OSDN Git Service

Merge tag 'probes-fixes-v6.5-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git...
[tomoyo/tomoyo-test1.git] / kernel / trace / trace_kprobe.c
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Kprobes-based tracing events
4  *
5  * Created by Masami Hiramatsu <mhiramat@redhat.com>
6  *
7  */
8 #define pr_fmt(fmt)     "trace_kprobe: " fmt
9
10 #include <linux/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 */
29 static char kprobe_boot_events_buf[COMMAND_LINE_SIZE] __initdata;
30
31 static 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
40 static int trace_kprobe_create(const char *raw_command);
41 static int trace_kprobe_show(struct seq_file *m, struct dyn_event *ev);
42 static int trace_kprobe_release(struct dyn_event *ev);
43 static bool trace_kprobe_is_busy(struct dyn_event *ev);
44 static bool trace_kprobe_match(const char *system, const char *event,
45                         int argc, const char **argv, struct dyn_event *ev);
46
47 static 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  */
58 struct 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
66 static bool is_trace_kprobe(struct dyn_event *ev)
67 {
68         return ev->ops == &trace_kprobe_ops;
69 }
70
71 static 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
85 static nokprobe_inline bool trace_kprobe_is_return(struct trace_kprobe *tk)
86 {
87         return tk->rp.handler != NULL;
88 }
89
90 static nokprobe_inline const char *trace_kprobe_symbol(struct trace_kprobe *tk)
91 {
92         return tk->symbol ? tk->symbol : "unknown";
93 }
94
95 static nokprobe_inline unsigned long trace_kprobe_offset(struct trace_kprobe *tk)
96 {
97         return tk->rp.kp.offset;
98 }
99
100 static nokprobe_inline bool trace_kprobe_has_gone(struct trace_kprobe *tk)
101 {
102         return kprobe_gone(&tk->rp.kp);
103 }
104
105 static nokprobe_inline bool trace_kprobe_within_module(struct trace_kprobe *tk,
106                                                  struct module *mod)
107 {
108         int len = strlen(module_name(mod));
109         const char *name = trace_kprobe_symbol(tk);
110
111         return strncmp(module_name(mod), name, len) == 0 && name[len] == ':';
112 }
113
114 static nokprobe_inline bool trace_kprobe_module_exist(struct trace_kprobe *tk)
115 {
116         char *p;
117         bool ret;
118
119         if (!tk->symbol)
120                 return false;
121         p = strchr(tk->symbol, ':');
122         if (!p)
123                 return true;
124         *p = '\0';
125         rcu_read_lock_sched();
126         ret = !!find_module(tk->symbol);
127         rcu_read_unlock_sched();
128         *p = ':';
129
130         return ret;
131 }
132
133 static bool trace_kprobe_is_busy(struct dyn_event *ev)
134 {
135         struct trace_kprobe *tk = to_trace_kprobe(ev);
136
137         return trace_probe_is_enabled(&tk->tp);
138 }
139
140 static bool trace_kprobe_match_command_head(struct trace_kprobe *tk,
141                                             int argc, const char **argv)
142 {
143         char buf[MAX_ARGSTR_LEN + 1];
144
145         if (!argc)
146                 return true;
147
148         if (!tk->symbol)
149                 snprintf(buf, sizeof(buf), "0x%p", tk->rp.kp.addr);
150         else if (tk->rp.kp.offset)
151                 snprintf(buf, sizeof(buf), "%s+%u",
152                          trace_kprobe_symbol(tk), tk->rp.kp.offset);
153         else
154                 snprintf(buf, sizeof(buf), "%s", trace_kprobe_symbol(tk));
155         if (strcmp(buf, argv[0]))
156                 return false;
157         argc--; argv++;
158
159         return trace_probe_match_command_args(&tk->tp, argc, argv);
160 }
161
162 static bool trace_kprobe_match(const char *system, const char *event,
163                         int argc, const char **argv, struct dyn_event *ev)
164 {
165         struct trace_kprobe *tk = to_trace_kprobe(ev);
166
167         return (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
173 static 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
184 static 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 */
191 static nokprobe_inline
192 unsigned 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
207 static nokprobe_inline struct trace_kprobe *
208 trace_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
219 bool 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
228 bool 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
236 static int register_kprobe_event(struct trace_kprobe *tk);
237 static int unregister_kprobe_event(struct trace_kprobe *tk);
238
239 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
240 static int kretprobe_dispatcher(struct kretprobe_instance *ri,
241                                 struct pt_regs *regs);
242
243 static 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  */
256 static struct trace_kprobe *alloc_trace_kprobe(const char *group,
257                                              const char *event,
258                                              void *addr,
259                                              const char *symbol,
260                                              unsigned long offs,
261                                              int maxactive,
262                                              int nargs, bool is_return)
263 {
264         struct trace_kprobe *tk;
265         int ret = -ENOMEM;
266
267         tk = kzalloc(struct_size(tk, tp.args, nargs), GFP_KERNEL);
268         if (!tk)
269                 return ERR_PTR(ret);
270
271         tk->nhit = alloc_percpu(unsigned long);
272         if (!tk->nhit)
273                 goto error;
274
275         if (symbol) {
276                 tk->symbol = kstrdup(symbol, GFP_KERNEL);
277                 if (!tk->symbol)
278                         goto error;
279                 tk->rp.kp.symbol_name = tk->symbol;
280                 tk->rp.kp.offset = offs;
281         } else
282                 tk->rp.kp.addr = addr;
283
284         if (is_return)
285                 tk->rp.handler = kretprobe_dispatcher;
286         else
287                 tk->rp.kp.pre_handler = kprobe_dispatcher;
288
289         tk->rp.maxactive = maxactive;
290         INIT_HLIST_NODE(&tk->rp.kp.hlist);
291         INIT_LIST_HEAD(&tk->rp.kp.list);
292
293         ret = trace_probe_init(&tk->tp, event, group, false);
294         if (ret < 0)
295                 goto error;
296
297         dyn_event_init(&tk->devent, &trace_kprobe_ops);
298         return tk;
299 error:
300         free_trace_kprobe(tk);
301         return ERR_PTR(ret);
302 }
303
304 static 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
317 static 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
331 static 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  */
349 static 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  */
399 static 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)
435 static 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
452 static 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 */
478 static 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 %s\n",
491                         trace_kprobe_symbol(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 */
516 static 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 */
532 static 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
550 unreg:
551         __unregister_trace_kprobe(tk);
552         dyn_event_remove(&tk->devent);
553         trace_probe_unlink(&tk->tp);
554
555         return 0;
556 }
557
558 static 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
587 static int append_trace_kprobe(struct trace_kprobe *tk, struct trace_kprobe *to)
588 {
589         int ret;
590
591         ret = trace_probe_compare_arg_type(&tk->tp, &to->tp);
592         if (ret) {
593                 /* Note that argument starts index = 2 */
594                 trace_probe_log_set_index(ret + 1);
595                 trace_probe_log_err(0, DIFF_ARG_TYPE);
596                 return -EEXIST;
597         }
598         if (trace_kprobe_has_same_kprobe(to, tk)) {
599                 trace_probe_log_set_index(0);
600                 trace_probe_log_err(0, SAME_PROBE);
601                 return -EEXIST;
602         }
603
604         /* Append to existing event */
605         ret = trace_probe_append(&tk->tp, &to->tp);
606         if (ret)
607                 return ret;
608
609         /* Register k*probe */
610         ret = __register_trace_kprobe(tk);
611         if (ret == -ENOENT && !trace_kprobe_module_exist(tk)) {
612                 pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
613                 ret = 0;
614         }
615
616         if (ret)
617                 trace_probe_unlink(&tk->tp);
618         else
619                 dyn_event_add(&tk->devent, trace_probe_event_call(&tk->tp));
620
621         return ret;
622 }
623
624 /* Register a trace_probe and probe_event */
625 static int register_trace_kprobe(struct trace_kprobe *tk)
626 {
627         struct trace_kprobe *old_tk;
628         int ret;
629
630         mutex_lock(&event_mutex);
631
632         old_tk = find_trace_kprobe(trace_probe_name(&tk->tp),
633                                    trace_probe_group_name(&tk->tp));
634         if (old_tk) {
635                 if (trace_kprobe_is_return(tk) != trace_kprobe_is_return(old_tk)) {
636                         trace_probe_log_set_index(0);
637                         trace_probe_log_err(0, DIFF_PROBE_TYPE);
638                         ret = -EEXIST;
639                 } else {
640                         ret = append_trace_kprobe(tk, old_tk);
641                 }
642                 goto end;
643         }
644
645         /* Register new event */
646         ret = register_kprobe_event(tk);
647         if (ret) {
648                 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
668 end:
669         mutex_unlock(&event_mutex);
670         return ret;
671 }
672
673 /* Module notifier call back, checking event on the module */
674 static 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
703 static struct notifier_block trace_kprobe_module_nb = {
704         .notifier_call = trace_kprobe_module_callback,
705         .priority = 1   /* Invoked after kprobe module callback */
706 };
707
708 static int __trace_kprobe_create(int argc, const char *argv[])
709 {
710         /*
711          * Argument syntax:
712          *  - Add kprobe:
713          *      p[:[GRP/][EVENT]] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
714          *  - Add kretprobe:
715          *      r[MAXACTIVE][:[GRP/][EVENT]] [MOD:]KSYM[+0] [FETCHARGS]
716          *    Or
717          *      p[:[GRP/][EVENT]] [MOD:]KSYM[+0]%return [FETCHARGS]
718          *
719          * Fetch args:
720          *  $retval     : fetch return value
721          *  $stack      : fetch stack address
722          *  $stackN     : fetch Nth of stack (N:0-)
723          *  $comm       : fetch current task comm
724          *  @ADDR       : fetch memory at ADDR (ADDR should be in kernel)
725          *  @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
726          *  %REG        : fetch register REG
727          * Dereferencing memory fetch:
728          *  +|-offs(ARG) : fetch memory at ARG +|- offs address.
729          * Alias name of args:
730          *  NAME=FETCHARG : set NAME as alias of FETCHARG.
731          * Type of args:
732          *  FETCHARG:TYPE : use TYPE instead of unsigned long.
733          */
734         struct trace_kprobe *tk = NULL;
735         int i, len, new_argc = 0, ret = 0;
736         bool is_return = false;
737         char *symbol = NULL, *tmp = NULL;
738         const char **new_argv = NULL;
739         const char *event = NULL, *group = KPROBE_EVENT_SYSTEM;
740         enum probe_print_type ptype;
741         int maxactive = 0;
742         long offset = 0;
743         void *addr = NULL;
744         char buf[MAX_EVENT_NAME_LEN];
745         char gbuf[MAX_EVENT_NAME_LEN];
746         char abuf[MAX_BTF_ARGS_LEN];
747         struct traceprobe_parse_context ctx = { .flags = TPARG_FL_KERNEL };
748
749         switch (argv[0][0]) {
750         case 'r':
751                 is_return = true;
752                 break;
753         case 'p':
754                 break;
755         default:
756                 return -ECANCELED;
757         }
758         if (argc < 2)
759                 return -ECANCELED;
760
761         trace_probe_log_init("trace_kprobe", argc, argv);
762
763         event = strchr(&argv[0][1], ':');
764         if (event)
765                 event++;
766
767         if (isdigit(argv[0][1])) {
768                 if (!is_return) {
769                         trace_probe_log_err(1, BAD_MAXACT_TYPE);
770                         goto parse_error;
771                 }
772                 if (event)
773                         len = event - &argv[0][1] - 1;
774                 else
775                         len = strlen(&argv[0][1]);
776                 if (len > MAX_EVENT_NAME_LEN - 1) {
777                         trace_probe_log_err(1, BAD_MAXACT);
778                         goto parse_error;
779                 }
780                 memcpy(buf, &argv[0][1], len);
781                 buf[len] = '\0';
782                 ret = kstrtouint(buf, 0, &maxactive);
783                 if (ret || !maxactive) {
784                         trace_probe_log_err(1, BAD_MAXACT);
785                         goto parse_error;
786                 }
787                 /* kretprobes instances are iterated over via a list. The
788                  * maximum should stay reasonable.
789                  */
790                 if (maxactive > KRETPROBE_MAXACTIVE_MAX) {
791                         trace_probe_log_err(1, MAXACT_TOO_BIG);
792                         goto parse_error;
793                 }
794         }
795
796         /* try to parse an address. if that fails, try to read the
797          * input as a symbol. */
798         if (kstrtoul(argv[1], 0, (unsigned long *)&addr)) {
799                 trace_probe_log_set_index(1);
800                 /* Check whether uprobe event specified */
801                 if (strchr(argv[1], '/') && strchr(argv[1], ':')) {
802                         ret = -ECANCELED;
803                         goto error;
804                 }
805                 /* a symbol specified */
806                 symbol = kstrdup(argv[1], GFP_KERNEL);
807                 if (!symbol)
808                         return -ENOMEM;
809
810                 tmp = strchr(symbol, '%');
811                 if (tmp) {
812                         if (!strcmp(tmp, "%return")) {
813                                 *tmp = '\0';
814                                 is_return = true;
815                         } else {
816                                 trace_probe_log_err(tmp - symbol, BAD_ADDR_SUFFIX);
817                                 goto parse_error;
818                         }
819                 }
820
821                 /* TODO: support .init module functions */
822                 ret = traceprobe_split_symbol_offset(symbol, &offset);
823                 if (ret || offset < 0 || offset > UINT_MAX) {
824                         trace_probe_log_err(0, BAD_PROBE_ADDR);
825                         goto parse_error;
826                 }
827                 if (is_return)
828                         ctx.flags |= TPARG_FL_RETURN;
829                 ret = kprobe_on_func_entry(NULL, symbol, offset);
830                 if (ret == 0 && !is_return)
831                         ctx.flags |= TPARG_FL_FENTRY;
832                 /* Defer the ENOENT case until register kprobe */
833                 if (ret == -EINVAL && is_return) {
834                         trace_probe_log_err(0, BAD_RETPROBE);
835                         goto parse_error;
836                 }
837         }
838
839         trace_probe_log_set_index(0);
840         if (event) {
841                 ret = traceprobe_parse_event_name(&event, &group, gbuf,
842                                                   event - argv[0]);
843                 if (ret)
844                         goto parse_error;
845         }
846
847         if (!event) {
848                 /* Make a new event name */
849                 if (symbol)
850                         snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
851                                  is_return ? 'r' : 'p', symbol, offset);
852                 else
853                         snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
854                                  is_return ? 'r' : 'p', addr);
855                 sanitize_event_name(buf);
856                 event = buf;
857         }
858
859         argc -= 2; argv += 2;
860         ctx.funcname = symbol;
861         new_argv = traceprobe_expand_meta_args(argc, argv, &new_argc,
862                                                abuf, MAX_BTF_ARGS_LEN, &ctx);
863         if (IS_ERR(new_argv)) {
864                 ret = PTR_ERR(new_argv);
865                 new_argv = NULL;
866                 goto out;
867         }
868         if (new_argv) {
869                 argc = new_argc;
870                 argv = new_argv;
871         }
872
873         /* setup a probe */
874         tk = alloc_trace_kprobe(group, event, addr, symbol, offset, maxactive,
875                                 argc, is_return);
876         if (IS_ERR(tk)) {
877                 ret = PTR_ERR(tk);
878                 /* This must return -ENOMEM, else there is a bug */
879                 WARN_ON_ONCE(ret != -ENOMEM);
880                 goto out;       /* We know tk is not allocated */
881         }
882
883         /* parse arguments */
884         for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
885                 trace_probe_log_set_index(i + 2);
886                 ctx.offset = 0;
887                 ret = traceprobe_parse_probe_arg(&tk->tp, i, argv[i], &ctx);
888                 if (ret)
889                         goto error;     /* This can be -ENOMEM */
890         }
891
892         ptype = is_return ? PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL;
893         ret = traceprobe_set_print_fmt(&tk->tp, ptype);
894         if (ret < 0)
895                 goto error;
896
897         ret = register_trace_kprobe(tk);
898         if (ret) {
899                 trace_probe_log_set_index(1);
900                 if (ret == -EILSEQ)
901                         trace_probe_log_err(0, BAD_INSN_BNDRY);
902                 else if (ret == -ENOENT)
903                         trace_probe_log_err(0, BAD_PROBE_ADDR);
904                 else if (ret != -ENOMEM && ret != -EEXIST)
905                         trace_probe_log_err(0, FAIL_REG_PROBE);
906                 goto error;
907         }
908
909 out:
910         trace_probe_log_clear();
911         kfree(new_argv);
912         kfree(symbol);
913         return ret;
914
915 parse_error:
916         ret = -EINVAL;
917 error:
918         free_trace_kprobe(tk);
919         goto out;
920 }
921
922 static int trace_kprobe_create(const char *raw_command)
923 {
924         return trace_probe_create(raw_command, __trace_kprobe_create);
925 }
926
927 static int create_or_delete_trace_kprobe(const char *raw_command)
928 {
929         int ret;
930
931         if (raw_command[0] == '-')
932                 return dyn_event_release(raw_command, &trace_kprobe_ops);
933
934         ret = trace_kprobe_create(raw_command);
935         return ret == -ECANCELED ? -EINVAL : ret;
936 }
937
938 static int trace_kprobe_run_command(struct dynevent_cmd *cmd)
939 {
940         return create_or_delete_trace_kprobe(cmd->seq.buffer);
941 }
942
943 /**
944  * kprobe_event_cmd_init - Initialize a kprobe event command object
945  * @cmd: A pointer to the dynevent_cmd struct representing the new event
946  * @buf: A pointer to the buffer used to build the command
947  * @maxlen: The length of the buffer passed in @buf
948  *
949  * Initialize a synthetic event command object.  Use this before
950  * calling any of the other kprobe_event functions.
951  */
952 void kprobe_event_cmd_init(struct dynevent_cmd *cmd, char *buf, int maxlen)
953 {
954         dynevent_cmd_init(cmd, buf, maxlen, DYNEVENT_TYPE_KPROBE,
955                           trace_kprobe_run_command);
956 }
957 EXPORT_SYMBOL_GPL(kprobe_event_cmd_init);
958
959 /**
960  * __kprobe_event_gen_cmd_start - Generate a kprobe event command from arg list
961  * @cmd: A pointer to the dynevent_cmd struct representing the new event
962  * @name: The name of the kprobe event
963  * @loc: The location of the kprobe event
964  * @kretprobe: Is this a return probe?
965  * @args: Variable number of arg (pairs), one pair for each field
966  *
967  * NOTE: Users normally won't want to call this function directly, but
968  * rather use the kprobe_event_gen_cmd_start() wrapper, which automatically
969  * adds a NULL to the end of the arg list.  If this function is used
970  * directly, make sure the last arg in the variable arg list is NULL.
971  *
972  * Generate a kprobe event command to be executed by
973  * kprobe_event_gen_cmd_end().  This function can be used to generate the
974  * complete command or only the first part of it; in the latter case,
975  * kprobe_event_add_fields() can be used to add more fields following this.
976  *
977  * Unlikely the synth_event_gen_cmd_start(), @loc must be specified. This
978  * returns -EINVAL if @loc == NULL.
979  *
980  * Return: 0 if successful, error otherwise.
981  */
982 int __kprobe_event_gen_cmd_start(struct dynevent_cmd *cmd, bool kretprobe,
983                                  const char *name, const char *loc, ...)
984 {
985         char buf[MAX_EVENT_NAME_LEN];
986         struct dynevent_arg arg;
987         va_list args;
988         int ret;
989
990         if (cmd->type != DYNEVENT_TYPE_KPROBE)
991                 return -EINVAL;
992
993         if (!loc)
994                 return -EINVAL;
995
996         if (kretprobe)
997                 snprintf(buf, MAX_EVENT_NAME_LEN, "r:kprobes/%s", name);
998         else
999                 snprintf(buf, MAX_EVENT_NAME_LEN, "p:kprobes/%s", name);
1000
1001         ret = dynevent_str_add(cmd, buf);
1002         if (ret)
1003                 return ret;
1004
1005         dynevent_arg_init(&arg, 0);
1006         arg.str = loc;
1007         ret = dynevent_arg_add(cmd, &arg, NULL);
1008         if (ret)
1009                 return ret;
1010
1011         va_start(args, loc);
1012         for (;;) {
1013                 const char *field;
1014
1015                 field = va_arg(args, const char *);
1016                 if (!field)
1017                         break;
1018
1019                 if (++cmd->n_fields > MAX_TRACE_ARGS) {
1020                         ret = -EINVAL;
1021                         break;
1022                 }
1023
1024                 arg.str = field;
1025                 ret = dynevent_arg_add(cmd, &arg, NULL);
1026                 if (ret)
1027                         break;
1028         }
1029         va_end(args);
1030
1031         return ret;
1032 }
1033 EXPORT_SYMBOL_GPL(__kprobe_event_gen_cmd_start);
1034
1035 /**
1036  * __kprobe_event_add_fields - Add probe fields to a kprobe command from arg list
1037  * @cmd: A pointer to the dynevent_cmd struct representing the new event
1038  * @args: Variable number of arg (pairs), one pair for each field
1039  *
1040  * NOTE: Users normally won't want to call this function directly, but
1041  * rather use the kprobe_event_add_fields() wrapper, which
1042  * automatically adds a NULL to the end of the arg list.  If this
1043  * function is used directly, make sure the last arg in the variable
1044  * arg list is NULL.
1045  *
1046  * Add probe fields to an existing kprobe command using a variable
1047  * list of args.  Fields are added in the same order they're listed.
1048  *
1049  * Return: 0 if successful, error otherwise.
1050  */
1051 int __kprobe_event_add_fields(struct dynevent_cmd *cmd, ...)
1052 {
1053         struct dynevent_arg arg;
1054         va_list args;
1055         int ret = 0;
1056
1057         if (cmd->type != DYNEVENT_TYPE_KPROBE)
1058                 return -EINVAL;
1059
1060         dynevent_arg_init(&arg, 0);
1061
1062         va_start(args, cmd);
1063         for (;;) {
1064                 const char *field;
1065
1066                 field = va_arg(args, const char *);
1067                 if (!field)
1068                         break;
1069
1070                 if (++cmd->n_fields > MAX_TRACE_ARGS) {
1071                         ret = -EINVAL;
1072                         break;
1073                 }
1074
1075                 arg.str = field;
1076                 ret = dynevent_arg_add(cmd, &arg, NULL);
1077                 if (ret)
1078                         break;
1079         }
1080         va_end(args);
1081
1082         return ret;
1083 }
1084 EXPORT_SYMBOL_GPL(__kprobe_event_add_fields);
1085
1086 /**
1087  * kprobe_event_delete - Delete a kprobe event
1088  * @name: The name of the kprobe event to delete
1089  *
1090  * Delete a kprobe event with the give @name from kernel code rather
1091  * than directly from the command line.
1092  *
1093  * Return: 0 if successful, error otherwise.
1094  */
1095 int kprobe_event_delete(const char *name)
1096 {
1097         char buf[MAX_EVENT_NAME_LEN];
1098
1099         snprintf(buf, MAX_EVENT_NAME_LEN, "-:%s", name);
1100
1101         return create_or_delete_trace_kprobe(buf);
1102 }
1103 EXPORT_SYMBOL_GPL(kprobe_event_delete);
1104
1105 static int trace_kprobe_release(struct dyn_event *ev)
1106 {
1107         struct trace_kprobe *tk = to_trace_kprobe(ev);
1108         int ret = unregister_trace_kprobe(tk);
1109
1110         if (!ret)
1111                 free_trace_kprobe(tk);
1112         return ret;
1113 }
1114
1115 static int trace_kprobe_show(struct seq_file *m, struct dyn_event *ev)
1116 {
1117         struct trace_kprobe *tk = to_trace_kprobe(ev);
1118         int i;
1119
1120         seq_putc(m, trace_kprobe_is_return(tk) ? 'r' : 'p');
1121         if (trace_kprobe_is_return(tk) && tk->rp.maxactive)
1122                 seq_printf(m, "%d", tk->rp.maxactive);
1123         seq_printf(m, ":%s/%s", trace_probe_group_name(&tk->tp),
1124                                 trace_probe_name(&tk->tp));
1125
1126         if (!tk->symbol)
1127                 seq_printf(m, " 0x%p", tk->rp.kp.addr);
1128         else if (tk->rp.kp.offset)
1129                 seq_printf(m, " %s+%u", trace_kprobe_symbol(tk),
1130                            tk->rp.kp.offset);
1131         else
1132                 seq_printf(m, " %s", trace_kprobe_symbol(tk));
1133
1134         for (i = 0; i < tk->tp.nr_args; i++)
1135                 seq_printf(m, " %s=%s", tk->tp.args[i].name, tk->tp.args[i].comm);
1136         seq_putc(m, '\n');
1137
1138         return 0;
1139 }
1140
1141 static int probes_seq_show(struct seq_file *m, void *v)
1142 {
1143         struct dyn_event *ev = v;
1144
1145         if (!is_trace_kprobe(ev))
1146                 return 0;
1147
1148         return trace_kprobe_show(m, ev);
1149 }
1150
1151 static const struct seq_operations probes_seq_op = {
1152         .start  = dyn_event_seq_start,
1153         .next   = dyn_event_seq_next,
1154         .stop   = dyn_event_seq_stop,
1155         .show   = probes_seq_show
1156 };
1157
1158 static int probes_open(struct inode *inode, struct file *file)
1159 {
1160         int ret;
1161
1162         ret = security_locked_down(LOCKDOWN_TRACEFS);
1163         if (ret)
1164                 return ret;
1165
1166         if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
1167                 ret = dyn_events_release_all(&trace_kprobe_ops);
1168                 if (ret < 0)
1169                         return ret;
1170         }
1171
1172         return seq_open(file, &probes_seq_op);
1173 }
1174
1175 static ssize_t probes_write(struct file *file, const char __user *buffer,
1176                             size_t count, loff_t *ppos)
1177 {
1178         return trace_parse_run_command(file, buffer, count, ppos,
1179                                        create_or_delete_trace_kprobe);
1180 }
1181
1182 static const struct file_operations kprobe_events_ops = {
1183         .owner          = THIS_MODULE,
1184         .open           = probes_open,
1185         .read           = seq_read,
1186         .llseek         = seq_lseek,
1187         .release        = seq_release,
1188         .write          = probes_write,
1189 };
1190
1191 /* Probes profiling interfaces */
1192 static int probes_profile_seq_show(struct seq_file *m, void *v)
1193 {
1194         struct dyn_event *ev = v;
1195         struct trace_kprobe *tk;
1196         unsigned long nmissed;
1197
1198         if (!is_trace_kprobe(ev))
1199                 return 0;
1200
1201         tk = to_trace_kprobe(ev);
1202         nmissed = trace_kprobe_is_return(tk) ?
1203                 tk->rp.kp.nmissed + tk->rp.nmissed : tk->rp.kp.nmissed;
1204         seq_printf(m, "  %-44s %15lu %15lu\n",
1205                    trace_probe_name(&tk->tp),
1206                    trace_kprobe_nhit(tk),
1207                    nmissed);
1208
1209         return 0;
1210 }
1211
1212 static const struct seq_operations profile_seq_op = {
1213         .start  = dyn_event_seq_start,
1214         .next   = dyn_event_seq_next,
1215         .stop   = dyn_event_seq_stop,
1216         .show   = probes_profile_seq_show
1217 };
1218
1219 static int profile_open(struct inode *inode, struct file *file)
1220 {
1221         int ret;
1222
1223         ret = security_locked_down(LOCKDOWN_TRACEFS);
1224         if (ret)
1225                 return ret;
1226
1227         return seq_open(file, &profile_seq_op);
1228 }
1229
1230 static const struct file_operations kprobe_profile_ops = {
1231         .owner          = THIS_MODULE,
1232         .open           = profile_open,
1233         .read           = seq_read,
1234         .llseek         = seq_lseek,
1235         .release        = seq_release,
1236 };
1237
1238 /* Note that we don't verify it, since the code does not come from user space */
1239 static int
1240 process_fetch_insn(struct fetch_insn *code, void *rec, void *dest,
1241                    void *base)
1242 {
1243         struct pt_regs *regs = rec;
1244         unsigned long val;
1245         int ret;
1246
1247 retry:
1248         /* 1st stage: get value from context */
1249         switch (code->op) {
1250         case FETCH_OP_REG:
1251                 val = regs_get_register(regs, code->param);
1252                 break;
1253         case FETCH_OP_STACK:
1254                 val = regs_get_kernel_stack_nth(regs, code->param);
1255                 break;
1256         case FETCH_OP_STACKP:
1257                 val = kernel_stack_pointer(regs);
1258                 break;
1259         case FETCH_OP_RETVAL:
1260                 val = regs_return_value(regs);
1261                 break;
1262 #ifdef CONFIG_HAVE_FUNCTION_ARG_ACCESS_API
1263         case FETCH_OP_ARG:
1264                 val = regs_get_kernel_argument(regs, code->param);
1265                 break;
1266 #endif
1267         case FETCH_NOP_SYMBOL:  /* Ignore a place holder */
1268                 code++;
1269                 goto retry;
1270         default:
1271                 ret = process_common_fetch_insn(code, &val);
1272                 if (ret < 0)
1273                         return ret;
1274         }
1275         code++;
1276
1277         return process_fetch_insn_bottom(code, val, dest, base);
1278 }
1279 NOKPROBE_SYMBOL(process_fetch_insn)
1280
1281 /* Kprobe handler */
1282 static nokprobe_inline void
1283 __kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs,
1284                     struct trace_event_file *trace_file)
1285 {
1286         struct kprobe_trace_entry_head *entry;
1287         struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1288         struct trace_event_buffer fbuffer;
1289         int dsize;
1290
1291         WARN_ON(call != trace_file->event_call);
1292
1293         if (trace_trigger_soft_disabled(trace_file))
1294                 return;
1295
1296         dsize = __get_data_size(&tk->tp, regs);
1297
1298         entry = trace_event_buffer_reserve(&fbuffer, trace_file,
1299                                            sizeof(*entry) + tk->tp.size + dsize);
1300         if (!entry)
1301                 return;
1302
1303         fbuffer.regs = regs;
1304         entry->ip = (unsigned long)tk->rp.kp.addr;
1305         store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1306
1307         trace_event_buffer_commit(&fbuffer);
1308 }
1309
1310 static void
1311 kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs)
1312 {
1313         struct event_file_link *link;
1314
1315         trace_probe_for_each_link_rcu(link, &tk->tp)
1316                 __kprobe_trace_func(tk, regs, link->file);
1317 }
1318 NOKPROBE_SYMBOL(kprobe_trace_func);
1319
1320 /* Kretprobe handler */
1321 static nokprobe_inline void
1322 __kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1323                        struct pt_regs *regs,
1324                        struct trace_event_file *trace_file)
1325 {
1326         struct kretprobe_trace_entry_head *entry;
1327         struct trace_event_buffer fbuffer;
1328         struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1329         int dsize;
1330
1331         WARN_ON(call != trace_file->event_call);
1332
1333         if (trace_trigger_soft_disabled(trace_file))
1334                 return;
1335
1336         dsize = __get_data_size(&tk->tp, regs);
1337
1338         entry = trace_event_buffer_reserve(&fbuffer, trace_file,
1339                                            sizeof(*entry) + tk->tp.size + dsize);
1340         if (!entry)
1341                 return;
1342
1343         fbuffer.regs = regs;
1344         entry->func = (unsigned long)tk->rp.kp.addr;
1345         entry->ret_ip = get_kretprobe_retaddr(ri);
1346         store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1347
1348         trace_event_buffer_commit(&fbuffer);
1349 }
1350
1351 static void
1352 kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1353                      struct pt_regs *regs)
1354 {
1355         struct event_file_link *link;
1356
1357         trace_probe_for_each_link_rcu(link, &tk->tp)
1358                 __kretprobe_trace_func(tk, ri, regs, link->file);
1359 }
1360 NOKPROBE_SYMBOL(kretprobe_trace_func);
1361
1362 /* Event entry printers */
1363 static enum print_line_t
1364 print_kprobe_event(struct trace_iterator *iter, int flags,
1365                    struct trace_event *event)
1366 {
1367         struct kprobe_trace_entry_head *field;
1368         struct trace_seq *s = &iter->seq;
1369         struct trace_probe *tp;
1370
1371         field = (struct kprobe_trace_entry_head *)iter->ent;
1372         tp = trace_probe_primary_from_call(
1373                 container_of(event, struct trace_event_call, event));
1374         if (WARN_ON_ONCE(!tp))
1375                 goto out;
1376
1377         trace_seq_printf(s, "%s: (", trace_probe_name(tp));
1378
1379         if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
1380                 goto out;
1381
1382         trace_seq_putc(s, ')');
1383
1384         if (trace_probe_print_args(s, tp->args, tp->nr_args,
1385                              (u8 *)&field[1], field) < 0)
1386                 goto out;
1387
1388         trace_seq_putc(s, '\n');
1389  out:
1390         return trace_handle_return(s);
1391 }
1392
1393 static enum print_line_t
1394 print_kretprobe_event(struct trace_iterator *iter, int flags,
1395                       struct trace_event *event)
1396 {
1397         struct kretprobe_trace_entry_head *field;
1398         struct trace_seq *s = &iter->seq;
1399         struct trace_probe *tp;
1400
1401         field = (struct kretprobe_trace_entry_head *)iter->ent;
1402         tp = trace_probe_primary_from_call(
1403                 container_of(event, struct trace_event_call, event));
1404         if (WARN_ON_ONCE(!tp))
1405                 goto out;
1406
1407         trace_seq_printf(s, "%s: (", trace_probe_name(tp));
1408
1409         if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1410                 goto out;
1411
1412         trace_seq_puts(s, " <- ");
1413
1414         if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1415                 goto out;
1416
1417         trace_seq_putc(s, ')');
1418
1419         if (trace_probe_print_args(s, tp->args, tp->nr_args,
1420                              (u8 *)&field[1], field) < 0)
1421                 goto out;
1422
1423         trace_seq_putc(s, '\n');
1424
1425  out:
1426         return trace_handle_return(s);
1427 }
1428
1429
1430 static int kprobe_event_define_fields(struct trace_event_call *event_call)
1431 {
1432         int ret;
1433         struct kprobe_trace_entry_head field;
1434         struct trace_probe *tp;
1435
1436         tp = trace_probe_primary_from_call(event_call);
1437         if (WARN_ON_ONCE(!tp))
1438                 return -ENOENT;
1439
1440         DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1441
1442         return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
1443 }
1444
1445 static int kretprobe_event_define_fields(struct trace_event_call *event_call)
1446 {
1447         int ret;
1448         struct kretprobe_trace_entry_head field;
1449         struct trace_probe *tp;
1450
1451         tp = trace_probe_primary_from_call(event_call);
1452         if (WARN_ON_ONCE(!tp))
1453                 return -ENOENT;
1454
1455         DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1456         DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1457
1458         return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
1459 }
1460
1461 #ifdef CONFIG_PERF_EVENTS
1462
1463 /* Kprobe profile handler */
1464 static int
1465 kprobe_perf_func(struct trace_kprobe *tk, struct pt_regs *regs)
1466 {
1467         struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1468         struct kprobe_trace_entry_head *entry;
1469         struct hlist_head *head;
1470         int size, __size, dsize;
1471         int rctx;
1472
1473         if (bpf_prog_array_valid(call)) {
1474                 unsigned long orig_ip = instruction_pointer(regs);
1475                 int ret;
1476
1477                 ret = trace_call_bpf(call, regs);
1478
1479                 /*
1480                  * We need to check and see if we modified the pc of the
1481                  * pt_regs, and if so return 1 so that we don't do the
1482                  * single stepping.
1483                  */
1484                 if (orig_ip != instruction_pointer(regs))
1485                         return 1;
1486                 if (!ret)
1487                         return 0;
1488         }
1489
1490         head = this_cpu_ptr(call->perf_events);
1491         if (hlist_empty(head))
1492                 return 0;
1493
1494         dsize = __get_data_size(&tk->tp, regs);
1495         __size = sizeof(*entry) + tk->tp.size + dsize;
1496         size = ALIGN(__size + sizeof(u32), sizeof(u64));
1497         size -= sizeof(u32);
1498
1499         entry = perf_trace_buf_alloc(size, NULL, &rctx);
1500         if (!entry)
1501                 return 0;
1502
1503         entry->ip = (unsigned long)tk->rp.kp.addr;
1504         memset(&entry[1], 0, dsize);
1505         store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1506         perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1507                               head, NULL);
1508         return 0;
1509 }
1510 NOKPROBE_SYMBOL(kprobe_perf_func);
1511
1512 /* Kretprobe profile handler */
1513 static void
1514 kretprobe_perf_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1515                     struct pt_regs *regs)
1516 {
1517         struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1518         struct kretprobe_trace_entry_head *entry;
1519         struct hlist_head *head;
1520         int size, __size, dsize;
1521         int rctx;
1522
1523         if (bpf_prog_array_valid(call) && !trace_call_bpf(call, regs))
1524                 return;
1525
1526         head = this_cpu_ptr(call->perf_events);
1527         if (hlist_empty(head))
1528                 return;
1529
1530         dsize = __get_data_size(&tk->tp, regs);
1531         __size = sizeof(*entry) + tk->tp.size + dsize;
1532         size = ALIGN(__size + sizeof(u32), sizeof(u64));
1533         size -= sizeof(u32);
1534
1535         entry = perf_trace_buf_alloc(size, NULL, &rctx);
1536         if (!entry)
1537                 return;
1538
1539         entry->func = (unsigned long)tk->rp.kp.addr;
1540         entry->ret_ip = get_kretprobe_retaddr(ri);
1541         store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1542         perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1543                               head, NULL);
1544 }
1545 NOKPROBE_SYMBOL(kretprobe_perf_func);
1546
1547 int bpf_get_kprobe_info(const struct perf_event *event, u32 *fd_type,
1548                         const char **symbol, u64 *probe_offset,
1549                         u64 *probe_addr, bool perf_type_tracepoint)
1550 {
1551         const char *pevent = trace_event_name(event->tp_event);
1552         const char *group = event->tp_event->class->system;
1553         struct trace_kprobe *tk;
1554
1555         if (perf_type_tracepoint)
1556                 tk = find_trace_kprobe(pevent, group);
1557         else
1558                 tk = trace_kprobe_primary_from_call(event->tp_event);
1559         if (!tk)
1560                 return -EINVAL;
1561
1562         *fd_type = trace_kprobe_is_return(tk) ? BPF_FD_TYPE_KRETPROBE
1563                                               : BPF_FD_TYPE_KPROBE;
1564         if (tk->symbol) {
1565                 *symbol = tk->symbol;
1566                 *probe_offset = tk->rp.kp.offset;
1567                 *probe_addr = 0;
1568         } else {
1569                 *symbol = NULL;
1570                 *probe_offset = 0;
1571                 *probe_addr = (unsigned long)tk->rp.kp.addr;
1572         }
1573         return 0;
1574 }
1575 #endif  /* CONFIG_PERF_EVENTS */
1576
1577 /*
1578  * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex.
1579  *
1580  * kprobe_trace_self_tests_init() does enable_trace_probe/disable_trace_probe
1581  * lockless, but we can't race with this __init function.
1582  */
1583 static int kprobe_register(struct trace_event_call *event,
1584                            enum trace_reg type, void *data)
1585 {
1586         struct trace_event_file *file = data;
1587
1588         switch (type) {
1589         case TRACE_REG_REGISTER:
1590                 return enable_trace_kprobe(event, file);
1591         case TRACE_REG_UNREGISTER:
1592                 return disable_trace_kprobe(event, file);
1593
1594 #ifdef CONFIG_PERF_EVENTS
1595         case TRACE_REG_PERF_REGISTER:
1596                 return enable_trace_kprobe(event, NULL);
1597         case TRACE_REG_PERF_UNREGISTER:
1598                 return disable_trace_kprobe(event, NULL);
1599         case TRACE_REG_PERF_OPEN:
1600         case TRACE_REG_PERF_CLOSE:
1601         case TRACE_REG_PERF_ADD:
1602         case TRACE_REG_PERF_DEL:
1603                 return 0;
1604 #endif
1605         }
1606         return 0;
1607 }
1608
1609 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1610 {
1611         struct trace_kprobe *tk = container_of(kp, struct trace_kprobe, rp.kp);
1612         int ret = 0;
1613
1614         raw_cpu_inc(*tk->nhit);
1615
1616         if (trace_probe_test_flag(&tk->tp, TP_FLAG_TRACE))
1617                 kprobe_trace_func(tk, regs);
1618 #ifdef CONFIG_PERF_EVENTS
1619         if (trace_probe_test_flag(&tk->tp, TP_FLAG_PROFILE))
1620                 ret = kprobe_perf_func(tk, regs);
1621 #endif
1622         return ret;
1623 }
1624 NOKPROBE_SYMBOL(kprobe_dispatcher);
1625
1626 static int
1627 kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1628 {
1629         struct kretprobe *rp = get_kretprobe(ri);
1630         struct trace_kprobe *tk;
1631
1632         /*
1633          * There is a small chance that get_kretprobe(ri) returns NULL when
1634          * the kretprobe is unregister on another CPU between kretprobe's
1635          * trampoline_handler and this function.
1636          */
1637         if (unlikely(!rp))
1638                 return 0;
1639
1640         tk = container_of(rp, struct trace_kprobe, rp);
1641         raw_cpu_inc(*tk->nhit);
1642
1643         if (trace_probe_test_flag(&tk->tp, TP_FLAG_TRACE))
1644                 kretprobe_trace_func(tk, ri, regs);
1645 #ifdef CONFIG_PERF_EVENTS
1646         if (trace_probe_test_flag(&tk->tp, TP_FLAG_PROFILE))
1647                 kretprobe_perf_func(tk, ri, regs);
1648 #endif
1649         return 0;       /* We don't tweak kernel, so just return 0 */
1650 }
1651 NOKPROBE_SYMBOL(kretprobe_dispatcher);
1652
1653 static struct trace_event_functions kretprobe_funcs = {
1654         .trace          = print_kretprobe_event
1655 };
1656
1657 static struct trace_event_functions kprobe_funcs = {
1658         .trace          = print_kprobe_event
1659 };
1660
1661 static struct trace_event_fields kretprobe_fields_array[] = {
1662         { .type = TRACE_FUNCTION_TYPE,
1663           .define_fields = kretprobe_event_define_fields },
1664         {}
1665 };
1666
1667 static struct trace_event_fields kprobe_fields_array[] = {
1668         { .type = TRACE_FUNCTION_TYPE,
1669           .define_fields = kprobe_event_define_fields },
1670         {}
1671 };
1672
1673 static inline void init_trace_event_call(struct trace_kprobe *tk)
1674 {
1675         struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1676
1677         if (trace_kprobe_is_return(tk)) {
1678                 call->event.funcs = &kretprobe_funcs;
1679                 call->class->fields_array = kretprobe_fields_array;
1680         } else {
1681                 call->event.funcs = &kprobe_funcs;
1682                 call->class->fields_array = kprobe_fields_array;
1683         }
1684
1685         call->flags = TRACE_EVENT_FL_KPROBE;
1686         call->class->reg = kprobe_register;
1687 }
1688
1689 static int register_kprobe_event(struct trace_kprobe *tk)
1690 {
1691         init_trace_event_call(tk);
1692
1693         return trace_probe_register_event_call(&tk->tp);
1694 }
1695
1696 static int unregister_kprobe_event(struct trace_kprobe *tk)
1697 {
1698         return trace_probe_unregister_event_call(&tk->tp);
1699 }
1700
1701 #ifdef CONFIG_PERF_EVENTS
1702 /* create a trace_kprobe, but don't add it to global lists */
1703 struct trace_event_call *
1704 create_local_trace_kprobe(char *func, void *addr, unsigned long offs,
1705                           bool is_return)
1706 {
1707         enum probe_print_type ptype;
1708         struct trace_kprobe *tk;
1709         int ret;
1710         char *event;
1711
1712         /*
1713          * local trace_kprobes are not added to dyn_event, so they are never
1714          * searched in find_trace_kprobe(). Therefore, there is no concern of
1715          * duplicated name here.
1716          */
1717         event = func ? func : "DUMMY_EVENT";
1718
1719         tk = alloc_trace_kprobe(KPROBE_EVENT_SYSTEM, event, (void *)addr, func,
1720                                 offs, 0 /* maxactive */, 0 /* nargs */,
1721                                 is_return);
1722
1723         if (IS_ERR(tk)) {
1724                 pr_info("Failed to allocate trace_probe.(%d)\n",
1725                         (int)PTR_ERR(tk));
1726                 return ERR_CAST(tk);
1727         }
1728
1729         init_trace_event_call(tk);
1730
1731         ptype = trace_kprobe_is_return(tk) ?
1732                 PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL;
1733         if (traceprobe_set_print_fmt(&tk->tp, ptype) < 0) {
1734                 ret = -ENOMEM;
1735                 goto error;
1736         }
1737
1738         ret = __register_trace_kprobe(tk);
1739         if (ret < 0)
1740                 goto error;
1741
1742         return trace_probe_event_call(&tk->tp);
1743 error:
1744         free_trace_kprobe(tk);
1745         return ERR_PTR(ret);
1746 }
1747
1748 void destroy_local_trace_kprobe(struct trace_event_call *event_call)
1749 {
1750         struct trace_kprobe *tk;
1751
1752         tk = trace_kprobe_primary_from_call(event_call);
1753         if (unlikely(!tk))
1754                 return;
1755
1756         if (trace_probe_is_enabled(&tk->tp)) {
1757                 WARN_ON(1);
1758                 return;
1759         }
1760
1761         __unregister_trace_kprobe(tk);
1762
1763         free_trace_kprobe(tk);
1764 }
1765 #endif /* CONFIG_PERF_EVENTS */
1766
1767 static __init void enable_boot_kprobe_events(void)
1768 {
1769         struct trace_array *tr = top_trace_array();
1770         struct trace_event_file *file;
1771         struct trace_kprobe *tk;
1772         struct dyn_event *pos;
1773
1774         mutex_lock(&event_mutex);
1775         for_each_trace_kprobe(tk, pos) {
1776                 list_for_each_entry(file, &tr->events, list)
1777                         if (file->event_call == trace_probe_event_call(&tk->tp))
1778                                 trace_event_enable_disable(file, 1, 0);
1779         }
1780         mutex_unlock(&event_mutex);
1781 }
1782
1783 static __init void setup_boot_kprobe_events(void)
1784 {
1785         char *p, *cmd = kprobe_boot_events_buf;
1786         int ret;
1787
1788         strreplace(kprobe_boot_events_buf, ',', ' ');
1789
1790         while (cmd && *cmd != '\0') {
1791                 p = strchr(cmd, ';');
1792                 if (p)
1793                         *p++ = '\0';
1794
1795                 ret = create_or_delete_trace_kprobe(cmd);
1796                 if (ret)
1797                         pr_warn("Failed to add event(%d): %s\n", ret, cmd);
1798
1799                 cmd = p;
1800         }
1801
1802         enable_boot_kprobe_events();
1803 }
1804
1805 /*
1806  * Register dynevent at core_initcall. This allows kernel to setup kprobe
1807  * events in postcore_initcall without tracefs.
1808  */
1809 static __init int init_kprobe_trace_early(void)
1810 {
1811         int ret;
1812
1813         ret = dyn_event_register(&trace_kprobe_ops);
1814         if (ret)
1815                 return ret;
1816
1817         if (register_module_notifier(&trace_kprobe_module_nb))
1818                 return -EINVAL;
1819
1820         return 0;
1821 }
1822 core_initcall(init_kprobe_trace_early);
1823
1824 /* Make a tracefs interface for controlling probe points */
1825 static __init int init_kprobe_trace(void)
1826 {
1827         int ret;
1828
1829         ret = tracing_init_dentry();
1830         if (ret)
1831                 return 0;
1832
1833         /* Event list interface */
1834         trace_create_file("kprobe_events", TRACE_MODE_WRITE,
1835                           NULL, NULL, &kprobe_events_ops);
1836
1837         /* Profile interface */
1838         trace_create_file("kprobe_profile", TRACE_MODE_READ,
1839                           NULL, NULL, &kprobe_profile_ops);
1840
1841         setup_boot_kprobe_events();
1842
1843         return 0;
1844 }
1845 fs_initcall(init_kprobe_trace);
1846
1847
1848 #ifdef CONFIG_FTRACE_STARTUP_TEST
1849 static __init struct trace_event_file *
1850 find_trace_probe_file(struct trace_kprobe *tk, struct trace_array *tr)
1851 {
1852         struct trace_event_file *file;
1853
1854         list_for_each_entry(file, &tr->events, list)
1855                 if (file->event_call == trace_probe_event_call(&tk->tp))
1856                         return file;
1857
1858         return NULL;
1859 }
1860
1861 /*
1862  * Nobody but us can call enable_trace_kprobe/disable_trace_kprobe at this
1863  * stage, we can do this lockless.
1864  */
1865 static __init int kprobe_trace_self_tests_init(void)
1866 {
1867         int ret, warn = 0;
1868         int (*target)(int, int, int, int, int, int);
1869         struct trace_kprobe *tk;
1870         struct trace_event_file *file;
1871
1872         if (tracing_is_disabled())
1873                 return -ENODEV;
1874
1875         if (tracing_selftest_disabled)
1876                 return 0;
1877
1878         target = kprobe_trace_selftest_target;
1879
1880         pr_info("Testing kprobe tracing: ");
1881
1882         ret = create_or_delete_trace_kprobe("p:testprobe kprobe_trace_selftest_target $stack $stack0 +0($stack)");
1883         if (WARN_ON_ONCE(ret)) {
1884                 pr_warn("error on probing function entry.\n");
1885                 warn++;
1886         } else {
1887                 /* Enable trace point */
1888                 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1889                 if (WARN_ON_ONCE(tk == NULL)) {
1890                         pr_warn("error on getting new probe.\n");
1891                         warn++;
1892                 } else {
1893                         file = find_trace_probe_file(tk, top_trace_array());
1894                         if (WARN_ON_ONCE(file == NULL)) {
1895                                 pr_warn("error on getting probe file.\n");
1896                                 warn++;
1897                         } else
1898                                 enable_trace_kprobe(
1899                                         trace_probe_event_call(&tk->tp), file);
1900                 }
1901         }
1902
1903         ret = create_or_delete_trace_kprobe("r:testprobe2 kprobe_trace_selftest_target $retval");
1904         if (WARN_ON_ONCE(ret)) {
1905                 pr_warn("error on probing function return.\n");
1906                 warn++;
1907         } else {
1908                 /* Enable trace point */
1909                 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1910                 if (WARN_ON_ONCE(tk == NULL)) {
1911                         pr_warn("error on getting 2nd new probe.\n");
1912                         warn++;
1913                 } else {
1914                         file = find_trace_probe_file(tk, top_trace_array());
1915                         if (WARN_ON_ONCE(file == NULL)) {
1916                                 pr_warn("error on getting probe file.\n");
1917                                 warn++;
1918                         } else
1919                                 enable_trace_kprobe(
1920                                         trace_probe_event_call(&tk->tp), file);
1921                 }
1922         }
1923
1924         if (warn)
1925                 goto end;
1926
1927         ret = target(1, 2, 3, 4, 5, 6);
1928
1929         /*
1930          * Not expecting an error here, the check is only to prevent the
1931          * optimizer from removing the call to target() as otherwise there
1932          * are no side-effects and the call is never performed.
1933          */
1934         if (ret != 21)
1935                 warn++;
1936
1937         /* Disable trace points before removing it */
1938         tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1939         if (WARN_ON_ONCE(tk == NULL)) {
1940                 pr_warn("error on getting test probe.\n");
1941                 warn++;
1942         } else {
1943                 if (trace_kprobe_nhit(tk) != 1) {
1944                         pr_warn("incorrect number of testprobe hits\n");
1945                         warn++;
1946                 }
1947
1948                 file = find_trace_probe_file(tk, top_trace_array());
1949                 if (WARN_ON_ONCE(file == NULL)) {
1950                         pr_warn("error on getting probe file.\n");
1951                         warn++;
1952                 } else
1953                         disable_trace_kprobe(
1954                                 trace_probe_event_call(&tk->tp), file);
1955         }
1956
1957         tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1958         if (WARN_ON_ONCE(tk == NULL)) {
1959                 pr_warn("error on getting 2nd test probe.\n");
1960                 warn++;
1961         } else {
1962                 if (trace_kprobe_nhit(tk) != 1) {
1963                         pr_warn("incorrect number of testprobe2 hits\n");
1964                         warn++;
1965                 }
1966
1967                 file = find_trace_probe_file(tk, top_trace_array());
1968                 if (WARN_ON_ONCE(file == NULL)) {
1969                         pr_warn("error on getting probe file.\n");
1970                         warn++;
1971                 } else
1972                         disable_trace_kprobe(
1973                                 trace_probe_event_call(&tk->tp), file);
1974         }
1975
1976         ret = create_or_delete_trace_kprobe("-:testprobe");
1977         if (WARN_ON_ONCE(ret)) {
1978                 pr_warn("error on deleting a probe.\n");
1979                 warn++;
1980         }
1981
1982         ret = create_or_delete_trace_kprobe("-:testprobe2");
1983         if (WARN_ON_ONCE(ret)) {
1984                 pr_warn("error on deleting a probe.\n");
1985                 warn++;
1986         }
1987
1988 end:
1989         ret = dyn_events_release_all(&trace_kprobe_ops);
1990         if (WARN_ON_ONCE(ret)) {
1991                 pr_warn("error on cleaning up probes.\n");
1992                 warn++;
1993         }
1994         /*
1995          * Wait for the optimizer work to finish. Otherwise it might fiddle
1996          * with probes in already freed __init text.
1997          */
1998         wait_for_kprobe_optimizer();
1999         if (warn)
2000                 pr_cont("NG: Some tests are failed. Please check them.\n");
2001         else
2002                 pr_cont("OK\n");
2003         return 0;
2004 }
2005
2006 late_initcall(kprobe_trace_self_tests_init);
2007
2008 #endif