Merge tag 'powerpc-5.11-3' of git://git.kernel.org/pub/scm/linux/kernel/git/powerpc...
[linux/fpc-iii.git] / kernel / trace / trace_kprobe.c
blob9c31f42245e93b6ae6386da8370274f61f1de173
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Kprobes-based tracing events
5 * Created by Masami Hiramatsu <mhiramat@redhat.com>
7 */
8 #define pr_fmt(fmt) "trace_kprobe: " fmt
10 #include <linux/security.h>
11 #include <linux/module.h>
12 #include <linux/uaccess.h>
13 #include <linux/rculist.h>
14 #include <linux/error-injection.h>
16 #include <asm/setup.h> /* for COMMAND_LINE_SIZE */
18 #include "trace_dynevent.h"
19 #include "trace_kprobe_selftest.h"
20 #include "trace_probe.h"
21 #include "trace_probe_tmpl.h"
23 #define KPROBE_EVENT_SYSTEM "kprobes"
24 #define KRETPROBE_MAXACTIVE_MAX 4096
26 /* Kprobe early definition from command line */
27 static char kprobe_boot_events_buf[COMMAND_LINE_SIZE] __initdata;
29 static int __init set_kprobe_boot_events(char *str)
31 strlcpy(kprobe_boot_events_buf, str, COMMAND_LINE_SIZE);
32 disable_tracing_selftest("running kprobe events");
34 return 0;
36 __setup("kprobe_event=", set_kprobe_boot_events);
38 static int trace_kprobe_create(int argc, const char **argv);
39 static int trace_kprobe_show(struct seq_file *m, struct dyn_event *ev);
40 static int trace_kprobe_release(struct dyn_event *ev);
41 static bool trace_kprobe_is_busy(struct dyn_event *ev);
42 static bool trace_kprobe_match(const char *system, const char *event,
43 int argc, const char **argv, struct dyn_event *ev);
45 static struct dyn_event_operations trace_kprobe_ops = {
46 .create = trace_kprobe_create,
47 .show = trace_kprobe_show,
48 .is_busy = trace_kprobe_is_busy,
49 .free = trace_kprobe_release,
50 .match = trace_kprobe_match,
54 * Kprobe event core functions
56 struct trace_kprobe {
57 struct dyn_event devent;
58 struct kretprobe rp; /* Use rp.kp for kprobe use */
59 unsigned long __percpu *nhit;
60 const char *symbol; /* symbol name */
61 struct trace_probe tp;
64 static bool is_trace_kprobe(struct dyn_event *ev)
66 return ev->ops == &trace_kprobe_ops;
69 static struct trace_kprobe *to_trace_kprobe(struct dyn_event *ev)
71 return container_of(ev, struct trace_kprobe, devent);
74 /**
75 * for_each_trace_kprobe - iterate over the trace_kprobe list
76 * @pos: the struct trace_kprobe * for each entry
77 * @dpos: the struct dyn_event * to use as a loop cursor
79 #define for_each_trace_kprobe(pos, dpos) \
80 for_each_dyn_event(dpos) \
81 if (is_trace_kprobe(dpos) && (pos = to_trace_kprobe(dpos)))
83 #define SIZEOF_TRACE_KPROBE(n) \
84 (offsetof(struct trace_kprobe, tp.args) + \
85 (sizeof(struct probe_arg) * (n)))
87 static nokprobe_inline bool trace_kprobe_is_return(struct trace_kprobe *tk)
89 return tk->rp.handler != NULL;
92 static nokprobe_inline const char *trace_kprobe_symbol(struct trace_kprobe *tk)
94 return tk->symbol ? tk->symbol : "unknown";
97 static nokprobe_inline unsigned long trace_kprobe_offset(struct trace_kprobe *tk)
99 return tk->rp.kp.offset;
102 static nokprobe_inline bool trace_kprobe_has_gone(struct trace_kprobe *tk)
104 return !!(kprobe_gone(&tk->rp.kp));
107 static nokprobe_inline bool trace_kprobe_within_module(struct trace_kprobe *tk,
108 struct module *mod)
110 int len = strlen(module_name(mod));
111 const char *name = trace_kprobe_symbol(tk);
113 return strncmp(module_name(mod), name, len) == 0 && name[len] == ':';
116 static nokprobe_inline bool trace_kprobe_module_exist(struct trace_kprobe *tk)
118 char *p;
119 bool ret;
121 if (!tk->symbol)
122 return false;
123 p = strchr(tk->symbol, ':');
124 if (!p)
125 return true;
126 *p = '\0';
127 mutex_lock(&module_mutex);
128 ret = !!find_module(tk->symbol);
129 mutex_unlock(&module_mutex);
130 *p = ':';
132 return ret;
135 static bool trace_kprobe_is_busy(struct dyn_event *ev)
137 struct trace_kprobe *tk = to_trace_kprobe(ev);
139 return trace_probe_is_enabled(&tk->tp);
142 static bool trace_kprobe_match_command_head(struct trace_kprobe *tk,
143 int argc, const char **argv)
145 char buf[MAX_ARGSTR_LEN + 1];
147 if (!argc)
148 return true;
150 if (!tk->symbol)
151 snprintf(buf, sizeof(buf), "0x%p", tk->rp.kp.addr);
152 else if (tk->rp.kp.offset)
153 snprintf(buf, sizeof(buf), "%s+%u",
154 trace_kprobe_symbol(tk), tk->rp.kp.offset);
155 else
156 snprintf(buf, sizeof(buf), "%s", trace_kprobe_symbol(tk));
157 if (strcmp(buf, argv[0]))
158 return false;
159 argc--; argv++;
161 return trace_probe_match_command_args(&tk->tp, argc, argv);
164 static bool trace_kprobe_match(const char *system, const char *event,
165 int argc, const char **argv, struct dyn_event *ev)
167 struct trace_kprobe *tk = to_trace_kprobe(ev);
169 return strcmp(trace_probe_name(&tk->tp), event) == 0 &&
170 (!system || strcmp(trace_probe_group_name(&tk->tp), system) == 0) &&
171 trace_kprobe_match_command_head(tk, argc, argv);
174 static nokprobe_inline unsigned long trace_kprobe_nhit(struct trace_kprobe *tk)
176 unsigned long nhit = 0;
177 int cpu;
179 for_each_possible_cpu(cpu)
180 nhit += *per_cpu_ptr(tk->nhit, cpu);
182 return nhit;
185 static nokprobe_inline bool trace_kprobe_is_registered(struct trace_kprobe *tk)
187 return !(list_empty(&tk->rp.kp.list) &&
188 hlist_unhashed(&tk->rp.kp.hlist));
191 /* Return 0 if it fails to find the symbol address */
192 static nokprobe_inline
193 unsigned long trace_kprobe_address(struct trace_kprobe *tk)
195 unsigned long addr;
197 if (tk->symbol) {
198 addr = (unsigned long)
199 kallsyms_lookup_name(trace_kprobe_symbol(tk));
200 if (addr)
201 addr += tk->rp.kp.offset;
202 } else {
203 addr = (unsigned long)tk->rp.kp.addr;
205 return addr;
208 static nokprobe_inline struct trace_kprobe *
209 trace_kprobe_primary_from_call(struct trace_event_call *call)
211 struct trace_probe *tp;
213 tp = trace_probe_primary_from_call(call);
214 if (WARN_ON_ONCE(!tp))
215 return NULL;
217 return container_of(tp, struct trace_kprobe, tp);
220 bool trace_kprobe_on_func_entry(struct trace_event_call *call)
222 struct trace_kprobe *tk = trace_kprobe_primary_from_call(call);
224 return tk ? kprobe_on_func_entry(tk->rp.kp.addr,
225 tk->rp.kp.addr ? NULL : tk->rp.kp.symbol_name,
226 tk->rp.kp.addr ? 0 : tk->rp.kp.offset) : false;
229 bool trace_kprobe_error_injectable(struct trace_event_call *call)
231 struct trace_kprobe *tk = trace_kprobe_primary_from_call(call);
233 return tk ? within_error_injection_list(trace_kprobe_address(tk)) :
234 false;
237 static int register_kprobe_event(struct trace_kprobe *tk);
238 static int unregister_kprobe_event(struct trace_kprobe *tk);
240 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
241 static int kretprobe_dispatcher(struct kretprobe_instance *ri,
242 struct pt_regs *regs);
244 static void free_trace_kprobe(struct trace_kprobe *tk)
246 if (tk) {
247 trace_probe_cleanup(&tk->tp);
248 kfree(tk->symbol);
249 free_percpu(tk->nhit);
250 kfree(tk);
255 * Allocate new trace_probe and initialize it (including kprobes).
257 static struct trace_kprobe *alloc_trace_kprobe(const char *group,
258 const char *event,
259 void *addr,
260 const char *symbol,
261 unsigned long offs,
262 int maxactive,
263 int nargs, bool is_return)
265 struct trace_kprobe *tk;
266 int ret = -ENOMEM;
268 tk = kzalloc(SIZEOF_TRACE_KPROBE(nargs), GFP_KERNEL);
269 if (!tk)
270 return ERR_PTR(ret);
272 tk->nhit = alloc_percpu(unsigned long);
273 if (!tk->nhit)
274 goto error;
276 if (symbol) {
277 tk->symbol = kstrdup(symbol, GFP_KERNEL);
278 if (!tk->symbol)
279 goto error;
280 tk->rp.kp.symbol_name = tk->symbol;
281 tk->rp.kp.offset = offs;
282 } else
283 tk->rp.kp.addr = addr;
285 if (is_return)
286 tk->rp.handler = kretprobe_dispatcher;
287 else
288 tk->rp.kp.pre_handler = kprobe_dispatcher;
290 tk->rp.maxactive = maxactive;
291 INIT_HLIST_NODE(&tk->rp.kp.hlist);
292 INIT_LIST_HEAD(&tk->rp.kp.list);
294 ret = trace_probe_init(&tk->tp, event, group, false);
295 if (ret < 0)
296 goto error;
298 dyn_event_init(&tk->devent, &trace_kprobe_ops);
299 return tk;
300 error:
301 free_trace_kprobe(tk);
302 return ERR_PTR(ret);
305 static struct trace_kprobe *find_trace_kprobe(const char *event,
306 const char *group)
308 struct dyn_event *pos;
309 struct trace_kprobe *tk;
311 for_each_trace_kprobe(tk, pos)
312 if (strcmp(trace_probe_name(&tk->tp), event) == 0 &&
313 strcmp(trace_probe_group_name(&tk->tp), group) == 0)
314 return tk;
315 return NULL;
318 static inline int __enable_trace_kprobe(struct trace_kprobe *tk)
320 int ret = 0;
322 if (trace_kprobe_is_registered(tk) && !trace_kprobe_has_gone(tk)) {
323 if (trace_kprobe_is_return(tk))
324 ret = enable_kretprobe(&tk->rp);
325 else
326 ret = enable_kprobe(&tk->rp.kp);
329 return ret;
332 static void __disable_trace_kprobe(struct trace_probe *tp)
334 struct trace_probe *pos;
335 struct trace_kprobe *tk;
337 list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
338 tk = container_of(pos, struct trace_kprobe, tp);
339 if (!trace_kprobe_is_registered(tk))
340 continue;
341 if (trace_kprobe_is_return(tk))
342 disable_kretprobe(&tk->rp);
343 else
344 disable_kprobe(&tk->rp.kp);
349 * Enable trace_probe
350 * if the file is NULL, enable "perf" handler, or enable "trace" handler.
352 static int enable_trace_kprobe(struct trace_event_call *call,
353 struct trace_event_file *file)
355 struct trace_probe *pos, *tp;
356 struct trace_kprobe *tk;
357 bool enabled;
358 int ret = 0;
360 tp = trace_probe_primary_from_call(call);
361 if (WARN_ON_ONCE(!tp))
362 return -ENODEV;
363 enabled = trace_probe_is_enabled(tp);
365 /* This also changes "enabled" state */
366 if (file) {
367 ret = trace_probe_add_file(tp, file);
368 if (ret)
369 return ret;
370 } else
371 trace_probe_set_flag(tp, TP_FLAG_PROFILE);
373 if (enabled)
374 return 0;
376 list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
377 tk = container_of(pos, struct trace_kprobe, tp);
378 if (trace_kprobe_has_gone(tk))
379 continue;
380 ret = __enable_trace_kprobe(tk);
381 if (ret)
382 break;
383 enabled = true;
386 if (ret) {
387 /* Failed to enable one of them. Roll back all */
388 if (enabled)
389 __disable_trace_kprobe(tp);
390 if (file)
391 trace_probe_remove_file(tp, file);
392 else
393 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
396 return ret;
400 * Disable trace_probe
401 * if the file is NULL, disable "perf" handler, or disable "trace" handler.
403 static int disable_trace_kprobe(struct trace_event_call *call,
404 struct trace_event_file *file)
406 struct trace_probe *tp;
408 tp = trace_probe_primary_from_call(call);
409 if (WARN_ON_ONCE(!tp))
410 return -ENODEV;
412 if (file) {
413 if (!trace_probe_get_file_link(tp, file))
414 return -ENOENT;
415 if (!trace_probe_has_single_file(tp))
416 goto out;
417 trace_probe_clear_flag(tp, TP_FLAG_TRACE);
418 } else
419 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
421 if (!trace_probe_is_enabled(tp))
422 __disable_trace_kprobe(tp);
424 out:
425 if (file)
427 * Synchronization is done in below function. For perf event,
428 * file == NULL and perf_trace_event_unreg() calls
429 * tracepoint_synchronize_unregister() to ensure synchronize
430 * event. We don't need to care about it.
432 trace_probe_remove_file(tp, file);
434 return 0;
437 #if defined(CONFIG_KPROBES_ON_FTRACE) && \
438 !defined(CONFIG_KPROBE_EVENTS_ON_NOTRACE)
439 static bool __within_notrace_func(unsigned long addr)
441 unsigned long offset, size;
443 if (!addr || !kallsyms_lookup_size_offset(addr, &size, &offset))
444 return false;
446 /* Get the entry address of the target function */
447 addr -= offset;
450 * Since ftrace_location_range() does inclusive range check, we need
451 * to subtract 1 byte from the end address.
453 return !ftrace_location_range(addr, addr + size - 1);
456 static bool within_notrace_func(struct trace_kprobe *tk)
458 unsigned long addr = trace_kprobe_address(tk);
459 char symname[KSYM_NAME_LEN], *p;
461 if (!__within_notrace_func(addr))
462 return false;
464 /* Check if the address is on a suffixed-symbol */
465 if (!lookup_symbol_name(addr, symname)) {
466 p = strchr(symname, '.');
467 if (!p)
468 return true;
469 *p = '\0';
470 addr = (unsigned long)kprobe_lookup_name(symname, 0);
471 if (addr)
472 return __within_notrace_func(addr);
475 return true;
477 #else
478 #define within_notrace_func(tk) (false)
479 #endif
481 /* Internal register function - just handle k*probes and flags */
482 static int __register_trace_kprobe(struct trace_kprobe *tk)
484 int i, ret;
486 ret = security_locked_down(LOCKDOWN_KPROBES);
487 if (ret)
488 return ret;
490 if (trace_kprobe_is_registered(tk))
491 return -EINVAL;
493 if (within_notrace_func(tk)) {
494 pr_warn("Could not probe notrace function %s\n",
495 trace_kprobe_symbol(tk));
496 return -EINVAL;
499 for (i = 0; i < tk->tp.nr_args; i++) {
500 ret = traceprobe_update_arg(&tk->tp.args[i]);
501 if (ret)
502 return ret;
505 /* Set/clear disabled flag according to tp->flag */
506 if (trace_probe_is_enabled(&tk->tp))
507 tk->rp.kp.flags &= ~KPROBE_FLAG_DISABLED;
508 else
509 tk->rp.kp.flags |= KPROBE_FLAG_DISABLED;
511 if (trace_kprobe_is_return(tk))
512 ret = register_kretprobe(&tk->rp);
513 else
514 ret = register_kprobe(&tk->rp.kp);
516 return ret;
519 /* Internal unregister function - just handle k*probes and flags */
520 static void __unregister_trace_kprobe(struct trace_kprobe *tk)
522 if (trace_kprobe_is_registered(tk)) {
523 if (trace_kprobe_is_return(tk))
524 unregister_kretprobe(&tk->rp);
525 else
526 unregister_kprobe(&tk->rp.kp);
527 /* Cleanup kprobe for reuse and mark it unregistered */
528 INIT_HLIST_NODE(&tk->rp.kp.hlist);
529 INIT_LIST_HEAD(&tk->rp.kp.list);
530 if (tk->rp.kp.symbol_name)
531 tk->rp.kp.addr = NULL;
535 /* Unregister a trace_probe and probe_event */
536 static int unregister_trace_kprobe(struct trace_kprobe *tk)
538 /* If other probes are on the event, just unregister kprobe */
539 if (trace_probe_has_sibling(&tk->tp))
540 goto unreg;
542 /* Enabled event can not be unregistered */
543 if (trace_probe_is_enabled(&tk->tp))
544 return -EBUSY;
546 /* Will fail if probe is being used by ftrace or perf */
547 if (unregister_kprobe_event(tk))
548 return -EBUSY;
550 unreg:
551 __unregister_trace_kprobe(tk);
552 dyn_event_remove(&tk->devent);
553 trace_probe_unlink(&tk->tp);
555 return 0;
558 static bool trace_kprobe_has_same_kprobe(struct trace_kprobe *orig,
559 struct trace_kprobe *comp)
561 struct trace_probe_event *tpe = orig->tp.event;
562 struct trace_probe *pos;
563 int i;
565 list_for_each_entry(pos, &tpe->probes, list) {
566 orig = container_of(pos, struct trace_kprobe, tp);
567 if (strcmp(trace_kprobe_symbol(orig),
568 trace_kprobe_symbol(comp)) ||
569 trace_kprobe_offset(orig) != trace_kprobe_offset(comp))
570 continue;
573 * trace_probe_compare_arg_type() ensured that nr_args and
574 * each argument name and type are same. Let's compare comm.
576 for (i = 0; i < orig->tp.nr_args; i++) {
577 if (strcmp(orig->tp.args[i].comm,
578 comp->tp.args[i].comm))
579 break;
582 if (i == orig->tp.nr_args)
583 return true;
586 return false;
589 static int append_trace_kprobe(struct trace_kprobe *tk, struct trace_kprobe *to)
591 int ret;
593 ret = trace_probe_compare_arg_type(&tk->tp, &to->tp);
594 if (ret) {
595 /* Note that argument starts index = 2 */
596 trace_probe_log_set_index(ret + 1);
597 trace_probe_log_err(0, DIFF_ARG_TYPE);
598 return -EEXIST;
600 if (trace_kprobe_has_same_kprobe(to, tk)) {
601 trace_probe_log_set_index(0);
602 trace_probe_log_err(0, SAME_PROBE);
603 return -EEXIST;
606 /* Append to existing event */
607 ret = trace_probe_append(&tk->tp, &to->tp);
608 if (ret)
609 return ret;
611 /* Register k*probe */
612 ret = __register_trace_kprobe(tk);
613 if (ret == -ENOENT && !trace_kprobe_module_exist(tk)) {
614 pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
615 ret = 0;
618 if (ret)
619 trace_probe_unlink(&tk->tp);
620 else
621 dyn_event_add(&tk->devent);
623 return ret;
626 /* Register a trace_probe and probe_event */
627 static int register_trace_kprobe(struct trace_kprobe *tk)
629 struct trace_kprobe *old_tk;
630 int ret;
632 mutex_lock(&event_mutex);
634 old_tk = find_trace_kprobe(trace_probe_name(&tk->tp),
635 trace_probe_group_name(&tk->tp));
636 if (old_tk) {
637 if (trace_kprobe_is_return(tk) != trace_kprobe_is_return(old_tk)) {
638 trace_probe_log_set_index(0);
639 trace_probe_log_err(0, DIFF_PROBE_TYPE);
640 ret = -EEXIST;
641 } else {
642 ret = append_trace_kprobe(tk, old_tk);
644 goto end;
647 /* Register new event */
648 ret = register_kprobe_event(tk);
649 if (ret) {
650 pr_warn("Failed to register probe event(%d)\n", ret);
651 goto end;
654 /* Register k*probe */
655 ret = __register_trace_kprobe(tk);
656 if (ret == -ENOENT && !trace_kprobe_module_exist(tk)) {
657 pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
658 ret = 0;
661 if (ret < 0)
662 unregister_kprobe_event(tk);
663 else
664 dyn_event_add(&tk->devent);
666 end:
667 mutex_unlock(&event_mutex);
668 return ret;
671 /* Module notifier call back, checking event on the module */
672 static int trace_kprobe_module_callback(struct notifier_block *nb,
673 unsigned long val, void *data)
675 struct module *mod = data;
676 struct dyn_event *pos;
677 struct trace_kprobe *tk;
678 int ret;
680 if (val != MODULE_STATE_COMING)
681 return NOTIFY_DONE;
683 /* Update probes on coming module */
684 mutex_lock(&event_mutex);
685 for_each_trace_kprobe(tk, pos) {
686 if (trace_kprobe_within_module(tk, mod)) {
687 /* Don't need to check busy - this should have gone. */
688 __unregister_trace_kprobe(tk);
689 ret = __register_trace_kprobe(tk);
690 if (ret)
691 pr_warn("Failed to re-register probe %s on %s: %d\n",
692 trace_probe_name(&tk->tp),
693 module_name(mod), ret);
696 mutex_unlock(&event_mutex);
698 return NOTIFY_DONE;
701 static struct notifier_block trace_kprobe_module_nb = {
702 .notifier_call = trace_kprobe_module_callback,
703 .priority = 1 /* Invoked after kprobe module callback */
706 /* Convert certain expected symbols into '_' when generating event names */
707 static inline void sanitize_event_name(char *name)
709 while (*name++ != '\0')
710 if (*name == ':' || *name == '.')
711 *name = '_';
714 static int trace_kprobe_create(int argc, const char *argv[])
717 * Argument syntax:
718 * - Add kprobe:
719 * p[:[GRP/]EVENT] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
720 * - Add kretprobe:
721 * r[MAXACTIVE][:[GRP/]EVENT] [MOD:]KSYM[+0] [FETCHARGS]
722 * Or
723 * p:[GRP/]EVENT] [MOD:]KSYM[+0]%return [FETCHARGS]
725 * Fetch args:
726 * $retval : fetch return value
727 * $stack : fetch stack address
728 * $stackN : fetch Nth of stack (N:0-)
729 * $comm : fetch current task comm
730 * @ADDR : fetch memory at ADDR (ADDR should be in kernel)
731 * @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
732 * %REG : fetch register REG
733 * Dereferencing memory fetch:
734 * +|-offs(ARG) : fetch memory at ARG +|- offs address.
735 * Alias name of args:
736 * NAME=FETCHARG : set NAME as alias of FETCHARG.
737 * Type of args:
738 * FETCHARG:TYPE : use TYPE instead of unsigned long.
740 struct trace_kprobe *tk = NULL;
741 int i, len, ret = 0;
742 bool is_return = false;
743 char *symbol = NULL, *tmp = NULL;
744 const char *event = NULL, *group = KPROBE_EVENT_SYSTEM;
745 int maxactive = 0;
746 long offset = 0;
747 void *addr = NULL;
748 char buf[MAX_EVENT_NAME_LEN];
749 unsigned int flags = TPARG_FL_KERNEL;
751 switch (argv[0][0]) {
752 case 'r':
753 is_return = true;
754 break;
755 case 'p':
756 break;
757 default:
758 return -ECANCELED;
760 if (argc < 2)
761 return -ECANCELED;
763 trace_probe_log_init("trace_kprobe", argc, argv);
765 event = strchr(&argv[0][1], ':');
766 if (event)
767 event++;
769 if (isdigit(argv[0][1])) {
770 if (!is_return) {
771 trace_probe_log_err(1, MAXACT_NO_KPROBE);
772 goto parse_error;
774 if (event)
775 len = event - &argv[0][1] - 1;
776 else
777 len = strlen(&argv[0][1]);
778 if (len > MAX_EVENT_NAME_LEN - 1) {
779 trace_probe_log_err(1, BAD_MAXACT);
780 goto parse_error;
782 memcpy(buf, &argv[0][1], len);
783 buf[len] = '\0';
784 ret = kstrtouint(buf, 0, &maxactive);
785 if (ret || !maxactive) {
786 trace_probe_log_err(1, BAD_MAXACT);
787 goto parse_error;
789 /* kretprobes instances are iterated over via a list. The
790 * maximum should stay reasonable.
792 if (maxactive > KRETPROBE_MAXACTIVE_MAX) {
793 trace_probe_log_err(1, MAXACT_TOO_BIG);
794 goto parse_error;
798 /* try to parse an address. if that fails, try to read the
799 * input as a symbol. */
800 if (kstrtoul(argv[1], 0, (unsigned long *)&addr)) {
801 trace_probe_log_set_index(1);
802 /* Check whether uprobe event specified */
803 if (strchr(argv[1], '/') && strchr(argv[1], ':')) {
804 ret = -ECANCELED;
805 goto error;
807 /* a symbol specified */
808 symbol = kstrdup(argv[1], GFP_KERNEL);
809 if (!symbol)
810 return -ENOMEM;
812 tmp = strchr(symbol, '%');
813 if (tmp) {
814 if (!strcmp(tmp, "%return")) {
815 *tmp = '\0';
816 is_return = true;
817 } else {
818 trace_probe_log_err(tmp - symbol, BAD_ADDR_SUFFIX);
819 goto parse_error;
823 /* TODO: support .init module functions */
824 ret = traceprobe_split_symbol_offset(symbol, &offset);
825 if (ret || offset < 0 || offset > UINT_MAX) {
826 trace_probe_log_err(0, BAD_PROBE_ADDR);
827 goto parse_error;
829 if (is_return)
830 flags |= TPARG_FL_RETURN;
831 if (kprobe_on_func_entry(NULL, symbol, offset))
832 flags |= TPARG_FL_FENTRY;
833 if (offset && is_return && !(flags & TPARG_FL_FENTRY)) {
834 trace_probe_log_err(0, BAD_RETPROBE);
835 goto parse_error;
839 trace_probe_log_set_index(0);
840 if (event) {
841 ret = traceprobe_parse_event_name(&event, &group, buf,
842 event - argv[0]);
843 if (ret)
844 goto parse_error;
845 } else {
846 /* Make a new event name */
847 if (symbol)
848 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
849 is_return ? 'r' : 'p', symbol, offset);
850 else
851 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
852 is_return ? 'r' : 'p', addr);
853 sanitize_event_name(buf);
854 event = buf;
857 /* setup a probe */
858 tk = alloc_trace_kprobe(group, event, addr, symbol, offset, maxactive,
859 argc - 2, is_return);
860 if (IS_ERR(tk)) {
861 ret = PTR_ERR(tk);
862 /* This must return -ENOMEM, else there is a bug */
863 WARN_ON_ONCE(ret != -ENOMEM);
864 goto out; /* We know tk is not allocated */
866 argc -= 2; argv += 2;
868 /* parse arguments */
869 for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
870 tmp = kstrdup(argv[i], GFP_KERNEL);
871 if (!tmp) {
872 ret = -ENOMEM;
873 goto error;
876 trace_probe_log_set_index(i + 2);
877 ret = traceprobe_parse_probe_arg(&tk->tp, i, tmp, flags);
878 kfree(tmp);
879 if (ret)
880 goto error; /* This can be -ENOMEM */
883 ret = traceprobe_set_print_fmt(&tk->tp, is_return);
884 if (ret < 0)
885 goto error;
887 ret = register_trace_kprobe(tk);
888 if (ret) {
889 trace_probe_log_set_index(1);
890 if (ret == -EILSEQ)
891 trace_probe_log_err(0, BAD_INSN_BNDRY);
892 else if (ret == -ENOENT)
893 trace_probe_log_err(0, BAD_PROBE_ADDR);
894 else if (ret != -ENOMEM && ret != -EEXIST)
895 trace_probe_log_err(0, FAIL_REG_PROBE);
896 goto error;
899 out:
900 trace_probe_log_clear();
901 kfree(symbol);
902 return ret;
904 parse_error:
905 ret = -EINVAL;
906 error:
907 free_trace_kprobe(tk);
908 goto out;
911 static int create_or_delete_trace_kprobe(int argc, char **argv)
913 int ret;
915 if (argv[0][0] == '-')
916 return dyn_event_release(argc, argv, &trace_kprobe_ops);
918 ret = trace_kprobe_create(argc, (const char **)argv);
919 return ret == -ECANCELED ? -EINVAL : ret;
922 static int trace_kprobe_run_command(struct dynevent_cmd *cmd)
924 return trace_run_command(cmd->seq.buffer, create_or_delete_trace_kprobe);
928 * kprobe_event_cmd_init - Initialize a kprobe event command object
929 * @cmd: A pointer to the dynevent_cmd struct representing the new event
930 * @buf: A pointer to the buffer used to build the command
931 * @maxlen: The length of the buffer passed in @buf
933 * Initialize a synthetic event command object. Use this before
934 * calling any of the other kprobe_event functions.
936 void kprobe_event_cmd_init(struct dynevent_cmd *cmd, char *buf, int maxlen)
938 dynevent_cmd_init(cmd, buf, maxlen, DYNEVENT_TYPE_KPROBE,
939 trace_kprobe_run_command);
941 EXPORT_SYMBOL_GPL(kprobe_event_cmd_init);
944 * __kprobe_event_gen_cmd_start - Generate a kprobe event command from arg list
945 * @cmd: A pointer to the dynevent_cmd struct representing the new event
946 * @name: The name of the kprobe event
947 * @loc: The location of the kprobe event
948 * @kretprobe: Is this a return probe?
949 * @args: Variable number of arg (pairs), one pair for each field
951 * NOTE: Users normally won't want to call this function directly, but
952 * rather use the kprobe_event_gen_cmd_start() wrapper, which automatically
953 * adds a NULL to the end of the arg list. If this function is used
954 * directly, make sure the last arg in the variable arg list is NULL.
956 * Generate a kprobe event command to be executed by
957 * kprobe_event_gen_cmd_end(). This function can be used to generate the
958 * complete command or only the first part of it; in the latter case,
959 * kprobe_event_add_fields() can be used to add more fields following this.
961 * Unlikely the synth_event_gen_cmd_start(), @loc must be specified. This
962 * returns -EINVAL if @loc == NULL.
964 * Return: 0 if successful, error otherwise.
966 int __kprobe_event_gen_cmd_start(struct dynevent_cmd *cmd, bool kretprobe,
967 const char *name, const char *loc, ...)
969 char buf[MAX_EVENT_NAME_LEN];
970 struct dynevent_arg arg;
971 va_list args;
972 int ret;
974 if (cmd->type != DYNEVENT_TYPE_KPROBE)
975 return -EINVAL;
977 if (!loc)
978 return -EINVAL;
980 if (kretprobe)
981 snprintf(buf, MAX_EVENT_NAME_LEN, "r:kprobes/%s", name);
982 else
983 snprintf(buf, MAX_EVENT_NAME_LEN, "p:kprobes/%s", name);
985 ret = dynevent_str_add(cmd, buf);
986 if (ret)
987 return ret;
989 dynevent_arg_init(&arg, 0);
990 arg.str = loc;
991 ret = dynevent_arg_add(cmd, &arg, NULL);
992 if (ret)
993 return ret;
995 va_start(args, loc);
996 for (;;) {
997 const char *field;
999 field = va_arg(args, const char *);
1000 if (!field)
1001 break;
1003 if (++cmd->n_fields > MAX_TRACE_ARGS) {
1004 ret = -EINVAL;
1005 break;
1008 arg.str = field;
1009 ret = dynevent_arg_add(cmd, &arg, NULL);
1010 if (ret)
1011 break;
1013 va_end(args);
1015 return ret;
1017 EXPORT_SYMBOL_GPL(__kprobe_event_gen_cmd_start);
1020 * __kprobe_event_add_fields - Add probe fields to a kprobe command from arg list
1021 * @cmd: A pointer to the dynevent_cmd struct representing the new event
1022 * @args: Variable number of arg (pairs), one pair for each field
1024 * NOTE: Users normally won't want to call this function directly, but
1025 * rather use the kprobe_event_add_fields() wrapper, which
1026 * automatically adds a NULL to the end of the arg list. If this
1027 * function is used directly, make sure the last arg in the variable
1028 * arg list is NULL.
1030 * Add probe fields to an existing kprobe command using a variable
1031 * list of args. Fields are added in the same order they're listed.
1033 * Return: 0 if successful, error otherwise.
1035 int __kprobe_event_add_fields(struct dynevent_cmd *cmd, ...)
1037 struct dynevent_arg arg;
1038 va_list args;
1039 int ret = 0;
1041 if (cmd->type != DYNEVENT_TYPE_KPROBE)
1042 return -EINVAL;
1044 dynevent_arg_init(&arg, 0);
1046 va_start(args, cmd);
1047 for (;;) {
1048 const char *field;
1050 field = va_arg(args, const char *);
1051 if (!field)
1052 break;
1054 if (++cmd->n_fields > MAX_TRACE_ARGS) {
1055 ret = -EINVAL;
1056 break;
1059 arg.str = field;
1060 ret = dynevent_arg_add(cmd, &arg, NULL);
1061 if (ret)
1062 break;
1064 va_end(args);
1066 return ret;
1068 EXPORT_SYMBOL_GPL(__kprobe_event_add_fields);
1071 * kprobe_event_delete - Delete a kprobe event
1072 * @name: The name of the kprobe event to delete
1074 * Delete a kprobe event with the give @name from kernel code rather
1075 * than directly from the command line.
1077 * Return: 0 if successful, error otherwise.
1079 int kprobe_event_delete(const char *name)
1081 char buf[MAX_EVENT_NAME_LEN];
1083 snprintf(buf, MAX_EVENT_NAME_LEN, "-:%s", name);
1085 return trace_run_command(buf, create_or_delete_trace_kprobe);
1087 EXPORT_SYMBOL_GPL(kprobe_event_delete);
1089 static int trace_kprobe_release(struct dyn_event *ev)
1091 struct trace_kprobe *tk = to_trace_kprobe(ev);
1092 int ret = unregister_trace_kprobe(tk);
1094 if (!ret)
1095 free_trace_kprobe(tk);
1096 return ret;
1099 static int trace_kprobe_show(struct seq_file *m, struct dyn_event *ev)
1101 struct trace_kprobe *tk = to_trace_kprobe(ev);
1102 int i;
1104 seq_putc(m, trace_kprobe_is_return(tk) ? 'r' : 'p');
1105 if (trace_kprobe_is_return(tk) && tk->rp.maxactive)
1106 seq_printf(m, "%d", tk->rp.maxactive);
1107 seq_printf(m, ":%s/%s", trace_probe_group_name(&tk->tp),
1108 trace_probe_name(&tk->tp));
1110 if (!tk->symbol)
1111 seq_printf(m, " 0x%p", tk->rp.kp.addr);
1112 else if (tk->rp.kp.offset)
1113 seq_printf(m, " %s+%u", trace_kprobe_symbol(tk),
1114 tk->rp.kp.offset);
1115 else
1116 seq_printf(m, " %s", trace_kprobe_symbol(tk));
1118 for (i = 0; i < tk->tp.nr_args; i++)
1119 seq_printf(m, " %s=%s", tk->tp.args[i].name, tk->tp.args[i].comm);
1120 seq_putc(m, '\n');
1122 return 0;
1125 static int probes_seq_show(struct seq_file *m, void *v)
1127 struct dyn_event *ev = v;
1129 if (!is_trace_kprobe(ev))
1130 return 0;
1132 return trace_kprobe_show(m, ev);
1135 static const struct seq_operations probes_seq_op = {
1136 .start = dyn_event_seq_start,
1137 .next = dyn_event_seq_next,
1138 .stop = dyn_event_seq_stop,
1139 .show = probes_seq_show
1142 static int probes_open(struct inode *inode, struct file *file)
1144 int ret;
1146 ret = security_locked_down(LOCKDOWN_TRACEFS);
1147 if (ret)
1148 return ret;
1150 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
1151 ret = dyn_events_release_all(&trace_kprobe_ops);
1152 if (ret < 0)
1153 return ret;
1156 return seq_open(file, &probes_seq_op);
1159 static ssize_t probes_write(struct file *file, const char __user *buffer,
1160 size_t count, loff_t *ppos)
1162 return trace_parse_run_command(file, buffer, count, ppos,
1163 create_or_delete_trace_kprobe);
1166 static const struct file_operations kprobe_events_ops = {
1167 .owner = THIS_MODULE,
1168 .open = probes_open,
1169 .read = seq_read,
1170 .llseek = seq_lseek,
1171 .release = seq_release,
1172 .write = probes_write,
1175 /* Probes profiling interfaces */
1176 static int probes_profile_seq_show(struct seq_file *m, void *v)
1178 struct dyn_event *ev = v;
1179 struct trace_kprobe *tk;
1181 if (!is_trace_kprobe(ev))
1182 return 0;
1184 tk = to_trace_kprobe(ev);
1185 seq_printf(m, " %-44s %15lu %15lu\n",
1186 trace_probe_name(&tk->tp),
1187 trace_kprobe_nhit(tk),
1188 tk->rp.kp.nmissed);
1190 return 0;
1193 static const struct seq_operations profile_seq_op = {
1194 .start = dyn_event_seq_start,
1195 .next = dyn_event_seq_next,
1196 .stop = dyn_event_seq_stop,
1197 .show = probes_profile_seq_show
1200 static int profile_open(struct inode *inode, struct file *file)
1202 int ret;
1204 ret = security_locked_down(LOCKDOWN_TRACEFS);
1205 if (ret)
1206 return ret;
1208 return seq_open(file, &profile_seq_op);
1211 static const struct file_operations kprobe_profile_ops = {
1212 .owner = THIS_MODULE,
1213 .open = profile_open,
1214 .read = seq_read,
1215 .llseek = seq_lseek,
1216 .release = seq_release,
1219 /* Kprobe specific fetch functions */
1221 /* Return the length of string -- including null terminal byte */
1222 static nokprobe_inline int
1223 fetch_store_strlen_user(unsigned long addr)
1225 const void __user *uaddr = (__force const void __user *)addr;
1227 return strnlen_user_nofault(uaddr, MAX_STRING_SIZE);
1230 /* Return the length of string -- including null terminal byte */
1231 static nokprobe_inline int
1232 fetch_store_strlen(unsigned long addr)
1234 int ret, len = 0;
1235 u8 c;
1237 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
1238 if (addr < TASK_SIZE)
1239 return fetch_store_strlen_user(addr);
1240 #endif
1242 do {
1243 ret = copy_from_kernel_nofault(&c, (u8 *)addr + len, 1);
1244 len++;
1245 } while (c && ret == 0 && len < MAX_STRING_SIZE);
1247 return (ret < 0) ? ret : len;
1251 * Fetch a null-terminated string from user. Caller MUST set *(u32 *)buf
1252 * with max length and relative data location.
1254 static nokprobe_inline int
1255 fetch_store_string_user(unsigned long addr, void *dest, void *base)
1257 const void __user *uaddr = (__force const void __user *)addr;
1258 int maxlen = get_loc_len(*(u32 *)dest);
1259 void *__dest;
1260 long ret;
1262 if (unlikely(!maxlen))
1263 return -ENOMEM;
1265 __dest = get_loc_data(dest, base);
1267 ret = strncpy_from_user_nofault(__dest, uaddr, maxlen);
1268 if (ret >= 0)
1269 *(u32 *)dest = make_data_loc(ret, __dest - base);
1271 return ret;
1275 * Fetch a null-terminated string. Caller MUST set *(u32 *)buf with max
1276 * length and relative data location.
1278 static nokprobe_inline int
1279 fetch_store_string(unsigned long addr, void *dest, void *base)
1281 int maxlen = get_loc_len(*(u32 *)dest);
1282 void *__dest;
1283 long ret;
1285 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
1286 if ((unsigned long)addr < TASK_SIZE)
1287 return fetch_store_string_user(addr, dest, base);
1288 #endif
1290 if (unlikely(!maxlen))
1291 return -ENOMEM;
1293 __dest = get_loc_data(dest, base);
1296 * Try to get string again, since the string can be changed while
1297 * probing.
1299 ret = strncpy_from_kernel_nofault(__dest, (void *)addr, maxlen);
1300 if (ret >= 0)
1301 *(u32 *)dest = make_data_loc(ret, __dest - base);
1303 return ret;
1306 static nokprobe_inline int
1307 probe_mem_read_user(void *dest, void *src, size_t size)
1309 const void __user *uaddr = (__force const void __user *)src;
1311 return copy_from_user_nofault(dest, uaddr, size);
1314 static nokprobe_inline int
1315 probe_mem_read(void *dest, void *src, size_t size)
1317 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
1318 if ((unsigned long)src < TASK_SIZE)
1319 return probe_mem_read_user(dest, src, size);
1320 #endif
1321 return copy_from_kernel_nofault(dest, src, size);
1324 /* Note that we don't verify it, since the code does not come from user space */
1325 static int
1326 process_fetch_insn(struct fetch_insn *code, struct pt_regs *regs, void *dest,
1327 void *base)
1329 unsigned long val;
1331 retry:
1332 /* 1st stage: get value from context */
1333 switch (code->op) {
1334 case FETCH_OP_REG:
1335 val = regs_get_register(regs, code->param);
1336 break;
1337 case FETCH_OP_STACK:
1338 val = regs_get_kernel_stack_nth(regs, code->param);
1339 break;
1340 case FETCH_OP_STACKP:
1341 val = kernel_stack_pointer(regs);
1342 break;
1343 case FETCH_OP_RETVAL:
1344 val = regs_return_value(regs);
1345 break;
1346 case FETCH_OP_IMM:
1347 val = code->immediate;
1348 break;
1349 case FETCH_OP_COMM:
1350 val = (unsigned long)current->comm;
1351 break;
1352 case FETCH_OP_DATA:
1353 val = (unsigned long)code->data;
1354 break;
1355 #ifdef CONFIG_HAVE_FUNCTION_ARG_ACCESS_API
1356 case FETCH_OP_ARG:
1357 val = regs_get_kernel_argument(regs, code->param);
1358 break;
1359 #endif
1360 case FETCH_NOP_SYMBOL: /* Ignore a place holder */
1361 code++;
1362 goto retry;
1363 default:
1364 return -EILSEQ;
1366 code++;
1368 return process_fetch_insn_bottom(code, val, dest, base);
1370 NOKPROBE_SYMBOL(process_fetch_insn)
1372 /* Kprobe handler */
1373 static nokprobe_inline void
1374 __kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs,
1375 struct trace_event_file *trace_file)
1377 struct kprobe_trace_entry_head *entry;
1378 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1379 struct trace_event_buffer fbuffer;
1380 int dsize;
1382 WARN_ON(call != trace_file->event_call);
1384 if (trace_trigger_soft_disabled(trace_file))
1385 return;
1387 local_save_flags(fbuffer.flags);
1388 fbuffer.pc = preempt_count();
1389 fbuffer.trace_file = trace_file;
1391 dsize = __get_data_size(&tk->tp, regs);
1393 fbuffer.event =
1394 trace_event_buffer_lock_reserve(&fbuffer.buffer, trace_file,
1395 call->event.type,
1396 sizeof(*entry) + tk->tp.size + dsize,
1397 fbuffer.flags, fbuffer.pc);
1398 if (!fbuffer.event)
1399 return;
1401 fbuffer.regs = regs;
1402 entry = fbuffer.entry = ring_buffer_event_data(fbuffer.event);
1403 entry->ip = (unsigned long)tk->rp.kp.addr;
1404 store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1406 trace_event_buffer_commit(&fbuffer);
1409 static void
1410 kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs)
1412 struct event_file_link *link;
1414 trace_probe_for_each_link_rcu(link, &tk->tp)
1415 __kprobe_trace_func(tk, regs, link->file);
1417 NOKPROBE_SYMBOL(kprobe_trace_func);
1419 /* Kretprobe handler */
1420 static nokprobe_inline void
1421 __kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1422 struct pt_regs *regs,
1423 struct trace_event_file *trace_file)
1425 struct kretprobe_trace_entry_head *entry;
1426 struct trace_event_buffer fbuffer;
1427 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1428 int dsize;
1430 WARN_ON(call != trace_file->event_call);
1432 if (trace_trigger_soft_disabled(trace_file))
1433 return;
1435 local_save_flags(fbuffer.flags);
1436 fbuffer.pc = preempt_count();
1437 fbuffer.trace_file = trace_file;
1439 dsize = __get_data_size(&tk->tp, regs);
1440 fbuffer.event =
1441 trace_event_buffer_lock_reserve(&fbuffer.buffer, trace_file,
1442 call->event.type,
1443 sizeof(*entry) + tk->tp.size + dsize,
1444 fbuffer.flags, fbuffer.pc);
1445 if (!fbuffer.event)
1446 return;
1448 fbuffer.regs = regs;
1449 entry = fbuffer.entry = ring_buffer_event_data(fbuffer.event);
1450 entry->func = (unsigned long)tk->rp.kp.addr;
1451 entry->ret_ip = (unsigned long)ri->ret_addr;
1452 store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1454 trace_event_buffer_commit(&fbuffer);
1457 static void
1458 kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1459 struct pt_regs *regs)
1461 struct event_file_link *link;
1463 trace_probe_for_each_link_rcu(link, &tk->tp)
1464 __kretprobe_trace_func(tk, ri, regs, link->file);
1466 NOKPROBE_SYMBOL(kretprobe_trace_func);
1468 /* Event entry printers */
1469 static enum print_line_t
1470 print_kprobe_event(struct trace_iterator *iter, int flags,
1471 struct trace_event *event)
1473 struct kprobe_trace_entry_head *field;
1474 struct trace_seq *s = &iter->seq;
1475 struct trace_probe *tp;
1477 field = (struct kprobe_trace_entry_head *)iter->ent;
1478 tp = trace_probe_primary_from_call(
1479 container_of(event, struct trace_event_call, event));
1480 if (WARN_ON_ONCE(!tp))
1481 goto out;
1483 trace_seq_printf(s, "%s: (", trace_probe_name(tp));
1485 if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
1486 goto out;
1488 trace_seq_putc(s, ')');
1490 if (print_probe_args(s, tp->args, tp->nr_args,
1491 (u8 *)&field[1], field) < 0)
1492 goto out;
1494 trace_seq_putc(s, '\n');
1495 out:
1496 return trace_handle_return(s);
1499 static enum print_line_t
1500 print_kretprobe_event(struct trace_iterator *iter, int flags,
1501 struct trace_event *event)
1503 struct kretprobe_trace_entry_head *field;
1504 struct trace_seq *s = &iter->seq;
1505 struct trace_probe *tp;
1507 field = (struct kretprobe_trace_entry_head *)iter->ent;
1508 tp = trace_probe_primary_from_call(
1509 container_of(event, struct trace_event_call, event));
1510 if (WARN_ON_ONCE(!tp))
1511 goto out;
1513 trace_seq_printf(s, "%s: (", trace_probe_name(tp));
1515 if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1516 goto out;
1518 trace_seq_puts(s, " <- ");
1520 if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1521 goto out;
1523 trace_seq_putc(s, ')');
1525 if (print_probe_args(s, tp->args, tp->nr_args,
1526 (u8 *)&field[1], field) < 0)
1527 goto out;
1529 trace_seq_putc(s, '\n');
1531 out:
1532 return trace_handle_return(s);
1536 static int kprobe_event_define_fields(struct trace_event_call *event_call)
1538 int ret;
1539 struct kprobe_trace_entry_head field;
1540 struct trace_probe *tp;
1542 tp = trace_probe_primary_from_call(event_call);
1543 if (WARN_ON_ONCE(!tp))
1544 return -ENOENT;
1546 DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1548 return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
1551 static int kretprobe_event_define_fields(struct trace_event_call *event_call)
1553 int ret;
1554 struct kretprobe_trace_entry_head field;
1555 struct trace_probe *tp;
1557 tp = trace_probe_primary_from_call(event_call);
1558 if (WARN_ON_ONCE(!tp))
1559 return -ENOENT;
1561 DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1562 DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1564 return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
1567 #ifdef CONFIG_PERF_EVENTS
1569 /* Kprobe profile handler */
1570 static int
1571 kprobe_perf_func(struct trace_kprobe *tk, struct pt_regs *regs)
1573 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1574 struct kprobe_trace_entry_head *entry;
1575 struct hlist_head *head;
1576 int size, __size, dsize;
1577 int rctx;
1579 if (bpf_prog_array_valid(call)) {
1580 unsigned long orig_ip = instruction_pointer(regs);
1581 int ret;
1583 ret = trace_call_bpf(call, regs);
1586 * We need to check and see if we modified the pc of the
1587 * pt_regs, and if so return 1 so that we don't do the
1588 * single stepping.
1590 if (orig_ip != instruction_pointer(regs))
1591 return 1;
1592 if (!ret)
1593 return 0;
1596 head = this_cpu_ptr(call->perf_events);
1597 if (hlist_empty(head))
1598 return 0;
1600 dsize = __get_data_size(&tk->tp, regs);
1601 __size = sizeof(*entry) + tk->tp.size + dsize;
1602 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1603 size -= sizeof(u32);
1605 entry = perf_trace_buf_alloc(size, NULL, &rctx);
1606 if (!entry)
1607 return 0;
1609 entry->ip = (unsigned long)tk->rp.kp.addr;
1610 memset(&entry[1], 0, dsize);
1611 store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1612 perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1613 head, NULL);
1614 return 0;
1616 NOKPROBE_SYMBOL(kprobe_perf_func);
1618 /* Kretprobe profile handler */
1619 static void
1620 kretprobe_perf_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1621 struct pt_regs *regs)
1623 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1624 struct kretprobe_trace_entry_head *entry;
1625 struct hlist_head *head;
1626 int size, __size, dsize;
1627 int rctx;
1629 if (bpf_prog_array_valid(call) && !trace_call_bpf(call, regs))
1630 return;
1632 head = this_cpu_ptr(call->perf_events);
1633 if (hlist_empty(head))
1634 return;
1636 dsize = __get_data_size(&tk->tp, regs);
1637 __size = sizeof(*entry) + tk->tp.size + dsize;
1638 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1639 size -= sizeof(u32);
1641 entry = perf_trace_buf_alloc(size, NULL, &rctx);
1642 if (!entry)
1643 return;
1645 entry->func = (unsigned long)tk->rp.kp.addr;
1646 entry->ret_ip = (unsigned long)ri->ret_addr;
1647 store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1648 perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1649 head, NULL);
1651 NOKPROBE_SYMBOL(kretprobe_perf_func);
1653 int bpf_get_kprobe_info(const struct perf_event *event, u32 *fd_type,
1654 const char **symbol, u64 *probe_offset,
1655 u64 *probe_addr, bool perf_type_tracepoint)
1657 const char *pevent = trace_event_name(event->tp_event);
1658 const char *group = event->tp_event->class->system;
1659 struct trace_kprobe *tk;
1661 if (perf_type_tracepoint)
1662 tk = find_trace_kprobe(pevent, group);
1663 else
1664 tk = trace_kprobe_primary_from_call(event->tp_event);
1665 if (!tk)
1666 return -EINVAL;
1668 *fd_type = trace_kprobe_is_return(tk) ? BPF_FD_TYPE_KRETPROBE
1669 : BPF_FD_TYPE_KPROBE;
1670 if (tk->symbol) {
1671 *symbol = tk->symbol;
1672 *probe_offset = tk->rp.kp.offset;
1673 *probe_addr = 0;
1674 } else {
1675 *symbol = NULL;
1676 *probe_offset = 0;
1677 *probe_addr = (unsigned long)tk->rp.kp.addr;
1679 return 0;
1681 #endif /* CONFIG_PERF_EVENTS */
1684 * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex.
1686 * kprobe_trace_self_tests_init() does enable_trace_probe/disable_trace_probe
1687 * lockless, but we can't race with this __init function.
1689 static int kprobe_register(struct trace_event_call *event,
1690 enum trace_reg type, void *data)
1692 struct trace_event_file *file = data;
1694 switch (type) {
1695 case TRACE_REG_REGISTER:
1696 return enable_trace_kprobe(event, file);
1697 case TRACE_REG_UNREGISTER:
1698 return disable_trace_kprobe(event, file);
1700 #ifdef CONFIG_PERF_EVENTS
1701 case TRACE_REG_PERF_REGISTER:
1702 return enable_trace_kprobe(event, NULL);
1703 case TRACE_REG_PERF_UNREGISTER:
1704 return disable_trace_kprobe(event, NULL);
1705 case TRACE_REG_PERF_OPEN:
1706 case TRACE_REG_PERF_CLOSE:
1707 case TRACE_REG_PERF_ADD:
1708 case TRACE_REG_PERF_DEL:
1709 return 0;
1710 #endif
1712 return 0;
1715 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1717 struct trace_kprobe *tk = container_of(kp, struct trace_kprobe, rp.kp);
1718 int ret = 0;
1720 raw_cpu_inc(*tk->nhit);
1722 if (trace_probe_test_flag(&tk->tp, TP_FLAG_TRACE))
1723 kprobe_trace_func(tk, regs);
1724 #ifdef CONFIG_PERF_EVENTS
1725 if (trace_probe_test_flag(&tk->tp, TP_FLAG_PROFILE))
1726 ret = kprobe_perf_func(tk, regs);
1727 #endif
1728 return ret;
1730 NOKPROBE_SYMBOL(kprobe_dispatcher);
1732 static int
1733 kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1735 struct kretprobe *rp = get_kretprobe(ri);
1736 struct trace_kprobe *tk = container_of(rp, struct trace_kprobe, rp);
1738 raw_cpu_inc(*tk->nhit);
1740 if (trace_probe_test_flag(&tk->tp, TP_FLAG_TRACE))
1741 kretprobe_trace_func(tk, ri, regs);
1742 #ifdef CONFIG_PERF_EVENTS
1743 if (trace_probe_test_flag(&tk->tp, TP_FLAG_PROFILE))
1744 kretprobe_perf_func(tk, ri, regs);
1745 #endif
1746 return 0; /* We don't tweek kernel, so just return 0 */
1748 NOKPROBE_SYMBOL(kretprobe_dispatcher);
1750 static struct trace_event_functions kretprobe_funcs = {
1751 .trace = print_kretprobe_event
1754 static struct trace_event_functions kprobe_funcs = {
1755 .trace = print_kprobe_event
1758 static struct trace_event_fields kretprobe_fields_array[] = {
1759 { .type = TRACE_FUNCTION_TYPE,
1760 .define_fields = kretprobe_event_define_fields },
1764 static struct trace_event_fields kprobe_fields_array[] = {
1765 { .type = TRACE_FUNCTION_TYPE,
1766 .define_fields = kprobe_event_define_fields },
1770 static inline void init_trace_event_call(struct trace_kprobe *tk)
1772 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1774 if (trace_kprobe_is_return(tk)) {
1775 call->event.funcs = &kretprobe_funcs;
1776 call->class->fields_array = kretprobe_fields_array;
1777 } else {
1778 call->event.funcs = &kprobe_funcs;
1779 call->class->fields_array = kprobe_fields_array;
1782 call->flags = TRACE_EVENT_FL_KPROBE;
1783 call->class->reg = kprobe_register;
1786 static int register_kprobe_event(struct trace_kprobe *tk)
1788 init_trace_event_call(tk);
1790 return trace_probe_register_event_call(&tk->tp);
1793 static int unregister_kprobe_event(struct trace_kprobe *tk)
1795 return trace_probe_unregister_event_call(&tk->tp);
1798 #ifdef CONFIG_PERF_EVENTS
1799 /* create a trace_kprobe, but don't add it to global lists */
1800 struct trace_event_call *
1801 create_local_trace_kprobe(char *func, void *addr, unsigned long offs,
1802 bool is_return)
1804 struct trace_kprobe *tk;
1805 int ret;
1806 char *event;
1809 * local trace_kprobes are not added to dyn_event, so they are never
1810 * searched in find_trace_kprobe(). Therefore, there is no concern of
1811 * duplicated name here.
1813 event = func ? func : "DUMMY_EVENT";
1815 tk = alloc_trace_kprobe(KPROBE_EVENT_SYSTEM, event, (void *)addr, func,
1816 offs, 0 /* maxactive */, 0 /* nargs */,
1817 is_return);
1819 if (IS_ERR(tk)) {
1820 pr_info("Failed to allocate trace_probe.(%d)\n",
1821 (int)PTR_ERR(tk));
1822 return ERR_CAST(tk);
1825 init_trace_event_call(tk);
1827 if (traceprobe_set_print_fmt(&tk->tp, trace_kprobe_is_return(tk)) < 0) {
1828 ret = -ENOMEM;
1829 goto error;
1832 ret = __register_trace_kprobe(tk);
1833 if (ret < 0)
1834 goto error;
1836 return trace_probe_event_call(&tk->tp);
1837 error:
1838 free_trace_kprobe(tk);
1839 return ERR_PTR(ret);
1842 void destroy_local_trace_kprobe(struct trace_event_call *event_call)
1844 struct trace_kprobe *tk;
1846 tk = trace_kprobe_primary_from_call(event_call);
1847 if (unlikely(!tk))
1848 return;
1850 if (trace_probe_is_enabled(&tk->tp)) {
1851 WARN_ON(1);
1852 return;
1855 __unregister_trace_kprobe(tk);
1857 free_trace_kprobe(tk);
1859 #endif /* CONFIG_PERF_EVENTS */
1861 static __init void enable_boot_kprobe_events(void)
1863 struct trace_array *tr = top_trace_array();
1864 struct trace_event_file *file;
1865 struct trace_kprobe *tk;
1866 struct dyn_event *pos;
1868 mutex_lock(&event_mutex);
1869 for_each_trace_kprobe(tk, pos) {
1870 list_for_each_entry(file, &tr->events, list)
1871 if (file->event_call == trace_probe_event_call(&tk->tp))
1872 trace_event_enable_disable(file, 1, 0);
1874 mutex_unlock(&event_mutex);
1877 static __init void setup_boot_kprobe_events(void)
1879 char *p, *cmd = kprobe_boot_events_buf;
1880 int ret;
1882 strreplace(kprobe_boot_events_buf, ',', ' ');
1884 while (cmd && *cmd != '\0') {
1885 p = strchr(cmd, ';');
1886 if (p)
1887 *p++ = '\0';
1889 ret = trace_run_command(cmd, create_or_delete_trace_kprobe);
1890 if (ret)
1891 pr_warn("Failed to add event(%d): %s\n", ret, cmd);
1893 cmd = p;
1896 enable_boot_kprobe_events();
1900 * Register dynevent at core_initcall. This allows kernel to setup kprobe
1901 * events in postcore_initcall without tracefs.
1903 static __init int init_kprobe_trace_early(void)
1905 int ret;
1907 ret = dyn_event_register(&trace_kprobe_ops);
1908 if (ret)
1909 return ret;
1911 if (register_module_notifier(&trace_kprobe_module_nb))
1912 return -EINVAL;
1914 return 0;
1916 core_initcall(init_kprobe_trace_early);
1918 /* Make a tracefs interface for controlling probe points */
1919 static __init int init_kprobe_trace(void)
1921 int ret;
1922 struct dentry *entry;
1924 ret = tracing_init_dentry();
1925 if (ret)
1926 return 0;
1928 entry = tracefs_create_file("kprobe_events", 0644, NULL,
1929 NULL, &kprobe_events_ops);
1931 /* Event list interface */
1932 if (!entry)
1933 pr_warn("Could not create tracefs 'kprobe_events' entry\n");
1935 /* Profile interface */
1936 entry = tracefs_create_file("kprobe_profile", 0444, NULL,
1937 NULL, &kprobe_profile_ops);
1939 if (!entry)
1940 pr_warn("Could not create tracefs 'kprobe_profile' entry\n");
1942 setup_boot_kprobe_events();
1944 return 0;
1946 fs_initcall(init_kprobe_trace);
1949 #ifdef CONFIG_FTRACE_STARTUP_TEST
1950 static __init struct trace_event_file *
1951 find_trace_probe_file(struct trace_kprobe *tk, struct trace_array *tr)
1953 struct trace_event_file *file;
1955 list_for_each_entry(file, &tr->events, list)
1956 if (file->event_call == trace_probe_event_call(&tk->tp))
1957 return file;
1959 return NULL;
1963 * Nobody but us can call enable_trace_kprobe/disable_trace_kprobe at this
1964 * stage, we can do this lockless.
1966 static __init int kprobe_trace_self_tests_init(void)
1968 int ret, warn = 0;
1969 int (*target)(int, int, int, int, int, int);
1970 struct trace_kprobe *tk;
1971 struct trace_event_file *file;
1973 if (tracing_is_disabled())
1974 return -ENODEV;
1976 if (tracing_selftest_disabled)
1977 return 0;
1979 target = kprobe_trace_selftest_target;
1981 pr_info("Testing kprobe tracing: ");
1983 ret = trace_run_command("p:testprobe kprobe_trace_selftest_target $stack $stack0 +0($stack)",
1984 create_or_delete_trace_kprobe);
1985 if (WARN_ON_ONCE(ret)) {
1986 pr_warn("error on probing function entry.\n");
1987 warn++;
1988 } else {
1989 /* Enable trace point */
1990 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1991 if (WARN_ON_ONCE(tk == NULL)) {
1992 pr_warn("error on getting new probe.\n");
1993 warn++;
1994 } else {
1995 file = find_trace_probe_file(tk, top_trace_array());
1996 if (WARN_ON_ONCE(file == NULL)) {
1997 pr_warn("error on getting probe file.\n");
1998 warn++;
1999 } else
2000 enable_trace_kprobe(
2001 trace_probe_event_call(&tk->tp), file);
2005 ret = trace_run_command("r:testprobe2 kprobe_trace_selftest_target $retval",
2006 create_or_delete_trace_kprobe);
2007 if (WARN_ON_ONCE(ret)) {
2008 pr_warn("error on probing function return.\n");
2009 warn++;
2010 } else {
2011 /* Enable trace point */
2012 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
2013 if (WARN_ON_ONCE(tk == NULL)) {
2014 pr_warn("error on getting 2nd new probe.\n");
2015 warn++;
2016 } else {
2017 file = find_trace_probe_file(tk, top_trace_array());
2018 if (WARN_ON_ONCE(file == NULL)) {
2019 pr_warn("error on getting probe file.\n");
2020 warn++;
2021 } else
2022 enable_trace_kprobe(
2023 trace_probe_event_call(&tk->tp), file);
2027 if (warn)
2028 goto end;
2030 ret = target(1, 2, 3, 4, 5, 6);
2033 * Not expecting an error here, the check is only to prevent the
2034 * optimizer from removing the call to target() as otherwise there
2035 * are no side-effects and the call is never performed.
2037 if (ret != 21)
2038 warn++;
2040 /* Disable trace points before removing it */
2041 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
2042 if (WARN_ON_ONCE(tk == NULL)) {
2043 pr_warn("error on getting test probe.\n");
2044 warn++;
2045 } else {
2046 if (trace_kprobe_nhit(tk) != 1) {
2047 pr_warn("incorrect number of testprobe hits\n");
2048 warn++;
2051 file = find_trace_probe_file(tk, top_trace_array());
2052 if (WARN_ON_ONCE(file == NULL)) {
2053 pr_warn("error on getting probe file.\n");
2054 warn++;
2055 } else
2056 disable_trace_kprobe(
2057 trace_probe_event_call(&tk->tp), file);
2060 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
2061 if (WARN_ON_ONCE(tk == NULL)) {
2062 pr_warn("error on getting 2nd test probe.\n");
2063 warn++;
2064 } else {
2065 if (trace_kprobe_nhit(tk) != 1) {
2066 pr_warn("incorrect number of testprobe2 hits\n");
2067 warn++;
2070 file = find_trace_probe_file(tk, top_trace_array());
2071 if (WARN_ON_ONCE(file == NULL)) {
2072 pr_warn("error on getting probe file.\n");
2073 warn++;
2074 } else
2075 disable_trace_kprobe(
2076 trace_probe_event_call(&tk->tp), file);
2079 ret = trace_run_command("-:testprobe", create_or_delete_trace_kprobe);
2080 if (WARN_ON_ONCE(ret)) {
2081 pr_warn("error on deleting a probe.\n");
2082 warn++;
2085 ret = trace_run_command("-:testprobe2", create_or_delete_trace_kprobe);
2086 if (WARN_ON_ONCE(ret)) {
2087 pr_warn("error on deleting a probe.\n");
2088 warn++;
2091 end:
2092 ret = dyn_events_release_all(&trace_kprobe_ops);
2093 if (WARN_ON_ONCE(ret)) {
2094 pr_warn("error on cleaning up probes.\n");
2095 warn++;
2098 * Wait for the optimizer work to finish. Otherwise it might fiddle
2099 * with probes in already freed __init text.
2101 wait_for_kprobe_optimizer();
2102 if (warn)
2103 pr_cont("NG: Some tests are failed. Please check them.\n");
2104 else
2105 pr_cont("OK\n");
2106 return 0;
2109 late_initcall(kprobe_trace_self_tests_init);
2111 #endif