~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/kernel/trace/trace_kprobe.c

Version: ~ [ linux-5.15-rc6 ] ~ [ linux-5.14.14 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.75 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.155 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.213 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.252 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.287 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.289 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.18.140 ] ~ [ linux-3.16.85 ] ~ [ linux-3.14.79 ] ~ [ linux-3.12.74 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 // SPDX-License-Identifier: GPL-2.0
  2 /*
  3  * Kprobes-based tracing events
  4  *
  5  * Created by Masami Hiramatsu <mhiramat@redhat.com>
  6  *
  7  */
  8 #define pr_fmt(fmt)     "trace_kprobe: " fmt
  9 
 10 #include <linux/security.h>
 11 #include <linux/module.h>
 12 #include <linux/uaccess.h>
 13 #include <linux/rculist.h>
 14 #include <linux/error-injection.h>
 15 
 16 #include <asm/setup.h>  /* for COMMAND_LINE_SIZE */
 17 
 18 #include "trace_dynevent.h"
 19 #include "trace_kprobe_selftest.h"
 20 #include "trace_probe.h"
 21 #include "trace_probe_tmpl.h"
 22 
 23 #define KPROBE_EVENT_SYSTEM "kprobes"
 24 #define KRETPROBE_MAXACTIVE_MAX 4096
 25 #define MAX_KPROBE_CMDLINE_SIZE 1024
 26 
 27 /* Kprobe early definition from command line */
 28 static char kprobe_boot_events_buf[COMMAND_LINE_SIZE] __initdata;
 29 static bool kprobe_boot_events_enabled __initdata;
 30 
 31 static int __init set_kprobe_boot_events(char *str)
 32 {
 33         strlcpy(kprobe_boot_events_buf, str, COMMAND_LINE_SIZE);
 34         return 0;
 35 }
 36 __setup("kprobe_event=", set_kprobe_boot_events);
 37 
 38 static int trace_kprobe_create(int argc, const char **argv);
 39 static int trace_kprobe_show(struct seq_file *m, struct dyn_event *ev);
 40 static int trace_kprobe_release(struct dyn_event *ev);
 41 static bool trace_kprobe_is_busy(struct dyn_event *ev);
 42 static bool trace_kprobe_match(const char *system, const char *event,
 43                         int argc, const char **argv, struct dyn_event *ev);
 44 
 45 static struct dyn_event_operations trace_kprobe_ops = {
 46         .create = trace_kprobe_create,
 47         .show = trace_kprobe_show,
 48         .is_busy = trace_kprobe_is_busy,
 49         .free = trace_kprobe_release,
 50         .match = trace_kprobe_match,
 51 };
 52 
 53 /*
 54  * Kprobe event core functions
 55  */
 56 struct trace_kprobe {
 57         struct dyn_event        devent;
 58         struct kretprobe        rp;     /* Use rp.kp for kprobe use */
 59         unsigned long __percpu *nhit;
 60         const char              *symbol;        /* symbol name */
 61         struct trace_probe      tp;
 62 };
 63 
 64 static bool is_trace_kprobe(struct dyn_event *ev)
 65 {
 66         return ev->ops == &trace_kprobe_ops;
 67 }
 68 
 69 static struct trace_kprobe *to_trace_kprobe(struct dyn_event *ev)
 70 {
 71         return container_of(ev, struct trace_kprobe, devent);
 72 }
 73 
 74 /**
 75  * for_each_trace_kprobe - iterate over the trace_kprobe list
 76  * @pos:        the struct trace_kprobe * for each entry
 77  * @dpos:       the struct dyn_event * to use as a loop cursor
 78  */
 79 #define for_each_trace_kprobe(pos, dpos)        \
 80         for_each_dyn_event(dpos)                \
 81                 if (is_trace_kprobe(dpos) && (pos = to_trace_kprobe(dpos)))
 82 
 83 #define SIZEOF_TRACE_KPROBE(n)                          \
 84         (offsetof(struct trace_kprobe, tp.args) +       \
 85         (sizeof(struct probe_arg) * (n)))
 86 
 87 static nokprobe_inline bool trace_kprobe_is_return(struct trace_kprobe *tk)
 88 {
 89         return tk->rp.handler != NULL;
 90 }
 91 
 92 static nokprobe_inline const char *trace_kprobe_symbol(struct trace_kprobe *tk)
 93 {
 94         return tk->symbol ? tk->symbol : "unknown";
 95 }
 96 
 97 static nokprobe_inline unsigned long trace_kprobe_offset(struct trace_kprobe *tk)
 98 {
 99         return tk->rp.kp.offset;
100 }
101 
102 static nokprobe_inline bool trace_kprobe_has_gone(struct trace_kprobe *tk)
103 {
104         return !!(kprobe_gone(&tk->rp.kp));
105 }
106 
107 static nokprobe_inline bool trace_kprobe_within_module(struct trace_kprobe *tk,
108                                                  struct module *mod)
109 {
110         int len = strlen(mod->name);
111         const char *name = trace_kprobe_symbol(tk);
112         return strncmp(mod->name, name, len) == 0 && name[len] == ':';
113 }
114 
115 static nokprobe_inline bool trace_kprobe_module_exist(struct trace_kprobe *tk)
116 {
117         char *p;
118         bool ret;
119 
120         if (!tk->symbol)
121                 return false;
122         p = strchr(tk->symbol, ':');
123         if (!p)
124                 return true;
125         *p = '\0';
126         mutex_lock(&module_mutex);
127         ret = !!find_module(tk->symbol);
128         mutex_unlock(&module_mutex);
129         *p = ':';
130 
131         return ret;
132 }
133 
134 static bool trace_kprobe_is_busy(struct dyn_event *ev)
135 {
136         struct trace_kprobe *tk = to_trace_kprobe(ev);
137 
138         return trace_probe_is_enabled(&tk->tp);
139 }
140 
141 static bool trace_kprobe_match_command_head(struct trace_kprobe *tk,
142                                             int argc, const char **argv)
143 {
144         char buf[MAX_ARGSTR_LEN + 1];
145 
146         if (!argc)
147                 return true;
148 
149         if (!tk->symbol)
150                 snprintf(buf, sizeof(buf), "0x%p", tk->rp.kp.addr);
151         else if (tk->rp.kp.offset)
152                 snprintf(buf, sizeof(buf), "%s+%u",
153                          trace_kprobe_symbol(tk), tk->rp.kp.offset);
154         else
155                 snprintf(buf, sizeof(buf), "%s", trace_kprobe_symbol(tk));
156         if (strcmp(buf, argv[0]))
157                 return false;
158         argc--; argv++;
159 
160         return trace_probe_match_command_args(&tk->tp, argc, argv);
161 }
162 
163 static bool trace_kprobe_match(const char *system, const char *event,
164                         int argc, const char **argv, struct dyn_event *ev)
165 {
166         struct trace_kprobe *tk = to_trace_kprobe(ev);
167 
168         return 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) : 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(SIZEOF_TRACE_KPROBE(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_probe *pos;
334         struct trace_kprobe *tk;
335 
336         list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
337                 tk = container_of(pos, struct trace_kprobe, tp);
338                 if (!trace_kprobe_is_registered(tk))
339                         continue;
340                 if (trace_kprobe_is_return(tk))
341                         disable_kretprobe(&tk->rp);
342                 else
343                         disable_kprobe(&tk->rp.kp);
344         }
345 }
346 
347 /*
348  * Enable trace_probe
349  * if the file is NULL, enable "perf" handler, or enable "trace" handler.
350  */
351 static int enable_trace_kprobe(struct trace_event_call *call,
352                                 struct trace_event_file *file)
353 {
354         struct trace_probe *pos, *tp;
355         struct trace_kprobe *tk;
356         bool enabled;
357         int ret = 0;
358 
359         tp = trace_probe_primary_from_call(call);
360         if (WARN_ON_ONCE(!tp))
361                 return -ENODEV;
362         enabled = trace_probe_is_enabled(tp);
363 
364         /* This also changes "enabled" state */
365         if (file) {
366                 ret = trace_probe_add_file(tp, file);
367                 if (ret)
368                         return ret;
369         } else
370                 trace_probe_set_flag(tp, TP_FLAG_PROFILE);
371 
372         if (enabled)
373                 return 0;
374 
375         list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
376                 tk = container_of(pos, struct trace_kprobe, tp);
377                 if (trace_kprobe_has_gone(tk))
378                         continue;
379                 ret = __enable_trace_kprobe(tk);
380                 if (ret)
381                         break;
382                 enabled = true;
383         }
384 
385         if (ret) {
386                 /* Failed to enable one of them. Roll back all */
387                 if (enabled)
388                         __disable_trace_kprobe(tp);
389                 if (file)
390                         trace_probe_remove_file(tp, file);
391                 else
392                         trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
393         }
394 
395         return ret;
396 }
397 
398 /*
399  * Disable trace_probe
400  * if the file is NULL, disable "perf" handler, or disable "trace" handler.
401  */
402 static int disable_trace_kprobe(struct trace_event_call *call,
403                                 struct trace_event_file *file)
404 {
405         struct trace_probe *tp;
406 
407         tp = trace_probe_primary_from_call(call);
408         if (WARN_ON_ONCE(!tp))
409                 return -ENODEV;
410 
411         if (file) {
412                 if (!trace_probe_get_file_link(tp, file))
413                         return -ENOENT;
414                 if (!trace_probe_has_single_file(tp))
415                         goto out;
416                 trace_probe_clear_flag(tp, TP_FLAG_TRACE);
417         } else
418                 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
419 
420         if (!trace_probe_is_enabled(tp))
421                 __disable_trace_kprobe(tp);
422 
423  out:
424         if (file)
425                 /*
426                  * Synchronization is done in below function. For perf event,
427                  * file == NULL and perf_trace_event_unreg() calls
428                  * tracepoint_synchronize_unregister() to ensure synchronize
429                  * event. We don't need to care about it.
430                  */
431                 trace_probe_remove_file(tp, file);
432 
433         return 0;
434 }
435 
436 #if defined(CONFIG_KPROBES_ON_FTRACE) && \
437         !defined(CONFIG_KPROBE_EVENTS_ON_NOTRACE)
438 static bool __within_notrace_func(unsigned long addr)
439 {
440         unsigned long offset, size;
441 
442         if (!addr || !kallsyms_lookup_size_offset(addr, &size, &offset))
443                 return false;
444 
445         /* Get the entry address of the target function */
446         addr -= offset;
447 
448         /*
449          * Since ftrace_location_range() does inclusive range check, we need
450          * to subtract 1 byte from the end address.
451          */
452         return !ftrace_location_range(addr, addr + size - 1);
453 }
454 
455 static bool within_notrace_func(struct trace_kprobe *tk)
456 {
457         unsigned long addr = addr = trace_kprobe_address(tk);
458         char symname[KSYM_NAME_LEN], *p;
459 
460         if (!__within_notrace_func(addr))
461                 return false;
462 
463         /* Check if the address is on a suffixed-symbol */
464         if (!lookup_symbol_name(addr, symname)) {
465                 p = strchr(symname, '.');
466                 if (!p)
467                         return true;
468                 *p = '\0';
469                 addr = (unsigned long)kprobe_lookup_name(symname, 0);
470                 if (addr)
471                         return __within_notrace_func(addr);
472         }
473 
474         return true;
475 }
476 #else
477 #define within_notrace_func(tk) (false)
478 #endif
479 
480 /* Internal register function - just handle k*probes and flags */
481 static int __register_trace_kprobe(struct trace_kprobe *tk)
482 {
483         int i, ret;
484 
485         ret = security_locked_down(LOCKDOWN_KPROBES);
486         if (ret)
487                 return ret;
488 
489         if (trace_kprobe_is_registered(tk))
490                 return -EINVAL;
491 
492         if (within_notrace_func(tk)) {
493                 pr_warn("Could not probe notrace function %s\n",
494                         trace_kprobe_symbol(tk));
495                 return -EINVAL;
496         }
497 
498         for (i = 0; i < tk->tp.nr_args; i++) {
499                 ret = traceprobe_update_arg(&tk->tp.args[i]);
500                 if (ret)
501                         return ret;
502         }
503 
504         /* Set/clear disabled flag according to tp->flag */
505         if (trace_probe_is_enabled(&tk->tp))
506                 tk->rp.kp.flags &= ~KPROBE_FLAG_DISABLED;
507         else
508                 tk->rp.kp.flags |= KPROBE_FLAG_DISABLED;
509 
510         if (trace_kprobe_is_return(tk))
511                 ret = register_kretprobe(&tk->rp);
512         else
513                 ret = register_kprobe(&tk->rp.kp);
514 
515         return ret;
516 }
517 
518 /* Internal unregister function - just handle k*probes and flags */
519 static void __unregister_trace_kprobe(struct trace_kprobe *tk)
520 {
521         if (trace_kprobe_is_registered(tk)) {
522                 if (trace_kprobe_is_return(tk))
523                         unregister_kretprobe(&tk->rp);
524                 else
525                         unregister_kprobe(&tk->rp.kp);
526                 /* Cleanup kprobe for reuse and mark it unregistered */
527                 INIT_HLIST_NODE(&tk->rp.kp.hlist);
528                 INIT_LIST_HEAD(&tk->rp.kp.list);
529                 if (tk->rp.kp.symbol_name)
530                         tk->rp.kp.addr = NULL;
531         }
532 }
533 
534 /* Unregister a trace_probe and probe_event */
535 static int unregister_trace_kprobe(struct trace_kprobe *tk)
536 {
537         /* If other probes are on the event, just unregister kprobe */
538         if (trace_probe_has_sibling(&tk->tp))
539                 goto unreg;
540 
541         /* Enabled event can not be unregistered */
542         if (trace_probe_is_enabled(&tk->tp))
543                 return -EBUSY;
544 
545         /* Will fail if probe is being used by ftrace or perf */
546         if (unregister_kprobe_event(tk))
547                 return -EBUSY;
548 
549 unreg:
550         __unregister_trace_kprobe(tk);
551         dyn_event_remove(&tk->devent);
552         trace_probe_unlink(&tk->tp);
553 
554         return 0;
555 }
556 
557 static bool trace_kprobe_has_same_kprobe(struct trace_kprobe *orig,
558                                          struct trace_kprobe *comp)
559 {
560         struct trace_probe_event *tpe = orig->tp.event;
561         struct trace_probe *pos;
562         int i;
563 
564         list_for_each_entry(pos, &tpe->probes, list) {
565                 orig = container_of(pos, struct trace_kprobe, tp);
566                 if (strcmp(trace_kprobe_symbol(orig),
567                            trace_kprobe_symbol(comp)) ||
568                     trace_kprobe_offset(orig) != trace_kprobe_offset(comp))
569                         continue;
570 
571                 /*
572                  * trace_probe_compare_arg_type() ensured that nr_args and
573                  * each argument name and type are same. Let's compare comm.
574                  */
575                 for (i = 0; i < orig->tp.nr_args; i++) {
576                         if (strcmp(orig->tp.args[i].comm,
577                                    comp->tp.args[i].comm))
578                                 break;
579                 }
580 
581                 if (i == orig->tp.nr_args)
582                         return true;
583         }
584 
585         return false;
586 }
587 
588 static int append_trace_kprobe(struct trace_kprobe *tk, struct trace_kprobe *to)
589 {
590         int ret;
591 
592         ret = trace_probe_compare_arg_type(&tk->tp, &to->tp);
593         if (ret) {
594                 /* Note that argument starts index = 2 */
595                 trace_probe_log_set_index(ret + 1);
596                 trace_probe_log_err(0, DIFF_ARG_TYPE);
597                 return -EEXIST;
598         }
599         if (trace_kprobe_has_same_kprobe(to, tk)) {
600                 trace_probe_log_set_index(0);
601                 trace_probe_log_err(0, SAME_PROBE);
602                 return -EEXIST;
603         }
604 
605         /* Append to existing event */
606         ret = trace_probe_append(&tk->tp, &to->tp);
607         if (ret)
608                 return ret;
609 
610         /* Register k*probe */
611         ret = __register_trace_kprobe(tk);
612         if (ret == -ENOENT && !trace_kprobe_module_exist(tk)) {
613                 pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
614                 ret = 0;
615         }
616 
617         if (ret)
618                 trace_probe_unlink(&tk->tp);
619         else
620                 dyn_event_add(&tk->devent);
621 
622         return ret;
623 }
624 
625 /* Register a trace_probe and probe_event */
626 static int register_trace_kprobe(struct trace_kprobe *tk)
627 {
628         struct trace_kprobe *old_tk;
629         int ret;
630 
631         mutex_lock(&event_mutex);
632 
633         old_tk = find_trace_kprobe(trace_probe_name(&tk->tp),
634                                    trace_probe_group_name(&tk->tp));
635         if (old_tk) {
636                 if (trace_kprobe_is_return(tk) != trace_kprobe_is_return(old_tk)) {
637                         trace_probe_log_set_index(0);
638                         trace_probe_log_err(0, DIFF_PROBE_TYPE);
639                         ret = -EEXIST;
640                 } else {
641                         ret = append_trace_kprobe(tk, old_tk);
642                 }
643                 goto end;
644         }
645 
646         /* Register new event */
647         ret = register_kprobe_event(tk);
648         if (ret) {
649                 pr_warn("Failed to register probe event(%d)\n", ret);
650                 goto end;
651         }
652 
653         /* Register k*probe */
654         ret = __register_trace_kprobe(tk);
655         if (ret == -ENOENT && !trace_kprobe_module_exist(tk)) {
656                 pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
657                 ret = 0;
658         }
659 
660         if (ret < 0)
661                 unregister_kprobe_event(tk);
662         else
663                 dyn_event_add(&tk->devent);
664 
665 end:
666         mutex_unlock(&event_mutex);
667         return ret;
668 }
669 
670 /* Module notifier call back, checking event on the module */
671 static int trace_kprobe_module_callback(struct notifier_block *nb,
672                                        unsigned long val, void *data)
673 {
674         struct module *mod = data;
675         struct dyn_event *pos;
676         struct trace_kprobe *tk;
677         int ret;
678 
679         if (val != MODULE_STATE_COMING)
680                 return NOTIFY_DONE;
681 
682         /* Update probes on coming module */
683         mutex_lock(&event_mutex);
684         for_each_trace_kprobe(tk, pos) {
685                 if (trace_kprobe_within_module(tk, mod)) {
686                         /* Don't need to check busy - this should have gone. */
687                         __unregister_trace_kprobe(tk);
688                         ret = __register_trace_kprobe(tk);
689                         if (ret)
690                                 pr_warn("Failed to re-register probe %s on %s: %d\n",
691                                         trace_probe_name(&tk->tp),
692                                         mod->name, ret);
693                 }
694         }
695         mutex_unlock(&event_mutex);
696 
697         return NOTIFY_DONE;
698 }
699 
700 static struct notifier_block trace_kprobe_module_nb = {
701         .notifier_call = trace_kprobe_module_callback,
702         .priority = 1   /* Invoked after kprobe module callback */
703 };
704 
705 /* Convert certain expected symbols into '_' when generating event names */
706 static inline void sanitize_event_name(char *name)
707 {
708         while (*name++ != '\0')
709                 if (*name == ':' || *name == '.')
710                         *name = '_';
711 }
712 
713 static int trace_kprobe_create(int argc, const char *argv[])
714 {
715         /*
716          * Argument syntax:
717          *  - Add kprobe:
718          *      p[:[GRP/]EVENT] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
719          *  - Add kretprobe:
720          *      r[MAXACTIVE][:[GRP/]EVENT] [MOD:]KSYM[+0] [FETCHARGS]
721          * Fetch args:
722          *  $retval     : fetch return value
723          *  $stack      : fetch stack address
724          *  $stackN     : fetch Nth of stack (N:0-)
725          *  $comm       : fetch current task comm
726          *  @ADDR       : fetch memory at ADDR (ADDR should be in kernel)
727          *  @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
728          *  %REG        : fetch register REG
729          * Dereferencing memory fetch:
730          *  +|-offs(ARG) : fetch memory at ARG +|- offs address.
731          * Alias name of args:
732          *  NAME=FETCHARG : set NAME as alias of FETCHARG.
733          * Type of args:
734          *  FETCHARG:TYPE : use TYPE instead of unsigned long.
735          */
736         struct trace_kprobe *tk = NULL;
737         int i, len, ret = 0;
738         bool is_return = false;
739         char *symbol = NULL, *tmp = NULL;
740         const char *event = NULL, *group = KPROBE_EVENT_SYSTEM;
741         int maxactive = 0;
742         long offset = 0;
743         void *addr = NULL;
744         char buf[MAX_EVENT_NAME_LEN];
745         unsigned int flags = TPARG_FL_KERNEL;
746 
747         switch (argv[0][0]) {
748         case 'r':
749                 is_return = true;
750                 flags |= TPARG_FL_RETURN;
751                 break;
752         case 'p':
753                 break;
754         default:
755                 return -ECANCELED;
756         }
757         if (argc < 2)
758                 return -ECANCELED;
759 
760         trace_probe_log_init("trace_kprobe", argc, argv);
761 
762         event = strchr(&argv[0][1], ':');
763         if (event)
764                 event++;
765 
766         if (isdigit(argv[0][1])) {
767                 if (!is_return) {
768                         trace_probe_log_err(1, MAXACT_NO_KPROBE);
769                         goto parse_error;
770                 }
771                 if (event)
772                         len = event - &argv[0][1] - 1;
773                 else
774                         len = strlen(&argv[0][1]);
775                 if (len > MAX_EVENT_NAME_LEN - 1) {
776                         trace_probe_log_err(1, BAD_MAXACT);
777                         goto parse_error;
778                 }
779                 memcpy(buf, &argv[0][1], len);
780                 buf[len] = '\0';
781                 ret = kstrtouint(buf, 0, &maxactive);
782                 if (ret || !maxactive) {
783                         trace_probe_log_err(1, BAD_MAXACT);
784                         goto parse_error;
785                 }
786                 /* kretprobes instances are iterated over via a list. The
787                  * maximum should stay reasonable.
788                  */
789                 if (maxactive > KRETPROBE_MAXACTIVE_MAX) {
790                         trace_probe_log_err(1, MAXACT_TOO_BIG);
791                         goto parse_error;
792                 }
793         }
794 
795         /* try to parse an address. if that fails, try to read the
796          * input as a symbol. */
797         if (kstrtoul(argv[1], 0, (unsigned long *)&addr)) {
798                 trace_probe_log_set_index(1);
799                 /* Check whether uprobe event specified */
800                 if (strchr(argv[1], '/') && strchr(argv[1], ':')) {
801                         ret = -ECANCELED;
802                         goto error;
803                 }
804                 /* a symbol specified */
805                 symbol = kstrdup(argv[1], GFP_KERNEL);
806                 if (!symbol)
807                         return -ENOMEM;
808                 /* TODO: support .init module functions */
809                 ret = traceprobe_split_symbol_offset(symbol, &offset);
810                 if (ret || offset < 0 || offset > UINT_MAX) {
811                         trace_probe_log_err(0, BAD_PROBE_ADDR);
812                         goto parse_error;
813                 }
814                 if (kprobe_on_func_entry(NULL, symbol, offset))
815                         flags |= TPARG_FL_FENTRY;
816                 if (offset && is_return && !(flags & TPARG_FL_FENTRY)) {
817                         trace_probe_log_err(0, BAD_RETPROBE);
818                         goto parse_error;
819                 }
820         }
821 
822         trace_probe_log_set_index(0);
823         if (event) {
824                 ret = traceprobe_parse_event_name(&event, &group, buf,
825                                                   event - argv[0]);
826                 if (ret)
827                         goto parse_error;
828         } else {
829                 /* Make a new event name */
830                 if (symbol)
831                         snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
832                                  is_return ? 'r' : 'p', symbol, offset);
833                 else
834                         snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
835                                  is_return ? 'r' : 'p', addr);
836                 sanitize_event_name(buf);
837                 event = buf;
838         }
839 
840         /* setup a probe */
841         tk = alloc_trace_kprobe(group, event, addr, symbol, offset, maxactive,
842                                argc - 2, is_return);
843         if (IS_ERR(tk)) {
844                 ret = PTR_ERR(tk);
845                 /* This must return -ENOMEM, else there is a bug */
846                 WARN_ON_ONCE(ret != -ENOMEM);
847                 goto out;       /* We know tk is not allocated */
848         }
849         argc -= 2; argv += 2;
850 
851         /* parse arguments */
852         for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
853                 tmp = kstrdup(argv[i], GFP_KERNEL);
854                 if (!tmp) {
855                         ret = -ENOMEM;
856                         goto error;
857                 }
858 
859                 trace_probe_log_set_index(i + 2);
860                 ret = traceprobe_parse_probe_arg(&tk->tp, i, tmp, flags);
861                 kfree(tmp);
862                 if (ret)
863                         goto error;     /* This can be -ENOMEM */
864         }
865 
866         ret = traceprobe_set_print_fmt(&tk->tp, is_return);
867         if (ret < 0)
868                 goto error;
869 
870         ret = register_trace_kprobe(tk);
871         if (ret) {
872                 trace_probe_log_set_index(1);
873                 if (ret == -EILSEQ)
874                         trace_probe_log_err(0, BAD_INSN_BNDRY);
875                 else if (ret == -ENOENT)
876                         trace_probe_log_err(0, BAD_PROBE_ADDR);
877                 else if (ret != -ENOMEM && ret != -EEXIST)
878                         trace_probe_log_err(0, FAIL_REG_PROBE);
879                 goto error;
880         }
881 
882 out:
883         trace_probe_log_clear();
884         kfree(symbol);
885         return ret;
886 
887 parse_error:
888         ret = -EINVAL;
889 error:
890         free_trace_kprobe(tk);
891         goto out;
892 }
893 
894 static int create_or_delete_trace_kprobe(int argc, char **argv)
895 {
896         int ret;
897 
898         if (argv[0][0] == '-')
899                 return dyn_event_release(argc, argv, &trace_kprobe_ops);
900 
901         ret = trace_kprobe_create(argc, (const char **)argv);
902         return ret == -ECANCELED ? -EINVAL : ret;
903 }
904 
905 static int trace_kprobe_release(struct dyn_event *ev)
906 {
907         struct trace_kprobe *tk = to_trace_kprobe(ev);
908         int ret = unregister_trace_kprobe(tk);
909 
910         if (!ret)
911                 free_trace_kprobe(tk);
912         return ret;
913 }
914 
915 static int trace_kprobe_show(struct seq_file *m, struct dyn_event *ev)
916 {
917         struct trace_kprobe *tk = to_trace_kprobe(ev);
918         int i;
919 
920         seq_putc(m, trace_kprobe_is_return(tk) ? 'r' : 'p');
921         if (trace_kprobe_is_return(tk) && tk->rp.maxactive)
922                 seq_printf(m, "%d", tk->rp.maxactive);
923         seq_printf(m, ":%s/%s", trace_probe_group_name(&tk->tp),
924                                 trace_probe_name(&tk->tp));
925 
926         if (!tk->symbol)
927                 seq_printf(m, " 0x%p", tk->rp.kp.addr);
928         else if (tk->rp.kp.offset)
929                 seq_printf(m, " %s+%u", trace_kprobe_symbol(tk),
930                            tk->rp.kp.offset);
931         else
932                 seq_printf(m, " %s", trace_kprobe_symbol(tk));
933 
934         for (i = 0; i < tk->tp.nr_args; i++)
935                 seq_printf(m, " %s=%s", tk->tp.args[i].name, tk->tp.args[i].comm);
936         seq_putc(m, '\n');
937 
938         return 0;
939 }
940 
941 static int probes_seq_show(struct seq_file *m, void *v)
942 {
943         struct dyn_event *ev = v;
944 
945         if (!is_trace_kprobe(ev))
946                 return 0;
947 
948         return trace_kprobe_show(m, ev);
949 }
950 
951 static const struct seq_operations probes_seq_op = {
952         .start  = dyn_event_seq_start,
953         .next   = dyn_event_seq_next,
954         .stop   = dyn_event_seq_stop,
955         .show   = probes_seq_show
956 };
957 
958 static int probes_open(struct inode *inode, struct file *file)
959 {
960         int ret;
961 
962         ret = security_locked_down(LOCKDOWN_TRACEFS);
963         if (ret)
964                 return ret;
965 
966         if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
967                 ret = dyn_events_release_all(&trace_kprobe_ops);
968                 if (ret < 0)
969                         return ret;
970         }
971 
972         return seq_open(file, &probes_seq_op);
973 }
974 
975 static ssize_t probes_write(struct file *file, const char __user *buffer,
976                             size_t count, loff_t *ppos)
977 {
978         return trace_parse_run_command(file, buffer, count, ppos,
979                                        create_or_delete_trace_kprobe);
980 }
981 
982 static const struct file_operations kprobe_events_ops = {
983         .owner          = THIS_MODULE,
984         .open           = probes_open,
985         .read           = seq_read,
986         .llseek         = seq_lseek,
987         .release        = seq_release,
988         .write          = probes_write,
989 };
990 
991 /* Probes profiling interfaces */
992 static int probes_profile_seq_show(struct seq_file *m, void *v)
993 {
994         struct dyn_event *ev = v;
995         struct trace_kprobe *tk;
996 
997         if (!is_trace_kprobe(ev))
998                 return 0;
999 
1000         tk = to_trace_kprobe(ev);
1001         seq_printf(m, "  %-44s %15lu %15lu\n",
1002                    trace_probe_name(&tk->tp),
1003                    trace_kprobe_nhit(tk),
1004                    tk->rp.kp.nmissed);
1005 
1006         return 0;
1007 }
1008 
1009 static const struct seq_operations profile_seq_op = {
1010         .start  = dyn_event_seq_start,
1011         .next   = dyn_event_seq_next,
1012         .stop   = dyn_event_seq_stop,
1013         .show   = probes_profile_seq_show
1014 };
1015 
1016 static int profile_open(struct inode *inode, struct file *file)
1017 {
1018         int ret;
1019 
1020         ret = security_locked_down(LOCKDOWN_TRACEFS);
1021         if (ret)
1022                 return ret;
1023 
1024         return seq_open(file, &profile_seq_op);
1025 }
1026 
1027 static const struct file_operations kprobe_profile_ops = {
1028         .owner          = THIS_MODULE,
1029         .open           = profile_open,
1030         .read           = seq_read,
1031         .llseek         = seq_lseek,
1032         .release        = seq_release,
1033 };
1034 
1035 /* Kprobe specific fetch functions */
1036 
1037 /* Return the length of string -- including null terminal byte */
1038 static nokprobe_inline int
1039 fetch_store_strlen(unsigned long addr)
1040 {
1041         int ret, len = 0;
1042         u8 c;
1043 
1044         do {
1045                 ret = probe_kernel_read(&c, (u8 *)addr + len, 1);
1046                 len++;
1047         } while (c && ret == 0 && len < MAX_STRING_SIZE);
1048 
1049         return (ret < 0) ? ret : len;
1050 }
1051 
1052 /* Return the length of string -- including null terminal byte */
1053 static nokprobe_inline int
1054 fetch_store_strlen_user(unsigned long addr)
1055 {
1056         const void __user *uaddr =  (__force const void __user *)addr;
1057 
1058         return strnlen_unsafe_user(uaddr, MAX_STRING_SIZE);
1059 }
1060 
1061 /*
1062  * Fetch a null-terminated string. Caller MUST set *(u32 *)buf with max
1063  * length and relative data location.
1064  */
1065 static nokprobe_inline int
1066 fetch_store_string(unsigned long addr, void *dest, void *base)
1067 {
1068         int maxlen = get_loc_len(*(u32 *)dest);
1069         void *__dest;
1070         long ret;
1071 
1072         if (unlikely(!maxlen))
1073                 return -ENOMEM;
1074 
1075         __dest = get_loc_data(dest, base);
1076 
1077         /*
1078          * Try to get string again, since the string can be changed while
1079          * probing.
1080          */
1081         ret = strncpy_from_unsafe(__dest, (void *)addr, maxlen);
1082         if (ret >= 0)
1083                 *(u32 *)dest = make_data_loc(ret, __dest - base);
1084 
1085         return ret;
1086 }
1087 
1088 /*
1089  * Fetch a null-terminated string from user. Caller MUST set *(u32 *)buf
1090  * with max length and relative data location.
1091  */
1092 static nokprobe_inline int
1093 fetch_store_string_user(unsigned long addr, void *dest, void *base)
1094 {
1095         const void __user *uaddr =  (__force const void __user *)addr;
1096         int maxlen = get_loc_len(*(u32 *)dest);
1097         void *__dest;
1098         long ret;
1099 
1100         if (unlikely(!maxlen))
1101                 return -ENOMEM;
1102 
1103         __dest = get_loc_data(dest, base);
1104 
1105         ret = strncpy_from_unsafe_user(__dest, uaddr, maxlen);
1106         if (ret >= 0)
1107                 *(u32 *)dest = make_data_loc(ret, __dest - base);
1108 
1109         return ret;
1110 }
1111 
1112 static nokprobe_inline int
1113 probe_mem_read(void *dest, void *src, size_t size)
1114 {
1115         return probe_kernel_read(dest, src, size);
1116 }
1117 
1118 static nokprobe_inline int
1119 probe_mem_read_user(void *dest, void *src, size_t size)
1120 {
1121         const void __user *uaddr =  (__force const void __user *)src;
1122 
1123         return probe_user_read(dest, uaddr, size);
1124 }
1125 
1126 /* Note that we don't verify it, since the code does not come from user space */
1127 static int
1128 process_fetch_insn(struct fetch_insn *code, struct pt_regs *regs, void *dest,
1129                    void *base)
1130 {
1131         unsigned long val;
1132 
1133 retry:
1134         /* 1st stage: get value from context */
1135         switch (code->op) {
1136         case FETCH_OP_REG:
1137                 val = regs_get_register(regs, code->param);
1138                 break;
1139         case FETCH_OP_STACK:
1140                 val = regs_get_kernel_stack_nth(regs, code->param);
1141                 break;
1142         case FETCH_OP_STACKP:
1143                 val = kernel_stack_pointer(regs);
1144                 break;
1145         case FETCH_OP_RETVAL:
1146                 val = regs_return_value(regs);
1147                 break;
1148         case FETCH_OP_IMM:
1149                 val = code->immediate;
1150                 break;
1151         case FETCH_OP_COMM:
1152                 val = (unsigned long)current->comm;
1153                 break;
1154         case FETCH_OP_DATA:
1155                 val = (unsigned long)code->data;
1156                 break;
1157 #ifdef CONFIG_HAVE_FUNCTION_ARG_ACCESS_API
1158         case FETCH_OP_ARG:
1159                 val = regs_get_kernel_argument(regs, code->param);
1160                 break;
1161 #endif
1162         case FETCH_NOP_SYMBOL:  /* Ignore a place holder */
1163                 code++;
1164                 goto retry;
1165         default:
1166                 return -EILSEQ;
1167         }
1168         code++;
1169 
1170         return process_fetch_insn_bottom(code, val, dest, base);
1171 }
1172 NOKPROBE_SYMBOL(process_fetch_insn)
1173 
1174 /* Kprobe handler */
1175 static nokprobe_inline void
1176 __kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs,
1177                     struct trace_event_file *trace_file)
1178 {
1179         struct kprobe_trace_entry_head *entry;
1180         struct ring_buffer_event *event;
1181         struct ring_buffer *buffer;
1182         int size, dsize, pc;
1183         unsigned long irq_flags;
1184         struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1185 
1186         WARN_ON(call != trace_file->event_call);
1187 
1188         if (trace_trigger_soft_disabled(trace_file))
1189                 return;
1190 
1191         local_save_flags(irq_flags);
1192         pc = preempt_count();
1193 
1194         dsize = __get_data_size(&tk->tp, regs);
1195         size = sizeof(*entry) + tk->tp.size + dsize;
1196 
1197         event = trace_event_buffer_lock_reserve(&buffer, trace_file,
1198                                                 call->event.type,
1199                                                 size, irq_flags, pc);
1200         if (!event)
1201                 return;
1202 
1203         entry = ring_buffer_event_data(event);
1204         entry->ip = (unsigned long)tk->rp.kp.addr;
1205         store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1206 
1207         event_trigger_unlock_commit_regs(trace_file, buffer, event,
1208                                          entry, irq_flags, pc, regs);
1209 }
1210 
1211 static void
1212 kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs)
1213 {
1214         struct event_file_link *link;
1215 
1216         trace_probe_for_each_link_rcu(link, &tk->tp)
1217                 __kprobe_trace_func(tk, regs, link->file);
1218 }
1219 NOKPROBE_SYMBOL(kprobe_trace_func);
1220 
1221 /* Kretprobe handler */
1222 static nokprobe_inline void
1223 __kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1224                        struct pt_regs *regs,
1225                        struct trace_event_file *trace_file)
1226 {
1227         struct kretprobe_trace_entry_head *entry;
1228         struct ring_buffer_event *event;
1229         struct ring_buffer *buffer;
1230         int size, pc, dsize;
1231         unsigned long irq_flags;
1232         struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1233 
1234         WARN_ON(call != trace_file->event_call);
1235 
1236         if (trace_trigger_soft_disabled(trace_file))
1237                 return;
1238 
1239         local_save_flags(irq_flags);
1240         pc = preempt_count();
1241 
1242         dsize = __get_data_size(&tk->tp, regs);
1243         size = sizeof(*entry) + tk->tp.size + dsize;
1244 
1245         event = trace_event_buffer_lock_reserve(&buffer, trace_file,
1246                                                 call->event.type,
1247                                                 size, irq_flags, pc);
1248         if (!event)
1249                 return;
1250 
1251         entry = ring_buffer_event_data(event);
1252         entry->func = (unsigned long)tk->rp.kp.addr;
1253         entry->ret_ip = (unsigned long)ri->ret_addr;
1254         store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1255 
1256         event_trigger_unlock_commit_regs(trace_file, buffer, event,
1257                                          entry, irq_flags, pc, regs);
1258 }
1259 
1260 static void
1261 kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1262                      struct pt_regs *regs)
1263 {
1264         struct event_file_link *link;
1265 
1266         trace_probe_for_each_link_rcu(link, &tk->tp)
1267                 __kretprobe_trace_func(tk, ri, regs, link->file);
1268 }
1269 NOKPROBE_SYMBOL(kretprobe_trace_func);
1270 
1271 /* Event entry printers */
1272 static enum print_line_t
1273 print_kprobe_event(struct trace_iterator *iter, int flags,
1274                    struct trace_event *event)
1275 {
1276         struct kprobe_trace_entry_head *field;
1277         struct trace_seq *s = &iter->seq;
1278         struct trace_probe *tp;
1279 
1280         field = (struct kprobe_trace_entry_head *)iter->ent;
1281         tp = trace_probe_primary_from_call(
1282                 container_of(event, struct trace_event_call, event));
1283         if (WARN_ON_ONCE(!tp))
1284                 goto out;
1285 
1286         trace_seq_printf(s, "%s: (", trace_probe_name(tp));
1287 
1288         if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
1289                 goto out;
1290 
1291         trace_seq_putc(s, ')');
1292 
1293         if (print_probe_args(s, tp->args, tp->nr_args,
1294                              (u8 *)&field[1], field) < 0)
1295                 goto out;
1296 
1297         trace_seq_putc(s, '\n');
1298  out:
1299         return trace_handle_return(s);
1300 }
1301 
1302 static enum print_line_t
1303 print_kretprobe_event(struct trace_iterator *iter, int flags,
1304                       struct trace_event *event)
1305 {
1306         struct kretprobe_trace_entry_head *field;
1307         struct trace_seq *s = &iter->seq;
1308         struct trace_probe *tp;
1309 
1310         field = (struct kretprobe_trace_entry_head *)iter->ent;
1311         tp = trace_probe_primary_from_call(
1312                 container_of(event, struct trace_event_call, event));
1313         if (WARN_ON_ONCE(!tp))
1314                 goto out;
1315 
1316         trace_seq_printf(s, "%s: (", trace_probe_name(tp));
1317 
1318         if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1319                 goto out;
1320 
1321         trace_seq_puts(s, " <- ");
1322 
1323         if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1324                 goto out;
1325 
1326         trace_seq_putc(s, ')');
1327 
1328         if (print_probe_args(s, tp->args, tp->nr_args,
1329                              (u8 *)&field[1], field) < 0)
1330                 goto out;
1331 
1332         trace_seq_putc(s, '\n');
1333 
1334  out:
1335         return trace_handle_return(s);
1336 }
1337 
1338 
1339 static int kprobe_event_define_fields(struct trace_event_call *event_call)
1340 {
1341         int ret;
1342         struct kprobe_trace_entry_head field;
1343         struct trace_probe *tp;
1344 
1345         tp = trace_probe_primary_from_call(event_call);
1346         if (WARN_ON_ONCE(!tp))
1347                 return -ENOENT;
1348 
1349         DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1350 
1351         return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
1352 }
1353 
1354 static int kretprobe_event_define_fields(struct trace_event_call *event_call)
1355 {
1356         int ret;
1357         struct kretprobe_trace_entry_head field;
1358         struct trace_probe *tp;
1359 
1360         tp = trace_probe_primary_from_call(event_call);
1361         if (WARN_ON_ONCE(!tp))
1362                 return -ENOENT;
1363 
1364         DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1365         DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1366 
1367         return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
1368 }
1369 
1370 #ifdef CONFIG_PERF_EVENTS
1371 
1372 /* Kprobe profile handler */
1373 static int
1374 kprobe_perf_func(struct trace_kprobe *tk, struct pt_regs *regs)
1375 {
1376         struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1377         struct kprobe_trace_entry_head *entry;
1378         struct hlist_head *head;
1379         int size, __size, dsize;
1380         int rctx;
1381 
1382         if (bpf_prog_array_valid(call)) {
1383                 unsigned long orig_ip = instruction_pointer(regs);
1384                 int ret;
1385 
1386                 ret = trace_call_bpf(call, regs);
1387 
1388                 /*
1389                  * We need to check and see if we modified the pc of the
1390                  * pt_regs, and if so return 1 so that we don't do the
1391                  * single stepping.
1392                  */
1393                 if (orig_ip != instruction_pointer(regs))
1394                         return 1;
1395                 if (!ret)
1396                         return 0;
1397         }
1398 
1399         head = this_cpu_ptr(call->perf_events);
1400         if (hlist_empty(head))
1401                 return 0;
1402 
1403         dsize = __get_data_size(&tk->tp, regs);
1404         __size = sizeof(*entry) + tk->tp.size + dsize;
1405         size = ALIGN(__size + sizeof(u32), sizeof(u64));
1406         size -= sizeof(u32);
1407 
1408         entry = perf_trace_buf_alloc(size, NULL, &rctx);
1409         if (!entry)
1410                 return 0;
1411 
1412         entry->ip = (unsigned long)tk->rp.kp.addr;
1413         memset(&entry[1], 0, dsize);
1414         store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1415         perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1416                               head, NULL);
1417         return 0;
1418 }
1419 NOKPROBE_SYMBOL(kprobe_perf_func);
1420 
1421 /* Kretprobe profile handler */
1422 static void
1423 kretprobe_perf_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1424                     struct pt_regs *regs)
1425 {
1426         struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1427         struct kretprobe_trace_entry_head *entry;
1428         struct hlist_head *head;
1429         int size, __size, dsize;
1430         int rctx;
1431 
1432         if (bpf_prog_array_valid(call) && !trace_call_bpf(call, regs))
1433                 return;
1434 
1435         head = this_cpu_ptr(call->perf_events);
1436         if (hlist_empty(head))
1437                 return;
1438 
1439         dsize = __get_data_size(&tk->tp, regs);
1440         __size = sizeof(*entry) + tk->tp.size + dsize;
1441         size = ALIGN(__size + sizeof(u32), sizeof(u64));
1442         size -= sizeof(u32);
1443 
1444         entry = perf_trace_buf_alloc(size, NULL, &rctx);
1445         if (!entry)
1446                 return;
1447 
1448         entry->func = (unsigned long)tk->rp.kp.addr;
1449         entry->ret_ip = (unsigned long)ri->ret_addr;
1450         store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1451         perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1452                               head, NULL);
1453 }
1454 NOKPROBE_SYMBOL(kretprobe_perf_func);
1455 
1456 int bpf_get_kprobe_info(const struct perf_event *event, u32 *fd_type,
1457                         const char **symbol, u64 *probe_offset,
1458                         u64 *probe_addr, bool perf_type_tracepoint)
1459 {
1460         const char *pevent = trace_event_name(event->tp_event);
1461         const char *group = event->tp_event->class->system;
1462         struct trace_kprobe *tk;
1463 
1464         if (perf_type_tracepoint)
1465                 tk = find_trace_kprobe(pevent, group);
1466         else
1467                 tk = event->tp_event->data;
1468         if (!tk)
1469                 return -EINVAL;
1470 
1471         *fd_type = trace_kprobe_is_return(tk) ? BPF_FD_TYPE_KRETPROBE
1472                                               : BPF_FD_TYPE_KPROBE;
1473         if (tk->symbol) {
1474                 *symbol = tk->symbol;
1475                 *probe_offset = tk->rp.kp.offset;
1476                 *probe_addr = 0;
1477         } else {
1478                 *symbol = NULL;
1479                 *probe_offset = 0;
1480                 *probe_addr = (unsigned long)tk->rp.kp.addr;
1481         }
1482         return 0;
1483 }
1484 #endif  /* CONFIG_PERF_EVENTS */
1485 
1486 /*
1487  * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex.
1488  *
1489  * kprobe_trace_self_tests_init() does enable_trace_probe/disable_trace_probe
1490  * lockless, but we can't race with this __init function.
1491  */
1492 static int kprobe_register(struct trace_event_call *event,
1493                            enum trace_reg type, void *data)
1494 {
1495         struct trace_event_file *file = data;
1496 
1497         switch (type) {
1498         case TRACE_REG_REGISTER:
1499                 return enable_trace_kprobe(event, file);
1500         case TRACE_REG_UNREGISTER:
1501                 return disable_trace_kprobe(event, file);
1502 
1503 #ifdef CONFIG_PERF_EVENTS
1504         case TRACE_REG_PERF_REGISTER:
1505                 return enable_trace_kprobe(event, NULL);
1506         case TRACE_REG_PERF_UNREGISTER:
1507                 return disable_trace_kprobe(event, NULL);
1508         case TRACE_REG_PERF_OPEN:
1509         case TRACE_REG_PERF_CLOSE:
1510         case TRACE_REG_PERF_ADD:
1511         case TRACE_REG_PERF_DEL:
1512                 return 0;
1513 #endif
1514         }
1515         return 0;
1516 }
1517 
1518 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1519 {
1520         struct trace_kprobe *tk = container_of(kp, struct trace_kprobe, rp.kp);
1521         int ret = 0;
1522 
1523         raw_cpu_inc(*tk->nhit);
1524 
1525         if (trace_probe_test_flag(&tk->tp, TP_FLAG_TRACE))
1526                 kprobe_trace_func(tk, regs);
1527 #ifdef CONFIG_PERF_EVENTS
1528         if (trace_probe_test_flag(&tk->tp, TP_FLAG_PROFILE))
1529                 ret = kprobe_perf_func(tk, regs);
1530 #endif
1531         return ret;
1532 }
1533 NOKPROBE_SYMBOL(kprobe_dispatcher);
1534 
1535 static int
1536 kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1537 {
1538         struct trace_kprobe *tk = container_of(ri->rp, struct trace_kprobe, rp);
1539 
1540         raw_cpu_inc(*tk->nhit);
1541 
1542         if (trace_probe_test_flag(&tk->tp, TP_FLAG_TRACE))
1543                 kretprobe_trace_func(tk, ri, regs);
1544 #ifdef CONFIG_PERF_EVENTS
1545         if (trace_probe_test_flag(&tk->tp, TP_FLAG_PROFILE))
1546                 kretprobe_perf_func(tk, ri, regs);
1547 #endif
1548         return 0;       /* We don't tweek kernel, so just return 0 */
1549 }
1550 NOKPROBE_SYMBOL(kretprobe_dispatcher);
1551 
1552 static struct trace_event_functions kretprobe_funcs = {
1553         .trace          = print_kretprobe_event
1554 };
1555 
1556 static struct trace_event_functions kprobe_funcs = {
1557         .trace          = print_kprobe_event
1558 };
1559 
1560 static inline void init_trace_event_call(struct trace_kprobe *tk)
1561 {
1562         struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1563 
1564         if (trace_kprobe_is_return(tk)) {
1565                 call->event.funcs = &kretprobe_funcs;
1566                 call->class->define_fields = kretprobe_event_define_fields;
1567         } else {
1568                 call->event.funcs = &kprobe_funcs;
1569                 call->class->define_fields = kprobe_event_define_fields;
1570         }
1571 
1572         call->flags = TRACE_EVENT_FL_KPROBE;
1573         call->class->reg = kprobe_register;
1574 }
1575 
1576 static int register_kprobe_event(struct trace_kprobe *tk)
1577 {
1578         init_trace_event_call(tk);
1579 
1580         return trace_probe_register_event_call(&tk->tp);
1581 }
1582 
1583 static int unregister_kprobe_event(struct trace_kprobe *tk)
1584 {
1585         return trace_probe_unregister_event_call(&tk->tp);
1586 }
1587 
1588 #ifdef CONFIG_PERF_EVENTS
1589 /* create a trace_kprobe, but don't add it to global lists */
1590 struct trace_event_call *
1591 create_local_trace_kprobe(char *func, void *addr, unsigned long offs,
1592                           bool is_return)
1593 {
1594         struct trace_kprobe *tk;
1595         int ret;
1596         char *event;
1597 
1598         /*
1599          * local trace_kprobes are not added to dyn_event, so they are never
1600          * searched in find_trace_kprobe(). Therefore, there is no concern of
1601          * duplicated name here.
1602          */
1603         event = func ? func : "DUMMY_EVENT";
1604 
1605         tk = alloc_trace_kprobe(KPROBE_EVENT_SYSTEM, event, (void *)addr, func,
1606                                 offs, 0 /* maxactive */, 0 /* nargs */,
1607                                 is_return);
1608 
1609         if (IS_ERR(tk)) {
1610                 pr_info("Failed to allocate trace_probe.(%d)\n",
1611                         (int)PTR_ERR(tk));
1612                 return ERR_CAST(tk);
1613         }
1614 
1615         init_trace_event_call(tk);
1616 
1617         if (traceprobe_set_print_fmt(&tk->tp, trace_kprobe_is_return(tk)) < 0) {
1618                 ret = -ENOMEM;
1619                 goto error;
1620         }
1621 
1622         ret = __register_trace_kprobe(tk);
1623         if (ret < 0)
1624                 goto error;
1625 
1626         return trace_probe_event_call(&tk->tp);
1627 error:
1628         free_trace_kprobe(tk);
1629         return ERR_PTR(ret);
1630 }
1631 
1632 void destroy_local_trace_kprobe(struct trace_event_call *event_call)
1633 {
1634         struct trace_kprobe *tk;
1635 
1636         tk = trace_kprobe_primary_from_call(event_call);
1637         if (unlikely(!tk))
1638                 return;
1639 
1640         if (trace_probe_is_enabled(&tk->tp)) {
1641                 WARN_ON(1);
1642                 return;
1643         }
1644 
1645         __unregister_trace_kprobe(tk);
1646 
1647         free_trace_kprobe(tk);
1648 }
1649 #endif /* CONFIG_PERF_EVENTS */
1650 
1651 static __init void enable_boot_kprobe_events(void)
1652 {
1653         struct trace_array *tr = top_trace_array();
1654         struct trace_event_file *file;
1655         struct trace_kprobe *tk;
1656         struct dyn_event *pos;
1657 
1658         mutex_lock(&event_mutex);
1659         for_each_trace_kprobe(tk, pos) {
1660                 list_for_each_entry(file, &tr->events, list)
1661                         if (file->event_call == trace_probe_event_call(&tk->tp))
1662                                 trace_event_enable_disable(file, 1, 0);
1663         }
1664         mutex_unlock(&event_mutex);
1665 }
1666 
1667 static __init void setup_boot_kprobe_events(void)
1668 {
1669         char *p, *cmd = kprobe_boot_events_buf;
1670         int ret;
1671 
1672         strreplace(kprobe_boot_events_buf, ',', ' ');
1673 
1674         while (cmd && *cmd != '\0') {
1675                 p = strchr(cmd, ';');
1676                 if (p)
1677                         *p++ = '\0';
1678 
1679                 ret = trace_run_command(cmd, create_or_delete_trace_kprobe);
1680                 if (ret)
1681                         pr_warn("Failed to add event(%d): %s\n", ret, cmd);
1682                 else
1683                         kprobe_boot_events_enabled = true;
1684 
1685                 cmd = p;
1686         }
1687 
1688         enable_boot_kprobe_events();
1689 }
1690 
1691 /* Make a tracefs interface for controlling probe points */
1692 static __init int init_kprobe_trace(void)
1693 {
1694         struct dentry *d_tracer;
1695         struct dentry *entry;
1696         int ret;
1697 
1698         ret = dyn_event_register(&trace_kprobe_ops);
1699         if (ret)
1700                 return ret;
1701 
1702         if (register_module_notifier(&trace_kprobe_module_nb))
1703                 return -EINVAL;
1704 
1705         d_tracer = tracing_init_dentry();
1706         if (IS_ERR(d_tracer))
1707                 return 0;
1708 
1709         entry = tracefs_create_file("kprobe_events", 0644, d_tracer,
1710                                     NULL, &kprobe_events_ops);
1711 
1712         /* Event list interface */
1713         if (!entry)
1714                 pr_warn("Could not create tracefs 'kprobe_events' entry\n");
1715 
1716         /* Profile interface */
1717         entry = tracefs_create_file("kprobe_profile", 0444, d_tracer,
1718                                     NULL, &kprobe_profile_ops);
1719 
1720         if (!entry)
1721                 pr_warn("Could not create tracefs 'kprobe_profile' entry\n");
1722 
1723         setup_boot_kprobe_events();
1724 
1725         return 0;
1726 }
1727 fs_initcall(init_kprobe_trace);
1728 
1729 
1730 #ifdef CONFIG_FTRACE_STARTUP_TEST
1731 static __init struct trace_event_file *
1732 find_trace_probe_file(struct trace_kprobe *tk, struct trace_array *tr)
1733 {
1734         struct trace_event_file *file;
1735 
1736         list_for_each_entry(file, &tr->events, list)
1737                 if (file->event_call == trace_probe_event_call(&tk->tp))
1738                         return file;
1739 
1740         return NULL;
1741 }
1742 
1743 /*
1744  * Nobody but us can call enable_trace_kprobe/disable_trace_kprobe at this
1745  * stage, we can do this lockless.
1746  */
1747 static __init int kprobe_trace_self_tests_init(void)
1748 {
1749         int ret, warn = 0;
1750         int (*target)(int, int, int, int, int, int);
1751         struct trace_kprobe *tk;
1752         struct trace_event_file *file;
1753 
1754         if (tracing_is_disabled())
1755                 return -ENODEV;
1756 
1757         if (kprobe_boot_events_enabled) {
1758                 pr_info("Skipping kprobe tests due to kprobe_event on cmdline\n");
1759                 return 0;
1760         }
1761 
1762         target = kprobe_trace_selftest_target;
1763 
1764         pr_info("Testing kprobe tracing: ");
1765 
1766         ret = trace_run_command("p:testprobe kprobe_trace_selftest_target $stack $stack0 +0($stack)",
1767                                 create_or_delete_trace_kprobe);
1768         if (WARN_ON_ONCE(ret)) {
1769                 pr_warn("error on probing function entry.\n");
1770                 warn++;
1771         } else {
1772                 /* Enable trace point */
1773                 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1774                 if (WARN_ON_ONCE(tk == NULL)) {
1775                         pr_warn("error on getting new probe.\n");
1776                         warn++;
1777                 } else {
1778                         file = find_trace_probe_file(tk, top_trace_array());
1779                         if (WARN_ON_ONCE(file == NULL)) {
1780                                 pr_warn("error on getting probe file.\n");
1781                                 warn++;
1782                         } else
1783                                 enable_trace_kprobe(
1784                                         trace_probe_event_call(&tk->tp), file);
1785                 }
1786         }
1787 
1788         ret = trace_run_command("r:testprobe2 kprobe_trace_selftest_target $retval",
1789                                 create_or_delete_trace_kprobe);
1790         if (WARN_ON_ONCE(ret)) {
1791                 pr_warn("error on probing function return.\n");
1792                 warn++;
1793         } else {
1794                 /* Enable trace point */
1795                 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1796                 if (WARN_ON_ONCE(tk == NULL)) {
1797                         pr_warn("error on getting 2nd new probe.\n");
1798                         warn++;
1799                 } else {
1800                         file = find_trace_probe_file(tk, top_trace_array());
1801                         if (WARN_ON_ONCE(file == NULL)) {
1802                                 pr_warn("error on getting probe file.\n");
1803                                 warn++;
1804                         } else
1805                                 enable_trace_kprobe(
1806                                         trace_probe_event_call(&tk->tp), file);
1807                 }
1808         }
1809 
1810         if (warn)
1811                 goto end;
1812 
1813         ret = target(1, 2, 3, 4, 5, 6);
1814 
1815         /*
1816          * Not expecting an error here, the check is only to prevent the
1817          * optimizer from removing the call to target() as otherwise there
1818          * are no side-effects and the call is never performed.
1819          */
1820         if (ret != 21)
1821                 warn++;
1822 
1823         /* Disable trace points before removing it */
1824         tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1825         if (WARN_ON_ONCE(tk == NULL)) {
1826                 pr_warn("error on getting test probe.\n");
1827                 warn++;
1828         } else {
1829                 if (trace_kprobe_nhit(tk) != 1) {
1830                         pr_warn("incorrect number of testprobe hits\n");
1831                         warn++;
1832                 }
1833 
1834                 file = find_trace_probe_file(tk, top_trace_array());
1835                 if (WARN_ON_ONCE(file == NULL)) {
1836                         pr_warn("error on getting probe file.\n");
1837                         warn++;
1838                 } else
1839                         disable_trace_kprobe(
1840                                 trace_probe_event_call(&tk->tp), file);
1841         }
1842 
1843         tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1844         if (WARN_ON_ONCE(tk == NULL)) {
1845                 pr_warn("error on getting 2nd test probe.\n");
1846                 warn++;
1847         } else {
1848                 if (trace_kprobe_nhit(tk) != 1) {
1849                         pr_warn("incorrect number of testprobe2 hits\n");
1850                         warn++;
1851                 }
1852 
1853                 file = find_trace_probe_file(tk, top_trace_array());
1854                 if (WARN_ON_ONCE(file == NULL)) {
1855                         pr_warn("error on getting probe file.\n");
1856                         warn++;
1857                 } else
1858                         disable_trace_kprobe(
1859                                 trace_probe_event_call(&tk->tp), file);
1860         }
1861 
1862         ret = trace_run_command("-:testprobe", create_or_delete_trace_kprobe);
1863         if (WARN_ON_ONCE(ret)) {
1864                 pr_warn("error on deleting a probe.\n");
1865                 warn++;
1866         }
1867 
1868         ret = trace_run_command("-:testprobe2", create_or_delete_trace_kprobe);
1869         if (WARN_ON_ONCE(ret)) {
1870                 pr_warn("error on deleting a probe.\n");
1871                 warn++;
1872         }
1873 
1874 end:
1875         ret = dyn_events_release_all(&trace_kprobe_ops);
1876         if (WARN_ON_ONCE(ret)) {
1877                 pr_warn("error on cleaning up probes.\n");
1878                 warn++;
1879         }
1880         /*
1881          * Wait for the optimizer work to finish. Otherwise it might fiddle
1882          * with probes in already freed __init text.
1883          */
1884         wait_for_kprobe_optimizer();
1885         if (warn)
1886                 pr_cont("NG: Some tests are failed. Please check them.\n");
1887         else
1888                 pr_cont("OK\n");
1889         return 0;
1890 }
1891 
1892 late_initcall(kprobe_trace_self_tests_init);
1893 
1894 #endif
1895 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp