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/cleanup.h>
  12 #include <linux/security.h>
  13 #include <linux/module.h>
  14 #include <linux/uaccess.h>
  15 #include <linux/rculist.h>
  16 #include <linux/error-injection.h>
  17
  18 #include <asm/setup.h>  /* for COMMAND_LINE_SIZE */
  19
  20 #include "trace_dynevent.h"
  21 #include "trace_kprobe_selftest.h"
  22 #include "trace_probe.h"
  23 #include "trace_probe_tmpl.h"
  24 #include "trace_probe_kernel.h"
  25
  26 #define KPROBE_EVENT_SYSTEM "kprobes"
  27 #define KRETPROBE_MAXACTIVE_MAX 4096
  28
  29 /* Kprobe early definition from command line */
  30 static char kprobe_boot_events_buf[COMMAND_LINE_SIZE] __initdata;
  31
  32 static int __init set_kprobe_boot_events(char *str)
  33 {
  34         strscpy(kprobe_boot_events_buf, str, COMMAND_LINE_SIZE);
  35         disable_tracing_selftest("running kprobe events");
  36
  37         return 1;
  38 }
  39 __setup("kprobe_event=", set_kprobe_boot_events);
  40
  41 static int trace_kprobe_create(const char *raw_command);
  42 static int trace_kprobe_show(struct seq_file *m, struct dyn_event *ev);
  43 static int trace_kprobe_release(struct dyn_event *ev);
  44 static bool trace_kprobe_is_busy(struct dyn_event *ev);
  45 static bool trace_kprobe_match(const char *system, const char *event,
  46                         int argc, const char **argv, struct dyn_event *ev);
  47
  48 static struct dyn_event_operations trace_kprobe_ops = {
  49         .create = trace_kprobe_create,
  50         .show = trace_kprobe_show,
  51         .is_busy = trace_kprobe_is_busy,
  52         .free = trace_kprobe_release,
  53         .match = trace_kprobe_match,
  54 };
  55
  56 /*
  57  * Kprobe event core functions
  58  */
  59 struct trace_kprobe {
  60         struct dyn_event        devent;
  61         struct kretprobe        rp;     /* Use rp.kp for kprobe use */
  62         unsigned long __percpu *nhit;
  63         const char              *symbol;        /* symbol name */
  64         struct trace_probe      tp;
  65 };
  66
  67 static bool is_trace_kprobe(struct dyn_event *ev)
  68 {
  69         return ev->ops == &trace_kprobe_ops;
  70 }
  71
  72 static struct trace_kprobe *to_trace_kprobe(struct dyn_event *ev)
  73 {
  74         return container_of(ev, struct trace_kprobe, devent);
  75 }
  76
  77 /**
  78  * for_each_trace_kprobe - iterate over the trace_kprobe list
  79  * @pos:        the struct trace_kprobe * for each entry
  80  * @dpos:       the struct dyn_event * to use as a loop cursor
  81  */
  82 #define for_each_trace_kprobe(pos, dpos)        \
  83         for_each_dyn_event(dpos)                \
  84                 if (is_trace_kprobe(dpos) && (pos = to_trace_kprobe(dpos)))
  85
  86 static nokprobe_inline bool trace_kprobe_is_return(struct trace_kprobe *tk)
  87 {
  88         return tk->rp.handler != NULL;
  89 }
  90
  91 static nokprobe_inline const char *trace_kprobe_symbol(struct trace_kprobe *tk)
  92 {
  93         return tk->symbol ? tk->symbol : "unknown";
  94 }
  95
  96 static nokprobe_inline unsigned long trace_kprobe_offset(struct trace_kprobe *tk)
  97 {
  98         return tk->rp.kp.offset;
  99 }
 100
 101 static nokprobe_inline bool trace_kprobe_has_gone(struct trace_kprobe *tk)
 102 {
 103         return kprobe_gone(&tk->rp.kp);
 104 }
 105
 106 static nokprobe_inline bool trace_kprobe_within_module(struct trace_kprobe *tk,
 107                                                  struct module *mod)
 108 {
 109         int len = strlen(module_name(mod));
 110         const char *name = trace_kprobe_symbol(tk);
 111
 112         return strncmp(module_name(mod), name, len) == 0 && name[len] == ':';
 113 }
 114
 115 #ifdef CONFIG_MODULES
 116 static nokprobe_inline bool trace_kprobe_module_exist(struct trace_kprobe *tk)
 117 {
 118         char *p;
 119         bool ret;
 120
 121         if (!tk->symbol)
 122                 return false;
 123         p = strchr(tk->symbol, ':');
 124         if (!p)
 125                 return true;
 126         *p = '\0';
 127         scoped_guard(rcu)
 128                 ret = !!find_module(tk->symbol);
 129         *p = ':';
 130
 131         return ret;
 132 }
 133 #else
 134 static inline bool trace_kprobe_module_exist(struct trace_kprobe *tk)
 135 {
 136         return false;
 137 }
 138 #endif
 139
 140 static bool trace_kprobe_is_busy(struct dyn_event *ev)
 141 {
 142         struct trace_kprobe *tk = to_trace_kprobe(ev);
 143
 144         return trace_probe_is_enabled(&tk->tp);
 145 }
 146
 147 static bool trace_kprobe_match_command_head(struct trace_kprobe *tk,
 148                                             int argc, const char **argv)
 149 {
 150         char buf[MAX_ARGSTR_LEN + 1];
 151
 152         if (!argc)
 153                 return true;
 154
 155         if (!tk->symbol)
 156                 snprintf(buf, sizeof(buf), "0x%p", tk->rp.kp.addr);
 157         else if (tk->rp.kp.offset)
 158                 snprintf(buf, sizeof(buf), "%s+%u",
 159                          trace_kprobe_symbol(tk), tk->rp.kp.offset);
 160         else
 161                 snprintf(buf, sizeof(buf), "%s", trace_kprobe_symbol(tk));
 162         if (strcmp(buf, argv[0]))
 163                 return false;
 164         argc--; argv++;
 165
 166         return trace_probe_match_command_args(&tk->tp, argc, argv);
 167 }
 168
 169 static bool trace_kprobe_match(const char *system, const char *event,
 170                         int argc, const char **argv, struct dyn_event *ev)
 171 {
 172         struct trace_kprobe *tk = to_trace_kprobe(ev);
 173
 174         return (event[0] == '\0' ||
 175                 strcmp(trace_probe_name(&tk->tp), event) == 0) &&
 176             (!system || strcmp(trace_probe_group_name(&tk->tp), system) == 0) &&
 177             trace_kprobe_match_command_head(tk, argc, argv);
 178 }
 179
 180 static nokprobe_inline unsigned long trace_kprobe_nhit(struct trace_kprobe *tk)
 181 {
 182         unsigned long nhit = 0;
 183         int cpu;
 184
 185         for_each_possible_cpu(cpu)
 186                 nhit += *per_cpu_ptr(tk->nhit, cpu);
 187
 188         return nhit;
 189 }
 190
 191 static nokprobe_inline bool trace_kprobe_is_registered(struct trace_kprobe *tk)
 192 {
 193         return !(list_empty(&tk->rp.kp.list) &&
 194                  hlist_unhashed(&tk->rp.kp.hlist));
 195 }
 196
 197 /* Return 0 if it fails to find the symbol address */
 198 static nokprobe_inline
 199 unsigned long trace_kprobe_address(struct trace_kprobe *tk)
 200 {
 201         unsigned long addr;
 202
 203         if (tk->symbol) {
 204                 addr = (unsigned long)
 205                         kallsyms_lookup_name(trace_kprobe_symbol(tk));
 206                 if (addr)
 207                         addr += tk->rp.kp.offset;
 208         } else {
 209                 addr = (unsigned long)tk->rp.kp.addr;
 210         }
 211         return addr;
 212 }
 213
 214 static nokprobe_inline struct trace_kprobe *
 215 trace_kprobe_primary_from_call(struct trace_event_call *call)
 216 {
 217         struct trace_probe *tp;
 218
 219         tp = trace_probe_primary_from_call(call);
 220         if (WARN_ON_ONCE(!tp))
 221                 return NULL;
 222
 223         return container_of(tp, struct trace_kprobe, tp);
 224 }
 225
 226 bool trace_kprobe_on_func_entry(struct trace_event_call *call)
 227 {
 228         struct trace_kprobe *tk = trace_kprobe_primary_from_call(call);
 229
 230         return tk ? (kprobe_on_func_entry(tk->rp.kp.addr,
 231                         tk->rp.kp.addr ? NULL : tk->rp.kp.symbol_name,
 232                         tk->rp.kp.addr ? 0 : tk->rp.kp.offset) == 0) : false;
 233 }
 234
 235 bool trace_kprobe_error_injectable(struct trace_event_call *call)
 236 {
 237         struct trace_kprobe *tk = trace_kprobe_primary_from_call(call);
 238
 239         return tk ? within_error_injection_list(trace_kprobe_address(tk)) :
 240                false;
 241 }
 242
 243 static int register_kprobe_event(struct trace_kprobe *tk);
 244 static int unregister_kprobe_event(struct trace_kprobe *tk);
 245
 246 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
 247 static int kretprobe_dispatcher(struct kretprobe_instance *ri,
 248                                 struct pt_regs *regs);
 249
 250 static void free_trace_kprobe(struct trace_kprobe *tk)
 251 {
 252         if (tk) {
 253                 trace_probe_cleanup(&tk->tp);
 254                 kfree(tk->symbol);
 255                 free_percpu(tk->nhit);
 256                 kfree(tk);
 257         }
 258 }
 259
 260 DEFINE_FREE(free_trace_kprobe, struct trace_kprobe *,
 261         if (!IS_ERR_OR_NULL(_T)) free_trace_kprobe(_T))
 262
 263 /*
 264  * Allocate new trace_probe and initialize it (including kprobes).
 265  */
 266 static struct trace_kprobe *alloc_trace_kprobe(const char *group,
 267                                              const char *event,
 268                                              void *addr,
 269                                              const char *symbol,
 270                                              unsigned long offs,
 271                                              int maxactive,
 272                                              int nargs, bool is_return)
 273 {
 274         struct trace_kprobe *tk __free(free_trace_kprobe) = NULL;
 275         int ret = -ENOMEM;
 276
 277         tk = kzalloc(struct_size(tk, tp.args, nargs), GFP_KERNEL);
 278         if (!tk)
 279                 return ERR_PTR(ret);
 280
 281         tk->nhit = alloc_percpu(unsigned long);
 282         if (!tk->nhit)
 283                 return ERR_PTR(ret);
 284
 285         if (symbol) {
 286                 tk->symbol = kstrdup(symbol, GFP_KERNEL);
 287                 if (!tk->symbol)
 288                         return ERR_PTR(ret);
 289                 tk->rp.kp.symbol_name = tk->symbol;
 290                 tk->rp.kp.offset = offs;
 291         } else
 292                 tk->rp.kp.addr = addr;
 293
 294         if (is_return)
 295                 tk->rp.handler = kretprobe_dispatcher;
 296         else
 297                 tk->rp.kp.pre_handler = kprobe_dispatcher;
 298
 299         tk->rp.maxactive = maxactive;
 300         INIT_HLIST_NODE(&tk->rp.kp.hlist);
 301         INIT_LIST_HEAD(&tk->rp.kp.list);
 302
 303         ret = trace_probe_init(&tk->tp, event, group, false, nargs);
 304         if (ret < 0)
 305                 return ERR_PTR(ret);
 306
 307         dyn_event_init(&tk->devent, &trace_kprobe_ops);
 308         return_ptr(tk);
 309 }
 310
 311 static struct trace_kprobe *find_trace_kprobe(const char *event,
 312                                               const char *group)
 313 {
 314         struct dyn_event *pos;
 315         struct trace_kprobe *tk;
 316
 317         for_each_trace_kprobe(tk, pos)
 318                 if (strcmp(trace_probe_name(&tk->tp), event) == 0 &&
 319                     strcmp(trace_probe_group_name(&tk->tp), group) == 0)
 320                         return tk;
 321         return NULL;
 322 }
 323
 324 static inline int __enable_trace_kprobe(struct trace_kprobe *tk)
 325 {
 326         int ret = 0;
 327
 328         if (trace_kprobe_is_registered(tk) && !trace_kprobe_has_gone(tk)) {
 329                 if (trace_kprobe_is_return(tk))
 330                         ret = enable_kretprobe(&tk->rp);
 331                 else
 332                         ret = enable_kprobe(&tk->rp.kp);
 333         }
 334
 335         return ret;
 336 }
 337
 338 static void __disable_trace_kprobe(struct trace_probe *tp)
 339 {
 340         struct trace_kprobe *tk;
 341
 342         list_for_each_entry(tk, trace_probe_probe_list(tp), tp.list) {
 343                 if (!trace_kprobe_is_registered(tk))
 344                         continue;
 345                 if (trace_kprobe_is_return(tk))
 346                         disable_kretprobe(&tk->rp);
 347                 else
 348                         disable_kprobe(&tk->rp.kp);
 349         }
 350 }
 351
 352 /*
 353  * Enable trace_probe
 354  * if the file is NULL, enable "perf" handler, or enable "trace" handler.
 355  */
 356 static int enable_trace_kprobe(struct trace_event_call *call,
 357                                 struct trace_event_file *file)
 358 {
 359         struct trace_probe *tp;
 360         struct trace_kprobe *tk;
 361         bool enabled;
 362         int ret = 0;
 363
 364         tp = trace_probe_primary_from_call(call);
 365         if (WARN_ON_ONCE(!tp))
 366                 return -ENODEV;
 367         enabled = trace_probe_is_enabled(tp);
 368
 369         /* This also changes "enabled" state */
 370         if (file) {
 371                 ret = trace_probe_add_file(tp, file);
 372                 if (ret)
 373                         return ret;
 374         } else
 375                 trace_probe_set_flag(tp, TP_FLAG_PROFILE);
 376
 377         if (enabled)
 378                 return 0;
 379
 380         list_for_each_entry(tk, trace_probe_probe_list(tp), tp.list) {
 381                 if (trace_kprobe_has_gone(tk))
 382                         continue;
 383                 ret = __enable_trace_kprobe(tk);
 384                 if (ret)
 385                         break;
 386                 enabled = true;
 387         }
 388
 389         if (ret) {
 390                 /* Failed to enable one of them. Roll back all */
 391                 if (enabled)
 392                         __disable_trace_kprobe(tp);
 393                 if (file)
 394                         trace_probe_remove_file(tp, file);
 395                 else
 396                         trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
 397         }
 398
 399         return ret;
 400 }
 401
 402 /*
 403  * Disable trace_probe
 404  * if the file is NULL, disable "perf" handler, or disable "trace" handler.
 405  */
 406 static int disable_trace_kprobe(struct trace_event_call *call,
 407                                 struct trace_event_file *file)
 408 {
 409         struct trace_probe *tp;
 410
 411         tp = trace_probe_primary_from_call(call);
 412         if (WARN_ON_ONCE(!tp))
 413                 return -ENODEV;
 414
 415         if (file) {
 416                 if (!trace_probe_get_file_link(tp, file))
 417                         return -ENOENT;
 418                 if (!trace_probe_has_single_file(tp))
 419                         goto out;
 420                 trace_probe_clear_flag(tp, TP_FLAG_TRACE);
 421         } else
 422                 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
 423
 424         if (!trace_probe_is_enabled(tp))
 425                 __disable_trace_kprobe(tp);
 426
 427  out:
 428         if (file)
 429                 /*
 430                  * Synchronization is done in below function. For perf event,
 431                  * file == NULL and perf_trace_event_unreg() calls
 432                  * tracepoint_synchronize_unregister() to ensure synchronize
 433                  * event. We don't need to care about it.
 434                  */
 435                 trace_probe_remove_file(tp, file);
 436
 437         return 0;
 438 }
 439
 440 #if defined(CONFIG_DYNAMIC_FTRACE) && \
 441         !defined(CONFIG_KPROBE_EVENTS_ON_NOTRACE)
 442 static bool __within_notrace_func(unsigned long addr)
 443 {
 444         unsigned long offset, size;
 445
 446         if (!addr || !kallsyms_lookup_size_offset(addr, &size, &offset))
 447                 return false;
 448
 449         /* Get the entry address of the target function */
 450         addr -= offset;
 451
 452         /*
 453          * Since ftrace_location_range() does inclusive range check, we need
 454          * to subtract 1 byte from the end address.
 455          */
 456         return !ftrace_location_range(addr, addr + size - 1);
 457 }
 458
 459 static bool within_notrace_func(struct trace_kprobe *tk)
 460 {
 461         unsigned long addr = trace_kprobe_address(tk);
 462         char symname[KSYM_NAME_LEN], *p;
 463
 464         if (!__within_notrace_func(addr))
 465                 return false;
 466
 467         /* Check if the address is on a suffixed-symbol */
 468         if (!lookup_symbol_name(addr, symname)) {
 469                 p = strchr(symname, '.');
 470                 if (!p)
 471                         return true;
 472                 *p = '\0';
 473                 addr = (unsigned long)kprobe_lookup_name(symname, 0);
 474                 if (addr)
 475                         return __within_notrace_func(addr);
 476         }
 477
 478         return true;
 479 }
 480 #else
 481 #define within_notrace_func(tk) (false)
 482 #endif
 483
 484 /* Internal register function - just handle k*probes and flags */
 485 static int __register_trace_kprobe(struct trace_kprobe *tk)
 486 {
 487         int i, ret;
 488
 489         ret = security_locked_down(LOCKDOWN_KPROBES);
 490         if (ret)
 491                 return ret;
 492
 493         if (trace_kprobe_is_registered(tk))
 494                 return -EINVAL;
 495
 496         if (within_notrace_func(tk)) {
 497                 pr_warn("Could not probe notrace function %ps\n",
 498                         (void *)trace_kprobe_address(tk));
 499                 return -EINVAL;
 500         }
 501
 502         for (i = 0; i < tk->tp.nr_args; i++) {
 503                 ret = traceprobe_update_arg(&tk->tp.args[i]);
 504                 if (ret)
 505                         return ret;
 506         }
 507
 508         /* Set/clear disabled flag according to tp->flag */
 509         if (trace_probe_is_enabled(&tk->tp))
 510                 tk->rp.kp.flags &= ~KPROBE_FLAG_DISABLED;
 511         else
 512                 tk->rp.kp.flags |= KPROBE_FLAG_DISABLED;
 513
 514         if (trace_kprobe_is_return(tk))
 515                 ret = register_kretprobe(&tk->rp);
 516         else
 517                 ret = register_kprobe(&tk->rp.kp);
 518
 519         return ret;
 520 }
 521
 522 /* Internal unregister function - just handle k*probes and flags */
 523 static void __unregister_trace_kprobe(struct trace_kprobe *tk)
 524 {
 525         if (trace_kprobe_is_registered(tk)) {
 526                 if (trace_kprobe_is_return(tk))
 527                         unregister_kretprobe(&tk->rp);
 528                 else
 529                         unregister_kprobe(&tk->rp.kp);
 530                 /* Cleanup kprobe for reuse and mark it unregistered */
 531                 INIT_HLIST_NODE(&tk->rp.kp.hlist);
 532                 INIT_LIST_HEAD(&tk->rp.kp.list);
 533                 if (tk->rp.kp.symbol_name)
 534                         tk->rp.kp.addr = NULL;
 535         }
 536 }
 537
 538 /* Unregister a trace_probe and probe_event */
 539 static int unregister_trace_kprobe(struct trace_kprobe *tk)
 540 {
 541         /* If other probes are on the event, just unregister kprobe */
 542         if (trace_probe_has_sibling(&tk->tp))
 543                 goto unreg;
 544
 545         /* Enabled event can not be unregistered */
 546         if (trace_probe_is_enabled(&tk->tp))
 547                 return -EBUSY;
 548
 549         /* If there's a reference to the dynamic event */
 550         if (trace_event_dyn_busy(trace_probe_event_call(&tk->tp)))
 551                 return -EBUSY;
 552
 553         /* Will fail if probe is being used by ftrace or perf */
 554         if (unregister_kprobe_event(tk))
 555                 return -EBUSY;
 556
 557 unreg:
 558         __unregister_trace_kprobe(tk);
 559         dyn_event_remove(&tk->devent);
 560         trace_probe_unlink(&tk->tp);
 561
 562         return 0;
 563 }
 564
 565 static bool trace_kprobe_has_same_kprobe(struct trace_kprobe *orig,
 566                                          struct trace_kprobe *comp)
 567 {
 568         struct trace_probe_event *tpe = orig->tp.event;
 569         int i;
 570
 571         list_for_each_entry(orig, &tpe->probes, tp.list) {
 572                 if (strcmp(trace_kprobe_symbol(orig),
 573                            trace_kprobe_symbol(comp)) ||
 574                     trace_kprobe_offset(orig) != trace_kprobe_offset(comp))
 575                         continue;
 576
 577                 /*
 578                  * trace_probe_compare_arg_type() ensured that nr_args and
 579                  * each argument name and type are same. Let's compare comm.
 580                  */
 581                 for (i = 0; i < orig->tp.nr_args; i++) {
 582                         if (strcmp(orig->tp.args[i].comm,
 583                                    comp->tp.args[i].comm))
 584                                 break;
 585                 }
 586
 587                 if (i == orig->tp.nr_args)
 588                         return true;
 589         }
 590
 591         return false;
 592 }
 593
 594 static int append_trace_kprobe(struct trace_kprobe *tk, struct trace_kprobe *to)
 595 {
 596         int ret;
 597
 598         ret = trace_probe_compare_arg_type(&tk->tp, &to->tp);
 599         if (ret) {
 600                 /* Note that argument starts index = 2 */
 601                 trace_probe_log_set_index(ret + 1);
 602                 trace_probe_log_err(0, DIFF_ARG_TYPE);
 603                 return -EEXIST;
 604         }
 605         if (trace_kprobe_has_same_kprobe(to, tk)) {
 606                 trace_probe_log_set_index(0);
 607                 trace_probe_log_err(0, SAME_PROBE);
 608                 return -EEXIST;
 609         }
 610
 611         /* Append to existing event */
 612         ret = trace_probe_append(&tk->tp, &to->tp);
 613         if (ret)
 614                 return ret;
 615
 616         /* Register k*probe */
 617         ret = __register_trace_kprobe(tk);
 618         if (ret == -ENOENT && !trace_kprobe_module_exist(tk)) {
 619                 pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
 620                 ret = 0;
 621         }
 622
 623         if (ret)
 624                 trace_probe_unlink(&tk->tp);
 625         else
 626                 dyn_event_add(&tk->devent, trace_probe_event_call(&tk->tp));
 627
 628         return ret;
 629 }
 630
 631 /* Register a trace_probe and probe_event */
 632 static int register_trace_kprobe(struct trace_kprobe *tk)
 633 {
 634         struct trace_kprobe *old_tk;
 635         int ret;
 636
 637         guard(mutex)(&event_mutex);
 638
 639         old_tk = find_trace_kprobe(trace_probe_name(&tk->tp),
 640                                    trace_probe_group_name(&tk->tp));
 641         if (old_tk) {
 642                 if (trace_kprobe_is_return(tk) != trace_kprobe_is_return(old_tk)) {
 643                         trace_probe_log_set_index(0);
 644                         trace_probe_log_err(0, DIFF_PROBE_TYPE);
 645                         return -EEXIST;
 646                 }
 647                 return append_trace_kprobe(tk, old_tk);
 648         }
 649
 650         /* Register new event */
 651         ret = register_kprobe_event(tk);
 652         if (ret) {
 653                 if (ret == -EEXIST) {
 654                         trace_probe_log_set_index(0);
 655                         trace_probe_log_err(0, EVENT_EXIST);
 656                 } else
 657                         pr_warn("Failed to register probe event(%d)\n", ret);
 658                 return ret;
 659         }
 660
 661         /* Register k*probe */
 662         ret = __register_trace_kprobe(tk);
 663         if (ret == -ENOENT && !trace_kprobe_module_exist(tk)) {
 664                 pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
 665                 ret = 0;
 666         }
 667
 668         if (ret < 0)
 669                 unregister_kprobe_event(tk);
 670         else
 671                 dyn_event_add(&tk->devent, trace_probe_event_call(&tk->tp));
 672
 673         return ret;
 674 }
 675
 676 #ifdef CONFIG_MODULES
 677 static int validate_module_probe_symbol(const char *modname, const char *symbol);
 678
 679 static int register_module_trace_kprobe(struct module *mod, struct trace_kprobe *tk)
 680 {
 681         const char *p;
 682         int ret = 0;
 683
 684         p = strchr(trace_kprobe_symbol(tk), ':');
 685         if (p)
 686                 ret = validate_module_probe_symbol(module_name(mod), p + 1);
 687         if (!ret)
 688                 ret = __register_trace_kprobe(tk);
 689         return ret;
 690 }
 691
 692 /* Module notifier call back, checking event on the module */
 693 static int trace_kprobe_module_callback(struct notifier_block *nb,
 694                                        unsigned long val, void *data)
 695 {
 696         struct module *mod = data;
 697         struct dyn_event *pos;
 698         struct trace_kprobe *tk;
 699         int ret;
 700
 701         if (val != MODULE_STATE_COMING)
 702                 return NOTIFY_DONE;
 703
 704         /* Update probes on coming module */
 705         guard(mutex)(&event_mutex);
 706         for_each_trace_kprobe(tk, pos) {
 707                 if (trace_kprobe_within_module(tk, mod)) {
 708                         /* Don't need to check busy - this should have gone. */
 709                         __unregister_trace_kprobe(tk);
 710                         ret = register_module_trace_kprobe(mod, tk);
 711                         if (ret)
 712                                 pr_warn("Failed to re-register probe %s on %s: %d\n",
 713                                         trace_probe_name(&tk->tp),
 714                                         module_name(mod), ret);
 715                 }
 716         }
 717
 718         return NOTIFY_DONE;
 719 }
 720
 721 static struct notifier_block trace_kprobe_module_nb = {
 722         .notifier_call = trace_kprobe_module_callback,
 723         .priority = 2   /* Invoked after kprobe and jump_label module callback */
 724 };
 725 static int trace_kprobe_register_module_notifier(void)
 726 {
 727         return register_module_notifier(&trace_kprobe_module_nb);
 728 }
 729 #else
 730 static int trace_kprobe_register_module_notifier(void)
 731 {
 732         return 0;
 733 }
 734 #endif /* CONFIG_MODULES */
 735
 736 static int count_symbols(void *data, unsigned long unused)
 737 {
 738         unsigned int *count = data;
 739
 740         (*count)++;
 741
 742         return 0;
 743 }
 744
 745 struct sym_count_ctx {
 746         unsigned int count;
 747         const char *name;
 748 };
 749
 750 static int count_mod_symbols(void *data, const char *name, unsigned long unused)
 751 {
 752         struct sym_count_ctx *ctx = data;
 753
 754         if (strcmp(name, ctx->name) == 0)
 755                 ctx->count++;
 756
 757         return 0;
 758 }
 759
 760 static unsigned int number_of_same_symbols(const char *mod, const char *func_name)
 761 {
 762         struct sym_count_ctx ctx = { .count = 0, .name = func_name };
 763
 764         if (!mod)
 765                 kallsyms_on_each_match_symbol(count_symbols, func_name, &ctx.count);
 766
 767         module_kallsyms_on_each_symbol(mod, count_mod_symbols, &ctx);
 768
 769         return ctx.count;
 770 }
 771
 772 static int validate_module_probe_symbol(const char *modname, const char *symbol)
 773 {
 774         unsigned int count = number_of_same_symbols(modname, symbol);
 775
 776         if (count > 1) {
 777                 /*
 778                  * Users should use ADDR to remove the ambiguity of
 779                  * using KSYM only.
 780                  */
 781                 return -EADDRNOTAVAIL;
 782         } else if (count == 0) {
 783                 /*
 784                  * We can return ENOENT earlier than when register the
 785                  * kprobe.
 786                  */
 787                 return -ENOENT;
 788         }
 789         return 0;
 790 }
 791
 792 #ifdef CONFIG_MODULES
 793 /* Return NULL if the module is not loaded or under unloading. */
 794 static struct module *try_module_get_by_name(const char *name)
 795 {
 796         struct module *mod;
 797
 798         guard(rcu)();
 799         mod = find_module(name);
 800         if (mod && !try_module_get(mod))
 801                 mod = NULL;
 802         return mod;
 803 }
 804 #else
 805 #define try_module_get_by_name(name)    (NULL)
 806 #endif
 807
 808 static int validate_probe_symbol(char *symbol)
 809 {
 810         struct module *mod = NULL;
 811         char *modname = NULL, *p;
 812         int ret = 0;
 813
 814         p = strchr(symbol, ':');
 815         if (p) {
 816                 modname = symbol;
 817                 symbol = p + 1;
 818                 *p = '\0';
 819                 mod = try_module_get_by_name(modname);
 820                 if (!mod)
 821                         goto out;
 822         }
 823
 824         ret = validate_module_probe_symbol(modname, symbol);
 825 out:
 826         if (p)
 827                 *p = ':';
 828         if (mod)
 829                 module_put(mod);
 830         return ret;
 831 }
 832
 833 static int trace_kprobe_entry_handler(struct kretprobe_instance *ri,
 834                                       struct pt_regs *regs);
 835
 836 static int trace_kprobe_create_internal(int argc, const char *argv[],
 837                                         struct traceprobe_parse_context *ctx)
 838 {
 839         /*
 840          * Argument syntax:
 841          *  - Add kprobe:
 842          *      p[:[GRP/][EVENT]] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
 843          *  - Add kretprobe:
 844          *      r[MAXACTIVE][:[GRP/][EVENT]] [MOD:]KSYM[+0] [FETCHARGS]
 845          *    Or
 846          *      p[:[GRP/][EVENT]] [MOD:]KSYM[+0]%return [FETCHARGS]
 847          *
 848          * Fetch args:
 849          *  $retval     : fetch return value
 850          *  $stack      : fetch stack address
 851          *  $stackN     : fetch Nth of stack (N:0-)
 852          *  $comm       : fetch current task comm
 853          *  @ADDR       : fetch memory at ADDR (ADDR should be in kernel)
 854          *  @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
 855          *  %REG        : fetch register REG
 856          * Dereferencing memory fetch:
 857          *  +|-offs(ARG) : fetch memory at ARG +|- offs address.
 858          * Alias name of args:
 859          *  NAME=FETCHARG : set NAME as alias of FETCHARG.
 860          * Type of args:
 861          *  FETCHARG:TYPE : use TYPE instead of unsigned long.
 862          */
 863         struct trace_kprobe *tk __free(free_trace_kprobe) = NULL;
 864         int i, len, new_argc = 0, ret = 0;
 865         bool is_return = false;
 866         char *symbol __free(kfree) = NULL;
 867         char *tmp = NULL;
 868         const char **new_argv __free(kfree) = NULL;
 869         const char *event = NULL, *group = KPROBE_EVENT_SYSTEM;
 870         enum probe_print_type ptype;
 871         int maxactive = 0;
 872         long offset = 0;
 873         void *addr = NULL;
 874         char buf[MAX_EVENT_NAME_LEN];
 875         char gbuf[MAX_EVENT_NAME_LEN];
 876         char abuf[MAX_BTF_ARGS_LEN];
 877         char *dbuf __free(kfree) = NULL;
 878
 879         switch (argv[0][0]) {
 880         case 'r':
 881                 is_return = true;
 882                 break;
 883         case 'p':
 884                 break;
 885         default:
 886                 return -ECANCELED;
 887         }
 888         if (argc < 2)
 889                 return -ECANCELED;
 890
 891         event = strchr(&argv[0][1], ':');
 892         if (event)
 893                 event++;
 894
 895         if (isdigit(argv[0][1])) {
 896                 if (!is_return) {
 897                         trace_probe_log_err(1, BAD_MAXACT_TYPE);
 898                         return -EINVAL;
 899                 }
 900                 if (event)
 901                         len = event - &argv[0][1] - 1;
 902                 else
 903                         len = strlen(&argv[0][1]);
 904                 if (len > MAX_EVENT_NAME_LEN - 1) {
 905                         trace_probe_log_err(1, BAD_MAXACT);
 906                         return -EINVAL;
 907                 }
 908                 memcpy(buf, &argv[0][1], len);
 909                 buf[len] = '\0';
 910                 ret = kstrtouint(buf, 0, &maxactive);
 911                 if (ret || !maxactive) {
 912                         trace_probe_log_err(1, BAD_MAXACT);
 913                         return -EINVAL;
 914                 }
 915                 /* kretprobes instances are iterated over via a list. The
 916                  * maximum should stay reasonable.
 917                  */
 918                 if (maxactive > KRETPROBE_MAXACTIVE_MAX) {
 919                         trace_probe_log_err(1, MAXACT_TOO_BIG);
 920                         return -EINVAL;
 921                 }
 922         }
 923
 924         /* try to parse an address. if that fails, try to read the
 925          * input as a symbol. */
 926         if (kstrtoul(argv[1], 0, (unsigned long *)&addr)) {
 927                 trace_probe_log_set_index(1);
 928                 /* Check whether uprobe event specified */
 929                 if (strchr(argv[1], '/') && strchr(argv[1], ':'))
 930                         return -ECANCELED;
 931
 932                 /* a symbol specified */
 933                 symbol = kstrdup(argv[1], GFP_KERNEL);
 934                 if (!symbol)
 935                         return -ENOMEM;
 936
 937                 tmp = strchr(symbol, '%');
 938                 if (tmp) {
 939                         if (!strcmp(tmp, "%return")) {
 940                                 *tmp = '\0';
 941                                 is_return = true;
 942                         } else {
 943                                 trace_probe_log_err(tmp - symbol, BAD_ADDR_SUFFIX);
 944                                 return -EINVAL;
 945                         }
 946                 }
 947
 948                 /* TODO: support .init module functions */
 949                 ret = traceprobe_split_symbol_offset(symbol, &offset);
 950                 if (ret || offset < 0 || offset > UINT_MAX) {
 951                         trace_probe_log_err(0, BAD_PROBE_ADDR);
 952                         return -EINVAL;
 953                 }
 954                 ret = validate_probe_symbol(symbol);
 955                 if (ret) {
 956                         if (ret == -EADDRNOTAVAIL)
 957                                 trace_probe_log_err(0, NON_UNIQ_SYMBOL);
 958                         else
 959                                 trace_probe_log_err(0, BAD_PROBE_ADDR);
 960                         return -EINVAL;
 961                 }
 962                 if (is_return)
 963                         ctx->flags |= TPARG_FL_RETURN;
 964                 ret = kprobe_on_func_entry(NULL, symbol, offset);
 965                 if (ret == 0 && !is_return)
 966                         ctx->flags |= TPARG_FL_FENTRY;
 967                 /* Defer the ENOENT case until register kprobe */
 968                 if (ret == -EINVAL && is_return) {
 969                         trace_probe_log_err(0, BAD_RETPROBE);
 970                         return -EINVAL;
 971                 }
 972         }
 973
 974         trace_probe_log_set_index(0);
 975         if (event) {
 976                 ret = traceprobe_parse_event_name(&event, &group, gbuf,
 977                                                   event - argv[0]);
 978                 if (ret)
 979                         return ret;
 980         }
 981
 982         if (!event) {
 983                 /* Make a new event name */
 984                 if (symbol)
 985                         snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
 986                                  is_return ? 'r' : 'p', symbol, offset);
 987                 else
 988                         snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
 989                                  is_return ? 'r' : 'p', addr);
 990                 sanitize_event_name(buf);
 991                 event = buf;
 992         }
 993
 994         argc -= 2; argv += 2;
 995         ctx->funcname = symbol;
 996         new_argv = traceprobe_expand_meta_args(argc, argv, &new_argc,
 997                                                abuf, MAX_BTF_ARGS_LEN, ctx);
 998         if (IS_ERR(new_argv)) {
 999                 ret = PTR_ERR(new_argv);
1000                 new_argv = NULL;
1001                 return ret;
1002         }
1003         if (new_argv) {
1004                 argc = new_argc;
1005                 argv = new_argv;
1006         }
1007         if (argc > MAX_TRACE_ARGS) {
1008                 trace_probe_log_set_index(2);
1009                 trace_probe_log_err(0, TOO_MANY_ARGS);
1010                 return -E2BIG;
1011         }
1012
1013         ret = traceprobe_expand_dentry_args(argc, argv, &dbuf);
1014         if (ret)
1015                 return ret;
1016
1017         /* setup a probe */
1018         tk = alloc_trace_kprobe(group, event, addr, symbol, offset, maxactive,
1019                                 argc, is_return);
1020         if (IS_ERR(tk)) {
1021                 ret = PTR_ERR(tk);
1022                 /* This must return -ENOMEM, else there is a bug */
1023                 WARN_ON_ONCE(ret != -ENOMEM);
1024                 return ret;     /* We know tk is not allocated */
1025         }
1026
1027         /* parse arguments */
1028         for (i = 0; i < argc; i++) {
1029                 trace_probe_log_set_index(i + 2);
1030                 ctx->offset = 0;
1031                 ret = traceprobe_parse_probe_arg(&tk->tp, i, argv[i], ctx);
1032                 if (ret)
1033                         return ret;     /* This can be -ENOMEM */
1034         }
1035         /* entry handler for kretprobe */
1036         if (is_return && tk->tp.entry_arg) {
1037                 tk->rp.entry_handler = trace_kprobe_entry_handler;
1038                 tk->rp.data_size = traceprobe_get_entry_data_size(&tk->tp);
1039         }
1040
1041         ptype = is_return ? PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL;
1042         ret = traceprobe_set_print_fmt(&tk->tp, ptype);
1043         if (ret < 0)
1044                 return ret;
1045
1046         ret = register_trace_kprobe(tk);
1047         if (ret) {
1048                 trace_probe_log_set_index(1);
1049                 if (ret == -EILSEQ)
1050                         trace_probe_log_err(0, BAD_INSN_BNDRY);
1051                 else if (ret == -ENOENT)
1052                         trace_probe_log_err(0, BAD_PROBE_ADDR);
1053                 else if (ret != -ENOMEM && ret != -EEXIST)
1054                         trace_probe_log_err(0, FAIL_REG_PROBE);
1055                 return ret;
1056         }
1057         /*
1058          * Here, 'tk' has been registered to the list successfully,
1059          * so we don't need to free it.
1060          */
1061         tk = NULL;
1062
1063         return 0;
1064 }
1065
1066 static int trace_kprobe_create_cb(int argc, const char *argv[])
1067 {
1068         struct traceprobe_parse_context ctx = { .flags = TPARG_FL_KERNEL };
1069         int ret;
1070
1071         trace_probe_log_init("trace_kprobe", argc, argv);
1072
1073         ret = trace_kprobe_create_internal(argc, argv, &ctx);
1074
1075         traceprobe_finish_parse(&ctx);
1076         trace_probe_log_clear();
1077         return ret;
1078 }
1079
1080 static int trace_kprobe_create(const char *raw_command)
1081 {
1082         return trace_probe_create(raw_command, trace_kprobe_create_cb);
1083 }
1084
1085 static int create_or_delete_trace_kprobe(const char *raw_command)
1086 {
1087         int ret;
1088
1089         if (raw_command[0] == '-')
1090                 return dyn_event_release(raw_command, &trace_kprobe_ops);
1091
1092         ret = dyn_event_create(raw_command, &trace_kprobe_ops);
1093         return ret == -ECANCELED ? -EINVAL : ret;
1094 }
1095
1096 static int trace_kprobe_run_command(struct dynevent_cmd *cmd)
1097 {
1098         return create_or_delete_trace_kprobe(cmd->seq.buffer);
1099 }
1100
1101 /**
1102  * kprobe_event_cmd_init - Initialize a kprobe event command object
1103  * @cmd: A pointer to the dynevent_cmd struct representing the new event
1104  * @buf: A pointer to the buffer used to build the command
1105  * @maxlen: The length of the buffer passed in @buf
1106  *
1107  * Initialize a synthetic event command object.  Use this before
1108  * calling any of the other kprobe_event functions.
1109  */
1110 void kprobe_event_cmd_init(struct dynevent_cmd *cmd, char *buf, int maxlen)
1111 {
1112         dynevent_cmd_init(cmd, buf, maxlen, DYNEVENT_TYPE_KPROBE,
1113                           trace_kprobe_run_command);
1114 }
1115 EXPORT_SYMBOL_GPL(kprobe_event_cmd_init);
1116
1117 /**
1118  * __kprobe_event_gen_cmd_start - Generate a kprobe event command from arg list
1119  * @cmd: A pointer to the dynevent_cmd struct representing the new event
1120  * @kretprobe: Is this a return probe?
1121  * @name: The name of the kprobe event
1122  * @loc: The location of the kprobe event
1123  * @...: Variable number of arg (pairs), one pair for each field
1124  *
1125  * NOTE: Users normally won't want to call this function directly, but
1126  * rather use the kprobe_event_gen_cmd_start() wrapper, which automatically
1127  * adds a NULL to the end of the arg list.  If this function is used
1128  * directly, make sure the last arg in the variable arg list is NULL.
1129  *
1130  * Generate a kprobe event command to be executed by
1131  * kprobe_event_gen_cmd_end().  This function can be used to generate the
1132  * complete command or only the first part of it; in the latter case,
1133  * kprobe_event_add_fields() can be used to add more fields following this.
1134  *
1135  * Unlikely the synth_event_gen_cmd_start(), @loc must be specified. This
1136  * returns -EINVAL if @loc == NULL.
1137  *
1138  * Return: 0 if successful, error otherwise.
1139  */
1140 int __kprobe_event_gen_cmd_start(struct dynevent_cmd *cmd, bool kretprobe,
1141                                  const char *name, const char *loc, ...)
1142 {
1143         char buf[MAX_EVENT_NAME_LEN];
1144         struct dynevent_arg arg;
1145         va_list args;
1146         int ret;
1147
1148         if (cmd->type != DYNEVENT_TYPE_KPROBE)
1149                 return -EINVAL;
1150
1151         if (!loc)
1152                 return -EINVAL;
1153
1154         if (kretprobe)
1155                 snprintf(buf, MAX_EVENT_NAME_LEN, "r:kprobes/%s", name);
1156         else
1157                 snprintf(buf, MAX_EVENT_NAME_LEN, "p:kprobes/%s", name);
1158
1159         ret = dynevent_str_add(cmd, buf);
1160         if (ret)
1161                 return ret;
1162
1163         dynevent_arg_init(&arg, 0);
1164         arg.str = loc;
1165         ret = dynevent_arg_add(cmd, &arg, NULL);
1166         if (ret)
1167                 return ret;
1168
1169         va_start(args, loc);
1170         for (;;) {
1171                 const char *field;
1172
1173                 field = va_arg(args, const char *);
1174                 if (!field)
1175                         break;
1176
1177                 if (++cmd->n_fields > MAX_TRACE_ARGS) {
1178                         ret = -EINVAL;
1179                         break;
1180                 }
1181
1182                 arg.str = field;
1183                 ret = dynevent_arg_add(cmd, &arg, NULL);
1184                 if (ret)
1185                         break;
1186         }
1187         va_end(args);
1188
1189         return ret;
1190 }
1191 EXPORT_SYMBOL_GPL(__kprobe_event_gen_cmd_start);
1192
1193 /**
1194  * __kprobe_event_add_fields - Add probe fields to a kprobe command from arg list
1195  * @cmd: A pointer to the dynevent_cmd struct representing the new event
1196  * @...: Variable number of arg (pairs), one pair for each field
1197  *
1198  * NOTE: Users normally won't want to call this function directly, but
1199  * rather use the kprobe_event_add_fields() wrapper, which
1200  * automatically adds a NULL to the end of the arg list.  If this
1201  * function is used directly, make sure the last arg in the variable
1202  * arg list is NULL.
1203  *
1204  * Add probe fields to an existing kprobe command using a variable
1205  * list of args.  Fields are added in the same order they're listed.
1206  *
1207  * Return: 0 if successful, error otherwise.
1208  */
1209 int __kprobe_event_add_fields(struct dynevent_cmd *cmd, ...)
1210 {
1211         struct dynevent_arg arg;
1212         va_list args;
1213         int ret = 0;
1214
1215         if (cmd->type != DYNEVENT_TYPE_KPROBE)
1216                 return -EINVAL;
1217
1218         dynevent_arg_init(&arg, 0);
1219
1220         va_start(args, cmd);
1221         for (;;) {
1222                 const char *field;
1223
1224                 field = va_arg(args, const char *);
1225                 if (!field)
1226                         break;
1227
1228                 if (++cmd->n_fields > MAX_TRACE_ARGS) {
1229                         ret = -EINVAL;
1230                         break;
1231                 }
1232
1233                 arg.str = field;
1234                 ret = dynevent_arg_add(cmd, &arg, NULL);
1235                 if (ret)
1236                         break;
1237         }
1238         va_end(args);
1239
1240         return ret;
1241 }
1242 EXPORT_SYMBOL_GPL(__kprobe_event_add_fields);
1243
1244 /**
1245  * kprobe_event_delete - Delete a kprobe event
1246  * @name: The name of the kprobe event to delete
1247  *
1248  * Delete a kprobe event with the give @name from kernel code rather
1249  * than directly from the command line.
1250  *
1251  * Return: 0 if successful, error otherwise.
1252  */
1253 int kprobe_event_delete(const char *name)
1254 {
1255         char buf[MAX_EVENT_NAME_LEN];
1256
1257         snprintf(buf, MAX_EVENT_NAME_LEN, "-:%s", name);
1258
1259         return create_or_delete_trace_kprobe(buf);
1260 }
1261 EXPORT_SYMBOL_GPL(kprobe_event_delete);
1262
1263 static int trace_kprobe_release(struct dyn_event *ev)
1264 {
1265         struct trace_kprobe *tk = to_trace_kprobe(ev);
1266         int ret = unregister_trace_kprobe(tk);
1267
1268         if (!ret)
1269                 free_trace_kprobe(tk);
1270         return ret;
1271 }
1272
1273 static int trace_kprobe_show(struct seq_file *m, struct dyn_event *ev)
1274 {
1275         struct trace_kprobe *tk = to_trace_kprobe(ev);
1276         int i;
1277
1278         seq_putc(m, trace_kprobe_is_return(tk) ? 'r' : 'p');
1279         if (trace_kprobe_is_return(tk) && tk->rp.maxactive)
1280                 seq_printf(m, "%d", tk->rp.maxactive);
1281         seq_printf(m, ":%s/%s", trace_probe_group_name(&tk->tp),
1282                                 trace_probe_name(&tk->tp));
1283
1284         if (!tk->symbol)
1285                 seq_printf(m, " 0x%p", tk->rp.kp.addr);
1286         else if (tk->rp.kp.offset)
1287                 seq_printf(m, " %s+%u", trace_kprobe_symbol(tk),
1288                            tk->rp.kp.offset);
1289         else
1290                 seq_printf(m, " %s", trace_kprobe_symbol(tk));
1291
1292         for (i = 0; i < tk->tp.nr_args; i++)
1293                 seq_printf(m, " %s=%s", tk->tp.args[i].name, tk->tp.args[i].comm);
1294         seq_putc(m, '\n');
1295
1296         return 0;
1297 }
1298
1299 static int probes_seq_show(struct seq_file *m, void *v)
1300 {
1301         struct dyn_event *ev = v;
1302
1303         if (!is_trace_kprobe(ev))
1304                 return 0;
1305
1306         return trace_kprobe_show(m, ev);
1307 }
1308
1309 static const struct seq_operations probes_seq_op = {
1310         .start  = dyn_event_seq_start,
1311         .next   = dyn_event_seq_next,
1312         .stop   = dyn_event_seq_stop,
1313         .show   = probes_seq_show
1314 };
1315
1316 static int probes_open(struct inode *inode, struct file *file)
1317 {
1318         int ret;
1319
1320         ret = security_locked_down(LOCKDOWN_TRACEFS);
1321         if (ret)
1322                 return ret;
1323
1324         if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
1325                 ret = dyn_events_release_all(&trace_kprobe_ops);
1326                 if (ret < 0)
1327                         return ret;
1328         }
1329
1330         return seq_open(file, &probes_seq_op);
1331 }
1332
1333 static ssize_t probes_write(struct file *file, const char __user *buffer,
1334                             size_t count, loff_t *ppos)
1335 {
1336         return trace_parse_run_command(file, buffer, count, ppos,
1337                                        create_or_delete_trace_kprobe);
1338 }
1339
1340 static const struct file_operations kprobe_events_ops = {
1341         .owner          = THIS_MODULE,
1342         .open           = probes_open,
1343         .read           = seq_read,
1344         .llseek         = seq_lseek,
1345         .release        = seq_release,
1346         .write          = probes_write,
1347 };
1348
1349 static unsigned long trace_kprobe_missed(struct trace_kprobe *tk)
1350 {
1351         return trace_kprobe_is_return(tk) ?
1352                 tk->rp.kp.nmissed + tk->rp.nmissed : tk->rp.kp.nmissed;
1353 }
1354
1355 /* Probes profiling interfaces */
1356 static int probes_profile_seq_show(struct seq_file *m, void *v)
1357 {
1358         struct dyn_event *ev = v;
1359         struct trace_kprobe *tk;
1360         unsigned long nmissed;
1361
1362         if (!is_trace_kprobe(ev))
1363                 return 0;
1364
1365         tk = to_trace_kprobe(ev);
1366         nmissed = trace_kprobe_missed(tk);
1367         seq_printf(m, "  %-44s %15lu %15lu\n",
1368                    trace_probe_name(&tk->tp),
1369                    trace_kprobe_nhit(tk),
1370                    nmissed);
1371
1372         return 0;
1373 }
1374
1375 static const struct seq_operations profile_seq_op = {
1376         .start  = dyn_event_seq_start,
1377         .next   = dyn_event_seq_next,
1378         .stop   = dyn_event_seq_stop,
1379         .show   = probes_profile_seq_show
1380 };
1381
1382 static int profile_open(struct inode *inode, struct file *file)
1383 {
1384         int ret;
1385
1386         ret = security_locked_down(LOCKDOWN_TRACEFS);
1387         if (ret)
1388                 return ret;
1389
1390         return seq_open(file, &profile_seq_op);
1391 }
1392
1393 static const struct file_operations kprobe_profile_ops = {
1394         .owner          = THIS_MODULE,
1395         .open           = profile_open,
1396         .read           = seq_read,
1397         .llseek         = seq_lseek,
1398         .release        = seq_release,
1399 };
1400
1401 /* Note that we don't verify it, since the code does not come from user space */
1402 static int
1403 process_fetch_insn(struct fetch_insn *code, void *rec, void *edata,
1404                    void *dest, void *base)
1405 {
1406         struct pt_regs *regs = rec;
1407         unsigned long val;
1408         int ret;
1409
1410 retry:
1411         /* 1st stage: get value from context */
1412         switch (code->op) {
1413         case FETCH_OP_REG:
1414                 val = regs_get_register(regs, code->param);
1415                 break;
1416         case FETCH_OP_STACK:
1417                 val = regs_get_kernel_stack_nth(regs, code->param);
1418                 break;
1419         case FETCH_OP_STACKP:
1420                 val = kernel_stack_pointer(regs);
1421                 break;
1422         case FETCH_OP_RETVAL:
1423                 val = regs_return_value(regs);
1424                 break;
1425 #ifdef CONFIG_HAVE_FUNCTION_ARG_ACCESS_API
1426         case FETCH_OP_ARG:
1427                 val = regs_get_kernel_argument(regs, code->param);
1428                 break;
1429         case FETCH_OP_EDATA:
1430                 val = *(unsigned long *)((unsigned long)edata + code->offset);
1431                 break;
1432 #endif
1433         case FETCH_NOP_SYMBOL:  /* Ignore a place holder */
1434                 code++;
1435                 goto retry;
1436         default:
1437                 ret = process_common_fetch_insn(code, &val);
1438                 if (ret < 0)
1439                         return ret;
1440         }
1441         code++;
1442
1443         return process_fetch_insn_bottom(code, val, dest, base);
1444 }
1445 NOKPROBE_SYMBOL(process_fetch_insn)
1446
1447 /* Kprobe handler */
1448 static nokprobe_inline void
1449 __kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs,
1450                     struct trace_event_file *trace_file)
1451 {
1452         struct kprobe_trace_entry_head *entry;
1453         struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1454         struct trace_event_buffer fbuffer;
1455         int dsize;
1456
1457         WARN_ON(call != trace_file->event_call);
1458
1459         if (trace_trigger_soft_disabled(trace_file))
1460                 return;
1461
1462         dsize = __get_data_size(&tk->tp, regs, NULL);
1463
1464         entry = trace_event_buffer_reserve(&fbuffer, trace_file,
1465                                            sizeof(*entry) + tk->tp.size + dsize);
1466         if (!entry)
1467                 return;
1468
1469         fbuffer.regs = regs;
1470         entry->ip = (unsigned long)tk->rp.kp.addr;
1471         store_trace_args(&entry[1], &tk->tp, regs, NULL, sizeof(*entry), dsize);
1472
1473         trace_event_buffer_commit(&fbuffer);
1474 }
1475
1476 static void
1477 kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs)
1478 {
1479         struct event_file_link *link;
1480
1481         trace_probe_for_each_link_rcu(link, &tk->tp)
1482                 __kprobe_trace_func(tk, regs, link->file);
1483 }
1484 NOKPROBE_SYMBOL(kprobe_trace_func);
1485
1486 /* Kretprobe handler */
1487
1488 static int trace_kprobe_entry_handler(struct kretprobe_instance *ri,
1489                                       struct pt_regs *regs)
1490 {
1491         struct kretprobe *rp = get_kretprobe(ri);
1492         struct trace_kprobe *tk;
1493
1494         /*
1495          * There is a small chance that get_kretprobe(ri) returns NULL when
1496          * the kretprobe is unregister on another CPU between kretprobe's
1497          * trampoline_handler and this function.
1498          */
1499         if (unlikely(!rp))
1500                 return -ENOENT;
1501
1502         tk = container_of(rp, struct trace_kprobe, rp);
1503
1504         /* store argument values into ri->data as entry data */
1505         if (tk->tp.entry_arg)
1506                 store_trace_entry_data(ri->data, &tk->tp, regs);
1507
1508         return 0;
1509 }
1510
1511
1512 static nokprobe_inline void
1513 __kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1514                        struct pt_regs *regs,
1515                        struct trace_event_file *trace_file)
1516 {
1517         struct kretprobe_trace_entry_head *entry;
1518         struct trace_event_buffer fbuffer;
1519         struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1520         int dsize;
1521
1522         WARN_ON(call != trace_file->event_call);
1523
1524         if (trace_trigger_soft_disabled(trace_file))
1525                 return;
1526
1527         dsize = __get_data_size(&tk->tp, regs, ri->data);
1528
1529         entry = trace_event_buffer_reserve(&fbuffer, trace_file,
1530                                            sizeof(*entry) + tk->tp.size + dsize);
1531         if (!entry)
1532                 return;
1533
1534         fbuffer.regs = regs;
1535         entry->func = (unsigned long)tk->rp.kp.addr;
1536         entry->ret_ip = get_kretprobe_retaddr(ri);
1537         store_trace_args(&entry[1], &tk->tp, regs, ri->data, sizeof(*entry), dsize);
1538
1539         trace_event_buffer_commit(&fbuffer);
1540 }
1541
1542 static void
1543 kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1544                      struct pt_regs *regs)
1545 {
1546         struct event_file_link *link;
1547
1548         trace_probe_for_each_link_rcu(link, &tk->tp)
1549                 __kretprobe_trace_func(tk, ri, regs, link->file);
1550 }
1551 NOKPROBE_SYMBOL(kretprobe_trace_func);
1552
1553 /* Event entry printers */
1554 static enum print_line_t
1555 print_kprobe_event(struct trace_iterator *iter, int flags,
1556                    struct trace_event *event)
1557 {
1558         struct kprobe_trace_entry_head *field;
1559         struct trace_seq *s = &iter->seq;
1560         struct trace_probe *tp;
1561
1562         field = (struct kprobe_trace_entry_head *)iter->ent;
1563         tp = trace_probe_primary_from_call(
1564                 container_of(event, struct trace_event_call, event));
1565         if (WARN_ON_ONCE(!tp))
1566                 goto out;
1567
1568         trace_seq_printf(s, "%s: (", trace_probe_name(tp));
1569
1570         if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
1571                 goto out;
1572
1573         trace_seq_putc(s, ')');
1574
1575         if (trace_probe_print_args(s, tp->args, tp->nr_args,
1576                              (u8 *)&field[1], field) < 0)
1577                 goto out;
1578
1579         trace_seq_putc(s, '\n');
1580  out:
1581         return trace_handle_return(s);
1582 }
1583
1584 static enum print_line_t
1585 print_kretprobe_event(struct trace_iterator *iter, int flags,
1586                       struct trace_event *event)
1587 {
1588         struct kretprobe_trace_entry_head *field;
1589         struct trace_seq *s = &iter->seq;
1590         struct trace_probe *tp;
1591
1592         field = (struct kretprobe_trace_entry_head *)iter->ent;
1593         tp = trace_probe_primary_from_call(
1594                 container_of(event, struct trace_event_call, event));
1595         if (WARN_ON_ONCE(!tp))
1596                 goto out;
1597
1598         trace_seq_printf(s, "%s: (", trace_probe_name(tp));
1599
1600         if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1601                 goto out;
1602
1603         trace_seq_puts(s, " <- ");
1604
1605         if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1606                 goto out;
1607
1608         trace_seq_putc(s, ')');
1609
1610         if (trace_probe_print_args(s, tp->args, tp->nr_args,
1611                              (u8 *)&field[1], field) < 0)
1612                 goto out;
1613
1614         trace_seq_putc(s, '\n');
1615
1616  out:
1617         return trace_handle_return(s);
1618 }
1619
1620
1621 static int kprobe_event_define_fields(struct trace_event_call *event_call)
1622 {
1623         int ret;
1624         struct kprobe_trace_entry_head field;
1625         struct trace_probe *tp;
1626
1627         tp = trace_probe_primary_from_call(event_call);
1628         if (WARN_ON_ONCE(!tp))
1629                 return -ENOENT;
1630
1631         DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1632
1633         return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
1634 }
1635
1636 static int kretprobe_event_define_fields(struct trace_event_call *event_call)
1637 {
1638         int ret;
1639         struct kretprobe_trace_entry_head field;
1640         struct trace_probe *tp;
1641
1642         tp = trace_probe_primary_from_call(event_call);
1643         if (WARN_ON_ONCE(!tp))
1644                 return -ENOENT;
1645
1646         DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1647         DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1648
1649         return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
1650 }
1651
1652 #ifdef CONFIG_PERF_EVENTS
1653
1654 /* Kprobe profile handler */
1655 static int
1656 kprobe_perf_func(struct trace_kprobe *tk, struct pt_regs *regs)
1657 {
1658         struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1659         struct kprobe_trace_entry_head *entry;
1660         struct hlist_head *head;
1661         int size, __size, dsize;
1662         int rctx;
1663
1664         if (bpf_prog_array_valid(call)) {
1665                 unsigned long orig_ip = instruction_pointer(regs);
1666                 int ret;
1667
1668                 ret = trace_call_bpf(call, regs);
1669
1670                 /*
1671                  * We need to check and see if we modified the pc of the
1672                  * pt_regs, and if so return 1 so that we don't do the
1673                  * single stepping.
1674                  */
1675                 if (orig_ip != instruction_pointer(regs))
1676                         return 1;
1677                 if (!ret)
1678                         return 0;
1679         }
1680
1681         head = this_cpu_ptr(call->perf_events);
1682         if (hlist_empty(head))
1683                 return 0;
1684
1685         dsize = __get_data_size(&tk->tp, regs, NULL);
1686         __size = sizeof(*entry) + tk->tp.size + dsize;
1687         size = ALIGN(__size + sizeof(u32), sizeof(u64));
1688         size -= sizeof(u32);
1689
1690         entry = perf_trace_buf_alloc(size, NULL, &rctx);
1691         if (!entry)
1692                 return 0;
1693
1694         entry->ip = (unsigned long)tk->rp.kp.addr;
1695         memset(&entry[1], 0, dsize);
1696         store_trace_args(&entry[1], &tk->tp, regs, NULL, sizeof(*entry), dsize);
1697         perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1698                               head, NULL);
1699         return 0;
1700 }
1701 NOKPROBE_SYMBOL(kprobe_perf_func);
1702
1703 /* Kretprobe profile handler */
1704 static void
1705 kretprobe_perf_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1706                     struct pt_regs *regs)
1707 {
1708         struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1709         struct kretprobe_trace_entry_head *entry;
1710         struct hlist_head *head;
1711         int size, __size, dsize;
1712         int rctx;
1713
1714         if (bpf_prog_array_valid(call) && !trace_call_bpf(call, regs))
1715                 return;
1716
1717         head = this_cpu_ptr(call->perf_events);
1718         if (hlist_empty(head))
1719                 return;
1720
1721         dsize = __get_data_size(&tk->tp, regs, ri->data);
1722         __size = sizeof(*entry) + tk->tp.size + dsize;
1723         size = ALIGN(__size + sizeof(u32), sizeof(u64));
1724         size -= sizeof(u32);
1725
1726         entry = perf_trace_buf_alloc(size, NULL, &rctx);
1727         if (!entry)
1728                 return;
1729
1730         entry->func = (unsigned long)tk->rp.kp.addr;
1731         entry->ret_ip = get_kretprobe_retaddr(ri);
1732         store_trace_args(&entry[1], &tk->tp, regs, ri->data, sizeof(*entry), dsize);
1733         perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1734                               head, NULL);
1735 }
1736 NOKPROBE_SYMBOL(kretprobe_perf_func);
1737
1738 int bpf_get_kprobe_info(const struct perf_event *event, u32 *fd_type,
1739                         const char **symbol, u64 *probe_offset,
1740                         u64 *probe_addr, unsigned long *missed,
1741                         bool perf_type_tracepoint)
1742 {
1743         const char *pevent = trace_event_name(event->tp_event);
1744         const char *group = event->tp_event->class->system;
1745         struct trace_kprobe *tk;
1746
1747         if (perf_type_tracepoint)
1748                 tk = find_trace_kprobe(pevent, group);
1749         else
1750                 tk = trace_kprobe_primary_from_call(event->tp_event);
1751         if (!tk)
1752                 return -EINVAL;
1753
1754         *fd_type = trace_kprobe_is_return(tk) ? BPF_FD_TYPE_KRETPROBE
1755                                               : BPF_FD_TYPE_KPROBE;
1756         *probe_offset = tk->rp.kp.offset;
1757         *probe_addr = kallsyms_show_value(current_cred()) ?
1758                       (unsigned long)tk->rp.kp.addr : 0;
1759         *symbol = tk->symbol;
1760         if (missed)
1761                 *missed = trace_kprobe_missed(tk);
1762         return 0;
1763 }
1764 #endif  /* CONFIG_PERF_EVENTS */
1765
1766 /*
1767  * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex.
1768  *
1769  * kprobe_trace_self_tests_init() does enable_trace_probe/disable_trace_probe
1770  * lockless, but we can't race with this __init function.
1771  */
1772 static int kprobe_register(struct trace_event_call *event,
1773                            enum trace_reg type, void *data)
1774 {
1775         struct trace_event_file *file = data;
1776
1777         switch (type) {
1778         case TRACE_REG_REGISTER:
1779                 return enable_trace_kprobe(event, file);
1780         case TRACE_REG_UNREGISTER:
1781                 return disable_trace_kprobe(event, file);
1782
1783 #ifdef CONFIG_PERF_EVENTS
1784         case TRACE_REG_PERF_REGISTER:
1785                 return enable_trace_kprobe(event, NULL);
1786         case TRACE_REG_PERF_UNREGISTER:
1787                 return disable_trace_kprobe(event, NULL);
1788         case TRACE_REG_PERF_OPEN:
1789         case TRACE_REG_PERF_CLOSE:
1790         case TRACE_REG_PERF_ADD:
1791         case TRACE_REG_PERF_DEL:
1792                 return 0;
1793 #endif
1794         }
1795         return 0;
1796 }
1797
1798 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1799 {
1800         struct trace_kprobe *tk = container_of(kp, struct trace_kprobe, rp.kp);
1801         int ret = 0;
1802
1803         raw_cpu_inc(*tk->nhit);
1804
1805         if (trace_probe_test_flag(&tk->tp, TP_FLAG_TRACE))
1806                 kprobe_trace_func(tk, regs);
1807 #ifdef CONFIG_PERF_EVENTS
1808         if (trace_probe_test_flag(&tk->tp, TP_FLAG_PROFILE))
1809                 ret = kprobe_perf_func(tk, regs);
1810 #endif
1811         return ret;
1812 }
1813 NOKPROBE_SYMBOL(kprobe_dispatcher);
1814
1815 static int
1816 kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1817 {
1818         struct kretprobe *rp = get_kretprobe(ri);
1819         struct trace_kprobe *tk;
1820
1821         /*
1822          * There is a small chance that get_kretprobe(ri) returns NULL when
1823          * the kretprobe is unregister on another CPU between kretprobe's
1824          * trampoline_handler and this function.
1825          */
1826         if (unlikely(!rp))
1827                 return 0;
1828
1829         tk = container_of(rp, struct trace_kprobe, rp);
1830         raw_cpu_inc(*tk->nhit);
1831
1832         if (trace_probe_test_flag(&tk->tp, TP_FLAG_TRACE))
1833                 kretprobe_trace_func(tk, ri, regs);
1834 #ifdef CONFIG_PERF_EVENTS
1835         if (trace_probe_test_flag(&tk->tp, TP_FLAG_PROFILE))
1836                 kretprobe_perf_func(tk, ri, regs);
1837 #endif
1838         return 0;       /* We don't tweak kernel, so just return 0 */
1839 }
1840 NOKPROBE_SYMBOL(kretprobe_dispatcher);
1841
1842 static struct trace_event_functions kretprobe_funcs = {
1843         .trace          = print_kretprobe_event
1844 };
1845
1846 static struct trace_event_functions kprobe_funcs = {
1847         .trace          = print_kprobe_event
1848 };
1849
1850 static struct trace_event_fields kretprobe_fields_array[] = {
1851         { .type = TRACE_FUNCTION_TYPE,
1852           .define_fields = kretprobe_event_define_fields },
1853         {}
1854 };
1855
1856 static struct trace_event_fields kprobe_fields_array[] = {
1857         { .type = TRACE_FUNCTION_TYPE,
1858           .define_fields = kprobe_event_define_fields },
1859         {}
1860 };
1861
1862 static inline void init_trace_event_call(struct trace_kprobe *tk)
1863 {
1864         struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1865
1866         if (trace_kprobe_is_return(tk)) {
1867                 call->event.funcs = &kretprobe_funcs;
1868                 call->class->fields_array = kretprobe_fields_array;
1869         } else {
1870                 call->event.funcs = &kprobe_funcs;
1871                 call->class->fields_array = kprobe_fields_array;
1872         }
1873
1874         call->flags = TRACE_EVENT_FL_KPROBE;
1875         call->class->reg = kprobe_register;
1876 }
1877
1878 static int register_kprobe_event(struct trace_kprobe *tk)
1879 {
1880         init_trace_event_call(tk);
1881
1882         return trace_probe_register_event_call(&tk->tp);
1883 }
1884
1885 static int unregister_kprobe_event(struct trace_kprobe *tk)
1886 {
1887         return trace_probe_unregister_event_call(&tk->tp);
1888 }
1889
1890 #ifdef CONFIG_PERF_EVENTS
1891
1892 /* create a trace_kprobe, but don't add it to global lists */
1893 struct trace_event_call *
1894 create_local_trace_kprobe(char *func, void *addr, unsigned long offs,
1895                           bool is_return)
1896 {
1897         enum probe_print_type ptype;
1898         struct trace_kprobe *tk __free(free_trace_kprobe) = NULL;
1899         int ret;
1900         char *event;
1901
1902         if (func) {
1903                 ret = validate_probe_symbol(func);
1904                 if (ret)
1905                         return ERR_PTR(ret);
1906         }
1907
1908         /*
1909          * local trace_kprobes are not added to dyn_event, so they are never
1910          * searched in find_trace_kprobe(). Therefore, there is no concern of
1911          * duplicated name here.
1912          */
1913         event = func ? func : "DUMMY_EVENT";
1914
1915         tk = alloc_trace_kprobe(KPROBE_EVENT_SYSTEM, event, (void *)addr, func,
1916                                 offs, 0 /* maxactive */, 0 /* nargs */,
1917                                 is_return);
1918
1919         if (IS_ERR(tk)) {
1920                 pr_info("Failed to allocate trace_probe.(%d)\n",
1921                         (int)PTR_ERR(tk));
1922                 return ERR_CAST(tk);
1923         }
1924
1925         init_trace_event_call(tk);
1926
1927         ptype = trace_kprobe_is_return(tk) ?
1928                 PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL;
1929         if (traceprobe_set_print_fmt(&tk->tp, ptype) < 0)
1930                 return ERR_PTR(-ENOMEM);
1931
1932         ret = __register_trace_kprobe(tk);
1933         if (ret < 0)
1934                 return ERR_PTR(ret);
1935
1936         return trace_probe_event_call(&(no_free_ptr(tk)->tp));
1937 }
1938
1939 void destroy_local_trace_kprobe(struct trace_event_call *event_call)
1940 {
1941         struct trace_kprobe *tk;
1942
1943         tk = trace_kprobe_primary_from_call(event_call);
1944         if (unlikely(!tk))
1945                 return;
1946
1947         if (trace_probe_is_enabled(&tk->tp)) {
1948                 WARN_ON(1);
1949                 return;
1950         }
1951
1952         __unregister_trace_kprobe(tk);
1953
1954         free_trace_kprobe(tk);
1955 }
1956 #endif /* CONFIG_PERF_EVENTS */
1957
1958 static __init void enable_boot_kprobe_events(void)
1959 {
1960         struct trace_array *tr = top_trace_array();
1961         struct trace_event_file *file;
1962         struct trace_kprobe *tk;
1963         struct dyn_event *pos;
1964
1965         guard(mutex)(&event_mutex);
1966         for_each_trace_kprobe(tk, pos) {
1967                 list_for_each_entry(file, &tr->events, list)
1968                         if (file->event_call == trace_probe_event_call(&tk->tp))
1969                                 trace_event_enable_disable(file, 1, 0);
1970         }
1971 }
1972
1973 static __init void setup_boot_kprobe_events(void)
1974 {
1975         char *p, *cmd = kprobe_boot_events_buf;
1976         int ret;
1977
1978         strreplace(kprobe_boot_events_buf, ',', ' ');
1979
1980         while (cmd && *cmd != '\0') {
1981                 p = strchr(cmd, ';');
1982                 if (p)
1983                         *p++ = '\0';
1984
1985                 ret = create_or_delete_trace_kprobe(cmd);
1986                 if (ret)
1987                         pr_warn("Failed to add event(%d): %s\n", ret, cmd);
1988
1989                 cmd = p;
1990         }
1991
1992         enable_boot_kprobe_events();
1993 }
1994
1995 /*
1996  * Register dynevent at core_initcall. This allows kernel to setup kprobe
1997  * events in postcore_initcall without tracefs.
1998  */
1999 static __init int init_kprobe_trace_early(void)
2000 {
2001         int ret;
2002
2003         ret = dyn_event_register(&trace_kprobe_ops);
2004         if (ret)
2005                 return ret;
2006
2007         if (trace_kprobe_register_module_notifier())
2008                 return -EINVAL;
2009
2010         return 0;
2011 }
2012 core_initcall(init_kprobe_trace_early);
2013
2014 /* Make a tracefs interface for controlling probe points */
2015 static __init int init_kprobe_trace(void)
2016 {
2017         int ret;
2018
2019         ret = tracing_init_dentry();
2020         if (ret)
2021                 return 0;
2022
2023         /* Event list interface */
2024         trace_create_file("kprobe_events", TRACE_MODE_WRITE,
2025                           NULL, NULL, &kprobe_events_ops);
2026
2027         /* Profile interface */
2028         trace_create_file("kprobe_profile", TRACE_MODE_READ,
2029                           NULL, NULL, &kprobe_profile_ops);
2030
2031         setup_boot_kprobe_events();
2032
2033         return 0;
2034 }
2035 fs_initcall(init_kprobe_trace);
2036
2037
2038 #ifdef CONFIG_FTRACE_STARTUP_TEST
2039 static __init struct trace_event_file *
2040 find_trace_probe_file(struct trace_kprobe *tk, struct trace_array *tr)
2041 {
2042         struct trace_event_file *file;
2043
2044         list_for_each_entry(file, &tr->events, list)
2045                 if (file->event_call == trace_probe_event_call(&tk->tp))
2046                         return file;
2047
2048         return NULL;
2049 }
2050
2051 /*
2052  * Nobody but us can call enable_trace_kprobe/disable_trace_kprobe at this
2053  * stage, we can do this lockless.
2054  */
2055 static __init int kprobe_trace_self_tests_init(void)
2056 {
2057         int ret, warn = 0;
2058         int (*target)(int, int, int, int, int, int);
2059         struct trace_kprobe *tk;
2060         struct trace_event_file *file;
2061
2062         if (tracing_is_disabled())
2063                 return -ENODEV;
2064
2065         if (tracing_selftest_disabled)
2066                 return 0;
2067
2068         target = kprobe_trace_selftest_target;
2069
2070         pr_info("Testing kprobe tracing: ");
2071
2072         ret = create_or_delete_trace_kprobe("p:testprobe kprobe_trace_selftest_target $stack $stack0 +0($stack)");
2073         if (WARN_ONCE(ret, "error on probing function entry.")) {
2074                 warn++;
2075         } else {
2076                 /* Enable trace point */
2077                 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
2078                 if (WARN_ONCE(tk == NULL, "error on probing function entry.")) {
2079                         warn++;
2080                 } else {
2081                         file = find_trace_probe_file(tk, top_trace_array());
2082                         if (WARN_ONCE(file == NULL, "error on getting probe file.")) {
2083                                 warn++;
2084                         } else
2085                                 enable_trace_kprobe(
2086                                         trace_probe_event_call(&tk->tp), file);
2087                 }
2088         }
2089
2090         ret = create_or_delete_trace_kprobe("r:testprobe2 kprobe_trace_selftest_target $retval");
2091         if (WARN_ONCE(ret, "error on probing function return.")) {
2092                 warn++;
2093         } else {
2094                 /* Enable trace point */
2095                 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
2096                 if (WARN_ONCE(tk == NULL, "error on getting 2nd new probe.")) {
2097                         warn++;
2098                 } else {
2099                         file = find_trace_probe_file(tk, top_trace_array());
2100                         if (WARN_ONCE(file == NULL, "error on getting probe file.")) {
2101                                 warn++;
2102                         } else
2103                                 enable_trace_kprobe(
2104                                         trace_probe_event_call(&tk->tp), file);
2105                 }
2106         }
2107
2108         if (warn)
2109                 goto end;
2110
2111         ret = target(1, 2, 3, 4, 5, 6);
2112
2113         /*
2114          * Not expecting an error here, the check is only to prevent the
2115          * optimizer from removing the call to target() as otherwise there
2116          * are no side-effects and the call is never performed.
2117          */
2118         if (ret != 21)
2119                 warn++;
2120
2121         /* Disable trace points before removing it */
2122         tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
2123         if (WARN_ONCE(tk == NULL, "error on getting test probe.")) {
2124                 warn++;
2125         } else {
2126                 if (WARN_ONCE(trace_kprobe_nhit(tk) != 1,
2127                                  "incorrect number of testprobe hits."))
2128                         warn++;
2129
2130                 file = find_trace_probe_file(tk, top_trace_array());
2131                 if (WARN_ONCE(file == NULL, "error on getting probe file.")) {
2132                         warn++;
2133                 } else
2134                         disable_trace_kprobe(
2135                                 trace_probe_event_call(&tk->tp), file);
2136         }
2137
2138         tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
2139         if (WARN_ONCE(tk == NULL, "error on getting 2nd test probe.")) {
2140                 warn++;
2141         } else {
2142                 if (WARN_ONCE(trace_kprobe_nhit(tk) != 1,
2143                                  "incorrect number of testprobe2 hits."))
2144                         warn++;
2145
2146                 file = find_trace_probe_file(tk, top_trace_array());
2147                 if (WARN_ONCE(file == NULL, "error on getting probe file.")) {
2148                         warn++;
2149                 } else
2150                         disable_trace_kprobe(
2151                                 trace_probe_event_call(&tk->tp), file);
2152         }
2153
2154         ret = create_or_delete_trace_kprobe("-:testprobe");
2155         if (WARN_ONCE(ret, "error on deleting a probe."))
2156                 warn++;
2157
2158         ret = create_or_delete_trace_kprobe("-:testprobe2");
2159         if (WARN_ONCE(ret, "error on deleting a probe."))
2160                 warn++;
2161
2162
2163 end:
2164         /*
2165          * Wait for the optimizer work to finish. Otherwise it might fiddle
2166          * with probes in already freed __init text.
2167          */
2168         wait_for_kprobe_optimizer();
2169         if (warn)
2170                 pr_cont("NG: Some tests are failed. Please check them.\n");
2171         else
2172                 pr_cont("OK\n");
2173         return 0;
2174 }
2175
2176 late_initcall(kprobe_trace_self_tests_init);
2177
2178 #endif