Loading...
1/*
2 * Kprobes-based tracing events
3 *
4 * Created by Masami Hiramatsu <mhiramat@redhat.com>
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 as
8 * published by the Free Software Foundation.
9 *
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
14 *
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write to the Free Software
17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
18 */
19
20#include <linux/module.h>
21#include <linux/uaccess.h>
22
23#include "trace_probe.h"
24
25#define KPROBE_EVENT_SYSTEM "kprobes"
26
27/**
28 * Kprobe event core functions
29 */
30
31struct trace_probe {
32 struct list_head list;
33 struct kretprobe rp; /* Use rp.kp for kprobe use */
34 unsigned long nhit;
35 unsigned int flags; /* For TP_FLAG_* */
36 const char *symbol; /* symbol name */
37 struct ftrace_event_class class;
38 struct ftrace_event_call call;
39 ssize_t size; /* trace entry size */
40 unsigned int nr_args;
41 struct probe_arg args[];
42};
43
44#define SIZEOF_TRACE_PROBE(n) \
45 (offsetof(struct trace_probe, args) + \
46 (sizeof(struct probe_arg) * (n)))
47
48
49static __kprobes int trace_probe_is_return(struct trace_probe *tp)
50{
51 return tp->rp.handler != NULL;
52}
53
54static __kprobes const char *trace_probe_symbol(struct trace_probe *tp)
55{
56 return tp->symbol ? tp->symbol : "unknown";
57}
58
59static __kprobes unsigned long trace_probe_offset(struct trace_probe *tp)
60{
61 return tp->rp.kp.offset;
62}
63
64static __kprobes bool trace_probe_is_enabled(struct trace_probe *tp)
65{
66 return !!(tp->flags & (TP_FLAG_TRACE | TP_FLAG_PROFILE));
67}
68
69static __kprobes bool trace_probe_is_registered(struct trace_probe *tp)
70{
71 return !!(tp->flags & TP_FLAG_REGISTERED);
72}
73
74static __kprobes bool trace_probe_has_gone(struct trace_probe *tp)
75{
76 return !!(kprobe_gone(&tp->rp.kp));
77}
78
79static __kprobes bool trace_probe_within_module(struct trace_probe *tp,
80 struct module *mod)
81{
82 int len = strlen(mod->name);
83 const char *name = trace_probe_symbol(tp);
84 return strncmp(mod->name, name, len) == 0 && name[len] == ':';
85}
86
87static __kprobes bool trace_probe_is_on_module(struct trace_probe *tp)
88{
89 return !!strchr(trace_probe_symbol(tp), ':');
90}
91
92static int register_probe_event(struct trace_probe *tp);
93static void unregister_probe_event(struct trace_probe *tp);
94
95static DEFINE_MUTEX(probe_lock);
96static LIST_HEAD(probe_list);
97
98static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
99static int kretprobe_dispatcher(struct kretprobe_instance *ri,
100 struct pt_regs *regs);
101
102/*
103 * Allocate new trace_probe and initialize it (including kprobes).
104 */
105static struct trace_probe *alloc_trace_probe(const char *group,
106 const char *event,
107 void *addr,
108 const char *symbol,
109 unsigned long offs,
110 int nargs, bool is_return)
111{
112 struct trace_probe *tp;
113 int ret = -ENOMEM;
114
115 tp = kzalloc(SIZEOF_TRACE_PROBE(nargs), GFP_KERNEL);
116 if (!tp)
117 return ERR_PTR(ret);
118
119 if (symbol) {
120 tp->symbol = kstrdup(symbol, GFP_KERNEL);
121 if (!tp->symbol)
122 goto error;
123 tp->rp.kp.symbol_name = tp->symbol;
124 tp->rp.kp.offset = offs;
125 } else
126 tp->rp.kp.addr = addr;
127
128 if (is_return)
129 tp->rp.handler = kretprobe_dispatcher;
130 else
131 tp->rp.kp.pre_handler = kprobe_dispatcher;
132
133 if (!event || !is_good_name(event)) {
134 ret = -EINVAL;
135 goto error;
136 }
137
138 tp->call.class = &tp->class;
139 tp->call.name = kstrdup(event, GFP_KERNEL);
140 if (!tp->call.name)
141 goto error;
142
143 if (!group || !is_good_name(group)) {
144 ret = -EINVAL;
145 goto error;
146 }
147
148 tp->class.system = kstrdup(group, GFP_KERNEL);
149 if (!tp->class.system)
150 goto error;
151
152 INIT_LIST_HEAD(&tp->list);
153 return tp;
154error:
155 kfree(tp->call.name);
156 kfree(tp->symbol);
157 kfree(tp);
158 return ERR_PTR(ret);
159}
160
161static void free_trace_probe(struct trace_probe *tp)
162{
163 int i;
164
165 for (i = 0; i < tp->nr_args; i++)
166 traceprobe_free_probe_arg(&tp->args[i]);
167
168 kfree(tp->call.class->system);
169 kfree(tp->call.name);
170 kfree(tp->symbol);
171 kfree(tp);
172}
173
174static struct trace_probe *find_trace_probe(const char *event,
175 const char *group)
176{
177 struct trace_probe *tp;
178
179 list_for_each_entry(tp, &probe_list, list)
180 if (strcmp(tp->call.name, event) == 0 &&
181 strcmp(tp->call.class->system, group) == 0)
182 return tp;
183 return NULL;
184}
185
186/* Enable trace_probe - @flag must be TP_FLAG_TRACE or TP_FLAG_PROFILE */
187static int enable_trace_probe(struct trace_probe *tp, int flag)
188{
189 int ret = 0;
190
191 tp->flags |= flag;
192 if (trace_probe_is_enabled(tp) && trace_probe_is_registered(tp) &&
193 !trace_probe_has_gone(tp)) {
194 if (trace_probe_is_return(tp))
195 ret = enable_kretprobe(&tp->rp);
196 else
197 ret = enable_kprobe(&tp->rp.kp);
198 }
199
200 return ret;
201}
202
203/* Disable trace_probe - @flag must be TP_FLAG_TRACE or TP_FLAG_PROFILE */
204static void disable_trace_probe(struct trace_probe *tp, int flag)
205{
206 tp->flags &= ~flag;
207 if (!trace_probe_is_enabled(tp) && trace_probe_is_registered(tp)) {
208 if (trace_probe_is_return(tp))
209 disable_kretprobe(&tp->rp);
210 else
211 disable_kprobe(&tp->rp.kp);
212 }
213}
214
215/* Internal register function - just handle k*probes and flags */
216static int __register_trace_probe(struct trace_probe *tp)
217{
218 int i, ret;
219
220 if (trace_probe_is_registered(tp))
221 return -EINVAL;
222
223 for (i = 0; i < tp->nr_args; i++)
224 traceprobe_update_arg(&tp->args[i]);
225
226 /* Set/clear disabled flag according to tp->flag */
227 if (trace_probe_is_enabled(tp))
228 tp->rp.kp.flags &= ~KPROBE_FLAG_DISABLED;
229 else
230 tp->rp.kp.flags |= KPROBE_FLAG_DISABLED;
231
232 if (trace_probe_is_return(tp))
233 ret = register_kretprobe(&tp->rp);
234 else
235 ret = register_kprobe(&tp->rp.kp);
236
237 if (ret == 0)
238 tp->flags |= TP_FLAG_REGISTERED;
239 else {
240 pr_warning("Could not insert probe at %s+%lu: %d\n",
241 trace_probe_symbol(tp), trace_probe_offset(tp), ret);
242 if (ret == -ENOENT && trace_probe_is_on_module(tp)) {
243 pr_warning("This probe might be able to register after"
244 "target module is loaded. Continue.\n");
245 ret = 0;
246 } else if (ret == -EILSEQ) {
247 pr_warning("Probing address(0x%p) is not an "
248 "instruction boundary.\n",
249 tp->rp.kp.addr);
250 ret = -EINVAL;
251 }
252 }
253
254 return ret;
255}
256
257/* Internal unregister function - just handle k*probes and flags */
258static void __unregister_trace_probe(struct trace_probe *tp)
259{
260 if (trace_probe_is_registered(tp)) {
261 if (trace_probe_is_return(tp))
262 unregister_kretprobe(&tp->rp);
263 else
264 unregister_kprobe(&tp->rp.kp);
265 tp->flags &= ~TP_FLAG_REGISTERED;
266 /* Cleanup kprobe for reuse */
267 if (tp->rp.kp.symbol_name)
268 tp->rp.kp.addr = NULL;
269 }
270}
271
272/* Unregister a trace_probe and probe_event: call with locking probe_lock */
273static int unregister_trace_probe(struct trace_probe *tp)
274{
275 /* Enabled event can not be unregistered */
276 if (trace_probe_is_enabled(tp))
277 return -EBUSY;
278
279 __unregister_trace_probe(tp);
280 list_del(&tp->list);
281 unregister_probe_event(tp);
282
283 return 0;
284}
285
286/* Register a trace_probe and probe_event */
287static int register_trace_probe(struct trace_probe *tp)
288{
289 struct trace_probe *old_tp;
290 int ret;
291
292 mutex_lock(&probe_lock);
293
294 /* Delete old (same name) event if exist */
295 old_tp = find_trace_probe(tp->call.name, tp->call.class->system);
296 if (old_tp) {
297 ret = unregister_trace_probe(old_tp);
298 if (ret < 0)
299 goto end;
300 free_trace_probe(old_tp);
301 }
302
303 /* Register new event */
304 ret = register_probe_event(tp);
305 if (ret) {
306 pr_warning("Failed to register probe event(%d)\n", ret);
307 goto end;
308 }
309
310 /* Register k*probe */
311 ret = __register_trace_probe(tp);
312 if (ret < 0)
313 unregister_probe_event(tp);
314 else
315 list_add_tail(&tp->list, &probe_list);
316
317end:
318 mutex_unlock(&probe_lock);
319 return ret;
320}
321
322/* Module notifier call back, checking event on the module */
323static int trace_probe_module_callback(struct notifier_block *nb,
324 unsigned long val, void *data)
325{
326 struct module *mod = data;
327 struct trace_probe *tp;
328 int ret;
329
330 if (val != MODULE_STATE_COMING)
331 return NOTIFY_DONE;
332
333 /* Update probes on coming module */
334 mutex_lock(&probe_lock);
335 list_for_each_entry(tp, &probe_list, list) {
336 if (trace_probe_within_module(tp, mod)) {
337 /* Don't need to check busy - this should have gone. */
338 __unregister_trace_probe(tp);
339 ret = __register_trace_probe(tp);
340 if (ret)
341 pr_warning("Failed to re-register probe %s on"
342 "%s: %d\n",
343 tp->call.name, mod->name, ret);
344 }
345 }
346 mutex_unlock(&probe_lock);
347
348 return NOTIFY_DONE;
349}
350
351static struct notifier_block trace_probe_module_nb = {
352 .notifier_call = trace_probe_module_callback,
353 .priority = 1 /* Invoked after kprobe module callback */
354};
355
356static int create_trace_probe(int argc, char **argv)
357{
358 /*
359 * Argument syntax:
360 * - Add kprobe: p[:[GRP/]EVENT] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
361 * - Add kretprobe: r[:[GRP/]EVENT] [MOD:]KSYM[+0] [FETCHARGS]
362 * Fetch args:
363 * $retval : fetch return value
364 * $stack : fetch stack address
365 * $stackN : fetch Nth of stack (N:0-)
366 * @ADDR : fetch memory at ADDR (ADDR should be in kernel)
367 * @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
368 * %REG : fetch register REG
369 * Dereferencing memory fetch:
370 * +|-offs(ARG) : fetch memory at ARG +|- offs address.
371 * Alias name of args:
372 * NAME=FETCHARG : set NAME as alias of FETCHARG.
373 * Type of args:
374 * FETCHARG:TYPE : use TYPE instead of unsigned long.
375 */
376 struct trace_probe *tp;
377 int i, ret = 0;
378 bool is_return = false, is_delete = false;
379 char *symbol = NULL, *event = NULL, *group = NULL;
380 char *arg;
381 unsigned long offset = 0;
382 void *addr = NULL;
383 char buf[MAX_EVENT_NAME_LEN];
384
385 /* argc must be >= 1 */
386 if (argv[0][0] == 'p')
387 is_return = false;
388 else if (argv[0][0] == 'r')
389 is_return = true;
390 else if (argv[0][0] == '-')
391 is_delete = true;
392 else {
393 pr_info("Probe definition must be started with 'p', 'r' or"
394 " '-'.\n");
395 return -EINVAL;
396 }
397
398 if (argv[0][1] == ':') {
399 event = &argv[0][2];
400 if (strchr(event, '/')) {
401 group = event;
402 event = strchr(group, '/') + 1;
403 event[-1] = '\0';
404 if (strlen(group) == 0) {
405 pr_info("Group name is not specified\n");
406 return -EINVAL;
407 }
408 }
409 if (strlen(event) == 0) {
410 pr_info("Event name is not specified\n");
411 return -EINVAL;
412 }
413 }
414 if (!group)
415 group = KPROBE_EVENT_SYSTEM;
416
417 if (is_delete) {
418 if (!event) {
419 pr_info("Delete command needs an event name.\n");
420 return -EINVAL;
421 }
422 mutex_lock(&probe_lock);
423 tp = find_trace_probe(event, group);
424 if (!tp) {
425 mutex_unlock(&probe_lock);
426 pr_info("Event %s/%s doesn't exist.\n", group, event);
427 return -ENOENT;
428 }
429 /* delete an event */
430 ret = unregister_trace_probe(tp);
431 if (ret == 0)
432 free_trace_probe(tp);
433 mutex_unlock(&probe_lock);
434 return ret;
435 }
436
437 if (argc < 2) {
438 pr_info("Probe point is not specified.\n");
439 return -EINVAL;
440 }
441 if (isdigit(argv[1][0])) {
442 if (is_return) {
443 pr_info("Return probe point must be a symbol.\n");
444 return -EINVAL;
445 }
446 /* an address specified */
447 ret = strict_strtoul(&argv[1][0], 0, (unsigned long *)&addr);
448 if (ret) {
449 pr_info("Failed to parse address.\n");
450 return ret;
451 }
452 } else {
453 /* a symbol specified */
454 symbol = argv[1];
455 /* TODO: support .init module functions */
456 ret = traceprobe_split_symbol_offset(symbol, &offset);
457 if (ret) {
458 pr_info("Failed to parse symbol.\n");
459 return ret;
460 }
461 if (offset && is_return) {
462 pr_info("Return probe must be used without offset.\n");
463 return -EINVAL;
464 }
465 }
466 argc -= 2; argv += 2;
467
468 /* setup a probe */
469 if (!event) {
470 /* Make a new event name */
471 if (symbol)
472 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
473 is_return ? 'r' : 'p', symbol, offset);
474 else
475 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
476 is_return ? 'r' : 'p', addr);
477 event = buf;
478 }
479 tp = alloc_trace_probe(group, event, addr, symbol, offset, argc,
480 is_return);
481 if (IS_ERR(tp)) {
482 pr_info("Failed to allocate trace_probe.(%d)\n",
483 (int)PTR_ERR(tp));
484 return PTR_ERR(tp);
485 }
486
487 /* parse arguments */
488 ret = 0;
489 for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
490 /* Increment count for freeing args in error case */
491 tp->nr_args++;
492
493 /* Parse argument name */
494 arg = strchr(argv[i], '=');
495 if (arg) {
496 *arg++ = '\0';
497 tp->args[i].name = kstrdup(argv[i], GFP_KERNEL);
498 } else {
499 arg = argv[i];
500 /* If argument name is omitted, set "argN" */
501 snprintf(buf, MAX_EVENT_NAME_LEN, "arg%d", i + 1);
502 tp->args[i].name = kstrdup(buf, GFP_KERNEL);
503 }
504
505 if (!tp->args[i].name) {
506 pr_info("Failed to allocate argument[%d] name.\n", i);
507 ret = -ENOMEM;
508 goto error;
509 }
510
511 if (!is_good_name(tp->args[i].name)) {
512 pr_info("Invalid argument[%d] name: %s\n",
513 i, tp->args[i].name);
514 ret = -EINVAL;
515 goto error;
516 }
517
518 if (traceprobe_conflict_field_name(tp->args[i].name,
519 tp->args, i)) {
520 pr_info("Argument[%d] name '%s' conflicts with "
521 "another field.\n", i, argv[i]);
522 ret = -EINVAL;
523 goto error;
524 }
525
526 /* Parse fetch argument */
527 ret = traceprobe_parse_probe_arg(arg, &tp->size, &tp->args[i],
528 is_return, true);
529 if (ret) {
530 pr_info("Parse error at argument[%d]. (%d)\n", i, ret);
531 goto error;
532 }
533 }
534
535 ret = register_trace_probe(tp);
536 if (ret)
537 goto error;
538 return 0;
539
540error:
541 free_trace_probe(tp);
542 return ret;
543}
544
545static int release_all_trace_probes(void)
546{
547 struct trace_probe *tp;
548 int ret = 0;
549
550 mutex_lock(&probe_lock);
551 /* Ensure no probe is in use. */
552 list_for_each_entry(tp, &probe_list, list)
553 if (trace_probe_is_enabled(tp)) {
554 ret = -EBUSY;
555 goto end;
556 }
557 /* TODO: Use batch unregistration */
558 while (!list_empty(&probe_list)) {
559 tp = list_entry(probe_list.next, struct trace_probe, list);
560 unregister_trace_probe(tp);
561 free_trace_probe(tp);
562 }
563
564end:
565 mutex_unlock(&probe_lock);
566
567 return ret;
568}
569
570/* Probes listing interfaces */
571static void *probes_seq_start(struct seq_file *m, loff_t *pos)
572{
573 mutex_lock(&probe_lock);
574 return seq_list_start(&probe_list, *pos);
575}
576
577static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
578{
579 return seq_list_next(v, &probe_list, pos);
580}
581
582static void probes_seq_stop(struct seq_file *m, void *v)
583{
584 mutex_unlock(&probe_lock);
585}
586
587static int probes_seq_show(struct seq_file *m, void *v)
588{
589 struct trace_probe *tp = v;
590 int i;
591
592 seq_printf(m, "%c", trace_probe_is_return(tp) ? 'r' : 'p');
593 seq_printf(m, ":%s/%s", tp->call.class->system, tp->call.name);
594
595 if (!tp->symbol)
596 seq_printf(m, " 0x%p", tp->rp.kp.addr);
597 else if (tp->rp.kp.offset)
598 seq_printf(m, " %s+%u", trace_probe_symbol(tp),
599 tp->rp.kp.offset);
600 else
601 seq_printf(m, " %s", trace_probe_symbol(tp));
602
603 for (i = 0; i < tp->nr_args; i++)
604 seq_printf(m, " %s=%s", tp->args[i].name, tp->args[i].comm);
605 seq_printf(m, "\n");
606
607 return 0;
608}
609
610static const struct seq_operations probes_seq_op = {
611 .start = probes_seq_start,
612 .next = probes_seq_next,
613 .stop = probes_seq_stop,
614 .show = probes_seq_show
615};
616
617static int probes_open(struct inode *inode, struct file *file)
618{
619 int ret;
620
621 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
622 ret = release_all_trace_probes();
623 if (ret < 0)
624 return ret;
625 }
626
627 return seq_open(file, &probes_seq_op);
628}
629
630static ssize_t probes_write(struct file *file, const char __user *buffer,
631 size_t count, loff_t *ppos)
632{
633 return traceprobe_probes_write(file, buffer, count, ppos,
634 create_trace_probe);
635}
636
637static const struct file_operations kprobe_events_ops = {
638 .owner = THIS_MODULE,
639 .open = probes_open,
640 .read = seq_read,
641 .llseek = seq_lseek,
642 .release = seq_release,
643 .write = probes_write,
644};
645
646/* Probes profiling interfaces */
647static int probes_profile_seq_show(struct seq_file *m, void *v)
648{
649 struct trace_probe *tp = v;
650
651 seq_printf(m, " %-44s %15lu %15lu\n", tp->call.name, tp->nhit,
652 tp->rp.kp.nmissed);
653
654 return 0;
655}
656
657static const struct seq_operations profile_seq_op = {
658 .start = probes_seq_start,
659 .next = probes_seq_next,
660 .stop = probes_seq_stop,
661 .show = probes_profile_seq_show
662};
663
664static int profile_open(struct inode *inode, struct file *file)
665{
666 return seq_open(file, &profile_seq_op);
667}
668
669static const struct file_operations kprobe_profile_ops = {
670 .owner = THIS_MODULE,
671 .open = profile_open,
672 .read = seq_read,
673 .llseek = seq_lseek,
674 .release = seq_release,
675};
676
677/* Sum up total data length for dynamic arraies (strings) */
678static __kprobes int __get_data_size(struct trace_probe *tp,
679 struct pt_regs *regs)
680{
681 int i, ret = 0;
682 u32 len;
683
684 for (i = 0; i < tp->nr_args; i++)
685 if (unlikely(tp->args[i].fetch_size.fn)) {
686 call_fetch(&tp->args[i].fetch_size, regs, &len);
687 ret += len;
688 }
689
690 return ret;
691}
692
693/* Store the value of each argument */
694static __kprobes void store_trace_args(int ent_size, struct trace_probe *tp,
695 struct pt_regs *regs,
696 u8 *data, int maxlen)
697{
698 int i;
699 u32 end = tp->size;
700 u32 *dl; /* Data (relative) location */
701
702 for (i = 0; i < tp->nr_args; i++) {
703 if (unlikely(tp->args[i].fetch_size.fn)) {
704 /*
705 * First, we set the relative location and
706 * maximum data length to *dl
707 */
708 dl = (u32 *)(data + tp->args[i].offset);
709 *dl = make_data_rloc(maxlen, end - tp->args[i].offset);
710 /* Then try to fetch string or dynamic array data */
711 call_fetch(&tp->args[i].fetch, regs, dl);
712 /* Reduce maximum length */
713 end += get_rloc_len(*dl);
714 maxlen -= get_rloc_len(*dl);
715 /* Trick here, convert data_rloc to data_loc */
716 *dl = convert_rloc_to_loc(*dl,
717 ent_size + tp->args[i].offset);
718 } else
719 /* Just fetching data normally */
720 call_fetch(&tp->args[i].fetch, regs,
721 data + tp->args[i].offset);
722 }
723}
724
725/* Kprobe handler */
726static __kprobes void kprobe_trace_func(struct kprobe *kp, struct pt_regs *regs)
727{
728 struct trace_probe *tp = container_of(kp, struct trace_probe, rp.kp);
729 struct kprobe_trace_entry_head *entry;
730 struct ring_buffer_event *event;
731 struct ring_buffer *buffer;
732 int size, dsize, pc;
733 unsigned long irq_flags;
734 struct ftrace_event_call *call = &tp->call;
735
736 tp->nhit++;
737
738 local_save_flags(irq_flags);
739 pc = preempt_count();
740
741 dsize = __get_data_size(tp, regs);
742 size = sizeof(*entry) + tp->size + dsize;
743
744 event = trace_current_buffer_lock_reserve(&buffer, call->event.type,
745 size, irq_flags, pc);
746 if (!event)
747 return;
748
749 entry = ring_buffer_event_data(event);
750 entry->ip = (unsigned long)kp->addr;
751 store_trace_args(sizeof(*entry), tp, regs, (u8 *)&entry[1], dsize);
752
753 if (!filter_current_check_discard(buffer, call, entry, event))
754 trace_nowake_buffer_unlock_commit_regs(buffer, event,
755 irq_flags, pc, regs);
756}
757
758/* Kretprobe handler */
759static __kprobes void kretprobe_trace_func(struct kretprobe_instance *ri,
760 struct pt_regs *regs)
761{
762 struct trace_probe *tp = container_of(ri->rp, struct trace_probe, rp);
763 struct kretprobe_trace_entry_head *entry;
764 struct ring_buffer_event *event;
765 struct ring_buffer *buffer;
766 int size, pc, dsize;
767 unsigned long irq_flags;
768 struct ftrace_event_call *call = &tp->call;
769
770 local_save_flags(irq_flags);
771 pc = preempt_count();
772
773 dsize = __get_data_size(tp, regs);
774 size = sizeof(*entry) + tp->size + dsize;
775
776 event = trace_current_buffer_lock_reserve(&buffer, call->event.type,
777 size, irq_flags, pc);
778 if (!event)
779 return;
780
781 entry = ring_buffer_event_data(event);
782 entry->func = (unsigned long)tp->rp.kp.addr;
783 entry->ret_ip = (unsigned long)ri->ret_addr;
784 store_trace_args(sizeof(*entry), tp, regs, (u8 *)&entry[1], dsize);
785
786 if (!filter_current_check_discard(buffer, call, entry, event))
787 trace_nowake_buffer_unlock_commit_regs(buffer, event,
788 irq_flags, pc, regs);
789}
790
791/* Event entry printers */
792enum print_line_t
793print_kprobe_event(struct trace_iterator *iter, int flags,
794 struct trace_event *event)
795{
796 struct kprobe_trace_entry_head *field;
797 struct trace_seq *s = &iter->seq;
798 struct trace_probe *tp;
799 u8 *data;
800 int i;
801
802 field = (struct kprobe_trace_entry_head *)iter->ent;
803 tp = container_of(event, struct trace_probe, call.event);
804
805 if (!trace_seq_printf(s, "%s: (", tp->call.name))
806 goto partial;
807
808 if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
809 goto partial;
810
811 if (!trace_seq_puts(s, ")"))
812 goto partial;
813
814 data = (u8 *)&field[1];
815 for (i = 0; i < tp->nr_args; i++)
816 if (!tp->args[i].type->print(s, tp->args[i].name,
817 data + tp->args[i].offset, field))
818 goto partial;
819
820 if (!trace_seq_puts(s, "\n"))
821 goto partial;
822
823 return TRACE_TYPE_HANDLED;
824partial:
825 return TRACE_TYPE_PARTIAL_LINE;
826}
827
828enum print_line_t
829print_kretprobe_event(struct trace_iterator *iter, int flags,
830 struct trace_event *event)
831{
832 struct kretprobe_trace_entry_head *field;
833 struct trace_seq *s = &iter->seq;
834 struct trace_probe *tp;
835 u8 *data;
836 int i;
837
838 field = (struct kretprobe_trace_entry_head *)iter->ent;
839 tp = container_of(event, struct trace_probe, call.event);
840
841 if (!trace_seq_printf(s, "%s: (", tp->call.name))
842 goto partial;
843
844 if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
845 goto partial;
846
847 if (!trace_seq_puts(s, " <- "))
848 goto partial;
849
850 if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
851 goto partial;
852
853 if (!trace_seq_puts(s, ")"))
854 goto partial;
855
856 data = (u8 *)&field[1];
857 for (i = 0; i < tp->nr_args; i++)
858 if (!tp->args[i].type->print(s, tp->args[i].name,
859 data + tp->args[i].offset, field))
860 goto partial;
861
862 if (!trace_seq_puts(s, "\n"))
863 goto partial;
864
865 return TRACE_TYPE_HANDLED;
866partial:
867 return TRACE_TYPE_PARTIAL_LINE;
868}
869
870
871static int kprobe_event_define_fields(struct ftrace_event_call *event_call)
872{
873 int ret, i;
874 struct kprobe_trace_entry_head field;
875 struct trace_probe *tp = (struct trace_probe *)event_call->data;
876
877 DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
878 /* Set argument names as fields */
879 for (i = 0; i < tp->nr_args; i++) {
880 ret = trace_define_field(event_call, tp->args[i].type->fmttype,
881 tp->args[i].name,
882 sizeof(field) + tp->args[i].offset,
883 tp->args[i].type->size,
884 tp->args[i].type->is_signed,
885 FILTER_OTHER);
886 if (ret)
887 return ret;
888 }
889 return 0;
890}
891
892static int kretprobe_event_define_fields(struct ftrace_event_call *event_call)
893{
894 int ret, i;
895 struct kretprobe_trace_entry_head field;
896 struct trace_probe *tp = (struct trace_probe *)event_call->data;
897
898 DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
899 DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
900 /* Set argument names as fields */
901 for (i = 0; i < tp->nr_args; i++) {
902 ret = trace_define_field(event_call, tp->args[i].type->fmttype,
903 tp->args[i].name,
904 sizeof(field) + tp->args[i].offset,
905 tp->args[i].type->size,
906 tp->args[i].type->is_signed,
907 FILTER_OTHER);
908 if (ret)
909 return ret;
910 }
911 return 0;
912}
913
914static int __set_print_fmt(struct trace_probe *tp, char *buf, int len)
915{
916 int i;
917 int pos = 0;
918
919 const char *fmt, *arg;
920
921 if (!trace_probe_is_return(tp)) {
922 fmt = "(%lx)";
923 arg = "REC->" FIELD_STRING_IP;
924 } else {
925 fmt = "(%lx <- %lx)";
926 arg = "REC->" FIELD_STRING_FUNC ", REC->" FIELD_STRING_RETIP;
927 }
928
929 /* When len=0, we just calculate the needed length */
930#define LEN_OR_ZERO (len ? len - pos : 0)
931
932 pos += snprintf(buf + pos, LEN_OR_ZERO, "\"%s", fmt);
933
934 for (i = 0; i < tp->nr_args; i++) {
935 pos += snprintf(buf + pos, LEN_OR_ZERO, " %s=%s",
936 tp->args[i].name, tp->args[i].type->fmt);
937 }
938
939 pos += snprintf(buf + pos, LEN_OR_ZERO, "\", %s", arg);
940
941 for (i = 0; i < tp->nr_args; i++) {
942 if (strcmp(tp->args[i].type->name, "string") == 0)
943 pos += snprintf(buf + pos, LEN_OR_ZERO,
944 ", __get_str(%s)",
945 tp->args[i].name);
946 else
947 pos += snprintf(buf + pos, LEN_OR_ZERO, ", REC->%s",
948 tp->args[i].name);
949 }
950
951#undef LEN_OR_ZERO
952
953 /* return the length of print_fmt */
954 return pos;
955}
956
957static int set_print_fmt(struct trace_probe *tp)
958{
959 int len;
960 char *print_fmt;
961
962 /* First: called with 0 length to calculate the needed length */
963 len = __set_print_fmt(tp, NULL, 0);
964 print_fmt = kmalloc(len + 1, GFP_KERNEL);
965 if (!print_fmt)
966 return -ENOMEM;
967
968 /* Second: actually write the @print_fmt */
969 __set_print_fmt(tp, print_fmt, len + 1);
970 tp->call.print_fmt = print_fmt;
971
972 return 0;
973}
974
975#ifdef CONFIG_PERF_EVENTS
976
977/* Kprobe profile handler */
978static __kprobes void kprobe_perf_func(struct kprobe *kp,
979 struct pt_regs *regs)
980{
981 struct trace_probe *tp = container_of(kp, struct trace_probe, rp.kp);
982 struct ftrace_event_call *call = &tp->call;
983 struct kprobe_trace_entry_head *entry;
984 struct hlist_head *head;
985 int size, __size, dsize;
986 int rctx;
987
988 dsize = __get_data_size(tp, regs);
989 __size = sizeof(*entry) + tp->size + dsize;
990 size = ALIGN(__size + sizeof(u32), sizeof(u64));
991 size -= sizeof(u32);
992 if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE,
993 "profile buffer not large enough"))
994 return;
995
996 entry = perf_trace_buf_prepare(size, call->event.type, regs, &rctx);
997 if (!entry)
998 return;
999
1000 entry->ip = (unsigned long)kp->addr;
1001 memset(&entry[1], 0, dsize);
1002 store_trace_args(sizeof(*entry), tp, regs, (u8 *)&entry[1], dsize);
1003
1004 head = this_cpu_ptr(call->perf_events);
1005 perf_trace_buf_submit(entry, size, rctx, entry->ip, 1, regs, head);
1006}
1007
1008/* Kretprobe profile handler */
1009static __kprobes void kretprobe_perf_func(struct kretprobe_instance *ri,
1010 struct pt_regs *regs)
1011{
1012 struct trace_probe *tp = container_of(ri->rp, struct trace_probe, rp);
1013 struct ftrace_event_call *call = &tp->call;
1014 struct kretprobe_trace_entry_head *entry;
1015 struct hlist_head *head;
1016 int size, __size, dsize;
1017 int rctx;
1018
1019 dsize = __get_data_size(tp, regs);
1020 __size = sizeof(*entry) + tp->size + dsize;
1021 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1022 size -= sizeof(u32);
1023 if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE,
1024 "profile buffer not large enough"))
1025 return;
1026
1027 entry = perf_trace_buf_prepare(size, call->event.type, regs, &rctx);
1028 if (!entry)
1029 return;
1030
1031 entry->func = (unsigned long)tp->rp.kp.addr;
1032 entry->ret_ip = (unsigned long)ri->ret_addr;
1033 store_trace_args(sizeof(*entry), tp, regs, (u8 *)&entry[1], dsize);
1034
1035 head = this_cpu_ptr(call->perf_events);
1036 perf_trace_buf_submit(entry, size, rctx, entry->ret_ip, 1, regs, head);
1037}
1038#endif /* CONFIG_PERF_EVENTS */
1039
1040static __kprobes
1041int kprobe_register(struct ftrace_event_call *event,
1042 enum trace_reg type, void *data)
1043{
1044 struct trace_probe *tp = (struct trace_probe *)event->data;
1045
1046 switch (type) {
1047 case TRACE_REG_REGISTER:
1048 return enable_trace_probe(tp, TP_FLAG_TRACE);
1049 case TRACE_REG_UNREGISTER:
1050 disable_trace_probe(tp, TP_FLAG_TRACE);
1051 return 0;
1052
1053#ifdef CONFIG_PERF_EVENTS
1054 case TRACE_REG_PERF_REGISTER:
1055 return enable_trace_probe(tp, TP_FLAG_PROFILE);
1056 case TRACE_REG_PERF_UNREGISTER:
1057 disable_trace_probe(tp, TP_FLAG_PROFILE);
1058 return 0;
1059 case TRACE_REG_PERF_OPEN:
1060 case TRACE_REG_PERF_CLOSE:
1061 case TRACE_REG_PERF_ADD:
1062 case TRACE_REG_PERF_DEL:
1063 return 0;
1064#endif
1065 }
1066 return 0;
1067}
1068
1069static __kprobes
1070int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1071{
1072 struct trace_probe *tp = container_of(kp, struct trace_probe, rp.kp);
1073
1074 if (tp->flags & TP_FLAG_TRACE)
1075 kprobe_trace_func(kp, regs);
1076#ifdef CONFIG_PERF_EVENTS
1077 if (tp->flags & TP_FLAG_PROFILE)
1078 kprobe_perf_func(kp, regs);
1079#endif
1080 return 0; /* We don't tweek kernel, so just return 0 */
1081}
1082
1083static __kprobes
1084int kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1085{
1086 struct trace_probe *tp = container_of(ri->rp, struct trace_probe, rp);
1087
1088 if (tp->flags & TP_FLAG_TRACE)
1089 kretprobe_trace_func(ri, regs);
1090#ifdef CONFIG_PERF_EVENTS
1091 if (tp->flags & TP_FLAG_PROFILE)
1092 kretprobe_perf_func(ri, regs);
1093#endif
1094 return 0; /* We don't tweek kernel, so just return 0 */
1095}
1096
1097static struct trace_event_functions kretprobe_funcs = {
1098 .trace = print_kretprobe_event
1099};
1100
1101static struct trace_event_functions kprobe_funcs = {
1102 .trace = print_kprobe_event
1103};
1104
1105static int register_probe_event(struct trace_probe *tp)
1106{
1107 struct ftrace_event_call *call = &tp->call;
1108 int ret;
1109
1110 /* Initialize ftrace_event_call */
1111 INIT_LIST_HEAD(&call->class->fields);
1112 if (trace_probe_is_return(tp)) {
1113 call->event.funcs = &kretprobe_funcs;
1114 call->class->define_fields = kretprobe_event_define_fields;
1115 } else {
1116 call->event.funcs = &kprobe_funcs;
1117 call->class->define_fields = kprobe_event_define_fields;
1118 }
1119 if (set_print_fmt(tp) < 0)
1120 return -ENOMEM;
1121 ret = register_ftrace_event(&call->event);
1122 if (!ret) {
1123 kfree(call->print_fmt);
1124 return -ENODEV;
1125 }
1126 call->flags = 0;
1127 call->class->reg = kprobe_register;
1128 call->data = tp;
1129 ret = trace_add_event_call(call);
1130 if (ret) {
1131 pr_info("Failed to register kprobe event: %s\n", call->name);
1132 kfree(call->print_fmt);
1133 unregister_ftrace_event(&call->event);
1134 }
1135 return ret;
1136}
1137
1138static void unregister_probe_event(struct trace_probe *tp)
1139{
1140 /* tp->event is unregistered in trace_remove_event_call() */
1141 trace_remove_event_call(&tp->call);
1142 kfree(tp->call.print_fmt);
1143}
1144
1145/* Make a debugfs interface for controlling probe points */
1146static __init int init_kprobe_trace(void)
1147{
1148 struct dentry *d_tracer;
1149 struct dentry *entry;
1150
1151 if (register_module_notifier(&trace_probe_module_nb))
1152 return -EINVAL;
1153
1154 d_tracer = tracing_init_dentry();
1155 if (!d_tracer)
1156 return 0;
1157
1158 entry = debugfs_create_file("kprobe_events", 0644, d_tracer,
1159 NULL, &kprobe_events_ops);
1160
1161 /* Event list interface */
1162 if (!entry)
1163 pr_warning("Could not create debugfs "
1164 "'kprobe_events' entry\n");
1165
1166 /* Profile interface */
1167 entry = debugfs_create_file("kprobe_profile", 0444, d_tracer,
1168 NULL, &kprobe_profile_ops);
1169
1170 if (!entry)
1171 pr_warning("Could not create debugfs "
1172 "'kprobe_profile' entry\n");
1173 return 0;
1174}
1175fs_initcall(init_kprobe_trace);
1176
1177
1178#ifdef CONFIG_FTRACE_STARTUP_TEST
1179
1180/*
1181 * The "__used" keeps gcc from removing the function symbol
1182 * from the kallsyms table.
1183 */
1184static __used int kprobe_trace_selftest_target(int a1, int a2, int a3,
1185 int a4, int a5, int a6)
1186{
1187 return a1 + a2 + a3 + a4 + a5 + a6;
1188}
1189
1190static __init int kprobe_trace_self_tests_init(void)
1191{
1192 int ret, warn = 0;
1193 int (*target)(int, int, int, int, int, int);
1194 struct trace_probe *tp;
1195
1196 target = kprobe_trace_selftest_target;
1197
1198 pr_info("Testing kprobe tracing: ");
1199
1200 ret = traceprobe_command("p:testprobe kprobe_trace_selftest_target "
1201 "$stack $stack0 +0($stack)",
1202 create_trace_probe);
1203 if (WARN_ON_ONCE(ret)) {
1204 pr_warning("error on probing function entry.\n");
1205 warn++;
1206 } else {
1207 /* Enable trace point */
1208 tp = find_trace_probe("testprobe", KPROBE_EVENT_SYSTEM);
1209 if (WARN_ON_ONCE(tp == NULL)) {
1210 pr_warning("error on getting new probe.\n");
1211 warn++;
1212 } else
1213 enable_trace_probe(tp, TP_FLAG_TRACE);
1214 }
1215
1216 ret = traceprobe_command("r:testprobe2 kprobe_trace_selftest_target "
1217 "$retval", create_trace_probe);
1218 if (WARN_ON_ONCE(ret)) {
1219 pr_warning("error on probing function return.\n");
1220 warn++;
1221 } else {
1222 /* Enable trace point */
1223 tp = find_trace_probe("testprobe2", KPROBE_EVENT_SYSTEM);
1224 if (WARN_ON_ONCE(tp == NULL)) {
1225 pr_warning("error on getting new probe.\n");
1226 warn++;
1227 } else
1228 enable_trace_probe(tp, TP_FLAG_TRACE);
1229 }
1230
1231 if (warn)
1232 goto end;
1233
1234 ret = target(1, 2, 3, 4, 5, 6);
1235
1236 /* Disable trace points before removing it */
1237 tp = find_trace_probe("testprobe", KPROBE_EVENT_SYSTEM);
1238 if (WARN_ON_ONCE(tp == NULL)) {
1239 pr_warning("error on getting test probe.\n");
1240 warn++;
1241 } else
1242 disable_trace_probe(tp, TP_FLAG_TRACE);
1243
1244 tp = find_trace_probe("testprobe2", KPROBE_EVENT_SYSTEM);
1245 if (WARN_ON_ONCE(tp == NULL)) {
1246 pr_warning("error on getting 2nd test probe.\n");
1247 warn++;
1248 } else
1249 disable_trace_probe(tp, TP_FLAG_TRACE);
1250
1251 ret = traceprobe_command("-:testprobe", create_trace_probe);
1252 if (WARN_ON_ONCE(ret)) {
1253 pr_warning("error on deleting a probe.\n");
1254 warn++;
1255 }
1256
1257 ret = traceprobe_command("-:testprobe2", create_trace_probe);
1258 if (WARN_ON_ONCE(ret)) {
1259 pr_warning("error on deleting a probe.\n");
1260 warn++;
1261 }
1262
1263end:
1264 release_all_trace_probes();
1265 if (warn)
1266 pr_cont("NG: Some tests are failed. Please check them.\n");
1267 else
1268 pr_cont("OK\n");
1269 return 0;
1270}
1271
1272late_initcall(kprobe_trace_self_tests_init);
1273
1274#endif
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Kprobes-based tracing events
4 *
5 * Created by Masami Hiramatsu <mhiramat@redhat.com>
6 *
7 */
8#define pr_fmt(fmt) "trace_kprobe: " fmt
9
10#include <linux/bpf-cgroup.h>
11#include <linux/security.h>
12#include <linux/module.h>
13#include <linux/uaccess.h>
14#include <linux/rculist.h>
15#include <linux/error-injection.h>
16
17#include <asm/setup.h> /* for COMMAND_LINE_SIZE */
18
19#include "trace_dynevent.h"
20#include "trace_kprobe_selftest.h"
21#include "trace_probe.h"
22#include "trace_probe_tmpl.h"
23#include "trace_probe_kernel.h"
24
25#define KPROBE_EVENT_SYSTEM "kprobes"
26#define KRETPROBE_MAXACTIVE_MAX 4096
27
28/* Kprobe early definition from command line */
29static char kprobe_boot_events_buf[COMMAND_LINE_SIZE] __initdata;
30
31static int __init set_kprobe_boot_events(char *str)
32{
33 strscpy(kprobe_boot_events_buf, str, COMMAND_LINE_SIZE);
34 disable_tracing_selftest("running kprobe events");
35
36 return 1;
37}
38__setup("kprobe_event=", set_kprobe_boot_events);
39
40static int trace_kprobe_create(const char *raw_command);
41static int trace_kprobe_show(struct seq_file *m, struct dyn_event *ev);
42static int trace_kprobe_release(struct dyn_event *ev);
43static bool trace_kprobe_is_busy(struct dyn_event *ev);
44static bool trace_kprobe_match(const char *system, const char *event,
45 int argc, const char **argv, struct dyn_event *ev);
46
47static struct dyn_event_operations trace_kprobe_ops = {
48 .create = trace_kprobe_create,
49 .show = trace_kprobe_show,
50 .is_busy = trace_kprobe_is_busy,
51 .free = trace_kprobe_release,
52 .match = trace_kprobe_match,
53};
54
55/*
56 * Kprobe event core functions
57 */
58struct trace_kprobe {
59 struct dyn_event devent;
60 struct kretprobe rp; /* Use rp.kp for kprobe use */
61 unsigned long __percpu *nhit;
62 const char *symbol; /* symbol name */
63 struct trace_probe tp;
64};
65
66static bool is_trace_kprobe(struct dyn_event *ev)
67{
68 return ev->ops == &trace_kprobe_ops;
69}
70
71static struct trace_kprobe *to_trace_kprobe(struct dyn_event *ev)
72{
73 return container_of(ev, struct trace_kprobe, devent);
74}
75
76/**
77 * for_each_trace_kprobe - iterate over the trace_kprobe list
78 * @pos: the struct trace_kprobe * for each entry
79 * @dpos: the struct dyn_event * to use as a loop cursor
80 */
81#define for_each_trace_kprobe(pos, dpos) \
82 for_each_dyn_event(dpos) \
83 if (is_trace_kprobe(dpos) && (pos = to_trace_kprobe(dpos)))
84
85static nokprobe_inline bool trace_kprobe_is_return(struct trace_kprobe *tk)
86{
87 return tk->rp.handler != NULL;
88}
89
90static nokprobe_inline const char *trace_kprobe_symbol(struct trace_kprobe *tk)
91{
92 return tk->symbol ? tk->symbol : "unknown";
93}
94
95static nokprobe_inline unsigned long trace_kprobe_offset(struct trace_kprobe *tk)
96{
97 return tk->rp.kp.offset;
98}
99
100static nokprobe_inline bool trace_kprobe_has_gone(struct trace_kprobe *tk)
101{
102 return kprobe_gone(&tk->rp.kp);
103}
104
105static nokprobe_inline bool trace_kprobe_within_module(struct trace_kprobe *tk,
106 struct module *mod)
107{
108 int len = strlen(module_name(mod));
109 const char *name = trace_kprobe_symbol(tk);
110
111 return strncmp(module_name(mod), name, len) == 0 && name[len] == ':';
112}
113
114static nokprobe_inline bool trace_kprobe_module_exist(struct trace_kprobe *tk)
115{
116 char *p;
117 bool ret;
118
119 if (!tk->symbol)
120 return false;
121 p = strchr(tk->symbol, ':');
122 if (!p)
123 return true;
124 *p = '\0';
125 rcu_read_lock_sched();
126 ret = !!find_module(tk->symbol);
127 rcu_read_unlock_sched();
128 *p = ':';
129
130 return ret;
131}
132
133static bool trace_kprobe_is_busy(struct dyn_event *ev)
134{
135 struct trace_kprobe *tk = to_trace_kprobe(ev);
136
137 return trace_probe_is_enabled(&tk->tp);
138}
139
140static bool trace_kprobe_match_command_head(struct trace_kprobe *tk,
141 int argc, const char **argv)
142{
143 char buf[MAX_ARGSTR_LEN + 1];
144
145 if (!argc)
146 return true;
147
148 if (!tk->symbol)
149 snprintf(buf, sizeof(buf), "0x%p", tk->rp.kp.addr);
150 else if (tk->rp.kp.offset)
151 snprintf(buf, sizeof(buf), "%s+%u",
152 trace_kprobe_symbol(tk), tk->rp.kp.offset);
153 else
154 snprintf(buf, sizeof(buf), "%s", trace_kprobe_symbol(tk));
155 if (strcmp(buf, argv[0]))
156 return false;
157 argc--; argv++;
158
159 return trace_probe_match_command_args(&tk->tp, argc, argv);
160}
161
162static bool trace_kprobe_match(const char *system, const char *event,
163 int argc, const char **argv, struct dyn_event *ev)
164{
165 struct trace_kprobe *tk = to_trace_kprobe(ev);
166
167 return (event[0] == '\0' ||
168 strcmp(trace_probe_name(&tk->tp), event) == 0) &&
169 (!system || strcmp(trace_probe_group_name(&tk->tp), system) == 0) &&
170 trace_kprobe_match_command_head(tk, argc, argv);
171}
172
173static nokprobe_inline unsigned long trace_kprobe_nhit(struct trace_kprobe *tk)
174{
175 unsigned long nhit = 0;
176 int cpu;
177
178 for_each_possible_cpu(cpu)
179 nhit += *per_cpu_ptr(tk->nhit, cpu);
180
181 return nhit;
182}
183
184static nokprobe_inline bool trace_kprobe_is_registered(struct trace_kprobe *tk)
185{
186 return !(list_empty(&tk->rp.kp.list) &&
187 hlist_unhashed(&tk->rp.kp.hlist));
188}
189
190/* Return 0 if it fails to find the symbol address */
191static nokprobe_inline
192unsigned long trace_kprobe_address(struct trace_kprobe *tk)
193{
194 unsigned long addr;
195
196 if (tk->symbol) {
197 addr = (unsigned long)
198 kallsyms_lookup_name(trace_kprobe_symbol(tk));
199 if (addr)
200 addr += tk->rp.kp.offset;
201 } else {
202 addr = (unsigned long)tk->rp.kp.addr;
203 }
204 return addr;
205}
206
207static nokprobe_inline struct trace_kprobe *
208trace_kprobe_primary_from_call(struct trace_event_call *call)
209{
210 struct trace_probe *tp;
211
212 tp = trace_probe_primary_from_call(call);
213 if (WARN_ON_ONCE(!tp))
214 return NULL;
215
216 return container_of(tp, struct trace_kprobe, tp);
217}
218
219bool trace_kprobe_on_func_entry(struct trace_event_call *call)
220{
221 struct trace_kprobe *tk = trace_kprobe_primary_from_call(call);
222
223 return tk ? (kprobe_on_func_entry(tk->rp.kp.addr,
224 tk->rp.kp.addr ? NULL : tk->rp.kp.symbol_name,
225 tk->rp.kp.addr ? 0 : tk->rp.kp.offset) == 0) : false;
226}
227
228bool trace_kprobe_error_injectable(struct trace_event_call *call)
229{
230 struct trace_kprobe *tk = trace_kprobe_primary_from_call(call);
231
232 return tk ? within_error_injection_list(trace_kprobe_address(tk)) :
233 false;
234}
235
236static int register_kprobe_event(struct trace_kprobe *tk);
237static int unregister_kprobe_event(struct trace_kprobe *tk);
238
239static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
240static int kretprobe_dispatcher(struct kretprobe_instance *ri,
241 struct pt_regs *regs);
242
243static void free_trace_kprobe(struct trace_kprobe *tk)
244{
245 if (tk) {
246 trace_probe_cleanup(&tk->tp);
247 kfree(tk->symbol);
248 free_percpu(tk->nhit);
249 kfree(tk);
250 }
251}
252
253/*
254 * Allocate new trace_probe and initialize it (including kprobes).
255 */
256static struct trace_kprobe *alloc_trace_kprobe(const char *group,
257 const char *event,
258 void *addr,
259 const char *symbol,
260 unsigned long offs,
261 int maxactive,
262 int nargs, bool is_return)
263{
264 struct trace_kprobe *tk;
265 int ret = -ENOMEM;
266
267 tk = kzalloc(struct_size(tk, tp.args, nargs), GFP_KERNEL);
268 if (!tk)
269 return ERR_PTR(ret);
270
271 tk->nhit = alloc_percpu(unsigned long);
272 if (!tk->nhit)
273 goto error;
274
275 if (symbol) {
276 tk->symbol = kstrdup(symbol, GFP_KERNEL);
277 if (!tk->symbol)
278 goto error;
279 tk->rp.kp.symbol_name = tk->symbol;
280 tk->rp.kp.offset = offs;
281 } else
282 tk->rp.kp.addr = addr;
283
284 if (is_return)
285 tk->rp.handler = kretprobe_dispatcher;
286 else
287 tk->rp.kp.pre_handler = kprobe_dispatcher;
288
289 tk->rp.maxactive = maxactive;
290 INIT_HLIST_NODE(&tk->rp.kp.hlist);
291 INIT_LIST_HEAD(&tk->rp.kp.list);
292
293 ret = trace_probe_init(&tk->tp, event, group, false, nargs);
294 if (ret < 0)
295 goto error;
296
297 dyn_event_init(&tk->devent, &trace_kprobe_ops);
298 return tk;
299error:
300 free_trace_kprobe(tk);
301 return ERR_PTR(ret);
302}
303
304static struct trace_kprobe *find_trace_kprobe(const char *event,
305 const char *group)
306{
307 struct dyn_event *pos;
308 struct trace_kprobe *tk;
309
310 for_each_trace_kprobe(tk, pos)
311 if (strcmp(trace_probe_name(&tk->tp), event) == 0 &&
312 strcmp(trace_probe_group_name(&tk->tp), group) == 0)
313 return tk;
314 return NULL;
315}
316
317static inline int __enable_trace_kprobe(struct trace_kprobe *tk)
318{
319 int ret = 0;
320
321 if (trace_kprobe_is_registered(tk) && !trace_kprobe_has_gone(tk)) {
322 if (trace_kprobe_is_return(tk))
323 ret = enable_kretprobe(&tk->rp);
324 else
325 ret = enable_kprobe(&tk->rp.kp);
326 }
327
328 return ret;
329}
330
331static void __disable_trace_kprobe(struct trace_probe *tp)
332{
333 struct trace_kprobe *tk;
334
335 list_for_each_entry(tk, trace_probe_probe_list(tp), tp.list) {
336 if (!trace_kprobe_is_registered(tk))
337 continue;
338 if (trace_kprobe_is_return(tk))
339 disable_kretprobe(&tk->rp);
340 else
341 disable_kprobe(&tk->rp.kp);
342 }
343}
344
345/*
346 * Enable trace_probe
347 * if the file is NULL, enable "perf" handler, or enable "trace" handler.
348 */
349static int enable_trace_kprobe(struct trace_event_call *call,
350 struct trace_event_file *file)
351{
352 struct trace_probe *tp;
353 struct trace_kprobe *tk;
354 bool enabled;
355 int ret = 0;
356
357 tp = trace_probe_primary_from_call(call);
358 if (WARN_ON_ONCE(!tp))
359 return -ENODEV;
360 enabled = trace_probe_is_enabled(tp);
361
362 /* This also changes "enabled" state */
363 if (file) {
364 ret = trace_probe_add_file(tp, file);
365 if (ret)
366 return ret;
367 } else
368 trace_probe_set_flag(tp, TP_FLAG_PROFILE);
369
370 if (enabled)
371 return 0;
372
373 list_for_each_entry(tk, trace_probe_probe_list(tp), tp.list) {
374 if (trace_kprobe_has_gone(tk))
375 continue;
376 ret = __enable_trace_kprobe(tk);
377 if (ret)
378 break;
379 enabled = true;
380 }
381
382 if (ret) {
383 /* Failed to enable one of them. Roll back all */
384 if (enabled)
385 __disable_trace_kprobe(tp);
386 if (file)
387 trace_probe_remove_file(tp, file);
388 else
389 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
390 }
391
392 return ret;
393}
394
395/*
396 * Disable trace_probe
397 * if the file is NULL, disable "perf" handler, or disable "trace" handler.
398 */
399static int disable_trace_kprobe(struct trace_event_call *call,
400 struct trace_event_file *file)
401{
402 struct trace_probe *tp;
403
404 tp = trace_probe_primary_from_call(call);
405 if (WARN_ON_ONCE(!tp))
406 return -ENODEV;
407
408 if (file) {
409 if (!trace_probe_get_file_link(tp, file))
410 return -ENOENT;
411 if (!trace_probe_has_single_file(tp))
412 goto out;
413 trace_probe_clear_flag(tp, TP_FLAG_TRACE);
414 } else
415 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
416
417 if (!trace_probe_is_enabled(tp))
418 __disable_trace_kprobe(tp);
419
420 out:
421 if (file)
422 /*
423 * Synchronization is done in below function. For perf event,
424 * file == NULL and perf_trace_event_unreg() calls
425 * tracepoint_synchronize_unregister() to ensure synchronize
426 * event. We don't need to care about it.
427 */
428 trace_probe_remove_file(tp, file);
429
430 return 0;
431}
432
433#if defined(CONFIG_DYNAMIC_FTRACE) && \
434 !defined(CONFIG_KPROBE_EVENTS_ON_NOTRACE)
435static bool __within_notrace_func(unsigned long addr)
436{
437 unsigned long offset, size;
438
439 if (!addr || !kallsyms_lookup_size_offset(addr, &size, &offset))
440 return false;
441
442 /* Get the entry address of the target function */
443 addr -= offset;
444
445 /*
446 * Since ftrace_location_range() does inclusive range check, we need
447 * to subtract 1 byte from the end address.
448 */
449 return !ftrace_location_range(addr, addr + size - 1);
450}
451
452static bool within_notrace_func(struct trace_kprobe *tk)
453{
454 unsigned long addr = trace_kprobe_address(tk);
455 char symname[KSYM_NAME_LEN], *p;
456
457 if (!__within_notrace_func(addr))
458 return false;
459
460 /* Check if the address is on a suffixed-symbol */
461 if (!lookup_symbol_name(addr, symname)) {
462 p = strchr(symname, '.');
463 if (!p)
464 return true;
465 *p = '\0';
466 addr = (unsigned long)kprobe_lookup_name(symname, 0);
467 if (addr)
468 return __within_notrace_func(addr);
469 }
470
471 return true;
472}
473#else
474#define within_notrace_func(tk) (false)
475#endif
476
477/* Internal register function - just handle k*probes and flags */
478static int __register_trace_kprobe(struct trace_kprobe *tk)
479{
480 int i, ret;
481
482 ret = security_locked_down(LOCKDOWN_KPROBES);
483 if (ret)
484 return ret;
485
486 if (trace_kprobe_is_registered(tk))
487 return -EINVAL;
488
489 if (within_notrace_func(tk)) {
490 pr_warn("Could not probe notrace function %ps\n",
491 (void *)trace_kprobe_address(tk));
492 return -EINVAL;
493 }
494
495 for (i = 0; i < tk->tp.nr_args; i++) {
496 ret = traceprobe_update_arg(&tk->tp.args[i]);
497 if (ret)
498 return ret;
499 }
500
501 /* Set/clear disabled flag according to tp->flag */
502 if (trace_probe_is_enabled(&tk->tp))
503 tk->rp.kp.flags &= ~KPROBE_FLAG_DISABLED;
504 else
505 tk->rp.kp.flags |= KPROBE_FLAG_DISABLED;
506
507 if (trace_kprobe_is_return(tk))
508 ret = register_kretprobe(&tk->rp);
509 else
510 ret = register_kprobe(&tk->rp.kp);
511
512 return ret;
513}
514
515/* Internal unregister function - just handle k*probes and flags */
516static void __unregister_trace_kprobe(struct trace_kprobe *tk)
517{
518 if (trace_kprobe_is_registered(tk)) {
519 if (trace_kprobe_is_return(tk))
520 unregister_kretprobe(&tk->rp);
521 else
522 unregister_kprobe(&tk->rp.kp);
523 /* Cleanup kprobe for reuse and mark it unregistered */
524 INIT_HLIST_NODE(&tk->rp.kp.hlist);
525 INIT_LIST_HEAD(&tk->rp.kp.list);
526 if (tk->rp.kp.symbol_name)
527 tk->rp.kp.addr = NULL;
528 }
529}
530
531/* Unregister a trace_probe and probe_event */
532static int unregister_trace_kprobe(struct trace_kprobe *tk)
533{
534 /* If other probes are on the event, just unregister kprobe */
535 if (trace_probe_has_sibling(&tk->tp))
536 goto unreg;
537
538 /* Enabled event can not be unregistered */
539 if (trace_probe_is_enabled(&tk->tp))
540 return -EBUSY;
541
542 /* If there's a reference to the dynamic event */
543 if (trace_event_dyn_busy(trace_probe_event_call(&tk->tp)))
544 return -EBUSY;
545
546 /* Will fail if probe is being used by ftrace or perf */
547 if (unregister_kprobe_event(tk))
548 return -EBUSY;
549
550unreg:
551 __unregister_trace_kprobe(tk);
552 dyn_event_remove(&tk->devent);
553 trace_probe_unlink(&tk->tp);
554
555 return 0;
556}
557
558static bool trace_kprobe_has_same_kprobe(struct trace_kprobe *orig,
559 struct trace_kprobe *comp)
560{
561 struct trace_probe_event *tpe = orig->tp.event;
562 int i;
563
564 list_for_each_entry(orig, &tpe->probes, tp.list) {
565 if (strcmp(trace_kprobe_symbol(orig),
566 trace_kprobe_symbol(comp)) ||
567 trace_kprobe_offset(orig) != trace_kprobe_offset(comp))
568 continue;
569
570 /*
571 * trace_probe_compare_arg_type() ensured that nr_args and
572 * each argument name and type are same. Let's compare comm.
573 */
574 for (i = 0; i < orig->tp.nr_args; i++) {
575 if (strcmp(orig->tp.args[i].comm,
576 comp->tp.args[i].comm))
577 break;
578 }
579
580 if (i == orig->tp.nr_args)
581 return true;
582 }
583
584 return false;
585}
586
587static int append_trace_kprobe(struct trace_kprobe *tk, struct trace_kprobe *to)
588{
589 int ret;
590
591 ret = trace_probe_compare_arg_type(&tk->tp, &to->tp);
592 if (ret) {
593 /* Note that argument starts index = 2 */
594 trace_probe_log_set_index(ret + 1);
595 trace_probe_log_err(0, DIFF_ARG_TYPE);
596 return -EEXIST;
597 }
598 if (trace_kprobe_has_same_kprobe(to, tk)) {
599 trace_probe_log_set_index(0);
600 trace_probe_log_err(0, SAME_PROBE);
601 return -EEXIST;
602 }
603
604 /* Append to existing event */
605 ret = trace_probe_append(&tk->tp, &to->tp);
606 if (ret)
607 return ret;
608
609 /* Register k*probe */
610 ret = __register_trace_kprobe(tk);
611 if (ret == -ENOENT && !trace_kprobe_module_exist(tk)) {
612 pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
613 ret = 0;
614 }
615
616 if (ret)
617 trace_probe_unlink(&tk->tp);
618 else
619 dyn_event_add(&tk->devent, trace_probe_event_call(&tk->tp));
620
621 return ret;
622}
623
624/* Register a trace_probe and probe_event */
625static int register_trace_kprobe(struct trace_kprobe *tk)
626{
627 struct trace_kprobe *old_tk;
628 int ret;
629
630 mutex_lock(&event_mutex);
631
632 old_tk = find_trace_kprobe(trace_probe_name(&tk->tp),
633 trace_probe_group_name(&tk->tp));
634 if (old_tk) {
635 if (trace_kprobe_is_return(tk) != trace_kprobe_is_return(old_tk)) {
636 trace_probe_log_set_index(0);
637 trace_probe_log_err(0, DIFF_PROBE_TYPE);
638 ret = -EEXIST;
639 } else {
640 ret = append_trace_kprobe(tk, old_tk);
641 }
642 goto end;
643 }
644
645 /* Register new event */
646 ret = register_kprobe_event(tk);
647 if (ret) {
648 if (ret == -EEXIST) {
649 trace_probe_log_set_index(0);
650 trace_probe_log_err(0, EVENT_EXIST);
651 } else
652 pr_warn("Failed to register probe event(%d)\n", ret);
653 goto end;
654 }
655
656 /* Register k*probe */
657 ret = __register_trace_kprobe(tk);
658 if (ret == -ENOENT && !trace_kprobe_module_exist(tk)) {
659 pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
660 ret = 0;
661 }
662
663 if (ret < 0)
664 unregister_kprobe_event(tk);
665 else
666 dyn_event_add(&tk->devent, trace_probe_event_call(&tk->tp));
667
668end:
669 mutex_unlock(&event_mutex);
670 return ret;
671}
672
673/* Module notifier call back, checking event on the module */
674static int trace_kprobe_module_callback(struct notifier_block *nb,
675 unsigned long val, void *data)
676{
677 struct module *mod = data;
678 struct dyn_event *pos;
679 struct trace_kprobe *tk;
680 int ret;
681
682 if (val != MODULE_STATE_COMING)
683 return NOTIFY_DONE;
684
685 /* Update probes on coming module */
686 mutex_lock(&event_mutex);
687 for_each_trace_kprobe(tk, pos) {
688 if (trace_kprobe_within_module(tk, mod)) {
689 /* Don't need to check busy - this should have gone. */
690 __unregister_trace_kprobe(tk);
691 ret = __register_trace_kprobe(tk);
692 if (ret)
693 pr_warn("Failed to re-register probe %s on %s: %d\n",
694 trace_probe_name(&tk->tp),
695 module_name(mod), ret);
696 }
697 }
698 mutex_unlock(&event_mutex);
699
700 return NOTIFY_DONE;
701}
702
703static struct notifier_block trace_kprobe_module_nb = {
704 .notifier_call = trace_kprobe_module_callback,
705 .priority = 1 /* Invoked after kprobe module callback */
706};
707
708static int count_symbols(void *data, unsigned long unused)
709{
710 unsigned int *count = data;
711
712 (*count)++;
713
714 return 0;
715}
716
717struct sym_count_ctx {
718 unsigned int count;
719 const char *name;
720};
721
722static int count_mod_symbols(void *data, const char *name, unsigned long unused)
723{
724 struct sym_count_ctx *ctx = data;
725
726 if (strcmp(name, ctx->name) == 0)
727 ctx->count++;
728
729 return 0;
730}
731
732static unsigned int number_of_same_symbols(char *func_name)
733{
734 struct sym_count_ctx ctx = { .count = 0, .name = func_name };
735
736 kallsyms_on_each_match_symbol(count_symbols, func_name, &ctx.count);
737
738 module_kallsyms_on_each_symbol(NULL, count_mod_symbols, &ctx);
739
740 return ctx.count;
741}
742
743static int trace_kprobe_entry_handler(struct kretprobe_instance *ri,
744 struct pt_regs *regs);
745
746static int __trace_kprobe_create(int argc, const char *argv[])
747{
748 /*
749 * Argument syntax:
750 * - Add kprobe:
751 * p[:[GRP/][EVENT]] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
752 * - Add kretprobe:
753 * r[MAXACTIVE][:[GRP/][EVENT]] [MOD:]KSYM[+0] [FETCHARGS]
754 * Or
755 * p[:[GRP/][EVENT]] [MOD:]KSYM[+0]%return [FETCHARGS]
756 *
757 * Fetch args:
758 * $retval : fetch return value
759 * $stack : fetch stack address
760 * $stackN : fetch Nth of stack (N:0-)
761 * $comm : fetch current task comm
762 * @ADDR : fetch memory at ADDR (ADDR should be in kernel)
763 * @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
764 * %REG : fetch register REG
765 * Dereferencing memory fetch:
766 * +|-offs(ARG) : fetch memory at ARG +|- offs address.
767 * Alias name of args:
768 * NAME=FETCHARG : set NAME as alias of FETCHARG.
769 * Type of args:
770 * FETCHARG:TYPE : use TYPE instead of unsigned long.
771 */
772 struct trace_kprobe *tk = NULL;
773 int i, len, new_argc = 0, ret = 0;
774 bool is_return = false;
775 char *symbol = NULL, *tmp = NULL;
776 const char **new_argv = NULL;
777 const char *event = NULL, *group = KPROBE_EVENT_SYSTEM;
778 enum probe_print_type ptype;
779 int maxactive = 0;
780 long offset = 0;
781 void *addr = NULL;
782 char buf[MAX_EVENT_NAME_LEN];
783 char gbuf[MAX_EVENT_NAME_LEN];
784 char abuf[MAX_BTF_ARGS_LEN];
785 struct traceprobe_parse_context ctx = { .flags = TPARG_FL_KERNEL };
786
787 switch (argv[0][0]) {
788 case 'r':
789 is_return = true;
790 break;
791 case 'p':
792 break;
793 default:
794 return -ECANCELED;
795 }
796 if (argc < 2)
797 return -ECANCELED;
798
799 trace_probe_log_init("trace_kprobe", argc, argv);
800
801 event = strchr(&argv[0][1], ':');
802 if (event)
803 event++;
804
805 if (isdigit(argv[0][1])) {
806 if (!is_return) {
807 trace_probe_log_err(1, BAD_MAXACT_TYPE);
808 goto parse_error;
809 }
810 if (event)
811 len = event - &argv[0][1] - 1;
812 else
813 len = strlen(&argv[0][1]);
814 if (len > MAX_EVENT_NAME_LEN - 1) {
815 trace_probe_log_err(1, BAD_MAXACT);
816 goto parse_error;
817 }
818 memcpy(buf, &argv[0][1], len);
819 buf[len] = '\0';
820 ret = kstrtouint(buf, 0, &maxactive);
821 if (ret || !maxactive) {
822 trace_probe_log_err(1, BAD_MAXACT);
823 goto parse_error;
824 }
825 /* kretprobes instances are iterated over via a list. The
826 * maximum should stay reasonable.
827 */
828 if (maxactive > KRETPROBE_MAXACTIVE_MAX) {
829 trace_probe_log_err(1, MAXACT_TOO_BIG);
830 goto parse_error;
831 }
832 }
833
834 /* try to parse an address. if that fails, try to read the
835 * input as a symbol. */
836 if (kstrtoul(argv[1], 0, (unsigned long *)&addr)) {
837 trace_probe_log_set_index(1);
838 /* Check whether uprobe event specified */
839 if (strchr(argv[1], '/') && strchr(argv[1], ':')) {
840 ret = -ECANCELED;
841 goto error;
842 }
843 /* a symbol specified */
844 symbol = kstrdup(argv[1], GFP_KERNEL);
845 if (!symbol)
846 return -ENOMEM;
847
848 tmp = strchr(symbol, '%');
849 if (tmp) {
850 if (!strcmp(tmp, "%return")) {
851 *tmp = '\0';
852 is_return = true;
853 } else {
854 trace_probe_log_err(tmp - symbol, BAD_ADDR_SUFFIX);
855 goto parse_error;
856 }
857 }
858
859 /* TODO: support .init module functions */
860 ret = traceprobe_split_symbol_offset(symbol, &offset);
861 if (ret || offset < 0 || offset > UINT_MAX) {
862 trace_probe_log_err(0, BAD_PROBE_ADDR);
863 goto parse_error;
864 }
865 if (is_return)
866 ctx.flags |= TPARG_FL_RETURN;
867 ret = kprobe_on_func_entry(NULL, symbol, offset);
868 if (ret == 0 && !is_return)
869 ctx.flags |= TPARG_FL_FENTRY;
870 /* Defer the ENOENT case until register kprobe */
871 if (ret == -EINVAL && is_return) {
872 trace_probe_log_err(0, BAD_RETPROBE);
873 goto parse_error;
874 }
875 }
876
877 if (symbol && !strchr(symbol, ':')) {
878 unsigned int count;
879
880 count = number_of_same_symbols(symbol);
881 if (count > 1) {
882 /*
883 * Users should use ADDR to remove the ambiguity of
884 * using KSYM only.
885 */
886 trace_probe_log_err(0, NON_UNIQ_SYMBOL);
887 ret = -EADDRNOTAVAIL;
888
889 goto error;
890 } else if (count == 0) {
891 /*
892 * We can return ENOENT earlier than when register the
893 * kprobe.
894 */
895 trace_probe_log_err(0, BAD_PROBE_ADDR);
896 ret = -ENOENT;
897
898 goto error;
899 }
900 }
901
902 trace_probe_log_set_index(0);
903 if (event) {
904 ret = traceprobe_parse_event_name(&event, &group, gbuf,
905 event - argv[0]);
906 if (ret)
907 goto parse_error;
908 }
909
910 if (!event) {
911 /* Make a new event name */
912 if (symbol)
913 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
914 is_return ? 'r' : 'p', symbol, offset);
915 else
916 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
917 is_return ? 'r' : 'p', addr);
918 sanitize_event_name(buf);
919 event = buf;
920 }
921
922 argc -= 2; argv += 2;
923 ctx.funcname = symbol;
924 new_argv = traceprobe_expand_meta_args(argc, argv, &new_argc,
925 abuf, MAX_BTF_ARGS_LEN, &ctx);
926 if (IS_ERR(new_argv)) {
927 ret = PTR_ERR(new_argv);
928 new_argv = NULL;
929 goto out;
930 }
931 if (new_argv) {
932 argc = new_argc;
933 argv = new_argv;
934 }
935
936 /* setup a probe */
937 tk = alloc_trace_kprobe(group, event, addr, symbol, offset, maxactive,
938 argc, is_return);
939 if (IS_ERR(tk)) {
940 ret = PTR_ERR(tk);
941 /* This must return -ENOMEM, else there is a bug */
942 WARN_ON_ONCE(ret != -ENOMEM);
943 goto out; /* We know tk is not allocated */
944 }
945
946 /* parse arguments */
947 for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
948 trace_probe_log_set_index(i + 2);
949 ctx.offset = 0;
950 ret = traceprobe_parse_probe_arg(&tk->tp, i, argv[i], &ctx);
951 if (ret)
952 goto error; /* This can be -ENOMEM */
953 }
954 /* entry handler for kretprobe */
955 if (is_return && tk->tp.entry_arg) {
956 tk->rp.entry_handler = trace_kprobe_entry_handler;
957 tk->rp.data_size = traceprobe_get_entry_data_size(&tk->tp);
958 }
959
960 ptype = is_return ? PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL;
961 ret = traceprobe_set_print_fmt(&tk->tp, ptype);
962 if (ret < 0)
963 goto error;
964
965 ret = register_trace_kprobe(tk);
966 if (ret) {
967 trace_probe_log_set_index(1);
968 if (ret == -EILSEQ)
969 trace_probe_log_err(0, BAD_INSN_BNDRY);
970 else if (ret == -ENOENT)
971 trace_probe_log_err(0, BAD_PROBE_ADDR);
972 else if (ret != -ENOMEM && ret != -EEXIST)
973 trace_probe_log_err(0, FAIL_REG_PROBE);
974 goto error;
975 }
976
977out:
978 traceprobe_finish_parse(&ctx);
979 trace_probe_log_clear();
980 kfree(new_argv);
981 kfree(symbol);
982 return ret;
983
984parse_error:
985 ret = -EINVAL;
986error:
987 free_trace_kprobe(tk);
988 goto out;
989}
990
991static int trace_kprobe_create(const char *raw_command)
992{
993 return trace_probe_create(raw_command, __trace_kprobe_create);
994}
995
996static int create_or_delete_trace_kprobe(const char *raw_command)
997{
998 int ret;
999
1000 if (raw_command[0] == '-')
1001 return dyn_event_release(raw_command, &trace_kprobe_ops);
1002
1003 ret = trace_kprobe_create(raw_command);
1004 return ret == -ECANCELED ? -EINVAL : ret;
1005}
1006
1007static int trace_kprobe_run_command(struct dynevent_cmd *cmd)
1008{
1009 return create_or_delete_trace_kprobe(cmd->seq.buffer);
1010}
1011
1012/**
1013 * kprobe_event_cmd_init - Initialize a kprobe event command object
1014 * @cmd: A pointer to the dynevent_cmd struct representing the new event
1015 * @buf: A pointer to the buffer used to build the command
1016 * @maxlen: The length of the buffer passed in @buf
1017 *
1018 * Initialize a synthetic event command object. Use this before
1019 * calling any of the other kprobe_event functions.
1020 */
1021void kprobe_event_cmd_init(struct dynevent_cmd *cmd, char *buf, int maxlen)
1022{
1023 dynevent_cmd_init(cmd, buf, maxlen, DYNEVENT_TYPE_KPROBE,
1024 trace_kprobe_run_command);
1025}
1026EXPORT_SYMBOL_GPL(kprobe_event_cmd_init);
1027
1028/**
1029 * __kprobe_event_gen_cmd_start - Generate a kprobe event command from arg list
1030 * @cmd: A pointer to the dynevent_cmd struct representing the new event
1031 * @kretprobe: Is this a return probe?
1032 * @name: The name of the kprobe event
1033 * @loc: The location of the kprobe event
1034 * @...: Variable number of arg (pairs), one pair for each field
1035 *
1036 * NOTE: Users normally won't want to call this function directly, but
1037 * rather use the kprobe_event_gen_cmd_start() wrapper, which automatically
1038 * adds a NULL to the end of the arg list. If this function is used
1039 * directly, make sure the last arg in the variable arg list is NULL.
1040 *
1041 * Generate a kprobe event command to be executed by
1042 * kprobe_event_gen_cmd_end(). This function can be used to generate the
1043 * complete command or only the first part of it; in the latter case,
1044 * kprobe_event_add_fields() can be used to add more fields following this.
1045 *
1046 * Unlikely the synth_event_gen_cmd_start(), @loc must be specified. This
1047 * returns -EINVAL if @loc == NULL.
1048 *
1049 * Return: 0 if successful, error otherwise.
1050 */
1051int __kprobe_event_gen_cmd_start(struct dynevent_cmd *cmd, bool kretprobe,
1052 const char *name, const char *loc, ...)
1053{
1054 char buf[MAX_EVENT_NAME_LEN];
1055 struct dynevent_arg arg;
1056 va_list args;
1057 int ret;
1058
1059 if (cmd->type != DYNEVENT_TYPE_KPROBE)
1060 return -EINVAL;
1061
1062 if (!loc)
1063 return -EINVAL;
1064
1065 if (kretprobe)
1066 snprintf(buf, MAX_EVENT_NAME_LEN, "r:kprobes/%s", name);
1067 else
1068 snprintf(buf, MAX_EVENT_NAME_LEN, "p:kprobes/%s", name);
1069
1070 ret = dynevent_str_add(cmd, buf);
1071 if (ret)
1072 return ret;
1073
1074 dynevent_arg_init(&arg, 0);
1075 arg.str = loc;
1076 ret = dynevent_arg_add(cmd, &arg, NULL);
1077 if (ret)
1078 return ret;
1079
1080 va_start(args, loc);
1081 for (;;) {
1082 const char *field;
1083
1084 field = va_arg(args, const char *);
1085 if (!field)
1086 break;
1087
1088 if (++cmd->n_fields > MAX_TRACE_ARGS) {
1089 ret = -EINVAL;
1090 break;
1091 }
1092
1093 arg.str = field;
1094 ret = dynevent_arg_add(cmd, &arg, NULL);
1095 if (ret)
1096 break;
1097 }
1098 va_end(args);
1099
1100 return ret;
1101}
1102EXPORT_SYMBOL_GPL(__kprobe_event_gen_cmd_start);
1103
1104/**
1105 * __kprobe_event_add_fields - Add probe fields to a kprobe command from arg list
1106 * @cmd: A pointer to the dynevent_cmd struct representing the new event
1107 * @...: Variable number of arg (pairs), one pair for each field
1108 *
1109 * NOTE: Users normally won't want to call this function directly, but
1110 * rather use the kprobe_event_add_fields() wrapper, which
1111 * automatically adds a NULL to the end of the arg list. If this
1112 * function is used directly, make sure the last arg in the variable
1113 * arg list is NULL.
1114 *
1115 * Add probe fields to an existing kprobe command using a variable
1116 * list of args. Fields are added in the same order they're listed.
1117 *
1118 * Return: 0 if successful, error otherwise.
1119 */
1120int __kprobe_event_add_fields(struct dynevent_cmd *cmd, ...)
1121{
1122 struct dynevent_arg arg;
1123 va_list args;
1124 int ret = 0;
1125
1126 if (cmd->type != DYNEVENT_TYPE_KPROBE)
1127 return -EINVAL;
1128
1129 dynevent_arg_init(&arg, 0);
1130
1131 va_start(args, cmd);
1132 for (;;) {
1133 const char *field;
1134
1135 field = va_arg(args, const char *);
1136 if (!field)
1137 break;
1138
1139 if (++cmd->n_fields > MAX_TRACE_ARGS) {
1140 ret = -EINVAL;
1141 break;
1142 }
1143
1144 arg.str = field;
1145 ret = dynevent_arg_add(cmd, &arg, NULL);
1146 if (ret)
1147 break;
1148 }
1149 va_end(args);
1150
1151 return ret;
1152}
1153EXPORT_SYMBOL_GPL(__kprobe_event_add_fields);
1154
1155/**
1156 * kprobe_event_delete - Delete a kprobe event
1157 * @name: The name of the kprobe event to delete
1158 *
1159 * Delete a kprobe event with the give @name from kernel code rather
1160 * than directly from the command line.
1161 *
1162 * Return: 0 if successful, error otherwise.
1163 */
1164int kprobe_event_delete(const char *name)
1165{
1166 char buf[MAX_EVENT_NAME_LEN];
1167
1168 snprintf(buf, MAX_EVENT_NAME_LEN, "-:%s", name);
1169
1170 return create_or_delete_trace_kprobe(buf);
1171}
1172EXPORT_SYMBOL_GPL(kprobe_event_delete);
1173
1174static int trace_kprobe_release(struct dyn_event *ev)
1175{
1176 struct trace_kprobe *tk = to_trace_kprobe(ev);
1177 int ret = unregister_trace_kprobe(tk);
1178
1179 if (!ret)
1180 free_trace_kprobe(tk);
1181 return ret;
1182}
1183
1184static int trace_kprobe_show(struct seq_file *m, struct dyn_event *ev)
1185{
1186 struct trace_kprobe *tk = to_trace_kprobe(ev);
1187 int i;
1188
1189 seq_putc(m, trace_kprobe_is_return(tk) ? 'r' : 'p');
1190 if (trace_kprobe_is_return(tk) && tk->rp.maxactive)
1191 seq_printf(m, "%d", tk->rp.maxactive);
1192 seq_printf(m, ":%s/%s", trace_probe_group_name(&tk->tp),
1193 trace_probe_name(&tk->tp));
1194
1195 if (!tk->symbol)
1196 seq_printf(m, " 0x%p", tk->rp.kp.addr);
1197 else if (tk->rp.kp.offset)
1198 seq_printf(m, " %s+%u", trace_kprobe_symbol(tk),
1199 tk->rp.kp.offset);
1200 else
1201 seq_printf(m, " %s", trace_kprobe_symbol(tk));
1202
1203 for (i = 0; i < tk->tp.nr_args; i++)
1204 seq_printf(m, " %s=%s", tk->tp.args[i].name, tk->tp.args[i].comm);
1205 seq_putc(m, '\n');
1206
1207 return 0;
1208}
1209
1210static int probes_seq_show(struct seq_file *m, void *v)
1211{
1212 struct dyn_event *ev = v;
1213
1214 if (!is_trace_kprobe(ev))
1215 return 0;
1216
1217 return trace_kprobe_show(m, ev);
1218}
1219
1220static const struct seq_operations probes_seq_op = {
1221 .start = dyn_event_seq_start,
1222 .next = dyn_event_seq_next,
1223 .stop = dyn_event_seq_stop,
1224 .show = probes_seq_show
1225};
1226
1227static int probes_open(struct inode *inode, struct file *file)
1228{
1229 int ret;
1230
1231 ret = security_locked_down(LOCKDOWN_TRACEFS);
1232 if (ret)
1233 return ret;
1234
1235 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
1236 ret = dyn_events_release_all(&trace_kprobe_ops);
1237 if (ret < 0)
1238 return ret;
1239 }
1240
1241 return seq_open(file, &probes_seq_op);
1242}
1243
1244static ssize_t probes_write(struct file *file, const char __user *buffer,
1245 size_t count, loff_t *ppos)
1246{
1247 return trace_parse_run_command(file, buffer, count, ppos,
1248 create_or_delete_trace_kprobe);
1249}
1250
1251static const struct file_operations kprobe_events_ops = {
1252 .owner = THIS_MODULE,
1253 .open = probes_open,
1254 .read = seq_read,
1255 .llseek = seq_lseek,
1256 .release = seq_release,
1257 .write = probes_write,
1258};
1259
1260static unsigned long trace_kprobe_missed(struct trace_kprobe *tk)
1261{
1262 return trace_kprobe_is_return(tk) ?
1263 tk->rp.kp.nmissed + tk->rp.nmissed : tk->rp.kp.nmissed;
1264}
1265
1266/* Probes profiling interfaces */
1267static int probes_profile_seq_show(struct seq_file *m, void *v)
1268{
1269 struct dyn_event *ev = v;
1270 struct trace_kprobe *tk;
1271 unsigned long nmissed;
1272
1273 if (!is_trace_kprobe(ev))
1274 return 0;
1275
1276 tk = to_trace_kprobe(ev);
1277 nmissed = trace_kprobe_missed(tk);
1278 seq_printf(m, " %-44s %15lu %15lu\n",
1279 trace_probe_name(&tk->tp),
1280 trace_kprobe_nhit(tk),
1281 nmissed);
1282
1283 return 0;
1284}
1285
1286static const struct seq_operations profile_seq_op = {
1287 .start = dyn_event_seq_start,
1288 .next = dyn_event_seq_next,
1289 .stop = dyn_event_seq_stop,
1290 .show = probes_profile_seq_show
1291};
1292
1293static int profile_open(struct inode *inode, struct file *file)
1294{
1295 int ret;
1296
1297 ret = security_locked_down(LOCKDOWN_TRACEFS);
1298 if (ret)
1299 return ret;
1300
1301 return seq_open(file, &profile_seq_op);
1302}
1303
1304static const struct file_operations kprobe_profile_ops = {
1305 .owner = THIS_MODULE,
1306 .open = profile_open,
1307 .read = seq_read,
1308 .llseek = seq_lseek,
1309 .release = seq_release,
1310};
1311
1312/* Note that we don't verify it, since the code does not come from user space */
1313static int
1314process_fetch_insn(struct fetch_insn *code, void *rec, void *edata,
1315 void *dest, void *base)
1316{
1317 struct pt_regs *regs = rec;
1318 unsigned long val;
1319 int ret;
1320
1321retry:
1322 /* 1st stage: get value from context */
1323 switch (code->op) {
1324 case FETCH_OP_REG:
1325 val = regs_get_register(regs, code->param);
1326 break;
1327 case FETCH_OP_STACK:
1328 val = regs_get_kernel_stack_nth(regs, code->param);
1329 break;
1330 case FETCH_OP_STACKP:
1331 val = kernel_stack_pointer(regs);
1332 break;
1333 case FETCH_OP_RETVAL:
1334 val = regs_return_value(regs);
1335 break;
1336#ifdef CONFIG_HAVE_FUNCTION_ARG_ACCESS_API
1337 case FETCH_OP_ARG:
1338 val = regs_get_kernel_argument(regs, code->param);
1339 break;
1340 case FETCH_OP_EDATA:
1341 val = *(unsigned long *)((unsigned long)edata + code->offset);
1342 break;
1343#endif
1344 case FETCH_NOP_SYMBOL: /* Ignore a place holder */
1345 code++;
1346 goto retry;
1347 default:
1348 ret = process_common_fetch_insn(code, &val);
1349 if (ret < 0)
1350 return ret;
1351 }
1352 code++;
1353
1354 return process_fetch_insn_bottom(code, val, dest, base);
1355}
1356NOKPROBE_SYMBOL(process_fetch_insn)
1357
1358/* Kprobe handler */
1359static nokprobe_inline void
1360__kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs,
1361 struct trace_event_file *trace_file)
1362{
1363 struct kprobe_trace_entry_head *entry;
1364 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1365 struct trace_event_buffer fbuffer;
1366 int dsize;
1367
1368 WARN_ON(call != trace_file->event_call);
1369
1370 if (trace_trigger_soft_disabled(trace_file))
1371 return;
1372
1373 dsize = __get_data_size(&tk->tp, regs, NULL);
1374
1375 entry = trace_event_buffer_reserve(&fbuffer, trace_file,
1376 sizeof(*entry) + tk->tp.size + dsize);
1377 if (!entry)
1378 return;
1379
1380 fbuffer.regs = regs;
1381 entry->ip = (unsigned long)tk->rp.kp.addr;
1382 store_trace_args(&entry[1], &tk->tp, regs, NULL, sizeof(*entry), dsize);
1383
1384 trace_event_buffer_commit(&fbuffer);
1385}
1386
1387static void
1388kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs)
1389{
1390 struct event_file_link *link;
1391
1392 trace_probe_for_each_link_rcu(link, &tk->tp)
1393 __kprobe_trace_func(tk, regs, link->file);
1394}
1395NOKPROBE_SYMBOL(kprobe_trace_func);
1396
1397/* Kretprobe handler */
1398
1399static int trace_kprobe_entry_handler(struct kretprobe_instance *ri,
1400 struct pt_regs *regs)
1401{
1402 struct kretprobe *rp = get_kretprobe(ri);
1403 struct trace_kprobe *tk;
1404
1405 /*
1406 * There is a small chance that get_kretprobe(ri) returns NULL when
1407 * the kretprobe is unregister on another CPU between kretprobe's
1408 * trampoline_handler and this function.
1409 */
1410 if (unlikely(!rp))
1411 return -ENOENT;
1412
1413 tk = container_of(rp, struct trace_kprobe, rp);
1414
1415 /* store argument values into ri->data as entry data */
1416 if (tk->tp.entry_arg)
1417 store_trace_entry_data(ri->data, &tk->tp, regs);
1418
1419 return 0;
1420}
1421
1422
1423static nokprobe_inline void
1424__kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1425 struct pt_regs *regs,
1426 struct trace_event_file *trace_file)
1427{
1428 struct kretprobe_trace_entry_head *entry;
1429 struct trace_event_buffer fbuffer;
1430 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1431 int dsize;
1432
1433 WARN_ON(call != trace_file->event_call);
1434
1435 if (trace_trigger_soft_disabled(trace_file))
1436 return;
1437
1438 dsize = __get_data_size(&tk->tp, regs, ri->data);
1439
1440 entry = trace_event_buffer_reserve(&fbuffer, trace_file,
1441 sizeof(*entry) + tk->tp.size + dsize);
1442 if (!entry)
1443 return;
1444
1445 fbuffer.regs = regs;
1446 entry->func = (unsigned long)tk->rp.kp.addr;
1447 entry->ret_ip = get_kretprobe_retaddr(ri);
1448 store_trace_args(&entry[1], &tk->tp, regs, ri->data, sizeof(*entry), dsize);
1449
1450 trace_event_buffer_commit(&fbuffer);
1451}
1452
1453static void
1454kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1455 struct pt_regs *regs)
1456{
1457 struct event_file_link *link;
1458
1459 trace_probe_for_each_link_rcu(link, &tk->tp)
1460 __kretprobe_trace_func(tk, ri, regs, link->file);
1461}
1462NOKPROBE_SYMBOL(kretprobe_trace_func);
1463
1464/* Event entry printers */
1465static enum print_line_t
1466print_kprobe_event(struct trace_iterator *iter, int flags,
1467 struct trace_event *event)
1468{
1469 struct kprobe_trace_entry_head *field;
1470 struct trace_seq *s = &iter->seq;
1471 struct trace_probe *tp;
1472
1473 field = (struct kprobe_trace_entry_head *)iter->ent;
1474 tp = trace_probe_primary_from_call(
1475 container_of(event, struct trace_event_call, event));
1476 if (WARN_ON_ONCE(!tp))
1477 goto out;
1478
1479 trace_seq_printf(s, "%s: (", trace_probe_name(tp));
1480
1481 if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
1482 goto out;
1483
1484 trace_seq_putc(s, ')');
1485
1486 if (trace_probe_print_args(s, tp->args, tp->nr_args,
1487 (u8 *)&field[1], field) < 0)
1488 goto out;
1489
1490 trace_seq_putc(s, '\n');
1491 out:
1492 return trace_handle_return(s);
1493}
1494
1495static enum print_line_t
1496print_kretprobe_event(struct trace_iterator *iter, int flags,
1497 struct trace_event *event)
1498{
1499 struct kretprobe_trace_entry_head *field;
1500 struct trace_seq *s = &iter->seq;
1501 struct trace_probe *tp;
1502
1503 field = (struct kretprobe_trace_entry_head *)iter->ent;
1504 tp = trace_probe_primary_from_call(
1505 container_of(event, struct trace_event_call, event));
1506 if (WARN_ON_ONCE(!tp))
1507 goto out;
1508
1509 trace_seq_printf(s, "%s: (", trace_probe_name(tp));
1510
1511 if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1512 goto out;
1513
1514 trace_seq_puts(s, " <- ");
1515
1516 if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1517 goto out;
1518
1519 trace_seq_putc(s, ')');
1520
1521 if (trace_probe_print_args(s, tp->args, tp->nr_args,
1522 (u8 *)&field[1], field) < 0)
1523 goto out;
1524
1525 trace_seq_putc(s, '\n');
1526
1527 out:
1528 return trace_handle_return(s);
1529}
1530
1531
1532static int kprobe_event_define_fields(struct trace_event_call *event_call)
1533{
1534 int ret;
1535 struct kprobe_trace_entry_head field;
1536 struct trace_probe *tp;
1537
1538 tp = trace_probe_primary_from_call(event_call);
1539 if (WARN_ON_ONCE(!tp))
1540 return -ENOENT;
1541
1542 DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1543
1544 return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
1545}
1546
1547static int kretprobe_event_define_fields(struct trace_event_call *event_call)
1548{
1549 int ret;
1550 struct kretprobe_trace_entry_head field;
1551 struct trace_probe *tp;
1552
1553 tp = trace_probe_primary_from_call(event_call);
1554 if (WARN_ON_ONCE(!tp))
1555 return -ENOENT;
1556
1557 DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1558 DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1559
1560 return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
1561}
1562
1563#ifdef CONFIG_PERF_EVENTS
1564
1565/* Kprobe profile handler */
1566static int
1567kprobe_perf_func(struct trace_kprobe *tk, struct pt_regs *regs)
1568{
1569 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1570 struct kprobe_trace_entry_head *entry;
1571 struct hlist_head *head;
1572 int size, __size, dsize;
1573 int rctx;
1574
1575 if (bpf_prog_array_valid(call)) {
1576 unsigned long orig_ip = instruction_pointer(regs);
1577 int ret;
1578
1579 ret = trace_call_bpf(call, regs);
1580
1581 /*
1582 * We need to check and see if we modified the pc of the
1583 * pt_regs, and if so return 1 so that we don't do the
1584 * single stepping.
1585 */
1586 if (orig_ip != instruction_pointer(regs))
1587 return 1;
1588 if (!ret)
1589 return 0;
1590 }
1591
1592 head = this_cpu_ptr(call->perf_events);
1593 if (hlist_empty(head))
1594 return 0;
1595
1596 dsize = __get_data_size(&tk->tp, regs, NULL);
1597 __size = sizeof(*entry) + tk->tp.size + dsize;
1598 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1599 size -= sizeof(u32);
1600
1601 entry = perf_trace_buf_alloc(size, NULL, &rctx);
1602 if (!entry)
1603 return 0;
1604
1605 entry->ip = (unsigned long)tk->rp.kp.addr;
1606 memset(&entry[1], 0, dsize);
1607 store_trace_args(&entry[1], &tk->tp, regs, NULL, sizeof(*entry), dsize);
1608 perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1609 head, NULL);
1610 return 0;
1611}
1612NOKPROBE_SYMBOL(kprobe_perf_func);
1613
1614/* Kretprobe profile handler */
1615static void
1616kretprobe_perf_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1617 struct pt_regs *regs)
1618{
1619 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1620 struct kretprobe_trace_entry_head *entry;
1621 struct hlist_head *head;
1622 int size, __size, dsize;
1623 int rctx;
1624
1625 if (bpf_prog_array_valid(call) && !trace_call_bpf(call, regs))
1626 return;
1627
1628 head = this_cpu_ptr(call->perf_events);
1629 if (hlist_empty(head))
1630 return;
1631
1632 dsize = __get_data_size(&tk->tp, regs, ri->data);
1633 __size = sizeof(*entry) + tk->tp.size + dsize;
1634 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1635 size -= sizeof(u32);
1636
1637 entry = perf_trace_buf_alloc(size, NULL, &rctx);
1638 if (!entry)
1639 return;
1640
1641 entry->func = (unsigned long)tk->rp.kp.addr;
1642 entry->ret_ip = get_kretprobe_retaddr(ri);
1643 store_trace_args(&entry[1], &tk->tp, regs, ri->data, sizeof(*entry), dsize);
1644 perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1645 head, NULL);
1646}
1647NOKPROBE_SYMBOL(kretprobe_perf_func);
1648
1649int bpf_get_kprobe_info(const struct perf_event *event, u32 *fd_type,
1650 const char **symbol, u64 *probe_offset,
1651 u64 *probe_addr, unsigned long *missed,
1652 bool perf_type_tracepoint)
1653{
1654 const char *pevent = trace_event_name(event->tp_event);
1655 const char *group = event->tp_event->class->system;
1656 struct trace_kprobe *tk;
1657
1658 if (perf_type_tracepoint)
1659 tk = find_trace_kprobe(pevent, group);
1660 else
1661 tk = trace_kprobe_primary_from_call(event->tp_event);
1662 if (!tk)
1663 return -EINVAL;
1664
1665 *fd_type = trace_kprobe_is_return(tk) ? BPF_FD_TYPE_KRETPROBE
1666 : BPF_FD_TYPE_KPROBE;
1667 *probe_offset = tk->rp.kp.offset;
1668 *probe_addr = kallsyms_show_value(current_cred()) ?
1669 (unsigned long)tk->rp.kp.addr : 0;
1670 *symbol = tk->symbol;
1671 if (missed)
1672 *missed = trace_kprobe_missed(tk);
1673 return 0;
1674}
1675#endif /* CONFIG_PERF_EVENTS */
1676
1677/*
1678 * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex.
1679 *
1680 * kprobe_trace_self_tests_init() does enable_trace_probe/disable_trace_probe
1681 * lockless, but we can't race with this __init function.
1682 */
1683static int kprobe_register(struct trace_event_call *event,
1684 enum trace_reg type, void *data)
1685{
1686 struct trace_event_file *file = data;
1687
1688 switch (type) {
1689 case TRACE_REG_REGISTER:
1690 return enable_trace_kprobe(event, file);
1691 case TRACE_REG_UNREGISTER:
1692 return disable_trace_kprobe(event, file);
1693
1694#ifdef CONFIG_PERF_EVENTS
1695 case TRACE_REG_PERF_REGISTER:
1696 return enable_trace_kprobe(event, NULL);
1697 case TRACE_REG_PERF_UNREGISTER:
1698 return disable_trace_kprobe(event, NULL);
1699 case TRACE_REG_PERF_OPEN:
1700 case TRACE_REG_PERF_CLOSE:
1701 case TRACE_REG_PERF_ADD:
1702 case TRACE_REG_PERF_DEL:
1703 return 0;
1704#endif
1705 }
1706 return 0;
1707}
1708
1709static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1710{
1711 struct trace_kprobe *tk = container_of(kp, struct trace_kprobe, rp.kp);
1712 int ret = 0;
1713
1714 raw_cpu_inc(*tk->nhit);
1715
1716 if (trace_probe_test_flag(&tk->tp, TP_FLAG_TRACE))
1717 kprobe_trace_func(tk, regs);
1718#ifdef CONFIG_PERF_EVENTS
1719 if (trace_probe_test_flag(&tk->tp, TP_FLAG_PROFILE))
1720 ret = kprobe_perf_func(tk, regs);
1721#endif
1722 return ret;
1723}
1724NOKPROBE_SYMBOL(kprobe_dispatcher);
1725
1726static int
1727kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1728{
1729 struct kretprobe *rp = get_kretprobe(ri);
1730 struct trace_kprobe *tk;
1731
1732 /*
1733 * There is a small chance that get_kretprobe(ri) returns NULL when
1734 * the kretprobe is unregister on another CPU between kretprobe's
1735 * trampoline_handler and this function.
1736 */
1737 if (unlikely(!rp))
1738 return 0;
1739
1740 tk = container_of(rp, struct trace_kprobe, rp);
1741 raw_cpu_inc(*tk->nhit);
1742
1743 if (trace_probe_test_flag(&tk->tp, TP_FLAG_TRACE))
1744 kretprobe_trace_func(tk, ri, regs);
1745#ifdef CONFIG_PERF_EVENTS
1746 if (trace_probe_test_flag(&tk->tp, TP_FLAG_PROFILE))
1747 kretprobe_perf_func(tk, ri, regs);
1748#endif
1749 return 0; /* We don't tweak kernel, so just return 0 */
1750}
1751NOKPROBE_SYMBOL(kretprobe_dispatcher);
1752
1753static struct trace_event_functions kretprobe_funcs = {
1754 .trace = print_kretprobe_event
1755};
1756
1757static struct trace_event_functions kprobe_funcs = {
1758 .trace = print_kprobe_event
1759};
1760
1761static struct trace_event_fields kretprobe_fields_array[] = {
1762 { .type = TRACE_FUNCTION_TYPE,
1763 .define_fields = kretprobe_event_define_fields },
1764 {}
1765};
1766
1767static struct trace_event_fields kprobe_fields_array[] = {
1768 { .type = TRACE_FUNCTION_TYPE,
1769 .define_fields = kprobe_event_define_fields },
1770 {}
1771};
1772
1773static inline void init_trace_event_call(struct trace_kprobe *tk)
1774{
1775 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1776
1777 if (trace_kprobe_is_return(tk)) {
1778 call->event.funcs = &kretprobe_funcs;
1779 call->class->fields_array = kretprobe_fields_array;
1780 } else {
1781 call->event.funcs = &kprobe_funcs;
1782 call->class->fields_array = kprobe_fields_array;
1783 }
1784
1785 call->flags = TRACE_EVENT_FL_KPROBE;
1786 call->class->reg = kprobe_register;
1787}
1788
1789static int register_kprobe_event(struct trace_kprobe *tk)
1790{
1791 init_trace_event_call(tk);
1792
1793 return trace_probe_register_event_call(&tk->tp);
1794}
1795
1796static int unregister_kprobe_event(struct trace_kprobe *tk)
1797{
1798 return trace_probe_unregister_event_call(&tk->tp);
1799}
1800
1801#ifdef CONFIG_PERF_EVENTS
1802
1803/* create a trace_kprobe, but don't add it to global lists */
1804struct trace_event_call *
1805create_local_trace_kprobe(char *func, void *addr, unsigned long offs,
1806 bool is_return)
1807{
1808 enum probe_print_type ptype;
1809 struct trace_kprobe *tk;
1810 int ret;
1811 char *event;
1812
1813 if (func) {
1814 unsigned int count;
1815
1816 count = number_of_same_symbols(func);
1817 if (count > 1)
1818 /*
1819 * Users should use addr to remove the ambiguity of
1820 * using func only.
1821 */
1822 return ERR_PTR(-EADDRNOTAVAIL);
1823 else if (count == 0)
1824 /*
1825 * We can return ENOENT earlier than when register the
1826 * kprobe.
1827 */
1828 return ERR_PTR(-ENOENT);
1829 }
1830
1831 /*
1832 * local trace_kprobes are not added to dyn_event, so they are never
1833 * searched in find_trace_kprobe(). Therefore, there is no concern of
1834 * duplicated name here.
1835 */
1836 event = func ? func : "DUMMY_EVENT";
1837
1838 tk = alloc_trace_kprobe(KPROBE_EVENT_SYSTEM, event, (void *)addr, func,
1839 offs, 0 /* maxactive */, 0 /* nargs */,
1840 is_return);
1841
1842 if (IS_ERR(tk)) {
1843 pr_info("Failed to allocate trace_probe.(%d)\n",
1844 (int)PTR_ERR(tk));
1845 return ERR_CAST(tk);
1846 }
1847
1848 init_trace_event_call(tk);
1849
1850 ptype = trace_kprobe_is_return(tk) ?
1851 PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL;
1852 if (traceprobe_set_print_fmt(&tk->tp, ptype) < 0) {
1853 ret = -ENOMEM;
1854 goto error;
1855 }
1856
1857 ret = __register_trace_kprobe(tk);
1858 if (ret < 0)
1859 goto error;
1860
1861 return trace_probe_event_call(&tk->tp);
1862error:
1863 free_trace_kprobe(tk);
1864 return ERR_PTR(ret);
1865}
1866
1867void destroy_local_trace_kprobe(struct trace_event_call *event_call)
1868{
1869 struct trace_kprobe *tk;
1870
1871 tk = trace_kprobe_primary_from_call(event_call);
1872 if (unlikely(!tk))
1873 return;
1874
1875 if (trace_probe_is_enabled(&tk->tp)) {
1876 WARN_ON(1);
1877 return;
1878 }
1879
1880 __unregister_trace_kprobe(tk);
1881
1882 free_trace_kprobe(tk);
1883}
1884#endif /* CONFIG_PERF_EVENTS */
1885
1886static __init void enable_boot_kprobe_events(void)
1887{
1888 struct trace_array *tr = top_trace_array();
1889 struct trace_event_file *file;
1890 struct trace_kprobe *tk;
1891 struct dyn_event *pos;
1892
1893 mutex_lock(&event_mutex);
1894 for_each_trace_kprobe(tk, pos) {
1895 list_for_each_entry(file, &tr->events, list)
1896 if (file->event_call == trace_probe_event_call(&tk->tp))
1897 trace_event_enable_disable(file, 1, 0);
1898 }
1899 mutex_unlock(&event_mutex);
1900}
1901
1902static __init void setup_boot_kprobe_events(void)
1903{
1904 char *p, *cmd = kprobe_boot_events_buf;
1905 int ret;
1906
1907 strreplace(kprobe_boot_events_buf, ',', ' ');
1908
1909 while (cmd && *cmd != '\0') {
1910 p = strchr(cmd, ';');
1911 if (p)
1912 *p++ = '\0';
1913
1914 ret = create_or_delete_trace_kprobe(cmd);
1915 if (ret)
1916 pr_warn("Failed to add event(%d): %s\n", ret, cmd);
1917
1918 cmd = p;
1919 }
1920
1921 enable_boot_kprobe_events();
1922}
1923
1924/*
1925 * Register dynevent at core_initcall. This allows kernel to setup kprobe
1926 * events in postcore_initcall without tracefs.
1927 */
1928static __init int init_kprobe_trace_early(void)
1929{
1930 int ret;
1931
1932 ret = dyn_event_register(&trace_kprobe_ops);
1933 if (ret)
1934 return ret;
1935
1936 if (register_module_notifier(&trace_kprobe_module_nb))
1937 return -EINVAL;
1938
1939 return 0;
1940}
1941core_initcall(init_kprobe_trace_early);
1942
1943/* Make a tracefs interface for controlling probe points */
1944static __init int init_kprobe_trace(void)
1945{
1946 int ret;
1947
1948 ret = tracing_init_dentry();
1949 if (ret)
1950 return 0;
1951
1952 /* Event list interface */
1953 trace_create_file("kprobe_events", TRACE_MODE_WRITE,
1954 NULL, NULL, &kprobe_events_ops);
1955
1956 /* Profile interface */
1957 trace_create_file("kprobe_profile", TRACE_MODE_READ,
1958 NULL, NULL, &kprobe_profile_ops);
1959
1960 setup_boot_kprobe_events();
1961
1962 return 0;
1963}
1964fs_initcall(init_kprobe_trace);
1965
1966
1967#ifdef CONFIG_FTRACE_STARTUP_TEST
1968static __init struct trace_event_file *
1969find_trace_probe_file(struct trace_kprobe *tk, struct trace_array *tr)
1970{
1971 struct trace_event_file *file;
1972
1973 list_for_each_entry(file, &tr->events, list)
1974 if (file->event_call == trace_probe_event_call(&tk->tp))
1975 return file;
1976
1977 return NULL;
1978}
1979
1980/*
1981 * Nobody but us can call enable_trace_kprobe/disable_trace_kprobe at this
1982 * stage, we can do this lockless.
1983 */
1984static __init int kprobe_trace_self_tests_init(void)
1985{
1986 int ret, warn = 0;
1987 int (*target)(int, int, int, int, int, int);
1988 struct trace_kprobe *tk;
1989 struct trace_event_file *file;
1990
1991 if (tracing_is_disabled())
1992 return -ENODEV;
1993
1994 if (tracing_selftest_disabled)
1995 return 0;
1996
1997 target = kprobe_trace_selftest_target;
1998
1999 pr_info("Testing kprobe tracing: ");
2000
2001 ret = create_or_delete_trace_kprobe("p:testprobe kprobe_trace_selftest_target $stack $stack0 +0($stack)");
2002 if (WARN_ON_ONCE(ret)) {
2003 pr_warn("error on probing function entry.\n");
2004 warn++;
2005 } else {
2006 /* Enable trace point */
2007 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
2008 if (WARN_ON_ONCE(tk == NULL)) {
2009 pr_warn("error on getting new probe.\n");
2010 warn++;
2011 } else {
2012 file = find_trace_probe_file(tk, top_trace_array());
2013 if (WARN_ON_ONCE(file == NULL)) {
2014 pr_warn("error on getting probe file.\n");
2015 warn++;
2016 } else
2017 enable_trace_kprobe(
2018 trace_probe_event_call(&tk->tp), file);
2019 }
2020 }
2021
2022 ret = create_or_delete_trace_kprobe("r:testprobe2 kprobe_trace_selftest_target $retval");
2023 if (WARN_ON_ONCE(ret)) {
2024 pr_warn("error on probing function return.\n");
2025 warn++;
2026 } else {
2027 /* Enable trace point */
2028 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
2029 if (WARN_ON_ONCE(tk == NULL)) {
2030 pr_warn("error on getting 2nd new probe.\n");
2031 warn++;
2032 } else {
2033 file = find_trace_probe_file(tk, top_trace_array());
2034 if (WARN_ON_ONCE(file == NULL)) {
2035 pr_warn("error on getting probe file.\n");
2036 warn++;
2037 } else
2038 enable_trace_kprobe(
2039 trace_probe_event_call(&tk->tp), file);
2040 }
2041 }
2042
2043 if (warn)
2044 goto end;
2045
2046 ret = target(1, 2, 3, 4, 5, 6);
2047
2048 /*
2049 * Not expecting an error here, the check is only to prevent the
2050 * optimizer from removing the call to target() as otherwise there
2051 * are no side-effects and the call is never performed.
2052 */
2053 if (ret != 21)
2054 warn++;
2055
2056 /* Disable trace points before removing it */
2057 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
2058 if (WARN_ON_ONCE(tk == NULL)) {
2059 pr_warn("error on getting test probe.\n");
2060 warn++;
2061 } else {
2062 if (trace_kprobe_nhit(tk) != 1) {
2063 pr_warn("incorrect number of testprobe hits\n");
2064 warn++;
2065 }
2066
2067 file = find_trace_probe_file(tk, top_trace_array());
2068 if (WARN_ON_ONCE(file == NULL)) {
2069 pr_warn("error on getting probe file.\n");
2070 warn++;
2071 } else
2072 disable_trace_kprobe(
2073 trace_probe_event_call(&tk->tp), file);
2074 }
2075
2076 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
2077 if (WARN_ON_ONCE(tk == NULL)) {
2078 pr_warn("error on getting 2nd test probe.\n");
2079 warn++;
2080 } else {
2081 if (trace_kprobe_nhit(tk) != 1) {
2082 pr_warn("incorrect number of testprobe2 hits\n");
2083 warn++;
2084 }
2085
2086 file = find_trace_probe_file(tk, top_trace_array());
2087 if (WARN_ON_ONCE(file == NULL)) {
2088 pr_warn("error on getting probe file.\n");
2089 warn++;
2090 } else
2091 disable_trace_kprobe(
2092 trace_probe_event_call(&tk->tp), file);
2093 }
2094
2095 ret = create_or_delete_trace_kprobe("-:testprobe");
2096 if (WARN_ON_ONCE(ret)) {
2097 pr_warn("error on deleting a probe.\n");
2098 warn++;
2099 }
2100
2101 ret = create_or_delete_trace_kprobe("-:testprobe2");
2102 if (WARN_ON_ONCE(ret)) {
2103 pr_warn("error on deleting a probe.\n");
2104 warn++;
2105 }
2106
2107end:
2108 ret = dyn_events_release_all(&trace_kprobe_ops);
2109 if (WARN_ON_ONCE(ret)) {
2110 pr_warn("error on cleaning up probes.\n");
2111 warn++;
2112 }
2113 /*
2114 * Wait for the optimizer work to finish. Otherwise it might fiddle
2115 * with probes in already freed __init text.
2116 */
2117 wait_for_kprobe_optimizer();
2118 if (warn)
2119 pr_cont("NG: Some tests are failed. Please check them.\n");
2120 else
2121 pr_cont("OK\n");
2122 return 0;
2123}
2124
2125late_initcall(kprobe_trace_self_tests_init);
2126
2127#endif