kernel/trace/trace_kprobe.c

   1 /*
   2  * Kprobes-based tracing events
   3  *
   4  * Created by Masami Hiramatsu <mhiramat@redhat.com>
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 as
   8  * published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful,
  11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  * GNU General Public License for more details.
  14  *
  15  * You should have received a copy of the GNU General Public License
  16  * along with this program; if not, write to the Free Software
  17  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
  18  */
  19
  20 #include <linux/module.h>
  21 #include <linux/uaccess.h>
  22
  23 #include "trace_probe.h"
  24
  25 #define KPROBE_EVENT_SYSTEM "kprobes"
  26
  27 /**
  28  * Kprobe event core functions
  29  */
  30 struct trace_probe {
  31         struct list_head        list;
  32         struct kretprobe        rp;     /* Use rp.kp for kprobe use */
  33         unsigned long           nhit;
  34         unsigned int            flags;  /* For TP_FLAG_* */
  35         const char              *symbol;        /* symbol name */
  36         struct ftrace_event_class       class;
  37         struct ftrace_event_call        call;
  38         struct ftrace_event_file * __rcu *files;
  39         ssize_t                 size;           /* trace entry size */
  40         unsigned int            nr_args;
  41         struct probe_arg        args[];
  42 };
  43
  44 #define SIZEOF_TRACE_PROBE(n)                   \
  45         (offsetof(struct trace_probe, args) +   \
  46         (sizeof(struct probe_arg) * (n)))
  47
  48
  49 static __kprobes bool trace_probe_is_return(struct trace_probe *tp)
  50 {
  51         return tp->rp.handler != NULL;
  52 }
  53
  54 static __kprobes const char *trace_probe_symbol(struct trace_probe *tp)
  55 {
  56         return tp->symbol ? tp->symbol : "unknown";
  57 }
  58
  59 static __kprobes unsigned long trace_probe_offset(struct trace_probe *tp)
  60 {
  61         return tp->rp.kp.offset;
  62 }
  63
  64 static __kprobes bool trace_probe_is_enabled(struct trace_probe *tp)
  65 {
  66         return !!(tp->flags & (TP_FLAG_TRACE | TP_FLAG_PROFILE));
  67 }
  68
  69 static __kprobes bool trace_probe_is_registered(struct trace_probe *tp)
  70 {
  71         return !!(tp->flags & TP_FLAG_REGISTERED);
  72 }
  73
  74 static __kprobes bool trace_probe_has_gone(struct trace_probe *tp)
  75 {
  76         return !!(kprobe_gone(&tp->rp.kp));
  77 }
  78
  79 static __kprobes bool trace_probe_within_module(struct trace_probe *tp,
  80                                                 struct module *mod)
  81 {
  82         int len = strlen(mod->name);
  83         const char *name = trace_probe_symbol(tp);
  84         return strncmp(mod->name, name, len) == 0 && name[len] == ':';
  85 }
  86
  87 static __kprobes bool trace_probe_is_on_module(struct trace_probe *tp)
  88 {
  89         return !!strchr(trace_probe_symbol(tp), ':');
  90 }
  91
  92 static int register_probe_event(struct trace_probe *tp);
  93 static int unregister_probe_event(struct trace_probe *tp);
  94
  95 static DEFINE_MUTEX(probe_lock);
  96 static LIST_HEAD(probe_list);
  97
  98 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
  99 static int kretprobe_dispatcher(struct kretprobe_instance *ri,
 100                                 struct pt_regs *regs);
 101
 102 /*
 103  * Allocate new trace_probe and initialize it (including kprobes).
 104  */
 105 static struct trace_probe *alloc_trace_probe(const char *group,
 106                                              const char *event,
 107                                              void *addr,
 108                                              const char *symbol,
 109                                              unsigned long offs,
 110                                              int nargs, bool is_return)
 111 {
 112         struct trace_probe *tp;
 113         int ret = -ENOMEM;
 114
 115         tp = kzalloc(SIZEOF_TRACE_PROBE(nargs), GFP_KERNEL);
 116         if (!tp)
 117                 return ERR_PTR(ret);
 118
 119         if (symbol) {
 120                 tp->symbol = kstrdup(symbol, GFP_KERNEL);
 121                 if (!tp->symbol)
 122                         goto error;
 123                 tp->rp.kp.symbol_name = tp->symbol;
 124                 tp->rp.kp.offset = offs;
 125         } else
 126                 tp->rp.kp.addr = addr;
 127
 128         if (is_return)
 129                 tp->rp.handler = kretprobe_dispatcher;
 130         else
 131                 tp->rp.kp.pre_handler = kprobe_dispatcher;
 132
 133         if (!event || !is_good_name(event)) {
 134                 ret = -EINVAL;
 135                 goto error;
 136         }
 137
 138         tp->call.class = &tp->class;
 139         tp->call.name = kstrdup(event, GFP_KERNEL);
 140         if (!tp->call.name)
 141                 goto error;
 142
 143         if (!group || !is_good_name(group)) {
 144                 ret = -EINVAL;
 145                 goto error;
 146         }
 147
 148         tp->class.system = kstrdup(group, GFP_KERNEL);
 149         if (!tp->class.system)
 150                 goto error;
 151
 152         INIT_LIST_HEAD(&tp->list);
 153         return tp;
 154 error:
 155         kfree(tp->call.name);
 156         kfree(tp->symbol);
 157         kfree(tp);
 158         return ERR_PTR(ret);
 159 }
 160
 161 static void free_trace_probe(struct trace_probe *tp)
 162 {
 163         int i;
 164
 165         for (i = 0; i < tp->nr_args; i++)
 166                 traceprobe_free_probe_arg(&tp->args[i]);
 167
 168         kfree(tp->call.class->system);
 169         kfree(tp->call.name);
 170         kfree(tp->symbol);
 171         kfree(tp);
 172 }
 173
 174 static struct trace_probe *find_trace_probe(const char *event,
 175                                             const char *group)
 176 {
 177         struct trace_probe *tp;
 178
 179         list_for_each_entry(tp, &probe_list, list)
 180                 if (strcmp(tp->call.name, event) == 0 &&
 181                     strcmp(tp->call.class->system, group) == 0)
 182                         return tp;
 183         return NULL;
 184 }
 185
 186 static int trace_probe_nr_files(struct trace_probe *tp)
 187 {
 188         struct ftrace_event_file **file;
 189         int ret = 0;
 190
 191         /*
 192          * Since all tp->files updater is protected by probe_enable_lock,
 193          * we don't need to lock an rcu_read_lock.
 194          */
 195         file = rcu_dereference_raw(tp->files);
 196         if (file)
 197                 while (*(file++))
 198                         ret++;
 199
 200         return ret;
 201 }
 202
 203 static DEFINE_MUTEX(probe_enable_lock);
 204
 205 /*
 206  * Enable trace_probe
 207  * if the file is NULL, enable "perf" handler, or enable "trace" handler.
 208  */
 209 static int
 210 enable_trace_probe(struct trace_probe *tp, struct ftrace_event_file *file)
 211 {
 212         int ret = 0;
 213
 214         mutex_lock(&probe_enable_lock);
 215
 216         if (file) {
 217                 struct ftrace_event_file **new, **old;
 218                 int n = trace_probe_nr_files(tp);
 219
 220                 old = rcu_dereference_raw(tp->files);
 221                 /* 1 is for new one and 1 is for stopper */
 222                 new = kzalloc((n + 2) * sizeof(struct ftrace_event_file *),
 223                               GFP_KERNEL);
 224                 if (!new) {
 225                         ret = -ENOMEM;
 226                         goto out_unlock;
 227                 }
 228                 memcpy(new, old, n * sizeof(struct ftrace_event_file *));
 229                 new[n] = file;
 230                 /* The last one keeps a NULL */
 231
 232                 rcu_assign_pointer(tp->files, new);
 233                 tp->flags |= TP_FLAG_TRACE;
 234
 235                 if (old) {
 236                         /* Make sure the probe is done with old files */
 237                         synchronize_sched();
 238                         kfree(old);
 239                 }
 240         } else
 241                 tp->flags |= TP_FLAG_PROFILE;
 242
 243         if (trace_probe_is_enabled(tp) && trace_probe_is_registered(tp) &&
 244             !trace_probe_has_gone(tp)) {
 245                 if (trace_probe_is_return(tp))
 246                         ret = enable_kretprobe(&tp->rp);
 247                 else
 248                         ret = enable_kprobe(&tp->rp.kp);
 249         }
 250
 251  out_unlock:
 252         mutex_unlock(&probe_enable_lock);
 253
 254         return ret;
 255 }
 256
 257 static int
 258 trace_probe_file_index(struct trace_probe *tp, struct ftrace_event_file *file)
 259 {
 260         struct ftrace_event_file **files;
 261         int i;
 262
 263         /*
 264          * Since all tp->files updater is protected by probe_enable_lock,
 265          * we don't need to lock an rcu_read_lock.
 266          */
 267         files = rcu_dereference_raw(tp->files);
 268         if (files) {
 269                 for (i = 0; files[i]; i++)
 270                         if (files[i] == file)
 271                                 return i;
 272         }
 273
 274         return -1;
 275 }
 276
 277 /*
 278  * Disable trace_probe
 279  * if the file is NULL, disable "perf" handler, or disable "trace" handler.
 280  */
 281 static int
 282 disable_trace_probe(struct trace_probe *tp, struct ftrace_event_file *file)
 283 {
 284         struct ftrace_event_file **old = NULL;
 285         int wait = 0;
 286         int ret = 0;
 287
 288         mutex_lock(&probe_enable_lock);
 289
 290         if (file) {
 291                 struct ftrace_event_file **new, **old;
 292                 int n = trace_probe_nr_files(tp);
 293                 int i, j;
 294
 295                 old = rcu_dereference_raw(tp->files);
 296                 if (n == 0 || trace_probe_file_index(tp, file) < 0) {
 297                         ret = -EINVAL;
 298                         goto out_unlock;
 299                 }
 300
 301                 if (n == 1) {   /* Remove the last file */
 302                         tp->flags &= ~TP_FLAG_TRACE;
 303                         new = NULL;
 304                 } else {
 305                         new = kzalloc(n * sizeof(struct ftrace_event_file *),
 306                                       GFP_KERNEL);
 307                         if (!new) {
 308                                 ret = -ENOMEM;
 309                                 goto out_unlock;
 310                         }
 311
 312                         /* This copy & check loop copies the NULL stopper too */
 313                         for (i = 0, j = 0; j < n && i < n + 1; i++)
 314                                 if (old[i] != file)
 315                                         new[j++] = old[i];
 316                 }
 317
 318                 rcu_assign_pointer(tp->files, new);
 319                 wait = 1;
 320         } else
 321                 tp->flags &= ~TP_FLAG_PROFILE;
 322
 323         if (!trace_probe_is_enabled(tp) && trace_probe_is_registered(tp)) {
 324                 if (trace_probe_is_return(tp))
 325                         disable_kretprobe(&tp->rp);
 326                 else
 327                         disable_kprobe(&tp->rp.kp);
 328                 wait = 1;
 329         }
 330
 331  out_unlock:
 332         mutex_unlock(&probe_enable_lock);
 333
 334         if (wait) {
 335                 /*
 336                  * Synchronize with kprobe_trace_func/kretprobe_trace_func
 337                  * to ensure disabled (all running handlers are finished).
 338                  * This is not only for kfree(), but also the caller,
 339                  * trace_remove_event_call() supposes it for releasing
 340                  * event_call related objects, which will be accessed in
 341                  * the kprobe_trace_func/kretprobe_trace_func.
 342                  */
 343                 synchronize_sched();
 344                 kfree(old);     /* Ignored if link == NULL */
 345         }
 346
 347         return ret;
 348 }
 349
 350 /* Internal register function - just handle k*probes and flags */
 351 static int __register_trace_probe(struct trace_probe *tp)
 352 {
 353         int i, ret;
 354
 355         if (trace_probe_is_registered(tp))
 356                 return -EINVAL;
 357
 358         for (i = 0; i < tp->nr_args; i++)
 359                 traceprobe_update_arg(&tp->args[i]);
 360
 361         /* Set/clear disabled flag according to tp->flag */
 362         if (trace_probe_is_enabled(tp))
 363                 tp->rp.kp.flags &= ~KPROBE_FLAG_DISABLED;
 364         else
 365                 tp->rp.kp.flags |= KPROBE_FLAG_DISABLED;
 366
 367         if (trace_probe_is_return(tp))
 368                 ret = register_kretprobe(&tp->rp);
 369         else
 370                 ret = register_kprobe(&tp->rp.kp);
 371
 372         if (ret == 0)
 373                 tp->flags |= TP_FLAG_REGISTERED;
 374         else {
 375                 pr_warning("Could not insert probe at %s+%lu: %d\n",
 376                            trace_probe_symbol(tp), trace_probe_offset(tp), ret);
 377                 if (ret == -ENOENT && trace_probe_is_on_module(tp)) {
 378                         pr_warning("This probe might be able to register after"
 379                                    "target module is loaded. Continue.\n");
 380                         ret = 0;
 381                 } else if (ret == -EILSEQ) {
 382                         pr_warning("Probing address(0x%p) is not an "
 383                                    "instruction boundary.\n",
 384                                    tp->rp.kp.addr);
 385                         ret = -EINVAL;
 386                 }
 387         }
 388
 389         return ret;
 390 }
 391
 392 /* Internal unregister function - just handle k*probes and flags */
 393 static void __unregister_trace_probe(struct trace_probe *tp)
 394 {
 395         if (trace_probe_is_registered(tp)) {
 396                 if (trace_probe_is_return(tp))
 397                         unregister_kretprobe(&tp->rp);
 398                 else
 399                         unregister_kprobe(&tp->rp.kp);
 400                 tp->flags &= ~TP_FLAG_REGISTERED;
 401                 /* Cleanup kprobe for reuse */
 402                 if (tp->rp.kp.symbol_name)
 403                         tp->rp.kp.addr = NULL;
 404         }
 405 }
 406
 407 /* Unregister a trace_probe and probe_event: call with locking probe_lock */
 408 static int unregister_trace_probe(struct trace_probe *tp)
 409 {
 410         /* Enabled event can not be unregistered */
 411         if (trace_probe_is_enabled(tp))
 412                 return -EBUSY;
 413
 414         /* Will fail if probe is being used by ftrace or perf */
 415         if (unregister_probe_event(tp))
 416                 return -EBUSY;
 417
 418         __unregister_trace_probe(tp);
 419         list_del(&tp->list);
 420
 421         return 0;
 422 }
 423
 424 /* Register a trace_probe and probe_event */
 425 static int register_trace_probe(struct trace_probe *tp)
 426 {
 427         struct trace_probe *old_tp;
 428         int ret;
 429
 430         mutex_lock(&probe_lock);
 431
 432         /* Delete old (same name) event if exist */
 433         old_tp = find_trace_probe(tp->call.name, tp->call.class->system);
 434         if (old_tp) {
 435                 ret = unregister_trace_probe(old_tp);
 436                 if (ret < 0)
 437                         goto end;
 438                 free_trace_probe(old_tp);
 439         }
 440
 441         /* Register new event */
 442         ret = register_probe_event(tp);
 443         if (ret) {
 444                 pr_warning("Failed to register probe event(%d)\n", ret);
 445                 goto end;
 446         }
 447
 448         /* Register k*probe */
 449         ret = __register_trace_probe(tp);
 450         if (ret < 0)
 451                 unregister_probe_event(tp);
 452         else
 453                 list_add_tail(&tp->list, &probe_list);
 454
 455 end:
 456         mutex_unlock(&probe_lock);
 457         return ret;
 458 }
 459
 460 /* Module notifier call back, checking event on the module */
 461 static int trace_probe_module_callback(struct notifier_block *nb,
 462                                        unsigned long val, void *data)
 463 {
 464         struct module *mod = data;
 465         struct trace_probe *tp;
 466         int ret;
 467
 468         if (val != MODULE_STATE_COMING)
 469                 return NOTIFY_DONE;
 470
 471         /* Update probes on coming module */
 472         mutex_lock(&probe_lock);
 473         list_for_each_entry(tp, &probe_list, list) {
 474                 if (trace_probe_within_module(tp, mod)) {
 475                         /* Don't need to check busy - this should have gone. */
 476                         __unregister_trace_probe(tp);
 477                         ret = __register_trace_probe(tp);
 478                         if (ret)
 479                                 pr_warning("Failed to re-register probe %s on"
 480                                            "%s: %d\n",
 481                                            tp->call.name, mod->name, ret);
 482                 }
 483         }
 484         mutex_unlock(&probe_lock);
 485
 486         return NOTIFY_DONE;
 487 }
 488
 489 static struct notifier_block trace_probe_module_nb = {
 490         .notifier_call = trace_probe_module_callback,
 491         .priority = 1   /* Invoked after kprobe module callback */
 492 };
 493
 494 static int create_trace_probe(int argc, char **argv)
 495 {
 496         /*
 497          * Argument syntax:
 498          *  - Add kprobe: p[:[GRP/]EVENT] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
 499          *  - Add kretprobe: r[:[GRP/]EVENT] [MOD:]KSYM[+0] [FETCHARGS]
 500          * Fetch args:
 501          *  $retval     : fetch return value
 502          *  $stack      : fetch stack address
 503          *  $stackN     : fetch Nth of stack (N:0-)
 504          *  @ADDR       : fetch memory at ADDR (ADDR should be in kernel)
 505          *  @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
 506          *  %REG        : fetch register REG
 507          * Dereferencing memory fetch:
 508          *  +|-offs(ARG) : fetch memory at ARG +|- offs address.
 509          * Alias name of args:
 510          *  NAME=FETCHARG : set NAME as alias of FETCHARG.
 511          * Type of args:
 512          *  FETCHARG:TYPE : use TYPE instead of unsigned long.
 513          */
 514         struct trace_probe *tp;
 515         int i, ret = 0;
 516         bool is_return = false, is_delete = false;
 517         char *symbol = NULL, *event = NULL, *group = NULL;
 518         char *arg;
 519         unsigned long offset = 0;
 520         void *addr = NULL;
 521         char buf[MAX_EVENT_NAME_LEN];
 522
 523         /* argc must be >= 1 */
 524         if (argv[0][0] == 'p')
 525                 is_return = false;
 526         else if (argv[0][0] == 'r')
 527                 is_return = true;
 528         else if (argv[0][0] == '-')
 529                 is_delete = true;
 530         else {
 531                 pr_info("Probe definition must be started with 'p', 'r' or"
 532                         " '-'.\n");
 533                 return -EINVAL;
 534         }
 535
 536         if (argv[0][1] == ':') {
 537                 event = &argv[0][2];
 538                 if (strchr(event, '/')) {
 539                         group = event;
 540                         event = strchr(group, '/') + 1;
 541                         event[-1] = '\0';
 542                         if (strlen(group) == 0) {
 543                                 pr_info("Group name is not specified\n");
 544                                 return -EINVAL;
 545                         }
 546                 }
 547                 if (strlen(event) == 0) {
 548                         pr_info("Event name is not specified\n");
 549                         return -EINVAL;
 550                 }
 551         }
 552         if (!group)
 553                 group = KPROBE_EVENT_SYSTEM;
 554
 555         if (is_delete) {
 556                 if (!event) {
 557                         pr_info("Delete command needs an event name.\n");
 558                         return -EINVAL;
 559                 }
 560                 mutex_lock(&probe_lock);
 561                 tp = find_trace_probe(event, group);
 562                 if (!tp) {
 563                         mutex_unlock(&probe_lock);
 564                         pr_info("Event %s/%s doesn't exist.\n", group, event);
 565                         return -ENOENT;
 566                 }
 567                 /* delete an event */
 568                 ret = unregister_trace_probe(tp);
 569                 if (ret == 0)
 570                         free_trace_probe(tp);
 571                 mutex_unlock(&probe_lock);
 572                 return ret;
 573         }
 574
 575         if (argc < 2) {
 576                 pr_info("Probe point is not specified.\n");
 577                 return -EINVAL;
 578         }
 579         if (isdigit(argv[1][0])) {
 580                 if (is_return) {
 581                         pr_info("Return probe point must be a symbol.\n");
 582                         return -EINVAL;
 583                 }
 584                 /* an address specified */
 585                 ret = kstrtoul(&argv[1][0], 0, (unsigned long *)&addr);
 586                 if (ret) {
 587                         pr_info("Failed to parse address.\n");
 588                         return ret;
 589                 }
 590         } else {
 591                 /* a symbol specified */
 592                 symbol = argv[1];
 593                 /* TODO: support .init module functions */
 594                 ret = traceprobe_split_symbol_offset(symbol, &offset);
 595                 if (ret) {
 596                         pr_info("Failed to parse symbol.\n");
 597                         return ret;
 598                 }
 599                 if (offset && is_return) {
 600                         pr_info("Return probe must be used without offset.\n");
 601                         return -EINVAL;
 602                 }
 603         }
 604         argc -= 2; argv += 2;
 605
 606         /* setup a probe */
 607         if (!event) {
 608                 /* Make a new event name */
 609                 if (symbol)
 610                         snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
 611                                  is_return ? 'r' : 'p', symbol, offset);
 612                 else
 613                         snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
 614                                  is_return ? 'r' : 'p', addr);
 615                 event = buf;
 616         }
 617         tp = alloc_trace_probe(group, event, addr, symbol, offset, argc,
 618                                is_return);
 619         if (IS_ERR(tp)) {
 620                 pr_info("Failed to allocate trace_probe.(%d)\n",
 621                         (int)PTR_ERR(tp));
 622                 return PTR_ERR(tp);
 623         }
 624
 625         /* parse arguments */
 626         ret = 0;
 627         for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
 628                 /* Increment count for freeing args in error case */
 629                 tp->nr_args++;
 630
 631                 /* Parse argument name */
 632                 arg = strchr(argv[i], '=');
 633                 if (arg) {
 634                         *arg++ = '\0';
 635                         tp->args[i].name = kstrdup(argv[i], GFP_KERNEL);
 636                 } else {
 637                         arg = argv[i];
 638                         /* If argument name is omitted, set "argN" */
 639                         snprintf(buf, MAX_EVENT_NAME_LEN, "arg%d", i + 1);
 640                         tp->args[i].name = kstrdup(buf, GFP_KERNEL);
 641                 }
 642
 643                 if (!tp->args[i].name) {
 644                         pr_info("Failed to allocate argument[%d] name.\n", i);
 645                         ret = -ENOMEM;
 646                         goto error;
 647                 }
 648
 649                 if (!is_good_name(tp->args[i].name)) {
 650                         pr_info("Invalid argument[%d] name: %s\n",
 651                                 i, tp->args[i].name);
 652                         ret = -EINVAL;
 653                         goto error;
 654                 }
 655
 656                 if (traceprobe_conflict_field_name(tp->args[i].name,
 657                                                         tp->args, i)) {
 658                         pr_info("Argument[%d] name '%s' conflicts with "
 659                                 "another field.\n", i, argv[i]);
 660                         ret = -EINVAL;
 661                         goto error;
 662                 }
 663
 664                 /* Parse fetch argument */
 665                 ret = traceprobe_parse_probe_arg(arg, &tp->size, &tp->args[i],
 666                                                 is_return, true);
 667                 if (ret) {
 668                         pr_info("Parse error at argument[%d]. (%d)\n", i, ret);
 669                         goto error;
 670                 }
 671         }
 672
 673         ret = register_trace_probe(tp);
 674         if (ret)
 675                 goto error;
 676         return 0;
 677
 678 error:
 679         free_trace_probe(tp);
 680         return ret;
 681 }
 682
 683 static int release_all_trace_probes(void)
 684 {
 685         struct trace_probe *tp;
 686         int ret = 0;
 687
 688         mutex_lock(&probe_lock);
 689         /* Ensure no probe is in use. */
 690         list_for_each_entry(tp, &probe_list, list)
 691                 if (trace_probe_is_enabled(tp)) {
 692                         ret = -EBUSY;
 693                         goto end;
 694                 }
 695         /* TODO: Use batch unregistration */
 696         while (!list_empty(&probe_list)) {
 697                 tp = list_entry(probe_list.next, struct trace_probe, list);
 698                 ret = unregister_trace_probe(tp);
 699                 if (ret)
 700                         goto end;
 701                 free_trace_probe(tp);
 702         }
 703
 704 end:
 705         mutex_unlock(&probe_lock);
 706
 707         return ret;
 708 }
 709
 710 /* Probes listing interfaces */
 711 static void *probes_seq_start(struct seq_file *m, loff_t *pos)
 712 {
 713         mutex_lock(&probe_lock);
 714         return seq_list_start(&probe_list, *pos);
 715 }
 716
 717 static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
 718 {
 719         return seq_list_next(v, &probe_list, pos);
 720 }
 721
 722 static void probes_seq_stop(struct seq_file *m, void *v)
 723 {
 724         mutex_unlock(&probe_lock);
 725 }
 726
 727 static int probes_seq_show(struct seq_file *m, void *v)
 728 {
 729         struct trace_probe *tp = v;
 730         int i;
 731
 732         seq_printf(m, "%c", trace_probe_is_return(tp) ? 'r' : 'p');
 733         seq_printf(m, ":%s/%s", tp->call.class->system, tp->call.name);
 734
 735         if (!tp->symbol)
 736                 seq_printf(m, " 0x%p", tp->rp.kp.addr);
 737         else if (tp->rp.kp.offset)
 738                 seq_printf(m, " %s+%u", trace_probe_symbol(tp),
 739                            tp->rp.kp.offset);
 740         else
 741                 seq_printf(m, " %s", trace_probe_symbol(tp));
 742
 743         for (i = 0; i < tp->nr_args; i++)
 744                 seq_printf(m, " %s=%s", tp->args[i].name, tp->args[i].comm);
 745         seq_printf(m, "\n");
 746
 747         return 0;
 748 }
 749
 750 static const struct seq_operations probes_seq_op = {
 751         .start  = probes_seq_start,
 752         .next   = probes_seq_next,
 753         .stop   = probes_seq_stop,
 754         .show   = probes_seq_show
 755 };
 756
 757 static int probes_open(struct inode *inode, struct file *file)
 758 {
 759         int ret;
 760
 761         if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
 762                 ret = release_all_trace_probes();
 763                 if (ret < 0)
 764                         return ret;
 765         }
 766
 767         return seq_open(file, &probes_seq_op);
 768 }
 769
 770 static ssize_t probes_write(struct file *file, const char __user *buffer,
 771                             size_t count, loff_t *ppos)
 772 {
 773         return traceprobe_probes_write(file, buffer, count, ppos,
 774                         create_trace_probe);
 775 }
 776
 777 static const struct file_operations kprobe_events_ops = {
 778         .owner          = THIS_MODULE,
 779         .open           = probes_open,
 780         .read           = seq_read,
 781         .llseek         = seq_lseek,
 782         .release        = seq_release,
 783         .write          = probes_write,
 784 };
 785
 786 /* Probes profiling interfaces */
 787 static int probes_profile_seq_show(struct seq_file *m, void *v)
 788 {
 789         struct trace_probe *tp = v;
 790
 791         seq_printf(m, "  %-44s %15lu %15lu\n", tp->call.name, tp->nhit,
 792                    tp->rp.kp.nmissed);
 793
 794         return 0;
 795 }
 796
 797 static const struct seq_operations profile_seq_op = {
 798         .start  = probes_seq_start,
 799         .next   = probes_seq_next,
 800         .stop   = probes_seq_stop,
 801         .show   = probes_profile_seq_show
 802 };
 803
 804 static int profile_open(struct inode *inode, struct file *file)
 805 {
 806         return seq_open(file, &profile_seq_op);
 807 }
 808
 809 static const struct file_operations kprobe_profile_ops = {
 810         .owner          = THIS_MODULE,
 811         .open           = profile_open,
 812         .read           = seq_read,
 813         .llseek         = seq_lseek,
 814         .release        = seq_release,
 815 };
 816
 817 /* Sum up total data length for dynamic arraies (strings) */
 818 static __kprobes int __get_data_size(struct trace_probe *tp,
 819                                      struct pt_regs *regs)
 820 {
 821         int i, ret = 0;
 822         u32 len;
 823
 824         for (i = 0; i < tp->nr_args; i++)
 825                 if (unlikely(tp->args[i].fetch_size.fn)) {
 826                         call_fetch(&tp->args[i].fetch_size, regs, &len);
 827                         ret += len;
 828                 }
 829
 830         return ret;
 831 }
 832
 833 /* Store the value of each argument */
 834 static __kprobes void store_trace_args(int ent_size, struct trace_probe *tp,
 835                                        struct pt_regs *regs,
 836                                        u8 *data, int maxlen)
 837 {
 838         int i;
 839         u32 end = tp->size;
 840         u32 *dl;        /* Data (relative) location */
 841
 842         for (i = 0; i < tp->nr_args; i++) {
 843                 if (unlikely(tp->args[i].fetch_size.fn)) {
 844                         /*
 845                          * First, we set the relative location and
 846                          * maximum data length to *dl
 847                          */
 848                         dl = (u32 *)(data + tp->args[i].offset);
 849                         *dl = make_data_rloc(maxlen, end - tp->args[i].offset);
 850                         /* Then try to fetch string or dynamic array data */
 851                         call_fetch(&tp->args[i].fetch, regs, dl);
 852                         /* Reduce maximum length */
 853                         end += get_rloc_len(*dl);
 854                         maxlen -= get_rloc_len(*dl);
 855                         /* Trick here, convert data_rloc to data_loc */
 856                         *dl = convert_rloc_to_loc(*dl,
 857                                  ent_size + tp->args[i].offset);
 858                 } else
 859                         /* Just fetching data normally */
 860                         call_fetch(&tp->args[i].fetch, regs,
 861                                    data + tp->args[i].offset);
 862         }
 863 }
 864
 865 /* Kprobe handler */
 866 static __kprobes void
 867 __kprobe_trace_func(struct trace_probe *tp, struct pt_regs *regs,
 868                     struct ftrace_event_file *ftrace_file)
 869 {
 870         struct kprobe_trace_entry_head *entry;
 871         struct ring_buffer_event *event;
 872         struct ring_buffer *buffer;
 873         int size, dsize, pc;
 874         unsigned long irq_flags;
 875         struct ftrace_event_call *call = &tp->call;
 876
 877         WARN_ON(call != ftrace_file->event_call);
 878
 879         if (test_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT, &ftrace_file->flags))
 880                 return;
 881
 882         local_save_flags(irq_flags);
 883         pc = preempt_count();
 884
 885         dsize = __get_data_size(tp, regs);
 886         size = sizeof(*entry) + tp->size + dsize;
 887
 888         event = trace_event_buffer_lock_reserve(&buffer, ftrace_file,
 889                                                 call->event.type,
 890                                                 size, irq_flags, pc);
 891         if (!event)
 892                 return;
 893
 894         entry = ring_buffer_event_data(event);
 895         entry->ip = (unsigned long)tp->rp.kp.addr;
 896         store_trace_args(sizeof(*entry), tp, regs, (u8 *)&entry[1], dsize);
 897
 898         if (!filter_current_check_discard(buffer, call, entry, event))
 899                 trace_buffer_unlock_commit_regs(buffer, event,
 900                                                 irq_flags, pc, regs);
 901 }
 902
 903 static __kprobes void
 904 kprobe_trace_func(struct trace_probe *tp, struct pt_regs *regs)
 905 {
 906         /*
 907          * Note: preempt is already disabled around the kprobe handler.
 908          * However, we still need an smp_read_barrier_depends() corresponding
 909          * to smp_wmb() in rcu_assign_pointer() to access the pointer.
 910          */
 911         struct ftrace_event_file **file = rcu_dereference_raw(tp->files);
 912
 913         if (unlikely(!file))
 914                 return;
 915
 916         while (*file) {
 917                 __kprobe_trace_func(tp, regs, *file);
 918                 file++;
 919         }
 920 }
 921
 922 /* Kretprobe handler */
 923 static __kprobes void
 924 __kretprobe_trace_func(struct trace_probe *tp, struct kretprobe_instance *ri,
 925                        struct pt_regs *regs,
 926                        struct ftrace_event_file *ftrace_file)
 927 {
 928         struct kretprobe_trace_entry_head *entry;
 929         struct ring_buffer_event *event;
 930         struct ring_buffer *buffer;
 931         int size, pc, dsize;
 932         unsigned long irq_flags;
 933         struct ftrace_event_call *call = &tp->call;
 934
 935         WARN_ON(call != ftrace_file->event_call);
 936
 937         if (test_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT, &ftrace_file->flags))
 938                 return;
 939
 940         local_save_flags(irq_flags);
 941         pc = preempt_count();
 942
 943         dsize = __get_data_size(tp, regs);
 944         size = sizeof(*entry) + tp->size + dsize;
 945
 946         event = trace_event_buffer_lock_reserve(&buffer, ftrace_file,
 947                                                 call->event.type,
 948                                                 size, irq_flags, pc);
 949         if (!event)
 950                 return;
 951
 952         entry = ring_buffer_event_data(event);
 953         entry->func = (unsigned long)tp->rp.kp.addr;
 954         entry->ret_ip = (unsigned long)ri->ret_addr;
 955         store_trace_args(sizeof(*entry), tp, regs, (u8 *)&entry[1], dsize);
 956
 957         if (!filter_current_check_discard(buffer, call, entry, event))
 958                 trace_buffer_unlock_commit_regs(buffer, event,
 959                                                 irq_flags, pc, regs);
 960 }
 961
 962 static __kprobes void
 963 kretprobe_trace_func(struct trace_probe *tp, struct kretprobe_instance *ri,
 964                      struct pt_regs *regs)
 965 {
 966         /*
 967          * Note: preempt is already disabled around the kprobe handler.
 968          * However, we still need an smp_read_barrier_depends() corresponding
 969          * to smp_wmb() in rcu_assign_pointer() to access the pointer.
 970          */
 971         struct ftrace_event_file **file = rcu_dereference_raw(tp->files);
 972
 973         if (unlikely(!file))
 974                 return;
 975
 976         while (*file) {
 977                 __kretprobe_trace_func(tp, ri, regs, *file);
 978                 file++;
 979         }
 980 }
 981
 982 /* Event entry printers */
 983 static enum print_line_t
 984 print_kprobe_event(struct trace_iterator *iter, int flags,
 985                    struct trace_event *event)
 986 {
 987         struct kprobe_trace_entry_head *field;
 988         struct trace_seq *s = &iter->seq;
 989         struct trace_probe *tp;
 990         u8 *data;
 991         int i;
 992
 993         field = (struct kprobe_trace_entry_head *)iter->ent;
 994         tp = container_of(event, struct trace_probe, call.event);
 995
 996         if (!trace_seq_printf(s, "%s: (", tp->call.name))
 997                 goto partial;
 998
 999         if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
1000                 goto partial;
1001
1002         if (!trace_seq_puts(s, ")"))
1003                 goto partial;
1004
1005         data = (u8 *)&field[1];
1006         for (i = 0; i < tp->nr_args; i++)
1007                 if (!tp->args[i].type->print(s, tp->args[i].name,
1008                                              data + tp->args[i].offset, field))
1009                         goto partial;
1010
1011         if (!trace_seq_puts(s, "\n"))
1012                 goto partial;
1013
1014         return TRACE_TYPE_HANDLED;
1015 partial:
1016         return TRACE_TYPE_PARTIAL_LINE;
1017 }
1018
1019 static enum print_line_t
1020 print_kretprobe_event(struct trace_iterator *iter, int flags,
1021                       struct trace_event *event)
1022 {
1023         struct kretprobe_trace_entry_head *field;
1024         struct trace_seq *s = &iter->seq;
1025         struct trace_probe *tp;
1026         u8 *data;
1027         int i;
1028
1029         field = (struct kretprobe_trace_entry_head *)iter->ent;
1030         tp = container_of(event, struct trace_probe, call.event);
1031
1032         if (!trace_seq_printf(s, "%s: (", tp->call.name))
1033                 goto partial;
1034
1035         if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1036                 goto partial;
1037
1038         if (!trace_seq_puts(s, " <- "))
1039                 goto partial;
1040
1041         if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1042                 goto partial;
1043
1044         if (!trace_seq_puts(s, ")"))
1045                 goto partial;
1046
1047         data = (u8 *)&field[1];
1048         for (i = 0; i < tp->nr_args; i++)
1049                 if (!tp->args[i].type->print(s, tp->args[i].name,
1050                                              data + tp->args[i].offset, field))
1051                         goto partial;
1052
1053         if (!trace_seq_puts(s, "\n"))
1054                 goto partial;
1055
1056         return TRACE_TYPE_HANDLED;
1057 partial:
1058         return TRACE_TYPE_PARTIAL_LINE;
1059 }
1060
1061
1062 static int kprobe_event_define_fields(struct ftrace_event_call *event_call)
1063 {
1064         int ret, i;
1065         struct kprobe_trace_entry_head field;
1066         struct trace_probe *tp = (struct trace_probe *)event_call->data;
1067
1068         DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1069         /* Set argument names as fields */
1070         for (i = 0; i < tp->nr_args; i++) {
1071                 ret = trace_define_field(event_call, tp->args[i].type->fmttype,
1072                                          tp->args[i].name,
1073                                          sizeof(field) + tp->args[i].offset,
1074                                          tp->args[i].type->size,
1075                                          tp->args[i].type->is_signed,
1076                                          FILTER_OTHER);
1077                 if (ret)
1078                         return ret;
1079         }
1080         return 0;
1081 }
1082
1083 static int kretprobe_event_define_fields(struct ftrace_event_call *event_call)
1084 {
1085         int ret, i;
1086         struct kretprobe_trace_entry_head field;
1087         struct trace_probe *tp = (struct trace_probe *)event_call->data;
1088
1089         DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1090         DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1091         /* Set argument names as fields */
1092         for (i = 0; i < tp->nr_args; i++) {
1093                 ret = trace_define_field(event_call, tp->args[i].type->fmttype,
1094                                          tp->args[i].name,
1095                                          sizeof(field) + tp->args[i].offset,
1096                                          tp->args[i].type->size,
1097                                          tp->args[i].type->is_signed,
1098                                          FILTER_OTHER);
1099                 if (ret)
1100                         return ret;
1101         }
1102         return 0;
1103 }
1104
1105 static int __set_print_fmt(struct trace_probe *tp, char *buf, int len)
1106 {
1107         int i;
1108         int pos = 0;
1109
1110         const char *fmt, *arg;
1111
1112         if (!trace_probe_is_return(tp)) {
1113                 fmt = "(%lx)";
1114                 arg = "REC->" FIELD_STRING_IP;
1115         } else {
1116                 fmt = "(%lx <- %lx)";
1117                 arg = "REC->" FIELD_STRING_FUNC ", REC->" FIELD_STRING_RETIP;
1118         }
1119
1120         /* When len=0, we just calculate the needed length */
1121 #define LEN_OR_ZERO (len ? len - pos : 0)
1122
1123         pos += snprintf(buf + pos, LEN_OR_ZERO, "\"%s", fmt);
1124
1125         for (i = 0; i < tp->nr_args; i++) {
1126                 pos += snprintf(buf + pos, LEN_OR_ZERO, " %s=%s",
1127                                 tp->args[i].name, tp->args[i].type->fmt);
1128         }
1129
1130         pos += snprintf(buf + pos, LEN_OR_ZERO, "\", %s", arg);
1131
1132         for (i = 0; i < tp->nr_args; i++) {
1133                 if (strcmp(tp->args[i].type->name, "string") == 0)
1134                         pos += snprintf(buf + pos, LEN_OR_ZERO,
1135                                         ", __get_str(%s)",
1136                                         tp->args[i].name);
1137                 else
1138                         pos += snprintf(buf + pos, LEN_OR_ZERO, ", REC->%s",
1139                                         tp->args[i].name);
1140         }
1141
1142 #undef LEN_OR_ZERO
1143
1144         /* return the length of print_fmt */
1145         return pos;
1146 }
1147
1148 static int set_print_fmt(struct trace_probe *tp)
1149 {
1150         int len;
1151         char *print_fmt;
1152
1153         /* First: called with 0 length to calculate the needed length */
1154         len = __set_print_fmt(tp, NULL, 0);
1155         print_fmt = kmalloc(len + 1, GFP_KERNEL);
1156         if (!print_fmt)
1157                 return -ENOMEM;
1158
1159         /* Second: actually write the @print_fmt */
1160         __set_print_fmt(tp, print_fmt, len + 1);
1161         tp->call.print_fmt = print_fmt;
1162
1163         return 0;
1164 }
1165
1166 #ifdef CONFIG_PERF_EVENTS
1167
1168 /* Kprobe profile handler */
1169 static __kprobes void
1170 kprobe_perf_func(struct trace_probe *tp, struct pt_regs *regs)
1171 {
1172         struct ftrace_event_call *call = &tp->call;
1173         struct kprobe_trace_entry_head *entry;
1174         struct hlist_head *head;
1175         int size, __size, dsize;
1176         int rctx;
1177
1178         dsize = __get_data_size(tp, regs);
1179         __size = sizeof(*entry) + tp->size + dsize;
1180         size = ALIGN(__size + sizeof(u32), sizeof(u64));
1181         size -= sizeof(u32);
1182         if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE,
1183                      "profile buffer not large enough"))
1184                 return;
1185
1186         entry = perf_trace_buf_prepare(size, call->event.type, regs, &rctx);
1187         if (!entry)
1188                 return;
1189
1190         entry->ip = (unsigned long)tp->rp.kp.addr;
1191         memset(&entry[1], 0, dsize);
1192         store_trace_args(sizeof(*entry), tp, regs, (u8 *)&entry[1], dsize);
1193
1194         head = this_cpu_ptr(call->perf_events);
1195         perf_trace_buf_submit(entry, size, rctx,
1196                                         entry->ip, 1, regs, head, NULL);
1197 }
1198
1199 /* Kretprobe profile handler */
1200 static __kprobes void
1201 kretprobe_perf_func(struct trace_probe *tp, struct kretprobe_instance *ri,
1202                     struct pt_regs *regs)
1203 {
1204         struct ftrace_event_call *call = &tp->call;
1205         struct kretprobe_trace_entry_head *entry;
1206         struct hlist_head *head;
1207         int size, __size, dsize;
1208         int rctx;
1209
1210         dsize = __get_data_size(tp, regs);
1211         __size = sizeof(*entry) + tp->size + dsize;
1212         size = ALIGN(__size + sizeof(u32), sizeof(u64));
1213         size -= sizeof(u32);
1214         if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE,
1215                      "profile buffer not large enough"))
1216                 return;
1217
1218         entry = perf_trace_buf_prepare(size, call->event.type, regs, &rctx);
1219         if (!entry)
1220                 return;
1221
1222         entry->func = (unsigned long)tp->rp.kp.addr;
1223         entry->ret_ip = (unsigned long)ri->ret_addr;
1224         store_trace_args(sizeof(*entry), tp, regs, (u8 *)&entry[1], dsize);
1225
1226         head = this_cpu_ptr(call->perf_events);
1227         perf_trace_buf_submit(entry, size, rctx,
1228                                         entry->ret_ip, 1, regs, head, NULL);
1229 }
1230 #endif  /* CONFIG_PERF_EVENTS */
1231
1232 static __kprobes
1233 int kprobe_register(struct ftrace_event_call *event,
1234                     enum trace_reg type, void *data)
1235 {
1236         struct trace_probe *tp = (struct trace_probe *)event->data;
1237         struct ftrace_event_file *file = data;
1238
1239         switch (type) {
1240         case TRACE_REG_REGISTER:
1241                 return enable_trace_probe(tp, file);
1242         case TRACE_REG_UNREGISTER:
1243                 return disable_trace_probe(tp, file);
1244
1245 #ifdef CONFIG_PERF_EVENTS
1246         case TRACE_REG_PERF_REGISTER:
1247                 return enable_trace_probe(tp, NULL);
1248         case TRACE_REG_PERF_UNREGISTER:
1249                 return disable_trace_probe(tp, NULL);
1250         case TRACE_REG_PERF_OPEN:
1251         case TRACE_REG_PERF_CLOSE:
1252         case TRACE_REG_PERF_ADD:
1253         case TRACE_REG_PERF_DEL:
1254                 return 0;
1255 #endif
1256         }
1257         return 0;
1258 }
1259
1260 static __kprobes
1261 int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1262 {
1263         struct trace_probe *tp = container_of(kp, struct trace_probe, rp.kp);
1264
1265         tp->nhit++;
1266
1267         if (tp->flags & TP_FLAG_TRACE)
1268                 kprobe_trace_func(tp, regs);
1269 #ifdef CONFIG_PERF_EVENTS
1270         if (tp->flags & TP_FLAG_PROFILE)
1271                 kprobe_perf_func(tp, regs);
1272 #endif
1273         return 0;       /* We don't tweek kernel, so just return 0 */
1274 }
1275
1276 static __kprobes
1277 int kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1278 {
1279         struct trace_probe *tp = container_of(ri->rp, struct trace_probe, rp);
1280
1281         tp->nhit++;
1282
1283         if (tp->flags & TP_FLAG_TRACE)
1284                 kretprobe_trace_func(tp, ri, regs);
1285 #ifdef CONFIG_PERF_EVENTS
1286         if (tp->flags & TP_FLAG_PROFILE)
1287                 kretprobe_perf_func(tp, ri, regs);
1288 #endif
1289         return 0;       /* We don't tweek kernel, so just return 0 */
1290 }
1291
1292 static struct trace_event_functions kretprobe_funcs = {
1293         .trace          = print_kretprobe_event
1294 };
1295
1296 static struct trace_event_functions kprobe_funcs = {
1297         .trace          = print_kprobe_event
1298 };
1299
1300 static int register_probe_event(struct trace_probe *tp)
1301 {
1302         struct ftrace_event_call *call = &tp->call;
1303         int ret;
1304
1305         /* Initialize ftrace_event_call */
1306         INIT_LIST_HEAD(&call->class->fields);
1307         if (trace_probe_is_return(tp)) {
1308                 call->event.funcs = &kretprobe_funcs;
1309                 call->class->define_fields = kretprobe_event_define_fields;
1310         } else {
1311                 call->event.funcs = &kprobe_funcs;
1312                 call->class->define_fields = kprobe_event_define_fields;
1313         }
1314         if (set_print_fmt(tp) < 0)
1315                 return -ENOMEM;
1316         ret = register_ftrace_event(&call->event);
1317         if (!ret) {
1318                 kfree(call->print_fmt);
1319                 return -ENODEV;
1320         }
1321         call->flags = 0;
1322         call->class->reg = kprobe_register;
1323         call->data = tp;
1324         ret = trace_add_event_call(call);
1325         if (ret) {
1326                 pr_info("Failed to register kprobe event: %s\n", call->name);
1327                 kfree(call->print_fmt);
1328                 unregister_ftrace_event(&call->event);
1329         }
1330         return ret;
1331 }
1332
1333 static int unregister_probe_event(struct trace_probe *tp)
1334 {
1335         int ret;
1336
1337         /* tp->event is unregistered in trace_remove_event_call() */
1338         ret = trace_remove_event_call(&tp->call);
1339         if (!ret)
1340                 kfree(tp->call.print_fmt);
1341         return ret;
1342 }
1343
1344 /* Make a debugfs interface for controlling probe points */
1345 static __init int init_kprobe_trace(void)
1346 {
1347         struct dentry *d_tracer;
1348         struct dentry *entry;
1349
1350         if (register_module_notifier(&trace_probe_module_nb))
1351                 return -EINVAL;
1352
1353         d_tracer = tracing_init_dentry();
1354         if (!d_tracer)
1355                 return 0;
1356
1357         entry = debugfs_create_file("kprobe_events", 0644, d_tracer,
1358                                     NULL, &kprobe_events_ops);
1359
1360         /* Event list interface */
1361         if (!entry)
1362                 pr_warning("Could not create debugfs "
1363                            "'kprobe_events' entry\n");
1364
1365         /* Profile interface */
1366         entry = debugfs_create_file("kprobe_profile", 0444, d_tracer,
1367                                     NULL, &kprobe_profile_ops);
1368
1369         if (!entry)
1370                 pr_warning("Could not create debugfs "
1371                            "'kprobe_profile' entry\n");
1372         return 0;
1373 }
1374 fs_initcall(init_kprobe_trace);
1375
1376
1377 #ifdef CONFIG_FTRACE_STARTUP_TEST
1378
1379 /*
1380  * The "__used" keeps gcc from removing the function symbol
1381  * from the kallsyms table.
1382  */
1383 static __used int kprobe_trace_selftest_target(int a1, int a2, int a3,
1384                                                int a4, int a5, int a6)
1385 {
1386         return a1 + a2 + a3 + a4 + a5 + a6;
1387 }
1388
1389 static struct ftrace_event_file *
1390 find_trace_probe_file(struct trace_probe *tp, struct trace_array *tr)
1391 {
1392         struct ftrace_event_file *file;
1393
1394         list_for_each_entry(file, &tr->events, list)
1395                 if (file->event_call == &tp->call)
1396                         return file;
1397
1398         return NULL;
1399 }
1400
1401 static __init int kprobe_trace_self_tests_init(void)
1402 {
1403         int ret, warn = 0;
1404         int (*target)(int, int, int, int, int, int);
1405         struct trace_probe *tp;
1406         struct ftrace_event_file *file;
1407
1408         target = kprobe_trace_selftest_target;
1409
1410         pr_info("Testing kprobe tracing: ");
1411
1412         ret = traceprobe_command("p:testprobe kprobe_trace_selftest_target "
1413                                   "$stack $stack0 +0($stack)",
1414                                   create_trace_probe);
1415         if (WARN_ON_ONCE(ret)) {
1416                 pr_warn("error on probing function entry.\n");
1417                 warn++;
1418         } else {
1419                 /* Enable trace point */
1420                 tp = find_trace_probe("testprobe", KPROBE_EVENT_SYSTEM);
1421                 if (WARN_ON_ONCE(tp == NULL)) {
1422                         pr_warn("error on getting new probe.\n");
1423                         warn++;
1424                 } else {
1425                         file = find_trace_probe_file(tp, top_trace_array());
1426                         if (WARN_ON_ONCE(file == NULL)) {
1427                                 pr_warn("error on getting probe file.\n");
1428                                 warn++;
1429                         } else
1430                                 enable_trace_probe(tp, file);
1431                 }
1432         }
1433
1434         ret = traceprobe_command("r:testprobe2 kprobe_trace_selftest_target "
1435                                   "$retval", create_trace_probe);
1436         if (WARN_ON_ONCE(ret)) {
1437                 pr_warn("error on probing function return.\n");
1438                 warn++;
1439         } else {
1440                 /* Enable trace point */
1441                 tp = find_trace_probe("testprobe2", KPROBE_EVENT_SYSTEM);
1442                 if (WARN_ON_ONCE(tp == NULL)) {
1443                         pr_warn("error on getting 2nd new probe.\n");
1444                         warn++;
1445                 } else {
1446                         file = find_trace_probe_file(tp, top_trace_array());
1447                         if (WARN_ON_ONCE(file == NULL)) {
1448                                 pr_warn("error on getting probe file.\n");
1449                                 warn++;
1450                         } else
1451                                 enable_trace_probe(tp, file);
1452                 }
1453         }
1454
1455         if (warn)
1456                 goto end;
1457
1458         ret = target(1, 2, 3, 4, 5, 6);
1459
1460         /* Disable trace points before removing it */
1461         tp = find_trace_probe("testprobe", KPROBE_EVENT_SYSTEM);
1462         if (WARN_ON_ONCE(tp == NULL)) {
1463                 pr_warn("error on getting test probe.\n");
1464                 warn++;
1465         } else {
1466                 file = find_trace_probe_file(tp, top_trace_array());
1467                 if (WARN_ON_ONCE(file == NULL)) {
1468                         pr_warn("error on getting probe file.\n");
1469                         warn++;
1470                 } else
1471                         disable_trace_probe(tp, file);
1472         }
1473
1474         tp = find_trace_probe("testprobe2", KPROBE_EVENT_SYSTEM);
1475         if (WARN_ON_ONCE(tp == NULL)) {
1476                 pr_warn("error on getting 2nd test probe.\n");
1477                 warn++;
1478         } else {
1479                 file = find_trace_probe_file(tp, top_trace_array());
1480                 if (WARN_ON_ONCE(file == NULL)) {
1481                         pr_warn("error on getting probe file.\n");
1482                         warn++;
1483                 } else
1484                         disable_trace_probe(tp, file);
1485         }
1486
1487         ret = traceprobe_command("-:testprobe", create_trace_probe);
1488         if (WARN_ON_ONCE(ret)) {
1489                 pr_warn("error on deleting a probe.\n");
1490                 warn++;
1491         }
1492
1493         ret = traceprobe_command("-:testprobe2", create_trace_probe);
1494         if (WARN_ON_ONCE(ret)) {
1495                 pr_warn("error on deleting a probe.\n");
1496                 warn++;
1497         }
1498
1499 end:
1500         release_all_trace_probes();
1501         if (warn)
1502                 pr_cont("NG: Some tests are failed. Please check them.\n");
1503         else
1504                 pr_cont("OK\n");
1505         return 0;
1506 }
1507
1508 late_initcall(kprobe_trace_self_tests_init);
1509
1510 #endif