobe_inline bool trace_kprobe_is_registered(struct trace_kprobe *tk)
185 return !(list_empty(&tk->rp.kp.list) &&
186 hlist_unhashed(&tk->rp.kp.hlist));
189 /* Return 0 if it fails to find the symbol address */
190 static nokprobe_inline
191 unsigned long trace_kprobe_address(struct trace_kprobe *tk)
196 addr = (unsigned long)
197 kallsyms_lookup_name(trace_kprobe_symbol(tk));
199 addr += tk->rp.kp.offset;
201 addr = (unsigned long)tk->rp.kp.addr;
206 static nokprobe_inline struct trace_kprobe *
207 trace_kprobe_primary_from_call(struct trace_event_call *call)
209 struct trace_probe *tp;
211 tp = trace_probe_primary_from_call(call);
212 if (WARN_ON_ONCE(!tp))
215 return container_of(tp, struct trace_kprobe, tp);
218 bool trace_kprobe_on_func_entry(struct trace_event_call *call)
220 struct trace_kprobe *tk = trace_kprobe_primary_from_call(call);
222 return tk ? kprobe_on_func_entry(tk->rp.kp.addr,
223 tk->rp.kp.addr ? NULL : tk->rp.kp.symbol_name,
224 tk->rp.kp.addr ? 0 : tk->rp.kp.offset) : false;
227 bool trace_kprobe_error_injectable(struct trace_event_call *call)
229 struct trace_kprobe *tk = trace_kprobe_primary_from_call(call);
231 return tk ? within_error_injection_list(trace_kprobe_address(tk)) :
235 static int register_kprobe_event(struct trace_kprobe *tk);
236 static int unregister_kprobe_event(struct trace_kprobe *tk);
238 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
239 static int kretprobe_dispatcher(struct kretprobe_instance *ri,
240 struct pt_regs *regs);
242 static void free_trace_kprobe(struct trace_kprobe *tk)
245 trace_probe_cleanup(&tk->tp);
247 free_percpu(tk->nhit);
253 * Allocate new trace_probe and initialize it (including kprobes).
255 static struct trace_kprobe *alloc_trace_kprobe(const char *group,
261 int nargs, bool is_return)
263 struct trace_kprobe *tk;
266 tk = kzalloc(SIZEOF_TRACE_KPROBE(nargs), GFP_KERNEL);
270 tk->nhit = alloc_percpu(unsigned long);
275 tk->symbol = kstrdup(symbol, GFP_KERNEL);
278 tk->rp.kp.symbol_name = tk->symbol;
279 tk->rp.kp.offset = offs;
281 tk->rp.kp.addr = addr;
284 tk->rp.handler = kretprobe_dispatcher;
286 tk->rp.kp.pre_handler = kprobe_dispatcher;
288 tk->rp.maxactive = maxactive;
289 INIT_HLIST_NODE(&tk->rp.kp.hlist);
290 INIT_LIST_HEAD(&tk->rp.kp.list);
292 ret = trace_probe_init(&tk->tp, event, group);
296 dyn_event_init(&tk->devent, &trace_kprobe_ops);
299 free_trace_kprobe(tk);
303 static struct trace_kprobe *find_trace_kprobe(const char *event,
306 struct dyn_event *pos;
307 struct trace_kprobe *tk;
309 for_each_trace_kprobe(tk, pos)
310 if (strcmp(trace_probe_name(&tk->tp), event) == 0 &&
311 strcmp(trace_probe_group_name(&tk->tp), group) == 0)
316 static inline int __enable_trace_kprobe(struct trace_kprobe *tk)
320 if (trace_kprobe_is_registered(tk) && !trace_kprobe_has_gone(tk)) {
321 if (trace_kprobe_is_return(tk))
322 ret = enable_kretprobe(&tk->rp);
324 ret = enable_kprobe(&tk->rp.kp);
330 static void __disable_trace_kprobe(struct trace_probe *tp)
332 struct trace_probe *pos;
333 struct trace_kprobe *tk;
335 list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
336 tk = container_of(pos, struct trace_kprobe, tp);
337 if (!trace_kprobe_is_registered(tk))
339 if (trace_kprobe_is_return(tk))
340 disable_kretprobe(&tk->rp);
342 disable_kprobe(&tk->rp.kp);
348 * if the file is NULL, enable "perf" handler, or enable "trace" handler.
350 static int enable_trace_kprobe(struct trace_event_call *call,
351 struct trace_event_file *file)
353 struct trace_probe *pos, *tp;
354 struct trace_kprobe *tk;
358 tp = trace_probe_primary_from_call(call);
359 if (WARN_ON_ONCE(!tp))
361 enabled = trace_probe_is_enabled(tp);
363 /* This also changes "enabled" state */
365 ret = trace_probe_add_file(tp, file);
369 trace_probe_set_flag(tp, TP_FLAG_PROFILE);
374 list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
375 tk = container_of(pos, struct trace_kprobe, tp);
376 if (trace_kprobe_has_gone(tk))
378 ret = __enable_trace_kprobe(tk);
385 /* Failed to enable one of them. Roll back all */
387 __disable_trace_kprobe(tp);
389 trace_probe_remove_file(tp, file);
391 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
398 * Disable trace_probe
399 * if the file is NULL, disable "perf" handler, or disable "trace" handler.
401 static int disable_trace_kprobe(struct trace_event_call *call,
402 struct trace_event_file *file)
404 struct trace_probe *tp;
406 tp = trace_probe_primary_from_call(call);
407 if (WARN_ON_ONCE(!tp))
411 if (!trace_probe_get_file_link(tp, file))
413 if (!trace_probe_has_single_file(tp))
415 trace_probe_clear_flag(tp, TP_FLAG_TRACE);
417 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
419 if (!trace_probe_is_enabled(tp))
420 __disable_trace_kprobe(tp);
425 * Synchronization is done in below function. For perf event,
426 * file == NULL and perf_trace_event_unreg() calls
427 * tracepoint_synchronize_unregister() to ensure synchronize
428 * event. We don't need to care about it.
430 trace_probe_remove_file(tp, file);
435 #if defined(CONFIG_KPROBES_ON_FTRACE) && \
436 !defined(CONFIG_KPROBE_EVENTS_ON_NOTRACE)
437 static bool __within_notrace_func(unsigned long addr)
439 unsigned long offset, size;
441 if (!addr || !kallsyms_lookup_size_offset(addr, &size, &offset))
444 /* Get the entry address of the target function */
448 * Since ftrace_location_range() does inclusive range check, we need
449 * to subtract 1 byte from the end address.
451 return !ftrace_location_range(addr, addr + size - 1);
454 static bool within_notrace_func(struct trace_kprobe *tk)
456 unsigned long addr = addr = trace_kprobe_address(tk);
457 char symname[KSYM_NAME_LEN], *p;
459 if (!__within_notrace_func(addr))
462 /* Check if the address is on a suffixed-symbol */
463 if (!lookup_symbol_name(addr, symname)) {
464 p = strchr(symname, '.');
468 addr = (unsigned long)kprobe_lookup_name(symname, 0);
470 return __within_notrace_func(addr);
476 #define within_notrace_func(tk) (false)
479 /* Internal register function - just handle k*probes and flags */
480 static int __register_trace_kprobe(struct trace_kprobe *tk)
484 ret = security_locked_down(LOCKDOWN_KPROBES);
488 if (trace_kprobe_is_registered(tk))
491 if (within_notrace_func(tk)) {
492 pr_warn("Could not probe notrace function %s\n",
493 trace_kprobe_symbol(tk));
497 for (i = 0; i < tk->tp.nr_args; i++) {
498 ret = traceprobe_update_arg(&tk->tp.args[i]);
503 /* Set/clear disabled flag according to tp->flag */
504 if (trace_probe_is_enabled(&tk->tp))
505 tk->rp.kp.flags &= ~KPROBE_FLAG_DISABLED;
507 tk->rp.kp.flags |= KPROBE_FLAG_DISABLED;
509 if (trace_kprobe_is_return(tk))
510 ret = register_kretprobe(&tk->rp);
512 ret = register_kprobe(&tk->rp.kp);
517 /* Internal unregister function - just handle k*probes and flags */
518 static void __unregister_trace_kprobe(struct trace_kprobe *tk)
520 if (trace_kprobe_is_registered(tk)) {
521 if (trace_kprobe_is_return(tk))
522 unregister_kretprobe(&tk->rp);
524 unregister_kprobe(&tk->rp.kp);
525 /* Cleanup kprobe for reuse and mark it unregistered */
526 INIT_HLIST_NODE(&tk->rp.kp.hlist);
527 INIT_LIST_HEAD(&tk->rp.kp.list);
528 if (tk->rp.kp.symbol_name)
529 tk->rp.kp.addr = NULL;
533 /* Unregister a trace_probe and probe_event */
534 static int unregister_trace_kprobe(struct trace_kprobe *tk)
536 /* If other probes are on the event, just unregister kprobe */
537 if (trace_probe_has_sibling(&tk->tp))
540 /* Enabled event can not be unregistered */
541 if (trace_probe_is_enabled(&tk->tp))
544 /* Will fail if probe is being used by ftrace or perf */
545 if (unregister_kprobe_event(tk))
549 __unregister_trace_kprobe(tk);
550 dyn_event_remove(&tk->devent);
551 trace_probe_unlink(&tk->tp);
556 static bool trace_kprobe_has_same_kprobe(struct trace_kprobe *orig,
557 struct trace_kprobe *comp)
559 struct trace_probe_event *tpe = orig->tp.event;
560 struct trace_probe *pos;
563 list_for_each_entry(pos, &tpe->probes, list) {
564 orig = container_of(pos, struct trace_kprobe, tp);
565 if (strcmp(trace_kprobe_symbol(orig),
566 trace_kprobe_symbol(comp)) ||
567 trace_kprobe_offset(orig) != trace_kprobe_offset(comp))
571 * trace_probe_compare_arg_type() ensured that nr_args and
572 * each argument name and type are same. Let's compare comm.
574 for (i = 0; i < orig->tp.nr_args; i++) {
575 if (strcmp(orig->tp.args[i].comm,
576 comp->tp.args[i].comm))
580 if (i == orig->tp.nr_args)
587 static int append_trace_kprobe(struct trace_kprobe *tk, struct trace_kprobe *to)
591 ret = trace_probe_compare_arg_type(&tk->tp, &to->tp);
593 /* Note that argument starts index = 2 */
594 trace_probe_log_set_index(ret + 1);
595 trace_probe_log_err(0, DIFF_ARG_TYPE);
598 if (trace_kprobe_has_same_kprobe(to, tk)) {
599 trace_probe_log_set_index(0);
600 trace_probe_log_err(0, SAME_PROBE);
604 /* Append to existing event */
605 ret = trace_probe_append(&tk->tp, &to->tp);
609 /* Register k*probe */
610 ret = __register_trace_kprobe(tk);
611 if (ret == -ENOENT && !trace_kprobe_module_exist(tk)) {
612 pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
617 trace_probe_unlink(&tk->tp);
619 dyn_event_add(&tk->devent);
624 /* Register a trace_probe and probe_event */
625 static int register_trace_kprobe(struct trace_kprobe *tk)
627 struct trace_kprobe *old_tk;
630 mutex_lock(&event_mutex);
632 old_tk = find_trace_kprobe(trace_probe_name(&tk->tp),
633 trace_probe_group_name(&tk->tp));
635 if (trace_kprobe_is_return(tk) != trace_kprobe_is_return(old_tk)) {
636 trace_probe_log_set_index(0);
637 trace_probe_log_err(0, DIFF_PROBE_TYPE);
640 ret = append_trace_kprobe(tk, old_tk);
645 /* Register new event */
646 ret = register_kprobe_event(tk);
648 pr_warn("Failed to register probe event(%d)\n", ret);
652 /* Register k*probe */
653 ret = __register_trace_kprobe(tk);
654 if (ret == -ENOENT && !trace_kprobe_module_exist(tk)) {
655 pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
660 unregister_kprobe_event(tk);
662 dyn_event_add(&tk->devent);
665 mutex_unlock(&event_mutex);
669 /* Module notifier call back, checking event on the module */
670 static int trace_kprobe_module_callback(struct notifier_block *nb,
671 unsigned long val, void *data)
673 struct module *mod = data;
674 struct dyn_event *pos;
675 struct trace_kprobe *tk;
678 if (val != MODULE_STATE_COMING)
681 /* Update probes on coming module */
682 mutex_lock(&event_mutex);
683 for_each_trace_kprobe(tk, pos) {
684 if (trace_kprobe_within_module(tk, mod)) {
685 /* Don't need to check busy - this should have gone. */
686 __unregister_trace_kprobe(tk);
687 ret = __register_trace_kprobe(tk);
689 pr_warn("Failed to re-register probe %s on %s: %d\n",
690 trace_probe_name(&tk->tp),
694 mutex_unlock(&event_mutex);
699 static struct notifier_block trace_kprobe_module_nb = {
700 .notifier_call = trace_kprobe_module_callback,
701 .priority = 1 /* Invoked after kprobe module callback */
704 /* Convert certain expected symbols into '_' when generating event names */
705 static inline void sanitize_event_name(char *name)
707 while (*name++ != '\0')
708 if (*name == ':' || *name == '.')
712 static int trace_kprobe_create(int argc, const char *argv[])
717 * p[:[GRP/]EVENT] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
719 * r[MAXACTIVE][:[GRP/]EVENT] [MOD:]KSYM[+0] [FETCHARGS]
721 * $retval : fetch return value
722 * $stack : fetch stack address
723 * $stackN : fetch Nth of stack (N:0-)
724 * $comm : fetch current task comm
725 * @ADDR : fetch memory at ADDR (ADDR should be in kernel)
726 * @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
727 * %REG : fetch register REG
728 * Dereferencing memory fetch:
729 * +|-offs(ARG) : fetch memory at ARG +|- offs address.
730 * Alias name of args:
731 * NAME=FETCHARG : set NAME as alias of FETCHARG.
733 * FETCHARG:TYPE : use TYPE instead of unsigned long.
735 struct trace_kprobe *tk = NULL;
737 bool is_return = false;
738 char *symbol = NULL, *tmp = NULL;
739 const char *event = NULL, *group = KPROBE_EVENT_SYSTEM;
743 char buf[MAX_EVENT_NAME_LEN];
744 unsigned int flags = TPARG_FL_KERNEL;
746 switch (argv[0][0]) {
749 flags |= TPARG_FL_RETURN;
759 trace_probe_log_init("trace_kprobe", argc, argv);
761 event = strchr(&argv[0][1], ':');
765 if (isdigit(argv[0][1])) {
767 trace_probe_log_err(1, MAXACT_NO_KPROBE);
771 len = event - &argv[0][1] - 1;
773 len = strlen(&argv[0][1]);
774 if (len > MAX_EVENT_NAME_LEN - 1) {
775 trace_probe_log_err(1, BAD_MAXACT);
778 memcpy(buf, &argv[0][1], len);
780 ret = kstrtouint(buf, 0, &maxactive);
781 if (ret || !maxactive) {
782 trace_probe_log_err(1, BAD_MAXACT);
785 /* kretprobes instances are iterated over via a list. The
786 * maximum should stay reasonable.
788 if (maxactive > KRETPROBE_MAXACTIVE_MAX) {
789 trace_probe_log_err(1, MAXACT_TOO_BIG);
794 /* try to parse an address. if that fails, try to read the
795 * input as a symbol. */
796 if (kstrtoul(argv[1], 0, (unsigned long *)&addr)) {
797 trace_probe_log_set_index(1);
798 /* Check whether uprobe event specified */
799 if (strchr(argv[1], '/') && strchr(argv[1], ':')) {
803 /* a symbol specified */
804 symbol = kstrdup(argv[1], GFP_KERNEL);
807 /* TODO: support .init module functions */
808 ret = traceprobe_split_symbol_offset(symbol, &offset);
809 if (ret || offset < 0 || offset > UINT_MAX) {
810 trace_probe_log_err(0, BAD_PROBE_ADDR);
813 if (kprobe_on_func_entry(NULL, symbol, offset))
814 flags |= TPARG_FL_FENTRY;
815 if (offset && is_return && !(flags & TPARG_FL_FENTRY)) {
816 trace_probe_log_err(0, BAD_RETPROBE);
821 trace_probe_log_set_index(0);
823 ret = traceprobe_parse_event_name(&event, &group, buf,
828 /* Make a new event name */
830 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
831 is_return ? 'r' : 'p', symbol, offset);
833 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
834 is_return ? 'r' : 'p', addr);
835 sanitize_event_name(buf);
840 tk = alloc_trace_kprobe(group, event, addr, symbol, offset, maxactive,
841 argc - 2, is_return);
844 /* This must return -ENOMEM, else there is a bug */
845 WARN_ON_ONCE(ret != -ENOMEM);
846 goto out; /* We know tk is not allocated */
848 argc -= 2; argv += 2;
850 /* parse arguments */
851 for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
852 tmp = kstrdup(argv[i], GFP_KERNEL);
858 trace_probe_log_set_index(i + 2);
859 ret = traceprobe_parse_probe_arg(&tk->tp, i, tmp, flags);
862 goto error; /* This can be -ENOMEM */
865 ret = traceprobe_set_print_fmt(&tk->tp, is_return);
869 ret = register_trace_kprobe(tk);
871 trace_probe_log_set_index(1);
873 trace_probe_log_err(0, BAD_INSN_BNDRY);
874 else if (ret == -ENOENT)
875 trace_probe_log_err(0, BAD_PROBE_ADDR);
876 else if (ret != -ENOMEM && ret != -EEXIST)
877 trace_probe_log_err(0, FAIL_REG_PROBE);
882 trace_probe_log_clear();
889 free_trace_kprobe(tk);
893 static int create_or_delete_trace_kprobe(int argc, char **argv)
897 if (argv[0][0] == '-')
898 return dyn_event_release(argc, argv, &trace_kprobe_ops);
900 ret = trace_kprobe_create(argc, (const char **)argv);
901 return ret == -ECANCELED ? -EINVAL : ret;
904 int trace_kprobe_run_command(const char *command)
906 return trace_run_command(command, create_or_delete_trace_kprobe);
909 static int trace_kprobe_run_cmd(struct dynevent_cmd *cmd)
911 return trace_run_command(cmd->buf, create_or_delete_trace_kprobe);
915 * kprobe_event_cmd_init - Initialize a kprobe event command object
916 * @cmd: A pointer to the dynevent_cmd struct representing the new event
917 * @buf: A pointer to the buffer used to build the command
918 * @maxlen: The length of the buffer passed in @buf
920 * Initialize a synthetic event command object. Use this before
921 * calling any of the other kprobe_event functions.
923 void kprobe_event_cmd_init(struct dynevent_cmd *cmd, char *buf, int maxlen)
925 dynevent_cmd_init(cmd, buf, maxlen, DYNEVENT_TYPE_KPROBE,
926 trace_kprobe_run_cmd);
928 EXPORT_SYMBOL_GPL(kprobe_event_cmd_init);
931 * __kprobe_event_gen_cmd_start - Generate a kprobe event command from arg list
932 * @cmd: A pointer to the dynevent_cmd struct representing the new event
933 * @name: The name of the kprobe event
934 * @loc: The location of the kprobe event
935 * @kretprobe: Is this a return probe?
936 * @args: Variable number of arg (pairs), one pair for each field
938 * NOTE: Users normally won't want to call this function directly, but
939 * rather use the kprobe_event_gen_cmd_start() wrapper, which automatically
940 * adds a NULL to the end of the arg list. If this function is used
941 * directly, make sure the last arg in the variable arg list is NULL.
943 * Generate a kprobe event command to be executed by
944 * kprobe_event_gen_cmd_end(). This function can be used to generate the
945 * complete command or only the first part of it; in the latter case,
946 * kprobe_event_add_fields() can be used to add more fields following this.
948 * Return: 0 if successful, error otherwise.
950 int __kprobe_event_gen_cmd_start(struct dynevent_cmd *cmd, bool kretprobe,
951 const char *name, const char *loc, ...)
953 char buf[MAX_EVENT_NAME_LEN];
954 struct dynevent_arg arg;
958 if (cmd->type != DYNEVENT_TYPE_KPROBE)
962 snprintf(buf, MAX_EVENT_NAME_LEN, "r:kprobes/%s", name);
964 snprintf(buf, MAX_EVENT_NAME_LEN, "p:kprobes/%s", name);
966 ret = dynevent_str_add(cmd, buf);
970 dynevent_arg_init(&arg, NULL, 0);
972 ret = dynevent_arg_add(cmd, &arg);
980 field = va_arg(args, const char *);
984 if (++cmd->n_fields > MAX_TRACE_ARGS) {
990 ret = dynevent_arg_add(cmd, &arg);
998 EXPORT_SYMBOL_GPL(__kprobe_event_gen_cmd_start);
1001 * __kprobe_event_add_fields - Add probe fields to a kprobe command from arg list
1002 * @cmd: A pointer to the dynevent_cmd struct representing the new event
1003 * @args: Variable number of arg (pairs), one pair for each field
1005 * NOTE: Users normally won't want to call this function directly, but
1006 * rather use the kprobe_event_add_fields() wrapper, which
1007 * automatically adds a NULL to the end of the arg list. If this
1008 * function is used directly, make sure the last arg in the variable
1011 * Add probe fields to an existing kprobe command using a variable
1012 * list of args. Fields are added in the same order they're listed.
1014 * Return: 0 if successful, error otherwise.
1016 int __kprobe_event_add_fields(struct dynevent_cmd *cmd, ...)
1018 struct dynevent_arg arg;
1022 if (cmd->type != DYNEVENT_TYPE_KPROBE)
1025 dynevent_arg_init(&arg, NULL, 0);
1027 va_start(args, cmd);
1031 field = va_arg(args, const char *);
1035 if (++cmd->n_fields > MAX_TRACE_ARGS) {
1041 ret = dynevent_arg_add(cmd, &arg);
1049 EXPORT_SYMBOL_GPL(__kprobe_event_add_fields);
1052 * kprobe_event_delete - Delete a kprobe event
1053 * @name: The name of the kprobe event to delete
1055 * Delete a kprobe event with the give @name from kernel code rather
1056 * than directly from the command line.
1058 * Return: 0 if successful, error otherwise.
1060 int kprobe_event_delete(const char *name)
1062 char buf[MAX_EVENT_NAME_LEN];
1064 snprintf(buf, MAX_EVENT_NAME_LEN, "-:%s", name);
1066 return trace_run_command(buf, create_or_delete_trace_kprobe);
1068 EXPORT_SYMBOL_GPL(kprobe_event_delete);
1070 static int trace_kprobe_release(struct dyn_event *ev)
1072 struct trace_kprobe *tk = to_trace_kprobe(ev);
1073 int ret = unregister_trace_kprobe(tk);
1076 free_trace_kprobe(tk);
1080 static int trace_kprobe_show(struct seq_file *m, struct dyn_event *ev)
1082 struct trace_kprobe *tk = to_trace_kprobe(ev);
1085 seq_putc(m, trace_kprobe_is_return(tk) ? 'r' : 'p');
1086 seq_printf(m, ":%s/%s", trace_probe_group_name(&tk->tp),
1087 trace_probe_name(&tk->tp));
1090 seq_printf(m, " 0x%p", tk->rp.kp.addr);
1091 else if (tk->rp.kp.offset)
1092 seq_printf(m, " %s+%u", trace_kprobe_symbol(tk),
1095 seq_printf(m, " %s", trace_kprobe_symbol(tk));
1097 for (i = 0; i < tk->tp.nr_args; i++)
1098 seq_printf(m, " %s=%s", tk->tp.args[i].name, tk->tp.args[i].comm);
1104 static int probes_seq_show(struct seq_file *m, void *v)
1106 struct dyn_event *ev = v;
1108 if (!is_trace_kprobe(ev))
1111 return trace_kprobe_show(m, ev);
1114 static const struct seq_operations probes_seq_op = {
1115 .start = dyn_event_seq_start,
1116 .next = dyn_event_seq_next,
1117 .stop = dyn_event_seq_stop,
1118 .show = probes_seq_show
1121 static int probes_open(struct inode *inode, struct file *file)
1125 ret = security_locked_down(LOCKDOWN_TRACEFS);
1129 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
1130 ret = dyn_events_release_all(&trace_kprobe_ops);
1135 return seq_open(file, &probes_seq_op);
1138 static ssize_t probes_write(struct file *file, const char __user *buffer,
1139 size_t count, loff_t *ppos)
1141 return trace_parse_run_command(file, buffer, count, ppos,
1142 create_or_delete_trace_kprobe);
1145 static const struct file_operations kprobe_events_ops = {
1146 .owner = THIS_MODULE,
1147 .open = probes_open,
1149 .llseek = seq_lseek,
1150 .release = seq_release,
1151 .write = probes_write,
1154 /* Probes profiling interfaces */
1155 static int probes_profile_seq_show(struct seq_file *m, void *v)
1157 struct dyn_event *ev = v;
1158 struct trace_kprobe *tk;
1160 if (!is_trace_kprobe(ev))
1163 tk = to_trace_kprobe(ev);
1164 seq_printf(m, " %-44s %15lu %15lu\n",
1165 trace_probe_name(&tk->tp),
1166 trace_kprobe_nhit(tk),
1172 static const struct seq_operations profile_seq_op = {
1173 .start = dyn_event_seq_start,
1174 .next = dyn_event_seq_next,
1175 .stop = dyn_event_seq_stop,
1176 .show = probes_profile_seq_show
1179 static int profile_open(struct inode *inode, struct file *file)
1183 ret = security_locked_down(LOCKDOWN_TRACEFS);
1187 return seq_open(file, &profile_seq_op);
1190 static const struct file_operations kprobe_profile_ops = {
1191 .owner = THIS_MODULE,
1192 .open = profile_open,
1194 .llseek = seq_lseek,
1195 .release = seq_release,
1198 /* Kprobe specific fetch functions */
1200 /* Return the length of string -- including null terminal byte */
1201 static nokprobe_inline int
1202 fetch_store_strlen(unsigned long addr)
1208 ret = probe_kernel_read(&c, (u8 *)addr + len, 1);
1210 } while (c && ret == 0 && len < MAX_STRING_SIZE);
1212 return (ret < 0) ? ret : len;
1215 /* Return the length of string -- including null terminal byte */
1216 static nokprobe_inline int
1217 fetch_store_strlen_user(unsigned long addr)
1219 const void __user *uaddr = (__force const void __user *)addr;
1221 return strnlen_unsafe_user(uaddr, MAX_STRING_SIZE);
1225 * Fetch a null-terminated string. Caller MUST set *(u32 *)buf with max
1226 * length and relative data location.
1228 static nokprobe_inline int
1229 fetch_store_string(unsigned long addr, void *dest, void *base)
1231 int maxlen = get_loc_len(*(u32 *)dest);
1235 if (unlikely(!maxlen))
1238 __dest = get_loc_data(dest, base);
1241 * Try to get string again, since the string can be changed while
1244 ret = strncpy_from_unsafe(__dest, (void *)addr, maxlen);
1246 *(u32 *)dest = make_data_loc(ret, __dest - base);
1252 * Fetch a null-terminated string from user. Caller MUST set *(u32 *)buf
1253 * with max length and relative data location.
1255 static nokprobe_inline int
1256 fetch_store_string_user(unsigned long addr, void *dest, void *base)
1258 const void __user *uaddr = (__force const void __user *)addr;
1259 int maxlen = get_loc_len(*(u32 *)dest);
1263 if (unlikely(!maxlen))
1266 __dest = get_loc_data(dest, base);
1268 ret = strncpy_from_unsafe_user(__dest, uaddr, maxlen);
1270 *(u32 *)dest = make_data_loc(ret, __dest - base);
1275 static nokprobe_inline int
1276 probe_mem_read(void *dest, void *src, size_t size)
1278 return probe_kernel_read(dest, src, size);
1281 static nokprobe_inline int
1282 probe_mem_read_user(void *dest, void *src, size_t size)
1284 const void __user *uaddr = (__force const void __user *)src;
1286 return probe_user_read(dest, uaddr, size);
1289 /* Note that we don't verify it, since the code does not come from user space */
1291 process_fetch_insn(struct fetch_insn *code, struct pt_regs *regs, void *dest,
1297 /* 1st stage: get value from context */
1300 val = regs_get_register(regs, code->param);
1302 case FETCH_OP_STACK:
1303 val = regs_get_kernel_stack_nth(regs, code->param);
1305 case FETCH_OP_STACKP:
1306 val = kernel_stack_pointer(regs);
1308 case FETCH_OP_RETVAL:
1309 val = regs_return_value(regs);
1312 val = code->immediate;
1315 val = (unsigned long)current->comm;
1318 val = (unsigned long)code->data;
1320 #ifdef CONFIG_HAVE_FUNCTION_ARG_ACCESS_API
1322 val = regs_get_kernel_argument(regs, code->param);
1325 case FETCH_NOP_SYMBOL: /* Ignore a place holder */
1333 return process_fetch_insn_bottom(code, val, dest, base);
1335 NOKPROBE_SYMBOL(process_fetch_insn)
1337 /* Kprobe handler */
1338 static nokprobe_inline void
1339 __kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs,
1340 struct trace_event_file *trace_file)
1342 struct kprobe_trace_entry_head *entry;
1343 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1344 struct trace_event_buffer fbuffer;
1347 WARN_ON(call != trace_file->event_call);
1349 if (trace_trigger_soft_disabled(trace_file))
1352 local_save_flags(fbuffer.flags);
1353 fbuffer.pc = preempt_count();
1354 fbuffer.trace_file = trace_file;
1356 dsize = __get_data_size(&tk->tp, regs);
1359 trace_event_buffer_lock_reserve(&fbuffer.buffer, trace_file,
1361 sizeof(*entry) + tk->tp.size + dsize,
1362 fbuffer.flags, fbuffer.pc);
1366 fbuffer.regs = regs;
1367 entry = fbuffer.entry = ring_buffer_event_data(fbuffer.event);
1368 entry->ip = (unsigned long)tk->rp.kp.addr;
1369 store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1371 trace_event_buffer_commit(&fbuffer);
1375 kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs)
1377 struct event_file_link *link;
1379 trace_probe_for_each_link_rcu(link, &tk->tp)
1380 __kprobe_trace_func(tk, regs, link->file);
1382 NOKPROBE_SYMBOL(kprobe_trace_func);
1384 /* Kretprobe handler */
1385 static nokprobe_inline void
1386 __kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1387 struct pt_regs *regs,
1388 struct trace_event_file *trace_file)
1390 struct kretprobe_trace_entry_head *entry;
1391 struct trace_event_buffer fbuffer;
1392 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1395 WARN_ON(call != trace_file->event_call);
1397 if (trace_trigger_soft_disabled(trace_file))
1400 local_save_flags(fbuffer.flags);
1401 fbuffer.pc = preempt_count();
1402 fbuffer.trace_file = trace_file;
1404 dsize = __get_data_size(&tk->tp, regs);
1406 trace_event_buffer_lock_reserve(&fbuffer.buffer, trace_file,
1408 sizeof(*entry) + tk->tp.size + dsize,
1409 fbuffer.flags, fbuffer.pc);
1413 fbuffer.regs = regs;
1414 entry = fbuffer.entry = ring_buffer_event_data(fbuffer.event);
1415 entry->func = (unsigned long)tk->rp.kp.addr;
1416 entry->ret_ip = (unsigned long)ri->ret_addr;
1417 store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1419 trace_event_buffer_commit(&fbuffer);
1423 kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1424 struct pt_regs *regs)
1426 struct event_file_link *link;
1428 trace_probe_for_each_link_rcu(link, &tk->tp)
1429 __kretprobe_trace_func(tk, ri, regs, link->file);
1431 NOKPROBE_SYMBOL(kretprobe_trace_func);
1433 /* Event entry printers */
1434 static enum print_line_t
1435 print_kprobe_event(struct trace_iterator *iter, int flags,
1436 struct trace_event *event)
1438 struct kprobe_trace_entry_head *field;
1439 struct trace_seq *s = &iter->seq;
1440 struct trace_probe *tp;
1442 field = (struct kprobe_trace_entry_head *)iter->ent;
1443 tp = trace_probe_primary_from_call(
1444 container_of(event, struct trace_event_call, event));
1445 if (WARN_ON_ONCE(!tp))
1448 trace_seq_printf(s, "%s: (", trace_probe_name(tp));
1450 if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
1453 trace_seq_putc(s, ')');
1455 if (print_probe_args(s, tp->args, tp->nr_args,
1456 (u8 *)&field[1], field) < 0)
1459 trace_seq_putc(s, '\n');
1461 return trace_handle_return(s);
1464 static enum print_line_t
1465 print_kretprobe_event(struct trace_iterator *iter, int flags,
1466 struct trace_event *event)
1468 struct kretprobe_trace_entry_head *field;
1469 struct trace_seq *s = &iter->seq;
1470 struct trace_probe *tp;
1472 field = (struct kretprobe_trace_entry_head *)iter->ent;
1473 tp = trace_probe_primary_from_call(
1474 container_of(event, struct trace_event_call, event));
1475 if (WARN_ON_ONCE(!tp))
1478 trace_seq_printf(s, "%s: (", trace_probe_name(tp));
1480 if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1483 trace_seq_puts(s, " <- ");
1485 if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1488 trace_seq_putc(s, ')');
1490 if (print_probe_args(s, tp->args, tp->nr_args,
1491 (u8 *)&field[1], field) < 0)
1494 trace_seq_putc(s, '\n');
1497 return trace_handle_return(s);
1501 static int kprobe_event_define_fields(struct trace_event_call *event_call)
1504 struct kprobe_trace_entry_head field;
1505 struct trace_probe *tp;
1507 tp = trace_probe_primary_from_call(event_call);
1508 if (WARN_ON_ONCE(!tp))
1511 DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1513 return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
1516 static int kretprobe_event_define_fields(struct trace_event_call *event_call)
1519 struct kretprobe_trace_entry_head field;
1520 struct trace_probe *tp;
1522 tp = trace_probe_primary_from_call(event_call);
1523 if (WARN_ON_ONCE(!tp))
1526 DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1527 DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1529 return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
1532 #ifdef CONFIG_PERF_EVENTS
1534 /* Kprobe profile handler */
1536 kprobe_perf_func(struct trace_kprobe *tk, struct pt_regs *regs)
1538 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1539 struct kprobe_trace_entry_head *entry;
1540 struct hlist_head *head;
1541 int size, __size, dsize;
1544 if (bpf_prog_array_valid(call)) {
1545 unsigned long orig_ip = instruction_pointer(regs);
1548 ret = trace_call_bpf(call, regs);
1551 * We need to check and see if we modified the pc of the
1552 * pt_regs, and if so return 1 so that we don't do the
1555 if (orig_ip != instruction_pointer(regs))
1561 head = this_cpu_ptr(call->perf_events);
1562 if (hlist_empty(head))
1565 dsize = __get_data_size(&tk->tp, regs);
1566 __size = sizeof(*entry) + tk->tp.size + dsize;
1567 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1568 size -= sizeof(u32);
1570 entry = perf_trace_buf_alloc(size, NULL, &rctx);
1574 entry->ip = (unsigned long)tk->rp.kp.addr;
1575 memset(&entry[1], 0, dsize);
1576 store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1577 perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1581 NOKPROBE_SYMBOL(kprobe_perf_func);
1583 /* Kretprobe profile handler */
1585 kretprobe_perf_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1586 struct pt_regs *regs)
1588 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1589 struct kretprobe_trace_entry_head *entry;
1590 struct hlist_head *head;
1591 int size, __size, dsize;
1594 if (bpf_prog_array_valid(call) && !trace_call_bpf(call, regs))
1597 head = this_cpu_ptr(call->perf_events);
1598 if (hlist_empty(head))
1601 dsize = __get_data_size(&tk->tp, regs);
1602 __size = sizeof(*entry) + tk->tp.size + dsize;
1603 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1604 size -= sizeof(u32);
1606 entry = perf_trace_buf_alloc(size, NULL, &rctx);
1610 entry->func = (unsigned long)tk->rp.kp.addr;
1611 entry->ret_ip = (unsigned long)ri->ret_addr;
1612 store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1613 perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1616 NOKPROBE_SYMBOL(kretprobe_perf_func);
1618 int bpf_get_kprobe_info(const struct perf_event *event, u32 *fd_type,
1619 const char **symbol, u64 *probe_offset,
1620 u64 *probe_addr, bool perf_type_tracepoint)
1622 const char *pevent = trace_event_name(event->tp_event);
1623 const char *group = event->tp_event->class->system;
1624 struct trace_kprobe *tk;
1626 if (perf_type_tracepoint)
1627 tk = find_trace_kprobe(pevent, group);
1629 tk = event->tp_event->data;
1633 *fd_type = trace_kprobe_is_return(tk) ? BPF_FD_TYPE_KRETPROBE
1634 : BPF_FD_TYPE_KPROBE;
1636 *symbol = tk->symbol;
1637 *probe_offset = tk->rp.kp.offset;
1642 *probe_addr = (unsigned long)tk->rp.kp.addr;
1646 #endif /* CONFIG_PERF_EVENTS */
1649 * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex.
1651 * kprobe_trace_self_tests_init() does enable_trace_probe/disable_trace_probe
1652 * lockless, but we can't race with this __init function.
1654 static int kprobe_register(struct trace_event_call *event,
1655 enum trace_reg type, void *data)
1657 struct trace_event_file *file = data;
1660 case TRACE_REG_REGISTER:
1661 return enable_trace_kprobe(event, file);
1662 case TRACE_REG_UNREGISTER:
1663 return disable_trace_kprobe(event, file);
1665 #ifdef CONFIG_PERF_EVENTS
1666 case TRACE_REG_PERF_REGISTER:
1667 return enable_trace_kprobe(event, NULL);
1668 case TRACE_REG_PERF_UNREGISTER:
1669 return disable_trace_kprobe(event, NULL);
1670 case TRACE_REG_PERF_OPEN:
1671 case TRACE_REG_PERF_CLOSE:
1672 case TRACE_REG_PERF_ADD:
1673 case TRACE_REG_PERF_DEL:
1680 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1682 struct trace_kprobe *tk = container_of(kp, struct trace_kprobe, rp.kp);
1685 raw_cpu_inc(*tk->nhit);
1687 if (trace_probe_test_flag(&tk->tp, TP_FLAG_TRACE))
1688 kprobe_trace_func(tk, regs);
1689 #ifdef CONFIG_PERF_EVENTS
1690 if (trace_probe_test_flag(&tk->tp, TP_FLAG_PROFILE))
1691 ret = kprobe_perf_func(tk, regs);
1695 NOKPROBE_SYMBOL(kprobe_dispatcher);
1698 kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1700 struct trace_kprobe *tk = container_of(ri->rp, struct trace_kprobe, rp);
1702 raw_cpu_inc(*tk->nhit);
1704 if (trace_probe_test_flag(&tk->tp, TP_FLAG_TRACE))
1705 kretprobe_trace_func(tk, ri, regs);
1706 #ifdef CONFIG_PERF_EVENTS
1707 if (trace_probe_test_flag(&tk->tp, TP_FLAG_PROFILE))
1708 kretprobe_perf_func(tk, ri, regs);
1710 return 0; /* We don't tweek kernel, so just return 0 */
1712 NOKPROBE_SYMBOL(kretprobe_dispatcher);
1714 static struct trace_event_functions kretprobe_funcs = {
1715 .trace = print_kretprobe_event
1718 static struct trace_event_functions kprobe_funcs = {
1719 .trace = print_kprobe_event
1722 static inline void init_trace_event_call(struct trace_kprobe *tk)
1724 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1726 if (trace_kprobe_is_return(tk)) {
1727 call->event.funcs = &kretprobe_funcs;
1728 call->class->define_fields = kretprobe_event_define_fields;
1730 call->event.funcs = &kprobe_funcs;
1731 call->class->define_fields = kprobe_event_define_fields;
1734 call->flags = TRACE_EVENT_FL_KPROBE;
1735 call->class->reg = kprobe_register;
1738 static int register_kprobe_event(struct trace_kprobe *tk)
1740 init_trace_event_call(tk);
1742 return trace_probe_register_event_call(&tk->tp);
1745 static int unregister_kprobe_event(struct trace_kprobe *tk)
1747 return trace_probe_unregister_event_call(&tk->tp);
1750 #ifdef CONFIG_PERF_EVENTS
1751 /* create a trace_kprobe, but don't add it to global lists */
1752 struct trace_event_call *
1753 create_local_trace_kprobe(char *func, void *addr, unsigned long offs,
1756 struct trace_kprobe *tk;
1761 * local trace_kprobes are not added to dyn_event, so they are never
1762 * searched in find_trace_kprobe(). Therefore, there is no concern of
1763 * duplicated name here.
1765 event = func ? func : "DUMMY_EVENT";
1767 tk = alloc_trace_kprobe(KPROBE_EVENT_SYSTEM, event, (void *)addr, func,
1768 offs, 0 /* maxactive */, 0 /* nargs */,
1772 pr_info("Failed to allocate trace_probe.(%d)\n",
1774 return ERR_CAST(tk);
1777 init_trace_event_call(tk);
1779 if (traceprobe_set_print_fmt(&tk->tp, trace_kprobe_is_return(tk)) < 0) {
1784 ret = __register_trace_kprobe(tk);
1788 return trace_probe_event_call(&tk->tp);
1790 free_trace_kprobe(tk);
1791 return ERR_PTR(ret);
1794 void destroy_local_trace_kprobe(struct trace_event_call *event_call)
1796 struct trace_kprobe *tk;
1798 tk = trace_kprobe_primary_from_call(event_call);
1802 if (trace_probe_is_enabled(&tk->tp)) {
1807 __unregister_trace_kprobe(tk);
1809 free_trace_kprobe(tk);
1811 #endif /* CONFIG_PERF_EVENTS */
1813 static __init void enable_boot_kprobe_events(void)
1815 struct trace_array *tr = top_trace_array();
1816 struct trace_event_file *file;
1817 struct trace_kprobe *tk;
1818 struct dyn_event *pos;
1820 mutex_lock(&event_mutex);
1821 for_each_trace_kprobe(tk, pos) {
1822 list_for_each_entry(file, &tr->events, list)
1823 if (file->event_call == trace_probe_event_call(&tk->tp))
1824 trace_event_enable_disable(file, 1, 0);
1826 mutex_unlock(&event_mutex);
1829 static __init void setup_boot_kprobe_events(void)
1831 char *p, *cmd = kprobe_boot_events_buf;
1834 strreplace(kprobe_boot_events_buf, ',', ' ');
1836 while (cmd && *cmd != '\0') {
1837 p = strchr(cmd, ';');
1841 ret = trace_run_command(cmd, create_or_delete_trace_kprobe);
1843 pr_warn("Failed to add event(%d): %s\n", ret, cmd);
1845 kprobe_boot_events_enabled = true;
1850 enable_boot_kprobe_events();
1854 * Register dynevent at subsys_initcall. This allows kernel to setup kprobe
1855 * events in fs_initcall without tracefs.
1857 static __init int init_kprobe_trace_early(void)
1861 ret = dyn_event_register(&trace_kprobe_ops);
1865 if (register_module_notifier(&trace_kprobe_module_nb))
1870 subsys_initcall(init_kprobe_trace_early);
1872 /* Make a tracefs interface for controlling probe points */
1873 static __init int init_kprobe_trace(void)
1875 struct dentry *d_tracer;
1876 struct dentry *entry;
1878 d_tracer = tracing_init_dentry();
1879 if (IS_ERR(d_tracer))
1882 entry = tracefs_create_file("kprobe_events", 0644, d_tracer,
1883 NULL, &kprobe_events_ops);
1885 /* Event list interface */
1887 pr_warn("Could not create tracefs 'kprobe_events' entry\n");
1889 /* Profile interface */
1890 entry = tracefs_create_file("kprobe_profile", 0444, d_tracer,
1891 NULL, &kprobe_profile_ops);
1894 pr_warn("Could not create tracefs 'kprobe_profile' entry\n");
1896 setup_boot_kprobe_events();
1900 fs_initcall(init_kprobe_trace);
1903 #ifdef CONFIG_FTRACE_STARTUP_TEST
1904 static __init struct trace_event_file *
1905 find_trace_probe_file(struct trace_kprobe *tk, struct trace_array *tr)
1907 struct trace_event_file *file;
1909 list_for_each_entry(file, &tr->events, list)
1910 if (file->event_call == trace_probe_event_call(&tk->tp))
1917 * Nobody but us can call enable_trace_kprobe/disable_trace_kprobe at this
1918 * stage, we can do this lockless.
1920 static __init int kprobe_trace_self_tests_init(void)
1923 int (*target)(int, int, int, int, int, int);
1924 struct trace_kprobe *tk;
1925 struct trace_event_file *file;
1927 if (tracing_is_disabled())
1930 if (kprobe_boot_events_enabled) {
1931 pr_info("Skipping kprobe tests due to kprobe_event on cmdline\n");
1935 target = kprobe_trace_selftest_target;
1937 pr_info("Testing kprobe tracing: ");
1939 ret = trace_run_command("p:testprobe kprobe_trace_selftest_target $stack $stack0 +0($stack)",
1940 create_or_delete_trace_kprobe);
1941 if (WARN_ON_ONCE(ret)) {
1942 pr_warn("error on probing function entry.\n");
1945 /* Enable trace point */
1946 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1947 if (WARN_ON_ONCE(tk == NULL)) {
1948 pr_warn("error on getting new probe.\n");
1951 file = find_trace_probe_file(tk, top_trace_array());
1952 if (WARN_ON_ONCE(file == NULL)) {
1953 pr_warn("error on getting probe file.\n");
1956 enable_trace_kprobe(
1957 trace_probe_event_call(&tk->tp), file);
1961 ret = trace_run_command("r:testprobe2 kprobe_trace_selftest_target $retval",
1962 create_or_delete_trace_kprobe);
1963 if (WARN_ON_ONCE(ret)) {
1964 pr_warn("error on probing function return.\n");
1967 /* Enable trace point */
1968 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1969 if (WARN_ON_ONCE(tk == NULL)) {
1970 pr_warn("error on getting 2nd new probe.\n");
1973 file = find_trace_probe_file(tk, top_trace_array());
1974 if (WARN_ON_ONCE(file == NULL)) {
1975 pr_warn("error on getting probe file.\n");
1978 enable_trace_kprobe(
1979 trace_probe_event_call(&tk->tp), file);
1986 ret = target(1, 2, 3, 4, 5, 6);
1989 * Not expecting an error here, the check is only to prevent the
1990 * optimizer from removing the call to target() as otherwise there
1991 * are no side-effects and the call is never performed.
1996 /* Disable trace points before removing it */
1997 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1998 if (WARN_ON_ONCE(tk == NULL)) {
1999 pr_warn("error on getting test probe.\n");
2002 if (trace_kprobe_nhit(tk) != 1) {
2003 pr_warn("incorrect number of testprobe hits\n");
2007 file = find_trace_probe_file(tk, top_trace_array());
2008 if (WARN_ON_ONCE(file == NULL)) {
2009 pr_warn("error on getting probe file.\n");
2012 disable_trace_kprobe(
2013 trace_probe_event_call(&tk->tp), file);
2016 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
2017 if (WARN_ON_ONCE(tk == NULL)) {
2018 pr_warn("error on getting 2nd test probe.\n");
2021 if (trace_kprobe_nhit(tk) != 1) {
2022 pr_warn("incorrect number of testprobe2 hits\n");
2026 file = find_trace_probe_file(tk, top_trace_array());
2027 if (WARN_ON_ONCE(file == NULL)) {
2028 pr_warn("error on getting probe file.\n");
2031 disable_trace_kprobe(
2032 trace_probe_event_call(&tk->tp), file);
2035 ret = trace_run_command("-:testprobe", create_or_delete_trace_kprobe);
2036 if (WARN_ON_ONCE(ret)) {
2037 pr_warn("error on deleting a probe.\n");
2041 ret = trace_run_command("-:testprobe2", create_or_delete_trace_kprobe);
2042 if (WARN_ON_ONCE(ret)) {
2043 pr_warn("error on deleting a probe.\n");
2048 ret = dyn_events_release_all(&trace_kprobe_ops);
2049 if (WARN_ON_ONCE(ret)) {
2050 pr_warn("error on cleaning up probes.\n");
2054 * Wait for the optimizer work to finish. Otherwise it might fiddle
2055 * with probes in already freed __init text.
2057 wait_for_kprobe_optimizer();
2059 pr_cont("NG: Some tests are failed. Please check them.\n");
2065 late_initcall(kprobe_trace_self_tests_init);