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/module.h>
  11 #include <linux/uaccess.h>
  12 #include <linux/rculist.h>
  13 #include <linux/error-injection.h>
  14
  15 #include "trace_dynevent.h"
  16 #include "trace_kprobe_selftest.h"
  17 #include "trace_probe.h"
  18 #include "trace_probe_tmpl.h"
  19
  20 #define KPROBE_EVENT_SYSTEM "kprobes"
  21 #define KRETPROBE_MAXACTIVE_MAX 4096
  22
  23 static int trace_kprobe_create(int argc, const char **argv);
  24 static int trace_kprobe_show(struct seq_file *m, struct dyn_event *ev);
  25 static int trace_kprobe_release(struct dyn_event *ev);
  26 static bool trace_kprobe_is_busy(struct dyn_event *ev);
  27 static bool trace_kprobe_match(const char *system, const char *event,
  28                                struct dyn_event *ev);
  29
  30 static struct dyn_event_operations trace_kprobe_ops = {
  31         .create = trace_kprobe_create,
  32         .show = trace_kprobe_show,
  33         .is_busy = trace_kprobe_is_busy,
  34         .free = trace_kprobe_release,
  35         .match = trace_kprobe_match,
  36 };
  37
  38 /*
  39  * Kprobe event core functions
  40  */
  41 struct trace_kprobe {
  42         struct dyn_event        devent;
  43         struct kretprobe        rp;     /* Use rp.kp for kprobe use */
  44         unsigned long __percpu *nhit;
  45         const char              *symbol;        /* symbol name */
  46         struct trace_probe      tp;
  47 };
  48
  49 static bool is_trace_kprobe(struct dyn_event *ev)
  50 {
  51         return ev->ops == &trace_kprobe_ops;
  52 }
  53
  54 static struct trace_kprobe *to_trace_kprobe(struct dyn_event *ev)
  55 {
  56         return container_of(ev, struct trace_kprobe, devent);
  57 }
  58
  59 /**
  60  * for_each_trace_kprobe - iterate over the trace_kprobe list
  61  * @pos:        the struct trace_kprobe * for each entry
  62  * @dpos:       the struct dyn_event * to use as a loop cursor
  63  */
  64 #define for_each_trace_kprobe(pos, dpos)        \
  65         for_each_dyn_event(dpos)                \
  66                 if (is_trace_kprobe(dpos) && (pos = to_trace_kprobe(dpos)))
  67
  68 #define SIZEOF_TRACE_KPROBE(n)                          \
  69         (offsetof(struct trace_kprobe, tp.args) +       \
  70         (sizeof(struct probe_arg) * (n)))
  71
  72 static nokprobe_inline bool trace_kprobe_is_return(struct trace_kprobe *tk)
  73 {
  74         return tk->rp.handler != NULL;
  75 }
  76
  77 static nokprobe_inline const char *trace_kprobe_symbol(struct trace_kprobe *tk)
  78 {
  79         return tk->symbol ? tk->symbol : "unknown";
  80 }
  81
  82 static nokprobe_inline unsigned long trace_kprobe_offset(struct trace_kprobe *tk)
  83 {
  84         return tk->rp.kp.offset;
  85 }
  86
  87 static nokprobe_inline bool trace_kprobe_has_gone(struct trace_kprobe *tk)
  88 {
  89         return !!(kprobe_gone(&tk->rp.kp));
  90 }
  91
  92 static nokprobe_inline bool trace_kprobe_within_module(struct trace_kprobe *tk,
  93                                                  struct module *mod)
  94 {
  95         int len = strlen(mod->name);
  96         const char *name = trace_kprobe_symbol(tk);
  97         return strncmp(mod->name, name, len) == 0 && name[len] == ':';
  98 }
  99
 100 static nokprobe_inline bool trace_kprobe_module_exist(struct trace_kprobe *tk)
 101 {
 102         char *p;
 103         bool ret;
 104
 105         if (!tk->symbol)
 106                 return false;
 107         p = strchr(tk->symbol, ':');
 108         if (!p)
 109                 return true;
 110         *p = '\0';
 111         mutex_lock(&module_mutex);
 112         ret = !!find_module(tk->symbol);
 113         mutex_unlock(&module_mutex);
 114         *p = ':';
 115
 116         return ret;
 117 }
 118
 119 static bool trace_kprobe_is_busy(struct dyn_event *ev)
 120 {
 121         struct trace_kprobe *tk = to_trace_kprobe(ev);
 122
 123         return trace_probe_is_enabled(&tk->tp);
 124 }
 125
 126 static bool trace_kprobe_match(const char *system, const char *event,
 127                                struct dyn_event *ev)
 128 {
 129         struct trace_kprobe *tk = to_trace_kprobe(ev);
 130
 131         return strcmp(trace_event_name(&tk->tp.call), event) == 0 &&
 132             (!system || strcmp(tk->tp.call.class->system, system) == 0);
 133 }
 134
 135 static nokprobe_inline unsigned long trace_kprobe_nhit(struct trace_kprobe *tk)
 136 {
 137         unsigned long nhit = 0;
 138         int cpu;
 139
 140         for_each_possible_cpu(cpu)
 141                 nhit += *per_cpu_ptr(tk->nhit, cpu);
 142
 143         return nhit;
 144 }
 145
 146 /* Return 0 if it fails to find the symbol address */
 147 static nokprobe_inline
 148 unsigned long trace_kprobe_address(struct trace_kprobe *tk)
 149 {
 150         unsigned long addr;
 151
 152         if (tk->symbol) {
 153                 addr = (unsigned long)
 154                         kallsyms_lookup_name(trace_kprobe_symbol(tk));
 155                 if (addr)
 156                         addr += tk->rp.kp.offset;
 157         } else {
 158                 addr = (unsigned long)tk->rp.kp.addr;
 159         }
 160         return addr;
 161 }
 162
 163 bool trace_kprobe_on_func_entry(struct trace_event_call *call)
 164 {
 165         struct trace_kprobe *tk = (struct trace_kprobe *)call->data;
 166
 167         return kprobe_on_func_entry(tk->rp.kp.addr,
 168                         tk->rp.kp.addr ? NULL : tk->rp.kp.symbol_name,
 169                         tk->rp.kp.addr ? 0 : tk->rp.kp.offset);
 170 }
 171
 172 bool trace_kprobe_error_injectable(struct trace_event_call *call)
 173 {
 174         struct trace_kprobe *tk = (struct trace_kprobe *)call->data;
 175
 176         return within_error_injection_list(trace_kprobe_address(tk));
 177 }
 178
 179 static int register_kprobe_event(struct trace_kprobe *tk);
 180 static int unregister_kprobe_event(struct trace_kprobe *tk);
 181
 182 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
 183 static int kretprobe_dispatcher(struct kretprobe_instance *ri,
 184                                 struct pt_regs *regs);
 185
 186 /*
 187  * Allocate new trace_probe and initialize it (including kprobes).
 188  */
 189 static struct trace_kprobe *alloc_trace_kprobe(const char *group,
 190                                              const char *event,
 191                                              void *addr,
 192                                              const char *symbol,
 193                                              unsigned long offs,
 194                                              int maxactive,
 195                                              int nargs, bool is_return)
 196 {
 197         struct trace_kprobe *tk;
 198         int ret = -ENOMEM;
 199
 200         tk = kzalloc(SIZEOF_TRACE_KPROBE(nargs), GFP_KERNEL);
 201         if (!tk)
 202                 return ERR_PTR(ret);
 203
 204         tk->nhit = alloc_percpu(unsigned long);
 205         if (!tk->nhit)
 206                 goto error;
 207
 208         if (symbol) {
 209                 tk->symbol = kstrdup(symbol, GFP_KERNEL);
 210                 if (!tk->symbol)
 211                         goto error;
 212                 tk->rp.kp.symbol_name = tk->symbol;
 213                 tk->rp.kp.offset = offs;
 214         } else
 215                 tk->rp.kp.addr = addr;
 216
 217         if (is_return)
 218                 tk->rp.handler = kretprobe_dispatcher;
 219         else
 220                 tk->rp.kp.pre_handler = kprobe_dispatcher;
 221
 222         tk->rp.maxactive = maxactive;
 223
 224         if (!event || !group) {
 225                 ret = -EINVAL;
 226                 goto error;
 227         }
 228
 229         tk->tp.call.class = &tk->tp.class;
 230         tk->tp.call.name = kstrdup(event, GFP_KERNEL);
 231         if (!tk->tp.call.name)
 232                 goto error;
 233
 234         tk->tp.class.system = kstrdup(group, GFP_KERNEL);
 235         if (!tk->tp.class.system)
 236                 goto error;
 237
 238         dyn_event_init(&tk->devent, &trace_kprobe_ops);
 239         INIT_LIST_HEAD(&tk->tp.files);
 240         return tk;
 241 error:
 242         kfree(tk->tp.call.name);
 243         kfree(tk->symbol);
 244         free_percpu(tk->nhit);
 245         kfree(tk);
 246         return ERR_PTR(ret);
 247 }
 248
 249 static void free_trace_kprobe(struct trace_kprobe *tk)
 250 {
 251         int i;
 252
 253         if (!tk)
 254                 return;
 255
 256         for (i = 0; i < tk->tp.nr_args; i++)
 257                 traceprobe_free_probe_arg(&tk->tp.args[i]);
 258
 259         kfree(tk->tp.call.class->system);
 260         kfree(tk->tp.call.name);
 261         kfree(tk->symbol);
 262         free_percpu(tk->nhit);
 263         kfree(tk);
 264 }
 265
 266 static struct trace_kprobe *find_trace_kprobe(const char *event,
 267                                               const char *group)
 268 {
 269         struct dyn_event *pos;
 270         struct trace_kprobe *tk;
 271
 272         for_each_trace_kprobe(tk, pos)
 273                 if (strcmp(trace_event_name(&tk->tp.call), event) == 0 &&
 274                     strcmp(tk->tp.call.class->system, group) == 0)
 275                         return tk;
 276         return NULL;
 277 }
 278
 279 static inline int __enable_trace_kprobe(struct trace_kprobe *tk)
 280 {
 281         int ret = 0;
 282
 283         if (trace_probe_is_registered(&tk->tp) && !trace_kprobe_has_gone(tk)) {
 284                 if (trace_kprobe_is_return(tk))
 285                         ret = enable_kretprobe(&tk->rp);
 286                 else
 287                         ret = enable_kprobe(&tk->rp.kp);
 288         }
 289
 290         return ret;
 291 }
 292
 293 /*
 294  * Enable trace_probe
 295  * if the file is NULL, enable "perf" handler, or enable "trace" handler.
 296  */
 297 static int
 298 enable_trace_kprobe(struct trace_kprobe *tk, struct trace_event_file *file)
 299 {
 300         struct event_file_link *link;
 301         int ret = 0;
 302
 303         if (file) {
 304                 link = kmalloc(sizeof(*link), GFP_KERNEL);
 305                 if (!link) {
 306                         ret = -ENOMEM;
 307                         goto out;
 308                 }
 309
 310                 link->file = file;
 311                 list_add_tail_rcu(&link->list, &tk->tp.files);
 312
 313                 tk->tp.flags |= TP_FLAG_TRACE;
 314                 ret = __enable_trace_kprobe(tk);
 315                 if (ret) {
 316                         list_del_rcu(&link->list);
 317                         kfree(link);
 318                         tk->tp.flags &= ~TP_FLAG_TRACE;
 319                 }
 320
 321         } else {
 322                 tk->tp.flags |= TP_FLAG_PROFILE;
 323                 ret = __enable_trace_kprobe(tk);
 324                 if (ret)
 325                         tk->tp.flags &= ~TP_FLAG_PROFILE;
 326         }
 327  out:
 328         return ret;
 329 }
 330
 331 /*
 332  * Disable trace_probe
 333  * if the file is NULL, disable "perf" handler, or disable "trace" handler.
 334  */
 335 static int
 336 disable_trace_kprobe(struct trace_kprobe *tk, struct trace_event_file *file)
 337 {
 338         struct event_file_link *link = NULL;
 339         int wait = 0;
 340         int ret = 0;
 341
 342         if (file) {
 343                 link = find_event_file_link(&tk->tp, file);
 344                 if (!link) {
 345                         ret = -EINVAL;
 346                         goto out;
 347                 }
 348
 349                 list_del_rcu(&link->list);
 350                 wait = 1;
 351                 if (!list_empty(&tk->tp.files))
 352                         goto out;
 353
 354                 tk->tp.flags &= ~TP_FLAG_TRACE;
 355         } else
 356                 tk->tp.flags &= ~TP_FLAG_PROFILE;
 357
 358         if (!trace_probe_is_enabled(&tk->tp) && trace_probe_is_registered(&tk->tp)) {
 359                 if (trace_kprobe_is_return(tk))
 360                         disable_kretprobe(&tk->rp);
 361                 else
 362                         disable_kprobe(&tk->rp.kp);
 363                 wait = 1;
 364         }
 365
 366         /*
 367          * if tk is not added to any list, it must be a local trace_kprobe
 368          * created with perf_event_open. We don't need to wait for these
 369          * trace_kprobes
 370          */
 371         if (list_empty(&tk->devent.list))
 372                 wait = 0;
 373  out:
 374         if (wait) {
 375                 /*
 376                  * Synchronize with kprobe_trace_func/kretprobe_trace_func
 377                  * to ensure disabled (all running handlers are finished).
 378                  * This is not only for kfree(), but also the caller,
 379                  * trace_remove_event_call() supposes it for releasing
 380                  * event_call related objects, which will be accessed in
 381                  * the kprobe_trace_func/kretprobe_trace_func.
 382                  */
 383                 synchronize_rcu();
 384                 kfree(link);    /* Ignored if link == NULL */
 385         }
 386
 387         return ret;
 388 }
 389
 390 #if defined(CONFIG_KPROBES_ON_FTRACE) && \
 391         !defined(CONFIG_KPROBE_EVENTS_ON_NOTRACE)
 392 static bool within_notrace_func(struct trace_kprobe *tk)
 393 {
 394         unsigned long offset, size, addr;
 395
 396         addr = trace_kprobe_address(tk);
 397         if (!addr || !kallsyms_lookup_size_offset(addr, &size, &offset))
 398                 return false;
 399
 400         /* Get the entry address of the target function */
 401         addr -= offset;
 402
 403         /*
 404          * Since ftrace_location_range() does inclusive range check, we need
 405          * to subtract 1 byte from the end address.
 406          */
 407         return !ftrace_location_range(addr, addr + size - 1);
 408 }
 409 #else
 410 #define within_notrace_func(tk) (false)
 411 #endif
 412
 413 /* Internal register function - just handle k*probes and flags */
 414 static int __register_trace_kprobe(struct trace_kprobe *tk)
 415 {
 416         int i, ret;
 417
 418         if (trace_probe_is_registered(&tk->tp))
 419                 return -EINVAL;
 420
 421         if (within_notrace_func(tk)) {
 422                 pr_warn("Could not probe notrace function %s\n",
 423                         trace_kprobe_symbol(tk));
 424                 return -EINVAL;
 425         }
 426
 427         for (i = 0; i < tk->tp.nr_args; i++) {
 428                 ret = traceprobe_update_arg(&tk->tp.args[i]);
 429                 if (ret)
 430                         return ret;
 431         }
 432
 433         /* Set/clear disabled flag according to tp->flag */
 434         if (trace_probe_is_enabled(&tk->tp))
 435                 tk->rp.kp.flags &= ~KPROBE_FLAG_DISABLED;
 436         else
 437                 tk->rp.kp.flags |= KPROBE_FLAG_DISABLED;
 438
 439         if (trace_kprobe_is_return(tk))
 440                 ret = register_kretprobe(&tk->rp);
 441         else
 442                 ret = register_kprobe(&tk->rp.kp);
 443
 444         if (ret == 0) {
 445                 tk->tp.flags |= TP_FLAG_REGISTERED;
 446         } else if (ret == -EILSEQ) {
 447                 pr_warn("Probing address(0x%p) is not an instruction boundary.\n",
 448                         tk->rp.kp.addr);
 449                 ret = -EINVAL;
 450         }
 451         return ret;
 452 }
 453
 454 /* Internal unregister function - just handle k*probes and flags */
 455 static void __unregister_trace_kprobe(struct trace_kprobe *tk)
 456 {
 457         if (trace_probe_is_registered(&tk->tp)) {
 458                 if (trace_kprobe_is_return(tk))
 459                         unregister_kretprobe(&tk->rp);
 460                 else
 461                         unregister_kprobe(&tk->rp.kp);
 462                 tk->tp.flags &= ~TP_FLAG_REGISTERED;
 463                 /* Cleanup kprobe for reuse */
 464                 if (tk->rp.kp.symbol_name)
 465                         tk->rp.kp.addr = NULL;
 466         }
 467 }
 468
 469 /* Unregister a trace_probe and probe_event */
 470 static int unregister_trace_kprobe(struct trace_kprobe *tk)
 471 {
 472         /* Enabled event can not be unregistered */
 473         if (trace_probe_is_enabled(&tk->tp))
 474                 return -EBUSY;
 475
 476         /* Will fail if probe is being used by ftrace or perf */
 477         if (unregister_kprobe_event(tk))
 478                 return -EBUSY;
 479
 480         __unregister_trace_kprobe(tk);
 481         dyn_event_remove(&tk->devent);
 482
 483         return 0;
 484 }
 485
 486 /* Register a trace_probe and probe_event */
 487 static int register_trace_kprobe(struct trace_kprobe *tk)
 488 {
 489         struct trace_kprobe *old_tk;
 490         int ret;
 491
 492         mutex_lock(&event_mutex);
 493
 494         /* Delete old (same name) event if exist */
 495         old_tk = find_trace_kprobe(trace_event_name(&tk->tp.call),
 496                         tk->tp.call.class->system);
 497         if (old_tk) {
 498                 ret = unregister_trace_kprobe(old_tk);
 499                 if (ret < 0)
 500                         goto end;
 501                 free_trace_kprobe(old_tk);
 502         }
 503
 504         /* Register new event */
 505         ret = register_kprobe_event(tk);
 506         if (ret) {
 507                 pr_warn("Failed to register probe event(%d)\n", ret);
 508                 goto end;
 509         }
 510
 511         /* Register k*probe */
 512         ret = __register_trace_kprobe(tk);
 513         if (ret == -ENOENT && !trace_kprobe_module_exist(tk)) {
 514                 pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
 515                 ret = 0;
 516         }
 517
 518         if (ret < 0)
 519                 unregister_kprobe_event(tk);
 520         else
 521                 dyn_event_add(&tk->devent);
 522
 523 end:
 524         mutex_unlock(&event_mutex);
 525         return ret;
 526 }
 527
 528 /* Module notifier call back, checking event on the module */
 529 static int trace_kprobe_module_callback(struct notifier_block *nb,
 530                                        unsigned long val, void *data)
 531 {
 532         struct module *mod = data;
 533         struct dyn_event *pos;
 534         struct trace_kprobe *tk;
 535         int ret;
 536
 537         if (val != MODULE_STATE_COMING)
 538                 return NOTIFY_DONE;
 539
 540         /* Update probes on coming module */
 541         mutex_lock(&event_mutex);
 542         for_each_trace_kprobe(tk, pos) {
 543                 if (trace_kprobe_within_module(tk, mod)) {
 544                         /* Don't need to check busy - this should have gone. */
 545                         __unregister_trace_kprobe(tk);
 546                         ret = __register_trace_kprobe(tk);
 547                         if (ret)
 548                                 pr_warn("Failed to re-register probe %s on %s: %d\n",
 549                                         trace_event_name(&tk->tp.call),
 550                                         mod->name, ret);
 551                 }
 552         }
 553         mutex_unlock(&event_mutex);
 554
 555         return NOTIFY_DONE;
 556 }
 557
 558 static struct notifier_block trace_kprobe_module_nb = {
 559         .notifier_call = trace_kprobe_module_callback,
 560         .priority = 1   /* Invoked after kprobe module callback */
 561 };
 562
 563 /* Convert certain expected symbols into '_' when generating event names */
 564 static inline void sanitize_event_name(char *name)
 565 {
 566         while (*name++ != '\0')
 567                 if (*name == ':' || *name == '.')
 568                         *name = '_';
 569 }
 570
 571 static int trace_kprobe_create(int argc, const char *argv[])
 572 {
 573         /*
 574          * Argument syntax:
 575          *  - Add kprobe:
 576          *      p[:[GRP/]EVENT] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
 577          *  - Add kretprobe:
 578          *      r[MAXACTIVE][:[GRP/]EVENT] [MOD:]KSYM[+0] [FETCHARGS]
 579          * Fetch args:
 580          *  $retval     : fetch return value
 581          *  $stack      : fetch stack address
 582          *  $stackN     : fetch Nth of stack (N:0-)
 583          *  $comm       : fetch current task comm
 584          *  @ADDR       : fetch memory at ADDR (ADDR should be in kernel)
 585          *  @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
 586          *  %REG        : fetch register REG
 587          * Dereferencing memory fetch:
 588          *  +|-offs(ARG) : fetch memory at ARG +|- offs address.
 589          * Alias name of args:
 590          *  NAME=FETCHARG : set NAME as alias of FETCHARG.
 591          * Type of args:
 592          *  FETCHARG:TYPE : use TYPE instead of unsigned long.
 593          */
 594         struct trace_kprobe *tk;
 595         int i, len, ret = 0;
 596         bool is_return = false;
 597         char *symbol = NULL, *tmp = NULL;
 598         const char *event = NULL, *group = KPROBE_EVENT_SYSTEM;
 599         int maxactive = 0;
 600         long offset = 0;
 601         void *addr = NULL;
 602         char buf[MAX_EVENT_NAME_LEN];
 603         unsigned int flags = TPARG_FL_KERNEL;
 604
 605         switch (argv[0][0]) {
 606         case 'r':
 607                 is_return = true;
 608                 flags |= TPARG_FL_RETURN;
 609                 break;
 610         case 'p':
 611                 break;
 612         default:
 613                 return -ECANCELED;
 614         }
 615         if (argc < 2)
 616                 return -ECANCELED;
 617
 618         event = strchr(&argv[0][1], ':');
 619         if (event)
 620                 event++;
 621
 622         if (isdigit(argv[0][1])) {
 623                 if (!is_return) {
 624                         pr_info("Maxactive is not for kprobe");
 625                         return -EINVAL;
 626                 }
 627                 if (event)
 628                         len = event - &argv[0][1] - 1;
 629                 else
 630                         len = strlen(&argv[0][1]);
 631                 if (len > MAX_EVENT_NAME_LEN - 1)
 632                         return -E2BIG;
 633                 memcpy(buf, &argv[0][1], len);
 634                 buf[len] = '\0';
 635                 ret = kstrtouint(buf, 0, &maxactive);
 636                 if (ret || !maxactive) {
 637                         pr_info("Invalid maxactive number\n");
 638                         return ret;
 639                 }
 640                 /* kretprobes instances are iterated over via a list. The
 641                  * maximum should stay reasonable.
 642                  */
 643                 if (maxactive > KRETPROBE_MAXACTIVE_MAX) {
 644                         pr_info("Maxactive is too big (%d > %d).\n",
 645                                 maxactive, KRETPROBE_MAXACTIVE_MAX);
 646                         return -E2BIG;
 647                 }
 648         }
 649
 650         /* try to parse an address. if that fails, try to read the
 651          * input as a symbol. */
 652         if (kstrtoul(argv[1], 0, (unsigned long *)&addr)) {
 653                 /* Check whether uprobe event specified */
 654                 if (strchr(argv[1], '/') && strchr(argv[1], ':'))
 655                         return -ECANCELED;
 656                 /* a symbol specified */
 657                 symbol = kstrdup(argv[1], GFP_KERNEL);
 658                 if (!symbol)
 659                         return -ENOMEM;
 660                 /* TODO: support .init module functions */
 661                 ret = traceprobe_split_symbol_offset(symbol, &offset);
 662                 if (ret || offset < 0 || offset > UINT_MAX) {
 663                         pr_info("Failed to parse either an address or a symbol.\n");
 664                         goto out;
 665                 }
 666                 if (kprobe_on_func_entry(NULL, symbol, offset))
 667                         flags |= TPARG_FL_FENTRY;
 668                 if (offset && is_return && !(flags & TPARG_FL_FENTRY)) {
 669                         pr_info("Given offset is not valid for return probe.\n");
 670                         ret = -EINVAL;
 671                         goto out;
 672                 }
 673         }
 674         argc -= 2; argv += 2;
 675
 676         if (event) {
 677                 ret = traceprobe_parse_event_name(&event, &group, buf);
 678                 if (ret)
 679                         goto out;
 680         } else {
 681                 /* Make a new event name */
 682                 if (symbol)
 683                         snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
 684                                  is_return ? 'r' : 'p', symbol, offset);
 685                 else
 686                         snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
 687                                  is_return ? 'r' : 'p', addr);
 688                 sanitize_event_name(buf);
 689                 event = buf;
 690         }
 691
 692         /* setup a probe */
 693         tk = alloc_trace_kprobe(group, event, addr, symbol, offset, maxactive,
 694                                argc, is_return);
 695         if (IS_ERR(tk)) {
 696                 pr_info("Failed to allocate trace_probe.(%d)\n",
 697                         (int)PTR_ERR(tk));
 698                 ret = PTR_ERR(tk);
 699                 goto out;
 700         }
 701
 702         /* parse arguments */
 703         for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
 704                 tmp = kstrdup(argv[i], GFP_KERNEL);
 705                 if (!tmp) {
 706                         ret = -ENOMEM;
 707                         goto error;
 708                 }
 709
 710                 ret = traceprobe_parse_probe_arg(&tk->tp, i, tmp, flags);
 711                 kfree(tmp);
 712                 if (ret)
 713                         goto error;
 714         }
 715
 716         ret = register_trace_kprobe(tk);
 717         if (ret)
 718                 goto error;
 719 out:
 720         kfree(symbol);
 721         return ret;
 722
 723 error:
 724         free_trace_kprobe(tk);
 725         goto out;
 726 }
 727
 728 static int create_or_delete_trace_kprobe(int argc, char **argv)
 729 {
 730         int ret;
 731
 732         if (argv[0][0] == '-')
 733                 return dyn_event_release(argc, argv, &trace_kprobe_ops);
 734
 735         ret = trace_kprobe_create(argc, (const char **)argv);
 736         return ret == -ECANCELED ? -EINVAL : ret;
 737 }
 738
 739 static int trace_kprobe_release(struct dyn_event *ev)
 740 {
 741         struct trace_kprobe *tk = to_trace_kprobe(ev);
 742         int ret = unregister_trace_kprobe(tk);
 743
 744         if (!ret)
 745                 free_trace_kprobe(tk);
 746         return ret;
 747 }
 748
 749 static int trace_kprobe_show(struct seq_file *m, struct dyn_event *ev)
 750 {
 751         struct trace_kprobe *tk = to_trace_kprobe(ev);
 752         int i;
 753
 754         seq_putc(m, trace_kprobe_is_return(tk) ? 'r' : 'p');
 755         seq_printf(m, ":%s/%s", tk->tp.call.class->system,
 756                         trace_event_name(&tk->tp.call));
 757
 758         if (!tk->symbol)
 759                 seq_printf(m, " 0x%p", tk->rp.kp.addr);
 760         else if (tk->rp.kp.offset)
 761                 seq_printf(m, " %s+%u", trace_kprobe_symbol(tk),
 762                            tk->rp.kp.offset);
 763         else
 764                 seq_printf(m, " %s", trace_kprobe_symbol(tk));
 765
 766         for (i = 0; i < tk->tp.nr_args; i++)
 767                 seq_printf(m, " %s=%s", tk->tp.args[i].name, tk->tp.args[i].comm);
 768         seq_putc(m, '\n');
 769
 770         return 0;
 771 }
 772
 773 static int probes_seq_show(struct seq_file *m, void *v)
 774 {
 775         struct dyn_event *ev = v;
 776
 777         if (!is_trace_kprobe(ev))
 778                 return 0;
 779
 780         return trace_kprobe_show(m, ev);
 781 }
 782
 783 static const struct seq_operations probes_seq_op = {
 784         .start  = dyn_event_seq_start,
 785         .next   = dyn_event_seq_next,
 786         .stop   = dyn_event_seq_stop,
 787         .show   = probes_seq_show
 788 };
 789
 790 static int probes_open(struct inode *inode, struct file *file)
 791 {
 792         int ret;
 793
 794         if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
 795                 ret = dyn_events_release_all(&trace_kprobe_ops);
 796                 if (ret < 0)
 797                         return ret;
 798         }
 799
 800         return seq_open(file, &probes_seq_op);
 801 }
 802
 803 static ssize_t probes_write(struct file *file, const char __user *buffer,
 804                             size_t count, loff_t *ppos)
 805 {
 806         return trace_parse_run_command(file, buffer, count, ppos,
 807                                        create_or_delete_trace_kprobe);
 808 }
 809
 810 static const struct file_operations kprobe_events_ops = {
 811         .owner          = THIS_MODULE,
 812         .open           = probes_open,
 813         .read           = seq_read,
 814         .llseek         = seq_lseek,
 815         .release        = seq_release,
 816         .write          = probes_write,
 817 };
 818
 819 /* Probes profiling interfaces */
 820 static int probes_profile_seq_show(struct seq_file *m, void *v)
 821 {
 822         struct dyn_event *ev = v;
 823         struct trace_kprobe *tk;
 824
 825         if (!is_trace_kprobe(ev))
 826                 return 0;
 827
 828         tk = to_trace_kprobe(ev);
 829         seq_printf(m, "  %-44s %15lu %15lu\n",
 830                    trace_event_name(&tk->tp.call),
 831                    trace_kprobe_nhit(tk),
 832                    tk->rp.kp.nmissed);
 833
 834         return 0;
 835 }
 836
 837 static const struct seq_operations profile_seq_op = {
 838         .start  = dyn_event_seq_start,
 839         .next   = dyn_event_seq_next,
 840         .stop   = dyn_event_seq_stop,
 841         .show   = probes_profile_seq_show
 842 };
 843
 844 static int profile_open(struct inode *inode, struct file *file)
 845 {
 846         return seq_open(file, &profile_seq_op);
 847 }
 848
 849 static const struct file_operations kprobe_profile_ops = {
 850         .owner          = THIS_MODULE,
 851         .open           = profile_open,
 852         .read           = seq_read,
 853         .llseek         = seq_lseek,
 854         .release        = seq_release,
 855 };
 856
 857 /* Kprobe specific fetch functions */
 858
 859 /* Return the length of string -- including null terminal byte */
 860 static nokprobe_inline int
 861 fetch_store_strlen(unsigned long addr)
 862 {
 863         mm_segment_t old_fs;
 864         int ret, len = 0;
 865         u8 c;
 866
 867         old_fs = get_fs();
 868         set_fs(KERNEL_DS);
 869         pagefault_disable();
 870
 871         do {
 872                 ret = __copy_from_user_inatomic(&c, (u8 *)addr + len, 1);
 873                 len++;
 874         } while (c && ret == 0 && len < MAX_STRING_SIZE);
 875
 876         pagefault_enable();
 877         set_fs(old_fs);
 878
 879         return (ret < 0) ? ret : len;
 880 }
 881
 882 /*
 883  * Fetch a null-terminated string. Caller MUST set *(u32 *)buf with max
 884  * length and relative data location.
 885  */
 886 static nokprobe_inline int
 887 fetch_store_string(unsigned long addr, void *dest, void *base)
 888 {
 889         int maxlen = get_loc_len(*(u32 *)dest);
 890         u8 *dst = get_loc_data(dest, base);
 891         long ret;
 892
 893         if (unlikely(!maxlen))
 894                 return -ENOMEM;
 895         /*
 896          * Try to get string again, since the string can be changed while
 897          * probing.
 898          */
 899         ret = strncpy_from_unsafe(dst, (void *)addr, maxlen);
 900
 901         if (ret >= 0)
 902                 *(u32 *)dest = make_data_loc(ret, (void *)dst - base);
 903         return ret;
 904 }
 905
 906 static nokprobe_inline int
 907 probe_mem_read(void *dest, void *src, size_t size)
 908 {
 909         return probe_kernel_read(dest, src, size);
 910 }
 911
 912 /* Note that we don't verify it, since the code does not come from user space */
 913 static int
 914 process_fetch_insn(struct fetch_insn *code, struct pt_regs *regs, void *dest,
 915                    void *base)
 916 {
 917         unsigned long val;
 918
 919 retry:
 920         /* 1st stage: get value from context */
 921         switch (code->op) {
 922         case FETCH_OP_REG:
 923                 val = regs_get_register(regs, code->param);
 924                 break;
 925         case FETCH_OP_STACK:
 926                 val = regs_get_kernel_stack_nth(regs, code->param);
 927                 break;
 928         case FETCH_OP_STACKP:
 929                 val = kernel_stack_pointer(regs);
 930                 break;
 931         case FETCH_OP_RETVAL:
 932                 val = regs_return_value(regs);
 933                 break;
 934         case FETCH_OP_IMM:
 935                 val = code->immediate;
 936                 break;
 937         case FETCH_OP_COMM:
 938                 val = (unsigned long)current->comm;
 939                 break;
 940 #ifdef CONFIG_HAVE_FUNCTION_ARG_ACCESS_API
 941         case FETCH_OP_ARG:
 942                 val = regs_get_kernel_argument(regs, code->param);
 943                 break;
 944 #endif
 945         case FETCH_NOP_SYMBOL:  /* Ignore a place holder */
 946                 code++;
 947                 goto retry;
 948         default:
 949                 return -EILSEQ;
 950         }
 951         code++;
 952
 953         return process_fetch_insn_bottom(code, val, dest, base);
 954 }
 955 NOKPROBE_SYMBOL(process_fetch_insn)
 956
 957 /* Kprobe handler */
 958 static nokprobe_inline void
 959 __kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs,
 960                     struct trace_event_file *trace_file)
 961 {
 962         struct kprobe_trace_entry_head *entry;
 963         struct ring_buffer_event *event;
 964         struct ring_buffer *buffer;
 965         int size, dsize, pc;
 966         unsigned long irq_flags;
 967         struct trace_event_call *call = &tk->tp.call;
 968
 969         WARN_ON(call != trace_file->event_call);
 970
 971         if (trace_trigger_soft_disabled(trace_file))
 972                 return;
 973
 974         local_save_flags(irq_flags);
 975         pc = preempt_count();
 976
 977         dsize = __get_data_size(&tk->tp, regs);
 978         size = sizeof(*entry) + tk->tp.size + dsize;
 979
 980         event = trace_event_buffer_lock_reserve(&buffer, trace_file,
 981                                                 call->event.type,
 982                                                 size, irq_flags, pc);
 983         if (!event)
 984                 return;
 985
 986         entry = ring_buffer_event_data(event);
 987         entry->ip = (unsigned long)tk->rp.kp.addr;
 988         store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
 989
 990         event_trigger_unlock_commit_regs(trace_file, buffer, event,
 991                                          entry, irq_flags, pc, regs);
 992 }
 993
 994 static void
 995 kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs)
 996 {
 997         struct event_file_link *link;
 998
 999         list_for_each_entry_rcu(link, &tk->tp.files, list)
1000                 __kprobe_trace_func(tk, regs, link->file);
1001 }
1002 NOKPROBE_SYMBOL(kprobe_trace_func);
1003
1004 /* Kretprobe handler */
1005 static nokprobe_inline void
1006 __kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1007                        struct pt_regs *regs,
1008                        struct trace_event_file *trace_file)
1009 {
1010         struct kretprobe_trace_entry_head *entry;
1011         struct ring_buffer_event *event;
1012         struct ring_buffer *buffer;
1013         int size, pc, dsize;
1014         unsigned long irq_flags;
1015         struct trace_event_call *call = &tk->tp.call;
1016
1017         WARN_ON(call != trace_file->event_call);
1018
1019         if (trace_trigger_soft_disabled(trace_file))
1020                 return;
1021
1022         local_save_flags(irq_flags);
1023         pc = preempt_count();
1024
1025         dsize = __get_data_size(&tk->tp, regs);
1026         size = sizeof(*entry) + tk->tp.size + dsize;
1027
1028         event = trace_event_buffer_lock_reserve(&buffer, trace_file,
1029                                                 call->event.type,
1030                                                 size, irq_flags, pc);
1031         if (!event)
1032                 return;
1033
1034         entry = ring_buffer_event_data(event);
1035         entry->func = (unsigned long)tk->rp.kp.addr;
1036         entry->ret_ip = (unsigned long)ri->ret_addr;
1037         store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1038
1039         event_trigger_unlock_commit_regs(trace_file, buffer, event,
1040                                          entry, irq_flags, pc, regs);
1041 }
1042
1043 static void
1044 kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1045                      struct pt_regs *regs)
1046 {
1047         struct event_file_link *link;
1048
1049         list_for_each_entry_rcu(link, &tk->tp.files, list)
1050                 __kretprobe_trace_func(tk, ri, regs, link->file);
1051 }
1052 NOKPROBE_SYMBOL(kretprobe_trace_func);
1053
1054 /* Event entry printers */
1055 static enum print_line_t
1056 print_kprobe_event(struct trace_iterator *iter, int flags,
1057                    struct trace_event *event)
1058 {
1059         struct kprobe_trace_entry_head *field;
1060         struct trace_seq *s = &iter->seq;
1061         struct trace_probe *tp;
1062
1063         field = (struct kprobe_trace_entry_head *)iter->ent;
1064         tp = container_of(event, struct trace_probe, call.event);
1065
1066         trace_seq_printf(s, "%s: (", trace_event_name(&tp->call));
1067
1068         if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
1069                 goto out;
1070
1071         trace_seq_putc(s, ')');
1072
1073         if (print_probe_args(s, tp->args, tp->nr_args,
1074                              (u8 *)&field[1], field) < 0)
1075                 goto out;
1076
1077         trace_seq_putc(s, '\n');
1078  out:
1079         return trace_handle_return(s);
1080 }
1081
1082 static enum print_line_t
1083 print_kretprobe_event(struct trace_iterator *iter, int flags,
1084                       struct trace_event *event)
1085 {
1086         struct kretprobe_trace_entry_head *field;
1087         struct trace_seq *s = &iter->seq;
1088         struct trace_probe *tp;
1089
1090         field = (struct kretprobe_trace_entry_head *)iter->ent;
1091         tp = container_of(event, struct trace_probe, call.event);
1092
1093         trace_seq_printf(s, "%s: (", trace_event_name(&tp->call));
1094
1095         if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1096                 goto out;
1097
1098         trace_seq_puts(s, " <- ");
1099
1100         if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1101                 goto out;
1102
1103         trace_seq_putc(s, ')');
1104
1105         if (print_probe_args(s, tp->args, tp->nr_args,
1106                              (u8 *)&field[1], field) < 0)
1107                 goto out;
1108
1109         trace_seq_putc(s, '\n');
1110
1111  out:
1112         return trace_handle_return(s);
1113 }
1114
1115
1116 static int kprobe_event_define_fields(struct trace_event_call *event_call)
1117 {
1118         int ret;
1119         struct kprobe_trace_entry_head field;
1120         struct trace_kprobe *tk = (struct trace_kprobe *)event_call->data;
1121
1122         DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1123
1124         return traceprobe_define_arg_fields(event_call, sizeof(field), &tk->tp);
1125 }
1126
1127 static int kretprobe_event_define_fields(struct trace_event_call *event_call)
1128 {
1129         int ret;
1130         struct kretprobe_trace_entry_head field;
1131         struct trace_kprobe *tk = (struct trace_kprobe *)event_call->data;
1132
1133         DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1134         DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1135
1136         return traceprobe_define_arg_fields(event_call, sizeof(field), &tk->tp);
1137 }
1138
1139 #ifdef CONFIG_PERF_EVENTS
1140
1141 /* Kprobe profile handler */
1142 static int
1143 kprobe_perf_func(struct trace_kprobe *tk, struct pt_regs *regs)
1144 {
1145         struct trace_event_call *call = &tk->tp.call;
1146         struct kprobe_trace_entry_head *entry;
1147         struct hlist_head *head;
1148         int size, __size, dsize;
1149         int rctx;
1150
1151         if (bpf_prog_array_valid(call)) {
1152                 unsigned long orig_ip = instruction_pointer(regs);
1153                 int ret;
1154
1155                 ret = trace_call_bpf(call, regs);
1156
1157                 /*
1158                  * We need to check and see if we modified the pc of the
1159                  * pt_regs, and if so return 1 so that we don't do the
1160                  * single stepping.
1161                  */
1162                 if (orig_ip != instruction_pointer(regs))
1163                         return 1;
1164                 if (!ret)
1165                         return 0;
1166         }
1167
1168         head = this_cpu_ptr(call->perf_events);
1169         if (hlist_empty(head))
1170                 return 0;
1171
1172         dsize = __get_data_size(&tk->tp, regs);
1173         __size = sizeof(*entry) + tk->tp.size + dsize;
1174         size = ALIGN(__size + sizeof(u32), sizeof(u64));
1175         size -= sizeof(u32);
1176
1177         entry = perf_trace_buf_alloc(size, NULL, &rctx);
1178         if (!entry)
1179                 return 0;
1180
1181         entry->ip = (unsigned long)tk->rp.kp.addr;
1182         memset(&entry[1], 0, dsize);
1183         store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1184         perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1185                               head, NULL);
1186         return 0;
1187 }
1188 NOKPROBE_SYMBOL(kprobe_perf_func);
1189
1190 /* Kretprobe profile handler */
1191 static void
1192 kretprobe_perf_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1193                     struct pt_regs *regs)
1194 {
1195         struct trace_event_call *call = &tk->tp.call;
1196         struct kretprobe_trace_entry_head *entry;
1197         struct hlist_head *head;
1198         int size, __size, dsize;
1199         int rctx;
1200
1201         if (bpf_prog_array_valid(call) && !trace_call_bpf(call, regs))
1202                 return;
1203
1204         head = this_cpu_ptr(call->perf_events);
1205         if (hlist_empty(head))
1206                 return;
1207
1208         dsize = __get_data_size(&tk->tp, regs);
1209         __size = sizeof(*entry) + tk->tp.size + dsize;
1210         size = ALIGN(__size + sizeof(u32), sizeof(u64));
1211         size -= sizeof(u32);
1212
1213         entry = perf_trace_buf_alloc(size, NULL, &rctx);
1214         if (!entry)
1215                 return;
1216
1217         entry->func = (unsigned long)tk->rp.kp.addr;
1218         entry->ret_ip = (unsigned long)ri->ret_addr;
1219         store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1220         perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1221                               head, NULL);
1222 }
1223 NOKPROBE_SYMBOL(kretprobe_perf_func);
1224
1225 int bpf_get_kprobe_info(const struct perf_event *event, u32 *fd_type,
1226                         const char **symbol, u64 *probe_offset,
1227                         u64 *probe_addr, bool perf_type_tracepoint)
1228 {
1229         const char *pevent = trace_event_name(event->tp_event);
1230         const char *group = event->tp_event->class->system;
1231         struct trace_kprobe *tk;
1232
1233         if (perf_type_tracepoint)
1234                 tk = find_trace_kprobe(pevent, group);
1235         else
1236                 tk = event->tp_event->data;
1237         if (!tk)
1238                 return -EINVAL;
1239
1240         *fd_type = trace_kprobe_is_return(tk) ? BPF_FD_TYPE_KRETPROBE
1241                                               : BPF_FD_TYPE_KPROBE;
1242         if (tk->symbol) {
1243                 *symbol = tk->symbol;
1244                 *probe_offset = tk->rp.kp.offset;
1245                 *probe_addr = 0;
1246         } else {
1247                 *symbol = NULL;
1248                 *probe_offset = 0;
1249                 *probe_addr = (unsigned long)tk->rp.kp.addr;
1250         }
1251         return 0;
1252 }
1253 #endif  /* CONFIG_PERF_EVENTS */
1254
1255 /*
1256  * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex.
1257  *
1258  * kprobe_trace_self_tests_init() does enable_trace_probe/disable_trace_probe
1259  * lockless, but we can't race with this __init function.
1260  */
1261 static int kprobe_register(struct trace_event_call *event,
1262                            enum trace_reg type, void *data)
1263 {
1264         struct trace_kprobe *tk = (struct trace_kprobe *)event->data;
1265         struct trace_event_file *file = data;
1266
1267         switch (type) {
1268         case TRACE_REG_REGISTER:
1269                 return enable_trace_kprobe(tk, file);
1270         case TRACE_REG_UNREGISTER:
1271                 return disable_trace_kprobe(tk, file);
1272
1273 #ifdef CONFIG_PERF_EVENTS
1274         case TRACE_REG_PERF_REGISTER:
1275                 return enable_trace_kprobe(tk, NULL);
1276         case TRACE_REG_PERF_UNREGISTER:
1277                 return disable_trace_kprobe(tk, NULL);
1278         case TRACE_REG_PERF_OPEN:
1279         case TRACE_REG_PERF_CLOSE:
1280         case TRACE_REG_PERF_ADD:
1281         case TRACE_REG_PERF_DEL:
1282                 return 0;
1283 #endif
1284         }
1285         return 0;
1286 }
1287
1288 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1289 {
1290         struct trace_kprobe *tk = container_of(kp, struct trace_kprobe, rp.kp);
1291         int ret = 0;
1292
1293         raw_cpu_inc(*tk->nhit);
1294
1295         if (tk->tp.flags & TP_FLAG_TRACE)
1296                 kprobe_trace_func(tk, regs);
1297 #ifdef CONFIG_PERF_EVENTS
1298         if (tk->tp.flags & TP_FLAG_PROFILE)
1299                 ret = kprobe_perf_func(tk, regs);
1300 #endif
1301         return ret;
1302 }
1303 NOKPROBE_SYMBOL(kprobe_dispatcher);
1304
1305 static int
1306 kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1307 {
1308         struct trace_kprobe *tk = container_of(ri->rp, struct trace_kprobe, rp);
1309
1310         raw_cpu_inc(*tk->nhit);
1311
1312         if (tk->tp.flags & TP_FLAG_TRACE)
1313                 kretprobe_trace_func(tk, ri, regs);
1314 #ifdef CONFIG_PERF_EVENTS
1315         if (tk->tp.flags & TP_FLAG_PROFILE)
1316                 kretprobe_perf_func(tk, ri, regs);
1317 #endif
1318         return 0;       /* We don't tweek kernel, so just return 0 */
1319 }
1320 NOKPROBE_SYMBOL(kretprobe_dispatcher);
1321
1322 static struct trace_event_functions kretprobe_funcs = {
1323         .trace          = print_kretprobe_event
1324 };
1325
1326 static struct trace_event_functions kprobe_funcs = {
1327         .trace          = print_kprobe_event
1328 };
1329
1330 static inline void init_trace_event_call(struct trace_kprobe *tk,
1331                                          struct trace_event_call *call)
1332 {
1333         INIT_LIST_HEAD(&call->class->fields);
1334         if (trace_kprobe_is_return(tk)) {
1335                 call->event.funcs = &kretprobe_funcs;
1336                 call->class->define_fields = kretprobe_event_define_fields;
1337         } else {
1338                 call->event.funcs = &kprobe_funcs;
1339                 call->class->define_fields = kprobe_event_define_fields;
1340         }
1341
1342         call->flags = TRACE_EVENT_FL_KPROBE;
1343         call->class->reg = kprobe_register;
1344         call->data = tk;
1345 }
1346
1347 static int register_kprobe_event(struct trace_kprobe *tk)
1348 {
1349         struct trace_event_call *call = &tk->tp.call;
1350         int ret = 0;
1351
1352         init_trace_event_call(tk, call);
1353
1354         if (traceprobe_set_print_fmt(&tk->tp, trace_kprobe_is_return(tk)) < 0)
1355                 return -ENOMEM;
1356         ret = register_trace_event(&call->event);
1357         if (!ret) {
1358                 kfree(call->print_fmt);
1359                 return -ENODEV;
1360         }
1361         ret = trace_add_event_call(call);
1362         if (ret) {
1363                 pr_info("Failed to register kprobe event: %s\n",
1364                         trace_event_name(call));
1365                 kfree(call->print_fmt);
1366                 unregister_trace_event(&call->event);
1367         }
1368         return ret;
1369 }
1370
1371 static int unregister_kprobe_event(struct trace_kprobe *tk)
1372 {
1373         int ret;
1374
1375         /* tp->event is unregistered in trace_remove_event_call() */
1376         ret = trace_remove_event_call(&tk->tp.call);
1377         if (!ret)
1378                 kfree(tk->tp.call.print_fmt);
1379         return ret;
1380 }
1381
1382 #ifdef CONFIG_PERF_EVENTS
1383 /* create a trace_kprobe, but don't add it to global lists */
1384 struct trace_event_call *
1385 create_local_trace_kprobe(char *func, void *addr, unsigned long offs,
1386                           bool is_return)
1387 {
1388         struct trace_kprobe *tk;
1389         int ret;
1390         char *event;
1391
1392         /*
1393          * local trace_kprobes are not added to dyn_event, so they are never
1394          * searched in find_trace_kprobe(). Therefore, there is no concern of
1395          * duplicated name here.
1396          */
1397         event = func ? func : "DUMMY_EVENT";
1398
1399         tk = alloc_trace_kprobe(KPROBE_EVENT_SYSTEM, event, (void *)addr, func,
1400                                 offs, 0 /* maxactive */, 0 /* nargs */,
1401                                 is_return);
1402
1403         if (IS_ERR(tk)) {
1404                 pr_info("Failed to allocate trace_probe.(%d)\n",
1405                         (int)PTR_ERR(tk));
1406                 return ERR_CAST(tk);
1407         }
1408
1409         init_trace_event_call(tk, &tk->tp.call);
1410
1411         if (traceprobe_set_print_fmt(&tk->tp, trace_kprobe_is_return(tk)) < 0) {
1412                 ret = -ENOMEM;
1413                 goto error;
1414         }
1415
1416         ret = __register_trace_kprobe(tk);
1417         if (ret < 0) {
1418                 kfree(tk->tp.call.print_fmt);
1419                 goto error;
1420         }
1421
1422         return &tk->tp.call;
1423 error:
1424         free_trace_kprobe(tk);
1425         return ERR_PTR(ret);
1426 }
1427
1428 void destroy_local_trace_kprobe(struct trace_event_call *event_call)
1429 {
1430         struct trace_kprobe *tk;
1431
1432         tk = container_of(event_call, struct trace_kprobe, tp.call);
1433
1434         if (trace_probe_is_enabled(&tk->tp)) {
1435                 WARN_ON(1);
1436                 return;
1437         }
1438
1439         __unregister_trace_kprobe(tk);
1440
1441         kfree(tk->tp.call.print_fmt);
1442         free_trace_kprobe(tk);
1443 }
1444 #endif /* CONFIG_PERF_EVENTS */
1445
1446 /* Make a tracefs interface for controlling probe points */
1447 static __init int init_kprobe_trace(void)
1448 {
1449         struct dentry *d_tracer;
1450         struct dentry *entry;
1451         int ret;
1452
1453         ret = dyn_event_register(&trace_kprobe_ops);
1454         if (ret)
1455                 return ret;
1456
1457         if (register_module_notifier(&trace_kprobe_module_nb))
1458                 return -EINVAL;
1459
1460         d_tracer = tracing_init_dentry();
1461         if (IS_ERR(d_tracer))
1462                 return 0;
1463
1464         entry = tracefs_create_file("kprobe_events", 0644, d_tracer,
1465                                     NULL, &kprobe_events_ops);
1466
1467         /* Event list interface */
1468         if (!entry)
1469                 pr_warn("Could not create tracefs 'kprobe_events' entry\n");
1470
1471         /* Profile interface */
1472         entry = tracefs_create_file("kprobe_profile", 0444, d_tracer,
1473                                     NULL, &kprobe_profile_ops);
1474
1475         if (!entry)
1476                 pr_warn("Could not create tracefs 'kprobe_profile' entry\n");
1477         return 0;
1478 }
1479 fs_initcall(init_kprobe_trace);
1480
1481
1482 #ifdef CONFIG_FTRACE_STARTUP_TEST
1483 static __init struct trace_event_file *
1484 find_trace_probe_file(struct trace_kprobe *tk, struct trace_array *tr)
1485 {
1486         struct trace_event_file *file;
1487
1488         list_for_each_entry(file, &tr->events, list)
1489                 if (file->event_call == &tk->tp.call)
1490                         return file;
1491
1492         return NULL;
1493 }
1494
1495 /*
1496  * Nobody but us can call enable_trace_kprobe/disable_trace_kprobe at this
1497  * stage, we can do this lockless.
1498  */
1499 static __init int kprobe_trace_self_tests_init(void)
1500 {
1501         int ret, warn = 0;
1502         int (*target)(int, int, int, int, int, int);
1503         struct trace_kprobe *tk;
1504         struct trace_event_file *file;
1505
1506         if (tracing_is_disabled())
1507                 return -ENODEV;
1508
1509         target = kprobe_trace_selftest_target;
1510
1511         pr_info("Testing kprobe tracing: ");
1512
1513         ret = trace_run_command("p:testprobe kprobe_trace_selftest_target $stack $stack0 +0($stack)",
1514                                 create_or_delete_trace_kprobe);
1515         if (WARN_ON_ONCE(ret)) {
1516                 pr_warn("error on probing function entry.\n");
1517                 warn++;
1518         } else {
1519                 /* Enable trace point */
1520                 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1521                 if (WARN_ON_ONCE(tk == NULL)) {
1522                         pr_warn("error on getting new probe.\n");
1523                         warn++;
1524                 } else {
1525                         file = find_trace_probe_file(tk, top_trace_array());
1526                         if (WARN_ON_ONCE(file == NULL)) {
1527                                 pr_warn("error on getting probe file.\n");
1528                                 warn++;
1529                         } else
1530                                 enable_trace_kprobe(tk, file);
1531                 }
1532         }
1533
1534         ret = trace_run_command("r:testprobe2 kprobe_trace_selftest_target $retval",
1535                                 create_or_delete_trace_kprobe);
1536         if (WARN_ON_ONCE(ret)) {
1537                 pr_warn("error on probing function return.\n");
1538                 warn++;
1539         } else {
1540                 /* Enable trace point */
1541                 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1542                 if (WARN_ON_ONCE(tk == NULL)) {
1543                         pr_warn("error on getting 2nd new probe.\n");
1544                         warn++;
1545                 } else {
1546                         file = find_trace_probe_file(tk, top_trace_array());
1547                         if (WARN_ON_ONCE(file == NULL)) {
1548                                 pr_warn("error on getting probe file.\n");
1549                                 warn++;
1550                         } else
1551                                 enable_trace_kprobe(tk, file);
1552                 }
1553         }
1554
1555         if (warn)
1556                 goto end;
1557
1558         ret = target(1, 2, 3, 4, 5, 6);
1559
1560         /*
1561          * Not expecting an error here, the check is only to prevent the
1562          * optimizer from removing the call to target() as otherwise there
1563          * are no side-effects and the call is never performed.
1564          */
1565         if (ret != 21)
1566                 warn++;
1567
1568         /* Disable trace points before removing it */
1569         tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1570         if (WARN_ON_ONCE(tk == NULL)) {
1571                 pr_warn("error on getting test probe.\n");
1572                 warn++;
1573         } else {
1574                 if (trace_kprobe_nhit(tk) != 1) {
1575                         pr_warn("incorrect number of testprobe hits\n");
1576                         warn++;
1577                 }
1578
1579                 file = find_trace_probe_file(tk, top_trace_array());
1580                 if (WARN_ON_ONCE(file == NULL)) {
1581                         pr_warn("error on getting probe file.\n");
1582                         warn++;
1583                 } else
1584                         disable_trace_kprobe(tk, file);
1585         }
1586
1587         tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1588         if (WARN_ON_ONCE(tk == NULL)) {
1589                 pr_warn("error on getting 2nd test probe.\n");
1590                 warn++;
1591         } else {
1592                 if (trace_kprobe_nhit(tk) != 1) {
1593                         pr_warn("incorrect number of testprobe2 hits\n");
1594                         warn++;
1595                 }
1596
1597                 file = find_trace_probe_file(tk, top_trace_array());
1598                 if (WARN_ON_ONCE(file == NULL)) {
1599                         pr_warn("error on getting probe file.\n");
1600                         warn++;
1601                 } else
1602                         disable_trace_kprobe(tk, file);
1603         }
1604
1605         ret = trace_run_command("-:testprobe", create_or_delete_trace_kprobe);
1606         if (WARN_ON_ONCE(ret)) {
1607                 pr_warn("error on deleting a probe.\n");
1608                 warn++;
1609         }
1610
1611         ret = trace_run_command("-:testprobe2", create_or_delete_trace_kprobe);
1612         if (WARN_ON_ONCE(ret)) {
1613                 pr_warn("error on deleting a probe.\n");
1614                 warn++;
1615         }
1616
1617 end:
1618         ret = dyn_events_release_all(&trace_kprobe_ops);
1619         if (WARN_ON_ONCE(ret)) {
1620                 pr_warn("error on cleaning up probes.\n");
1621                 warn++;
1622         }
1623         /*
1624          * Wait for the optimizer work to finish. Otherwise it might fiddle
1625          * with probes in already freed __init text.
1626          */
1627         wait_for_kprobe_optimizer();
1628         if (warn)
1629                 pr_cont("NG: Some tests are failed. Please check them.\n");
1630         else
1631                 pr_cont("OK\n");
1632         return 0;
1633 }
1634
1635 late_initcall(kprobe_trace_self_tests_init);
1636
1637 #endif