]> git.ipfire.org Git - thirdparty/linux.git/blob - kernel/trace/trace_kprobe.c
projects / thirdparty / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
io_uring: reset -EBUSY error when io sq thread is waken up
[thirdparty/linux.git] / kernel / trace / trace_kprobe.c
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 = 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 * Return: 0 if successful, error otherwise.
944 */
945 int __kprobe_event_gen_cmd_start(struct dynevent_cmd *cmd, bool kretprobe,
946 const char *name, const char *loc, ...)
947 {
948 char buf[MAX_EVENT_NAME_LEN];
949 struct dynevent_arg arg;
950 va_list args;
951 int ret;
952
953 if (cmd->type != DYNEVENT_TYPE_KPROBE)
954 return -EINVAL;
955
956 if (kretprobe)
957 snprintf(buf, MAX_EVENT_NAME_LEN, "r:kprobes/%s", name);
958 else
959 snprintf(buf, MAX_EVENT_NAME_LEN, "p:kprobes/%s", name);
960
961 ret = dynevent_str_add(cmd, buf);
962 if (ret)
963 return ret;
964
965 dynevent_arg_init(&arg, 0);
966 arg.str = loc;
967 ret = dynevent_arg_add(cmd, &arg, NULL);
968 if (ret)
969 return ret;
970
971 va_start(args, loc);
972 for (;;) {
973 const char *field;
974
975 field = va_arg(args, const char *);
976 if (!field)
977 break;
978
979 if (++cmd->n_fields > MAX_TRACE_ARGS) {
980 ret = -EINVAL;
981 break;
982 }
983
984 arg.str = field;
985 ret = dynevent_arg_add(cmd, &arg, NULL);
986 if (ret)
987 break;
988 }
989 va_end(args);
990
991 return ret;
992 }
993 EXPORT_SYMBOL_GPL(__kprobe_event_gen_cmd_start);
994
995 /**
996 * __kprobe_event_add_fields - Add probe fields to a kprobe command from arg list
997 * @cmd: A pointer to the dynevent_cmd struct representing the new event
998 * @args: Variable number of arg (pairs), one pair for each field
999 *
1000 * NOTE: Users normally won't want to call this function directly, but
1001 * rather use the kprobe_event_add_fields() wrapper, which
1002 * automatically adds a NULL to the end of the arg list. If this
1003 * function is used directly, make sure the last arg in the variable
1004 * arg list is NULL.
1005 *
1006 * Add probe fields to an existing kprobe command using a variable
1007 * list of args. Fields are added in the same order they're listed.
1008 *
1009 * Return: 0 if successful, error otherwise.
1010 */
1011 int __kprobe_event_add_fields(struct dynevent_cmd *cmd, ...)
1012 {
1013 struct dynevent_arg arg;
1014 va_list args;
1015 int ret = 0;
1016
1017 if (cmd->type != DYNEVENT_TYPE_KPROBE)
1018 return -EINVAL;
1019
1020 dynevent_arg_init(&arg, 0);
1021
1022 va_start(args, cmd);
1023 for (;;) {
1024 const char *field;
1025
1026 field = va_arg(args, const char *);
1027 if (!field)
1028 break;
1029
1030 if (++cmd->n_fields > MAX_TRACE_ARGS) {
1031 ret = -EINVAL;
1032 break;
1033 }
1034
1035 arg.str = field;
1036 ret = dynevent_arg_add(cmd, &arg, NULL);
1037 if (ret)
1038 break;
1039 }
1040 va_end(args);
1041
1042 return ret;
1043 }
1044 EXPORT_SYMBOL_GPL(__kprobe_event_add_fields);
1045
1046 /**
1047 * kprobe_event_delete - Delete a kprobe event
1048 * @name: The name of the kprobe event to delete
1049 *
1050 * Delete a kprobe event with the give @name from kernel code rather
1051 * than directly from the command line.
1052 *
1053 * Return: 0 if successful, error otherwise.
1054 */
1055 int kprobe_event_delete(const char *name)
1056 {
1057 char buf[MAX_EVENT_NAME_LEN];
1058
1059 snprintf(buf, MAX_EVENT_NAME_LEN, "-:%s", name);
1060
1061 return trace_run_command(buf, create_or_delete_trace_kprobe);
1062 }
1063 EXPORT_SYMBOL_GPL(kprobe_event_delete);
1064
1065 static int trace_kprobe_release(struct dyn_event *ev)
1066 {
1067 struct trace_kprobe *tk = to_trace_kprobe(ev);
1068 int ret = unregister_trace_kprobe(tk);
1069
1070 if (!ret)
1071 free_trace_kprobe(tk);
1072 return ret;
1073 }
1074
1075 static int trace_kprobe_show(struct seq_file *m, struct dyn_event *ev)
1076 {
1077 struct trace_kprobe *tk = to_trace_kprobe(ev);
1078 int i;
1079
1080 seq_putc(m, trace_kprobe_is_return(tk) ? 'r' : 'p');
1081 if (trace_kprobe_is_return(tk) && tk->rp.maxactive)
1082 seq_printf(m, "%d", tk->rp.maxactive);
1083 seq_printf(m, ":%s/%s", trace_probe_group_name(&tk->tp),
1084 trace_probe_name(&tk->tp));
1085
1086 if (!tk->symbol)
1087 seq_printf(m, " 0x%p", tk->rp.kp.addr);
1088 else if (tk->rp.kp.offset)
1089 seq_printf(m, " %s+%u", trace_kprobe_symbol(tk),
1090 tk->rp.kp.offset);
1091 else
1092 seq_printf(m, " %s", trace_kprobe_symbol(tk));
1093
1094 for (i = 0; i < tk->tp.nr_args; i++)
1095 seq_printf(m, " %s=%s", tk->tp.args[i].name, tk->tp.args[i].comm);
1096 seq_putc(m, '\n');
1097
1098 return 0;
1099 }
1100
1101 static int probes_seq_show(struct seq_file *m, void *v)
1102 {
1103 struct dyn_event *ev = v;
1104
1105 if (!is_trace_kprobe(ev))
1106 return 0;
1107
1108 return trace_kprobe_show(m, ev);
1109 }
1110
1111 static const struct seq_operations probes_seq_op = {
1112 .start = dyn_event_seq_start,
1113 .next = dyn_event_seq_next,
1114 .stop = dyn_event_seq_stop,
1115 .show = probes_seq_show
1116 };
1117
1118 static int probes_open(struct inode *inode, struct file *file)
1119 {
1120 int ret;
1121
1122 ret = security_locked_down(LOCKDOWN_TRACEFS);
1123 if (ret)
1124 return ret;
1125
1126 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
1127 ret = dyn_events_release_all(&trace_kprobe_ops);
1128 if (ret < 0)
1129 return ret;
1130 }
1131
1132 return seq_open(file, &probes_seq_op);
1133 }
1134
1135 static ssize_t probes_write(struct file *file, const char __user *buffer,
1136 size_t count, loff_t *ppos)
1137 {
1138 return trace_parse_run_command(file, buffer, count, ppos,
1139 create_or_delete_trace_kprobe);
1140 }
1141
1142 static const struct file_operations kprobe_events_ops = {
1143 .owner = THIS_MODULE,
1144 .open = probes_open,
1145 .read = seq_read,
1146 .llseek = seq_lseek,
1147 .release = seq_release,
1148 .write = probes_write,
1149 };
1150
1151 /* Probes profiling interfaces */
1152 static int probes_profile_seq_show(struct seq_file *m, void *v)
1153 {
1154 struct dyn_event *ev = v;
1155 struct trace_kprobe *tk;
1156
1157 if (!is_trace_kprobe(ev))
1158 return 0;
1159
1160 tk = to_trace_kprobe(ev);
1161 seq_printf(m, " %-44s %15lu %15lu\n",
1162 trace_probe_name(&tk->tp),
1163 trace_kprobe_nhit(tk),
1164 tk->rp.kp.nmissed);
1165
1166 return 0;
1167 }
1168
1169 static const struct seq_operations profile_seq_op = {
1170 .start = dyn_event_seq_start,
1171 .next = dyn_event_seq_next,
1172 .stop = dyn_event_seq_stop,
1173 .show = probes_profile_seq_show
1174 };
1175
1176 static int profile_open(struct inode *inode, struct file *file)
1177 {
1178 int ret;
1179
1180 ret = security_locked_down(LOCKDOWN_TRACEFS);
1181 if (ret)
1182 return ret;
1183
1184 return seq_open(file, &profile_seq_op);
1185 }
1186
1187 static const struct file_operations kprobe_profile_ops = {
1188 .owner = THIS_MODULE,
1189 .open = profile_open,
1190 .read = seq_read,
1191 .llseek = seq_lseek,
1192 .release = seq_release,
1193 };
1194
1195 /* Kprobe specific fetch functions */
1196
1197 /* Return the length of string -- including null terminal byte */
1198 static nokprobe_inline int
1199 fetch_store_strlen(unsigned long addr)
1200 {
1201 int ret, len = 0;
1202 u8 c;
1203
1204 do {
1205 ret = probe_kernel_read(&c, (u8 *)addr + len, 1);
1206 len++;
1207 } while (c && ret == 0 && len < MAX_STRING_SIZE);
1208
1209 return (ret < 0) ? ret : len;
1210 }
1211
1212 /* Return the length of string -- including null terminal byte */
1213 static nokprobe_inline int
1214 fetch_store_strlen_user(unsigned long addr)
1215 {
1216 const void __user *uaddr = (__force const void __user *)addr;
1217
1218 return strnlen_unsafe_user(uaddr, MAX_STRING_SIZE);
1219 }
1220
1221 /*
1222 * Fetch a null-terminated string. Caller MUST set *(u32 *)buf with max
1223 * length and relative data location.
1224 */
1225 static nokprobe_inline int
1226 fetch_store_string(unsigned long addr, void *dest, void *base)
1227 {
1228 int maxlen = get_loc_len(*(u32 *)dest);
1229 void *__dest;
1230 long ret;
1231
1232 if (unlikely(!maxlen))
1233 return -ENOMEM;
1234
1235 __dest = get_loc_data(dest, base);
1236
1237 /*
1238 * Try to get string again, since the string can be changed while
1239 * probing.
1240 */
1241 ret = strncpy_from_unsafe(__dest, (void *)addr, maxlen);
1242 if (ret >= 0)
1243 *(u32 *)dest = make_data_loc(ret, __dest - base);
1244
1245 return ret;
1246 }
1247
1248 /*
1249 * Fetch a null-terminated string from user. Caller MUST set *(u32 *)buf
1250 * with max length and relative data location.
1251 */
1252 static nokprobe_inline int
1253 fetch_store_string_user(unsigned long addr, void *dest, void *base)
1254 {
1255 const void __user *uaddr = (__force const void __user *)addr;
1256 int maxlen = get_loc_len(*(u32 *)dest);
1257 void *__dest;
1258 long ret;
1259
1260 if (unlikely(!maxlen))
1261 return -ENOMEM;
1262
1263 __dest = get_loc_data(dest, base);
1264
1265 ret = strncpy_from_unsafe_user(__dest, uaddr, maxlen);
1266 if (ret >= 0)
1267 *(u32 *)dest = make_data_loc(ret, __dest - base);
1268
1269 return ret;
1270 }
1271
1272 static nokprobe_inline int
1273 probe_mem_read(void *dest, void *src, size_t size)
1274 {
1275 return probe_kernel_read(dest, src, size);
1276 }
1277
1278 static nokprobe_inline int
1279 probe_mem_read_user(void *dest, void *src, size_t size)
1280 {
1281 const void __user *uaddr = (__force const void __user *)src;
1282
1283 return probe_user_read(dest, uaddr, size);
1284 }
1285
1286 /* Note that we don't verify it, since the code does not come from user space */
1287 static int
1288 process_fetch_insn(struct fetch_insn *code, struct pt_regs *regs, void *dest,
1289 void *base)
1290 {
1291 unsigned long val;
1292
1293 retry:
1294 /* 1st stage: get value from context */
1295 switch (code->op) {
1296 case FETCH_OP_REG:
1297 val = regs_get_register(regs, code->param);
1298 break;
1299 case FETCH_OP_STACK:
1300 val = regs_get_kernel_stack_nth(regs, code->param);
1301 break;
1302 case FETCH_OP_STACKP:
1303 val = kernel_stack_pointer(regs);
1304 break;
1305 case FETCH_OP_RETVAL:
1306 val = regs_return_value(regs);
1307 break;
1308 case FETCH_OP_IMM:
1309 val = code->immediate;
1310 break;
1311 case FETCH_OP_COMM:
1312 val = (unsigned long)current->comm;
1313 break;
1314 case FETCH_OP_DATA:
1315 val = (unsigned long)code->data;
1316 break;
1317 #ifdef CONFIG_HAVE_FUNCTION_ARG_ACCESS_API
1318 case FETCH_OP_ARG:
1319 val = regs_get_kernel_argument(regs, code->param);
1320 break;
1321 #endif
1322 case FETCH_NOP_SYMBOL: /* Ignore a place holder */
1323 code++;
1324 goto retry;
1325 default:
1326 return -EILSEQ;
1327 }
1328 code++;
1329
1330 return process_fetch_insn_bottom(code, val, dest, base);
1331 }
1332 NOKPROBE_SYMBOL(process_fetch_insn)
1333
1334 /* Kprobe handler */
1335 static nokprobe_inline void
1336 __kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs,
1337 struct trace_event_file *trace_file)
1338 {
1339 struct kprobe_trace_entry_head *entry;
1340 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1341 struct trace_event_buffer fbuffer;
1342 int dsize;
1343
1344 WARN_ON(call != trace_file->event_call);
1345
1346 if (trace_trigger_soft_disabled(trace_file))
1347 return;
1348
1349 local_save_flags(fbuffer.flags);
1350 fbuffer.pc = preempt_count();
1351 fbuffer.trace_file = trace_file;
1352
1353 dsize = __get_data_size(&tk->tp, regs);
1354
1355 fbuffer.event =
1356 trace_event_buffer_lock_reserve(&fbuffer.buffer, trace_file,
1357 call->event.type,
1358 sizeof(*entry) + tk->tp.size + dsize,
1359 fbuffer.flags, fbuffer.pc);
1360 if (!fbuffer.event)
1361 return;
1362
1363 fbuffer.regs = regs;
1364 entry = fbuffer.entry = ring_buffer_event_data(fbuffer.event);
1365 entry->ip = (unsigned long)tk->rp.kp.addr;
1366 store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1367
1368 trace_event_buffer_commit(&fbuffer);
1369 }
1370
1371 static void
1372 kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs)
1373 {
1374 struct event_file_link *link;
1375
1376 trace_probe_for_each_link_rcu(link, &tk->tp)
1377 __kprobe_trace_func(tk, regs, link->file);
1378 }
1379 NOKPROBE_SYMBOL(kprobe_trace_func);
1380
1381 /* Kretprobe handler */
1382 static nokprobe_inline void
1383 __kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1384 struct pt_regs *regs,
1385 struct trace_event_file *trace_file)
1386 {
1387 struct kretprobe_trace_entry_head *entry;
1388 struct trace_event_buffer fbuffer;
1389 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1390 int dsize;
1391
1392 WARN_ON(call != trace_file->event_call);
1393
1394 if (trace_trigger_soft_disabled(trace_file))
1395 return;
1396
1397 local_save_flags(fbuffer.flags);
1398 fbuffer.pc = preempt_count();
1399 fbuffer.trace_file = trace_file;
1400
1401 dsize = __get_data_size(&tk->tp, regs);
1402 fbuffer.event =
1403 trace_event_buffer_lock_reserve(&fbuffer.buffer, trace_file,
1404 call->event.type,
1405 sizeof(*entry) + tk->tp.size + dsize,
1406 fbuffer.flags, fbuffer.pc);
1407 if (!fbuffer.event)
1408 return;
1409
1410 fbuffer.regs = regs;
1411 entry = fbuffer.entry = ring_buffer_event_data(fbuffer.event);
1412 entry->func = (unsigned long)tk->rp.kp.addr;
1413 entry->ret_ip = (unsigned long)ri->ret_addr;
1414 store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1415
1416 trace_event_buffer_commit(&fbuffer);
1417 }
1418
1419 static void
1420 kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1421 struct pt_regs *regs)
1422 {
1423 struct event_file_link *link;
1424
1425 trace_probe_for_each_link_rcu(link, &tk->tp)
1426 __kretprobe_trace_func(tk, ri, regs, link->file);
1427 }
1428 NOKPROBE_SYMBOL(kretprobe_trace_func);
1429
1430 /* Event entry printers */
1431 static enum print_line_t
1432 print_kprobe_event(struct trace_iterator *iter, int flags,
1433 struct trace_event *event)
1434 {
1435 struct kprobe_trace_entry_head *field;
1436 struct trace_seq *s = &iter->seq;
1437 struct trace_probe *tp;
1438
1439 field = (struct kprobe_trace_entry_head *)iter->ent;
1440 tp = trace_probe_primary_from_call(
1441 container_of(event, struct trace_event_call, event));
1442 if (WARN_ON_ONCE(!tp))
1443 goto out;
1444
1445 trace_seq_printf(s, "%s: (", trace_probe_name(tp));
1446
1447 if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
1448 goto out;
1449
1450 trace_seq_putc(s, ')');
1451
1452 if (print_probe_args(s, tp->args, tp->nr_args,
1453 (u8 *)&field[1], field) < 0)
1454 goto out;
1455
1456 trace_seq_putc(s, '\n');
1457 out:
1458 return trace_handle_return(s);
1459 }
1460
1461 static enum print_line_t
1462 print_kretprobe_event(struct trace_iterator *iter, int flags,
1463 struct trace_event *event)
1464 {
1465 struct kretprobe_trace_entry_head *field;
1466 struct trace_seq *s = &iter->seq;
1467 struct trace_probe *tp;
1468
1469 field = (struct kretprobe_trace_entry_head *)iter->ent;
1470 tp = trace_probe_primary_from_call(
1471 container_of(event, struct trace_event_call, event));
1472 if (WARN_ON_ONCE(!tp))
1473 goto out;
1474
1475 trace_seq_printf(s, "%s: (", trace_probe_name(tp));
1476
1477 if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1478 goto out;
1479
1480 trace_seq_puts(s, " <- ");
1481
1482 if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1483 goto out;
1484
1485 trace_seq_putc(s, ')');
1486
1487 if (print_probe_args(s, tp->args, tp->nr_args,
1488 (u8 *)&field[1], field) < 0)
1489 goto out;
1490
1491 trace_seq_putc(s, '\n');
1492
1493 out:
1494 return trace_handle_return(s);
1495 }
1496
1497
1498 static int kprobe_event_define_fields(struct trace_event_call *event_call)
1499 {
1500 int ret;
1501 struct kprobe_trace_entry_head field;
1502 struct trace_probe *tp;
1503
1504 tp = trace_probe_primary_from_call(event_call);
1505 if (WARN_ON_ONCE(!tp))
1506 return -ENOENT;
1507
1508 DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1509
1510 return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
1511 }
1512
1513 static int kretprobe_event_define_fields(struct trace_event_call *event_call)
1514 {
1515 int ret;
1516 struct kretprobe_trace_entry_head field;
1517 struct trace_probe *tp;
1518
1519 tp = trace_probe_primary_from_call(event_call);
1520 if (WARN_ON_ONCE(!tp))
1521 return -ENOENT;
1522
1523 DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1524 DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1525
1526 return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
1527 }
1528
1529 #ifdef CONFIG_PERF_EVENTS
1530
1531 /* Kprobe profile handler */
1532 static int
1533 kprobe_perf_func(struct trace_kprobe *tk, struct pt_regs *regs)
1534 {
1535 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1536 struct kprobe_trace_entry_head *entry;
1537 struct hlist_head *head;
1538 int size, __size, dsize;
1539 int rctx;
1540
1541 if (bpf_prog_array_valid(call)) {
1542 unsigned long orig_ip = instruction_pointer(regs);
1543 int ret;
1544
1545 ret = trace_call_bpf(call, regs);
1546
1547 /*
1548 * We need to check and see if we modified the pc of the
1549 * pt_regs, and if so return 1 so that we don't do the
1550 * single stepping.
1551 */
1552 if (orig_ip != instruction_pointer(regs))
1553 return 1;
1554 if (!ret)
1555 return 0;
1556 }
1557
1558 head = this_cpu_ptr(call->perf_events);
1559 if (hlist_empty(head))
1560 return 0;
1561
1562 dsize = __get_data_size(&tk->tp, regs);
1563 __size = sizeof(*entry) + tk->tp.size + dsize;
1564 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1565 size -= sizeof(u32);
1566
1567 entry = perf_trace_buf_alloc(size, NULL, &rctx);
1568 if (!entry)
1569 return 0;
1570
1571 entry->ip = (unsigned long)tk->rp.kp.addr;
1572 memset(&entry[1], 0, dsize);
1573 store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1574 perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1575 head, NULL);
1576 return 0;
1577 }
1578 NOKPROBE_SYMBOL(kprobe_perf_func);
1579
1580 /* Kretprobe profile handler */
1581 static void
1582 kretprobe_perf_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1583 struct pt_regs *regs)
1584 {
1585 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1586 struct kretprobe_trace_entry_head *entry;
1587 struct hlist_head *head;
1588 int size, __size, dsize;
1589 int rctx;
1590
1591 if (bpf_prog_array_valid(call) && !trace_call_bpf(call, regs))
1592 return;
1593
1594 head = this_cpu_ptr(call->perf_events);
1595 if (hlist_empty(head))
1596 return;
1597
1598 dsize = __get_data_size(&tk->tp, regs);
1599 __size = sizeof(*entry) + tk->tp.size + dsize;
1600 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1601 size -= sizeof(u32);
1602
1603 entry = perf_trace_buf_alloc(size, NULL, &rctx);
1604 if (!entry)
1605 return;
1606
1607 entry->func = (unsigned long)tk->rp.kp.addr;
1608 entry->ret_ip = (unsigned long)ri->ret_addr;
1609 store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1610 perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1611 head, NULL);
1612 }
1613 NOKPROBE_SYMBOL(kretprobe_perf_func);
1614
1615 int bpf_get_kprobe_info(const struct perf_event *event, u32 *fd_type,
1616 const char **symbol, u64 *probe_offset,
1617 u64 *probe_addr, bool perf_type_tracepoint)
1618 {
1619 const char *pevent = trace_event_name(event->tp_event);
1620 const char *group = event->tp_event->class->system;
1621 struct trace_kprobe *tk;
1622
1623 if (perf_type_tracepoint)
1624 tk = find_trace_kprobe(pevent, group);
1625 else
1626 tk = event->tp_event->data;
1627 if (!tk)
1628 return -EINVAL;
1629
1630 *fd_type = trace_kprobe_is_return(tk) ? BPF_FD_TYPE_KRETPROBE
1631 : BPF_FD_TYPE_KPROBE;
1632 if (tk->symbol) {
1633 *symbol = tk->symbol;
1634 *probe_offset = tk->rp.kp.offset;
1635 *probe_addr = 0;
1636 } else {
1637 *symbol = NULL;
1638 *probe_offset = 0;
1639 *probe_addr = (unsigned long)tk->rp.kp.addr;
1640 }
1641 return 0;
1642 }
1643 #endif /* CONFIG_PERF_EVENTS */
1644
1645 /*
1646 * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex.
1647 *
1648 * kprobe_trace_self_tests_init() does enable_trace_probe/disable_trace_probe
1649 * lockless, but we can't race with this __init function.
1650 */
1651 static int kprobe_register(struct trace_event_call *event,
1652 enum trace_reg type, void *data)
1653 {
1654 struct trace_event_file *file = data;
1655
1656 switch (type) {
1657 case TRACE_REG_REGISTER:
1658 return enable_trace_kprobe(event, file);
1659 case TRACE_REG_UNREGISTER:
1660 return disable_trace_kprobe(event, file);
1661
1662 #ifdef CONFIG_PERF_EVENTS
1663 case TRACE_REG_PERF_REGISTER:
1664 return enable_trace_kprobe(event, NULL);
1665 case TRACE_REG_PERF_UNREGISTER:
1666 return disable_trace_kprobe(event, NULL);
1667 case TRACE_REG_PERF_OPEN:
1668 case TRACE_REG_PERF_CLOSE:
1669 case TRACE_REG_PERF_ADD:
1670 case TRACE_REG_PERF_DEL:
1671 return 0;
1672 #endif
1673 }
1674 return 0;
1675 }
1676
1677 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1678 {
1679 struct trace_kprobe *tk = container_of(kp, struct trace_kprobe, rp.kp);
1680 int ret = 0;
1681
1682 raw_cpu_inc(*tk->nhit);
1683
1684 if (trace_probe_test_flag(&tk->tp, TP_FLAG_TRACE))
1685 kprobe_trace_func(tk, regs);
1686 #ifdef CONFIG_PERF_EVENTS
1687 if (trace_probe_test_flag(&tk->tp, TP_FLAG_PROFILE))
1688 ret = kprobe_perf_func(tk, regs);
1689 #endif
1690 return ret;
1691 }
1692 NOKPROBE_SYMBOL(kprobe_dispatcher);
1693
1694 static int
1695 kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1696 {
1697 struct trace_kprobe *tk = container_of(ri->rp, struct trace_kprobe, rp);
1698
1699 raw_cpu_inc(*tk->nhit);
1700
1701 if (trace_probe_test_flag(&tk->tp, TP_FLAG_TRACE))
1702 kretprobe_trace_func(tk, ri, regs);
1703 #ifdef CONFIG_PERF_EVENTS
1704 if (trace_probe_test_flag(&tk->tp, TP_FLAG_PROFILE))
1705 kretprobe_perf_func(tk, ri, regs);
1706 #endif
1707 return 0; /* We don't tweek kernel, so just return 0 */
1708 }
1709 NOKPROBE_SYMBOL(kretprobe_dispatcher);
1710
1711 static struct trace_event_functions kretprobe_funcs = {
1712 .trace = print_kretprobe_event
1713 };
1714
1715 static struct trace_event_functions kprobe_funcs = {
1716 .trace = print_kprobe_event
1717 };
1718
1719 static struct trace_event_fields kretprobe_fields_array[] = {
1720 { .type = TRACE_FUNCTION_TYPE,
1721 .define_fields = kretprobe_event_define_fields },
1722 {}
1723 };
1724
1725 static struct trace_event_fields kprobe_fields_array[] = {
1726 { .type = TRACE_FUNCTION_TYPE,
1727 .define_fields = kprobe_event_define_fields },
1728 {}
1729 };
1730
1731 static inline void init_trace_event_call(struct trace_kprobe *tk)
1732 {
1733 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1734
1735 if (trace_kprobe_is_return(tk)) {
1736 call->event.funcs = &kretprobe_funcs;
1737 call->class->fields_array = kretprobe_fields_array;
1738 } else {
1739 call->event.funcs = &kprobe_funcs;
1740 call->class->fields_array = kprobe_fields_array;
1741 }
1742
1743 call->flags = TRACE_EVENT_FL_KPROBE;
1744 call->class->reg = kprobe_register;
1745 }
1746
1747 static int register_kprobe_event(struct trace_kprobe *tk)
1748 {
1749 init_trace_event_call(tk);
1750
1751 return trace_probe_register_event_call(&tk->tp);
1752 }
1753
1754 static int unregister_kprobe_event(struct trace_kprobe *tk)
1755 {
1756 return trace_probe_unregister_event_call(&tk->tp);
1757 }
1758
1759 #ifdef CONFIG_PERF_EVENTS
1760 /* create a trace_kprobe, but don't add it to global lists */
1761 struct trace_event_call *
1762 create_local_trace_kprobe(char *func, void *addr, unsigned long offs,
1763 bool is_return)
1764 {
1765 struct trace_kprobe *tk;
1766 int ret;
1767 char *event;
1768
1769 /*
1770 * local trace_kprobes are not added to dyn_event, so they are never
1771 * searched in find_trace_kprobe(). Therefore, there is no concern of
1772 * duplicated name here.
1773 */
1774 event = func ? func : "DUMMY_EVENT";
1775
1776 tk = alloc_trace_kprobe(KPROBE_EVENT_SYSTEM, event, (void *)addr, func,
1777 offs, 0 /* maxactive */, 0 /* nargs */,
1778 is_return);
1779
1780 if (IS_ERR(tk)) {
1781 pr_info("Failed to allocate trace_probe.(%d)\n",
1782 (int)PTR_ERR(tk));
1783 return ERR_CAST(tk);
1784 }
1785
1786 init_trace_event_call(tk);
1787
1788 if (traceprobe_set_print_fmt(&tk->tp, trace_kprobe_is_return(tk)) < 0) {
1789 ret = -ENOMEM;
1790 goto error;
1791 }
1792
1793 ret = __register_trace_kprobe(tk);
1794 if (ret < 0)
1795 goto error;
1796
1797 return trace_probe_event_call(&tk->tp);
1798 error:
1799 free_trace_kprobe(tk);
1800 return ERR_PTR(ret);
1801 }
1802
1803 void destroy_local_trace_kprobe(struct trace_event_call *event_call)
1804 {
1805 struct trace_kprobe *tk;
1806
1807 tk = trace_kprobe_primary_from_call(event_call);
1808 if (unlikely(!tk))
1809 return;
1810
1811 if (trace_probe_is_enabled(&tk->tp)) {
1812 WARN_ON(1);
1813 return;
1814 }
1815
1816 __unregister_trace_kprobe(tk);
1817
1818 free_trace_kprobe(tk);
1819 }
1820 #endif /* CONFIG_PERF_EVENTS */
1821
1822 static __init void enable_boot_kprobe_events(void)
1823 {
1824 struct trace_array *tr = top_trace_array();
1825 struct trace_event_file *file;
1826 struct trace_kprobe *tk;
1827 struct dyn_event *pos;
1828
1829 mutex_lock(&event_mutex);
1830 for_each_trace_kprobe(tk, pos) {
1831 list_for_each_entry(file, &tr->events, list)
1832 if (file->event_call == trace_probe_event_call(&tk->tp))
1833 trace_event_enable_disable(file, 1, 0);
1834 }
1835 mutex_unlock(&event_mutex);
1836 }
1837
1838 static __init void setup_boot_kprobe_events(void)
1839 {
1840 char *p, *cmd = kprobe_boot_events_buf;
1841 int ret;
1842
1843 strreplace(kprobe_boot_events_buf, ',', ' ');
1844
1845 while (cmd && *cmd != '\0') {
1846 p = strchr(cmd, ';');
1847 if (p)
1848 *p++ = '\0';
1849
1850 ret = trace_run_command(cmd, create_or_delete_trace_kprobe);
1851 if (ret)
1852 pr_warn("Failed to add event(%d): %s\n", ret, cmd);
1853 else
1854 kprobe_boot_events_enabled = true;
1855
1856 cmd = p;
1857 }
1858
1859 enable_boot_kprobe_events();
1860 }
1861
1862 /*
1863 * Register dynevent at subsys_initcall. This allows kernel to setup kprobe
1864 * events in fs_initcall without tracefs.
1865 */
1866 static __init int init_kprobe_trace_early(void)
1867 {
1868 int ret;
1869
1870 ret = dyn_event_register(&trace_kprobe_ops);
1871 if (ret)
1872 return ret;
1873
1874 if (register_module_notifier(&trace_kprobe_module_nb))
1875 return -EINVAL;
1876
1877 return 0;
1878 }
1879 subsys_initcall(init_kprobe_trace_early);
1880
1881 /* Make a tracefs interface for controlling probe points */
1882 static __init int init_kprobe_trace(void)
1883 {
1884 struct dentry *d_tracer;
1885 struct dentry *entry;
1886
1887 d_tracer = tracing_init_dentry();
1888 if (IS_ERR(d_tracer))
1889 return 0;
1890
1891 entry = tracefs_create_file("kprobe_events", 0644, d_tracer,
1892 NULL, &kprobe_events_ops);
1893
1894 /* Event list interface */
1895 if (!entry)
1896 pr_warn("Could not create tracefs 'kprobe_events' entry\n");
1897
1898 /* Profile interface */
1899 entry = tracefs_create_file("kprobe_profile", 0444, d_tracer,
1900 NULL, &kprobe_profile_ops);
1901
1902 if (!entry)
1903 pr_warn("Could not create tracefs 'kprobe_profile' entry\n");
1904
1905 setup_boot_kprobe_events();
1906
1907 return 0;
1908 }
1909 fs_initcall(init_kprobe_trace);
1910
1911
1912 #ifdef CONFIG_FTRACE_STARTUP_TEST
1913 static __init struct trace_event_file *
1914 find_trace_probe_file(struct trace_kprobe *tk, struct trace_array *tr)
1915 {
1916 struct trace_event_file *file;
1917
1918 list_for_each_entry(file, &tr->events, list)
1919 if (file->event_call == trace_probe_event_call(&tk->tp))
1920 return file;
1921
1922 return NULL;
1923 }
1924
1925 /*
1926 * Nobody but us can call enable_trace_kprobe/disable_trace_kprobe at this
1927 * stage, we can do this lockless.
1928 */
1929 static __init int kprobe_trace_self_tests_init(void)
1930 {
1931 int ret, warn = 0;
1932 int (*target)(int, int, int, int, int, int);
1933 struct trace_kprobe *tk;
1934 struct trace_event_file *file;
1935
1936 if (tracing_is_disabled())
1937 return -ENODEV;
1938
1939 if (kprobe_boot_events_enabled) {
1940 pr_info("Skipping kprobe tests due to kprobe_event on cmdline\n");
1941 return 0;
1942 }
1943
1944 target = kprobe_trace_selftest_target;
1945
1946 pr_info("Testing kprobe tracing: ");
1947
1948 ret = trace_run_command("p:testprobe kprobe_trace_selftest_target $stack $stack0 +0($stack)",
1949 create_or_delete_trace_kprobe);
1950 if (WARN_ON_ONCE(ret)) {
1951 pr_warn("error on probing function entry.\n");
1952 warn++;
1953 } else {
1954 /* Enable trace point */
1955 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1956 if (WARN_ON_ONCE(tk == NULL)) {
1957 pr_warn("error on getting new probe.\n");
1958 warn++;
1959 } else {
1960 file = find_trace_probe_file(tk, top_trace_array());
1961 if (WARN_ON_ONCE(file == NULL)) {
1962 pr_warn("error on getting probe file.\n");
1963 warn++;
1964 } else
1965 enable_trace_kprobe(
1966 trace_probe_event_call(&tk->tp), file);
1967 }
1968 }
1969
1970 ret = trace_run_command("r:testprobe2 kprobe_trace_selftest_target $retval",
1971 create_or_delete_trace_kprobe);
1972 if (WARN_ON_ONCE(ret)) {
1973 pr_warn("error on probing function return.\n");
1974 warn++;
1975 } else {
1976 /* Enable trace point */
1977 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1978 if (WARN_ON_ONCE(tk == NULL)) {
1979 pr_warn("error on getting 2nd new probe.\n");
1980 warn++;
1981 } else {
1982 file = find_trace_probe_file(tk, top_trace_array());
1983 if (WARN_ON_ONCE(file == NULL)) {
1984 pr_warn("error on getting probe file.\n");
1985 warn++;
1986 } else
1987 enable_trace_kprobe(
1988 trace_probe_event_call(&tk->tp), file);
1989 }
1990 }
1991
1992 if (warn)
1993 goto end;
1994
1995 ret = target(1, 2, 3, 4, 5, 6);
1996
1997 /*
1998 * Not expecting an error here, the check is only to prevent the
1999 * optimizer from removing the call to target() as otherwise there
2000 * are no side-effects and the call is never performed.
2001 */
2002 if (ret != 21)
2003 warn++;
2004
2005 /* Disable trace points before removing it */
2006 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
2007 if (WARN_ON_ONCE(tk == NULL)) {
2008 pr_warn("error on getting test probe.\n");
2009 warn++;
2010 } else {
2011 if (trace_kprobe_nhit(tk) != 1) {
2012 pr_warn("incorrect number of testprobe hits\n");
2013 warn++;
2014 }
2015
2016 file = find_trace_probe_file(tk, top_trace_array());
2017 if (WARN_ON_ONCE(file == NULL)) {
2018 pr_warn("error on getting probe file.\n");
2019 warn++;
2020 } else
2021 disable_trace_kprobe(
2022 trace_probe_event_call(&tk->tp), file);
2023 }
2024
2025 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
2026 if (WARN_ON_ONCE(tk == NULL)) {
2027 pr_warn("error on getting 2nd test probe.\n");
2028 warn++;
2029 } else {
2030 if (trace_kprobe_nhit(tk) != 1) {
2031 pr_warn("incorrect number of testprobe2 hits\n");
2032 warn++;
2033 }
2034
2035 file = find_trace_probe_file(tk, top_trace_array());
2036 if (WARN_ON_ONCE(file == NULL)) {
2037 pr_warn("error on getting probe file.\n");
2038 warn++;
2039 } else
2040 disable_trace_kprobe(
2041 trace_probe_event_call(&tk->tp), file);
2042 }
2043
2044 ret = trace_run_command("-:testprobe", create_or_delete_trace_kprobe);
2045 if (WARN_ON_ONCE(ret)) {
2046 pr_warn("error on deleting a probe.\n");
2047 warn++;
2048 }
2049
2050 ret = trace_run_command("-:testprobe2", 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 end:
2057 ret = dyn_events_release_all(&trace_kprobe_ops);
2058 if (WARN_ON_ONCE(ret)) {
2059 pr_warn("error on cleaning up probes.\n");
2060 warn++;
2061 }
2062 /*
2063 * Wait for the optimizer work to finish. Otherwise it might fiddle
2064 * with probes in already freed __init text.
2065 */
2066 wait_for_kprobe_optimizer();
2067 if (warn)
2068 pr_cont("NG: Some tests are failed. Please check them.\n");
2069 else
2070 pr_cont("OK\n");
2071 return 0;
2072 }
2073
2074 late_initcall(kprobe_trace_self_tests_init);
2075
2076 #endif
A mirror of Linus' kernel repository