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/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
114#ifdef CONFIG_MODULES
115static nokprobe_inline bool trace_kprobe_module_exist(struct trace_kprobe *tk)
116{
117 char *p;
118 bool ret;
119
120 if (!tk->symbol)
121 return false;
122 p = strchr(tk->symbol, ':');
123 if (!p)
124 return true;
125 *p = '\0';
126 rcu_read_lock_sched();
127 ret = !!find_module(tk->symbol);
128 rcu_read_unlock_sched();
129 *p = ':';
130
131 return ret;
132}
133#else
134static inline bool trace_kprobe_module_exist(struct trace_kprobe *tk)
135{
136 return false;
137}
138#endif
139
140static bool trace_kprobe_is_busy(struct dyn_event *ev)
141{
142 struct trace_kprobe *tk = to_trace_kprobe(ev);
143
144 return trace_probe_is_enabled(&tk->tp);
145}
146
147static bool trace_kprobe_match_command_head(struct trace_kprobe *tk,
148 int argc, const char **argv)
149{
150 char buf[MAX_ARGSTR_LEN + 1];
151
152 if (!argc)
153 return true;
154
155 if (!tk->symbol)
156 snprintf(buf, sizeof(buf), "0x%p", tk->rp.kp.addr);
157 else if (tk->rp.kp.offset)
158 snprintf(buf, sizeof(buf), "%s+%u",
159 trace_kprobe_symbol(tk), tk->rp.kp.offset);
160 else
161 snprintf(buf, sizeof(buf), "%s", trace_kprobe_symbol(tk));
162 if (strcmp(buf, argv[0]))
163 return false;
164 argc--; argv++;
165
166 return trace_probe_match_command_args(&tk->tp, argc, argv);
167}
168
169static bool trace_kprobe_match(const char *system, const char *event,
170 int argc, const char **argv, struct dyn_event *ev)
171{
172 struct trace_kprobe *tk = to_trace_kprobe(ev);
173
174 return (event[0] == '\0' ||
175 strcmp(trace_probe_name(&tk->tp), event) == 0) &&
176 (!system || strcmp(trace_probe_group_name(&tk->tp), system) == 0) &&
177 trace_kprobe_match_command_head(tk, argc, argv);
178}
179
180static nokprobe_inline unsigned long trace_kprobe_nhit(struct trace_kprobe *tk)
181{
182 unsigned long nhit = 0;
183 int cpu;
184
185 for_each_possible_cpu(cpu)
186 nhit += *per_cpu_ptr(tk->nhit, cpu);
187
188 return nhit;
189}
190
191static nokprobe_inline bool trace_kprobe_is_registered(struct trace_kprobe *tk)
192{
193 return !(list_empty(&tk->rp.kp.list) &&
194 hlist_unhashed(&tk->rp.kp.hlist));
195}
196
197/* Return 0 if it fails to find the symbol address */
198static nokprobe_inline
199unsigned long trace_kprobe_address(struct trace_kprobe *tk)
200{
201 unsigned long addr;
202
203 if (tk->symbol) {
204 addr = (unsigned long)
205 kallsyms_lookup_name(trace_kprobe_symbol(tk));
206 if (addr)
207 addr += tk->rp.kp.offset;
208 } else {
209 addr = (unsigned long)tk->rp.kp.addr;
210 }
211 return addr;
212}
213
214static nokprobe_inline struct trace_kprobe *
215trace_kprobe_primary_from_call(struct trace_event_call *call)
216{
217 struct trace_probe *tp;
218
219 tp = trace_probe_primary_from_call(call);
220 if (WARN_ON_ONCE(!tp))
221 return NULL;
222
223 return container_of(tp, struct trace_kprobe, tp);
224}
225
226bool trace_kprobe_on_func_entry(struct trace_event_call *call)
227{
228 struct trace_kprobe *tk = trace_kprobe_primary_from_call(call);
229
230 return tk ? (kprobe_on_func_entry(tk->rp.kp.addr,
231 tk->rp.kp.addr ? NULL : tk->rp.kp.symbol_name,
232 tk->rp.kp.addr ? 0 : tk->rp.kp.offset) == 0) : false;
233}
234
235bool trace_kprobe_error_injectable(struct trace_event_call *call)
236{
237 struct trace_kprobe *tk = trace_kprobe_primary_from_call(call);
238
239 return tk ? within_error_injection_list(trace_kprobe_address(tk)) :
240 false;
241}
242
243static int register_kprobe_event(struct trace_kprobe *tk);
244static int unregister_kprobe_event(struct trace_kprobe *tk);
245
246static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
247static int kretprobe_dispatcher(struct kretprobe_instance *ri,
248 struct pt_regs *regs);
249
250static void free_trace_kprobe(struct trace_kprobe *tk)
251{
252 if (tk) {
253 trace_probe_cleanup(&tk->tp);
254 kfree(tk->symbol);
255 free_percpu(tk->nhit);
256 kfree(tk);
257 }
258}
259
260/*
261 * Allocate new trace_probe and initialize it (including kprobes).
262 */
263static struct trace_kprobe *alloc_trace_kprobe(const char *group,
264 const char *event,
265 void *addr,
266 const char *symbol,
267 unsigned long offs,
268 int maxactive,
269 int nargs, bool is_return)
270{
271 struct trace_kprobe *tk;
272 int ret = -ENOMEM;
273
274 tk = kzalloc(struct_size(tk, tp.args, nargs), GFP_KERNEL);
275 if (!tk)
276 return ERR_PTR(ret);
277
278 tk->nhit = alloc_percpu(unsigned long);
279 if (!tk->nhit)
280 goto error;
281
282 if (symbol) {
283 tk->symbol = kstrdup(symbol, GFP_KERNEL);
284 if (!tk->symbol)
285 goto error;
286 tk->rp.kp.symbol_name = tk->symbol;
287 tk->rp.kp.offset = offs;
288 } else
289 tk->rp.kp.addr = addr;
290
291 if (is_return)
292 tk->rp.handler = kretprobe_dispatcher;
293 else
294 tk->rp.kp.pre_handler = kprobe_dispatcher;
295
296 tk->rp.maxactive = maxactive;
297 INIT_HLIST_NODE(&tk->rp.kp.hlist);
298 INIT_LIST_HEAD(&tk->rp.kp.list);
299
300 ret = trace_probe_init(&tk->tp, event, group, false, nargs);
301 if (ret < 0)
302 goto error;
303
304 dyn_event_init(&tk->devent, &trace_kprobe_ops);
305 return tk;
306error:
307 free_trace_kprobe(tk);
308 return ERR_PTR(ret);
309}
310
311static struct trace_kprobe *find_trace_kprobe(const char *event,
312 const char *group)
313{
314 struct dyn_event *pos;
315 struct trace_kprobe *tk;
316
317 for_each_trace_kprobe(tk, pos)
318 if (strcmp(trace_probe_name(&tk->tp), event) == 0 &&
319 strcmp(trace_probe_group_name(&tk->tp), group) == 0)
320 return tk;
321 return NULL;
322}
323
324static inline int __enable_trace_kprobe(struct trace_kprobe *tk)
325{
326 int ret = 0;
327
328 if (trace_kprobe_is_registered(tk) && !trace_kprobe_has_gone(tk)) {
329 if (trace_kprobe_is_return(tk))
330 ret = enable_kretprobe(&tk->rp);
331 else
332 ret = enable_kprobe(&tk->rp.kp);
333 }
334
335 return ret;
336}
337
338static void __disable_trace_kprobe(struct trace_probe *tp)
339{
340 struct trace_kprobe *tk;
341
342 list_for_each_entry(tk, trace_probe_probe_list(tp), tp.list) {
343 if (!trace_kprobe_is_registered(tk))
344 continue;
345 if (trace_kprobe_is_return(tk))
346 disable_kretprobe(&tk->rp);
347 else
348 disable_kprobe(&tk->rp.kp);
349 }
350}
351
352/*
353 * Enable trace_probe
354 * if the file is NULL, enable "perf" handler, or enable "trace" handler.
355 */
356static int enable_trace_kprobe(struct trace_event_call *call,
357 struct trace_event_file *file)
358{
359 struct trace_probe *tp;
360 struct trace_kprobe *tk;
361 bool enabled;
362 int ret = 0;
363
364 tp = trace_probe_primary_from_call(call);
365 if (WARN_ON_ONCE(!tp))
366 return -ENODEV;
367 enabled = trace_probe_is_enabled(tp);
368
369 /* This also changes "enabled" state */
370 if (file) {
371 ret = trace_probe_add_file(tp, file);
372 if (ret)
373 return ret;
374 } else
375 trace_probe_set_flag(tp, TP_FLAG_PROFILE);
376
377 if (enabled)
378 return 0;
379
380 list_for_each_entry(tk, trace_probe_probe_list(tp), tp.list) {
381 if (trace_kprobe_has_gone(tk))
382 continue;
383 ret = __enable_trace_kprobe(tk);
384 if (ret)
385 break;
386 enabled = true;
387 }
388
389 if (ret) {
390 /* Failed to enable one of them. Roll back all */
391 if (enabled)
392 __disable_trace_kprobe(tp);
393 if (file)
394 trace_probe_remove_file(tp, file);
395 else
396 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
397 }
398
399 return ret;
400}
401
402/*
403 * Disable trace_probe
404 * if the file is NULL, disable "perf" handler, or disable "trace" handler.
405 */
406static int disable_trace_kprobe(struct trace_event_call *call,
407 struct trace_event_file *file)
408{
409 struct trace_probe *tp;
410
411 tp = trace_probe_primary_from_call(call);
412 if (WARN_ON_ONCE(!tp))
413 return -ENODEV;
414
415 if (file) {
416 if (!trace_probe_get_file_link(tp, file))
417 return -ENOENT;
418 if (!trace_probe_has_single_file(tp))
419 goto out;
420 trace_probe_clear_flag(tp, TP_FLAG_TRACE);
421 } else
422 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
423
424 if (!trace_probe_is_enabled(tp))
425 __disable_trace_kprobe(tp);
426
427 out:
428 if (file)
429 /*
430 * Synchronization is done in below function. For perf event,
431 * file == NULL and perf_trace_event_unreg() calls
432 * tracepoint_synchronize_unregister() to ensure synchronize
433 * event. We don't need to care about it.
434 */
435 trace_probe_remove_file(tp, file);
436
437 return 0;
438}
439
440#if defined(CONFIG_DYNAMIC_FTRACE) && \
441 !defined(CONFIG_KPROBE_EVENTS_ON_NOTRACE)
442static bool __within_notrace_func(unsigned long addr)
443{
444 unsigned long offset, size;
445
446 if (!addr || !kallsyms_lookup_size_offset(addr, &size, &offset))
447 return false;
448
449 /* Get the entry address of the target function */
450 addr -= offset;
451
452 /*
453 * Since ftrace_location_range() does inclusive range check, we need
454 * to subtract 1 byte from the end address.
455 */
456 return !ftrace_location_range(addr, addr + size - 1);
457}
458
459static bool within_notrace_func(struct trace_kprobe *tk)
460{
461 unsigned long addr = trace_kprobe_address(tk);
462 char symname[KSYM_NAME_LEN], *p;
463
464 if (!__within_notrace_func(addr))
465 return false;
466
467 /* Check if the address is on a suffixed-symbol */
468 if (!lookup_symbol_name(addr, symname)) {
469 p = strchr(symname, '.');
470 if (!p)
471 return true;
472 *p = '\0';
473 addr = (unsigned long)kprobe_lookup_name(symname, 0);
474 if (addr)
475 return __within_notrace_func(addr);
476 }
477
478 return true;
479}
480#else
481#define within_notrace_func(tk) (false)
482#endif
483
484/* Internal register function - just handle k*probes and flags */
485static int __register_trace_kprobe(struct trace_kprobe *tk)
486{
487 int i, ret;
488
489 ret = security_locked_down(LOCKDOWN_KPROBES);
490 if (ret)
491 return ret;
492
493 if (trace_kprobe_is_registered(tk))
494 return -EINVAL;
495
496 if (within_notrace_func(tk)) {
497 pr_warn("Could not probe notrace function %ps\n",
498 (void *)trace_kprobe_address(tk));
499 return -EINVAL;
500 }
501
502 for (i = 0; i < tk->tp.nr_args; i++) {
503 ret = traceprobe_update_arg(&tk->tp.args[i]);
504 if (ret)
505 return ret;
506 }
507
508 /* Set/clear disabled flag according to tp->flag */
509 if (trace_probe_is_enabled(&tk->tp))
510 tk->rp.kp.flags &= ~KPROBE_FLAG_DISABLED;
511 else
512 tk->rp.kp.flags |= KPROBE_FLAG_DISABLED;
513
514 if (trace_kprobe_is_return(tk))
515 ret = register_kretprobe(&tk->rp);
516 else
517 ret = register_kprobe(&tk->rp.kp);
518
519 return ret;
520}
521
522/* Internal unregister function - just handle k*probes and flags */
523static void __unregister_trace_kprobe(struct trace_kprobe *tk)
524{
525 if (trace_kprobe_is_registered(tk)) {
526 if (trace_kprobe_is_return(tk))
527 unregister_kretprobe(&tk->rp);
528 else
529 unregister_kprobe(&tk->rp.kp);
530 /* Cleanup kprobe for reuse and mark it unregistered */
531 INIT_HLIST_NODE(&tk->rp.kp.hlist);
532 INIT_LIST_HEAD(&tk->rp.kp.list);
533 if (tk->rp.kp.symbol_name)
534 tk->rp.kp.addr = NULL;
535 }
536}
537
538/* Unregister a trace_probe and probe_event */
539static int unregister_trace_kprobe(struct trace_kprobe *tk)
540{
541 /* If other probes are on the event, just unregister kprobe */
542 if (trace_probe_has_sibling(&tk->tp))
543 goto unreg;
544
545 /* Enabled event can not be unregistered */
546 if (trace_probe_is_enabled(&tk->tp))
547 return -EBUSY;
548
549 /* If there's a reference to the dynamic event */
550 if (trace_event_dyn_busy(trace_probe_event_call(&tk->tp)))
551 return -EBUSY;
552
553 /* Will fail if probe is being used by ftrace or perf */
554 if (unregister_kprobe_event(tk))
555 return -EBUSY;
556
557unreg:
558 __unregister_trace_kprobe(tk);
559 dyn_event_remove(&tk->devent);
560 trace_probe_unlink(&tk->tp);
561
562 return 0;
563}
564
565static bool trace_kprobe_has_same_kprobe(struct trace_kprobe *orig,
566 struct trace_kprobe *comp)
567{
568 struct trace_probe_event *tpe = orig->tp.event;
569 int i;
570
571 list_for_each_entry(orig, &tpe->probes, tp.list) {
572 if (strcmp(trace_kprobe_symbol(orig),
573 trace_kprobe_symbol(comp)) ||
574 trace_kprobe_offset(orig) != trace_kprobe_offset(comp))
575 continue;
576
577 /*
578 * trace_probe_compare_arg_type() ensured that nr_args and
579 * each argument name and type are same. Let's compare comm.
580 */
581 for (i = 0; i < orig->tp.nr_args; i++) {
582 if (strcmp(orig->tp.args[i].comm,
583 comp->tp.args[i].comm))
584 break;
585 }
586
587 if (i == orig->tp.nr_args)
588 return true;
589 }
590
591 return false;
592}
593
594static int append_trace_kprobe(struct trace_kprobe *tk, struct trace_kprobe *to)
595{
596 int ret;
597
598 ret = trace_probe_compare_arg_type(&tk->tp, &to->tp);
599 if (ret) {
600 /* Note that argument starts index = 2 */
601 trace_probe_log_set_index(ret + 1);
602 trace_probe_log_err(0, DIFF_ARG_TYPE);
603 return -EEXIST;
604 }
605 if (trace_kprobe_has_same_kprobe(to, tk)) {
606 trace_probe_log_set_index(0);
607 trace_probe_log_err(0, SAME_PROBE);
608 return -EEXIST;
609 }
610
611 /* Append to existing event */
612 ret = trace_probe_append(&tk->tp, &to->tp);
613 if (ret)
614 return ret;
615
616 /* Register k*probe */
617 ret = __register_trace_kprobe(tk);
618 if (ret == -ENOENT && !trace_kprobe_module_exist(tk)) {
619 pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
620 ret = 0;
621 }
622
623 if (ret)
624 trace_probe_unlink(&tk->tp);
625 else
626 dyn_event_add(&tk->devent, trace_probe_event_call(&tk->tp));
627
628 return ret;
629}
630
631/* Register a trace_probe and probe_event */
632static int register_trace_kprobe(struct trace_kprobe *tk)
633{
634 struct trace_kprobe *old_tk;
635 int ret;
636
637 mutex_lock(&event_mutex);
638
639 old_tk = find_trace_kprobe(trace_probe_name(&tk->tp),
640 trace_probe_group_name(&tk->tp));
641 if (old_tk) {
642 if (trace_kprobe_is_return(tk) != trace_kprobe_is_return(old_tk)) {
643 trace_probe_log_set_index(0);
644 trace_probe_log_err(0, DIFF_PROBE_TYPE);
645 ret = -EEXIST;
646 } else {
647 ret = append_trace_kprobe(tk, old_tk);
648 }
649 goto end;
650 }
651
652 /* Register new event */
653 ret = register_kprobe_event(tk);
654 if (ret) {
655 if (ret == -EEXIST) {
656 trace_probe_log_set_index(0);
657 trace_probe_log_err(0, EVENT_EXIST);
658 } else
659 pr_warn("Failed to register probe event(%d)\n", ret);
660 goto end;
661 }
662
663 /* Register k*probe */
664 ret = __register_trace_kprobe(tk);
665 if (ret == -ENOENT && !trace_kprobe_module_exist(tk)) {
666 pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
667 ret = 0;
668 }
669
670 if (ret < 0)
671 unregister_kprobe_event(tk);
672 else
673 dyn_event_add(&tk->devent, trace_probe_event_call(&tk->tp));
674
675end:
676 mutex_unlock(&event_mutex);
677 return ret;
678}
679
680#ifdef CONFIG_MODULES
681static int validate_module_probe_symbol(const char *modname, const char *symbol);
682
683static int register_module_trace_kprobe(struct module *mod, struct trace_kprobe *tk)
684{
685 const char *p;
686 int ret = 0;
687
688 p = strchr(trace_kprobe_symbol(tk), ':');
689 if (p)
690 ret = validate_module_probe_symbol(module_name(mod), p + 1);
691 if (!ret)
692 ret = __register_trace_kprobe(tk);
693 return ret;
694}
695
696/* Module notifier call back, checking event on the module */
697static int trace_kprobe_module_callback(struct notifier_block *nb,
698 unsigned long val, void *data)
699{
700 struct module *mod = data;
701 struct dyn_event *pos;
702 struct trace_kprobe *tk;
703 int ret;
704
705 if (val != MODULE_STATE_COMING)
706 return NOTIFY_DONE;
707
708 /* Update probes on coming module */
709 mutex_lock(&event_mutex);
710 for_each_trace_kprobe(tk, pos) {
711 if (trace_kprobe_within_module(tk, mod)) {
712 /* Don't need to check busy - this should have gone. */
713 __unregister_trace_kprobe(tk);
714 ret = register_module_trace_kprobe(mod, tk);
715 if (ret)
716 pr_warn("Failed to re-register probe %s on %s: %d\n",
717 trace_probe_name(&tk->tp),
718 module_name(mod), ret);
719 }
720 }
721 mutex_unlock(&event_mutex);
722
723 return NOTIFY_DONE;
724}
725
726static struct notifier_block trace_kprobe_module_nb = {
727 .notifier_call = trace_kprobe_module_callback,
728 .priority = 2 /* Invoked after kprobe and jump_label module callback */
729};
730static int trace_kprobe_register_module_notifier(void)
731{
732 return register_module_notifier(&trace_kprobe_module_nb);
733}
734#else
735static int trace_kprobe_register_module_notifier(void)
736{
737 return 0;
738}
739#endif /* CONFIG_MODULES */
740
741static int count_symbols(void *data, unsigned long unused)
742{
743 unsigned int *count = data;
744
745 (*count)++;
746
747 return 0;
748}
749
750struct sym_count_ctx {
751 unsigned int count;
752 const char *name;
753};
754
755static int count_mod_symbols(void *data, const char *name, unsigned long unused)
756{
757 struct sym_count_ctx *ctx = data;
758
759 if (strcmp(name, ctx->name) == 0)
760 ctx->count++;
761
762 return 0;
763}
764
765static unsigned int number_of_same_symbols(const char *mod, const char *func_name)
766{
767 struct sym_count_ctx ctx = { .count = 0, .name = func_name };
768
769 if (!mod)
770 kallsyms_on_each_match_symbol(count_symbols, func_name, &ctx.count);
771
772 module_kallsyms_on_each_symbol(mod, count_mod_symbols, &ctx);
773
774 return ctx.count;
775}
776
777static int validate_module_probe_symbol(const char *modname, const char *symbol)
778{
779 unsigned int count = number_of_same_symbols(modname, symbol);
780
781 if (count > 1) {
782 /*
783 * Users should use ADDR to remove the ambiguity of
784 * using KSYM only.
785 */
786 return -EADDRNOTAVAIL;
787 } else if (count == 0) {
788 /*
789 * We can return ENOENT earlier than when register the
790 * kprobe.
791 */
792 return -ENOENT;
793 }
794 return 0;
795}
796
797#ifdef CONFIG_MODULES
798/* Return NULL if the module is not loaded or under unloading. */
799static struct module *try_module_get_by_name(const char *name)
800{
801 struct module *mod;
802
803 rcu_read_lock_sched();
804 mod = find_module(name);
805 if (mod && !try_module_get(mod))
806 mod = NULL;
807 rcu_read_unlock_sched();
808
809 return mod;
810}
811#else
812#define try_module_get_by_name(name) (NULL)
813#endif
814
815static int validate_probe_symbol(char *symbol)
816{
817 struct module *mod = NULL;
818 char *modname = NULL, *p;
819 int ret = 0;
820
821 p = strchr(symbol, ':');
822 if (p) {
823 modname = symbol;
824 symbol = p + 1;
825 *p = '\0';
826 mod = try_module_get_by_name(modname);
827 if (!mod)
828 goto out;
829 }
830
831 ret = validate_module_probe_symbol(modname, symbol);
832out:
833 if (p)
834 *p = ':';
835 if (mod)
836 module_put(mod);
837 return ret;
838}
839
840static int trace_kprobe_entry_handler(struct kretprobe_instance *ri,
841 struct pt_regs *regs);
842
843static int __trace_kprobe_create(int argc, const char *argv[])
844{
845 /*
846 * Argument syntax:
847 * - Add kprobe:
848 * p[:[GRP/][EVENT]] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
849 * - Add kretprobe:
850 * r[MAXACTIVE][:[GRP/][EVENT]] [MOD:]KSYM[+0] [FETCHARGS]
851 * Or
852 * p[:[GRP/][EVENT]] [MOD:]KSYM[+0]%return [FETCHARGS]
853 *
854 * Fetch args:
855 * $retval : fetch return value
856 * $stack : fetch stack address
857 * $stackN : fetch Nth of stack (N:0-)
858 * $comm : fetch current task comm
859 * @ADDR : fetch memory at ADDR (ADDR should be in kernel)
860 * @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
861 * %REG : fetch register REG
862 * Dereferencing memory fetch:
863 * +|-offs(ARG) : fetch memory at ARG +|- offs address.
864 * Alias name of args:
865 * NAME=FETCHARG : set NAME as alias of FETCHARG.
866 * Type of args:
867 * FETCHARG:TYPE : use TYPE instead of unsigned long.
868 */
869 struct trace_kprobe *tk = NULL;
870 int i, len, new_argc = 0, ret = 0;
871 bool is_return = false;
872 char *symbol = NULL, *tmp = NULL;
873 const char **new_argv = NULL;
874 const char *event = NULL, *group = KPROBE_EVENT_SYSTEM;
875 enum probe_print_type ptype;
876 int maxactive = 0;
877 long offset = 0;
878 void *addr = NULL;
879 char buf[MAX_EVENT_NAME_LEN];
880 char gbuf[MAX_EVENT_NAME_LEN];
881 char abuf[MAX_BTF_ARGS_LEN];
882 char *dbuf = NULL;
883 struct traceprobe_parse_context ctx = { .flags = TPARG_FL_KERNEL };
884
885 switch (argv[0][0]) {
886 case 'r':
887 is_return = true;
888 break;
889 case 'p':
890 break;
891 default:
892 return -ECANCELED;
893 }
894 if (argc < 2)
895 return -ECANCELED;
896
897 trace_probe_log_init("trace_kprobe", argc, argv);
898
899 event = strchr(&argv[0][1], ':');
900 if (event)
901 event++;
902
903 if (isdigit(argv[0][1])) {
904 if (!is_return) {
905 trace_probe_log_err(1, BAD_MAXACT_TYPE);
906 goto parse_error;
907 }
908 if (event)
909 len = event - &argv[0][1] - 1;
910 else
911 len = strlen(&argv[0][1]);
912 if (len > MAX_EVENT_NAME_LEN - 1) {
913 trace_probe_log_err(1, BAD_MAXACT);
914 goto parse_error;
915 }
916 memcpy(buf, &argv[0][1], len);
917 buf[len] = '\0';
918 ret = kstrtouint(buf, 0, &maxactive);
919 if (ret || !maxactive) {
920 trace_probe_log_err(1, BAD_MAXACT);
921 goto parse_error;
922 }
923 /* kretprobes instances are iterated over via a list. The
924 * maximum should stay reasonable.
925 */
926 if (maxactive > KRETPROBE_MAXACTIVE_MAX) {
927 trace_probe_log_err(1, MAXACT_TOO_BIG);
928 goto parse_error;
929 }
930 }
931
932 /* try to parse an address. if that fails, try to read the
933 * input as a symbol. */
934 if (kstrtoul(argv[1], 0, (unsigned long *)&addr)) {
935 trace_probe_log_set_index(1);
936 /* Check whether uprobe event specified */
937 if (strchr(argv[1], '/') && strchr(argv[1], ':')) {
938 ret = -ECANCELED;
939 goto error;
940 }
941 /* a symbol specified */
942 symbol = kstrdup(argv[1], GFP_KERNEL);
943 if (!symbol) {
944 ret = -ENOMEM;
945 goto error;
946 }
947
948 tmp = strchr(symbol, '%');
949 if (tmp) {
950 if (!strcmp(tmp, "%return")) {
951 *tmp = '\0';
952 is_return = true;
953 } else {
954 trace_probe_log_err(tmp - symbol, BAD_ADDR_SUFFIX);
955 goto parse_error;
956 }
957 }
958
959 /* TODO: support .init module functions */
960 ret = traceprobe_split_symbol_offset(symbol, &offset);
961 if (ret || offset < 0 || offset > UINT_MAX) {
962 trace_probe_log_err(0, BAD_PROBE_ADDR);
963 goto parse_error;
964 }
965 ret = validate_probe_symbol(symbol);
966 if (ret) {
967 if (ret == -EADDRNOTAVAIL)
968 trace_probe_log_err(0, NON_UNIQ_SYMBOL);
969 else
970 trace_probe_log_err(0, BAD_PROBE_ADDR);
971 goto parse_error;
972 }
973 if (is_return)
974 ctx.flags |= TPARG_FL_RETURN;
975 ret = kprobe_on_func_entry(NULL, symbol, offset);
976 if (ret == 0 && !is_return)
977 ctx.flags |= TPARG_FL_FENTRY;
978 /* Defer the ENOENT case until register kprobe */
979 if (ret == -EINVAL && is_return) {
980 trace_probe_log_err(0, BAD_RETPROBE);
981 goto parse_error;
982 }
983 }
984
985 trace_probe_log_set_index(0);
986 if (event) {
987 ret = traceprobe_parse_event_name(&event, &group, gbuf,
988 event - argv[0]);
989 if (ret)
990 goto parse_error;
991 }
992
993 if (!event) {
994 /* Make a new event name */
995 if (symbol)
996 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
997 is_return ? 'r' : 'p', symbol, offset);
998 else
999 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
1000 is_return ? 'r' : 'p', addr);
1001 sanitize_event_name(buf);
1002 event = buf;
1003 }
1004
1005 argc -= 2; argv += 2;
1006 ctx.funcname = symbol;
1007 new_argv = traceprobe_expand_meta_args(argc, argv, &new_argc,
1008 abuf, MAX_BTF_ARGS_LEN, &ctx);
1009 if (IS_ERR(new_argv)) {
1010 ret = PTR_ERR(new_argv);
1011 new_argv = NULL;
1012 goto out;
1013 }
1014 if (new_argv) {
1015 argc = new_argc;
1016 argv = new_argv;
1017 }
1018 if (argc > MAX_TRACE_ARGS) {
1019 ret = -E2BIG;
1020 goto out;
1021 }
1022
1023 ret = traceprobe_expand_dentry_args(argc, argv, &dbuf);
1024 if (ret)
1025 goto out;
1026
1027 /* setup a probe */
1028 tk = alloc_trace_kprobe(group, event, addr, symbol, offset, maxactive,
1029 argc, is_return);
1030 if (IS_ERR(tk)) {
1031 ret = PTR_ERR(tk);
1032 /* This must return -ENOMEM, else there is a bug */
1033 WARN_ON_ONCE(ret != -ENOMEM);
1034 goto out; /* We know tk is not allocated */
1035 }
1036
1037 /* parse arguments */
1038 for (i = 0; i < argc; i++) {
1039 trace_probe_log_set_index(i + 2);
1040 ctx.offset = 0;
1041 ret = traceprobe_parse_probe_arg(&tk->tp, i, argv[i], &ctx);
1042 if (ret)
1043 goto error; /* This can be -ENOMEM */
1044 }
1045 /* entry handler for kretprobe */
1046 if (is_return && tk->tp.entry_arg) {
1047 tk->rp.entry_handler = trace_kprobe_entry_handler;
1048 tk->rp.data_size = traceprobe_get_entry_data_size(&tk->tp);
1049 }
1050
1051 ptype = is_return ? PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL;
1052 ret = traceprobe_set_print_fmt(&tk->tp, ptype);
1053 if (ret < 0)
1054 goto error;
1055
1056 ret = register_trace_kprobe(tk);
1057 if (ret) {
1058 trace_probe_log_set_index(1);
1059 if (ret == -EILSEQ)
1060 trace_probe_log_err(0, BAD_INSN_BNDRY);
1061 else if (ret == -ENOENT)
1062 trace_probe_log_err(0, BAD_PROBE_ADDR);
1063 else if (ret != -ENOMEM && ret != -EEXIST)
1064 trace_probe_log_err(0, FAIL_REG_PROBE);
1065 goto error;
1066 }
1067
1068out:
1069 traceprobe_finish_parse(&ctx);
1070 trace_probe_log_clear();
1071 kfree(new_argv);
1072 kfree(symbol);
1073 kfree(dbuf);
1074 return ret;
1075
1076parse_error:
1077 ret = -EINVAL;
1078error:
1079 free_trace_kprobe(tk);
1080 goto out;
1081}
1082
1083static int trace_kprobe_create(const char *raw_command)
1084{
1085 return trace_probe_create(raw_command, __trace_kprobe_create);
1086}
1087
1088static int create_or_delete_trace_kprobe(const char *raw_command)
1089{
1090 int ret;
1091
1092 if (raw_command[0] == '-')
1093 return dyn_event_release(raw_command, &trace_kprobe_ops);
1094
1095 ret = trace_kprobe_create(raw_command);
1096 return ret == -ECANCELED ? -EINVAL : ret;
1097}
1098
1099static int trace_kprobe_run_command(struct dynevent_cmd *cmd)
1100{
1101 return create_or_delete_trace_kprobe(cmd->seq.buffer);
1102}
1103
1104/**
1105 * kprobe_event_cmd_init - Initialize a kprobe event command object
1106 * @cmd: A pointer to the dynevent_cmd struct representing the new event
1107 * @buf: A pointer to the buffer used to build the command
1108 * @maxlen: The length of the buffer passed in @buf
1109 *
1110 * Initialize a synthetic event command object. Use this before
1111 * calling any of the other kprobe_event functions.
1112 */
1113void kprobe_event_cmd_init(struct dynevent_cmd *cmd, char *buf, int maxlen)
1114{
1115 dynevent_cmd_init(cmd, buf, maxlen, DYNEVENT_TYPE_KPROBE,
1116 trace_kprobe_run_command);
1117}
1118EXPORT_SYMBOL_GPL(kprobe_event_cmd_init);
1119
1120/**
1121 * __kprobe_event_gen_cmd_start - Generate a kprobe event command from arg list
1122 * @cmd: A pointer to the dynevent_cmd struct representing the new event
1123 * @kretprobe: Is this a return probe?
1124 * @name: The name of the kprobe event
1125 * @loc: The location of the kprobe event
1126 * @...: Variable number of arg (pairs), one pair for each field
1127 *
1128 * NOTE: Users normally won't want to call this function directly, but
1129 * rather use the kprobe_event_gen_cmd_start() wrapper, which automatically
1130 * adds a NULL to the end of the arg list. If this function is used
1131 * directly, make sure the last arg in the variable arg list is NULL.
1132 *
1133 * Generate a kprobe event command to be executed by
1134 * kprobe_event_gen_cmd_end(). This function can be used to generate the
1135 * complete command or only the first part of it; in the latter case,
1136 * kprobe_event_add_fields() can be used to add more fields following this.
1137 *
1138 * Unlikely the synth_event_gen_cmd_start(), @loc must be specified. This
1139 * returns -EINVAL if @loc == NULL.
1140 *
1141 * Return: 0 if successful, error otherwise.
1142 */
1143int __kprobe_event_gen_cmd_start(struct dynevent_cmd *cmd, bool kretprobe,
1144 const char *name, const char *loc, ...)
1145{
1146 char buf[MAX_EVENT_NAME_LEN];
1147 struct dynevent_arg arg;
1148 va_list args;
1149 int ret;
1150
1151 if (cmd->type != DYNEVENT_TYPE_KPROBE)
1152 return -EINVAL;
1153
1154 if (!loc)
1155 return -EINVAL;
1156
1157 if (kretprobe)
1158 snprintf(buf, MAX_EVENT_NAME_LEN, "r:kprobes/%s", name);
1159 else
1160 snprintf(buf, MAX_EVENT_NAME_LEN, "p:kprobes/%s", name);
1161
1162 ret = dynevent_str_add(cmd, buf);
1163 if (ret)
1164 return ret;
1165
1166 dynevent_arg_init(&arg, 0);
1167 arg.str = loc;
1168 ret = dynevent_arg_add(cmd, &arg, NULL);
1169 if (ret)
1170 return ret;
1171
1172 va_start(args, loc);
1173 for (;;) {
1174 const char *field;
1175
1176 field = va_arg(args, const char *);
1177 if (!field)
1178 break;
1179
1180 if (++cmd->n_fields > MAX_TRACE_ARGS) {
1181 ret = -EINVAL;
1182 break;
1183 }
1184
1185 arg.str = field;
1186 ret = dynevent_arg_add(cmd, &arg, NULL);
1187 if (ret)
1188 break;
1189 }
1190 va_end(args);
1191
1192 return ret;
1193}
1194EXPORT_SYMBOL_GPL(__kprobe_event_gen_cmd_start);
1195
1196/**
1197 * __kprobe_event_add_fields - Add probe fields to a kprobe command from arg list
1198 * @cmd: A pointer to the dynevent_cmd struct representing the new event
1199 * @...: Variable number of arg (pairs), one pair for each field
1200 *
1201 * NOTE: Users normally won't want to call this function directly, but
1202 * rather use the kprobe_event_add_fields() wrapper, which
1203 * automatically adds a NULL to the end of the arg list. If this
1204 * function is used directly, make sure the last arg in the variable
1205 * arg list is NULL.
1206 *
1207 * Add probe fields to an existing kprobe command using a variable
1208 * list of args. Fields are added in the same order they're listed.
1209 *
1210 * Return: 0 if successful, error otherwise.
1211 */
1212int __kprobe_event_add_fields(struct dynevent_cmd *cmd, ...)
1213{
1214 struct dynevent_arg arg;
1215 va_list args;
1216 int ret = 0;
1217
1218 if (cmd->type != DYNEVENT_TYPE_KPROBE)
1219 return -EINVAL;
1220
1221 dynevent_arg_init(&arg, 0);
1222
1223 va_start(args, cmd);
1224 for (;;) {
1225 const char *field;
1226
1227 field = va_arg(args, const char *);
1228 if (!field)
1229 break;
1230
1231 if (++cmd->n_fields > MAX_TRACE_ARGS) {
1232 ret = -EINVAL;
1233 break;
1234 }
1235
1236 arg.str = field;
1237 ret = dynevent_arg_add(cmd, &arg, NULL);
1238 if (ret)
1239 break;
1240 }
1241 va_end(args);
1242
1243 return ret;
1244}
1245EXPORT_SYMBOL_GPL(__kprobe_event_add_fields);
1246
1247/**
1248 * kprobe_event_delete - Delete a kprobe event
1249 * @name: The name of the kprobe event to delete
1250 *
1251 * Delete a kprobe event with the give @name from kernel code rather
1252 * than directly from the command line.
1253 *
1254 * Return: 0 if successful, error otherwise.
1255 */
1256int kprobe_event_delete(const char *name)
1257{
1258 char buf[MAX_EVENT_NAME_LEN];
1259
1260 snprintf(buf, MAX_EVENT_NAME_LEN, "-:%s", name);
1261
1262 return create_or_delete_trace_kprobe(buf);
1263}
1264EXPORT_SYMBOL_GPL(kprobe_event_delete);
1265
1266static int trace_kprobe_release(struct dyn_event *ev)
1267{
1268 struct trace_kprobe *tk = to_trace_kprobe(ev);
1269 int ret = unregister_trace_kprobe(tk);
1270
1271 if (!ret)
1272 free_trace_kprobe(tk);
1273 return ret;
1274}
1275
1276static int trace_kprobe_show(struct seq_file *m, struct dyn_event *ev)
1277{
1278 struct trace_kprobe *tk = to_trace_kprobe(ev);
1279 int i;
1280
1281 seq_putc(m, trace_kprobe_is_return(tk) ? 'r' : 'p');
1282 if (trace_kprobe_is_return(tk) && tk->rp.maxactive)
1283 seq_printf(m, "%d", tk->rp.maxactive);
1284 seq_printf(m, ":%s/%s", trace_probe_group_name(&tk->tp),
1285 trace_probe_name(&tk->tp));
1286
1287 if (!tk->symbol)
1288 seq_printf(m, " 0x%p", tk->rp.kp.addr);
1289 else if (tk->rp.kp.offset)
1290 seq_printf(m, " %s+%u", trace_kprobe_symbol(tk),
1291 tk->rp.kp.offset);
1292 else
1293 seq_printf(m, " %s", trace_kprobe_symbol(tk));
1294
1295 for (i = 0; i < tk->tp.nr_args; i++)
1296 seq_printf(m, " %s=%s", tk->tp.args[i].name, tk->tp.args[i].comm);
1297 seq_putc(m, '\n');
1298
1299 return 0;
1300}
1301
1302static int probes_seq_show(struct seq_file *m, void *v)
1303{
1304 struct dyn_event *ev = v;
1305
1306 if (!is_trace_kprobe(ev))
1307 return 0;
1308
1309 return trace_kprobe_show(m, ev);
1310}
1311
1312static const struct seq_operations probes_seq_op = {
1313 .start = dyn_event_seq_start,
1314 .next = dyn_event_seq_next,
1315 .stop = dyn_event_seq_stop,
1316 .show = probes_seq_show
1317};
1318
1319static int probes_open(struct inode *inode, struct file *file)
1320{
1321 int ret;
1322
1323 ret = security_locked_down(LOCKDOWN_TRACEFS);
1324 if (ret)
1325 return ret;
1326
1327 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
1328 ret = dyn_events_release_all(&trace_kprobe_ops);
1329 if (ret < 0)
1330 return ret;
1331 }
1332
1333 return seq_open(file, &probes_seq_op);
1334}
1335
1336static ssize_t probes_write(struct file *file, const char __user *buffer,
1337 size_t count, loff_t *ppos)
1338{
1339 return trace_parse_run_command(file, buffer, count, ppos,
1340 create_or_delete_trace_kprobe);
1341}
1342
1343static const struct file_operations kprobe_events_ops = {
1344 .owner = THIS_MODULE,
1345 .open = probes_open,
1346 .read = seq_read,
1347 .llseek = seq_lseek,
1348 .release = seq_release,
1349 .write = probes_write,
1350};
1351
1352static unsigned long trace_kprobe_missed(struct trace_kprobe *tk)
1353{
1354 return trace_kprobe_is_return(tk) ?
1355 tk->rp.kp.nmissed + tk->rp.nmissed : tk->rp.kp.nmissed;
1356}
1357
1358/* Probes profiling interfaces */
1359static int probes_profile_seq_show(struct seq_file *m, void *v)
1360{
1361 struct dyn_event *ev = v;
1362 struct trace_kprobe *tk;
1363 unsigned long nmissed;
1364
1365 if (!is_trace_kprobe(ev))
1366 return 0;
1367
1368 tk = to_trace_kprobe(ev);
1369 nmissed = trace_kprobe_missed(tk);
1370 seq_printf(m, " %-44s %15lu %15lu\n",
1371 trace_probe_name(&tk->tp),
1372 trace_kprobe_nhit(tk),
1373 nmissed);
1374
1375 return 0;
1376}
1377
1378static const struct seq_operations profile_seq_op = {
1379 .start = dyn_event_seq_start,
1380 .next = dyn_event_seq_next,
1381 .stop = dyn_event_seq_stop,
1382 .show = probes_profile_seq_show
1383};
1384
1385static int profile_open(struct inode *inode, struct file *file)
1386{
1387 int ret;
1388
1389 ret = security_locked_down(LOCKDOWN_TRACEFS);
1390 if (ret)
1391 return ret;
1392
1393 return seq_open(file, &profile_seq_op);
1394}
1395
1396static const struct file_operations kprobe_profile_ops = {
1397 .owner = THIS_MODULE,
1398 .open = profile_open,
1399 .read = seq_read,
1400 .llseek = seq_lseek,
1401 .release = seq_release,
1402};
1403
1404/* Note that we don't verify it, since the code does not come from user space */
1405static int
1406process_fetch_insn(struct fetch_insn *code, void *rec, void *edata,
1407 void *dest, void *base)
1408{
1409 struct pt_regs *regs = rec;
1410 unsigned long val;
1411 int ret;
1412
1413retry:
1414 /* 1st stage: get value from context */
1415 switch (code->op) {
1416 case FETCH_OP_REG:
1417 val = regs_get_register(regs, code->param);
1418 break;
1419 case FETCH_OP_STACK:
1420 val = regs_get_kernel_stack_nth(regs, code->param);
1421 break;
1422 case FETCH_OP_STACKP:
1423 val = kernel_stack_pointer(regs);
1424 break;
1425 case FETCH_OP_RETVAL:
1426 val = regs_return_value(regs);
1427 break;
1428#ifdef CONFIG_HAVE_FUNCTION_ARG_ACCESS_API
1429 case FETCH_OP_ARG:
1430 val = regs_get_kernel_argument(regs, code->param);
1431 break;
1432 case FETCH_OP_EDATA:
1433 val = *(unsigned long *)((unsigned long)edata + code->offset);
1434 break;
1435#endif
1436 case FETCH_NOP_SYMBOL: /* Ignore a place holder */
1437 code++;
1438 goto retry;
1439 default:
1440 ret = process_common_fetch_insn(code, &val);
1441 if (ret < 0)
1442 return ret;
1443 }
1444 code++;
1445
1446 return process_fetch_insn_bottom(code, val, dest, base);
1447}
1448NOKPROBE_SYMBOL(process_fetch_insn)
1449
1450/* Kprobe handler */
1451static nokprobe_inline void
1452__kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs,
1453 struct trace_event_file *trace_file)
1454{
1455 struct kprobe_trace_entry_head *entry;
1456 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1457 struct trace_event_buffer fbuffer;
1458 int dsize;
1459
1460 WARN_ON(call != trace_file->event_call);
1461
1462 if (trace_trigger_soft_disabled(trace_file))
1463 return;
1464
1465 dsize = __get_data_size(&tk->tp, regs, NULL);
1466
1467 entry = trace_event_buffer_reserve(&fbuffer, trace_file,
1468 sizeof(*entry) + tk->tp.size + dsize);
1469 if (!entry)
1470 return;
1471
1472 fbuffer.regs = regs;
1473 entry->ip = (unsigned long)tk->rp.kp.addr;
1474 store_trace_args(&entry[1], &tk->tp, regs, NULL, sizeof(*entry), dsize);
1475
1476 trace_event_buffer_commit(&fbuffer);
1477}
1478
1479static void
1480kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs)
1481{
1482 struct event_file_link *link;
1483
1484 trace_probe_for_each_link_rcu(link, &tk->tp)
1485 __kprobe_trace_func(tk, regs, link->file);
1486}
1487NOKPROBE_SYMBOL(kprobe_trace_func);
1488
1489/* Kretprobe handler */
1490
1491static int trace_kprobe_entry_handler(struct kretprobe_instance *ri,
1492 struct pt_regs *regs)
1493{
1494 struct kretprobe *rp = get_kretprobe(ri);
1495 struct trace_kprobe *tk;
1496
1497 /*
1498 * There is a small chance that get_kretprobe(ri) returns NULL when
1499 * the kretprobe is unregister on another CPU between kretprobe's
1500 * trampoline_handler and this function.
1501 */
1502 if (unlikely(!rp))
1503 return -ENOENT;
1504
1505 tk = container_of(rp, struct trace_kprobe, rp);
1506
1507 /* store argument values into ri->data as entry data */
1508 if (tk->tp.entry_arg)
1509 store_trace_entry_data(ri->data, &tk->tp, regs);
1510
1511 return 0;
1512}
1513
1514
1515static nokprobe_inline void
1516__kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1517 struct pt_regs *regs,
1518 struct trace_event_file *trace_file)
1519{
1520 struct kretprobe_trace_entry_head *entry;
1521 struct trace_event_buffer fbuffer;
1522 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1523 int dsize;
1524
1525 WARN_ON(call != trace_file->event_call);
1526
1527 if (trace_trigger_soft_disabled(trace_file))
1528 return;
1529
1530 dsize = __get_data_size(&tk->tp, regs, ri->data);
1531
1532 entry = trace_event_buffer_reserve(&fbuffer, trace_file,
1533 sizeof(*entry) + tk->tp.size + dsize);
1534 if (!entry)
1535 return;
1536
1537 fbuffer.regs = regs;
1538 entry->func = (unsigned long)tk->rp.kp.addr;
1539 entry->ret_ip = get_kretprobe_retaddr(ri);
1540 store_trace_args(&entry[1], &tk->tp, regs, ri->data, sizeof(*entry), dsize);
1541
1542 trace_event_buffer_commit(&fbuffer);
1543}
1544
1545static void
1546kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1547 struct pt_regs *regs)
1548{
1549 struct event_file_link *link;
1550
1551 trace_probe_for_each_link_rcu(link, &tk->tp)
1552 __kretprobe_trace_func(tk, ri, regs, link->file);
1553}
1554NOKPROBE_SYMBOL(kretprobe_trace_func);
1555
1556/* Event entry printers */
1557static enum print_line_t
1558print_kprobe_event(struct trace_iterator *iter, int flags,
1559 struct trace_event *event)
1560{
1561 struct kprobe_trace_entry_head *field;
1562 struct trace_seq *s = &iter->seq;
1563 struct trace_probe *tp;
1564
1565 field = (struct kprobe_trace_entry_head *)iter->ent;
1566 tp = trace_probe_primary_from_call(
1567 container_of(event, struct trace_event_call, event));
1568 if (WARN_ON_ONCE(!tp))
1569 goto out;
1570
1571 trace_seq_printf(s, "%s: (", trace_probe_name(tp));
1572
1573 if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
1574 goto out;
1575
1576 trace_seq_putc(s, ')');
1577
1578 if (trace_probe_print_args(s, tp->args, tp->nr_args,
1579 (u8 *)&field[1], field) < 0)
1580 goto out;
1581
1582 trace_seq_putc(s, '\n');
1583 out:
1584 return trace_handle_return(s);
1585}
1586
1587static enum print_line_t
1588print_kretprobe_event(struct trace_iterator *iter, int flags,
1589 struct trace_event *event)
1590{
1591 struct kretprobe_trace_entry_head *field;
1592 struct trace_seq *s = &iter->seq;
1593 struct trace_probe *tp;
1594
1595 field = (struct kretprobe_trace_entry_head *)iter->ent;
1596 tp = trace_probe_primary_from_call(
1597 container_of(event, struct trace_event_call, event));
1598 if (WARN_ON_ONCE(!tp))
1599 goto out;
1600
1601 trace_seq_printf(s, "%s: (", trace_probe_name(tp));
1602
1603 if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1604 goto out;
1605
1606 trace_seq_puts(s, " <- ");
1607
1608 if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1609 goto out;
1610
1611 trace_seq_putc(s, ')');
1612
1613 if (trace_probe_print_args(s, tp->args, tp->nr_args,
1614 (u8 *)&field[1], field) < 0)
1615 goto out;
1616
1617 trace_seq_putc(s, '\n');
1618
1619 out:
1620 return trace_handle_return(s);
1621}
1622
1623
1624static int kprobe_event_define_fields(struct trace_event_call *event_call)
1625{
1626 int ret;
1627 struct kprobe_trace_entry_head field;
1628 struct trace_probe *tp;
1629
1630 tp = trace_probe_primary_from_call(event_call);
1631 if (WARN_ON_ONCE(!tp))
1632 return -ENOENT;
1633
1634 DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1635
1636 return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
1637}
1638
1639static int kretprobe_event_define_fields(struct trace_event_call *event_call)
1640{
1641 int ret;
1642 struct kretprobe_trace_entry_head field;
1643 struct trace_probe *tp;
1644
1645 tp = trace_probe_primary_from_call(event_call);
1646 if (WARN_ON_ONCE(!tp))
1647 return -ENOENT;
1648
1649 DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1650 DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1651
1652 return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
1653}
1654
1655#ifdef CONFIG_PERF_EVENTS
1656
1657/* Kprobe profile handler */
1658static int
1659kprobe_perf_func(struct trace_kprobe *tk, struct pt_regs *regs)
1660{
1661 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1662 struct kprobe_trace_entry_head *entry;
1663 struct hlist_head *head;
1664 int size, __size, dsize;
1665 int rctx;
1666
1667 if (bpf_prog_array_valid(call)) {
1668 unsigned long orig_ip = instruction_pointer(regs);
1669 int ret;
1670
1671 ret = trace_call_bpf(call, regs);
1672
1673 /*
1674 * We need to check and see if we modified the pc of the
1675 * pt_regs, and if so return 1 so that we don't do the
1676 * single stepping.
1677 */
1678 if (orig_ip != instruction_pointer(regs))
1679 return 1;
1680 if (!ret)
1681 return 0;
1682 }
1683
1684 head = this_cpu_ptr(call->perf_events);
1685 if (hlist_empty(head))
1686 return 0;
1687
1688 dsize = __get_data_size(&tk->tp, regs, NULL);
1689 __size = sizeof(*entry) + tk->tp.size + dsize;
1690 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1691 size -= sizeof(u32);
1692
1693 entry = perf_trace_buf_alloc(size, NULL, &rctx);
1694 if (!entry)
1695 return 0;
1696
1697 entry->ip = (unsigned long)tk->rp.kp.addr;
1698 memset(&entry[1], 0, dsize);
1699 store_trace_args(&entry[1], &tk->tp, regs, NULL, sizeof(*entry), dsize);
1700 perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1701 head, NULL);
1702 return 0;
1703}
1704NOKPROBE_SYMBOL(kprobe_perf_func);
1705
1706/* Kretprobe profile handler */
1707static void
1708kretprobe_perf_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1709 struct pt_regs *regs)
1710{
1711 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1712 struct kretprobe_trace_entry_head *entry;
1713 struct hlist_head *head;
1714 int size, __size, dsize;
1715 int rctx;
1716
1717 if (bpf_prog_array_valid(call) && !trace_call_bpf(call, regs))
1718 return;
1719
1720 head = this_cpu_ptr(call->perf_events);
1721 if (hlist_empty(head))
1722 return;
1723
1724 dsize = __get_data_size(&tk->tp, regs, ri->data);
1725 __size = sizeof(*entry) + tk->tp.size + dsize;
1726 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1727 size -= sizeof(u32);
1728
1729 entry = perf_trace_buf_alloc(size, NULL, &rctx);
1730 if (!entry)
1731 return;
1732
1733 entry->func = (unsigned long)tk->rp.kp.addr;
1734 entry->ret_ip = get_kretprobe_retaddr(ri);
1735 store_trace_args(&entry[1], &tk->tp, regs, ri->data, sizeof(*entry), dsize);
1736 perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1737 head, NULL);
1738}
1739NOKPROBE_SYMBOL(kretprobe_perf_func);
1740
1741int bpf_get_kprobe_info(const struct perf_event *event, u32 *fd_type,
1742 const char **symbol, u64 *probe_offset,
1743 u64 *probe_addr, unsigned long *missed,
1744 bool perf_type_tracepoint)
1745{
1746 const char *pevent = trace_event_name(event->tp_event);
1747 const char *group = event->tp_event->class->system;
1748 struct trace_kprobe *tk;
1749
1750 if (perf_type_tracepoint)
1751 tk = find_trace_kprobe(pevent, group);
1752 else
1753 tk = trace_kprobe_primary_from_call(event->tp_event);
1754 if (!tk)
1755 return -EINVAL;
1756
1757 *fd_type = trace_kprobe_is_return(tk) ? BPF_FD_TYPE_KRETPROBE
1758 : BPF_FD_TYPE_KPROBE;
1759 *probe_offset = tk->rp.kp.offset;
1760 *probe_addr = kallsyms_show_value(current_cred()) ?
1761 (unsigned long)tk->rp.kp.addr : 0;
1762 *symbol = tk->symbol;
1763 if (missed)
1764 *missed = trace_kprobe_missed(tk);
1765 return 0;
1766}
1767#endif /* CONFIG_PERF_EVENTS */
1768
1769/*
1770 * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex.
1771 *
1772 * kprobe_trace_self_tests_init() does enable_trace_probe/disable_trace_probe
1773 * lockless, but we can't race with this __init function.
1774 */
1775static int kprobe_register(struct trace_event_call *event,
1776 enum trace_reg type, void *data)
1777{
1778 struct trace_event_file *file = data;
1779
1780 switch (type) {
1781 case TRACE_REG_REGISTER:
1782 return enable_trace_kprobe(event, file);
1783 case TRACE_REG_UNREGISTER:
1784 return disable_trace_kprobe(event, file);
1785
1786#ifdef CONFIG_PERF_EVENTS
1787 case TRACE_REG_PERF_REGISTER:
1788 return enable_trace_kprobe(event, NULL);
1789 case TRACE_REG_PERF_UNREGISTER:
1790 return disable_trace_kprobe(event, NULL);
1791 case TRACE_REG_PERF_OPEN:
1792 case TRACE_REG_PERF_CLOSE:
1793 case TRACE_REG_PERF_ADD:
1794 case TRACE_REG_PERF_DEL:
1795 return 0;
1796#endif
1797 }
1798 return 0;
1799}
1800
1801static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1802{
1803 struct trace_kprobe *tk = container_of(kp, struct trace_kprobe, rp.kp);
1804 int ret = 0;
1805
1806 raw_cpu_inc(*tk->nhit);
1807
1808 if (trace_probe_test_flag(&tk->tp, TP_FLAG_TRACE))
1809 kprobe_trace_func(tk, regs);
1810#ifdef CONFIG_PERF_EVENTS
1811 if (trace_probe_test_flag(&tk->tp, TP_FLAG_PROFILE))
1812 ret = kprobe_perf_func(tk, regs);
1813#endif
1814 return ret;
1815}
1816NOKPROBE_SYMBOL(kprobe_dispatcher);
1817
1818static int
1819kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1820{
1821 struct kretprobe *rp = get_kretprobe(ri);
1822 struct trace_kprobe *tk;
1823
1824 /*
1825 * There is a small chance that get_kretprobe(ri) returns NULL when
1826 * the kretprobe is unregister on another CPU between kretprobe's
1827 * trampoline_handler and this function.
1828 */
1829 if (unlikely(!rp))
1830 return 0;
1831
1832 tk = container_of(rp, struct trace_kprobe, rp);
1833 raw_cpu_inc(*tk->nhit);
1834
1835 if (trace_probe_test_flag(&tk->tp, TP_FLAG_TRACE))
1836 kretprobe_trace_func(tk, ri, regs);
1837#ifdef CONFIG_PERF_EVENTS
1838 if (trace_probe_test_flag(&tk->tp, TP_FLAG_PROFILE))
1839 kretprobe_perf_func(tk, ri, regs);
1840#endif
1841 return 0; /* We don't tweak kernel, so just return 0 */
1842}
1843NOKPROBE_SYMBOL(kretprobe_dispatcher);
1844
1845static struct trace_event_functions kretprobe_funcs = {
1846 .trace = print_kretprobe_event
1847};
1848
1849static struct trace_event_functions kprobe_funcs = {
1850 .trace = print_kprobe_event
1851};
1852
1853static struct trace_event_fields kretprobe_fields_array[] = {
1854 { .type = TRACE_FUNCTION_TYPE,
1855 .define_fields = kretprobe_event_define_fields },
1856 {}
1857};
1858
1859static struct trace_event_fields kprobe_fields_array[] = {
1860 { .type = TRACE_FUNCTION_TYPE,
1861 .define_fields = kprobe_event_define_fields },
1862 {}
1863};
1864
1865static inline void init_trace_event_call(struct trace_kprobe *tk)
1866{
1867 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1868
1869 if (trace_kprobe_is_return(tk)) {
1870 call->event.funcs = &kretprobe_funcs;
1871 call->class->fields_array = kretprobe_fields_array;
1872 } else {
1873 call->event.funcs = &kprobe_funcs;
1874 call->class->fields_array = kprobe_fields_array;
1875 }
1876
1877 call->flags = TRACE_EVENT_FL_KPROBE;
1878 call->class->reg = kprobe_register;
1879}
1880
1881static int register_kprobe_event(struct trace_kprobe *tk)
1882{
1883 init_trace_event_call(tk);
1884
1885 return trace_probe_register_event_call(&tk->tp);
1886}
1887
1888static int unregister_kprobe_event(struct trace_kprobe *tk)
1889{
1890 return trace_probe_unregister_event_call(&tk->tp);
1891}
1892
1893#ifdef CONFIG_PERF_EVENTS
1894
1895/* create a trace_kprobe, but don't add it to global lists */
1896struct trace_event_call *
1897create_local_trace_kprobe(char *func, void *addr, unsigned long offs,
1898 bool is_return)
1899{
1900 enum probe_print_type ptype;
1901 struct trace_kprobe *tk;
1902 int ret;
1903 char *event;
1904
1905 if (func) {
1906 ret = validate_probe_symbol(func);
1907 if (ret)
1908 return ERR_PTR(ret);
1909 }
1910
1911 /*
1912 * local trace_kprobes are not added to dyn_event, so they are never
1913 * searched in find_trace_kprobe(). Therefore, there is no concern of
1914 * duplicated name here.
1915 */
1916 event = func ? func : "DUMMY_EVENT";
1917
1918 tk = alloc_trace_kprobe(KPROBE_EVENT_SYSTEM, event, (void *)addr, func,
1919 offs, 0 /* maxactive */, 0 /* nargs */,
1920 is_return);
1921
1922 if (IS_ERR(tk)) {
1923 pr_info("Failed to allocate trace_probe.(%d)\n",
1924 (int)PTR_ERR(tk));
1925 return ERR_CAST(tk);
1926 }
1927
1928 init_trace_event_call(tk);
1929
1930 ptype = trace_kprobe_is_return(tk) ?
1931 PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL;
1932 if (traceprobe_set_print_fmt(&tk->tp, ptype) < 0) {
1933 ret = -ENOMEM;
1934 goto error;
1935 }
1936
1937 ret = __register_trace_kprobe(tk);
1938 if (ret < 0)
1939 goto error;
1940
1941 return trace_probe_event_call(&tk->tp);
1942error:
1943 free_trace_kprobe(tk);
1944 return ERR_PTR(ret);
1945}
1946
1947void destroy_local_trace_kprobe(struct trace_event_call *event_call)
1948{
1949 struct trace_kprobe *tk;
1950
1951 tk = trace_kprobe_primary_from_call(event_call);
1952 if (unlikely(!tk))
1953 return;
1954
1955 if (trace_probe_is_enabled(&tk->tp)) {
1956 WARN_ON(1);
1957 return;
1958 }
1959
1960 __unregister_trace_kprobe(tk);
1961
1962 free_trace_kprobe(tk);
1963}
1964#endif /* CONFIG_PERF_EVENTS */
1965
1966static __init void enable_boot_kprobe_events(void)
1967{
1968 struct trace_array *tr = top_trace_array();
1969 struct trace_event_file *file;
1970 struct trace_kprobe *tk;
1971 struct dyn_event *pos;
1972
1973 mutex_lock(&event_mutex);
1974 for_each_trace_kprobe(tk, pos) {
1975 list_for_each_entry(file, &tr->events, list)
1976 if (file->event_call == trace_probe_event_call(&tk->tp))
1977 trace_event_enable_disable(file, 1, 0);
1978 }
1979 mutex_unlock(&event_mutex);
1980}
1981
1982static __init void setup_boot_kprobe_events(void)
1983{
1984 char *p, *cmd = kprobe_boot_events_buf;
1985 int ret;
1986
1987 strreplace(kprobe_boot_events_buf, ',', ' ');
1988
1989 while (cmd && *cmd != '\0') {
1990 p = strchr(cmd, ';');
1991 if (p)
1992 *p++ = '\0';
1993
1994 ret = create_or_delete_trace_kprobe(cmd);
1995 if (ret)
1996 pr_warn("Failed to add event(%d): %s\n", ret, cmd);
1997
1998 cmd = p;
1999 }
2000
2001 enable_boot_kprobe_events();
2002}
2003
2004/*
2005 * Register dynevent at core_initcall. This allows kernel to setup kprobe
2006 * events in postcore_initcall without tracefs.
2007 */
2008static __init int init_kprobe_trace_early(void)
2009{
2010 int ret;
2011
2012 ret = dyn_event_register(&trace_kprobe_ops);
2013 if (ret)
2014 return ret;
2015
2016 if (trace_kprobe_register_module_notifier())
2017 return -EINVAL;
2018
2019 return 0;
2020}
2021core_initcall(init_kprobe_trace_early);
2022
2023/* Make a tracefs interface for controlling probe points */
2024static __init int init_kprobe_trace(void)
2025{
2026 int ret;
2027
2028 ret = tracing_init_dentry();
2029 if (ret)
2030 return 0;
2031
2032 /* Event list interface */
2033 trace_create_file("kprobe_events", TRACE_MODE_WRITE,
2034 NULL, NULL, &kprobe_events_ops);
2035
2036 /* Profile interface */
2037 trace_create_file("kprobe_profile", TRACE_MODE_READ,
2038 NULL, NULL, &kprobe_profile_ops);
2039
2040 setup_boot_kprobe_events();
2041
2042 return 0;
2043}
2044fs_initcall(init_kprobe_trace);
2045
2046
2047#ifdef CONFIG_FTRACE_STARTUP_TEST
2048static __init struct trace_event_file *
2049find_trace_probe_file(struct trace_kprobe *tk, struct trace_array *tr)
2050{
2051 struct trace_event_file *file;
2052
2053 list_for_each_entry(file, &tr->events, list)
2054 if (file->event_call == trace_probe_event_call(&tk->tp))
2055 return file;
2056
2057 return NULL;
2058}
2059
2060/*
2061 * Nobody but us can call enable_trace_kprobe/disable_trace_kprobe at this
2062 * stage, we can do this lockless.
2063 */
2064static __init int kprobe_trace_self_tests_init(void)
2065{
2066 int ret, warn = 0;
2067 int (*target)(int, int, int, int, int, int);
2068 struct trace_kprobe *tk;
2069 struct trace_event_file *file;
2070
2071 if (tracing_is_disabled())
2072 return -ENODEV;
2073
2074 if (tracing_selftest_disabled)
2075 return 0;
2076
2077 target = kprobe_trace_selftest_target;
2078
2079 pr_info("Testing kprobe tracing: ");
2080
2081 ret = create_or_delete_trace_kprobe("p:testprobe kprobe_trace_selftest_target $stack $stack0 +0($stack)");
2082 if (WARN_ONCE(ret, "error on probing function entry.")) {
2083 warn++;
2084 } else {
2085 /* Enable trace point */
2086 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
2087 if (WARN_ONCE(tk == NULL, "error on probing function entry.")) {
2088 warn++;
2089 } else {
2090 file = find_trace_probe_file(tk, top_trace_array());
2091 if (WARN_ONCE(file == NULL, "error on getting probe file.")) {
2092 warn++;
2093 } else
2094 enable_trace_kprobe(
2095 trace_probe_event_call(&tk->tp), file);
2096 }
2097 }
2098
2099 ret = create_or_delete_trace_kprobe("r:testprobe2 kprobe_trace_selftest_target $retval");
2100 if (WARN_ONCE(ret, "error on probing function return.")) {
2101 warn++;
2102 } else {
2103 /* Enable trace point */
2104 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
2105 if (WARN_ONCE(tk == NULL, "error on getting 2nd new probe.")) {
2106 warn++;
2107 } else {
2108 file = find_trace_probe_file(tk, top_trace_array());
2109 if (WARN_ONCE(file == NULL, "error on getting probe file.")) {
2110 warn++;
2111 } else
2112 enable_trace_kprobe(
2113 trace_probe_event_call(&tk->tp), file);
2114 }
2115 }
2116
2117 if (warn)
2118 goto end;
2119
2120 ret = target(1, 2, 3, 4, 5, 6);
2121
2122 /*
2123 * Not expecting an error here, the check is only to prevent the
2124 * optimizer from removing the call to target() as otherwise there
2125 * are no side-effects and the call is never performed.
2126 */
2127 if (ret != 21)
2128 warn++;
2129
2130 /* Disable trace points before removing it */
2131 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
2132 if (WARN_ONCE(tk == NULL, "error on getting test probe.")) {
2133 warn++;
2134 } else {
2135 if (WARN_ONCE(trace_kprobe_nhit(tk) != 1,
2136 "incorrect number of testprobe hits."))
2137 warn++;
2138
2139 file = find_trace_probe_file(tk, top_trace_array());
2140 if (WARN_ONCE(file == NULL, "error on getting probe file.")) {
2141 warn++;
2142 } else
2143 disable_trace_kprobe(
2144 trace_probe_event_call(&tk->tp), file);
2145 }
2146
2147 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
2148 if (WARN_ONCE(tk == NULL, "error on getting 2nd test probe.")) {
2149 warn++;
2150 } else {
2151 if (WARN_ONCE(trace_kprobe_nhit(tk) != 1,
2152 "incorrect number of testprobe2 hits."))
2153 warn++;
2154
2155 file = find_trace_probe_file(tk, top_trace_array());
2156 if (WARN_ONCE(file == NULL, "error on getting probe file.")) {
2157 warn++;
2158 } else
2159 disable_trace_kprobe(
2160 trace_probe_event_call(&tk->tp), file);
2161 }
2162
2163 ret = create_or_delete_trace_kprobe("-:testprobe");
2164 if (WARN_ONCE(ret, "error on deleting a probe."))
2165 warn++;
2166
2167 ret = create_or_delete_trace_kprobe("-:testprobe2");
2168 if (WARN_ONCE(ret, "error on deleting a probe."))
2169 warn++;
2170
2171
2172end:
2173 /*
2174 * Wait for the optimizer work to finish. Otherwise it might fiddle
2175 * with probes in already freed __init text.
2176 */
2177 wait_for_kprobe_optimizer();
2178 if (warn)
2179 pr_cont("NG: Some tests are failed. Please check them.\n");
2180 else
2181 pr_cont("OK\n");
2182 return 0;
2183}
2184
2185late_initcall(kprobe_trace_self_tests_init);
2186
2187#endif