?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Linux 5.6.13
[linux/fpc-iii.git] / kernel / trace / trace_kprobe.c
blob35989383ae1131e11e45a39e0a5333a57f4493e0
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Kprobes-based tracing events
4 *
5 * Created by Masami Hiramatsu <mhiramat@redhat.com>
6 *
7 */
8 #define pr_fmt(fmt) "trace_kprobe: " fmt
9
10 #include <linux/security.h>
11 #include <linux/module.h>
12 #include <linux/uaccess.h>
13 #include <linux/rculist.h>
14 #include <linux/error-injection.h>
15
16 #include <asm/setup.h> /* for COMMAND_LINE_SIZE */
17
18 #include "trace_dynevent.h"
19 #include "trace_kprobe_selftest.h"
20 #include "trace_probe.h"
21 #include "trace_probe_tmpl.h"
22
23 #define KPROBE_EVENT_SYSTEM "kprobes"
24 #define KRETPROBE_MAXACTIVE_MAX 4096
25
26 /* Kprobe early definition from command line */
27 static char kprobe_boot_events_buf[COMMAND_LINE_SIZE] __initdata;
28 static bool kprobe_boot_events_enabled __initdata;
29
30 static int __init set_kprobe_boot_events(char *str)
31 {
32 strlcpy(kprobe_boot_events_buf, str, COMMAND_LINE_SIZE);
33 return 0;
34 }
35 __setup("kprobe_event=", set_kprobe_boot_events);
36
37 static int trace_kprobe_create(int argc, const char **argv);
38 static int trace_kprobe_show(struct seq_file *m, struct dyn_event *ev);
39 static int trace_kprobe_release(struct dyn_event *ev);
40 static bool trace_kprobe_is_busy(struct dyn_event *ev);
41 static bool trace_kprobe_match(const char *system, const char *event,
42 int argc, const char **argv, struct dyn_event *ev);
43
44 static struct dyn_event_operations trace_kprobe_ops = {
45 .create = trace_kprobe_create,
46 .show = trace_kprobe_show,
47 .is_busy = trace_kprobe_is_busy,
48 .free = trace_kprobe_release,
49 .match = trace_kprobe_match,
50 };
51
52 /*
53 * Kprobe event core functions
54 */
55 struct trace_kprobe {
56 struct dyn_event devent;
57 struct kretprobe rp; /* Use rp.kp for kprobe use */
58 unsigned long __percpu *nhit;
59 const char *symbol; /* symbol name */
60 struct trace_probe tp;
61 };
62
63 static bool is_trace_kprobe(struct dyn_event *ev)
64 {
65 return ev->ops == &trace_kprobe_ops;
66 }
67
68 static struct trace_kprobe *to_trace_kprobe(struct dyn_event *ev)
69 {
70 return container_of(ev, struct trace_kprobe, devent);
71 }
72
73 /**
74 * for_each_trace_kprobe - iterate over the trace_kprobe list
75 * @pos: the struct trace_kprobe * for each entry
76 * @dpos: the struct dyn_event * to use as a loop cursor
77 */
78 #define for_each_trace_kprobe(pos, dpos) \
79 for_each_dyn_event(dpos) \
80 if (is_trace_kprobe(dpos) && (pos = to_trace_kprobe(dpos)))
81
82 #define SIZEOF_TRACE_KPROBE(n) \
83 (offsetof(struct trace_kprobe, tp.args) + \
84 (sizeof(struct probe_arg) * (n)))
85
86 static nokprobe_inline bool trace_kprobe_is_return(struct trace_kprobe *tk)
87 {
88 return tk->rp.handler != NULL;
89 }
90
91 static nokprobe_inline const char *trace_kprobe_symbol(struct trace_kprobe *tk)
92 {
93 return tk->symbol ? tk->symbol : "unknown";
94 }
95
96 static nokprobe_inline unsigned long trace_kprobe_offset(struct trace_kprobe *tk)
97 {
98 return tk->rp.kp.offset;
99 }
100
101 static nokprobe_inline bool trace_kprobe_has_gone(struct trace_kprobe *tk)
102 {
103 return !!(kprobe_gone(&tk->rp.kp));
104 }
105
106 static nokprobe_inline bool trace_kprobe_within_module(struct trace_kprobe *tk,
107 struct module *mod)
108 {
109 int len = strlen(mod->name);
110 const char *name = trace_kprobe_symbol(tk);
111 return strncmp(mod->name, name, len) == 0 && name[len] == ':';
112 }
113
114 static nokprobe_inline bool trace_kprobe_module_exist(struct trace_kprobe *tk)
115 {
116 char *p;
117 bool ret;
118
119 if (!tk->symbol)
120 return false;
121 p = strchr(tk->symbol, ':');
122 if (!p)
123 return true;
124 *p = '\0';
125 mutex_lock(&module_mutex);
126 ret = !!find_module(tk->symbol);
127 mutex_unlock(&module_mutex);
128 *p = ':';
129
130 return ret;
131 }
132
133 static 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
140 static 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
162 static bool trace_kprobe_match(const char *system, const char *event,
163 int argc, const char **argv, struct dyn_event *ev)
164 {
165 struct trace_kprobe *tk = to_trace_kprobe(ev);
166
167 return strcmp(trace_probe_name(&tk->tp), event) == 0 &&
168 (!system || strcmp(trace_probe_group_name(&tk->tp), system) == 0) &&
169 trace_kprobe_match_command_head(tk, argc, argv);
170 }
171
172 static nokprobe_inline unsigned long trace_kprobe_nhit(struct trace_kprobe *tk)
173 {
174 unsigned long nhit = 0;
175 int cpu;
176
177 for_each_possible_cpu(cpu)
178 nhit += *per_cpu_ptr(tk->nhit, cpu);
179
180 return nhit;
181 }
182
183 static nokprobe_inline bool trace_kprobe_is_registered(struct trace_kprobe *tk)
184 {
185 return !(list_empty(&tk->rp.kp.list) &&
186 hlist_unhashed(&tk->rp.kp.hlist));
187 }
188
189 /* Return 0 if it fails to find the symbol address */
190 static nokprobe_inline
191 unsigned long trace_kprobe_address(struct trace_kprobe *tk)
192 {
193 unsigned long addr;
194
195 if (tk->symbol) {
196 addr = (unsigned long)
197 kallsyms_lookup_name(trace_kprobe_symbol(tk));
198 if (addr)
199 addr += tk->rp.kp.offset;
200 } else {
201 addr = (unsigned long)tk->rp.kp.addr;
202 }
203 return addr;
204 }
205
206 static nokprobe_inline struct trace_kprobe *
207 trace_kprobe_primary_from_call(struct trace_event_call *call)
208 {
209 struct trace_probe *tp;
210
211 tp = trace_probe_primary_from_call(call);
212 if (WARN_ON_ONCE(!tp))
213 return NULL;
214
215 return container_of(tp, struct trace_kprobe, tp);
216 }
217
218 bool trace_kprobe_on_func_entry(struct trace_event_call *call)
219 {
220 struct trace_kprobe *tk = trace_kprobe_primary_from_call(call);
221
222 return tk ? kprobe_on_func_entry(tk->rp.kp.addr,
223 tk->rp.kp.addr ? NULL : tk->rp.kp.symbol_name,
224 tk->rp.kp.addr ? 0 : tk->rp.kp.offset) : false;
225 }
226
227 bool trace_kprobe_error_injectable(struct trace_event_call *call)
228 {
229 struct trace_kprobe *tk = trace_kprobe_primary_from_call(call);
230
231 return tk ? within_error_injection_list(trace_kprobe_address(tk)) :
232 false;
233 }
234
235 static int register_kprobe_event(struct trace_kprobe *tk);
236 static int unregister_kprobe_event(struct trace_kprobe *tk);
237
238 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
239 static int kretprobe_dispatcher(struct kretprobe_instance *ri,
240 struct pt_regs *regs);
241
242 static void free_trace_kprobe(struct trace_kprobe *tk)
243 {
244 if (tk) {
245 trace_probe_cleanup(&tk->tp);
246 kfree(tk->symbol);
247 free_percpu(tk->nhit);
248 kfree(tk);
249 }
250 }
251
252 /*
253 * Allocate new trace_probe and initialize it (including kprobes).
254 */
255 static struct trace_kprobe *alloc_trace_kprobe(const char *group,
256 const char *event,
257 void *addr,
258 const char *symbol,
259 unsigned long offs,
260 int maxactive,
261 int nargs, bool is_return)
262 {
263 struct trace_kprobe *tk;
264 int ret = -ENOMEM;
265
266 tk = kzalloc(SIZEOF_TRACE_KPROBE(nargs), GFP_KERNEL);
267 if (!tk)
268 return ERR_PTR(ret);
269
270 tk->nhit = alloc_percpu(unsigned long);
271 if (!tk->nhit)
272 goto error;
273
274 if (symbol) {
275 tk->symbol = kstrdup(symbol, GFP_KERNEL);
276 if (!tk->symbol)
277 goto error;
278 tk->rp.kp.symbol_name = tk->symbol;
279 tk->rp.kp.offset = offs;
280 } else
281 tk->rp.kp.addr = addr;
282
283 if (is_return)
284 tk->rp.handler = kretprobe_dispatcher;
285 else
286 tk->rp.kp.pre_handler = kprobe_dispatcher;
287
288 tk->rp.maxactive = maxactive;
289 INIT_HLIST_NODE(&tk->rp.kp.hlist);
290 INIT_LIST_HEAD(&tk->rp.kp.list);
291
292 ret = trace_probe_init(&tk->tp, event, group, false);
293 if (ret < 0)
294 goto error;
295
296 dyn_event_init(&tk->devent, &trace_kprobe_ops);
297 return tk;
298 error:
299 free_trace_kprobe(tk);
300 return ERR_PTR(ret);
301 }
302
303 static struct trace_kprobe *find_trace_kprobe(const char *event,
304 const char *group)
305 {
306 struct dyn_event *pos;
307 struct trace_kprobe *tk;
308
309 for_each_trace_kprobe(tk, pos)
310 if (strcmp(trace_probe_name(&tk->tp), event) == 0 &&
311 strcmp(trace_probe_group_name(&tk->tp), group) == 0)
312 return tk;
313 return NULL;
314 }
315
316 static inline int __enable_trace_kprobe(struct trace_kprobe *tk)
317 {
318 int ret = 0;
319
320 if (trace_kprobe_is_registered(tk) && !trace_kprobe_has_gone(tk)) {
321 if (trace_kprobe_is_return(tk))
322 ret = enable_kretprobe(&tk->rp);
323 else
324 ret = enable_kprobe(&tk->rp.kp);
325 }
326
327 return ret;
328 }
329
330 static void __disable_trace_kprobe(struct trace_probe *tp)
331 {
332 struct trace_probe *pos;
333 struct trace_kprobe *tk;
334
335 list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
336 tk = container_of(pos, struct trace_kprobe, tp);
337 if (!trace_kprobe_is_registered(tk))
338 continue;
339 if (trace_kprobe_is_return(tk))
340 disable_kretprobe(&tk->rp);
341 else
342 disable_kprobe(&tk->rp.kp);
343 }
344 }
345
346 /*
347 * Enable trace_probe
348 * if the file is NULL, enable "perf" handler, or enable "trace" handler.
349 */
350 static int enable_trace_kprobe(struct trace_event_call *call,
351 struct trace_event_file *file)
352 {
353 struct trace_probe *pos, *tp;
354 struct trace_kprobe *tk;
355 bool enabled;
356 int ret = 0;
357
358 tp = trace_probe_primary_from_call(call);
359 if (WARN_ON_ONCE(!tp))
360 return -ENODEV;
361 enabled = trace_probe_is_enabled(tp);
362
363 /* This also changes "enabled" state */
364 if (file) {
365 ret = trace_probe_add_file(tp, file);
366 if (ret)
367 return ret;
368 } else
369 trace_probe_set_flag(tp, TP_FLAG_PROFILE);
370
371 if (enabled)
372 return 0;
373
374 list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
375 tk = container_of(pos, struct trace_kprobe, tp);
376 if (trace_kprobe_has_gone(tk))
377 continue;
378 ret = __enable_trace_kprobe(tk);
379 if (ret)
380 break;
381 enabled = true;
382 }
383
384 if (ret) {
385 /* Failed to enable one of them. Roll back all */
386 if (enabled)
387 __disable_trace_kprobe(tp);
388 if (file)
389 trace_probe_remove_file(tp, file);
390 else
391 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
392 }
393
394 return ret;
395 }
396
397 /*
398 * Disable trace_probe
399 * if the file is NULL, disable "perf" handler, or disable "trace" handler.
400 */
401 static int disable_trace_kprobe(struct trace_event_call *call,
402 struct trace_event_file *file)
403 {
404 struct trace_probe *tp;
405
406 tp = trace_probe_primary_from_call(call);
407 if (WARN_ON_ONCE(!tp))
408 return -ENODEV;
409
410 if (file) {
411 if (!trace_probe_get_file_link(tp, file))
412 return -ENOENT;
413 if (!trace_probe_has_single_file(tp))
414 goto out;
415 trace_probe_clear_flag(tp, TP_FLAG_TRACE);
416 } else
417 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
418
419 if (!trace_probe_is_enabled(tp))
420 __disable_trace_kprobe(tp);
421
422 out:
423 if (file)
424 /*
425 * Synchronization is done in below function. For perf event,
426 * file == NULL and perf_trace_event_unreg() calls
427 * tracepoint_synchronize_unregister() to ensure synchronize
428 * event. We don't need to care about it.
429 */
430 trace_probe_remove_file(tp, file);
431
432 return 0;
433 }
434
435 #if defined(CONFIG_KPROBES_ON_FTRACE) && \
436 !defined(CONFIG_KPROBE_EVENTS_ON_NOTRACE)
437 static bool __within_notrace_func(unsigned long addr)
438 {
439 unsigned long offset, size;
440
441 if (!addr || !kallsyms_lookup_size_offset(addr, &size, &offset))
442 return false;
443
444 /* Get the entry address of the target function */
445 addr -= offset;
446
447 /*
448 * Since ftrace_location_range() does inclusive range check, we need
449 * to subtract 1 byte from the end address.
450 */
451 return !ftrace_location_range(addr, addr + size - 1);
452 }
453
454 static bool within_notrace_func(struct trace_kprobe *tk)
455 {
456 unsigned long addr = trace_kprobe_address(tk);
457 char symname[KSYM_NAME_LEN], *p;
458
459 if (!__within_notrace_func(addr))
460 return false;
461
462 /* Check if the address is on a suffixed-symbol */
463 if (!lookup_symbol_name(addr, symname)) {
464 p = strchr(symname, '.');
465 if (!p)
466 return true;
467 *p = '\0';
468 addr = (unsigned long)kprobe_lookup_name(symname, 0);
469 if (addr)
470 return __within_notrace_func(addr);
471 }
472
473 return true;
474 }
475 #else
476 #define within_notrace_func(tk) (false)
477 #endif
478
479 /* Internal register function - just handle k*probes and flags */
480 static int __register_trace_kprobe(struct trace_kprobe *tk)
481 {
482 int i, ret;
483
484 ret = security_locked_down(LOCKDOWN_KPROBES);
485 if (ret)
486 return ret;
487
488 if (trace_kprobe_is_registered(tk))
489 return -EINVAL;
490
491 if (within_notrace_func(tk)) {
492 pr_warn("Could not probe notrace function %s\n",
493 trace_kprobe_symbol(tk));
494 return -EINVAL;
495 }
496
497 for (i = 0; i < tk->tp.nr_args; i++) {
498 ret = traceprobe_update_arg(&tk->tp.args[i]);
499 if (ret)
500 return ret;
501 }
502
503 /* Set/clear disabled flag according to tp->flag */
504 if (trace_probe_is_enabled(&tk->tp))
505 tk->rp.kp.flags &= ~KPROBE_FLAG_DISABLED;
506 else
507 tk->rp.kp.flags |= KPROBE_FLAG_DISABLED;
508
509 if (trace_kprobe_is_return(tk))
510 ret = register_kretprobe(&tk->rp);
511 else
512 ret = register_kprobe(&tk->rp.kp);
513
514 return ret;
515 }
516
517 /* Internal unregister function - just handle k*probes and flags */
518 static void __unregister_trace_kprobe(struct trace_kprobe *tk)
519 {
520 if (trace_kprobe_is_registered(tk)) {
521 if (trace_kprobe_is_return(tk))
522 unregister_kretprobe(&tk->rp);
523 else
524 unregister_kprobe(&tk->rp.kp);
525 /* Cleanup kprobe for reuse and mark it unregistered */
526 INIT_HLIST_NODE(&tk->rp.kp.hlist);
527 INIT_LIST_HEAD(&tk->rp.kp.list);
528 if (tk->rp.kp.symbol_name)
529 tk->rp.kp.addr = NULL;
530 }
531 }
532
533 /* Unregister a trace_probe and probe_event */
534 static int unregister_trace_kprobe(struct trace_kprobe *tk)
535 {
536 /* If other probes are on the event, just unregister kprobe */
537 if (trace_probe_has_sibling(&tk->tp))
538 goto unreg;
539
540 /* Enabled event can not be unregistered */
541 if (trace_probe_is_enabled(&tk->tp))
542 return -EBUSY;
543
544 /* Will fail if probe is being used by ftrace or perf */
545 if (unregister_kprobe_event(tk))
546 return -EBUSY;
547
548 unreg:
549 __unregister_trace_kprobe(tk);
550 dyn_event_remove(&tk->devent);
551 trace_probe_unlink(&tk->tp);
552
553 return 0;
554 }
555
556 static bool trace_kprobe_has_same_kprobe(struct trace_kprobe *orig,
557 struct trace_kprobe *comp)
558 {
559 struct trace_probe_event *tpe = orig->tp.event;
560 struct trace_probe *pos;
561 int i;
562
563 list_for_each_entry(pos, &tpe->probes, list) {
564 orig = container_of(pos, struct trace_kprobe, tp);
565 if (strcmp(trace_kprobe_symbol(orig),
566 trace_kprobe_symbol(comp)) ||
567 trace_kprobe_offset(orig) != trace_kprobe_offset(comp))
568 continue;
569
570 /*
571 * trace_probe_compare_arg_type() ensured that nr_args and
572 * each argument name and type are same. Let's compare comm.
573 */
574 for (i = 0; i < orig->tp.nr_args; i++) {
575 if (strcmp(orig->tp.args[i].comm,
576 comp->tp.args[i].comm))
577 break;
578 }
579
580 if (i == orig->tp.nr_args)
581 return true;
582 }
583
584 return false;
585 }
586
587 static int append_trace_kprobe(struct trace_kprobe *tk, struct trace_kprobe *to)
588 {
589 int ret;
590
591 ret = trace_probe_compare_arg_type(&tk->tp, &to->tp);
592 if (ret) {
593 /* Note that argument starts index = 2 */
594 trace_probe_log_set_index(ret + 1);
595 trace_probe_log_err(0, DIFF_ARG_TYPE);
596 return -EEXIST;
597 }
598 if (trace_kprobe_has_same_kprobe(to, tk)) {
599 trace_probe_log_set_index(0);
600 trace_probe_log_err(0, SAME_PROBE);
601 return -EEXIST;
602 }
603
604 /* Append to existing event */
605 ret = trace_probe_append(&tk->tp, &to->tp);
606 if (ret)
607 return ret;
608
609 /* Register k*probe */
610 ret = __register_trace_kprobe(tk);
611 if (ret == -ENOENT && !trace_kprobe_module_exist(tk)) {
612 pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
613 ret = 0;
614 }
615
616 if (ret)
617 trace_probe_unlink(&tk->tp);
618 else
619 dyn_event_add(&tk->devent);
620
621 return ret;
622 }
623
624 /* Register a trace_probe and probe_event */
625 static int register_trace_kprobe(struct trace_kprobe *tk)
626 {
627 struct trace_kprobe *old_tk;
628 int ret;
629
630 mutex_lock(&event_mutex);
631
632 old_tk = find_trace_kprobe(trace_probe_name(&tk->tp),
633 trace_probe_group_name(&tk->tp));
634 if (old_tk) {
635 if (trace_kprobe_is_return(tk) != trace_kprobe_is_return(old_tk)) {
636 trace_probe_log_set_index(0);
637 trace_probe_log_err(0, DIFF_PROBE_TYPE);
638 ret = -EEXIST;
639 } else {
640 ret = append_trace_kprobe(tk, old_tk);
641 }
642 goto end;
643 }
644
645 /* Register new event */
646 ret = register_kprobe_event(tk);
647 if (ret) {
648 pr_warn("Failed to register probe event(%d)\n", ret);
649 goto end;
650 }
651
652 /* Register k*probe */
653 ret = __register_trace_kprobe(tk);
654 if (ret == -ENOENT && !trace_kprobe_module_exist(tk)) {
655 pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
656 ret = 0;
657 }
658
659 if (ret < 0)
660 unregister_kprobe_event(tk);
661 else
662 dyn_event_add(&tk->devent);
663
664 end:
665 mutex_unlock(&event_mutex);
666 return ret;
667 }
668
669 /* Module notifier call back, checking event on the module */
670 static int trace_kprobe_module_callback(struct notifier_block *nb,
671 unsigned long val, void *data)
672 {
673 struct module *mod = data;
674 struct dyn_event *pos;
675 struct trace_kprobe *tk;
676 int ret;
677
678 if (val != MODULE_STATE_COMING)
679 return NOTIFY_DONE;
680
681 /* Update probes on coming module */
682 mutex_lock(&event_mutex);
683 for_each_trace_kprobe(tk, pos) {
684 if (trace_kprobe_within_module(tk, mod)) {
685 /* Don't need to check busy - this should have gone. */
686 __unregister_trace_kprobe(tk);
687 ret = __register_trace_kprobe(tk);
688 if (ret)
689 pr_warn("Failed to re-register probe %s on %s: %d\n",
690 trace_probe_name(&tk->tp),
691 mod->name, ret);
692 }
693 }
694 mutex_unlock(&event_mutex);
695
696 return NOTIFY_DONE;
697 }
698
699 static struct notifier_block trace_kprobe_module_nb = {
700 .notifier_call = trace_kprobe_module_callback,
701 .priority = 1 /* Invoked after kprobe module callback */
702 };
703
704 /* Convert certain expected symbols into '_' when generating event names */
705 static inline void sanitize_event_name(char *name)
706 {
707 while (*name++ != '\0')
708 if (*name == ':' || *name == '.')
709 *name = '_';
710 }
711
712 static int trace_kprobe_create(int argc, const char *argv[])
713 {
714 /*
715 * Argument syntax:
716 * - Add kprobe:
717 * p[:[GRP/]EVENT] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
718 * - Add kretprobe:
719 * r[MAXACTIVE][:[GRP/]EVENT] [MOD:]KSYM[+0] [FETCHARGS]
720 * Fetch args:
721 * $retval : fetch return value
722 * $stack : fetch stack address
723 * $stackN : fetch Nth of stack (N:0-)
724 * $comm : fetch current task comm
725 * @ADDR : fetch memory at ADDR (ADDR should be in kernel)
726 * @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
727 * %REG : fetch register REG
728 * Dereferencing memory fetch:
729 * +|-offs(ARG) : fetch memory at ARG +|- offs address.
730 * Alias name of args:
731 * NAME=FETCHARG : set NAME as alias of FETCHARG.
732 * Type of args:
733 * FETCHARG:TYPE : use TYPE instead of unsigned long.
734 */
735 struct trace_kprobe *tk = NULL;
736 int i, len, ret = 0;
737 bool is_return = false;
738 char *symbol = NULL, *tmp = NULL;
739 const char *event = NULL, *group = KPROBE_EVENT_SYSTEM;
740 int maxactive = 0;
741 long offset = 0;
742 void *addr = NULL;
743 char buf[MAX_EVENT_NAME_LEN];
744 unsigned int flags = TPARG_FL_KERNEL;
745
746 switch (argv[0][0]) {
747 case 'r':
748 is_return = true;
749 flags |= TPARG_FL_RETURN;
750 break;
751 case 'p':
752 break;
753 default:
754 return -ECANCELED;
755 }
756 if (argc < 2)
757 return -ECANCELED;
758
759 trace_probe_log_init("trace_kprobe", argc, argv);
760
761 event = strchr(&argv[0][1], ':');
762 if (event)
763 event++;
764
765 if (isdigit(argv[0][1])) {
766 if (!is_return) {
767 trace_probe_log_err(1, MAXACT_NO_KPROBE);
768 goto parse_error;
769 }
770 if (event)
771 len = event - &argv[0][1] - 1;
772 else
773 len = strlen(&argv[0][1]);
774 if (len > MAX_EVENT_NAME_LEN - 1) {
775 trace_probe_log_err(1, BAD_MAXACT);
776 goto parse_error;
777 }
778 memcpy(buf, &argv[0][1], len);
779 buf[len] = '\0';
780 ret = kstrtouint(buf, 0, &maxactive);
781 if (ret || !maxactive) {
782 trace_probe_log_err(1, BAD_MAXACT);
783 goto parse_error;
784 }
785 /* kretprobes instances are iterated over via a list. The
786 * maximum should stay reasonable.
787 */
788 if (maxactive > KRETPROBE_MAXACTIVE_MAX) {
789 trace_probe_log_err(1, MAXACT_TOO_BIG);
790 goto parse_error;
791 }
792 }
793
794 /* try to parse an address. if that fails, try to read the
795 * input as a symbol. */
796 if (kstrtoul(argv[1], 0, (unsigned long *)&addr)) {
797 trace_probe_log_set_index(1);
798 /* Check whether uprobe event specified */
799 if (strchr(argv[1], '/') && strchr(argv[1], ':')) {
800 ret = -ECANCELED;
801 goto error;
802 }
803 /* a symbol specified */
804 symbol = kstrdup(argv[1], GFP_KERNEL);
805 if (!symbol)
806 return -ENOMEM;
807 /* TODO: support .init module functions */
808 ret = traceprobe_split_symbol_offset(symbol, &offset);
809 if (ret || offset < 0 || offset > UINT_MAX) {
810 trace_probe_log_err(0, BAD_PROBE_ADDR);
811 goto parse_error;
812 }
813 if (kprobe_on_func_entry(NULL, symbol, offset))
814 flags |= TPARG_FL_FENTRY;
815 if (offset && is_return && !(flags & TPARG_FL_FENTRY)) {
816 trace_probe_log_err(0, BAD_RETPROBE);
817 goto parse_error;
818 }
819 }
820
821 trace_probe_log_set_index(0);
822 if (event) {
823 ret = traceprobe_parse_event_name(&event, &group, buf,
824 event - argv[0]);
825 if (ret)
826 goto parse_error;
827 } else {
828 /* Make a new event name */
829 if (symbol)
830 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
831 is_return ? 'r' : 'p', symbol, offset);
832 else
833 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
834 is_return ? 'r' : 'p', addr);
835 sanitize_event_name(buf);
836 event = buf;
837 }
838
839 /* setup a probe */
840 tk = alloc_trace_kprobe(group, event, addr, symbol, offset, maxactive,
841 argc - 2, is_return);
842 if (IS_ERR(tk)) {
843 ret = PTR_ERR(tk);
844 /* This must return -ENOMEM, else there is a bug */
845 WARN_ON_ONCE(ret != -ENOMEM);
846 goto out; /* We know tk is not allocated */
847 }
848 argc -= 2; argv += 2;
849
850 /* parse arguments */
851 for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
852 tmp = kstrdup(argv[i], GFP_KERNEL);
853 if (!tmp) {
854 ret = -ENOMEM;
855 goto error;
856 }
857
858 trace_probe_log_set_index(i + 2);
859 ret = traceprobe_parse_probe_arg(&tk->tp, i, tmp, flags);
860 kfree(tmp);
861 if (ret)
862 goto error; /* This can be -ENOMEM */
863 }
864
865 ret = traceprobe_set_print_fmt(&tk->tp, is_return);
866 if (ret < 0)
867 goto error;
868
869 ret = register_trace_kprobe(tk);
870 if (ret) {
871 trace_probe_log_set_index(1);
872 if (ret == -EILSEQ)
873 trace_probe_log_err(0, BAD_INSN_BNDRY);
874 else if (ret == -ENOENT)
875 trace_probe_log_err(0, BAD_PROBE_ADDR);
876 else if (ret != -ENOMEM && ret != -EEXIST)
877 trace_probe_log_err(0, FAIL_REG_PROBE);
878 goto error;
879 }
880
881 out:
882 trace_probe_log_clear();
883 kfree(symbol);
884 return ret;
885
886 parse_error:
887 ret = -EINVAL;
888 error:
889 free_trace_kprobe(tk);
890 goto out;
891 }
892
893 static int create_or_delete_trace_kprobe(int argc, char **argv)
894 {
895 int ret;
896
897 if (argv[0][0] == '-')
898 return dyn_event_release(argc, argv, &trace_kprobe_ops);
899
900 ret = trace_kprobe_create(argc, (const char **)argv);
901 return ret == -ECANCELED ? -EINVAL : ret;
902 }
903
904 static int trace_kprobe_run_command(struct dynevent_cmd *cmd)
905 {
906 return trace_run_command(cmd->seq.buffer, create_or_delete_trace_kprobe);
907 }
908
909 /**
910 * kprobe_event_cmd_init - Initialize a kprobe event command object
911 * @cmd: A pointer to the dynevent_cmd struct representing the new event
912 * @buf: A pointer to the buffer used to build the command
913 * @maxlen: The length of the buffer passed in @buf
914 *
915 * Initialize a synthetic event command object. Use this before
916 * calling any of the other kprobe_event functions.
917 */
918 void kprobe_event_cmd_init(struct dynevent_cmd *cmd, char *buf, int maxlen)
919 {
920 dynevent_cmd_init(cmd, buf, maxlen, DYNEVENT_TYPE_KPROBE,
921 trace_kprobe_run_command);
922 }
923 EXPORT_SYMBOL_GPL(kprobe_event_cmd_init);
924
925 /**
926 * __kprobe_event_gen_cmd_start - Generate a kprobe event command from arg list
927 * @cmd: A pointer to the dynevent_cmd struct representing the new event
928 * @name: The name of the kprobe event
929 * @loc: The location of the kprobe event
930 * @kretprobe: Is this a return probe?
931 * @args: Variable number of arg (pairs), one pair for each field
932 *
933 * NOTE: Users normally won't want to call this function directly, but
934 * rather use the kprobe_event_gen_cmd_start() wrapper, which automatically
935 * adds a NULL to the end of the arg list. If this function is used
936 * directly, make sure the last arg in the variable arg list is NULL.
937 *
938 * Generate a kprobe event command to be executed by
939 * kprobe_event_gen_cmd_end(). This function can be used to generate the
940 * complete command or only the first part of it; in the latter case,
941 * kprobe_event_add_fields() can be used to add more fields following this.
942 *
943 * Unlikely the synth_event_gen_cmd_start(), @loc must be specified. This
944 * returns -EINVAL if @loc == NULL.
945 *
946 * Return: 0 if successful, error otherwise.
947 */
948 int __kprobe_event_gen_cmd_start(struct dynevent_cmd *cmd, bool kretprobe,
949 const char *name, const char *loc, ...)
950 {
951 char buf[MAX_EVENT_NAME_LEN];
952 struct dynevent_arg arg;
953 va_list args;
954 int ret;
955
956 if (cmd->type != DYNEVENT_TYPE_KPROBE)
957 return -EINVAL;
958
959 if (!loc)
960 return -EINVAL;
961
962 if (kretprobe)
963 snprintf(buf, MAX_EVENT_NAME_LEN, "r:kprobes/%s", name);
964 else
965 snprintf(buf, MAX_EVENT_NAME_LEN, "p:kprobes/%s", name);
966
967 ret = dynevent_str_add(cmd, buf);
968 if (ret)
969 return ret;
970
971 dynevent_arg_init(&arg, 0);
972 arg.str = loc;
973 ret = dynevent_arg_add(cmd, &arg, NULL);
974 if (ret)
975 return ret;
976
977 va_start(args, loc);
978 for (;;) {
979 const char *field;
980
981 field = va_arg(args, const char *);
982 if (!field)
983 break;
984
985 if (++cmd->n_fields > MAX_TRACE_ARGS) {
986 ret = -EINVAL;
987 break;
988 }
989
990 arg.str = field;
991 ret = dynevent_arg_add(cmd, &arg, NULL);
992 if (ret)
993 break;
994 }
995 va_end(args);
996
997 return ret;
998 }
999 EXPORT_SYMBOL_GPL(__kprobe_event_gen_cmd_start);
1000
1001 /**
1002 * __kprobe_event_add_fields - Add probe fields to a kprobe command from arg list
1003 * @cmd: A pointer to the dynevent_cmd struct representing the new event
1004 * @args: Variable number of arg (pairs), one pair for each field
1005 *
1006 * NOTE: Users normally won't want to call this function directly, but
1007 * rather use the kprobe_event_add_fields() wrapper, which
1008 * automatically adds a NULL to the end of the arg list. If this
1009 * function is used directly, make sure the last arg in the variable
1010 * arg list is NULL.
1011 *
1012 * Add probe fields to an existing kprobe command using a variable
1013 * list of args. Fields are added in the same order they're listed.
1014 *
1015 * Return: 0 if successful, error otherwise.
1016 */
1017 int __kprobe_event_add_fields(struct dynevent_cmd *cmd, ...)
1018 {
1019 struct dynevent_arg arg;
1020 va_list args;
1021 int ret = 0;
1022
1023 if (cmd->type != DYNEVENT_TYPE_KPROBE)
1024 return -EINVAL;
1025
1026 dynevent_arg_init(&arg, 0);
1027
1028 va_start(args, cmd);
1029 for (;;) {
1030 const char *field;
1031
1032 field = va_arg(args, const char *);
1033 if (!field)
1034 break;
1035
1036 if (++cmd->n_fields > MAX_TRACE_ARGS) {
1037 ret = -EINVAL;
1038 break;
1039 }
1040
1041 arg.str = field;
1042 ret = dynevent_arg_add(cmd, &arg, NULL);
1043 if (ret)
1044 break;
1045 }
1046 va_end(args);
1047
1048 return ret;
1049 }
1050 EXPORT_SYMBOL_GPL(__kprobe_event_add_fields);
1051
1052 /**
1053 * kprobe_event_delete - Delete a kprobe event
1054 * @name: The name of the kprobe event to delete
1055 *
1056 * Delete a kprobe event with the give @name from kernel code rather
1057 * than directly from the command line.
1058 *
1059 * Return: 0 if successful, error otherwise.
1060 */
1061 int kprobe_event_delete(const char *name)
1062 {
1063 char buf[MAX_EVENT_NAME_LEN];
1064
1065 snprintf(buf, MAX_EVENT_NAME_LEN, "-:%s", name);
1066
1067 return trace_run_command(buf, create_or_delete_trace_kprobe);
1068 }
1069 EXPORT_SYMBOL_GPL(kprobe_event_delete);
1070
1071 static int trace_kprobe_release(struct dyn_event *ev)
1072 {
1073 struct trace_kprobe *tk = to_trace_kprobe(ev);
1074 int ret = unregister_trace_kprobe(tk);
1075
1076 if (!ret)
1077 free_trace_kprobe(tk);
1078 return ret;
1079 }
1080
1081 static int trace_kprobe_show(struct seq_file *m, struct dyn_event *ev)
1082 {
1083 struct trace_kprobe *tk = to_trace_kprobe(ev);
1084 int i;
1085
1086 seq_putc(m, trace_kprobe_is_return(tk) ? 'r' : 'p');
1087 if (trace_kprobe_is_return(tk) && tk->rp.maxactive)
1088 seq_printf(m, "%d", tk->rp.maxactive);
1089 seq_printf(m, ":%s/%s", trace_probe_group_name(&tk->tp),
1090 trace_probe_name(&tk->tp));
1091
1092 if (!tk->symbol)
1093 seq_printf(m, " 0x%p", tk->rp.kp.addr);
1094 else if (tk->rp.kp.offset)
1095 seq_printf(m, " %s+%u", trace_kprobe_symbol(tk),
1096 tk->rp.kp.offset);
1097 else
1098 seq_printf(m, " %s", trace_kprobe_symbol(tk));
1099
1100 for (i = 0; i < tk->tp.nr_args; i++)
1101 seq_printf(m, " %s=%s", tk->tp.args[i].name, tk->tp.args[i].comm);
1102 seq_putc(m, '\n');
1103
1104 return 0;
1105 }
1106
1107 static int probes_seq_show(struct seq_file *m, void *v)
1108 {
1109 struct dyn_event *ev = v;
1110
1111 if (!is_trace_kprobe(ev))
1112 return 0;
1113
1114 return trace_kprobe_show(m, ev);
1115 }
1116
1117 static const struct seq_operations probes_seq_op = {
1118 .start = dyn_event_seq_start,
1119 .next = dyn_event_seq_next,
1120 .stop = dyn_event_seq_stop,
1121 .show = probes_seq_show
1122 };
1123
1124 static int probes_open(struct inode *inode, struct file *file)
1125 {
1126 int ret;
1127
1128 ret = security_locked_down(LOCKDOWN_TRACEFS);
1129 if (ret)
1130 return ret;
1131
1132 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
1133 ret = dyn_events_release_all(&trace_kprobe_ops);
1134 if (ret < 0)
1135 return ret;
1136 }
1137
1138 return seq_open(file, &probes_seq_op);
1139 }
1140
1141 static ssize_t probes_write(struct file *file, const char __user *buffer,
1142 size_t count, loff_t *ppos)
1143 {
1144 return trace_parse_run_command(file, buffer, count, ppos,
1145 create_or_delete_trace_kprobe);
1146 }
1147
1148 static const struct file_operations kprobe_events_ops = {
1149 .owner = THIS_MODULE,
1150 .open = probes_open,
1151 .read = seq_read,
1152 .llseek = seq_lseek,
1153 .release = seq_release,
1154 .write = probes_write,
1155 };
1156
1157 /* Probes profiling interfaces */
1158 static int probes_profile_seq_show(struct seq_file *m, void *v)
1159 {
1160 struct dyn_event *ev = v;
1161 struct trace_kprobe *tk;
1162
1163 if (!is_trace_kprobe(ev))
1164 return 0;
1165
1166 tk = to_trace_kprobe(ev);
1167 seq_printf(m, " %-44s %15lu %15lu\n",
1168 trace_probe_name(&tk->tp),
1169 trace_kprobe_nhit(tk),
1170 tk->rp.kp.nmissed);
1171
1172 return 0;
1173 }
1174
1175 static const struct seq_operations profile_seq_op = {
1176 .start = dyn_event_seq_start,
1177 .next = dyn_event_seq_next,
1178 .stop = dyn_event_seq_stop,
1179 .show = probes_profile_seq_show
1180 };
1181
1182 static int profile_open(struct inode *inode, struct file *file)
1183 {
1184 int ret;
1185
1186 ret = security_locked_down(LOCKDOWN_TRACEFS);
1187 if (ret)
1188 return ret;
1189
1190 return seq_open(file, &profile_seq_op);
1191 }
1192
1193 static const struct file_operations kprobe_profile_ops = {
1194 .owner = THIS_MODULE,
1195 .open = profile_open,
1196 .read = seq_read,
1197 .llseek = seq_lseek,
1198 .release = seq_release,
1199 };
1200
1201 /* Kprobe specific fetch functions */
1202
1203 /* Return the length of string -- including null terminal byte */
1204 static nokprobe_inline int
1205 fetch_store_strlen(unsigned long addr)
1206 {
1207 int ret, len = 0;
1208 u8 c;
1209
1210 do {
1211 ret = probe_kernel_read(&c, (u8 *)addr + len, 1);
1212 len++;
1213 } while (c && ret == 0 && len < MAX_STRING_SIZE);
1214
1215 return (ret < 0) ? ret : len;
1216 }
1217
1218 /* Return the length of string -- including null terminal byte */
1219 static nokprobe_inline int
1220 fetch_store_strlen_user(unsigned long addr)
1221 {
1222 const void __user *uaddr = (__force const void __user *)addr;
1223
1224 return strnlen_unsafe_user(uaddr, MAX_STRING_SIZE);
1225 }
1226
1227 /*
1228 * Fetch a null-terminated string. Caller MUST set *(u32 *)buf with max
1229 * length and relative data location.
1230 */
1231 static nokprobe_inline int
1232 fetch_store_string(unsigned long addr, void *dest, void *base)
1233 {
1234 int maxlen = get_loc_len(*(u32 *)dest);
1235 void *__dest;
1236 long ret;
1237
1238 if (unlikely(!maxlen))
1239 return -ENOMEM;
1240
1241 __dest = get_loc_data(dest, base);
1242
1243 /*
1244 * Try to get string again, since the string can be changed while
1245 * probing.
1246 */
1247 ret = strncpy_from_unsafe(__dest, (void *)addr, maxlen);
1248 if (ret >= 0)
1249 *(u32 *)dest = make_data_loc(ret, __dest - base);
1250
1251 return ret;
1252 }
1253
1254 /*
1255 * Fetch a null-terminated string from user. Caller MUST set *(u32 *)buf
1256 * with max length and relative data location.
1257 */
1258 static nokprobe_inline int
1259 fetch_store_string_user(unsigned long addr, void *dest, void *base)
1260 {
1261 const void __user *uaddr = (__force const void __user *)addr;
1262 int maxlen = get_loc_len(*(u32 *)dest);
1263 void *__dest;
1264 long ret;
1265
1266 if (unlikely(!maxlen))
1267 return -ENOMEM;
1268
1269 __dest = get_loc_data(dest, base);
1270
1271 ret = strncpy_from_unsafe_user(__dest, uaddr, maxlen);
1272 if (ret >= 0)
1273 *(u32 *)dest = make_data_loc(ret, __dest - base);
1274
1275 return ret;
1276 }
1277
1278 static nokprobe_inline int
1279 probe_mem_read(void *dest, void *src, size_t size)
1280 {
1281 return probe_kernel_read(dest, src, size);
1282 }
1283
1284 static nokprobe_inline int
1285 probe_mem_read_user(void *dest, void *src, size_t size)
1286 {
1287 const void __user *uaddr = (__force const void __user *)src;
1288
1289 return probe_user_read(dest, uaddr, size);
1290 }
1291
1292 /* Note that we don't verify it, since the code does not come from user space */
1293 static int
1294 process_fetch_insn(struct fetch_insn *code, struct pt_regs *regs, void *dest,
1295 void *base)
1296 {
1297 unsigned long val;
1298
1299 retry:
1300 /* 1st stage: get value from context */
1301 switch (code->op) {
1302 case FETCH_OP_REG:
1303 val = regs_get_register(regs, code->param);
1304 break;
1305 case FETCH_OP_STACK:
1306 val = regs_get_kernel_stack_nth(regs, code->param);
1307 break;
1308 case FETCH_OP_STACKP:
1309 val = kernel_stack_pointer(regs);
1310 break;
1311 case FETCH_OP_RETVAL:
1312 val = regs_return_value(regs);
1313 break;
1314 case FETCH_OP_IMM:
1315 val = code->immediate;
1316 break;
1317 case FETCH_OP_COMM:
1318 val = (unsigned long)current->comm;
1319 break;
1320 case FETCH_OP_DATA:
1321 val = (unsigned long)code->data;
1322 break;
1323 #ifdef CONFIG_HAVE_FUNCTION_ARG_ACCESS_API
1324 case FETCH_OP_ARG:
1325 val = regs_get_kernel_argument(regs, code->param);
1326 break;
1327 #endif
1328 case FETCH_NOP_SYMBOL: /* Ignore a place holder */
1329 code++;
1330 goto retry;
1331 default:
1332 return -EILSEQ;
1333 }
1334 code++;
1335
1336 return process_fetch_insn_bottom(code, val, dest, base);
1337 }
1338 NOKPROBE_SYMBOL(process_fetch_insn)
1339
1340 /* Kprobe handler */
1341 static nokprobe_inline void
1342 __kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs,
1343 struct trace_event_file *trace_file)
1344 {
1345 struct kprobe_trace_entry_head *entry;
1346 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1347 struct trace_event_buffer fbuffer;
1348 int dsize;
1349
1350 WARN_ON(call != trace_file->event_call);
1351
1352 if (trace_trigger_soft_disabled(trace_file))
1353 return;
1354
1355 local_save_flags(fbuffer.flags);
1356 fbuffer.pc = preempt_count();
1357 fbuffer.trace_file = trace_file;
1358
1359 dsize = __get_data_size(&tk->tp, regs);
1360
1361 fbuffer.event =
1362 trace_event_buffer_lock_reserve(&fbuffer.buffer, trace_file,
1363 call->event.type,
1364 sizeof(*entry) + tk->tp.size + dsize,
1365 fbuffer.flags, fbuffer.pc);
1366 if (!fbuffer.event)
1367 return;
1368
1369 fbuffer.regs = regs;
1370 entry = fbuffer.entry = ring_buffer_event_data(fbuffer.event);
1371 entry->ip = (unsigned long)tk->rp.kp.addr;
1372 store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1373
1374 trace_event_buffer_commit(&fbuffer);
1375 }
1376
1377 static void
1378 kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs)
1379 {
1380 struct event_file_link *link;
1381
1382 trace_probe_for_each_link_rcu(link, &tk->tp)
1383 __kprobe_trace_func(tk, regs, link->file);
1384 }
1385 NOKPROBE_SYMBOL(kprobe_trace_func);
1386
1387 /* Kretprobe handler */
1388 static nokprobe_inline void
1389 __kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1390 struct pt_regs *regs,
1391 struct trace_event_file *trace_file)
1392 {
1393 struct kretprobe_trace_entry_head *entry;
1394 struct trace_event_buffer fbuffer;
1395 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1396 int dsize;
1397
1398 WARN_ON(call != trace_file->event_call);
1399
1400 if (trace_trigger_soft_disabled(trace_file))
1401 return;
1402
1403 local_save_flags(fbuffer.flags);
1404 fbuffer.pc = preempt_count();
1405 fbuffer.trace_file = trace_file;
1406
1407 dsize = __get_data_size(&tk->tp, regs);
1408 fbuffer.event =
1409 trace_event_buffer_lock_reserve(&fbuffer.buffer, trace_file,
1410 call->event.type,
1411 sizeof(*entry) + tk->tp.size + dsize,
1412 fbuffer.flags, fbuffer.pc);
1413 if (!fbuffer.event)
1414 return;
1415
1416 fbuffer.regs = regs;
1417 entry = fbuffer.entry = ring_buffer_event_data(fbuffer.event);
1418 entry->func = (unsigned long)tk->rp.kp.addr;
1419 entry->ret_ip = (unsigned long)ri->ret_addr;
1420 store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1421
1422 trace_event_buffer_commit(&fbuffer);
1423 }
1424
1425 static void
1426 kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1427 struct pt_regs *regs)
1428 {
1429 struct event_file_link *link;
1430
1431 trace_probe_for_each_link_rcu(link, &tk->tp)
1432 __kretprobe_trace_func(tk, ri, regs, link->file);
1433 }
1434 NOKPROBE_SYMBOL(kretprobe_trace_func);
1435
1436 /* Event entry printers */
1437 static enum print_line_t
1438 print_kprobe_event(struct trace_iterator *iter, int flags,
1439 struct trace_event *event)
1440 {
1441 struct kprobe_trace_entry_head *field;
1442 struct trace_seq *s = &iter->seq;
1443 struct trace_probe *tp;
1444
1445 field = (struct kprobe_trace_entry_head *)iter->ent;
1446 tp = trace_probe_primary_from_call(
1447 container_of(event, struct trace_event_call, event));
1448 if (WARN_ON_ONCE(!tp))
1449 goto out;
1450
1451 trace_seq_printf(s, "%s: (", trace_probe_name(tp));
1452
1453 if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
1454 goto out;
1455
1456 trace_seq_putc(s, ')');
1457
1458 if (print_probe_args(s, tp->args, tp->nr_args,
1459 (u8 *)&field[1], field) < 0)
1460 goto out;
1461
1462 trace_seq_putc(s, '\n');
1463 out:
1464 return trace_handle_return(s);
1465 }
1466
1467 static enum print_line_t
1468 print_kretprobe_event(struct trace_iterator *iter, int flags,
1469 struct trace_event *event)
1470 {
1471 struct kretprobe_trace_entry_head *field;
1472 struct trace_seq *s = &iter->seq;
1473 struct trace_probe *tp;
1474
1475 field = (struct kretprobe_trace_entry_head *)iter->ent;
1476 tp = trace_probe_primary_from_call(
1477 container_of(event, struct trace_event_call, event));
1478 if (WARN_ON_ONCE(!tp))
1479 goto out;
1480
1481 trace_seq_printf(s, "%s: (", trace_probe_name(tp));
1482
1483 if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1484 goto out;
1485
1486 trace_seq_puts(s, " <- ");
1487
1488 if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1489 goto out;
1490
1491 trace_seq_putc(s, ')');
1492
1493 if (print_probe_args(s, tp->args, tp->nr_args,
1494 (u8 *)&field[1], field) < 0)
1495 goto out;
1496
1497 trace_seq_putc(s, '\n');
1498
1499 out:
1500 return trace_handle_return(s);
1501 }
1502
1503
1504 static int kprobe_event_define_fields(struct trace_event_call *event_call)
1505 {
1506 int ret;
1507 struct kprobe_trace_entry_head field;
1508 struct trace_probe *tp;
1509
1510 tp = trace_probe_primary_from_call(event_call);
1511 if (WARN_ON_ONCE(!tp))
1512 return -ENOENT;
1513
1514 DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1515
1516 return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
1517 }
1518
1519 static int kretprobe_event_define_fields(struct trace_event_call *event_call)
1520 {
1521 int ret;
1522 struct kretprobe_trace_entry_head field;
1523 struct trace_probe *tp;
1524
1525 tp = trace_probe_primary_from_call(event_call);
1526 if (WARN_ON_ONCE(!tp))
1527 return -ENOENT;
1528
1529 DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1530 DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1531
1532 return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
1533 }
1534
1535 #ifdef CONFIG_PERF_EVENTS
1536
1537 /* Kprobe profile handler */
1538 static int
1539 kprobe_perf_func(struct trace_kprobe *tk, struct pt_regs *regs)
1540 {
1541 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1542 struct kprobe_trace_entry_head *entry;
1543 struct hlist_head *head;
1544 int size, __size, dsize;
1545 int rctx;
1546
1547 if (bpf_prog_array_valid(call)) {
1548 unsigned long orig_ip = instruction_pointer(regs);
1549 int ret;
1550
1551 ret = trace_call_bpf(call, regs);
1552
1553 /*
1554 * We need to check and see if we modified the pc of the
1555 * pt_regs, and if so return 1 so that we don't do the
1556 * single stepping.
1557 */
1558 if (orig_ip != instruction_pointer(regs))
1559 return 1;
1560 if (!ret)
1561 return 0;
1562 }
1563
1564 head = this_cpu_ptr(call->perf_events);
1565 if (hlist_empty(head))
1566 return 0;
1567
1568 dsize = __get_data_size(&tk->tp, regs);
1569 __size = sizeof(*entry) + tk->tp.size + dsize;
1570 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1571 size -= sizeof(u32);
1572
1573 entry = perf_trace_buf_alloc(size, NULL, &rctx);
1574 if (!entry)
1575 return 0;
1576
1577 entry->ip = (unsigned long)tk->rp.kp.addr;
1578 memset(&entry[1], 0, dsize);
1579 store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1580 perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1581 head, NULL);
1582 return 0;
1583 }
1584 NOKPROBE_SYMBOL(kprobe_perf_func);
1585
1586 /* Kretprobe profile handler */
1587 static void
1588 kretprobe_perf_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1589 struct pt_regs *regs)
1590 {
1591 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1592 struct kretprobe_trace_entry_head *entry;
1593 struct hlist_head *head;
1594 int size, __size, dsize;
1595 int rctx;
1596
1597 if (bpf_prog_array_valid(call) && !trace_call_bpf(call, regs))
1598 return;
1599
1600 head = this_cpu_ptr(call->perf_events);
1601 if (hlist_empty(head))
1602 return;
1603
1604 dsize = __get_data_size(&tk->tp, regs);
1605 __size = sizeof(*entry) + tk->tp.size + dsize;
1606 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1607 size -= sizeof(u32);
1608
1609 entry = perf_trace_buf_alloc(size, NULL, &rctx);
1610 if (!entry)
1611 return;
1612
1613 entry->func = (unsigned long)tk->rp.kp.addr;
1614 entry->ret_ip = (unsigned long)ri->ret_addr;
1615 store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1616 perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1617 head, NULL);
1618 }
1619 NOKPROBE_SYMBOL(kretprobe_perf_func);
1620
1621 int bpf_get_kprobe_info(const struct perf_event *event, u32 *fd_type,
1622 const char **symbol, u64 *probe_offset,
1623 u64 *probe_addr, bool perf_type_tracepoint)
1624 {
1625 const char *pevent = trace_event_name(event->tp_event);
1626 const char *group = event->tp_event->class->system;
1627 struct trace_kprobe *tk;
1628
1629 if (perf_type_tracepoint)
1630 tk = find_trace_kprobe(pevent, group);
1631 else
1632 tk = event->tp_event->data;
1633 if (!tk)
1634 return -EINVAL;
1635
1636 *fd_type = trace_kprobe_is_return(tk) ? BPF_FD_TYPE_KRETPROBE
1637 : BPF_FD_TYPE_KPROBE;
1638 if (tk->symbol) {
1639 *symbol = tk->symbol;
1640 *probe_offset = tk->rp.kp.offset;
1641 *probe_addr = 0;
1642 } else {
1643 *symbol = NULL;
1644 *probe_offset = 0;
1645 *probe_addr = (unsigned long)tk->rp.kp.addr;
1646 }
1647 return 0;
1648 }
1649 #endif /* CONFIG_PERF_EVENTS */
1650
1651 /*
1652 * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex.
1653 *
1654 * kprobe_trace_self_tests_init() does enable_trace_probe/disable_trace_probe
1655 * lockless, but we can't race with this __init function.
1656 */
1657 static int kprobe_register(struct trace_event_call *event,
1658 enum trace_reg type, void *data)
1659 {
1660 struct trace_event_file *file = data;
1661
1662 switch (type) {
1663 case TRACE_REG_REGISTER:
1664 return enable_trace_kprobe(event, file);
1665 case TRACE_REG_UNREGISTER:
1666 return disable_trace_kprobe(event, file);
1667
1668 #ifdef CONFIG_PERF_EVENTS
1669 case TRACE_REG_PERF_REGISTER:
1670 return enable_trace_kprobe(event, NULL);
1671 case TRACE_REG_PERF_UNREGISTER:
1672 return disable_trace_kprobe(event, NULL);
1673 case TRACE_REG_PERF_OPEN:
1674 case TRACE_REG_PERF_CLOSE:
1675 case TRACE_REG_PERF_ADD:
1676 case TRACE_REG_PERF_DEL:
1677 return 0;
1678 #endif
1679 }
1680 return 0;
1681 }
1682
1683 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1684 {
1685 struct trace_kprobe *tk = container_of(kp, struct trace_kprobe, rp.kp);
1686 int ret = 0;
1687
1688 raw_cpu_inc(*tk->nhit);
1689
1690 if (trace_probe_test_flag(&tk->tp, TP_FLAG_TRACE))
1691 kprobe_trace_func(tk, regs);
1692 #ifdef CONFIG_PERF_EVENTS
1693 if (trace_probe_test_flag(&tk->tp, TP_FLAG_PROFILE))
1694 ret = kprobe_perf_func(tk, regs);
1695 #endif
1696 return ret;
1697 }
1698 NOKPROBE_SYMBOL(kprobe_dispatcher);
1699
1700 static int
1701 kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1702 {
1703 struct trace_kprobe *tk = container_of(ri->rp, struct trace_kprobe, rp);
1704
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 tweek kernel, so just return 0 */
1714 }
1715 NOKPROBE_SYMBOL(kretprobe_dispatcher);
1716
1717 static struct trace_event_functions kretprobe_funcs = {
1718 .trace = print_kretprobe_event
1719 };
1720
1721 static struct trace_event_functions kprobe_funcs = {
1722 .trace = print_kprobe_event
1723 };
1724
1725 static struct trace_event_fields kretprobe_fields_array[] = {
1726 { .type = TRACE_FUNCTION_TYPE,
1727 .define_fields = kretprobe_event_define_fields },
1728 {}
1729 };
1730
1731 static struct trace_event_fields kprobe_fields_array[] = {
1732 { .type = TRACE_FUNCTION_TYPE,
1733 .define_fields = kprobe_event_define_fields },
1734 {}
1735 };
1736
1737 static 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
1753 static 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
1760 static 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 /* create a trace_kprobe, but don't add it to global lists */
1767 struct trace_event_call *
1768 create_local_trace_kprobe(char *func, void *addr, unsigned long offs,
1769 bool is_return)
1770 {
1771 struct trace_kprobe *tk;
1772 int ret;
1773 char *event;
1774
1775 /*
1776 * local trace_kprobes are not added to dyn_event, so they are never
1777 * searched in find_trace_kprobe(). Therefore, there is no concern of
1778 * duplicated name here.
1779 */
1780 event = func ? func : "DUMMY_EVENT";
1781
1782 tk = alloc_trace_kprobe(KPROBE_EVENT_SYSTEM, event, (void *)addr, func,
1783 offs, 0 /* maxactive */, 0 /* nargs */,
1784 is_return);
1785
1786 if (IS_ERR(tk)) {
1787 pr_info("Failed to allocate trace_probe.(%d)\n",
1788 (int)PTR_ERR(tk));
1789 return ERR_CAST(tk);
1790 }
1791
1792 init_trace_event_call(tk);
1793
1794 if (traceprobe_set_print_fmt(&tk->tp, trace_kprobe_is_return(tk)) < 0) {
1795 ret = -ENOMEM;
1796 goto error;
1797 }
1798
1799 ret = __register_trace_kprobe(tk);
1800 if (ret < 0)
1801 goto error;
1802
1803 return trace_probe_event_call(&tk->tp);
1804 error:
1805 free_trace_kprobe(tk);
1806 return ERR_PTR(ret);
1807 }
1808
1809 void destroy_local_trace_kprobe(struct trace_event_call *event_call)
1810 {
1811 struct trace_kprobe *tk;
1812
1813 tk = trace_kprobe_primary_from_call(event_call);
1814 if (unlikely(!tk))
1815 return;
1816
1817 if (trace_probe_is_enabled(&tk->tp)) {
1818 WARN_ON(1);
1819 return;
1820 }
1821
1822 __unregister_trace_kprobe(tk);
1823
1824 free_trace_kprobe(tk);
1825 }
1826 #endif /* CONFIG_PERF_EVENTS */
1827
1828 static __init void enable_boot_kprobe_events(void)
1829 {
1830 struct trace_array *tr = top_trace_array();
1831 struct trace_event_file *file;
1832 struct trace_kprobe *tk;
1833 struct dyn_event *pos;
1834
1835 mutex_lock(&event_mutex);
1836 for_each_trace_kprobe(tk, pos) {
1837 list_for_each_entry(file, &tr->events, list)
1838 if (file->event_call == trace_probe_event_call(&tk->tp))
1839 trace_event_enable_disable(file, 1, 0);
1840 }
1841 mutex_unlock(&event_mutex);
1842 }
1843
1844 static __init void setup_boot_kprobe_events(void)
1845 {
1846 char *p, *cmd = kprobe_boot_events_buf;
1847 int ret;
1848
1849 strreplace(kprobe_boot_events_buf, ',', ' ');
1850
1851 while (cmd && *cmd != '\0') {
1852 p = strchr(cmd, ';');
1853 if (p)
1854 *p++ = '\0';
1855
1856 ret = trace_run_command(cmd, create_or_delete_trace_kprobe);
1857 if (ret)
1858 pr_warn("Failed to add event(%d): %s\n", ret, cmd);
1859 else
1860 kprobe_boot_events_enabled = true;
1861
1862 cmd = p;
1863 }
1864
1865 enable_boot_kprobe_events();
1866 }
1867
1868 /*
1869 * Register dynevent at subsys_initcall. This allows kernel to setup kprobe
1870 * events in fs_initcall without tracefs.
1871 */
1872 static __init int init_kprobe_trace_early(void)
1873 {
1874 int ret;
1875
1876 ret = dyn_event_register(&trace_kprobe_ops);
1877 if (ret)
1878 return ret;
1879
1880 if (register_module_notifier(&trace_kprobe_module_nb))
1881 return -EINVAL;
1882
1883 return 0;
1884 }
1885 subsys_initcall(init_kprobe_trace_early);
1886
1887 /* Make a tracefs interface for controlling probe points */
1888 static __init int init_kprobe_trace(void)
1889 {
1890 struct dentry *d_tracer;
1891 struct dentry *entry;
1892
1893 d_tracer = tracing_init_dentry();
1894 if (IS_ERR(d_tracer))
1895 return 0;
1896
1897 entry = tracefs_create_file("kprobe_events", 0644, d_tracer,
1898 NULL, &kprobe_events_ops);
1899
1900 /* Event list interface */
1901 if (!entry)
1902 pr_warn("Could not create tracefs 'kprobe_events' entry\n");
1903
1904 /* Profile interface */
1905 entry = tracefs_create_file("kprobe_profile", 0444, d_tracer,
1906 NULL, &kprobe_profile_ops);
1907
1908 if (!entry)
1909 pr_warn("Could not create tracefs 'kprobe_profile' entry\n");
1910
1911 setup_boot_kprobe_events();
1912
1913 return 0;
1914 }
1915 fs_initcall(init_kprobe_trace);
1916
1917
1918 #ifdef CONFIG_FTRACE_STARTUP_TEST
1919 static __init struct trace_event_file *
1920 find_trace_probe_file(struct trace_kprobe *tk, struct trace_array *tr)
1921 {
1922 struct trace_event_file *file;
1923
1924 list_for_each_entry(file, &tr->events, list)
1925 if (file->event_call == trace_probe_event_call(&tk->tp))
1926 return file;
1927
1928 return NULL;
1929 }
1930
1931 /*
1932 * Nobody but us can call enable_trace_kprobe/disable_trace_kprobe at this
1933 * stage, we can do this lockless.
1934 */
1935 static __init int kprobe_trace_self_tests_init(void)
1936 {
1937 int ret, warn = 0;
1938 int (*target)(int, int, int, int, int, int);
1939 struct trace_kprobe *tk;
1940 struct trace_event_file *file;
1941
1942 if (tracing_is_disabled())
1943 return -ENODEV;
1944
1945 if (kprobe_boot_events_enabled) {
1946 pr_info("Skipping kprobe tests due to kprobe_event on cmdline\n");
1947 return 0;
1948 }
1949
1950 target = kprobe_trace_selftest_target;
1951
1952 pr_info("Testing kprobe tracing: ");
1953
1954 ret = trace_run_command("p:testprobe kprobe_trace_selftest_target $stack $stack0 +0($stack)",
1955 create_or_delete_trace_kprobe);
1956 if (WARN_ON_ONCE(ret)) {
1957 pr_warn("error on probing function entry.\n");
1958 warn++;
1959 } else {
1960 /* Enable trace point */
1961 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1962 if (WARN_ON_ONCE(tk == NULL)) {
1963 pr_warn("error on getting new probe.\n");
1964 warn++;
1965 } else {
1966 file = find_trace_probe_file(tk, top_trace_array());
1967 if (WARN_ON_ONCE(file == NULL)) {
1968 pr_warn("error on getting probe file.\n");
1969 warn++;
1970 } else
1971 enable_trace_kprobe(
1972 trace_probe_event_call(&tk->tp), file);
1973 }
1974 }
1975
1976 ret = trace_run_command("r:testprobe2 kprobe_trace_selftest_target $retval",
1977 create_or_delete_trace_kprobe);
1978 if (WARN_ON_ONCE(ret)) {
1979 pr_warn("error on probing function return.\n");
1980 warn++;
1981 } else {
1982 /* Enable trace point */
1983 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1984 if (WARN_ON_ONCE(tk == NULL)) {
1985 pr_warn("error on getting 2nd new probe.\n");
1986 warn++;
1987 } else {
1988 file = find_trace_probe_file(tk, top_trace_array());
1989 if (WARN_ON_ONCE(file == NULL)) {
1990 pr_warn("error on getting probe file.\n");
1991 warn++;
1992 } else
1993 enable_trace_kprobe(
1994 trace_probe_event_call(&tk->tp), file);
1995 }
1996 }
1997
1998 if (warn)
1999 goto end;
2000
2001 ret = target(1, 2, 3, 4, 5, 6);
2002
2003 /*
2004 * Not expecting an error here, the check is only to prevent the
2005 * optimizer from removing the call to target() as otherwise there
2006 * are no side-effects and the call is never performed.
2007 */
2008 if (ret != 21)
2009 warn++;
2010
2011 /* Disable trace points before removing it */
2012 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
2013 if (WARN_ON_ONCE(tk == NULL)) {
2014 pr_warn("error on getting test probe.\n");
2015 warn++;
2016 } else {
2017 if (trace_kprobe_nhit(tk) != 1) {
2018 pr_warn("incorrect number of testprobe hits\n");
2019 warn++;
2020 }
2021
2022 file = find_trace_probe_file(tk, top_trace_array());
2023 if (WARN_ON_ONCE(file == NULL)) {
2024 pr_warn("error on getting probe file.\n");
2025 warn++;
2026 } else
2027 disable_trace_kprobe(
2028 trace_probe_event_call(&tk->tp), file);
2029 }
2030
2031 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
2032 if (WARN_ON_ONCE(tk == NULL)) {
2033 pr_warn("error on getting 2nd test probe.\n");
2034 warn++;
2035 } else {
2036 if (trace_kprobe_nhit(tk) != 1) {
2037 pr_warn("incorrect number of testprobe2 hits\n");
2038 warn++;
2039 }
2040
2041 file = find_trace_probe_file(tk, top_trace_array());
2042 if (WARN_ON_ONCE(file == NULL)) {
2043 pr_warn("error on getting probe file.\n");
2044 warn++;
2045 } else
2046 disable_trace_kprobe(
2047 trace_probe_event_call(&tk->tp), file);
2048 }
2049
2050 ret = trace_run_command("-:testprobe", create_or_delete_trace_kprobe);
2051 if (WARN_ON_ONCE(ret)) {
2052 pr_warn("error on deleting a probe.\n");
2053 warn++;
2054 }
2055
2056 ret = trace_run_command("-:testprobe2", create_or_delete_trace_kprobe);
2057 if (WARN_ON_ONCE(ret)) {
2058 pr_warn("error on deleting a probe.\n");
2059 warn++;
2060 }
2061
2062 end:
2063 ret = dyn_events_release_all(&trace_kprobe_ops);
2064 if (WARN_ON_ONCE(ret)) {
2065 pr_warn("error on cleaning up probes.\n");
2066 warn++;
2067 }
2068 /*
2069 * Wait for the optimizer work to finish. Otherwise it might fiddle
2070 * with probes in already freed __init text.
2071 */
2072 wait_for_kprobe_optimizer();
2073 if (warn)
2074 pr_cont("NG: Some tests are failed. Please check them.\n");
2075 else
2076 pr_cont("OK\n");
2077 return 0;
2078 }
2079
2080 late_initcall(kprobe_trace_self_tests_init);
2081
2082 #endif
Linux with added FPC-III support