Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Kprobes-based tracing events
4 *
5 * Created by Masami Hiramatsu <mhiramat@redhat.com>
6 *
7 */
8#define pr_fmt(fmt) "trace_kprobe: " fmt
9
10#include <linux/bpf-cgroup.h>
11#include <linux/security.h>
12#include <linux/module.h>
13#include <linux/uaccess.h>
14#include <linux/rculist.h>
15#include <linux/error-injection.h>
16
17#include <asm/setup.h> /* for COMMAND_LINE_SIZE */
18
19#include "trace_dynevent.h"
20#include "trace_kprobe_selftest.h"
21#include "trace_probe.h"
22#include "trace_probe_tmpl.h"
23#include "trace_probe_kernel.h"
24
25#define KPROBE_EVENT_SYSTEM "kprobes"
26#define KRETPROBE_MAXACTIVE_MAX 4096
27
28/* Kprobe early definition from command line */
29static char kprobe_boot_events_buf[COMMAND_LINE_SIZE] __initdata;
30
31static int __init set_kprobe_boot_events(char *str)
32{
33 strscpy(kprobe_boot_events_buf, str, COMMAND_LINE_SIZE);
34 disable_tracing_selftest("running kprobe events");
35
36 return 1;
37}
38__setup("kprobe_event=", set_kprobe_boot_events);
39
40static int trace_kprobe_create(const char *raw_command);
41static int trace_kprobe_show(struct seq_file *m, struct dyn_event *ev);
42static int trace_kprobe_release(struct dyn_event *ev);
43static bool trace_kprobe_is_busy(struct dyn_event *ev);
44static bool trace_kprobe_match(const char *system, const char *event,
45 int argc, const char **argv, struct dyn_event *ev);
46
47static struct dyn_event_operations trace_kprobe_ops = {
48 .create = trace_kprobe_create,
49 .show = trace_kprobe_show,
50 .is_busy = trace_kprobe_is_busy,
51 .free = trace_kprobe_release,
52 .match = trace_kprobe_match,
53};
54
55/*
56 * Kprobe event core functions
57 */
58struct trace_kprobe {
59 struct dyn_event devent;
60 struct kretprobe rp; /* Use rp.kp for kprobe use */
61 unsigned long __percpu *nhit;
62 const char *symbol; /* symbol name */
63 struct trace_probe tp;
64};
65
66static bool is_trace_kprobe(struct dyn_event *ev)
67{
68 return ev->ops == &trace_kprobe_ops;
69}
70
71static struct trace_kprobe *to_trace_kprobe(struct dyn_event *ev)
72{
73 return container_of(ev, struct trace_kprobe, devent);
74}
75
76/**
77 * for_each_trace_kprobe - iterate over the trace_kprobe list
78 * @pos: the struct trace_kprobe * for each entry
79 * @dpos: the struct dyn_event * to use as a loop cursor
80 */
81#define for_each_trace_kprobe(pos, dpos) \
82 for_each_dyn_event(dpos) \
83 if (is_trace_kprobe(dpos) && (pos = to_trace_kprobe(dpos)))
84
85static nokprobe_inline bool trace_kprobe_is_return(struct trace_kprobe *tk)
86{
87 return tk->rp.handler != NULL;
88}
89
90static nokprobe_inline const char *trace_kprobe_symbol(struct trace_kprobe *tk)
91{
92 return tk->symbol ? tk->symbol : "unknown";
93}
94
95static nokprobe_inline unsigned long trace_kprobe_offset(struct trace_kprobe *tk)
96{
97 return tk->rp.kp.offset;
98}
99
100static nokprobe_inline bool trace_kprobe_has_gone(struct trace_kprobe *tk)
101{
102 return kprobe_gone(&tk->rp.kp);
103}
104
105static nokprobe_inline bool trace_kprobe_within_module(struct trace_kprobe *tk,
106 struct module *mod)
107{
108 int len = strlen(module_name(mod));
109 const char *name = trace_kprobe_symbol(tk);
110
111 return strncmp(module_name(mod), name, len) == 0 && name[len] == ':';
112}
113
114static nokprobe_inline bool trace_kprobe_module_exist(struct trace_kprobe *tk)
115{
116 char *p;
117 bool ret;
118
119 if (!tk->symbol)
120 return false;
121 p = strchr(tk->symbol, ':');
122 if (!p)
123 return true;
124 *p = '\0';
125 rcu_read_lock_sched();
126 ret = !!find_module(tk->symbol);
127 rcu_read_unlock_sched();
128 *p = ':';
129
130 return ret;
131}
132
133static bool trace_kprobe_is_busy(struct dyn_event *ev)
134{
135 struct trace_kprobe *tk = to_trace_kprobe(ev);
136
137 return trace_probe_is_enabled(&tk->tp);
138}
139
140static bool trace_kprobe_match_command_head(struct trace_kprobe *tk,
141 int argc, const char **argv)
142{
143 char buf[MAX_ARGSTR_LEN + 1];
144
145 if (!argc)
146 return true;
147
148 if (!tk->symbol)
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);
153 else
154 snprintf(buf, sizeof(buf), "%s", trace_kprobe_symbol(tk));
155 if (strcmp(buf, argv[0]))
156 return false;
157 argc--; argv++;
158
159 return trace_probe_match_command_args(&tk->tp, argc, argv);
160}
161
162static bool trace_kprobe_match(const char *system, const char *event,
163 int argc, const char **argv, struct dyn_event *ev)
164{
165 struct trace_kprobe *tk = to_trace_kprobe(ev);
166
167 return (event[0] == '\0' ||
168 strcmp(trace_probe_name(&tk->tp), event) == 0) &&
169 (!system || strcmp(trace_probe_group_name(&tk->tp), system) == 0) &&
170 trace_kprobe_match_command_head(tk, argc, argv);
171}
172
173static nokprobe_inline unsigned long trace_kprobe_nhit(struct trace_kprobe *tk)
174{
175 unsigned long nhit = 0;
176 int cpu;
177
178 for_each_possible_cpu(cpu)
179 nhit += *per_cpu_ptr(tk->nhit, cpu);
180
181 return nhit;
182}
183
184static nokprobe_inline bool trace_kprobe_is_registered(struct trace_kprobe *tk)
185{
186 return !(list_empty(&tk->rp.kp.list) &&
187 hlist_unhashed(&tk->rp.kp.hlist));
188}
189
190/* Return 0 if it fails to find the symbol address */
191static nokprobe_inline
192unsigned long trace_kprobe_address(struct trace_kprobe *tk)
193{
194 unsigned long addr;
195
196 if (tk->symbol) {
197 addr = (unsigned long)
198 kallsyms_lookup_name(trace_kprobe_symbol(tk));
199 if (addr)
200 addr += tk->rp.kp.offset;
201 } else {
202 addr = (unsigned long)tk->rp.kp.addr;
203 }
204 return addr;
205}
206
207static nokprobe_inline struct trace_kprobe *
208trace_kprobe_primary_from_call(struct trace_event_call *call)
209{
210 struct trace_probe *tp;
211
212 tp = trace_probe_primary_from_call(call);
213 if (WARN_ON_ONCE(!tp))
214 return NULL;
215
216 return container_of(tp, struct trace_kprobe, tp);
217}
218
219bool trace_kprobe_on_func_entry(struct trace_event_call *call)
220{
221 struct trace_kprobe *tk = trace_kprobe_primary_from_call(call);
222
223 return tk ? (kprobe_on_func_entry(tk->rp.kp.addr,
224 tk->rp.kp.addr ? NULL : tk->rp.kp.symbol_name,
225 tk->rp.kp.addr ? 0 : tk->rp.kp.offset) == 0) : false;
226}
227
228bool trace_kprobe_error_injectable(struct trace_event_call *call)
229{
230 struct trace_kprobe *tk = trace_kprobe_primary_from_call(call);
231
232 return tk ? within_error_injection_list(trace_kprobe_address(tk)) :
233 false;
234}
235
236static int register_kprobe_event(struct trace_kprobe *tk);
237static int unregister_kprobe_event(struct trace_kprobe *tk);
238
239static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
240static int kretprobe_dispatcher(struct kretprobe_instance *ri,
241 struct pt_regs *regs);
242
243static void free_trace_kprobe(struct trace_kprobe *tk)
244{
245 if (tk) {
246 trace_probe_cleanup(&tk->tp);
247 kfree(tk->symbol);
248 free_percpu(tk->nhit);
249 kfree(tk);
250 }
251}
252
253/*
254 * Allocate new trace_probe and initialize it (including kprobes).
255 */
256static struct trace_kprobe *alloc_trace_kprobe(const char *group,
257 const char *event,
258 void *addr,
259 const char *symbol,
260 unsigned long offs,
261 int maxactive,
262 int nargs, bool is_return)
263{
264 struct trace_kprobe *tk;
265 int ret = -ENOMEM;
266
267 tk = kzalloc(struct_size(tk, tp.args, nargs), GFP_KERNEL);
268 if (!tk)
269 return ERR_PTR(ret);
270
271 tk->nhit = alloc_percpu(unsigned long);
272 if (!tk->nhit)
273 goto error;
274
275 if (symbol) {
276 tk->symbol = kstrdup(symbol, GFP_KERNEL);
277 if (!tk->symbol)
278 goto error;
279 tk->rp.kp.symbol_name = tk->symbol;
280 tk->rp.kp.offset = offs;
281 } else
282 tk->rp.kp.addr = addr;
283
284 if (is_return)
285 tk->rp.handler = kretprobe_dispatcher;
286 else
287 tk->rp.kp.pre_handler = kprobe_dispatcher;
288
289 tk->rp.maxactive = maxactive;
290 INIT_HLIST_NODE(&tk->rp.kp.hlist);
291 INIT_LIST_HEAD(&tk->rp.kp.list);
292
293 ret = trace_probe_init(&tk->tp, event, group, false);
294 if (ret < 0)
295 goto error;
296
297 dyn_event_init(&tk->devent, &trace_kprobe_ops);
298 return tk;
299error:
300 free_trace_kprobe(tk);
301 return ERR_PTR(ret);
302}
303
304static struct trace_kprobe *find_trace_kprobe(const char *event,
305 const char *group)
306{
307 struct dyn_event *pos;
308 struct trace_kprobe *tk;
309
310 for_each_trace_kprobe(tk, pos)
311 if (strcmp(trace_probe_name(&tk->tp), event) == 0 &&
312 strcmp(trace_probe_group_name(&tk->tp), group) == 0)
313 return tk;
314 return NULL;
315}
316
317static inline int __enable_trace_kprobe(struct trace_kprobe *tk)
318{
319 int ret = 0;
320
321 if (trace_kprobe_is_registered(tk) && !trace_kprobe_has_gone(tk)) {
322 if (trace_kprobe_is_return(tk))
323 ret = enable_kretprobe(&tk->rp);
324 else
325 ret = enable_kprobe(&tk->rp.kp);
326 }
327
328 return ret;
329}
330
331static void __disable_trace_kprobe(struct trace_probe *tp)
332{
333 struct trace_kprobe *tk;
334
335 list_for_each_entry(tk, trace_probe_probe_list(tp), tp.list) {
336 if (!trace_kprobe_is_registered(tk))
337 continue;
338 if (trace_kprobe_is_return(tk))
339 disable_kretprobe(&tk->rp);
340 else
341 disable_kprobe(&tk->rp.kp);
342 }
343}
344
345/*
346 * Enable trace_probe
347 * if the file is NULL, enable "perf" handler, or enable "trace" handler.
348 */
349static int enable_trace_kprobe(struct trace_event_call *call,
350 struct trace_event_file *file)
351{
352 struct trace_probe *tp;
353 struct trace_kprobe *tk;
354 bool enabled;
355 int ret = 0;
356
357 tp = trace_probe_primary_from_call(call);
358 if (WARN_ON_ONCE(!tp))
359 return -ENODEV;
360 enabled = trace_probe_is_enabled(tp);
361
362 /* This also changes "enabled" state */
363 if (file) {
364 ret = trace_probe_add_file(tp, file);
365 if (ret)
366 return ret;
367 } else
368 trace_probe_set_flag(tp, TP_FLAG_PROFILE);
369
370 if (enabled)
371 return 0;
372
373 list_for_each_entry(tk, trace_probe_probe_list(tp), tp.list) {
374 if (trace_kprobe_has_gone(tk))
375 continue;
376 ret = __enable_trace_kprobe(tk);
377 if (ret)
378 break;
379 enabled = true;
380 }
381
382 if (ret) {
383 /* Failed to enable one of them. Roll back all */
384 if (enabled)
385 __disable_trace_kprobe(tp);
386 if (file)
387 trace_probe_remove_file(tp, file);
388 else
389 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
390 }
391
392 return ret;
393}
394
395/*
396 * Disable trace_probe
397 * if the file is NULL, disable "perf" handler, or disable "trace" handler.
398 */
399static int disable_trace_kprobe(struct trace_event_call *call,
400 struct trace_event_file *file)
401{
402 struct trace_probe *tp;
403
404 tp = trace_probe_primary_from_call(call);
405 if (WARN_ON_ONCE(!tp))
406 return -ENODEV;
407
408 if (file) {
409 if (!trace_probe_get_file_link(tp, file))
410 return -ENOENT;
411 if (!trace_probe_has_single_file(tp))
412 goto out;
413 trace_probe_clear_flag(tp, TP_FLAG_TRACE);
414 } else
415 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
416
417 if (!trace_probe_is_enabled(tp))
418 __disable_trace_kprobe(tp);
419
420 out:
421 if (file)
422 /*
423 * Synchronization is done in below function. For perf event,
424 * file == NULL and perf_trace_event_unreg() calls
425 * tracepoint_synchronize_unregister() to ensure synchronize
426 * event. We don't need to care about it.
427 */
428 trace_probe_remove_file(tp, file);
429
430 return 0;
431}
432
433#if defined(CONFIG_DYNAMIC_FTRACE) && \
434 !defined(CONFIG_KPROBE_EVENTS_ON_NOTRACE)
435static bool __within_notrace_func(unsigned long addr)
436{
437 unsigned long offset, size;
438
439 if (!addr || !kallsyms_lookup_size_offset(addr, &size, &offset))
440 return false;
441
442 /* Get the entry address of the target function */
443 addr -= offset;
444
445 /*
446 * Since ftrace_location_range() does inclusive range check, we need
447 * to subtract 1 byte from the end address.
448 */
449 return !ftrace_location_range(addr, addr + size - 1);
450}
451
452static bool within_notrace_func(struct trace_kprobe *tk)
453{
454 unsigned long addr = trace_kprobe_address(tk);
455 char symname[KSYM_NAME_LEN], *p;
456
457 if (!__within_notrace_func(addr))
458 return false;
459
460 /* Check if the address is on a suffixed-symbol */
461 if (!lookup_symbol_name(addr, symname)) {
462 p = strchr(symname, '.');
463 if (!p)
464 return true;
465 *p = '\0';
466 addr = (unsigned long)kprobe_lookup_name(symname, 0);
467 if (addr)
468 return __within_notrace_func(addr);
469 }
470
471 return true;
472}
473#else
474#define within_notrace_func(tk) (false)
475#endif
476
477/* Internal register function - just handle k*probes and flags */
478static int __register_trace_kprobe(struct trace_kprobe *tk)
479{
480 int i, ret;
481
482 ret = security_locked_down(LOCKDOWN_KPROBES);
483 if (ret)
484 return ret;
485
486 if (trace_kprobe_is_registered(tk))
487 return -EINVAL;
488
489 if (within_notrace_func(tk)) {
490 pr_warn("Could not probe notrace function %ps\n",
491 (void *)trace_kprobe_address(tk));
492 return -EINVAL;
493 }
494
495 for (i = 0; i < tk->tp.nr_args; i++) {
496 ret = traceprobe_update_arg(&tk->tp.args[i]);
497 if (ret)
498 return ret;
499 }
500
501 /* Set/clear disabled flag according to tp->flag */
502 if (trace_probe_is_enabled(&tk->tp))
503 tk->rp.kp.flags &= ~KPROBE_FLAG_DISABLED;
504 else
505 tk->rp.kp.flags |= KPROBE_FLAG_DISABLED;
506
507 if (trace_kprobe_is_return(tk))
508 ret = register_kretprobe(&tk->rp);
509 else
510 ret = register_kprobe(&tk->rp.kp);
511
512 return ret;
513}
514
515/* Internal unregister function - just handle k*probes and flags */
516static void __unregister_trace_kprobe(struct trace_kprobe *tk)
517{
518 if (trace_kprobe_is_registered(tk)) {
519 if (trace_kprobe_is_return(tk))
520 unregister_kretprobe(&tk->rp);
521 else
522 unregister_kprobe(&tk->rp.kp);
523 /* Cleanup kprobe for reuse and mark it unregistered */
524 INIT_HLIST_NODE(&tk->rp.kp.hlist);
525 INIT_LIST_HEAD(&tk->rp.kp.list);
526 if (tk->rp.kp.symbol_name)
527 tk->rp.kp.addr = NULL;
528 }
529}
530
531/* Unregister a trace_probe and probe_event */
532static int unregister_trace_kprobe(struct trace_kprobe *tk)
533{
534 /* If other probes are on the event, just unregister kprobe */
535 if (trace_probe_has_sibling(&tk->tp))
536 goto unreg;
537
538 /* Enabled event can not be unregistered */
539 if (trace_probe_is_enabled(&tk->tp))
540 return -EBUSY;
541
542 /* If there's a reference to the dynamic event */
543 if (trace_event_dyn_busy(trace_probe_event_call(&tk->tp)))
544 return -EBUSY;
545
546 /* Will fail if probe is being used by ftrace or perf */
547 if (unregister_kprobe_event(tk))
548 return -EBUSY;
549
550unreg:
551 __unregister_trace_kprobe(tk);
552 dyn_event_remove(&tk->devent);
553 trace_probe_unlink(&tk->tp);
554
555 return 0;
556}
557
558static bool trace_kprobe_has_same_kprobe(struct trace_kprobe *orig,
559 struct trace_kprobe *comp)
560{
561 struct trace_probe_event *tpe = orig->tp.event;
562 int i;
563
564 list_for_each_entry(orig, &tpe->probes, tp.list) {
565 if (strcmp(trace_kprobe_symbol(orig),
566 trace_kprobe_symbol(comp)) ||
567 trace_kprobe_offset(orig) != trace_kprobe_offset(comp))
568 continue;
569
570 /*
571 * trace_probe_compare_arg_type() ensured that nr_args and
572 * each argument name and type are same. Let's compare comm.
573 */
574 for (i = 0; i < orig->tp.nr_args; i++) {
575 if (strcmp(orig->tp.args[i].comm,
576 comp->tp.args[i].comm))
577 break;
578 }
579
580 if (i == orig->tp.nr_args)
581 return true;
582 }
583
584 return false;
585}
586
587static int append_trace_kprobe(struct trace_kprobe *tk, struct trace_kprobe *to)
588{
589 int ret;
590
591 ret = trace_probe_compare_arg_type(&tk->tp, &to->tp);
592 if (ret) {
593 /* Note that argument starts index = 2 */
594 trace_probe_log_set_index(ret + 1);
595 trace_probe_log_err(0, DIFF_ARG_TYPE);
596 return -EEXIST;
597 }
598 if (trace_kprobe_has_same_kprobe(to, tk)) {
599 trace_probe_log_set_index(0);
600 trace_probe_log_err(0, SAME_PROBE);
601 return -EEXIST;
602 }
603
604 /* Append to existing event */
605 ret = trace_probe_append(&tk->tp, &to->tp);
606 if (ret)
607 return ret;
608
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");
613 ret = 0;
614 }
615
616 if (ret)
617 trace_probe_unlink(&tk->tp);
618 else
619 dyn_event_add(&tk->devent, trace_probe_event_call(&tk->tp));
620
621 return ret;
622}
623
624/* Register a trace_probe and probe_event */
625static int register_trace_kprobe(struct trace_kprobe *tk)
626{
627 struct trace_kprobe *old_tk;
628 int ret;
629
630 mutex_lock(&event_mutex);
631
632 old_tk = find_trace_kprobe(trace_probe_name(&tk->tp),
633 trace_probe_group_name(&tk->tp));
634 if (old_tk) {
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);
638 ret = -EEXIST;
639 } else {
640 ret = append_trace_kprobe(tk, old_tk);
641 }
642 goto end;
643 }
644
645 /* Register new event */
646 ret = register_kprobe_event(tk);
647 if (ret) {
648 if (ret == -EEXIST) {
649 trace_probe_log_set_index(0);
650 trace_probe_log_err(0, EVENT_EXIST);
651 } else
652 pr_warn("Failed to register probe event(%d)\n", ret);
653 goto end;
654 }
655
656 /* Register k*probe */
657 ret = __register_trace_kprobe(tk);
658 if (ret == -ENOENT && !trace_kprobe_module_exist(tk)) {
659 pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
660 ret = 0;
661 }
662
663 if (ret < 0)
664 unregister_kprobe_event(tk);
665 else
666 dyn_event_add(&tk->devent, trace_probe_event_call(&tk->tp));
667
668end:
669 mutex_unlock(&event_mutex);
670 return ret;
671}
672
673/* Module notifier call back, checking event on the module */
674static int trace_kprobe_module_callback(struct notifier_block *nb,
675 unsigned long val, void *data)
676{
677 struct module *mod = data;
678 struct dyn_event *pos;
679 struct trace_kprobe *tk;
680 int ret;
681
682 if (val != MODULE_STATE_COMING)
683 return NOTIFY_DONE;
684
685 /* Update probes on coming module */
686 mutex_lock(&event_mutex);
687 for_each_trace_kprobe(tk, pos) {
688 if (trace_kprobe_within_module(tk, mod)) {
689 /* Don't need to check busy - this should have gone. */
690 __unregister_trace_kprobe(tk);
691 ret = __register_trace_kprobe(tk);
692 if (ret)
693 pr_warn("Failed to re-register probe %s on %s: %d\n",
694 trace_probe_name(&tk->tp),
695 module_name(mod), ret);
696 }
697 }
698 mutex_unlock(&event_mutex);
699
700 return NOTIFY_DONE;
701}
702
703static struct notifier_block trace_kprobe_module_nb = {
704 .notifier_call = trace_kprobe_module_callback,
705 .priority = 1 /* Invoked after kprobe module callback */
706};
707
708static int count_symbols(void *data, unsigned long unused)
709{
710 unsigned int *count = data;
711
712 (*count)++;
713
714 return 0;
715}
716
717struct sym_count_ctx {
718 unsigned int count;
719 const char *name;
720};
721
722static int count_mod_symbols(void *data, const char *name, unsigned long unused)
723{
724 struct sym_count_ctx *ctx = data;
725
726 if (strcmp(name, ctx->name) == 0)
727 ctx->count++;
728
729 return 0;
730}
731
732static unsigned int number_of_same_symbols(char *func_name)
733{
734 struct sym_count_ctx ctx = { .count = 0, .name = func_name };
735
736 kallsyms_on_each_match_symbol(count_symbols, func_name, &ctx.count);
737
738 module_kallsyms_on_each_symbol(NULL, count_mod_symbols, &ctx);
739
740 return ctx.count;
741}
742
743static int __trace_kprobe_create(int argc, const char *argv[])
744{
745 /*
746 * Argument syntax:
747 * - Add kprobe:
748 * p[:[GRP/][EVENT]] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
749 * - Add kretprobe:
750 * r[MAXACTIVE][:[GRP/][EVENT]] [MOD:]KSYM[+0] [FETCHARGS]
751 * Or
752 * p[:[GRP/][EVENT]] [MOD:]KSYM[+0]%return [FETCHARGS]
753 *
754 * Fetch args:
755 * $retval : fetch return value
756 * $stack : fetch stack address
757 * $stackN : fetch Nth of stack (N:0-)
758 * $comm : fetch current task comm
759 * @ADDR : fetch memory at ADDR (ADDR should be in kernel)
760 * @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
761 * %REG : fetch register REG
762 * Dereferencing memory fetch:
763 * +|-offs(ARG) : fetch memory at ARG +|- offs address.
764 * Alias name of args:
765 * NAME=FETCHARG : set NAME as alias of FETCHARG.
766 * Type of args:
767 * FETCHARG:TYPE : use TYPE instead of unsigned long.
768 */
769 struct trace_kprobe *tk = NULL;
770 int i, len, new_argc = 0, ret = 0;
771 bool is_return = false;
772 char *symbol = NULL, *tmp = NULL;
773 const char **new_argv = NULL;
774 const char *event = NULL, *group = KPROBE_EVENT_SYSTEM;
775 enum probe_print_type ptype;
776 int maxactive = 0;
777 long offset = 0;
778 void *addr = NULL;
779 char buf[MAX_EVENT_NAME_LEN];
780 char gbuf[MAX_EVENT_NAME_LEN];
781 char abuf[MAX_BTF_ARGS_LEN];
782 struct traceprobe_parse_context ctx = { .flags = TPARG_FL_KERNEL };
783
784 switch (argv[0][0]) {
785 case 'r':
786 is_return = true;
787 break;
788 case 'p':
789 break;
790 default:
791 return -ECANCELED;
792 }
793 if (argc < 2)
794 return -ECANCELED;
795
796 trace_probe_log_init("trace_kprobe", argc, argv);
797
798 event = strchr(&argv[0][1], ':');
799 if (event)
800 event++;
801
802 if (isdigit(argv[0][1])) {
803 if (!is_return) {
804 trace_probe_log_err(1, BAD_MAXACT_TYPE);
805 goto parse_error;
806 }
807 if (event)
808 len = event - &argv[0][1] - 1;
809 else
810 len = strlen(&argv[0][1]);
811 if (len > MAX_EVENT_NAME_LEN - 1) {
812 trace_probe_log_err(1, BAD_MAXACT);
813 goto parse_error;
814 }
815 memcpy(buf, &argv[0][1], len);
816 buf[len] = '\0';
817 ret = kstrtouint(buf, 0, &maxactive);
818 if (ret || !maxactive) {
819 trace_probe_log_err(1, BAD_MAXACT);
820 goto parse_error;
821 }
822 /* kretprobes instances are iterated over via a list. The
823 * maximum should stay reasonable.
824 */
825 if (maxactive > KRETPROBE_MAXACTIVE_MAX) {
826 trace_probe_log_err(1, MAXACT_TOO_BIG);
827 goto parse_error;
828 }
829 }
830
831 /* try to parse an address. if that fails, try to read the
832 * input as a symbol. */
833 if (kstrtoul(argv[1], 0, (unsigned long *)&addr)) {
834 trace_probe_log_set_index(1);
835 /* Check whether uprobe event specified */
836 if (strchr(argv[1], '/') && strchr(argv[1], ':')) {
837 ret = -ECANCELED;
838 goto error;
839 }
840 /* a symbol specified */
841 symbol = kstrdup(argv[1], GFP_KERNEL);
842 if (!symbol)
843 return -ENOMEM;
844
845 tmp = strchr(symbol, '%');
846 if (tmp) {
847 if (!strcmp(tmp, "%return")) {
848 *tmp = '\0';
849 is_return = true;
850 } else {
851 trace_probe_log_err(tmp - symbol, BAD_ADDR_SUFFIX);
852 goto parse_error;
853 }
854 }
855
856 /* TODO: support .init module functions */
857 ret = traceprobe_split_symbol_offset(symbol, &offset);
858 if (ret || offset < 0 || offset > UINT_MAX) {
859 trace_probe_log_err(0, BAD_PROBE_ADDR);
860 goto parse_error;
861 }
862 if (is_return)
863 ctx.flags |= TPARG_FL_RETURN;
864 ret = kprobe_on_func_entry(NULL, symbol, offset);
865 if (ret == 0 && !is_return)
866 ctx.flags |= TPARG_FL_FENTRY;
867 /* Defer the ENOENT case until register kprobe */
868 if (ret == -EINVAL && is_return) {
869 trace_probe_log_err(0, BAD_RETPROBE);
870 goto parse_error;
871 }
872 }
873
874 if (symbol && !strchr(symbol, ':')) {
875 unsigned int count;
876
877 count = number_of_same_symbols(symbol);
878 if (count > 1) {
879 /*
880 * Users should use ADDR to remove the ambiguity of
881 * using KSYM only.
882 */
883 trace_probe_log_err(0, NON_UNIQ_SYMBOL);
884 ret = -EADDRNOTAVAIL;
885
886 goto error;
887 } else if (count == 0) {
888 /*
889 * We can return ENOENT earlier than when register the
890 * kprobe.
891 */
892 trace_probe_log_err(0, BAD_PROBE_ADDR);
893 ret = -ENOENT;
894
895 goto error;
896 }
897 }
898
899 trace_probe_log_set_index(0);
900 if (event) {
901 ret = traceprobe_parse_event_name(&event, &group, gbuf,
902 event - argv[0]);
903 if (ret)
904 goto parse_error;
905 }
906
907 if (!event) {
908 /* Make a new event name */
909 if (symbol)
910 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
911 is_return ? 'r' : 'p', symbol, offset);
912 else
913 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
914 is_return ? 'r' : 'p', addr);
915 sanitize_event_name(buf);
916 event = buf;
917 }
918
919 argc -= 2; argv += 2;
920 ctx.funcname = symbol;
921 new_argv = traceprobe_expand_meta_args(argc, argv, &new_argc,
922 abuf, MAX_BTF_ARGS_LEN, &ctx);
923 if (IS_ERR(new_argv)) {
924 ret = PTR_ERR(new_argv);
925 new_argv = NULL;
926 goto out;
927 }
928 if (new_argv) {
929 argc = new_argc;
930 argv = new_argv;
931 }
932
933 /* setup a probe */
934 tk = alloc_trace_kprobe(group, event, addr, symbol, offset, maxactive,
935 argc, is_return);
936 if (IS_ERR(tk)) {
937 ret = PTR_ERR(tk);
938 /* This must return -ENOMEM, else there is a bug */
939 WARN_ON_ONCE(ret != -ENOMEM);
940 goto out; /* We know tk is not allocated */
941 }
942
943 /* parse arguments */
944 for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
945 trace_probe_log_set_index(i + 2);
946 ctx.offset = 0;
947 ret = traceprobe_parse_probe_arg(&tk->tp, i, argv[i], &ctx);
948 if (ret)
949 goto error; /* This can be -ENOMEM */
950 }
951
952 ptype = is_return ? PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL;
953 ret = traceprobe_set_print_fmt(&tk->tp, ptype);
954 if (ret < 0)
955 goto error;
956
957 ret = register_trace_kprobe(tk);
958 if (ret) {
959 trace_probe_log_set_index(1);
960 if (ret == -EILSEQ)
961 trace_probe_log_err(0, BAD_INSN_BNDRY);
962 else if (ret == -ENOENT)
963 trace_probe_log_err(0, BAD_PROBE_ADDR);
964 else if (ret != -ENOMEM && ret != -EEXIST)
965 trace_probe_log_err(0, FAIL_REG_PROBE);
966 goto error;
967 }
968
969out:
970 traceprobe_finish_parse(&ctx);
971 trace_probe_log_clear();
972 kfree(new_argv);
973 kfree(symbol);
974 return ret;
975
976parse_error:
977 ret = -EINVAL;
978error:
979 free_trace_kprobe(tk);
980 goto out;
981}
982
983static int trace_kprobe_create(const char *raw_command)
984{
985 return trace_probe_create(raw_command, __trace_kprobe_create);
986}
987
988static int create_or_delete_trace_kprobe(const char *raw_command)
989{
990 int ret;
991
992 if (raw_command[0] == '-')
993 return dyn_event_release(raw_command, &trace_kprobe_ops);
994
995 ret = trace_kprobe_create(raw_command);
996 return ret == -ECANCELED ? -EINVAL : ret;
997}
998
999static int trace_kprobe_run_command(struct dynevent_cmd *cmd)
1000{
1001 return create_or_delete_trace_kprobe(cmd->seq.buffer);
1002}
1003
1004/**
1005 * kprobe_event_cmd_init - Initialize a kprobe event command object
1006 * @cmd: A pointer to the dynevent_cmd struct representing the new event
1007 * @buf: A pointer to the buffer used to build the command
1008 * @maxlen: The length of the buffer passed in @buf
1009 *
1010 * Initialize a synthetic event command object. Use this before
1011 * calling any of the other kprobe_event functions.
1012 */
1013void kprobe_event_cmd_init(struct dynevent_cmd *cmd, char *buf, int maxlen)
1014{
1015 dynevent_cmd_init(cmd, buf, maxlen, DYNEVENT_TYPE_KPROBE,
1016 trace_kprobe_run_command);
1017}
1018EXPORT_SYMBOL_GPL(kprobe_event_cmd_init);
1019
1020/**
1021 * __kprobe_event_gen_cmd_start - Generate a kprobe event command from arg list
1022 * @cmd: A pointer to the dynevent_cmd struct representing the new event
1023 * @kretprobe: Is this a return probe?
1024 * @name: The name of the kprobe event
1025 * @loc: The location of the kprobe event
1026 * @...: Variable number of arg (pairs), one pair for each field
1027 *
1028 * NOTE: Users normally won't want to call this function directly, but
1029 * rather use the kprobe_event_gen_cmd_start() wrapper, which automatically
1030 * adds a NULL to the end of the arg list. If this function is used
1031 * directly, make sure the last arg in the variable arg list is NULL.
1032 *
1033 * Generate a kprobe event command to be executed by
1034 * kprobe_event_gen_cmd_end(). This function can be used to generate the
1035 * complete command or only the first part of it; in the latter case,
1036 * kprobe_event_add_fields() can be used to add more fields following this.
1037 *
1038 * Unlikely the synth_event_gen_cmd_start(), @loc must be specified. This
1039 * returns -EINVAL if @loc == NULL.
1040 *
1041 * Return: 0 if successful, error otherwise.
1042 */
1043int __kprobe_event_gen_cmd_start(struct dynevent_cmd *cmd, bool kretprobe,
1044 const char *name, const char *loc, ...)
1045{
1046 char buf[MAX_EVENT_NAME_LEN];
1047 struct dynevent_arg arg;
1048 va_list args;
1049 int ret;
1050
1051 if (cmd->type != DYNEVENT_TYPE_KPROBE)
1052 return -EINVAL;
1053
1054 if (!loc)
1055 return -EINVAL;
1056
1057 if (kretprobe)
1058 snprintf(buf, MAX_EVENT_NAME_LEN, "r:kprobes/%s", name);
1059 else
1060 snprintf(buf, MAX_EVENT_NAME_LEN, "p:kprobes/%s", name);
1061
1062 ret = dynevent_str_add(cmd, buf);
1063 if (ret)
1064 return ret;
1065
1066 dynevent_arg_init(&arg, 0);
1067 arg.str = loc;
1068 ret = dynevent_arg_add(cmd, &arg, NULL);
1069 if (ret)
1070 return ret;
1071
1072 va_start(args, loc);
1073 for (;;) {
1074 const char *field;
1075
1076 field = va_arg(args, const char *);
1077 if (!field)
1078 break;
1079
1080 if (++cmd->n_fields > MAX_TRACE_ARGS) {
1081 ret = -EINVAL;
1082 break;
1083 }
1084
1085 arg.str = field;
1086 ret = dynevent_arg_add(cmd, &arg, NULL);
1087 if (ret)
1088 break;
1089 }
1090 va_end(args);
1091
1092 return ret;
1093}
1094EXPORT_SYMBOL_GPL(__kprobe_event_gen_cmd_start);
1095
1096/**
1097 * __kprobe_event_add_fields - Add probe fields to a kprobe command from arg list
1098 * @cmd: A pointer to the dynevent_cmd struct representing the new event
1099 * @...: Variable number of arg (pairs), one pair for each field
1100 *
1101 * NOTE: Users normally won't want to call this function directly, but
1102 * rather use the kprobe_event_add_fields() wrapper, which
1103 * automatically adds a NULL to the end of the arg list. If this
1104 * function is used directly, make sure the last arg in the variable
1105 * arg list is NULL.
1106 *
1107 * Add probe fields to an existing kprobe command using a variable
1108 * list of args. Fields are added in the same order they're listed.
1109 *
1110 * Return: 0 if successful, error otherwise.
1111 */
1112int __kprobe_event_add_fields(struct dynevent_cmd *cmd, ...)
1113{
1114 struct dynevent_arg arg;
1115 va_list args;
1116 int ret = 0;
1117
1118 if (cmd->type != DYNEVENT_TYPE_KPROBE)
1119 return -EINVAL;
1120
1121 dynevent_arg_init(&arg, 0);
1122
1123 va_start(args, cmd);
1124 for (;;) {
1125 const char *field;
1126
1127 field = va_arg(args, const char *);
1128 if (!field)
1129 break;
1130
1131 if (++cmd->n_fields > MAX_TRACE_ARGS) {
1132 ret = -EINVAL;
1133 break;
1134 }
1135
1136 arg.str = field;
1137 ret = dynevent_arg_add(cmd, &arg, NULL);
1138 if (ret)
1139 break;
1140 }
1141 va_end(args);
1142
1143 return ret;
1144}
1145EXPORT_SYMBOL_GPL(__kprobe_event_add_fields);
1146
1147/**
1148 * kprobe_event_delete - Delete a kprobe event
1149 * @name: The name of the kprobe event to delete
1150 *
1151 * Delete a kprobe event with the give @name from kernel code rather
1152 * than directly from the command line.
1153 *
1154 * Return: 0 if successful, error otherwise.
1155 */
1156int kprobe_event_delete(const char *name)
1157{
1158 char buf[MAX_EVENT_NAME_LEN];
1159
1160 snprintf(buf, MAX_EVENT_NAME_LEN, "-:%s", name);
1161
1162 return create_or_delete_trace_kprobe(buf);
1163}
1164EXPORT_SYMBOL_GPL(kprobe_event_delete);
1165
1166static int trace_kprobe_release(struct dyn_event *ev)
1167{
1168 struct trace_kprobe *tk = to_trace_kprobe(ev);
1169 int ret = unregister_trace_kprobe(tk);
1170
1171 if (!ret)
1172 free_trace_kprobe(tk);
1173 return ret;
1174}
1175
1176static int trace_kprobe_show(struct seq_file *m, struct dyn_event *ev)
1177{
1178 struct trace_kprobe *tk = to_trace_kprobe(ev);
1179 int i;
1180
1181 seq_putc(m, trace_kprobe_is_return(tk) ? 'r' : 'p');
1182 if (trace_kprobe_is_return(tk) && tk->rp.maxactive)
1183 seq_printf(m, "%d", tk->rp.maxactive);
1184 seq_printf(m, ":%s/%s", trace_probe_group_name(&tk->tp),
1185 trace_probe_name(&tk->tp));
1186
1187 if (!tk->symbol)
1188 seq_printf(m, " 0x%p", tk->rp.kp.addr);
1189 else if (tk->rp.kp.offset)
1190 seq_printf(m, " %s+%u", trace_kprobe_symbol(tk),
1191 tk->rp.kp.offset);
1192 else
1193 seq_printf(m, " %s", trace_kprobe_symbol(tk));
1194
1195 for (i = 0; i < tk->tp.nr_args; i++)
1196 seq_printf(m, " %s=%s", tk->tp.args[i].name, tk->tp.args[i].comm);
1197 seq_putc(m, '\n');
1198
1199 return 0;
1200}
1201
1202static int probes_seq_show(struct seq_file *m, void *v)
1203{
1204 struct dyn_event *ev = v;
1205
1206 if (!is_trace_kprobe(ev))
1207 return 0;
1208
1209 return trace_kprobe_show(m, ev);
1210}
1211
1212static const struct seq_operations probes_seq_op = {
1213 .start = dyn_event_seq_start,
1214 .next = dyn_event_seq_next,
1215 .stop = dyn_event_seq_stop,
1216 .show = probes_seq_show
1217};
1218
1219static int probes_open(struct inode *inode, struct file *file)
1220{
1221 int ret;
1222
1223 ret = security_locked_down(LOCKDOWN_TRACEFS);
1224 if (ret)
1225 return ret;
1226
1227 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
1228 ret = dyn_events_release_all(&trace_kprobe_ops);
1229 if (ret < 0)
1230 return ret;
1231 }
1232
1233 return seq_open(file, &probes_seq_op);
1234}
1235
1236static ssize_t probes_write(struct file *file, const char __user *buffer,
1237 size_t count, loff_t *ppos)
1238{
1239 return trace_parse_run_command(file, buffer, count, ppos,
1240 create_or_delete_trace_kprobe);
1241}
1242
1243static const struct file_operations kprobe_events_ops = {
1244 .owner = THIS_MODULE,
1245 .open = probes_open,
1246 .read = seq_read,
1247 .llseek = seq_lseek,
1248 .release = seq_release,
1249 .write = probes_write,
1250};
1251
1252static unsigned long trace_kprobe_missed(struct trace_kprobe *tk)
1253{
1254 return trace_kprobe_is_return(tk) ?
1255 tk->rp.kp.nmissed + tk->rp.nmissed : tk->rp.kp.nmissed;
1256}
1257
1258/* Probes profiling interfaces */
1259static int probes_profile_seq_show(struct seq_file *m, void *v)
1260{
1261 struct dyn_event *ev = v;
1262 struct trace_kprobe *tk;
1263 unsigned long nmissed;
1264
1265 if (!is_trace_kprobe(ev))
1266 return 0;
1267
1268 tk = to_trace_kprobe(ev);
1269 nmissed = trace_kprobe_missed(tk);
1270 seq_printf(m, " %-44s %15lu %15lu\n",
1271 trace_probe_name(&tk->tp),
1272 trace_kprobe_nhit(tk),
1273 nmissed);
1274
1275 return 0;
1276}
1277
1278static const struct seq_operations profile_seq_op = {
1279 .start = dyn_event_seq_start,
1280 .next = dyn_event_seq_next,
1281 .stop = dyn_event_seq_stop,
1282 .show = probes_profile_seq_show
1283};
1284
1285static int profile_open(struct inode *inode, struct file *file)
1286{
1287 int ret;
1288
1289 ret = security_locked_down(LOCKDOWN_TRACEFS);
1290 if (ret)
1291 return ret;
1292
1293 return seq_open(file, &profile_seq_op);
1294}
1295
1296static const struct file_operations kprobe_profile_ops = {
1297 .owner = THIS_MODULE,
1298 .open = profile_open,
1299 .read = seq_read,
1300 .llseek = seq_lseek,
1301 .release = seq_release,
1302};
1303
1304/* Note that we don't verify it, since the code does not come from user space */
1305static int
1306process_fetch_insn(struct fetch_insn *code, void *rec, void *dest,
1307 void *base)
1308{
1309 struct pt_regs *regs = rec;
1310 unsigned long val;
1311 int ret;
1312
1313retry:
1314 /* 1st stage: get value from context */
1315 switch (code->op) {
1316 case FETCH_OP_REG:
1317 val = regs_get_register(regs, code->param);
1318 break;
1319 case FETCH_OP_STACK:
1320 val = regs_get_kernel_stack_nth(regs, code->param);
1321 break;
1322 case FETCH_OP_STACKP:
1323 val = kernel_stack_pointer(regs);
1324 break;
1325 case FETCH_OP_RETVAL:
1326 val = regs_return_value(regs);
1327 break;
1328#ifdef CONFIG_HAVE_FUNCTION_ARG_ACCESS_API
1329 case FETCH_OP_ARG:
1330 val = regs_get_kernel_argument(regs, code->param);
1331 break;
1332#endif
1333 case FETCH_NOP_SYMBOL: /* Ignore a place holder */
1334 code++;
1335 goto retry;
1336 default:
1337 ret = process_common_fetch_insn(code, &val);
1338 if (ret < 0)
1339 return ret;
1340 }
1341 code++;
1342
1343 return process_fetch_insn_bottom(code, val, dest, base);
1344}
1345NOKPROBE_SYMBOL(process_fetch_insn)
1346
1347/* Kprobe handler */
1348static nokprobe_inline void
1349__kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs,
1350 struct trace_event_file *trace_file)
1351{
1352 struct kprobe_trace_entry_head *entry;
1353 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1354 struct trace_event_buffer fbuffer;
1355 int dsize;
1356
1357 WARN_ON(call != trace_file->event_call);
1358
1359 if (trace_trigger_soft_disabled(trace_file))
1360 return;
1361
1362 dsize = __get_data_size(&tk->tp, regs);
1363
1364 entry = trace_event_buffer_reserve(&fbuffer, trace_file,
1365 sizeof(*entry) + tk->tp.size + dsize);
1366 if (!entry)
1367 return;
1368
1369 fbuffer.regs = regs;
1370 entry->ip = (unsigned long)tk->rp.kp.addr;
1371 store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1372
1373 trace_event_buffer_commit(&fbuffer);
1374}
1375
1376static void
1377kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs)
1378{
1379 struct event_file_link *link;
1380
1381 trace_probe_for_each_link_rcu(link, &tk->tp)
1382 __kprobe_trace_func(tk, regs, link->file);
1383}
1384NOKPROBE_SYMBOL(kprobe_trace_func);
1385
1386/* Kretprobe handler */
1387static nokprobe_inline void
1388__kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1389 struct pt_regs *regs,
1390 struct trace_event_file *trace_file)
1391{
1392 struct kretprobe_trace_entry_head *entry;
1393 struct trace_event_buffer fbuffer;
1394 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1395 int dsize;
1396
1397 WARN_ON(call != trace_file->event_call);
1398
1399 if (trace_trigger_soft_disabled(trace_file))
1400 return;
1401
1402 dsize = __get_data_size(&tk->tp, regs);
1403
1404 entry = trace_event_buffer_reserve(&fbuffer, trace_file,
1405 sizeof(*entry) + tk->tp.size + dsize);
1406 if (!entry)
1407 return;
1408
1409 fbuffer.regs = regs;
1410 entry->func = (unsigned long)tk->rp.kp.addr;
1411 entry->ret_ip = get_kretprobe_retaddr(ri);
1412 store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1413
1414 trace_event_buffer_commit(&fbuffer);
1415}
1416
1417static void
1418kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1419 struct pt_regs *regs)
1420{
1421 struct event_file_link *link;
1422
1423 trace_probe_for_each_link_rcu(link, &tk->tp)
1424 __kretprobe_trace_func(tk, ri, regs, link->file);
1425}
1426NOKPROBE_SYMBOL(kretprobe_trace_func);
1427
1428/* Event entry printers */
1429static enum print_line_t
1430print_kprobe_event(struct trace_iterator *iter, int flags,
1431 struct trace_event *event)
1432{
1433 struct kprobe_trace_entry_head *field;
1434 struct trace_seq *s = &iter->seq;
1435 struct trace_probe *tp;
1436
1437 field = (struct kprobe_trace_entry_head *)iter->ent;
1438 tp = trace_probe_primary_from_call(
1439 container_of(event, struct trace_event_call, event));
1440 if (WARN_ON_ONCE(!tp))
1441 goto out;
1442
1443 trace_seq_printf(s, "%s: (", trace_probe_name(tp));
1444
1445 if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
1446 goto out;
1447
1448 trace_seq_putc(s, ')');
1449
1450 if (trace_probe_print_args(s, tp->args, tp->nr_args,
1451 (u8 *)&field[1], field) < 0)
1452 goto out;
1453
1454 trace_seq_putc(s, '\n');
1455 out:
1456 return trace_handle_return(s);
1457}
1458
1459static enum print_line_t
1460print_kretprobe_event(struct trace_iterator *iter, int flags,
1461 struct trace_event *event)
1462{
1463 struct kretprobe_trace_entry_head *field;
1464 struct trace_seq *s = &iter->seq;
1465 struct trace_probe *tp;
1466
1467 field = (struct kretprobe_trace_entry_head *)iter->ent;
1468 tp = trace_probe_primary_from_call(
1469 container_of(event, struct trace_event_call, event));
1470 if (WARN_ON_ONCE(!tp))
1471 goto out;
1472
1473 trace_seq_printf(s, "%s: (", trace_probe_name(tp));
1474
1475 if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1476 goto out;
1477
1478 trace_seq_puts(s, " <- ");
1479
1480 if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1481 goto out;
1482
1483 trace_seq_putc(s, ')');
1484
1485 if (trace_probe_print_args(s, tp->args, tp->nr_args,
1486 (u8 *)&field[1], field) < 0)
1487 goto out;
1488
1489 trace_seq_putc(s, '\n');
1490
1491 out:
1492 return trace_handle_return(s);
1493}
1494
1495
1496static int kprobe_event_define_fields(struct trace_event_call *event_call)
1497{
1498 int ret;
1499 struct kprobe_trace_entry_head field;
1500 struct trace_probe *tp;
1501
1502 tp = trace_probe_primary_from_call(event_call);
1503 if (WARN_ON_ONCE(!tp))
1504 return -ENOENT;
1505
1506 DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1507
1508 return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
1509}
1510
1511static int kretprobe_event_define_fields(struct trace_event_call *event_call)
1512{
1513 int ret;
1514 struct kretprobe_trace_entry_head field;
1515 struct trace_probe *tp;
1516
1517 tp = trace_probe_primary_from_call(event_call);
1518 if (WARN_ON_ONCE(!tp))
1519 return -ENOENT;
1520
1521 DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1522 DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1523
1524 return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
1525}
1526
1527#ifdef CONFIG_PERF_EVENTS
1528
1529/* Kprobe profile handler */
1530static int
1531kprobe_perf_func(struct trace_kprobe *tk, struct pt_regs *regs)
1532{
1533 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1534 struct kprobe_trace_entry_head *entry;
1535 struct hlist_head *head;
1536 int size, __size, dsize;
1537 int rctx;
1538
1539 if (bpf_prog_array_valid(call)) {
1540 unsigned long orig_ip = instruction_pointer(regs);
1541 int ret;
1542
1543 ret = trace_call_bpf(call, regs);
1544
1545 /*
1546 * We need to check and see if we modified the pc of the
1547 * pt_regs, and if so return 1 so that we don't do the
1548 * single stepping.
1549 */
1550 if (orig_ip != instruction_pointer(regs))
1551 return 1;
1552 if (!ret)
1553 return 0;
1554 }
1555
1556 head = this_cpu_ptr(call->perf_events);
1557 if (hlist_empty(head))
1558 return 0;
1559
1560 dsize = __get_data_size(&tk->tp, regs);
1561 __size = sizeof(*entry) + tk->tp.size + dsize;
1562 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1563 size -= sizeof(u32);
1564
1565 entry = perf_trace_buf_alloc(size, NULL, &rctx);
1566 if (!entry)
1567 return 0;
1568
1569 entry->ip = (unsigned long)tk->rp.kp.addr;
1570 memset(&entry[1], 0, dsize);
1571 store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1572 perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1573 head, NULL);
1574 return 0;
1575}
1576NOKPROBE_SYMBOL(kprobe_perf_func);
1577
1578/* Kretprobe profile handler */
1579static void
1580kretprobe_perf_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1581 struct pt_regs *regs)
1582{
1583 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1584 struct kretprobe_trace_entry_head *entry;
1585 struct hlist_head *head;
1586 int size, __size, dsize;
1587 int rctx;
1588
1589 if (bpf_prog_array_valid(call) && !trace_call_bpf(call, regs))
1590 return;
1591
1592 head = this_cpu_ptr(call->perf_events);
1593 if (hlist_empty(head))
1594 return;
1595
1596 dsize = __get_data_size(&tk->tp, regs);
1597 __size = sizeof(*entry) + tk->tp.size + dsize;
1598 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1599 size -= sizeof(u32);
1600
1601 entry = perf_trace_buf_alloc(size, NULL, &rctx);
1602 if (!entry)
1603 return;
1604
1605 entry->func = (unsigned long)tk->rp.kp.addr;
1606 entry->ret_ip = get_kretprobe_retaddr(ri);
1607 store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1608 perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1609 head, NULL);
1610}
1611NOKPROBE_SYMBOL(kretprobe_perf_func);
1612
1613int bpf_get_kprobe_info(const struct perf_event *event, u32 *fd_type,
1614 const char **symbol, u64 *probe_offset,
1615 u64 *probe_addr, unsigned long *missed,
1616 bool perf_type_tracepoint)
1617{
1618 const char *pevent = trace_event_name(event->tp_event);
1619 const char *group = event->tp_event->class->system;
1620 struct trace_kprobe *tk;
1621
1622 if (perf_type_tracepoint)
1623 tk = find_trace_kprobe(pevent, group);
1624 else
1625 tk = trace_kprobe_primary_from_call(event->tp_event);
1626 if (!tk)
1627 return -EINVAL;
1628
1629 *fd_type = trace_kprobe_is_return(tk) ? BPF_FD_TYPE_KRETPROBE
1630 : BPF_FD_TYPE_KPROBE;
1631 *probe_offset = tk->rp.kp.offset;
1632 *probe_addr = kallsyms_show_value(current_cred()) ?
1633 (unsigned long)tk->rp.kp.addr : 0;
1634 *symbol = tk->symbol;
1635 if (missed)
1636 *missed = trace_kprobe_missed(tk);
1637 return 0;
1638}
1639#endif /* CONFIG_PERF_EVENTS */
1640
1641/*
1642 * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex.
1643 *
1644 * kprobe_trace_self_tests_init() does enable_trace_probe/disable_trace_probe
1645 * lockless, but we can't race with this __init function.
1646 */
1647static int kprobe_register(struct trace_event_call *event,
1648 enum trace_reg type, void *data)
1649{
1650 struct trace_event_file *file = data;
1651
1652 switch (type) {
1653 case TRACE_REG_REGISTER:
1654 return enable_trace_kprobe(event, file);
1655 case TRACE_REG_UNREGISTER:
1656 return disable_trace_kprobe(event, file);
1657
1658#ifdef CONFIG_PERF_EVENTS
1659 case TRACE_REG_PERF_REGISTER:
1660 return enable_trace_kprobe(event, NULL);
1661 case TRACE_REG_PERF_UNREGISTER:
1662 return disable_trace_kprobe(event, NULL);
1663 case TRACE_REG_PERF_OPEN:
1664 case TRACE_REG_PERF_CLOSE:
1665 case TRACE_REG_PERF_ADD:
1666 case TRACE_REG_PERF_DEL:
1667 return 0;
1668#endif
1669 }
1670 return 0;
1671}
1672
1673static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1674{
1675 struct trace_kprobe *tk = container_of(kp, struct trace_kprobe, rp.kp);
1676 int ret = 0;
1677
1678 raw_cpu_inc(*tk->nhit);
1679
1680 if (trace_probe_test_flag(&tk->tp, TP_FLAG_TRACE))
1681 kprobe_trace_func(tk, regs);
1682#ifdef CONFIG_PERF_EVENTS
1683 if (trace_probe_test_flag(&tk->tp, TP_FLAG_PROFILE))
1684 ret = kprobe_perf_func(tk, regs);
1685#endif
1686 return ret;
1687}
1688NOKPROBE_SYMBOL(kprobe_dispatcher);
1689
1690static int
1691kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1692{
1693 struct kretprobe *rp = get_kretprobe(ri);
1694 struct trace_kprobe *tk;
1695
1696 /*
1697 * There is a small chance that get_kretprobe(ri) returns NULL when
1698 * the kretprobe is unregister on another CPU between kretprobe's
1699 * trampoline_handler and this function.
1700 */
1701 if (unlikely(!rp))
1702 return 0;
1703
1704 tk = container_of(rp, struct trace_kprobe, rp);
1705 raw_cpu_inc(*tk->nhit);
1706
1707 if (trace_probe_test_flag(&tk->tp, TP_FLAG_TRACE))
1708 kretprobe_trace_func(tk, ri, regs);
1709#ifdef CONFIG_PERF_EVENTS
1710 if (trace_probe_test_flag(&tk->tp, TP_FLAG_PROFILE))
1711 kretprobe_perf_func(tk, ri, regs);
1712#endif
1713 return 0; /* We don't tweak kernel, so just return 0 */
1714}
1715NOKPROBE_SYMBOL(kretprobe_dispatcher);
1716
1717static struct trace_event_functions kretprobe_funcs = {
1718 .trace = print_kretprobe_event
1719};
1720
1721static struct trace_event_functions kprobe_funcs = {
1722 .trace = print_kprobe_event
1723};
1724
1725static struct trace_event_fields kretprobe_fields_array[] = {
1726 { .type = TRACE_FUNCTION_TYPE,
1727 .define_fields = kretprobe_event_define_fields },
1728 {}
1729};
1730
1731static struct trace_event_fields kprobe_fields_array[] = {
1732 { .type = TRACE_FUNCTION_TYPE,
1733 .define_fields = kprobe_event_define_fields },
1734 {}
1735};
1736
1737static inline void init_trace_event_call(struct trace_kprobe *tk)
1738{
1739 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1740
1741 if (trace_kprobe_is_return(tk)) {
1742 call->event.funcs = &kretprobe_funcs;
1743 call->class->fields_array = kretprobe_fields_array;
1744 } else {
1745 call->event.funcs = &kprobe_funcs;
1746 call->class->fields_array = kprobe_fields_array;
1747 }
1748
1749 call->flags = TRACE_EVENT_FL_KPROBE;
1750 call->class->reg = kprobe_register;
1751}
1752
1753static int register_kprobe_event(struct trace_kprobe *tk)
1754{
1755 init_trace_event_call(tk);
1756
1757 return trace_probe_register_event_call(&tk->tp);
1758}
1759
1760static int unregister_kprobe_event(struct trace_kprobe *tk)
1761{
1762 return trace_probe_unregister_event_call(&tk->tp);
1763}
1764
1765#ifdef CONFIG_PERF_EVENTS
1766
1767/* create a trace_kprobe, but don't add it to global lists */
1768struct trace_event_call *
1769create_local_trace_kprobe(char *func, void *addr, unsigned long offs,
1770 bool is_return)
1771{
1772 enum probe_print_type ptype;
1773 struct trace_kprobe *tk;
1774 int ret;
1775 char *event;
1776
1777 if (func) {
1778 unsigned int count;
1779
1780 count = number_of_same_symbols(func);
1781 if (count > 1)
1782 /*
1783 * Users should use addr to remove the ambiguity of
1784 * using func only.
1785 */
1786 return ERR_PTR(-EADDRNOTAVAIL);
1787 else if (count == 0)
1788 /*
1789 * We can return ENOENT earlier than when register the
1790 * kprobe.
1791 */
1792 return ERR_PTR(-ENOENT);
1793 }
1794
1795 /*
1796 * local trace_kprobes are not added to dyn_event, so they are never
1797 * searched in find_trace_kprobe(). Therefore, there is no concern of
1798 * duplicated name here.
1799 */
1800 event = func ? func : "DUMMY_EVENT";
1801
1802 tk = alloc_trace_kprobe(KPROBE_EVENT_SYSTEM, event, (void *)addr, func,
1803 offs, 0 /* maxactive */, 0 /* nargs */,
1804 is_return);
1805
1806 if (IS_ERR(tk)) {
1807 pr_info("Failed to allocate trace_probe.(%d)\n",
1808 (int)PTR_ERR(tk));
1809 return ERR_CAST(tk);
1810 }
1811
1812 init_trace_event_call(tk);
1813
1814 ptype = trace_kprobe_is_return(tk) ?
1815 PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL;
1816 if (traceprobe_set_print_fmt(&tk->tp, ptype) < 0) {
1817 ret = -ENOMEM;
1818 goto error;
1819 }
1820
1821 ret = __register_trace_kprobe(tk);
1822 if (ret < 0)
1823 goto error;
1824
1825 return trace_probe_event_call(&tk->tp);
1826error:
1827 free_trace_kprobe(tk);
1828 return ERR_PTR(ret);
1829}
1830
1831void destroy_local_trace_kprobe(struct trace_event_call *event_call)
1832{
1833 struct trace_kprobe *tk;
1834
1835 tk = trace_kprobe_primary_from_call(event_call);
1836 if (unlikely(!tk))
1837 return;
1838
1839 if (trace_probe_is_enabled(&tk->tp)) {
1840 WARN_ON(1);
1841 return;
1842 }
1843
1844 __unregister_trace_kprobe(tk);
1845
1846 free_trace_kprobe(tk);
1847}
1848#endif /* CONFIG_PERF_EVENTS */
1849
1850static __init void enable_boot_kprobe_events(void)
1851{
1852 struct trace_array *tr = top_trace_array();
1853 struct trace_event_file *file;
1854 struct trace_kprobe *tk;
1855 struct dyn_event *pos;
1856
1857 mutex_lock(&event_mutex);
1858 for_each_trace_kprobe(tk, pos) {
1859 list_for_each_entry(file, &tr->events, list)
1860 if (file->event_call == trace_probe_event_call(&tk->tp))
1861 trace_event_enable_disable(file, 1, 0);
1862 }
1863 mutex_unlock(&event_mutex);
1864}
1865
1866static __init void setup_boot_kprobe_events(void)
1867{
1868 char *p, *cmd = kprobe_boot_events_buf;
1869 int ret;
1870
1871 strreplace(kprobe_boot_events_buf, ',', ' ');
1872
1873 while (cmd && *cmd != '\0') {
1874 p = strchr(cmd, ';');
1875 if (p)
1876 *p++ = '\0';
1877
1878 ret = create_or_delete_trace_kprobe(cmd);
1879 if (ret)
1880 pr_warn("Failed to add event(%d): %s\n", ret, cmd);
1881
1882 cmd = p;
1883 }
1884
1885 enable_boot_kprobe_events();
1886}
1887
1888/*
1889 * Register dynevent at core_initcall. This allows kernel to setup kprobe
1890 * events in postcore_initcall without tracefs.
1891 */
1892static __init int init_kprobe_trace_early(void)
1893{
1894 int ret;
1895
1896 ret = dyn_event_register(&trace_kprobe_ops);
1897 if (ret)
1898 return ret;
1899
1900 if (register_module_notifier(&trace_kprobe_module_nb))
1901 return -EINVAL;
1902
1903 return 0;
1904}
1905core_initcall(init_kprobe_trace_early);
1906
1907/* Make a tracefs interface for controlling probe points */
1908static __init int init_kprobe_trace(void)
1909{
1910 int ret;
1911
1912 ret = tracing_init_dentry();
1913 if (ret)
1914 return 0;
1915
1916 /* Event list interface */
1917 trace_create_file("kprobe_events", TRACE_MODE_WRITE,
1918 NULL, NULL, &kprobe_events_ops);
1919
1920 /* Profile interface */
1921 trace_create_file("kprobe_profile", TRACE_MODE_READ,
1922 NULL, NULL, &kprobe_profile_ops);
1923
1924 setup_boot_kprobe_events();
1925
1926 return 0;
1927}
1928fs_initcall(init_kprobe_trace);
1929
1930
1931#ifdef CONFIG_FTRACE_STARTUP_TEST
1932static __init struct trace_event_file *
1933find_trace_probe_file(struct trace_kprobe *tk, struct trace_array *tr)
1934{
1935 struct trace_event_file *file;
1936
1937 list_for_each_entry(file, &tr->events, list)
1938 if (file->event_call == trace_probe_event_call(&tk->tp))
1939 return file;
1940
1941 return NULL;
1942}
1943
1944/*
1945 * Nobody but us can call enable_trace_kprobe/disable_trace_kprobe at this
1946 * stage, we can do this lockless.
1947 */
1948static __init int kprobe_trace_self_tests_init(void)
1949{
1950 int ret, warn = 0;
1951 int (*target)(int, int, int, int, int, int);
1952 struct trace_kprobe *tk;
1953 struct trace_event_file *file;
1954
1955 if (tracing_is_disabled())
1956 return -ENODEV;
1957
1958 if (tracing_selftest_disabled)
1959 return 0;
1960
1961 target = kprobe_trace_selftest_target;
1962
1963 pr_info("Testing kprobe tracing: ");
1964
1965 ret = create_or_delete_trace_kprobe("p:testprobe kprobe_trace_selftest_target $stack $stack0 +0($stack)");
1966 if (WARN_ON_ONCE(ret)) {
1967 pr_warn("error on probing function entry.\n");
1968 warn++;
1969 } else {
1970 /* Enable trace point */
1971 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1972 if (WARN_ON_ONCE(tk == NULL)) {
1973 pr_warn("error on getting new probe.\n");
1974 warn++;
1975 } else {
1976 file = find_trace_probe_file(tk, top_trace_array());
1977 if (WARN_ON_ONCE(file == NULL)) {
1978 pr_warn("error on getting probe file.\n");
1979 warn++;
1980 } else
1981 enable_trace_kprobe(
1982 trace_probe_event_call(&tk->tp), file);
1983 }
1984 }
1985
1986 ret = create_or_delete_trace_kprobe("r:testprobe2 kprobe_trace_selftest_target $retval");
1987 if (WARN_ON_ONCE(ret)) {
1988 pr_warn("error on probing function return.\n");
1989 warn++;
1990 } else {
1991 /* Enable trace point */
1992 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1993 if (WARN_ON_ONCE(tk == NULL)) {
1994 pr_warn("error on getting 2nd new probe.\n");
1995 warn++;
1996 } else {
1997 file = find_trace_probe_file(tk, top_trace_array());
1998 if (WARN_ON_ONCE(file == NULL)) {
1999 pr_warn("error on getting probe file.\n");
2000 warn++;
2001 } else
2002 enable_trace_kprobe(
2003 trace_probe_event_call(&tk->tp), file);
2004 }
2005 }
2006
2007 if (warn)
2008 goto end;
2009
2010 ret = target(1, 2, 3, 4, 5, 6);
2011
2012 /*
2013 * Not expecting an error here, the check is only to prevent the
2014 * optimizer from removing the call to target() as otherwise there
2015 * are no side-effects and the call is never performed.
2016 */
2017 if (ret != 21)
2018 warn++;
2019
2020 /* Disable trace points before removing it */
2021 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
2022 if (WARN_ON_ONCE(tk == NULL)) {
2023 pr_warn("error on getting test probe.\n");
2024 warn++;
2025 } else {
2026 if (trace_kprobe_nhit(tk) != 1) {
2027 pr_warn("incorrect number of testprobe hits\n");
2028 warn++;
2029 }
2030
2031 file = find_trace_probe_file(tk, top_trace_array());
2032 if (WARN_ON_ONCE(file == NULL)) {
2033 pr_warn("error on getting probe file.\n");
2034 warn++;
2035 } else
2036 disable_trace_kprobe(
2037 trace_probe_event_call(&tk->tp), file);
2038 }
2039
2040 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
2041 if (WARN_ON_ONCE(tk == NULL)) {
2042 pr_warn("error on getting 2nd test probe.\n");
2043 warn++;
2044 } else {
2045 if (trace_kprobe_nhit(tk) != 1) {
2046 pr_warn("incorrect number of testprobe2 hits\n");
2047 warn++;
2048 }
2049
2050 file = find_trace_probe_file(tk, top_trace_array());
2051 if (WARN_ON_ONCE(file == NULL)) {
2052 pr_warn("error on getting probe file.\n");
2053 warn++;
2054 } else
2055 disable_trace_kprobe(
2056 trace_probe_event_call(&tk->tp), file);
2057 }
2058
2059 ret = create_or_delete_trace_kprobe("-:testprobe");
2060 if (WARN_ON_ONCE(ret)) {
2061 pr_warn("error on deleting a probe.\n");
2062 warn++;
2063 }
2064
2065 ret = create_or_delete_trace_kprobe("-:testprobe2");
2066 if (WARN_ON_ONCE(ret)) {
2067 pr_warn("error on deleting a probe.\n");
2068 warn++;
2069 }
2070
2071end:
2072 ret = dyn_events_release_all(&trace_kprobe_ops);
2073 if (WARN_ON_ONCE(ret)) {
2074 pr_warn("error on cleaning up probes.\n");
2075 warn++;
2076 }
2077 /*
2078 * Wait for the optimizer work to finish. Otherwise it might fiddle
2079 * with probes in already freed __init text.
2080 */
2081 wait_for_kprobe_optimizer();
2082 if (warn)
2083 pr_cont("NG: Some tests are failed. Please check them.\n");
2084 else
2085 pr_cont("OK\n");
2086 return 0;
2087}
2088
2089late_initcall(kprobe_trace_self_tests_init);
2090
2091#endif
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Kprobes-based tracing events
4 *
5 * Created by Masami Hiramatsu <mhiramat@redhat.com>
6 *
7 */
8#define pr_fmt(fmt) "trace_kprobe: " fmt
9
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>
15
16#include <asm/setup.h> /* for COMMAND_LINE_SIZE */
17
18#include "trace_dynevent.h"
19#include "trace_kprobe_selftest.h"
20#include "trace_probe.h"
21#include "trace_probe_tmpl.h"
22
23#define KPROBE_EVENT_SYSTEM "kprobes"
24#define KRETPROBE_MAXACTIVE_MAX 4096
25
26/* Kprobe early definition from command line */
27static char kprobe_boot_events_buf[COMMAND_LINE_SIZE] __initdata;
28static bool kprobe_boot_events_enabled __initdata;
29
30static int __init set_kprobe_boot_events(char *str)
31{
32 strlcpy(kprobe_boot_events_buf, str, COMMAND_LINE_SIZE);
33 return 0;
34}
35__setup("kprobe_event=", set_kprobe_boot_events);
36
37static int trace_kprobe_create(int argc, const char **argv);
38static int trace_kprobe_show(struct seq_file *m, struct dyn_event *ev);
39static int trace_kprobe_release(struct dyn_event *ev);
40static bool trace_kprobe_is_busy(struct dyn_event *ev);
41static bool trace_kprobe_match(const char *system, const char *event,
42 int argc, const char **argv, struct dyn_event *ev);
43
44static 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,
50};
51
52/*
53 * Kprobe event core functions
54 */
55struct trace_kprobe {
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;
61};
62
63static bool is_trace_kprobe(struct dyn_event *ev)
64{
65 return ev->ops == &trace_kprobe_ops;
66}
67
68static struct trace_kprobe *to_trace_kprobe(struct dyn_event *ev)
69{
70 return container_of(ev, struct trace_kprobe, devent);
71}
72
73/**
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
77 */
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)))
81
82#define SIZEOF_TRACE_KPROBE(n) \
83 (offsetof(struct trace_kprobe, tp.args) + \
84 (sizeof(struct probe_arg) * (n)))
85
86static nokprobe_inline bool trace_kprobe_is_return(struct trace_kprobe *tk)
87{
88 return tk->rp.handler != NULL;
89}
90
91static nokprobe_inline const char *trace_kprobe_symbol(struct trace_kprobe *tk)
92{
93 return tk->symbol ? tk->symbol : "unknown";
94}
95
96static nokprobe_inline unsigned long trace_kprobe_offset(struct trace_kprobe *tk)
97{
98 return tk->rp.kp.offset;
99}
100
101static nokprobe_inline bool trace_kprobe_has_gone(struct trace_kprobe *tk)
102{
103 return !!(kprobe_gone(&tk->rp.kp));
104}
105
106static nokprobe_inline bool trace_kprobe_within_module(struct trace_kprobe *tk,
107 struct module *mod)
108{
109 int len = strlen(mod->name);
110 const char *name = trace_kprobe_symbol(tk);
111 return strncmp(mod->name, name, len) == 0 && name[len] == ':';
112}
113
114static nokprobe_inline bool trace_kprobe_module_exist(struct trace_kprobe *tk)
115{
116 char *p;
117 bool ret;
118
119 if (!tk->symbol)
120 return false;
121 p = strchr(tk->symbol, ':');
122 if (!p)
123 return true;
124 *p = '\0';
125 mutex_lock(&module_mutex);
126 ret = !!find_module(tk->symbol);
127 mutex_unlock(&module_mutex);
128 *p = ':';
129
130 return ret;
131}
132
133static bool trace_kprobe_is_busy(struct dyn_event *ev)
134{
135 struct trace_kprobe *tk = to_trace_kprobe(ev);
136
137 return trace_probe_is_enabled(&tk->tp);
138}
139
140static bool trace_kprobe_match_command_head(struct trace_kprobe *tk,
141 int argc, const char **argv)
142{
143 char buf[MAX_ARGSTR_LEN + 1];
144
145 if (!argc)
146 return true;
147
148 if (!tk->symbol)
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);
153 else
154 snprintf(buf, sizeof(buf), "%s", trace_kprobe_symbol(tk));
155 if (strcmp(buf, argv[0]))
156 return false;
157 argc--; argv++;
158
159 return trace_probe_match_command_args(&tk->tp, argc, argv);
160}
161
162static bool trace_kprobe_match(const char *system, const char *event,
163 int argc, const char **argv, struct dyn_event *ev)
164{
165 struct trace_kprobe *tk = to_trace_kprobe(ev);
166
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);
170}
171
172static nokprobe_inline unsigned long trace_kprobe_nhit(struct trace_kprobe *tk)
173{
174 unsigned long nhit = 0;
175 int cpu;
176
177 for_each_possible_cpu(cpu)
178 nhit += *per_cpu_ptr(tk->nhit, cpu);
179
180 return nhit;
181}
182
183static nokprobe_inline bool trace_kprobe_is_registered(struct trace_kprobe *tk)
184{
185 return !(list_empty(&tk->rp.kp.list) &&
186 hlist_unhashed(&tk->rp.kp.hlist));
187}
188
189/* Return 0 if it fails to find the symbol address */
190static nokprobe_inline
191unsigned long trace_kprobe_address(struct trace_kprobe *tk)
192{
193 unsigned long addr;
194
195 if (tk->symbol) {
196 addr = (unsigned long)
197 kallsyms_lookup_name(trace_kprobe_symbol(tk));
198 if (addr)
199 addr += tk->rp.kp.offset;
200 } else {
201 addr = (unsigned long)tk->rp.kp.addr;
202 }
203 return addr;
204}
205
206static nokprobe_inline struct trace_kprobe *
207trace_kprobe_primary_from_call(struct trace_event_call *call)
208{
209 struct trace_probe *tp;
210
211 tp = trace_probe_primary_from_call(call);
212 if (WARN_ON_ONCE(!tp))
213 return NULL;
214
215 return container_of(tp, struct trace_kprobe, tp);
216}
217
218bool trace_kprobe_on_func_entry(struct trace_event_call *call)
219{
220 struct trace_kprobe *tk = trace_kprobe_primary_from_call(call);
221
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;
225}
226
227bool trace_kprobe_error_injectable(struct trace_event_call *call)
228{
229 struct trace_kprobe *tk = trace_kprobe_primary_from_call(call);
230
231 return tk ? within_error_injection_list(trace_kprobe_address(tk)) :
232 false;
233}
234
235static int register_kprobe_event(struct trace_kprobe *tk);
236static int unregister_kprobe_event(struct trace_kprobe *tk);
237
238static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
239static int kretprobe_dispatcher(struct kretprobe_instance *ri,
240 struct pt_regs *regs);
241
242static void free_trace_kprobe(struct trace_kprobe *tk)
243{
244 if (tk) {
245 trace_probe_cleanup(&tk->tp);
246 kfree(tk->symbol);
247 free_percpu(tk->nhit);
248 kfree(tk);
249 }
250}
251
252/*
253 * Allocate new trace_probe and initialize it (including kprobes).
254 */
255static struct trace_kprobe *alloc_trace_kprobe(const char *group,
256 const char *event,
257 void *addr,
258 const char *symbol,
259 unsigned long offs,
260 int maxactive,
261 int nargs, bool is_return)
262{
263 struct trace_kprobe *tk;
264 int ret = -ENOMEM;
265
266 tk = kzalloc(SIZEOF_TRACE_KPROBE(nargs), GFP_KERNEL);
267 if (!tk)
268 return ERR_PTR(ret);
269
270 tk->nhit = alloc_percpu(unsigned long);
271 if (!tk->nhit)
272 goto error;
273
274 if (symbol) {
275 tk->symbol = kstrdup(symbol, GFP_KERNEL);
276 if (!tk->symbol)
277 goto error;
278 tk->rp.kp.symbol_name = tk->symbol;
279 tk->rp.kp.offset = offs;
280 } else
281 tk->rp.kp.addr = addr;
282
283 if (is_return)
284 tk->rp.handler = kretprobe_dispatcher;
285 else
286 tk->rp.kp.pre_handler = kprobe_dispatcher;
287
288 tk->rp.maxactive = maxactive;
289 INIT_HLIST_NODE(&tk->rp.kp.hlist);
290 INIT_LIST_HEAD(&tk->rp.kp.list);
291
292 ret = trace_probe_init(&tk->tp, event, group, false);
293 if (ret < 0)
294 goto error;
295
296 dyn_event_init(&tk->devent, &trace_kprobe_ops);
297 return tk;
298error:
299 free_trace_kprobe(tk);
300 return ERR_PTR(ret);
301}
302
303static struct trace_kprobe *find_trace_kprobe(const char *event,
304 const char *group)
305{
306 struct dyn_event *pos;
307 struct trace_kprobe *tk;
308
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)
312 return tk;
313 return NULL;
314}
315
316static inline int __enable_trace_kprobe(struct trace_kprobe *tk)
317{
318 int ret = 0;
319
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);
323 else
324 ret = enable_kprobe(&tk->rp.kp);
325 }
326
327 return ret;
328}
329
330static void __disable_trace_kprobe(struct trace_probe *tp)
331{
332 struct trace_probe *pos;
333 struct trace_kprobe *tk;
334
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))
338 continue;
339 if (trace_kprobe_is_return(tk))
340 disable_kretprobe(&tk->rp);
341 else
342 disable_kprobe(&tk->rp.kp);
343 }
344}
345
346/*
347 * Enable trace_probe
348 * if the file is NULL, enable "perf" handler, or enable "trace" handler.
349 */
350static int enable_trace_kprobe(struct trace_event_call *call,
351 struct trace_event_file *file)
352{
353 struct trace_probe *pos, *tp;
354 struct trace_kprobe *tk;
355 bool enabled;
356 int ret = 0;
357
358 tp = trace_probe_primary_from_call(call);
359 if (WARN_ON_ONCE(!tp))
360 return -ENODEV;
361 enabled = trace_probe_is_enabled(tp);
362
363 /* This also changes "enabled" state */
364 if (file) {
365 ret = trace_probe_add_file(tp, file);
366 if (ret)
367 return ret;
368 } else
369 trace_probe_set_flag(tp, TP_FLAG_PROFILE);
370
371 if (enabled)
372 return 0;
373
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))
377 continue;
378 ret = __enable_trace_kprobe(tk);
379 if (ret)
380 break;
381 enabled = true;
382 }
383
384 if (ret) {
385 /* Failed to enable one of them. Roll back all */
386 if (enabled)
387 __disable_trace_kprobe(tp);
388 if (file)
389 trace_probe_remove_file(tp, file);
390 else
391 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
392 }
393
394 return ret;
395}
396
397/*
398 * Disable trace_probe
399 * if the file is NULL, disable "perf" handler, or disable "trace" handler.
400 */
401static int disable_trace_kprobe(struct trace_event_call *call,
402 struct trace_event_file *file)
403{
404 struct trace_probe *tp;
405
406 tp = trace_probe_primary_from_call(call);
407 if (WARN_ON_ONCE(!tp))
408 return -ENODEV;
409
410 if (file) {
411 if (!trace_probe_get_file_link(tp, file))
412 return -ENOENT;
413 if (!trace_probe_has_single_file(tp))
414 goto out;
415 trace_probe_clear_flag(tp, TP_FLAG_TRACE);
416 } else
417 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
418
419 if (!trace_probe_is_enabled(tp))
420 __disable_trace_kprobe(tp);
421
422 out:
423 if (file)
424 /*
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.
429 */
430 trace_probe_remove_file(tp, file);
431
432 return 0;
433}
434
435#if defined(CONFIG_KPROBES_ON_FTRACE) && \
436 !defined(CONFIG_KPROBE_EVENTS_ON_NOTRACE)
437static bool __within_notrace_func(unsigned long addr)
438{
439 unsigned long offset, size;
440
441 if (!addr || !kallsyms_lookup_size_offset(addr, &size, &offset))
442 return false;
443
444 /* Get the entry address of the target function */
445 addr -= offset;
446
447 /*
448 * Since ftrace_location_range() does inclusive range check, we need
449 * to subtract 1 byte from the end address.
450 */
451 return !ftrace_location_range(addr, addr + size - 1);
452}
453
454static bool within_notrace_func(struct trace_kprobe *tk)
455{
456 unsigned long addr = trace_kprobe_address(tk);
457 char symname[KSYM_NAME_LEN], *p;
458
459 if (!__within_notrace_func(addr))
460 return false;
461
462 /* Check if the address is on a suffixed-symbol */
463 if (!lookup_symbol_name(addr, symname)) {
464 p = strchr(symname, '.');
465 if (!p)
466 return true;
467 *p = '\0';
468 addr = (unsigned long)kprobe_lookup_name(symname, 0);
469 if (addr)
470 return __within_notrace_func(addr);
471 }
472
473 return true;
474}
475#else
476#define within_notrace_func(tk) (false)
477#endif
478
479/* Internal register function - just handle k*probes and flags */
480static int __register_trace_kprobe(struct trace_kprobe *tk)
481{
482 int i, ret;
483
484 ret = security_locked_down(LOCKDOWN_KPROBES);
485 if (ret)
486 return ret;
487
488 if (trace_kprobe_is_registered(tk))
489 return -EINVAL;
490
491 if (within_notrace_func(tk)) {
492 pr_warn("Could not probe notrace function %s\n",
493 trace_kprobe_symbol(tk));
494 return -EINVAL;
495 }
496
497 for (i = 0; i < tk->tp.nr_args; i++) {
498 ret = traceprobe_update_arg(&tk->tp.args[i]);
499 if (ret)
500 return ret;
501 }
502
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;
506 else
507 tk->rp.kp.flags |= KPROBE_FLAG_DISABLED;
508
509 if (trace_kprobe_is_return(tk))
510 ret = register_kretprobe(&tk->rp);
511 else
512 ret = register_kprobe(&tk->rp.kp);
513
514 return ret;
515}
516
517/* Internal unregister function - just handle k*probes and flags */
518static void __unregister_trace_kprobe(struct trace_kprobe *tk)
519{
520 if (trace_kprobe_is_registered(tk)) {
521 if (trace_kprobe_is_return(tk))
522 unregister_kretprobe(&tk->rp);
523 else
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;
530 }
531}
532
533/* Unregister a trace_probe and probe_event */
534static int unregister_trace_kprobe(struct trace_kprobe *tk)
535{
536 /* If other probes are on the event, just unregister kprobe */
537 if (trace_probe_has_sibling(&tk->tp))
538 goto unreg;
539
540 /* Enabled event can not be unregistered */
541 if (trace_probe_is_enabled(&tk->tp))
542 return -EBUSY;
543
544 /* Will fail if probe is being used by ftrace or perf */
545 if (unregister_kprobe_event(tk))
546 return -EBUSY;
547
548unreg:
549 __unregister_trace_kprobe(tk);
550 dyn_event_remove(&tk->devent);
551 trace_probe_unlink(&tk->tp);
552
553 return 0;
554}
555
556static bool trace_kprobe_has_same_kprobe(struct trace_kprobe *orig,
557 struct trace_kprobe *comp)
558{
559 struct trace_probe_event *tpe = orig->tp.event;
560 struct trace_probe *pos;
561 int i;
562
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))
568 continue;
569
570 /*
571 * trace_probe_compare_arg_type() ensured that nr_args and
572 * each argument name and type are same. Let's compare comm.
573 */
574 for (i = 0; i < orig->tp.nr_args; i++) {
575 if (strcmp(orig->tp.args[i].comm,
576 comp->tp.args[i].comm))
577 break;
578 }
579
580 if (i == orig->tp.nr_args)
581 return true;
582 }
583
584 return false;
585}
586
587static int append_trace_kprobe(struct trace_kprobe *tk, struct trace_kprobe *to)
588{
589 int ret;
590
591 ret = trace_probe_compare_arg_type(&tk->tp, &to->tp);
592 if (ret) {
593 /* Note that argument starts index = 2 */
594 trace_probe_log_set_index(ret + 1);
595 trace_probe_log_err(0, DIFF_ARG_TYPE);
596 return -EEXIST;
597 }
598 if (trace_kprobe_has_same_kprobe(to, tk)) {
599 trace_probe_log_set_index(0);
600 trace_probe_log_err(0, SAME_PROBE);
601 return -EEXIST;
602 }
603
604 /* Append to existing event */
605 ret = trace_probe_append(&tk->tp, &to->tp);
606 if (ret)
607 return ret;
608
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");
613 ret = 0;
614 }
615
616 if (ret)
617 trace_probe_unlink(&tk->tp);
618 else
619 dyn_event_add(&tk->devent);
620
621 return ret;
622}
623
624/* Register a trace_probe and probe_event */
625static int register_trace_kprobe(struct trace_kprobe *tk)
626{
627 struct trace_kprobe *old_tk;
628 int ret;
629
630 mutex_lock(&event_mutex);
631
632 old_tk = find_trace_kprobe(trace_probe_name(&tk->tp),
633 trace_probe_group_name(&tk->tp));
634 if (old_tk) {
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);
638 ret = -EEXIST;
639 } else {
640 ret = append_trace_kprobe(tk, old_tk);
641 }
642 goto end;
643 }
644
645 /* Register new event */
646 ret = register_kprobe_event(tk);
647 if (ret) {
648 pr_warn("Failed to register probe event(%d)\n", ret);
649 goto end;
650 }
651
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");
656 ret = 0;
657 }
658
659 if (ret < 0)
660 unregister_kprobe_event(tk);
661 else
662 dyn_event_add(&tk->devent);
663
664end:
665 mutex_unlock(&event_mutex);
666 return ret;
667}
668
669/* Module notifier call back, checking event on the module */
670static int trace_kprobe_module_callback(struct notifier_block *nb,
671 unsigned long val, void *data)
672{
673 struct module *mod = data;
674 struct dyn_event *pos;
675 struct trace_kprobe *tk;
676 int ret;
677
678 if (val != MODULE_STATE_COMING)
679 return NOTIFY_DONE;
680
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);
688 if (ret)
689 pr_warn("Failed to re-register probe %s on %s: %d\n",
690 trace_probe_name(&tk->tp),
691 mod->name, ret);
692 }
693 }
694 mutex_unlock(&event_mutex);
695
696 return NOTIFY_DONE;
697}
698
699static struct notifier_block trace_kprobe_module_nb = {
700 .notifier_call = trace_kprobe_module_callback,
701 .priority = 1 /* Invoked after kprobe module callback */
702};
703
704/* Convert certain expected symbols into '_' when generating event names */
705static inline void sanitize_event_name(char *name)
706{
707 while (*name++ != '\0')
708 if (*name == ':' || *name == '.')
709 *name = '_';
710}
711
712static int trace_kprobe_create(int argc, const char *argv[])
713{
714 /*
715 * Argument syntax:
716 * - Add kprobe:
717 * p[:[GRP/]EVENT] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
718 * - Add kretprobe:
719 * r[MAXACTIVE][:[GRP/]EVENT] [MOD:]KSYM[+0] [FETCHARGS]
720 * Fetch args:
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.
732 * Type of args:
733 * FETCHARG:TYPE : use TYPE instead of unsigned long.
734 */
735 struct trace_kprobe *tk = NULL;
736 int i, len, ret = 0;
737 bool is_return = false;
738 char *symbol = NULL, *tmp = NULL;
739 const char *event = NULL, *group = KPROBE_EVENT_SYSTEM;
740 int maxactive = 0;
741 long offset = 0;
742 void *addr = NULL;
743 char buf[MAX_EVENT_NAME_LEN];
744 unsigned int flags = TPARG_FL_KERNEL;
745
746 switch (argv[0][0]) {
747 case 'r':
748 is_return = true;
749 flags |= TPARG_FL_RETURN;
750 break;
751 case 'p':
752 break;
753 default:
754 return -ECANCELED;
755 }
756 if (argc < 2)
757 return -ECANCELED;
758
759 trace_probe_log_init("trace_kprobe", argc, argv);
760
761 event = strchr(&argv[0][1], ':');
762 if (event)
763 event++;
764
765 if (isdigit(argv[0][1])) {
766 if (!is_return) {
767 trace_probe_log_err(1, MAXACT_NO_KPROBE);
768 goto parse_error;
769 }
770 if (event)
771 len = event - &argv[0][1] - 1;
772 else
773 len = strlen(&argv[0][1]);
774 if (len > MAX_EVENT_NAME_LEN - 1) {
775 trace_probe_log_err(1, BAD_MAXACT);
776 goto parse_error;
777 }
778 memcpy(buf, &argv[0][1], len);
779 buf[len] = '\0';
780 ret = kstrtouint(buf, 0, &maxactive);
781 if (ret || !maxactive) {
782 trace_probe_log_err(1, BAD_MAXACT);
783 goto parse_error;
784 }
785 /* kretprobes instances are iterated over via a list. The
786 * maximum should stay reasonable.
787 */
788 if (maxactive > KRETPROBE_MAXACTIVE_MAX) {
789 trace_probe_log_err(1, MAXACT_TOO_BIG);
790 goto parse_error;
791 }
792 }
793
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], ':')) {
800 ret = -ECANCELED;
801 goto error;
802 }
803 /* a symbol specified */
804 symbol = kstrdup(argv[1], GFP_KERNEL);
805 if (!symbol)
806 return -ENOMEM;
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);
811 goto parse_error;
812 }
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);
817 goto parse_error;
818 }
819 }
820
821 trace_probe_log_set_index(0);
822 if (event) {
823 ret = traceprobe_parse_event_name(&event, &group, buf,
824 event - argv[0]);
825 if (ret)
826 goto parse_error;
827 } else {
828 /* Make a new event name */
829 if (symbol)
830 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
831 is_return ? 'r' : 'p', symbol, offset);
832 else
833 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
834 is_return ? 'r' : 'p', addr);
835 sanitize_event_name(buf);
836 event = buf;
837 }
838
839 /* setup a probe */
840 tk = alloc_trace_kprobe(group, event, addr, symbol, offset, maxactive,
841 argc - 2, is_return);
842 if (IS_ERR(tk)) {
843 ret = PTR_ERR(tk);
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 */
847 }
848 argc -= 2; argv += 2;
849
850 /* parse arguments */
851 for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
852 tmp = kstrdup(argv[i], GFP_KERNEL);
853 if (!tmp) {
854 ret = -ENOMEM;
855 goto error;
856 }
857
858 trace_probe_log_set_index(i + 2);
859 ret = traceprobe_parse_probe_arg(&tk->tp, i, tmp, flags);
860 kfree(tmp);
861 if (ret)
862 goto error; /* This can be -ENOMEM */
863 }
864
865 ret = traceprobe_set_print_fmt(&tk->tp, is_return);
866 if (ret < 0)
867 goto error;
868
869 ret = register_trace_kprobe(tk);
870 if (ret) {
871 trace_probe_log_set_index(1);
872 if (ret == -EILSEQ)
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);
878 goto error;
879 }
880
881out:
882 trace_probe_log_clear();
883 kfree(symbol);
884 return ret;
885
886parse_error:
887 ret = -EINVAL;
888error:
889 free_trace_kprobe(tk);
890 goto out;
891}
892
893static int create_or_delete_trace_kprobe(int argc, char **argv)
894{
895 int ret;
896
897 if (argv[0][0] == '-')
898 return dyn_event_release(argc, argv, &trace_kprobe_ops);
899
900 ret = trace_kprobe_create(argc, (const char **)argv);
901 return ret == -ECANCELED ? -EINVAL : ret;
902}
903
904static int trace_kprobe_run_command(struct dynevent_cmd *cmd)
905{
906 return trace_run_command(cmd->seq.buffer, create_or_delete_trace_kprobe);
907}
908
909/**
910 * kprobe_event_cmd_init - Initialize a kprobe event command object
911 * @cmd: A pointer to the dynevent_cmd struct representing the new event
912 * @buf: A pointer to the buffer used to build the command
913 * @maxlen: The length of the buffer passed in @buf
914 *
915 * Initialize a synthetic event command object. Use this before
916 * calling any of the other kprobe_event functions.
917 */
918void kprobe_event_cmd_init(struct dynevent_cmd *cmd, char *buf, int maxlen)
919{
920 dynevent_cmd_init(cmd, buf, maxlen, DYNEVENT_TYPE_KPROBE,
921 trace_kprobe_run_command);
922}
923EXPORT_SYMBOL_GPL(kprobe_event_cmd_init);
924
925/**
926 * __kprobe_event_gen_cmd_start - Generate a kprobe event command from arg list
927 * @cmd: A pointer to the dynevent_cmd struct representing the new event
928 * @name: The name of the kprobe event
929 * @loc: The location of the kprobe event
930 * @kretprobe: Is this a return probe?
931 * @args: Variable number of arg (pairs), one pair for each field
932 *
933 * NOTE: Users normally won't want to call this function directly, but
934 * rather use the kprobe_event_gen_cmd_start() wrapper, which automatically
935 * adds a NULL to the end of the arg list. If this function is used
936 * directly, make sure the last arg in the variable arg list is NULL.
937 *
938 * Generate a kprobe event command to be executed by
939 * kprobe_event_gen_cmd_end(). This function can be used to generate the
940 * complete command or only the first part of it; in the latter case,
941 * kprobe_event_add_fields() can be used to add more fields following this.
942 *
943 * Unlikely the synth_event_gen_cmd_start(), @loc must be specified. This
944 * returns -EINVAL if @loc == NULL.
945 *
946 * Return: 0 if successful, error otherwise.
947 */
948int __kprobe_event_gen_cmd_start(struct dynevent_cmd *cmd, bool kretprobe,
949 const char *name, const char *loc, ...)
950{
951 char buf[MAX_EVENT_NAME_LEN];
952 struct dynevent_arg arg;
953 va_list args;
954 int ret;
955
956 if (cmd->type != DYNEVENT_TYPE_KPROBE)
957 return -EINVAL;
958
959 if (!loc)
960 return -EINVAL;
961
962 if (kretprobe)
963 snprintf(buf, MAX_EVENT_NAME_LEN, "r:kprobes/%s", name);
964 else
965 snprintf(buf, MAX_EVENT_NAME_LEN, "p:kprobes/%s", name);
966
967 ret = dynevent_str_add(cmd, buf);
968 if (ret)
969 return ret;
970
971 dynevent_arg_init(&arg, 0);
972 arg.str = loc;
973 ret = dynevent_arg_add(cmd, &arg, NULL);
974 if (ret)
975 return ret;
976
977 va_start(args, loc);
978 for (;;) {
979 const char *field;
980
981 field = va_arg(args, const char *);
982 if (!field)
983 break;
984
985 if (++cmd->n_fields > MAX_TRACE_ARGS) {
986 ret = -EINVAL;
987 break;
988 }
989
990 arg.str = field;
991 ret = dynevent_arg_add(cmd, &arg, NULL);
992 if (ret)
993 break;
994 }
995 va_end(args);
996
997 return ret;
998}
999EXPORT_SYMBOL_GPL(__kprobe_event_gen_cmd_start);
1000
1001/**
1002 * __kprobe_event_add_fields - Add probe fields to a kprobe command from arg list
1003 * @cmd: A pointer to the dynevent_cmd struct representing the new event
1004 * @args: Variable number of arg (pairs), one pair for each field
1005 *
1006 * NOTE: Users normally won't want to call this function directly, but
1007 * rather use the kprobe_event_add_fields() wrapper, which
1008 * automatically adds a NULL to the end of the arg list. If this
1009 * function is used directly, make sure the last arg in the variable
1010 * arg list is NULL.
1011 *
1012 * Add probe fields to an existing kprobe command using a variable
1013 * list of args. Fields are added in the same order they're listed.
1014 *
1015 * Return: 0 if successful, error otherwise.
1016 */
1017int __kprobe_event_add_fields(struct dynevent_cmd *cmd, ...)
1018{
1019 struct dynevent_arg arg;
1020 va_list args;
1021 int ret = 0;
1022
1023 if (cmd->type != DYNEVENT_TYPE_KPROBE)
1024 return -EINVAL;
1025
1026 dynevent_arg_init(&arg, 0);
1027
1028 va_start(args, cmd);
1029 for (;;) {
1030 const char *field;
1031
1032 field = va_arg(args, const char *);
1033 if (!field)
1034 break;
1035
1036 if (++cmd->n_fields > MAX_TRACE_ARGS) {
1037 ret = -EINVAL;
1038 break;
1039 }
1040
1041 arg.str = field;
1042 ret = dynevent_arg_add(cmd, &arg, NULL);
1043 if (ret)
1044 break;
1045 }
1046 va_end(args);
1047
1048 return ret;
1049}
1050EXPORT_SYMBOL_GPL(__kprobe_event_add_fields);
1051
1052/**
1053 * kprobe_event_delete - Delete a kprobe event
1054 * @name: The name of the kprobe event to delete
1055 *
1056 * Delete a kprobe event with the give @name from kernel code rather
1057 * than directly from the command line.
1058 *
1059 * Return: 0 if successful, error otherwise.
1060 */
1061int kprobe_event_delete(const char *name)
1062{
1063 char buf[MAX_EVENT_NAME_LEN];
1064
1065 snprintf(buf, MAX_EVENT_NAME_LEN, "-:%s", name);
1066
1067 return trace_run_command(buf, create_or_delete_trace_kprobe);
1068}
1069EXPORT_SYMBOL_GPL(kprobe_event_delete);
1070
1071static int trace_kprobe_release(struct dyn_event *ev)
1072{
1073 struct trace_kprobe *tk = to_trace_kprobe(ev);
1074 int ret = unregister_trace_kprobe(tk);
1075
1076 if (!ret)
1077 free_trace_kprobe(tk);
1078 return ret;
1079}
1080
1081static int trace_kprobe_show(struct seq_file *m, struct dyn_event *ev)
1082{
1083 struct trace_kprobe *tk = to_trace_kprobe(ev);
1084 int i;
1085
1086 seq_putc(m, trace_kprobe_is_return(tk) ? 'r' : 'p');
1087 if (trace_kprobe_is_return(tk) && tk->rp.maxactive)
1088 seq_printf(m, "%d", tk->rp.maxactive);
1089 seq_printf(m, ":%s/%s", trace_probe_group_name(&tk->tp),
1090 trace_probe_name(&tk->tp));
1091
1092 if (!tk->symbol)
1093 seq_printf(m, " 0x%p", tk->rp.kp.addr);
1094 else if (tk->rp.kp.offset)
1095 seq_printf(m, " %s+%u", trace_kprobe_symbol(tk),
1096 tk->rp.kp.offset);
1097 else
1098 seq_printf(m, " %s", trace_kprobe_symbol(tk));
1099
1100 for (i = 0; i < tk->tp.nr_args; i++)
1101 seq_printf(m, " %s=%s", tk->tp.args[i].name, tk->tp.args[i].comm);
1102 seq_putc(m, '\n');
1103
1104 return 0;
1105}
1106
1107static int probes_seq_show(struct seq_file *m, void *v)
1108{
1109 struct dyn_event *ev = v;
1110
1111 if (!is_trace_kprobe(ev))
1112 return 0;
1113
1114 return trace_kprobe_show(m, ev);
1115}
1116
1117static const struct seq_operations probes_seq_op = {
1118 .start = dyn_event_seq_start,
1119 .next = dyn_event_seq_next,
1120 .stop = dyn_event_seq_stop,
1121 .show = probes_seq_show
1122};
1123
1124static int probes_open(struct inode *inode, struct file *file)
1125{
1126 int ret;
1127
1128 ret = security_locked_down(LOCKDOWN_TRACEFS);
1129 if (ret)
1130 return ret;
1131
1132 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
1133 ret = dyn_events_release_all(&trace_kprobe_ops);
1134 if (ret < 0)
1135 return ret;
1136 }
1137
1138 return seq_open(file, &probes_seq_op);
1139}
1140
1141static ssize_t probes_write(struct file *file, const char __user *buffer,
1142 size_t count, loff_t *ppos)
1143{
1144 return trace_parse_run_command(file, buffer, count, ppos,
1145 create_or_delete_trace_kprobe);
1146}
1147
1148static const struct file_operations kprobe_events_ops = {
1149 .owner = THIS_MODULE,
1150 .open = probes_open,
1151 .read = seq_read,
1152 .llseek = seq_lseek,
1153 .release = seq_release,
1154 .write = probes_write,
1155};
1156
1157/* Probes profiling interfaces */
1158static int probes_profile_seq_show(struct seq_file *m, void *v)
1159{
1160 struct dyn_event *ev = v;
1161 struct trace_kprobe *tk;
1162
1163 if (!is_trace_kprobe(ev))
1164 return 0;
1165
1166 tk = to_trace_kprobe(ev);
1167 seq_printf(m, " %-44s %15lu %15lu\n",
1168 trace_probe_name(&tk->tp),
1169 trace_kprobe_nhit(tk),
1170 tk->rp.kp.nmissed);
1171
1172 return 0;
1173}
1174
1175static const struct seq_operations profile_seq_op = {
1176 .start = dyn_event_seq_start,
1177 .next = dyn_event_seq_next,
1178 .stop = dyn_event_seq_stop,
1179 .show = probes_profile_seq_show
1180};
1181
1182static int profile_open(struct inode *inode, struct file *file)
1183{
1184 int ret;
1185
1186 ret = security_locked_down(LOCKDOWN_TRACEFS);
1187 if (ret)
1188 return ret;
1189
1190 return seq_open(file, &profile_seq_op);
1191}
1192
1193static const struct file_operations kprobe_profile_ops = {
1194 .owner = THIS_MODULE,
1195 .open = profile_open,
1196 .read = seq_read,
1197 .llseek = seq_lseek,
1198 .release = seq_release,
1199};
1200
1201/* Kprobe specific fetch functions */
1202
1203/* Return the length of string -- including null terminal byte */
1204static nokprobe_inline int
1205fetch_store_strlen_user(unsigned long addr)
1206{
1207 const void __user *uaddr = (__force const void __user *)addr;
1208
1209 return strnlen_user_nofault(uaddr, MAX_STRING_SIZE);
1210}
1211
1212/* Return the length of string -- including null terminal byte */
1213static nokprobe_inline int
1214fetch_store_strlen(unsigned long addr)
1215{
1216 int ret, len = 0;
1217 u8 c;
1218
1219#ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
1220 if (addr < TASK_SIZE)
1221 return fetch_store_strlen_user(addr);
1222#endif
1223
1224 do {
1225 ret = copy_from_kernel_nofault(&c, (u8 *)addr + len, 1);
1226 len++;
1227 } while (c && ret == 0 && len < MAX_STRING_SIZE);
1228
1229 return (ret < 0) ? ret : len;
1230}
1231
1232/*
1233 * Fetch a null-terminated string from user. Caller MUST set *(u32 *)buf
1234 * with max length and relative data location.
1235 */
1236static nokprobe_inline int
1237fetch_store_string_user(unsigned long addr, void *dest, void *base)
1238{
1239 const void __user *uaddr = (__force const void __user *)addr;
1240 int maxlen = get_loc_len(*(u32 *)dest);
1241 void *__dest;
1242 long ret;
1243
1244 if (unlikely(!maxlen))
1245 return -ENOMEM;
1246
1247 __dest = get_loc_data(dest, base);
1248
1249 ret = strncpy_from_user_nofault(__dest, uaddr, maxlen);
1250 if (ret >= 0)
1251 *(u32 *)dest = make_data_loc(ret, __dest - base);
1252
1253 return ret;
1254}
1255
1256/*
1257 * Fetch a null-terminated string. Caller MUST set *(u32 *)buf with max
1258 * length and relative data location.
1259 */
1260static nokprobe_inline int
1261fetch_store_string(unsigned long addr, void *dest, void *base)
1262{
1263 int maxlen = get_loc_len(*(u32 *)dest);
1264 void *__dest;
1265 long ret;
1266
1267#ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
1268 if ((unsigned long)addr < TASK_SIZE)
1269 return fetch_store_string_user(addr, dest, base);
1270#endif
1271
1272 if (unlikely(!maxlen))
1273 return -ENOMEM;
1274
1275 __dest = get_loc_data(dest, base);
1276
1277 /*
1278 * Try to get string again, since the string can be changed while
1279 * probing.
1280 */
1281 ret = strncpy_from_kernel_nofault(__dest, (void *)addr, maxlen);
1282 if (ret >= 0)
1283 *(u32 *)dest = make_data_loc(ret, __dest - base);
1284
1285 return ret;
1286}
1287
1288static nokprobe_inline int
1289probe_mem_read_user(void *dest, void *src, size_t size)
1290{
1291 const void __user *uaddr = (__force const void __user *)src;
1292
1293 return copy_from_user_nofault(dest, uaddr, size);
1294}
1295
1296static nokprobe_inline int
1297probe_mem_read(void *dest, void *src, size_t size)
1298{
1299#ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
1300 if ((unsigned long)src < TASK_SIZE)
1301 return probe_mem_read_user(dest, src, size);
1302#endif
1303 return copy_from_kernel_nofault(dest, src, size);
1304}
1305
1306/* Note that we don't verify it, since the code does not come from user space */
1307static int
1308process_fetch_insn(struct fetch_insn *code, struct pt_regs *regs, void *dest,
1309 void *base)
1310{
1311 unsigned long val;
1312
1313retry:
1314 /* 1st stage: get value from context */
1315 switch (code->op) {
1316 case FETCH_OP_REG:
1317 val = regs_get_register(regs, code->param);
1318 break;
1319 case FETCH_OP_STACK:
1320 val = regs_get_kernel_stack_nth(regs, code->param);
1321 break;
1322 case FETCH_OP_STACKP:
1323 val = kernel_stack_pointer(regs);
1324 break;
1325 case FETCH_OP_RETVAL:
1326 val = regs_return_value(regs);
1327 break;
1328 case FETCH_OP_IMM:
1329 val = code->immediate;
1330 break;
1331 case FETCH_OP_COMM:
1332 val = (unsigned long)current->comm;
1333 break;
1334 case FETCH_OP_DATA:
1335 val = (unsigned long)code->data;
1336 break;
1337#ifdef CONFIG_HAVE_FUNCTION_ARG_ACCESS_API
1338 case FETCH_OP_ARG:
1339 val = regs_get_kernel_argument(regs, code->param);
1340 break;
1341#endif
1342 case FETCH_NOP_SYMBOL: /* Ignore a place holder */
1343 code++;
1344 goto retry;
1345 default:
1346 return -EILSEQ;
1347 }
1348 code++;
1349
1350 return process_fetch_insn_bottom(code, val, dest, base);
1351}
1352NOKPROBE_SYMBOL(process_fetch_insn)
1353
1354/* Kprobe handler */
1355static nokprobe_inline void
1356__kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs,
1357 struct trace_event_file *trace_file)
1358{
1359 struct kprobe_trace_entry_head *entry;
1360 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1361 struct trace_event_buffer fbuffer;
1362 int dsize;
1363
1364 WARN_ON(call != trace_file->event_call);
1365
1366 if (trace_trigger_soft_disabled(trace_file))
1367 return;
1368
1369 local_save_flags(fbuffer.flags);
1370 fbuffer.pc = preempt_count();
1371 fbuffer.trace_file = trace_file;
1372
1373 dsize = __get_data_size(&tk->tp, regs);
1374
1375 fbuffer.event =
1376 trace_event_buffer_lock_reserve(&fbuffer.buffer, trace_file,
1377 call->event.type,
1378 sizeof(*entry) + tk->tp.size + dsize,
1379 fbuffer.flags, fbuffer.pc);
1380 if (!fbuffer.event)
1381 return;
1382
1383 fbuffer.regs = regs;
1384 entry = fbuffer.entry = ring_buffer_event_data(fbuffer.event);
1385 entry->ip = (unsigned long)tk->rp.kp.addr;
1386 store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1387
1388 trace_event_buffer_commit(&fbuffer);
1389}
1390
1391static void
1392kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs)
1393{
1394 struct event_file_link *link;
1395
1396 trace_probe_for_each_link_rcu(link, &tk->tp)
1397 __kprobe_trace_func(tk, regs, link->file);
1398}
1399NOKPROBE_SYMBOL(kprobe_trace_func);
1400
1401/* Kretprobe handler */
1402static nokprobe_inline void
1403__kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1404 struct pt_regs *regs,
1405 struct trace_event_file *trace_file)
1406{
1407 struct kretprobe_trace_entry_head *entry;
1408 struct trace_event_buffer fbuffer;
1409 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1410 int dsize;
1411
1412 WARN_ON(call != trace_file->event_call);
1413
1414 if (trace_trigger_soft_disabled(trace_file))
1415 return;
1416
1417 local_save_flags(fbuffer.flags);
1418 fbuffer.pc = preempt_count();
1419 fbuffer.trace_file = trace_file;
1420
1421 dsize = __get_data_size(&tk->tp, regs);
1422 fbuffer.event =
1423 trace_event_buffer_lock_reserve(&fbuffer.buffer, trace_file,
1424 call->event.type,
1425 sizeof(*entry) + tk->tp.size + dsize,
1426 fbuffer.flags, fbuffer.pc);
1427 if (!fbuffer.event)
1428 return;
1429
1430 fbuffer.regs = regs;
1431 entry = fbuffer.entry = ring_buffer_event_data(fbuffer.event);
1432 entry->func = (unsigned long)tk->rp.kp.addr;
1433 entry->ret_ip = (unsigned long)ri->ret_addr;
1434 store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1435
1436 trace_event_buffer_commit(&fbuffer);
1437}
1438
1439static void
1440kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1441 struct pt_regs *regs)
1442{
1443 struct event_file_link *link;
1444
1445 trace_probe_for_each_link_rcu(link, &tk->tp)
1446 __kretprobe_trace_func(tk, ri, regs, link->file);
1447}
1448NOKPROBE_SYMBOL(kretprobe_trace_func);
1449
1450/* Event entry printers */
1451static enum print_line_t
1452print_kprobe_event(struct trace_iterator *iter, int flags,
1453 struct trace_event *event)
1454{
1455 struct kprobe_trace_entry_head *field;
1456 struct trace_seq *s = &iter->seq;
1457 struct trace_probe *tp;
1458
1459 field = (struct kprobe_trace_entry_head *)iter->ent;
1460 tp = trace_probe_primary_from_call(
1461 container_of(event, struct trace_event_call, event));
1462 if (WARN_ON_ONCE(!tp))
1463 goto out;
1464
1465 trace_seq_printf(s, "%s: (", trace_probe_name(tp));
1466
1467 if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
1468 goto out;
1469
1470 trace_seq_putc(s, ')');
1471
1472 if (print_probe_args(s, tp->args, tp->nr_args,
1473 (u8 *)&field[1], field) < 0)
1474 goto out;
1475
1476 trace_seq_putc(s, '\n');
1477 out:
1478 return trace_handle_return(s);
1479}
1480
1481static enum print_line_t
1482print_kretprobe_event(struct trace_iterator *iter, int flags,
1483 struct trace_event *event)
1484{
1485 struct kretprobe_trace_entry_head *field;
1486 struct trace_seq *s = &iter->seq;
1487 struct trace_probe *tp;
1488
1489 field = (struct kretprobe_trace_entry_head *)iter->ent;
1490 tp = trace_probe_primary_from_call(
1491 container_of(event, struct trace_event_call, event));
1492 if (WARN_ON_ONCE(!tp))
1493 goto out;
1494
1495 trace_seq_printf(s, "%s: (", trace_probe_name(tp));
1496
1497 if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1498 goto out;
1499
1500 trace_seq_puts(s, " <- ");
1501
1502 if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1503 goto out;
1504
1505 trace_seq_putc(s, ')');
1506
1507 if (print_probe_args(s, tp->args, tp->nr_args,
1508 (u8 *)&field[1], field) < 0)
1509 goto out;
1510
1511 trace_seq_putc(s, '\n');
1512
1513 out:
1514 return trace_handle_return(s);
1515}
1516
1517
1518static int kprobe_event_define_fields(struct trace_event_call *event_call)
1519{
1520 int ret;
1521 struct kprobe_trace_entry_head field;
1522 struct trace_probe *tp;
1523
1524 tp = trace_probe_primary_from_call(event_call);
1525 if (WARN_ON_ONCE(!tp))
1526 return -ENOENT;
1527
1528 DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1529
1530 return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
1531}
1532
1533static int kretprobe_event_define_fields(struct trace_event_call *event_call)
1534{
1535 int ret;
1536 struct kretprobe_trace_entry_head field;
1537 struct trace_probe *tp;
1538
1539 tp = trace_probe_primary_from_call(event_call);
1540 if (WARN_ON_ONCE(!tp))
1541 return -ENOENT;
1542
1543 DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1544 DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1545
1546 return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
1547}
1548
1549#ifdef CONFIG_PERF_EVENTS
1550
1551/* Kprobe profile handler */
1552static int
1553kprobe_perf_func(struct trace_kprobe *tk, struct pt_regs *regs)
1554{
1555 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1556 struct kprobe_trace_entry_head *entry;
1557 struct hlist_head *head;
1558 int size, __size, dsize;
1559 int rctx;
1560
1561 if (bpf_prog_array_valid(call)) {
1562 unsigned long orig_ip = instruction_pointer(regs);
1563 int ret;
1564
1565 ret = trace_call_bpf(call, regs);
1566
1567 /*
1568 * We need to check and see if we modified the pc of the
1569 * pt_regs, and if so return 1 so that we don't do the
1570 * single stepping.
1571 */
1572 if (orig_ip != instruction_pointer(regs))
1573 return 1;
1574 if (!ret)
1575 return 0;
1576 }
1577
1578 head = this_cpu_ptr(call->perf_events);
1579 if (hlist_empty(head))
1580 return 0;
1581
1582 dsize = __get_data_size(&tk->tp, regs);
1583 __size = sizeof(*entry) + tk->tp.size + dsize;
1584 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1585 size -= sizeof(u32);
1586
1587 entry = perf_trace_buf_alloc(size, NULL, &rctx);
1588 if (!entry)
1589 return 0;
1590
1591 entry->ip = (unsigned long)tk->rp.kp.addr;
1592 memset(&entry[1], 0, dsize);
1593 store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1594 perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1595 head, NULL);
1596 return 0;
1597}
1598NOKPROBE_SYMBOL(kprobe_perf_func);
1599
1600/* Kretprobe profile handler */
1601static void
1602kretprobe_perf_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1603 struct pt_regs *regs)
1604{
1605 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1606 struct kretprobe_trace_entry_head *entry;
1607 struct hlist_head *head;
1608 int size, __size, dsize;
1609 int rctx;
1610
1611 if (bpf_prog_array_valid(call) && !trace_call_bpf(call, regs))
1612 return;
1613
1614 head = this_cpu_ptr(call->perf_events);
1615 if (hlist_empty(head))
1616 return;
1617
1618 dsize = __get_data_size(&tk->tp, regs);
1619 __size = sizeof(*entry) + tk->tp.size + dsize;
1620 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1621 size -= sizeof(u32);
1622
1623 entry = perf_trace_buf_alloc(size, NULL, &rctx);
1624 if (!entry)
1625 return;
1626
1627 entry->func = (unsigned long)tk->rp.kp.addr;
1628 entry->ret_ip = (unsigned long)ri->ret_addr;
1629 store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1630 perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1631 head, NULL);
1632}
1633NOKPROBE_SYMBOL(kretprobe_perf_func);
1634
1635int bpf_get_kprobe_info(const struct perf_event *event, u32 *fd_type,
1636 const char **symbol, u64 *probe_offset,
1637 u64 *probe_addr, bool perf_type_tracepoint)
1638{
1639 const char *pevent = trace_event_name(event->tp_event);
1640 const char *group = event->tp_event->class->system;
1641 struct trace_kprobe *tk;
1642
1643 if (perf_type_tracepoint)
1644 tk = find_trace_kprobe(pevent, group);
1645 else
1646 tk = trace_kprobe_primary_from_call(event->tp_event);
1647 if (!tk)
1648 return -EINVAL;
1649
1650 *fd_type = trace_kprobe_is_return(tk) ? BPF_FD_TYPE_KRETPROBE
1651 : BPF_FD_TYPE_KPROBE;
1652 if (tk->symbol) {
1653 *symbol = tk->symbol;
1654 *probe_offset = tk->rp.kp.offset;
1655 *probe_addr = 0;
1656 } else {
1657 *symbol = NULL;
1658 *probe_offset = 0;
1659 *probe_addr = (unsigned long)tk->rp.kp.addr;
1660 }
1661 return 0;
1662}
1663#endif /* CONFIG_PERF_EVENTS */
1664
1665/*
1666 * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex.
1667 *
1668 * kprobe_trace_self_tests_init() does enable_trace_probe/disable_trace_probe
1669 * lockless, but we can't race with this __init function.
1670 */
1671static int kprobe_register(struct trace_event_call *event,
1672 enum trace_reg type, void *data)
1673{
1674 struct trace_event_file *file = data;
1675
1676 switch (type) {
1677 case TRACE_REG_REGISTER:
1678 return enable_trace_kprobe(event, file);
1679 case TRACE_REG_UNREGISTER:
1680 return disable_trace_kprobe(event, file);
1681
1682#ifdef CONFIG_PERF_EVENTS
1683 case TRACE_REG_PERF_REGISTER:
1684 return enable_trace_kprobe(event, NULL);
1685 case TRACE_REG_PERF_UNREGISTER:
1686 return disable_trace_kprobe(event, NULL);
1687 case TRACE_REG_PERF_OPEN:
1688 case TRACE_REG_PERF_CLOSE:
1689 case TRACE_REG_PERF_ADD:
1690 case TRACE_REG_PERF_DEL:
1691 return 0;
1692#endif
1693 }
1694 return 0;
1695}
1696
1697static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1698{
1699 struct trace_kprobe *tk = container_of(kp, struct trace_kprobe, rp.kp);
1700 int ret = 0;
1701
1702 raw_cpu_inc(*tk->nhit);
1703
1704 if (trace_probe_test_flag(&tk->tp, TP_FLAG_TRACE))
1705 kprobe_trace_func(tk, regs);
1706#ifdef CONFIG_PERF_EVENTS
1707 if (trace_probe_test_flag(&tk->tp, TP_FLAG_PROFILE))
1708 ret = kprobe_perf_func(tk, regs);
1709#endif
1710 return ret;
1711}
1712NOKPROBE_SYMBOL(kprobe_dispatcher);
1713
1714static int
1715kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1716{
1717 struct trace_kprobe *tk = container_of(ri->rp, struct trace_kprobe, rp);
1718
1719 raw_cpu_inc(*tk->nhit);
1720
1721 if (trace_probe_test_flag(&tk->tp, TP_FLAG_TRACE))
1722 kretprobe_trace_func(tk, ri, regs);
1723#ifdef CONFIG_PERF_EVENTS
1724 if (trace_probe_test_flag(&tk->tp, TP_FLAG_PROFILE))
1725 kretprobe_perf_func(tk, ri, regs);
1726#endif
1727 return 0; /* We don't tweek kernel, so just return 0 */
1728}
1729NOKPROBE_SYMBOL(kretprobe_dispatcher);
1730
1731static struct trace_event_functions kretprobe_funcs = {
1732 .trace = print_kretprobe_event
1733};
1734
1735static struct trace_event_functions kprobe_funcs = {
1736 .trace = print_kprobe_event
1737};
1738
1739static struct trace_event_fields kretprobe_fields_array[] = {
1740 { .type = TRACE_FUNCTION_TYPE,
1741 .define_fields = kretprobe_event_define_fields },
1742 {}
1743};
1744
1745static struct trace_event_fields kprobe_fields_array[] = {
1746 { .type = TRACE_FUNCTION_TYPE,
1747 .define_fields = kprobe_event_define_fields },
1748 {}
1749};
1750
1751static inline void init_trace_event_call(struct trace_kprobe *tk)
1752{
1753 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1754
1755 if (trace_kprobe_is_return(tk)) {
1756 call->event.funcs = &kretprobe_funcs;
1757 call->class->fields_array = kretprobe_fields_array;
1758 } else {
1759 call->event.funcs = &kprobe_funcs;
1760 call->class->fields_array = kprobe_fields_array;
1761 }
1762
1763 call->flags = TRACE_EVENT_FL_KPROBE;
1764 call->class->reg = kprobe_register;
1765}
1766
1767static int register_kprobe_event(struct trace_kprobe *tk)
1768{
1769 init_trace_event_call(tk);
1770
1771 return trace_probe_register_event_call(&tk->tp);
1772}
1773
1774static int unregister_kprobe_event(struct trace_kprobe *tk)
1775{
1776 return trace_probe_unregister_event_call(&tk->tp);
1777}
1778
1779#ifdef CONFIG_PERF_EVENTS
1780/* create a trace_kprobe, but don't add it to global lists */
1781struct trace_event_call *
1782create_local_trace_kprobe(char *func, void *addr, unsigned long offs,
1783 bool is_return)
1784{
1785 struct trace_kprobe *tk;
1786 int ret;
1787 char *event;
1788
1789 /*
1790 * local trace_kprobes are not added to dyn_event, so they are never
1791 * searched in find_trace_kprobe(). Therefore, there is no concern of
1792 * duplicated name here.
1793 */
1794 event = func ? func : "DUMMY_EVENT";
1795
1796 tk = alloc_trace_kprobe(KPROBE_EVENT_SYSTEM, event, (void *)addr, func,
1797 offs, 0 /* maxactive */, 0 /* nargs */,
1798 is_return);
1799
1800 if (IS_ERR(tk)) {
1801 pr_info("Failed to allocate trace_probe.(%d)\n",
1802 (int)PTR_ERR(tk));
1803 return ERR_CAST(tk);
1804 }
1805
1806 init_trace_event_call(tk);
1807
1808 if (traceprobe_set_print_fmt(&tk->tp, trace_kprobe_is_return(tk)) < 0) {
1809 ret = -ENOMEM;
1810 goto error;
1811 }
1812
1813 ret = __register_trace_kprobe(tk);
1814 if (ret < 0)
1815 goto error;
1816
1817 return trace_probe_event_call(&tk->tp);
1818error:
1819 free_trace_kprobe(tk);
1820 return ERR_PTR(ret);
1821}
1822
1823void destroy_local_trace_kprobe(struct trace_event_call *event_call)
1824{
1825 struct trace_kprobe *tk;
1826
1827 tk = trace_kprobe_primary_from_call(event_call);
1828 if (unlikely(!tk))
1829 return;
1830
1831 if (trace_probe_is_enabled(&tk->tp)) {
1832 WARN_ON(1);
1833 return;
1834 }
1835
1836 __unregister_trace_kprobe(tk);
1837
1838 free_trace_kprobe(tk);
1839}
1840#endif /* CONFIG_PERF_EVENTS */
1841
1842static __init void enable_boot_kprobe_events(void)
1843{
1844 struct trace_array *tr = top_trace_array();
1845 struct trace_event_file *file;
1846 struct trace_kprobe *tk;
1847 struct dyn_event *pos;
1848
1849 mutex_lock(&event_mutex);
1850 for_each_trace_kprobe(tk, pos) {
1851 list_for_each_entry(file, &tr->events, list)
1852 if (file->event_call == trace_probe_event_call(&tk->tp))
1853 trace_event_enable_disable(file, 1, 0);
1854 }
1855 mutex_unlock(&event_mutex);
1856}
1857
1858static __init void setup_boot_kprobe_events(void)
1859{
1860 char *p, *cmd = kprobe_boot_events_buf;
1861 int ret;
1862
1863 strreplace(kprobe_boot_events_buf, ',', ' ');
1864
1865 while (cmd && *cmd != '\0') {
1866 p = strchr(cmd, ';');
1867 if (p)
1868 *p++ = '\0';
1869
1870 ret = trace_run_command(cmd, create_or_delete_trace_kprobe);
1871 if (ret)
1872 pr_warn("Failed to add event(%d): %s\n", ret, cmd);
1873 else
1874 kprobe_boot_events_enabled = true;
1875
1876 cmd = p;
1877 }
1878
1879 enable_boot_kprobe_events();
1880}
1881
1882/*
1883 * Register dynevent at subsys_initcall. This allows kernel to setup kprobe
1884 * events in fs_initcall without tracefs.
1885 */
1886static __init int init_kprobe_trace_early(void)
1887{
1888 int ret;
1889
1890 ret = dyn_event_register(&trace_kprobe_ops);
1891 if (ret)
1892 return ret;
1893
1894 if (register_module_notifier(&trace_kprobe_module_nb))
1895 return -EINVAL;
1896
1897 return 0;
1898}
1899subsys_initcall(init_kprobe_trace_early);
1900
1901/* Make a tracefs interface for controlling probe points */
1902static __init int init_kprobe_trace(void)
1903{
1904 struct dentry *d_tracer;
1905 struct dentry *entry;
1906
1907 d_tracer = tracing_init_dentry();
1908 if (IS_ERR(d_tracer))
1909 return 0;
1910
1911 entry = tracefs_create_file("kprobe_events", 0644, d_tracer,
1912 NULL, &kprobe_events_ops);
1913
1914 /* Event list interface */
1915 if (!entry)
1916 pr_warn("Could not create tracefs 'kprobe_events' entry\n");
1917
1918 /* Profile interface */
1919 entry = tracefs_create_file("kprobe_profile", 0444, d_tracer,
1920 NULL, &kprobe_profile_ops);
1921
1922 if (!entry)
1923 pr_warn("Could not create tracefs 'kprobe_profile' entry\n");
1924
1925 setup_boot_kprobe_events();
1926
1927 return 0;
1928}
1929fs_initcall(init_kprobe_trace);
1930
1931
1932#ifdef CONFIG_FTRACE_STARTUP_TEST
1933static __init struct trace_event_file *
1934find_trace_probe_file(struct trace_kprobe *tk, struct trace_array *tr)
1935{
1936 struct trace_event_file *file;
1937
1938 list_for_each_entry(file, &tr->events, list)
1939 if (file->event_call == trace_probe_event_call(&tk->tp))
1940 return file;
1941
1942 return NULL;
1943}
1944
1945/*
1946 * Nobody but us can call enable_trace_kprobe/disable_trace_kprobe at this
1947 * stage, we can do this lockless.
1948 */
1949static __init int kprobe_trace_self_tests_init(void)
1950{
1951 int ret, warn = 0;
1952 int (*target)(int, int, int, int, int, int);
1953 struct trace_kprobe *tk;
1954 struct trace_event_file *file;
1955
1956 if (tracing_is_disabled())
1957 return -ENODEV;
1958
1959 if (kprobe_boot_events_enabled) {
1960 pr_info("Skipping kprobe tests due to kprobe_event on cmdline\n");
1961 return 0;
1962 }
1963
1964 target = kprobe_trace_selftest_target;
1965
1966 pr_info("Testing kprobe tracing: ");
1967
1968 ret = trace_run_command("p:testprobe kprobe_trace_selftest_target $stack $stack0 +0($stack)",
1969 create_or_delete_trace_kprobe);
1970 if (WARN_ON_ONCE(ret)) {
1971 pr_warn("error on probing function entry.\n");
1972 warn++;
1973 } else {
1974 /* Enable trace point */
1975 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1976 if (WARN_ON_ONCE(tk == NULL)) {
1977 pr_warn("error on getting new probe.\n");
1978 warn++;
1979 } else {
1980 file = find_trace_probe_file(tk, top_trace_array());
1981 if (WARN_ON_ONCE(file == NULL)) {
1982 pr_warn("error on getting probe file.\n");
1983 warn++;
1984 } else
1985 enable_trace_kprobe(
1986 trace_probe_event_call(&tk->tp), file);
1987 }
1988 }
1989
1990 ret = trace_run_command("r:testprobe2 kprobe_trace_selftest_target $retval",
1991 create_or_delete_trace_kprobe);
1992 if (WARN_ON_ONCE(ret)) {
1993 pr_warn("error on probing function return.\n");
1994 warn++;
1995 } else {
1996 /* Enable trace point */
1997 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1998 if (WARN_ON_ONCE(tk == NULL)) {
1999 pr_warn("error on getting 2nd new probe.\n");
2000 warn++;
2001 } else {
2002 file = find_trace_probe_file(tk, top_trace_array());
2003 if (WARN_ON_ONCE(file == NULL)) {
2004 pr_warn("error on getting probe file.\n");
2005 warn++;
2006 } else
2007 enable_trace_kprobe(
2008 trace_probe_event_call(&tk->tp), file);
2009 }
2010 }
2011
2012 if (warn)
2013 goto end;
2014
2015 ret = target(1, 2, 3, 4, 5, 6);
2016
2017 /*
2018 * Not expecting an error here, the check is only to prevent the
2019 * optimizer from removing the call to target() as otherwise there
2020 * are no side-effects and the call is never performed.
2021 */
2022 if (ret != 21)
2023 warn++;
2024
2025 /* Disable trace points before removing it */
2026 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
2027 if (WARN_ON_ONCE(tk == NULL)) {
2028 pr_warn("error on getting test probe.\n");
2029 warn++;
2030 } else {
2031 if (trace_kprobe_nhit(tk) != 1) {
2032 pr_warn("incorrect number of testprobe hits\n");
2033 warn++;
2034 }
2035
2036 file = find_trace_probe_file(tk, top_trace_array());
2037 if (WARN_ON_ONCE(file == NULL)) {
2038 pr_warn("error on getting probe file.\n");
2039 warn++;
2040 } else
2041 disable_trace_kprobe(
2042 trace_probe_event_call(&tk->tp), file);
2043 }
2044
2045 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
2046 if (WARN_ON_ONCE(tk == NULL)) {
2047 pr_warn("error on getting 2nd test probe.\n");
2048 warn++;
2049 } else {
2050 if (trace_kprobe_nhit(tk) != 1) {
2051 pr_warn("incorrect number of testprobe2 hits\n");
2052 warn++;
2053 }
2054
2055 file = find_trace_probe_file(tk, top_trace_array());
2056 if (WARN_ON_ONCE(file == NULL)) {
2057 pr_warn("error on getting probe file.\n");
2058 warn++;
2059 } else
2060 disable_trace_kprobe(
2061 trace_probe_event_call(&tk->tp), file);
2062 }
2063
2064 ret = trace_run_command("-:testprobe", create_or_delete_trace_kprobe);
2065 if (WARN_ON_ONCE(ret)) {
2066 pr_warn("error on deleting a probe.\n");
2067 warn++;
2068 }
2069
2070 ret = trace_run_command("-:testprobe2", create_or_delete_trace_kprobe);
2071 if (WARN_ON_ONCE(ret)) {
2072 pr_warn("error on deleting a probe.\n");
2073 warn++;
2074 }
2075
2076end:
2077 ret = dyn_events_release_all(&trace_kprobe_ops);
2078 if (WARN_ON_ONCE(ret)) {
2079 pr_warn("error on cleaning up probes.\n");
2080 warn++;
2081 }
2082 /*
2083 * Wait for the optimizer work to finish. Otherwise it might fiddle
2084 * with probes in already freed __init text.
2085 */
2086 wait_for_kprobe_optimizer();
2087 if (warn)
2088 pr_cont("NG: Some tests are failed. Please check them.\n");
2089 else
2090 pr_cont("OK\n");
2091 return 0;
2092}
2093
2094late_initcall(kprobe_trace_self_tests_init);
2095
2096#endif