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/error-injection.h>
  13 #include <linux/module.h>
  14 #include <linux/rculist.h>
  15 #include <linux/security.h>
  16 #include <linux/uaccess.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_kernel.h"
  24 #include "trace_probe_tmpl.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         const char *event = NULL, *group = KPROBE_EVENT_SYSTEM;
 865         const char **new_argv __free(kfree) = NULL;
 866         int i, len, new_argc = 0, ret = 0;
 867         char *symbol __free(kfree) = NULL;
 868         char *ebuf __free(kfree) = NULL;
 869         char *gbuf __free(kfree) = NULL;
 870         char *abuf __free(kfree) = NULL;
 871         char *dbuf __free(kfree) = NULL;
 872         enum probe_print_type ptype;
 873         bool is_return = false;
 874         int maxactive = 0;
 875         void *addr = NULL;
 876         char *tmp = NULL;
 877         long offset = 0;
 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                 char *buf __free(kfree) = NULL;
 897
 898                 if (!is_return) {
 899                         trace_probe_log_err(1, BAD_MAXACT_TYPE);
 900                         return -EINVAL;
 901                 }
 902                 if (event)
 903                         len = event - &argv[0][1] - 1;
 904                 else
 905                         len = strlen(&argv[0][1]);
 906                 if (len > MAX_EVENT_NAME_LEN - 1) {
 907                         trace_probe_log_err(1, BAD_MAXACT);
 908                         return -EINVAL;
 909                 }
 910                 buf = kmemdup(&argv[0][1], len + 1, GFP_KERNEL);
 911                 buf[len] = '\0';
 912                 ret = kstrtouint(buf, 0, &maxactive);
 913                 if (ret || !maxactive) {
 914                         trace_probe_log_err(1, BAD_MAXACT);
 915                         return -EINVAL;
 916                 }
 917                 /* kretprobes instances are iterated over via a list. The
 918                  * maximum should stay reasonable.
 919                  */
 920                 if (maxactive > KRETPROBE_MAXACTIVE_MAX) {
 921                         trace_probe_log_err(1, MAXACT_TOO_BIG);
 922                         return -EINVAL;
 923                 }
 924         }
 925
 926         /* try to parse an address. if that fails, try to read the
 927          * input as a symbol. */
 928         if (kstrtoul(argv[1], 0, (unsigned long *)&addr)) {
 929                 trace_probe_log_set_index(1);
 930                 /* Check whether uprobe event specified */
 931                 if (strchr(argv[1], '/') && strchr(argv[1], ':'))
 932                         return -ECANCELED;
 933
 934                 /* a symbol specified */
 935                 symbol = kstrdup(argv[1], GFP_KERNEL);
 936                 if (!symbol)
 937                         return -ENOMEM;
 938
 939                 tmp = strchr(symbol, '%');
 940                 if (tmp) {
 941                         if (!strcmp(tmp, "%return")) {
 942                                 *tmp = '\0';
 943                                 is_return = true;
 944                         } else {
 945                                 trace_probe_log_err(tmp - symbol, BAD_ADDR_SUFFIX);
 946                                 return -EINVAL;
 947                         }
 948                 }
 949
 950                 /* TODO: support .init module functions */
 951                 ret = traceprobe_split_symbol_offset(symbol, &offset);
 952                 if (ret || offset < 0 || offset > UINT_MAX) {
 953                         trace_probe_log_err(0, BAD_PROBE_ADDR);
 954                         return -EINVAL;
 955                 }
 956                 ret = validate_probe_symbol(symbol);
 957                 if (ret) {
 958                         if (ret == -EADDRNOTAVAIL)
 959                                 trace_probe_log_err(0, NON_UNIQ_SYMBOL);
 960                         else
 961                                 trace_probe_log_err(0, BAD_PROBE_ADDR);
 962                         return -EINVAL;
 963                 }
 964                 if (is_return)
 965                         ctx->flags |= TPARG_FL_RETURN;
 966                 ret = kprobe_on_func_entry(NULL, symbol, offset);
 967                 if (ret == 0 && !is_return)
 968                         ctx->flags |= TPARG_FL_FENTRY;
 969                 /* Defer the ENOENT case until register kprobe */
 970                 if (ret == -EINVAL && is_return) {
 971                         trace_probe_log_err(0, BAD_RETPROBE);
 972                         return -EINVAL;
 973                 }
 974         }
 975
 976         trace_probe_log_set_index(0);
 977         if (event) {
 978                 gbuf = kmalloc(MAX_EVENT_NAME_LEN, GFP_KERNEL);
 979                 if (!gbuf)
 980                         return -ENOMEM;
 981                 ret = traceprobe_parse_event_name(&event, &group, gbuf,
 982                                                   event - argv[0]);
 983                 if (ret)
 984                         return ret;
 985         }
 986
 987         if (!event) {
 988                 /* Make a new event name */
 989                 ebuf = kmalloc(MAX_EVENT_NAME_LEN, GFP_KERNEL);
 990                 if (!ebuf)
 991                         return -ENOMEM;
 992                 if (symbol)
 993                         snprintf(ebuf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
 994                                  is_return ? 'r' : 'p', symbol, offset);
 995                 else
 996                         snprintf(ebuf, MAX_EVENT_NAME_LEN, "%c_0x%p",
 997                                  is_return ? 'r' : 'p', addr);
 998                 sanitize_event_name(ebuf);
 999                 event = ebuf;
1000         }
1001
1002         abuf = kmalloc(MAX_BTF_ARGS_LEN, GFP_KERNEL);
1003         if (!abuf)
1004                 return -ENOMEM;
1005         argc -= 2; argv += 2;
1006         ctx->funcname = symbol;
1007         new_argv = traceprobe_expand_meta_args(argc, argv, &new_argc,
1008                                                abuf, MAX_BTF_ARGS_LEN, ctx);
1009         if (IS_ERR(new_argv)) {
1010                 ret = PTR_ERR(new_argv);
1011                 new_argv = NULL;
1012                 return ret;
1013         }
1014         if (new_argv) {
1015                 argc = new_argc;
1016                 argv = new_argv;
1017         }
1018         if (argc > MAX_TRACE_ARGS) {
1019                 trace_probe_log_set_index(2);
1020                 trace_probe_log_err(0, TOO_MANY_ARGS);
1021                 return -E2BIG;
1022         }
1023
1024         ret = traceprobe_expand_dentry_args(argc, argv, &dbuf);
1025         if (ret)
1026                 return ret;
1027
1028         /* setup a probe */
1029         tk = alloc_trace_kprobe(group, event, addr, symbol, offset, maxactive,
1030                                 argc, is_return);
1031         if (IS_ERR(tk)) {
1032                 ret = PTR_ERR(tk);
1033                 /* This must return -ENOMEM, else there is a bug */
1034                 WARN_ON_ONCE(ret != -ENOMEM);
1035                 return ret;     /* We know tk is not allocated */
1036         }
1037
1038         /* parse arguments */
1039         for (i = 0; i < argc; i++) {
1040                 trace_probe_log_set_index(i + 2);
1041                 ctx->offset = 0;
1042                 ret = traceprobe_parse_probe_arg(&tk->tp, i, argv[i], ctx);
1043                 if (ret)
1044                         return ret;     /* This can be -ENOMEM */
1045         }
1046         /* entry handler for kretprobe */
1047         if (is_return && tk->tp.entry_arg) {
1048                 tk->rp.entry_handler = trace_kprobe_entry_handler;
1049                 tk->rp.data_size = traceprobe_get_entry_data_size(&tk->tp);
1050         }
1051
1052         ptype = is_return ? PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL;
1053         ret = traceprobe_set_print_fmt(&tk->tp, ptype);
1054         if (ret < 0)
1055                 return ret;
1056
1057         ret = register_trace_kprobe(tk);
1058         if (ret) {
1059                 trace_probe_log_set_index(1);
1060                 if (ret == -EILSEQ)
1061                         trace_probe_log_err(0, BAD_INSN_BNDRY);
1062                 else if (ret == -ENOENT)
1063                         trace_probe_log_err(0, BAD_PROBE_ADDR);
1064                 else if (ret != -ENOMEM && ret != -EEXIST)
1065                         trace_probe_log_err(0, FAIL_REG_PROBE);
1066                 return ret;
1067         }
1068         /*
1069          * Here, 'tk' has been registered to the list successfully,
1070          * so we don't need to free it.
1071          */
1072         tk = NULL;
1073
1074         return 0;
1075 }
1076
1077 static int trace_kprobe_create_cb(int argc, const char *argv[])
1078 {
1079         struct traceprobe_parse_context *ctx __free(traceprobe_parse_context) = NULL;
1080         int ret;
1081
1082         ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
1083         if (!ctx)
1084                 return -ENOMEM;
1085         ctx->flags = TPARG_FL_KERNEL;
1086
1087         trace_probe_log_init("trace_kprobe", argc, argv);
1088
1089         ret = trace_kprobe_create_internal(argc, argv, ctx);
1090
1091         trace_probe_log_clear();
1092         return ret;
1093 }
1094
1095 static int trace_kprobe_create(const char *raw_command)
1096 {
1097         return trace_probe_create(raw_command, trace_kprobe_create_cb);
1098 }
1099
1100 static int create_or_delete_trace_kprobe(const char *raw_command)
1101 {
1102         int ret;
1103
1104         if (raw_command[0] == '-')
1105                 return dyn_event_release(raw_command, &trace_kprobe_ops);
1106
1107         ret = dyn_event_create(raw_command, &trace_kprobe_ops);
1108         return ret == -ECANCELED ? -EINVAL : ret;
1109 }
1110
1111 static int trace_kprobe_run_command(struct dynevent_cmd *cmd)
1112 {
1113         return create_or_delete_trace_kprobe(cmd->seq.buffer);
1114 }
1115
1116 /**
1117  * kprobe_event_cmd_init - Initialize a kprobe event command object
1118  * @cmd: A pointer to the dynevent_cmd struct representing the new event
1119  * @buf: A pointer to the buffer used to build the command
1120  * @maxlen: The length of the buffer passed in @buf
1121  *
1122  * Initialize a synthetic event command object.  Use this before
1123  * calling any of the other kprobe_event functions.
1124  */
1125 void kprobe_event_cmd_init(struct dynevent_cmd *cmd, char *buf, int maxlen)
1126 {
1127         dynevent_cmd_init(cmd, buf, maxlen, DYNEVENT_TYPE_KPROBE,
1128                           trace_kprobe_run_command);
1129 }
1130 EXPORT_SYMBOL_GPL(kprobe_event_cmd_init);
1131
1132 /**
1133  * __kprobe_event_gen_cmd_start - Generate a kprobe event command from arg list
1134  * @cmd: A pointer to the dynevent_cmd struct representing the new event
1135  * @kretprobe: Is this a return probe?
1136  * @name: The name of the kprobe event
1137  * @loc: The location of the kprobe event
1138  * @...: Variable number of arg (pairs), one pair for each field
1139  *
1140  * NOTE: Users normally won't want to call this function directly, but
1141  * rather use the kprobe_event_gen_cmd_start() wrapper, which automatically
1142  * adds a NULL to the end of the arg list.  If this function is used
1143  * directly, make sure the last arg in the variable arg list is NULL.
1144  *
1145  * Generate a kprobe event command to be executed by
1146  * kprobe_event_gen_cmd_end().  This function can be used to generate the
1147  * complete command or only the first part of it; in the latter case,
1148  * kprobe_event_add_fields() can be used to add more fields following this.
1149  *
1150  * Unlikely the synth_event_gen_cmd_start(), @loc must be specified. This
1151  * returns -EINVAL if @loc == NULL.
1152  *
1153  * Return: 0 if successful, error otherwise.
1154  */
1155 int __kprobe_event_gen_cmd_start(struct dynevent_cmd *cmd, bool kretprobe,
1156                                  const char *name, const char *loc, ...)
1157 {
1158         char buf[MAX_EVENT_NAME_LEN];
1159         struct dynevent_arg arg;
1160         va_list args;
1161         int ret;
1162
1163         if (cmd->type != DYNEVENT_TYPE_KPROBE)
1164                 return -EINVAL;
1165
1166         if (!loc)
1167                 return -EINVAL;
1168
1169         if (kretprobe)
1170                 snprintf(buf, MAX_EVENT_NAME_LEN, "r:kprobes/%s", name);
1171         else
1172                 snprintf(buf, MAX_EVENT_NAME_LEN, "p:kprobes/%s", name);
1173
1174         ret = dynevent_str_add(cmd, buf);
1175         if (ret)
1176                 return ret;
1177
1178         dynevent_arg_init(&arg, 0);
1179         arg.str = loc;
1180         ret = dynevent_arg_add(cmd, &arg, NULL);
1181         if (ret)
1182                 return ret;
1183
1184         va_start(args, loc);
1185         for (;;) {
1186                 const char *field;
1187
1188                 field = va_arg(args, const char *);
1189                 if (!field)
1190                         break;
1191
1192                 if (++cmd->n_fields > MAX_TRACE_ARGS) {
1193                         ret = -EINVAL;
1194                         break;
1195                 }
1196
1197                 arg.str = field;
1198                 ret = dynevent_arg_add(cmd, &arg, NULL);
1199                 if (ret)
1200                         break;
1201         }
1202         va_end(args);
1203
1204         return ret;
1205 }
1206 EXPORT_SYMBOL_GPL(__kprobe_event_gen_cmd_start);
1207
1208 /**
1209  * __kprobe_event_add_fields - Add probe fields to a kprobe command from arg list
1210  * @cmd: A pointer to the dynevent_cmd struct representing the new event
1211  * @...: Variable number of arg (pairs), one pair for each field
1212  *
1213  * NOTE: Users normally won't want to call this function directly, but
1214  * rather use the kprobe_event_add_fields() wrapper, which
1215  * automatically adds a NULL to the end of the arg list.  If this
1216  * function is used directly, make sure the last arg in the variable
1217  * arg list is NULL.
1218  *
1219  * Add probe fields to an existing kprobe command using a variable
1220  * list of args.  Fields are added in the same order they're listed.
1221  *
1222  * Return: 0 if successful, error otherwise.
1223  */
1224 int __kprobe_event_add_fields(struct dynevent_cmd *cmd, ...)
1225 {
1226         struct dynevent_arg arg;
1227         va_list args;
1228         int ret = 0;
1229
1230         if (cmd->type != DYNEVENT_TYPE_KPROBE)
1231                 return -EINVAL;
1232
1233         dynevent_arg_init(&arg, 0);
1234
1235         va_start(args, cmd);
1236         for (;;) {
1237                 const char *field;
1238
1239                 field = va_arg(args, const char *);
1240                 if (!field)
1241                         break;
1242
1243                 if (++cmd->n_fields > MAX_TRACE_ARGS) {
1244                         ret = -EINVAL;
1245                         break;
1246                 }
1247
1248                 arg.str = field;
1249                 ret = dynevent_arg_add(cmd, &arg, NULL);
1250                 if (ret)
1251                         break;
1252         }
1253         va_end(args);
1254
1255         return ret;
1256 }
1257 EXPORT_SYMBOL_GPL(__kprobe_event_add_fields);
1258
1259 /**
1260  * kprobe_event_delete - Delete a kprobe event
1261  * @name: The name of the kprobe event to delete
1262  *
1263  * Delete a kprobe event with the give @name from kernel code rather
1264  * than directly from the command line.
1265  *
1266  * Return: 0 if successful, error otherwise.
1267  */
1268 int kprobe_event_delete(const char *name)
1269 {
1270         char buf[MAX_EVENT_NAME_LEN];
1271
1272         snprintf(buf, MAX_EVENT_NAME_LEN, "-:%s", name);
1273
1274         return create_or_delete_trace_kprobe(buf);
1275 }
1276 EXPORT_SYMBOL_GPL(kprobe_event_delete);
1277
1278 static int trace_kprobe_release(struct dyn_event *ev)
1279 {
1280         struct trace_kprobe *tk = to_trace_kprobe(ev);
1281         int ret = unregister_trace_kprobe(tk);
1282
1283         if (!ret)
1284                 free_trace_kprobe(tk);
1285         return ret;
1286 }
1287
1288 static int trace_kprobe_show(struct seq_file *m, struct dyn_event *ev)
1289 {
1290         struct trace_kprobe *tk = to_trace_kprobe(ev);
1291         int i;
1292
1293         seq_putc(m, trace_kprobe_is_return(tk) ? 'r' : 'p');
1294         if (trace_kprobe_is_return(tk) && tk->rp.maxactive)
1295                 seq_printf(m, "%d", tk->rp.maxactive);
1296         seq_printf(m, ":%s/%s", trace_probe_group_name(&tk->tp),
1297                                 trace_probe_name(&tk->tp));
1298
1299         if (!tk->symbol)
1300                 seq_printf(m, " 0x%p", tk->rp.kp.addr);
1301         else if (tk->rp.kp.offset)
1302                 seq_printf(m, " %s+%u", trace_kprobe_symbol(tk),
1303                            tk->rp.kp.offset);
1304         else
1305                 seq_printf(m, " %s", trace_kprobe_symbol(tk));
1306
1307         for (i = 0; i < tk->tp.nr_args; i++)
1308                 seq_printf(m, " %s=%s", tk->tp.args[i].name, tk->tp.args[i].comm);
1309         seq_putc(m, '\n');
1310
1311         return 0;
1312 }
1313
1314 static int probes_seq_show(struct seq_file *m, void *v)
1315 {
1316         struct dyn_event *ev = v;
1317
1318         if (!is_trace_kprobe(ev))
1319                 return 0;
1320
1321         return trace_kprobe_show(m, ev);
1322 }
1323
1324 static const struct seq_operations probes_seq_op = {
1325         .start  = dyn_event_seq_start,
1326         .next   = dyn_event_seq_next,
1327         .stop   = dyn_event_seq_stop,
1328         .show   = probes_seq_show
1329 };
1330
1331 static int probes_open(struct inode *inode, struct file *file)
1332 {
1333         int ret;
1334
1335         ret = security_locked_down(LOCKDOWN_TRACEFS);
1336         if (ret)
1337                 return ret;
1338
1339         if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
1340                 ret = dyn_events_release_all(&trace_kprobe_ops);
1341                 if (ret < 0)
1342                         return ret;
1343         }
1344
1345         return seq_open(file, &probes_seq_op);
1346 }
1347
1348 static ssize_t probes_write(struct file *file, const char __user *buffer,
1349                             size_t count, loff_t *ppos)
1350 {
1351         return trace_parse_run_command(file, buffer, count, ppos,
1352                                        create_or_delete_trace_kprobe);
1353 }
1354
1355 static const struct file_operations kprobe_events_ops = {
1356         .owner          = THIS_MODULE,
1357         .open           = probes_open,
1358         .read           = seq_read,
1359         .llseek         = seq_lseek,
1360         .release        = seq_release,
1361         .write          = probes_write,
1362 };
1363
1364 static unsigned long trace_kprobe_missed(struct trace_kprobe *tk)
1365 {
1366         return trace_kprobe_is_return(tk) ?
1367                 tk->rp.kp.nmissed + tk->rp.nmissed : tk->rp.kp.nmissed;
1368 }
1369
1370 /* Probes profiling interfaces */
1371 static int probes_profile_seq_show(struct seq_file *m, void *v)
1372 {
1373         struct dyn_event *ev = v;
1374         struct trace_kprobe *tk;
1375         unsigned long nmissed;
1376
1377         if (!is_trace_kprobe(ev))
1378                 return 0;
1379
1380         tk = to_trace_kprobe(ev);
1381         nmissed = trace_kprobe_missed(tk);
1382         seq_printf(m, "  %-44s %15lu %15lu\n",
1383                    trace_probe_name(&tk->tp),
1384                    trace_kprobe_nhit(tk),
1385                    nmissed);
1386
1387         return 0;
1388 }
1389
1390 static const struct seq_operations profile_seq_op = {
1391         .start  = dyn_event_seq_start,
1392         .next   = dyn_event_seq_next,
1393         .stop   = dyn_event_seq_stop,
1394         .show   = probes_profile_seq_show
1395 };
1396
1397 static int profile_open(struct inode *inode, struct file *file)
1398 {
1399         int ret;
1400
1401         ret = security_locked_down(LOCKDOWN_TRACEFS);
1402         if (ret)
1403                 return ret;
1404
1405         return seq_open(file, &profile_seq_op);
1406 }
1407
1408 static const struct file_operations kprobe_profile_ops = {
1409         .owner          = THIS_MODULE,
1410         .open           = profile_open,
1411         .read           = seq_read,
1412         .llseek         = seq_lseek,
1413         .release        = seq_release,
1414 };
1415
1416 /* Note that we don't verify it, since the code does not come from user space */
1417 static int
1418 process_fetch_insn(struct fetch_insn *code, void *rec, void *edata,
1419                    void *dest, void *base)
1420 {
1421         struct pt_regs *regs = rec;
1422         unsigned long val;
1423         int ret;
1424
1425 retry:
1426         /* 1st stage: get value from context */
1427         switch (code->op) {
1428         case FETCH_OP_REG:
1429                 val = regs_get_register(regs, code->param);
1430                 break;
1431         case FETCH_OP_STACK:
1432                 val = regs_get_kernel_stack_nth(regs, code->param);
1433                 break;
1434         case FETCH_OP_STACKP:
1435                 val = kernel_stack_pointer(regs);
1436                 break;
1437         case FETCH_OP_RETVAL:
1438                 val = regs_return_value(regs);
1439                 break;
1440 #ifdef CONFIG_HAVE_FUNCTION_ARG_ACCESS_API
1441         case FETCH_OP_ARG:
1442                 val = regs_get_kernel_argument(regs, code->param);
1443                 break;
1444         case FETCH_OP_EDATA:
1445                 val = *(unsigned long *)((unsigned long)edata + code->offset);
1446                 break;
1447 #endif
1448         case FETCH_NOP_SYMBOL:  /* Ignore a place holder */
1449                 code++;
1450                 goto retry;
1451         default:
1452                 ret = process_common_fetch_insn(code, &val);
1453                 if (ret < 0)
1454                         return ret;
1455         }
1456         code++;
1457
1458         return process_fetch_insn_bottom(code, val, dest, base);
1459 }
1460 NOKPROBE_SYMBOL(process_fetch_insn)
1461
1462 /* Kprobe handler */
1463 static nokprobe_inline void
1464 __kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs,
1465                     struct trace_event_file *trace_file)
1466 {
1467         struct kprobe_trace_entry_head *entry;
1468         struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1469         struct trace_event_buffer fbuffer;
1470         int dsize;
1471
1472         WARN_ON(call != trace_file->event_call);
1473
1474         if (trace_trigger_soft_disabled(trace_file))
1475                 return;
1476
1477         dsize = __get_data_size(&tk->tp, regs, NULL);
1478
1479         entry = trace_event_buffer_reserve(&fbuffer, trace_file,
1480                                            sizeof(*entry) + tk->tp.size + dsize);
1481         if (!entry)
1482                 return;
1483
1484         fbuffer.regs = regs;
1485         entry->ip = (unsigned long)tk->rp.kp.addr;
1486         store_trace_args(&entry[1], &tk->tp, regs, NULL, sizeof(*entry), dsize);
1487
1488         trace_event_buffer_commit(&fbuffer);
1489 }
1490
1491 static void
1492 kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs)
1493 {
1494         struct event_file_link *link;
1495
1496         trace_probe_for_each_link_rcu(link, &tk->tp)
1497                 __kprobe_trace_func(tk, regs, link->file);
1498 }
1499 NOKPROBE_SYMBOL(kprobe_trace_func);
1500
1501 /* Kretprobe handler */
1502
1503 static int trace_kprobe_entry_handler(struct kretprobe_instance *ri,
1504                                       struct pt_regs *regs)
1505 {
1506         struct kretprobe *rp = get_kretprobe(ri);
1507         struct trace_kprobe *tk;
1508
1509         /*
1510          * There is a small chance that get_kretprobe(ri) returns NULL when
1511          * the kretprobe is unregister on another CPU between kretprobe's
1512          * trampoline_handler and this function.
1513          */
1514         if (unlikely(!rp))
1515                 return -ENOENT;
1516
1517         tk = container_of(rp, struct trace_kprobe, rp);
1518
1519         /* store argument values into ri->data as entry data */
1520         if (tk->tp.entry_arg)
1521                 store_trace_entry_data(ri->data, &tk->tp, regs);
1522
1523         return 0;
1524 }
1525
1526
1527 static nokprobe_inline void
1528 __kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1529                        struct pt_regs *regs,
1530                        struct trace_event_file *trace_file)
1531 {
1532         struct kretprobe_trace_entry_head *entry;
1533         struct trace_event_buffer fbuffer;
1534         struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1535         int dsize;
1536
1537         WARN_ON(call != trace_file->event_call);
1538
1539         if (trace_trigger_soft_disabled(trace_file))
1540                 return;
1541
1542         dsize = __get_data_size(&tk->tp, regs, ri->data);
1543
1544         entry = trace_event_buffer_reserve(&fbuffer, trace_file,
1545                                            sizeof(*entry) + tk->tp.size + dsize);
1546         if (!entry)
1547                 return;
1548
1549         fbuffer.regs = regs;
1550         entry->func = (unsigned long)tk->rp.kp.addr;
1551         entry->ret_ip = get_kretprobe_retaddr(ri);
1552         store_trace_args(&entry[1], &tk->tp, regs, ri->data, sizeof(*entry), dsize);
1553
1554         trace_event_buffer_commit(&fbuffer);
1555 }
1556
1557 static void
1558 kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1559                      struct pt_regs *regs)
1560 {
1561         struct event_file_link *link;
1562
1563         trace_probe_for_each_link_rcu(link, &tk->tp)
1564                 __kretprobe_trace_func(tk, ri, regs, link->file);
1565 }
1566 NOKPROBE_SYMBOL(kretprobe_trace_func);
1567
1568 /* Event entry printers */
1569 static enum print_line_t
1570 print_kprobe_event(struct trace_iterator *iter, int flags,
1571                    struct trace_event *event)
1572 {
1573         struct kprobe_trace_entry_head *field;
1574         struct trace_seq *s = &iter->seq;
1575         struct trace_probe *tp;
1576
1577         field = (struct kprobe_trace_entry_head *)iter->ent;
1578         tp = trace_probe_primary_from_call(
1579                 container_of(event, struct trace_event_call, event));
1580         if (WARN_ON_ONCE(!tp))
1581                 goto out;
1582
1583         trace_seq_printf(s, "%s: (", trace_probe_name(tp));
1584
1585         if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
1586                 goto out;
1587
1588         trace_seq_putc(s, ')');
1589
1590         if (trace_probe_print_args(s, tp->args, tp->nr_args,
1591                              (u8 *)&field[1], field) < 0)
1592                 goto out;
1593
1594         trace_seq_putc(s, '\n');
1595  out:
1596         return trace_handle_return(s);
1597 }
1598
1599 static enum print_line_t
1600 print_kretprobe_event(struct trace_iterator *iter, int flags,
1601                       struct trace_event *event)
1602 {
1603         struct kretprobe_trace_entry_head *field;
1604         struct trace_seq *s = &iter->seq;
1605         struct trace_probe *tp;
1606
1607         field = (struct kretprobe_trace_entry_head *)iter->ent;
1608         tp = trace_probe_primary_from_call(
1609                 container_of(event, struct trace_event_call, event));
1610         if (WARN_ON_ONCE(!tp))
1611                 goto out;
1612
1613         trace_seq_printf(s, "%s: (", trace_probe_name(tp));
1614
1615         if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1616                 goto out;
1617
1618         trace_seq_puts(s, " <- ");
1619
1620         if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1621                 goto out;
1622
1623         trace_seq_putc(s, ')');
1624
1625         if (trace_probe_print_args(s, tp->args, tp->nr_args,
1626                              (u8 *)&field[1], field) < 0)
1627                 goto out;
1628
1629         trace_seq_putc(s, '\n');
1630
1631  out:
1632         return trace_handle_return(s);
1633 }
1634
1635
1636 static int kprobe_event_define_fields(struct trace_event_call *event_call)
1637 {
1638         int ret;
1639         struct kprobe_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, ip, FIELD_STRING_IP, 0);
1647
1648         return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
1649 }
1650
1651 static int kretprobe_event_define_fields(struct trace_event_call *event_call)
1652 {
1653         int ret;
1654         struct kretprobe_trace_entry_head field;
1655         struct trace_probe *tp;
1656
1657         tp = trace_probe_primary_from_call(event_call);
1658         if (WARN_ON_ONCE(!tp))
1659                 return -ENOENT;
1660
1661         DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1662         DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1663
1664         return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
1665 }
1666
1667 #ifdef CONFIG_PERF_EVENTS
1668
1669 /* Kprobe profile handler */
1670 static int
1671 kprobe_perf_func(struct trace_kprobe *tk, struct pt_regs *regs)
1672 {
1673         struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1674         struct kprobe_trace_entry_head *entry;
1675         struct hlist_head *head;
1676         int size, __size, dsize;
1677         int rctx;
1678
1679         if (bpf_prog_array_valid(call)) {
1680                 unsigned long orig_ip = instruction_pointer(regs);
1681                 int ret;
1682
1683                 ret = trace_call_bpf(call, regs);
1684
1685                 /*
1686                  * We need to check and see if we modified the pc of the
1687                  * pt_regs, and if so return 1 so that we don't do the
1688                  * single stepping.
1689                  */
1690                 if (orig_ip != instruction_pointer(regs))
1691                         return 1;
1692                 if (!ret)
1693                         return 0;
1694         }
1695
1696         head = this_cpu_ptr(call->perf_events);
1697         if (hlist_empty(head))
1698                 return 0;
1699
1700         dsize = __get_data_size(&tk->tp, regs, NULL);
1701         __size = sizeof(*entry) + tk->tp.size + dsize;
1702         size = ALIGN(__size + sizeof(u32), sizeof(u64));
1703         size -= sizeof(u32);
1704
1705         entry = perf_trace_buf_alloc(size, NULL, &rctx);
1706         if (!entry)
1707                 return 0;
1708
1709         entry->ip = (unsigned long)tk->rp.kp.addr;
1710         memset(&entry[1], 0, dsize);
1711         store_trace_args(&entry[1], &tk->tp, regs, NULL, sizeof(*entry), dsize);
1712         perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1713                               head, NULL);
1714         return 0;
1715 }
1716 NOKPROBE_SYMBOL(kprobe_perf_func);
1717
1718 /* Kretprobe profile handler */
1719 static void
1720 kretprobe_perf_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1721                     struct pt_regs *regs)
1722 {
1723         struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1724         struct kretprobe_trace_entry_head *entry;
1725         struct hlist_head *head;
1726         int size, __size, dsize;
1727         int rctx;
1728
1729         if (bpf_prog_array_valid(call) && !trace_call_bpf(call, regs))
1730                 return;
1731
1732         head = this_cpu_ptr(call->perf_events);
1733         if (hlist_empty(head))
1734                 return;
1735
1736         dsize = __get_data_size(&tk->tp, regs, ri->data);
1737         __size = sizeof(*entry) + tk->tp.size + dsize;
1738         size = ALIGN(__size + sizeof(u32), sizeof(u64));
1739         size -= sizeof(u32);
1740
1741         entry = perf_trace_buf_alloc(size, NULL, &rctx);
1742         if (!entry)
1743                 return;
1744
1745         entry->func = (unsigned long)tk->rp.kp.addr;
1746         entry->ret_ip = get_kretprobe_retaddr(ri);
1747         store_trace_args(&entry[1], &tk->tp, regs, ri->data, sizeof(*entry), dsize);
1748         perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1749                               head, NULL);
1750 }
1751 NOKPROBE_SYMBOL(kretprobe_perf_func);
1752
1753 int bpf_get_kprobe_info(const struct perf_event *event, u32 *fd_type,
1754                         const char **symbol, u64 *probe_offset,
1755                         u64 *probe_addr, unsigned long *missed,
1756                         bool perf_type_tracepoint)
1757 {
1758         const char *pevent = trace_event_name(event->tp_event);
1759         const char *group = event->tp_event->class->system;
1760         struct trace_kprobe *tk;
1761
1762         if (perf_type_tracepoint)
1763                 tk = find_trace_kprobe(pevent, group);
1764         else
1765                 tk = trace_kprobe_primary_from_call(event->tp_event);
1766         if (!tk)
1767                 return -EINVAL;
1768
1769         *fd_type = trace_kprobe_is_return(tk) ? BPF_FD_TYPE_KRETPROBE
1770                                               : BPF_FD_TYPE_KPROBE;
1771         *probe_offset = tk->rp.kp.offset;
1772         *probe_addr = kallsyms_show_value(current_cred()) ?
1773                       (unsigned long)tk->rp.kp.addr : 0;
1774         *symbol = tk->symbol;
1775         if (missed)
1776                 *missed = trace_kprobe_missed(tk);
1777         return 0;
1778 }
1779 #endif  /* CONFIG_PERF_EVENTS */
1780
1781 /*
1782  * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex.
1783  *
1784  * kprobe_trace_self_tests_init() does enable_trace_probe/disable_trace_probe
1785  * lockless, but we can't race with this __init function.
1786  */
1787 static int kprobe_register(struct trace_event_call *event,
1788                            enum trace_reg type, void *data)
1789 {
1790         struct trace_event_file *file = data;
1791
1792         switch (type) {
1793         case TRACE_REG_REGISTER:
1794                 return enable_trace_kprobe(event, file);
1795         case TRACE_REG_UNREGISTER:
1796                 return disable_trace_kprobe(event, file);
1797
1798 #ifdef CONFIG_PERF_EVENTS
1799         case TRACE_REG_PERF_REGISTER:
1800                 return enable_trace_kprobe(event, NULL);
1801         case TRACE_REG_PERF_UNREGISTER:
1802                 return disable_trace_kprobe(event, NULL);
1803         case TRACE_REG_PERF_OPEN:
1804         case TRACE_REG_PERF_CLOSE:
1805         case TRACE_REG_PERF_ADD:
1806         case TRACE_REG_PERF_DEL:
1807                 return 0;
1808 #endif
1809         }
1810         return 0;
1811 }
1812
1813 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1814 {
1815         struct trace_kprobe *tk = container_of(kp, struct trace_kprobe, rp.kp);
1816         int ret = 0;
1817
1818         raw_cpu_inc(*tk->nhit);
1819
1820         if (trace_probe_test_flag(&tk->tp, TP_FLAG_TRACE))
1821                 kprobe_trace_func(tk, regs);
1822 #ifdef CONFIG_PERF_EVENTS
1823         if (trace_probe_test_flag(&tk->tp, TP_FLAG_PROFILE))
1824                 ret = kprobe_perf_func(tk, regs);
1825 #endif
1826         return ret;
1827 }
1828 NOKPROBE_SYMBOL(kprobe_dispatcher);
1829
1830 static int
1831 kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1832 {
1833         struct kretprobe *rp = get_kretprobe(ri);
1834         struct trace_kprobe *tk;
1835
1836         /*
1837          * There is a small chance that get_kretprobe(ri) returns NULL when
1838          * the kretprobe is unregister on another CPU between kretprobe's
1839          * trampoline_handler and this function.
1840          */
1841         if (unlikely(!rp))
1842                 return 0;
1843
1844         tk = container_of(rp, struct trace_kprobe, rp);
1845         raw_cpu_inc(*tk->nhit);
1846
1847         if (trace_probe_test_flag(&tk->tp, TP_FLAG_TRACE))
1848                 kretprobe_trace_func(tk, ri, regs);
1849 #ifdef CONFIG_PERF_EVENTS
1850         if (trace_probe_test_flag(&tk->tp, TP_FLAG_PROFILE))
1851                 kretprobe_perf_func(tk, ri, regs);
1852 #endif
1853         return 0;       /* We don't tweak kernel, so just return 0 */
1854 }
1855 NOKPROBE_SYMBOL(kretprobe_dispatcher);
1856
1857 static struct trace_event_functions kretprobe_funcs = {
1858         .trace          = print_kretprobe_event
1859 };
1860
1861 static struct trace_event_functions kprobe_funcs = {
1862         .trace          = print_kprobe_event
1863 };
1864
1865 static struct trace_event_fields kretprobe_fields_array[] = {
1866         { .type = TRACE_FUNCTION_TYPE,
1867           .define_fields = kretprobe_event_define_fields },
1868         {}
1869 };
1870
1871 static struct trace_event_fields kprobe_fields_array[] = {
1872         { .type = TRACE_FUNCTION_TYPE,
1873           .define_fields = kprobe_event_define_fields },
1874         {}
1875 };
1876
1877 static inline void init_trace_event_call(struct trace_kprobe *tk)
1878 {
1879         struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1880
1881         if (trace_kprobe_is_return(tk)) {
1882                 call->event.funcs = &kretprobe_funcs;
1883                 call->class->fields_array = kretprobe_fields_array;
1884         } else {
1885                 call->event.funcs = &kprobe_funcs;
1886                 call->class->fields_array = kprobe_fields_array;
1887         }
1888
1889         call->flags = TRACE_EVENT_FL_KPROBE;
1890         call->class->reg = kprobe_register;
1891 }
1892
1893 static int register_kprobe_event(struct trace_kprobe *tk)
1894 {
1895         init_trace_event_call(tk);
1896
1897         return trace_probe_register_event_call(&tk->tp);
1898 }
1899
1900 static int unregister_kprobe_event(struct trace_kprobe *tk)
1901 {
1902         return trace_probe_unregister_event_call(&tk->tp);
1903 }
1904
1905 #ifdef CONFIG_PERF_EVENTS
1906
1907 /* create a trace_kprobe, but don't add it to global lists */
1908 struct trace_event_call *
1909 create_local_trace_kprobe(char *func, void *addr, unsigned long offs,
1910                           bool is_return)
1911 {
1912         enum probe_print_type ptype;
1913         struct trace_kprobe *tk __free(free_trace_kprobe) = NULL;
1914         int ret;
1915         char *event;
1916
1917         if (func) {
1918                 ret = validate_probe_symbol(func);
1919                 if (ret)
1920                         return ERR_PTR(ret);
1921         }
1922
1923         /*
1924          * local trace_kprobes are not added to dyn_event, so they are never
1925          * searched in find_trace_kprobe(). Therefore, there is no concern of
1926          * duplicated name here.
1927          */
1928         event = func ? func : "DUMMY_EVENT";
1929
1930         tk = alloc_trace_kprobe(KPROBE_EVENT_SYSTEM, event, (void *)addr, func,
1931                                 offs, 0 /* maxactive */, 0 /* nargs */,
1932                                 is_return);
1933
1934         if (IS_ERR(tk)) {
1935                 pr_info("Failed to allocate trace_probe.(%d)\n",
1936                         (int)PTR_ERR(tk));
1937                 return ERR_CAST(tk);
1938         }
1939
1940         init_trace_event_call(tk);
1941
1942         ptype = trace_kprobe_is_return(tk) ?
1943                 PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL;
1944         if (traceprobe_set_print_fmt(&tk->tp, ptype) < 0)
1945                 return ERR_PTR(-ENOMEM);
1946
1947         ret = __register_trace_kprobe(tk);
1948         if (ret < 0)
1949                 return ERR_PTR(ret);
1950
1951         return trace_probe_event_call(&(no_free_ptr(tk)->tp));
1952 }
1953
1954 void destroy_local_trace_kprobe(struct trace_event_call *event_call)
1955 {
1956         struct trace_kprobe *tk;
1957
1958         tk = trace_kprobe_primary_from_call(event_call);
1959         if (unlikely(!tk))
1960                 return;
1961
1962         if (trace_probe_is_enabled(&tk->tp)) {
1963                 WARN_ON(1);
1964                 return;
1965         }
1966
1967         __unregister_trace_kprobe(tk);
1968
1969         free_trace_kprobe(tk);
1970 }
1971 #endif /* CONFIG_PERF_EVENTS */
1972
1973 static __init void enable_boot_kprobe_events(void)
1974 {
1975         struct trace_array *tr = top_trace_array();
1976         struct trace_event_file *file;
1977         struct trace_kprobe *tk;
1978         struct dyn_event *pos;
1979
1980         guard(mutex)(&event_mutex);
1981         for_each_trace_kprobe(tk, pos) {
1982                 list_for_each_entry(file, &tr->events, list)
1983                         if (file->event_call == trace_probe_event_call(&tk->tp))
1984                                 trace_event_enable_disable(file, 1, 0);
1985         }
1986 }
1987
1988 static __init void setup_boot_kprobe_events(void)
1989 {
1990         char *p, *cmd = kprobe_boot_events_buf;
1991         int ret;
1992
1993         strreplace(kprobe_boot_events_buf, ',', ' ');
1994
1995         while (cmd && *cmd != '\0') {
1996                 p = strchr(cmd, ';');
1997                 if (p)
1998                         *p++ = '\0';
1999
2000                 ret = create_or_delete_trace_kprobe(cmd);
2001                 if (ret)
2002                         pr_warn("Failed to add event(%d): %s\n", ret, cmd);
2003
2004                 cmd = p;
2005         }
2006
2007         enable_boot_kprobe_events();
2008 }
2009
2010 /*
2011  * Register dynevent at core_initcall. This allows kernel to setup kprobe
2012  * events in postcore_initcall without tracefs.
2013  */
2014 static __init int init_kprobe_trace_early(void)
2015 {
2016         int ret;
2017
2018         ret = dyn_event_register(&trace_kprobe_ops);
2019         if (ret)
2020                 return ret;
2021
2022         if (trace_kprobe_register_module_notifier())
2023                 return -EINVAL;
2024
2025         return 0;
2026 }
2027 core_initcall(init_kprobe_trace_early);
2028
2029 /* Make a tracefs interface for controlling probe points */
2030 static __init int init_kprobe_trace(void)
2031 {
2032         int ret;
2033
2034         ret = tracing_init_dentry();
2035         if (ret)
2036                 return 0;
2037
2038         /* Event list interface */
2039         trace_create_file("kprobe_events", TRACE_MODE_WRITE,
2040                           NULL, NULL, &kprobe_events_ops);
2041
2042         /* Profile interface */
2043         trace_create_file("kprobe_profile", TRACE_MODE_READ,
2044                           NULL, NULL, &kprobe_profile_ops);
2045
2046         setup_boot_kprobe_events();
2047
2048         return 0;
2049 }
2050 fs_initcall(init_kprobe_trace);
2051
2052
2053 #ifdef CONFIG_FTRACE_STARTUP_TEST
2054 static __init struct trace_event_file *
2055 find_trace_probe_file(struct trace_kprobe *tk, struct trace_array *tr)
2056 {
2057         struct trace_event_file *file;
2058
2059         list_for_each_entry(file, &tr->events, list)
2060                 if (file->event_call == trace_probe_event_call(&tk->tp))
2061                         return file;
2062
2063         return NULL;
2064 }
2065
2066 /*
2067  * Nobody but us can call enable_trace_kprobe/disable_trace_kprobe at this
2068  * stage, we can do this lockless.
2069  */
2070 static __init int kprobe_trace_self_tests_init(void)
2071 {
2072         int ret, warn = 0;
2073         int (*target)(int, int, int, int, int, int);
2074         struct trace_kprobe *tk;
2075         struct trace_event_file *file;
2076
2077         if (tracing_is_disabled())
2078                 return -ENODEV;
2079
2080         if (tracing_selftest_disabled)
2081                 return 0;
2082
2083         target = kprobe_trace_selftest_target;
2084
2085         pr_info("Testing kprobe tracing: ");
2086
2087         ret = create_or_delete_trace_kprobe("p:testprobe kprobe_trace_selftest_target $stack $stack0 +0($stack)");
2088         if (WARN_ONCE(ret, "error on probing function entry.")) {
2089                 warn++;
2090         } else {
2091                 /* Enable trace point */
2092                 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
2093                 if (WARN_ONCE(tk == NULL, "error on probing function entry.")) {
2094                         warn++;
2095                 } else {
2096                         file = find_trace_probe_file(tk, top_trace_array());
2097                         if (WARN_ONCE(file == NULL, "error on getting probe file.")) {
2098                                 warn++;
2099                         } else
2100                                 enable_trace_kprobe(
2101                                         trace_probe_event_call(&tk->tp), file);
2102                 }
2103         }
2104
2105         ret = create_or_delete_trace_kprobe("r:testprobe2 kprobe_trace_selftest_target $retval");
2106         if (WARN_ONCE(ret, "error on probing function return.")) {
2107                 warn++;
2108         } else {
2109                 /* Enable trace point */
2110                 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
2111                 if (WARN_ONCE(tk == NULL, "error on getting 2nd new probe.")) {
2112                         warn++;
2113                 } else {
2114                         file = find_trace_probe_file(tk, top_trace_array());
2115                         if (WARN_ONCE(file == NULL, "error on getting probe file.")) {
2116                                 warn++;
2117                         } else
2118                                 enable_trace_kprobe(
2119                                         trace_probe_event_call(&tk->tp), file);
2120                 }
2121         }
2122
2123         if (warn)
2124                 goto end;
2125
2126         ret = target(1, 2, 3, 4, 5, 6);
2127
2128         /*
2129          * Not expecting an error here, the check is only to prevent the
2130          * optimizer from removing the call to target() as otherwise there
2131          * are no side-effects and the call is never performed.
2132          */
2133         if (ret != 21)
2134                 warn++;
2135
2136         /* Disable trace points before removing it */
2137         tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
2138         if (WARN_ONCE(tk == NULL, "error on getting test probe.")) {
2139                 warn++;
2140         } else {
2141                 if (WARN_ONCE(trace_kprobe_nhit(tk) != 1,
2142                                  "incorrect number of testprobe hits."))
2143                         warn++;
2144
2145                 file = find_trace_probe_file(tk, top_trace_array());
2146                 if (WARN_ONCE(file == NULL, "error on getting probe file.")) {
2147                         warn++;
2148                 } else
2149                         disable_trace_kprobe(
2150                                 trace_probe_event_call(&tk->tp), file);
2151         }
2152
2153         tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
2154         if (WARN_ONCE(tk == NULL, "error on getting 2nd test probe.")) {
2155                 warn++;
2156         } else {
2157                 if (WARN_ONCE(trace_kprobe_nhit(tk) != 1,
2158                                  "incorrect number of testprobe2 hits."))
2159                         warn++;
2160
2161                 file = find_trace_probe_file(tk, top_trace_array());
2162                 if (WARN_ONCE(file == NULL, "error on getting probe file.")) {
2163                         warn++;
2164                 } else
2165                         disable_trace_kprobe(
2166                                 trace_probe_event_call(&tk->tp), file);
2167         }
2168
2169         ret = create_or_delete_trace_kprobe("-:testprobe");
2170         if (WARN_ONCE(ret, "error on deleting a probe."))
2171                 warn++;
2172
2173         ret = create_or_delete_trace_kprobe("-:testprobe2");
2174         if (WARN_ONCE(ret, "error on deleting a probe."))
2175                 warn++;
2176
2177
2178 end:
2179         /*
2180          * Wait for the optimizer work to finish. Otherwise it might fiddle
2181          * with probes in already freed __init text.
2182          */
2183         wait_for_kprobe_optimizer();
2184         if (warn)
2185                 pr_cont("NG: Some tests are failed. Please check them.\n");
2186         else
2187                 pr_cont("OK\n");
2188         return 0;
2189 }
2190
2191 late_initcall(kprobe_trace_self_tests_init);
2192
2193 #endif