SAFE public projects git trees. - safe/jmp/linux-2.6/blob - kernel/trace/trace_kprobe.c

   1 /*
   2  * kprobe based kernel tracer
   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 #include <linux/kprobes.h>
  23 #include <linux/seq_file.h>
  24 #include <linux/slab.h>
  25 #include <linux/smp.h>
  26 #include <linux/debugfs.h>
  27 #include <linux/types.h>
  28 #include <linux/string.h>
  29 #include <linux/ctype.h>
  30 #include <linux/ptrace.h>
  31 #include <linux/perf_event.h>
  32
  33 #include "trace.h"
  34 #include "trace_output.h"
  35
  36 #define MAX_TRACE_ARGS 128
  37 #define MAX_ARGSTR_LEN 63
  38 #define MAX_EVENT_NAME_LEN 64
  39 #define KPROBE_EVENT_SYSTEM "kprobes"
  40
  41 /* Reserved field names */
  42 #define FIELD_STRING_IP "__probe_ip"
  43 #define FIELD_STRING_NARGS "__probe_nargs"
  44 #define FIELD_STRING_RETIP "__probe_ret_ip"
  45 #define FIELD_STRING_FUNC "__probe_func"
  46
  47 const char *reserved_field_names[] = {
  48         "common_type",
  49         "common_flags",
  50         "common_preempt_count",
  51         "common_pid",
  52         "common_tgid",
  53         "common_lock_depth",
  54         FIELD_STRING_IP,
  55         FIELD_STRING_NARGS,
  56         FIELD_STRING_RETIP,
  57         FIELD_STRING_FUNC,
  58 };
  59
  60 /* currently, trace_kprobe only supports X86. */
  61
  62 struct fetch_func {
  63         unsigned long (*func)(struct pt_regs *, void *);
  64         void *data;
  65 };
  66
  67 static __kprobes unsigned long call_fetch(struct fetch_func *f,
  68                                           struct pt_regs *regs)
  69 {
  70         return f->func(regs, f->data);
  71 }
  72
  73 /* fetch handlers */
  74 static __kprobes unsigned long fetch_register(struct pt_regs *regs,
  75                                               void *offset)
  76 {
  77         return regs_get_register(regs, (unsigned int)((unsigned long)offset));
  78 }
  79
  80 static __kprobes unsigned long fetch_stack(struct pt_regs *regs,
  81                                            void *num)
  82 {
  83         return regs_get_kernel_stack_nth(regs,
  84                                          (unsigned int)((unsigned long)num));
  85 }
  86
  87 static __kprobes unsigned long fetch_memory(struct pt_regs *regs, void *addr)
  88 {
  89         unsigned long retval;
  90
  91         if (probe_kernel_address(addr, retval))
  92                 return 0;
  93         return retval;
  94 }
  95
  96 static __kprobes unsigned long fetch_argument(struct pt_regs *regs, void *num)
  97 {
  98         return regs_get_argument_nth(regs, (unsigned int)((unsigned long)num));
  99 }
 100
 101 static __kprobes unsigned long fetch_retvalue(struct pt_regs *regs,
 102                                               void *dummy)
 103 {
 104         return regs_return_value(regs);
 105 }
 106
 107 static __kprobes unsigned long fetch_stack_address(struct pt_regs *regs,
 108                                                    void *dummy)
 109 {
 110         return kernel_stack_pointer(regs);
 111 }
 112
 113 /* Memory fetching by symbol */
 114 struct symbol_cache {
 115         char *symbol;
 116         long offset;
 117         unsigned long addr;
 118 };
 119
 120 static unsigned long update_symbol_cache(struct symbol_cache *sc)
 121 {
 122         sc->addr = (unsigned long)kallsyms_lookup_name(sc->symbol);
 123         if (sc->addr)
 124                 sc->addr += sc->offset;
 125         return sc->addr;
 126 }
 127
 128 static void free_symbol_cache(struct symbol_cache *sc)
 129 {
 130         kfree(sc->symbol);
 131         kfree(sc);
 132 }
 133
 134 static struct symbol_cache *alloc_symbol_cache(const char *sym, long offset)
 135 {
 136         struct symbol_cache *sc;
 137
 138         if (!sym || strlen(sym) == 0)
 139                 return NULL;
 140         sc = kzalloc(sizeof(struct symbol_cache), GFP_KERNEL);
 141         if (!sc)
 142                 return NULL;
 143
 144         sc->symbol = kstrdup(sym, GFP_KERNEL);
 145         if (!sc->symbol) {
 146                 kfree(sc);
 147                 return NULL;
 148         }
 149         sc->offset = offset;
 150
 151         update_symbol_cache(sc);
 152         return sc;
 153 }
 154
 155 static __kprobes unsigned long fetch_symbol(struct pt_regs *regs, void *data)
 156 {
 157         struct symbol_cache *sc = data;
 158
 159         if (sc->addr)
 160                 return fetch_memory(regs, (void *)sc->addr);
 161         else
 162                 return 0;
 163 }
 164
 165 /* Special indirect memory access interface */
 166 struct indirect_fetch_data {
 167         struct fetch_func orig;
 168         long offset;
 169 };
 170
 171 static __kprobes unsigned long fetch_indirect(struct pt_regs *regs, void *data)
 172 {
 173         struct indirect_fetch_data *ind = data;
 174         unsigned long addr;
 175
 176         addr = call_fetch(&ind->orig, regs);
 177         if (addr) {
 178                 addr += ind->offset;
 179                 return fetch_memory(regs, (void *)addr);
 180         } else
 181                 return 0;
 182 }
 183
 184 static __kprobes void free_indirect_fetch_data(struct indirect_fetch_data *data)
 185 {
 186         if (data->orig.func == fetch_indirect)
 187                 free_indirect_fetch_data(data->orig.data);
 188         else if (data->orig.func == fetch_symbol)
 189                 free_symbol_cache(data->orig.data);
 190         kfree(data);
 191 }
 192
 193 /**
 194  * Kprobe tracer core functions
 195  */
 196
 197 struct probe_arg {
 198         struct fetch_func       fetch;
 199         const char              *name;
 200 };
 201
 202 /* Flags for trace_probe */
 203 #define TP_FLAG_TRACE   1
 204 #define TP_FLAG_PROFILE 2
 205
 206 struct trace_probe {
 207         struct list_head        list;
 208         struct kretprobe        rp;     /* Use rp.kp for kprobe use */
 209         unsigned long           nhit;
 210         unsigned int            flags;  /* For TP_FLAG_* */
 211         const char              *symbol;        /* symbol name */
 212         struct ftrace_event_call        call;
 213         struct trace_event              event;
 214         unsigned int            nr_args;
 215         struct probe_arg        args[];
 216 };
 217
 218 #define SIZEOF_TRACE_PROBE(n)                   \
 219         (offsetof(struct trace_probe, args) +   \
 220         (sizeof(struct probe_arg) * (n)))
 221
 222 static __kprobes int probe_is_return(struct trace_probe *tp)
 223 {
 224         return tp->rp.handler != NULL;
 225 }
 226
 227 static __kprobes const char *probe_symbol(struct trace_probe *tp)
 228 {
 229         return tp->symbol ? tp->symbol : "unknown";
 230 }
 231
 232 static int probe_arg_string(char *buf, size_t n, struct fetch_func *ff)
 233 {
 234         int ret = -EINVAL;
 235
 236         if (ff->func == fetch_argument)
 237                 ret = snprintf(buf, n, "$arg%lu", (unsigned long)ff->data);
 238         else if (ff->func == fetch_register) {
 239                 const char *name;
 240                 name = regs_query_register_name((unsigned int)((long)ff->data));
 241                 ret = snprintf(buf, n, "%%%s", name);
 242         } else if (ff->func == fetch_stack)
 243                 ret = snprintf(buf, n, "$stack%lu", (unsigned long)ff->data);
 244         else if (ff->func == fetch_memory)
 245                 ret = snprintf(buf, n, "@0x%p", ff->data);
 246         else if (ff->func == fetch_symbol) {
 247                 struct symbol_cache *sc = ff->data;
 248                 ret = snprintf(buf, n, "@%s%+ld", sc->symbol, sc->offset);
 249         } else if (ff->func == fetch_retvalue)
 250                 ret = snprintf(buf, n, "$retval");
 251         else if (ff->func == fetch_stack_address)
 252                 ret = snprintf(buf, n, "$stack");
 253         else if (ff->func == fetch_indirect) {
 254                 struct indirect_fetch_data *id = ff->data;
 255                 size_t l = 0;
 256                 ret = snprintf(buf, n, "%+ld(", id->offset);
 257                 if (ret >= n)
 258                         goto end;
 259                 l += ret;
 260                 ret = probe_arg_string(buf + l, n - l, &id->orig);
 261                 if (ret < 0)
 262                         goto end;
 263                 l += ret;
 264                 ret = snprintf(buf + l, n - l, ")");
 265                 ret += l;
 266         }
 267 end:
 268         if (ret >= n)
 269                 return -ENOSPC;
 270         return ret;
 271 }
 272
 273 static int register_probe_event(struct trace_probe *tp);
 274 static void unregister_probe_event(struct trace_probe *tp);
 275
 276 static DEFINE_MUTEX(probe_lock);
 277 static LIST_HEAD(probe_list);
 278
 279 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
 280 static int kretprobe_dispatcher(struct kretprobe_instance *ri,
 281                                 struct pt_regs *regs);
 282
 283 /*
 284  * Allocate new trace_probe and initialize it (including kprobes).
 285  */
 286 static struct trace_probe *alloc_trace_probe(const char *group,
 287                                              const char *event,
 288                                              void *addr,
 289                                              const char *symbol,
 290                                              unsigned long offs,
 291                                              int nargs, int is_return)
 292 {
 293         struct trace_probe *tp;
 294
 295         tp = kzalloc(SIZEOF_TRACE_PROBE(nargs), GFP_KERNEL);
 296         if (!tp)
 297                 return ERR_PTR(-ENOMEM);
 298
 299         if (symbol) {
 300                 tp->symbol = kstrdup(symbol, GFP_KERNEL);
 301                 if (!tp->symbol)
 302                         goto error;
 303                 tp->rp.kp.symbol_name = tp->symbol;
 304                 tp->rp.kp.offset = offs;
 305         } else
 306                 tp->rp.kp.addr = addr;
 307
 308         if (is_return)
 309                 tp->rp.handler = kretprobe_dispatcher;
 310         else
 311                 tp->rp.kp.pre_handler = kprobe_dispatcher;
 312
 313         if (!event)
 314                 goto error;
 315         tp->call.name = kstrdup(event, GFP_KERNEL);
 316         if (!tp->call.name)
 317                 goto error;
 318
 319         if (!group)
 320                 goto error;
 321         tp->call.system = kstrdup(group, GFP_KERNEL);
 322         if (!tp->call.system)
 323                 goto error;
 324
 325         INIT_LIST_HEAD(&tp->list);
 326         return tp;
 327 error:
 328         kfree(tp->call.name);
 329         kfree(tp->symbol);
 330         kfree(tp);
 331         return ERR_PTR(-ENOMEM);
 332 }
 333
 334 static void free_probe_arg(struct probe_arg *arg)
 335 {
 336         if (arg->fetch.func == fetch_symbol)
 337                 free_symbol_cache(arg->fetch.data);
 338         else if (arg->fetch.func == fetch_indirect)
 339                 free_indirect_fetch_data(arg->fetch.data);
 340         kfree(arg->name);
 341 }
 342
 343 static void free_trace_probe(struct trace_probe *tp)
 344 {
 345         int i;
 346
 347         for (i = 0; i < tp->nr_args; i++)
 348                 free_probe_arg(&tp->args[i]);
 349
 350         kfree(tp->call.system);
 351         kfree(tp->call.name);
 352         kfree(tp->symbol);
 353         kfree(tp);
 354 }
 355
 356 static struct trace_probe *find_probe_event(const char *event)
 357 {
 358         struct trace_probe *tp;
 359
 360         list_for_each_entry(tp, &probe_list, list)
 361                 if (!strcmp(tp->call.name, event))
 362                         return tp;
 363         return NULL;
 364 }
 365
 366 /* Unregister a trace_probe and probe_event: call with locking probe_lock */
 367 static void unregister_trace_probe(struct trace_probe *tp)
 368 {
 369         if (probe_is_return(tp))
 370                 unregister_kretprobe(&tp->rp);
 371         else
 372                 unregister_kprobe(&tp->rp.kp);
 373         list_del(&tp->list);
 374         unregister_probe_event(tp);
 375 }
 376
 377 /* Register a trace_probe and probe_event */
 378 static int register_trace_probe(struct trace_probe *tp)
 379 {
 380         struct trace_probe *old_tp;
 381         int ret;
 382
 383         mutex_lock(&probe_lock);
 384
 385         /* register as an event */
 386         old_tp = find_probe_event(tp->call.name);
 387         if (old_tp) {
 388                 /* delete old event */
 389                 unregister_trace_probe(old_tp);
 390                 free_trace_probe(old_tp);
 391         }
 392         ret = register_probe_event(tp);
 393         if (ret) {
 394                 pr_warning("Faild to register probe event(%d)\n", ret);
 395                 goto end;
 396         }
 397
 398         tp->rp.kp.flags |= KPROBE_FLAG_DISABLED;
 399         if (probe_is_return(tp))
 400                 ret = register_kretprobe(&tp->rp);
 401         else
 402                 ret = register_kprobe(&tp->rp.kp);
 403
 404         if (ret) {
 405                 pr_warning("Could not insert probe(%d)\n", ret);
 406                 if (ret == -EILSEQ) {
 407                         pr_warning("Probing address(0x%p) is not an "
 408                                    "instruction boundary.\n",
 409                                    tp->rp.kp.addr);
 410                         ret = -EINVAL;
 411                 }
 412                 unregister_probe_event(tp);
 413         } else
 414                 list_add_tail(&tp->list, &probe_list);
 415 end:
 416         mutex_unlock(&probe_lock);
 417         return ret;
 418 }
 419
 420 /* Split symbol and offset. */
 421 static int split_symbol_offset(char *symbol, unsigned long *offset)
 422 {
 423         char *tmp;
 424         int ret;
 425
 426         if (!offset)
 427                 return -EINVAL;
 428
 429         tmp = strchr(symbol, '+');
 430         if (tmp) {
 431                 /* skip sign because strict_strtol doesn't accept '+' */
 432                 ret = strict_strtoul(tmp + 1, 0, offset);
 433                 if (ret)
 434                         return ret;
 435                 *tmp = '\0';
 436         } else
 437                 *offset = 0;
 438         return 0;
 439 }
 440
 441 #define PARAM_MAX_ARGS 16
 442 #define PARAM_MAX_STACK (THREAD_SIZE / sizeof(unsigned long))
 443
 444 static int parse_probe_vars(char *arg, struct fetch_func *ff, int is_return)
 445 {
 446         int ret = 0;
 447         unsigned long param;
 448
 449         if (strcmp(arg, "retval") == 0) {
 450                 if (is_return) {
 451                         ff->func = fetch_retvalue;
 452                         ff->data = NULL;
 453                 } else
 454                         ret = -EINVAL;
 455         } else if (strncmp(arg, "stack", 5) == 0) {
 456                 if (arg[5] == '\0') {
 457                         ff->func = fetch_stack_address;
 458                         ff->data = NULL;
 459                 } else if (isdigit(arg[5])) {
 460                         ret = strict_strtoul(arg + 5, 10, &param);
 461                         if (ret || param > PARAM_MAX_STACK)
 462                                 ret = -EINVAL;
 463                         else {
 464                                 ff->func = fetch_stack;
 465                                 ff->data = (void *)param;
 466                         }
 467                 } else
 468                         ret = -EINVAL;
 469         } else if (strncmp(arg, "arg", 3) == 0 && isdigit(arg[3])) {
 470                 ret = strict_strtoul(arg + 3, 10, &param);
 471                 if (ret || param > PARAM_MAX_ARGS)
 472                         ret = -EINVAL;
 473                 else {
 474                         ff->func = fetch_argument;
 475                         ff->data = (void *)param;
 476                 }
 477         } else
 478                 ret = -EINVAL;
 479         return ret;
 480 }
 481
 482 static int parse_probe_arg(char *arg, struct fetch_func *ff, int is_return)
 483 {
 484         int ret = 0;
 485         unsigned long param;
 486         long offset;
 487         char *tmp;
 488
 489         switch (arg[0]) {
 490         case '$':
 491                 ret = parse_probe_vars(arg + 1, ff, is_return);
 492                 break;
 493         case '%':       /* named register */
 494                 ret = regs_query_register_offset(arg + 1);
 495                 if (ret >= 0) {
 496                         ff->func = fetch_register;
 497                         ff->data = (void *)(unsigned long)ret;
 498                         ret = 0;
 499                 }
 500                 break;
 501         case '@':       /* memory or symbol */
 502                 if (isdigit(arg[1])) {
 503                         ret = strict_strtoul(arg + 1, 0, &param);
 504                         if (ret)
 505                                 break;
 506                         ff->func = fetch_memory;
 507                         ff->data = (void *)param;
 508                 } else {
 509                         ret = split_symbol_offset(arg + 1, &offset);
 510                         if (ret)
 511                                 break;
 512                         ff->data = alloc_symbol_cache(arg + 1, offset);
 513                         if (ff->data)
 514                                 ff->func = fetch_symbol;
 515                         else
 516                                 ret = -EINVAL;
 517                 }
 518                 break;
 519         case '+':       /* indirect memory */
 520         case '-':
 521                 tmp = strchr(arg, '(');
 522                 if (!tmp) {
 523                         ret = -EINVAL;
 524                         break;
 525                 }
 526                 *tmp = '\0';
 527                 ret = strict_strtol(arg + 1, 0, &offset);
 528                 if (ret)
 529                         break;
 530                 if (arg[0] == '-')
 531                         offset = -offset;
 532                 arg = tmp + 1;
 533                 tmp = strrchr(arg, ')');
 534                 if (tmp) {
 535                         struct indirect_fetch_data *id;
 536                         *tmp = '\0';
 537                         id = kzalloc(sizeof(struct indirect_fetch_data),
 538                                      GFP_KERNEL);
 539                         if (!id)
 540                                 return -ENOMEM;
 541                         id->offset = offset;
 542                         ret = parse_probe_arg(arg, &id->orig, is_return);
 543                         if (ret)
 544                                 kfree(id);
 545                         else {
 546                                 ff->func = fetch_indirect;
 547                                 ff->data = (void *)id;
 548                         }
 549                 } else
 550                         ret = -EINVAL;
 551                 break;
 552         default:
 553                 /* TODO: support custom handler */
 554                 ret = -EINVAL;
 555         }
 556         return ret;
 557 }
 558
 559 /* Return 1 if name is reserved or already used by another argument */
 560 static int conflict_field_name(const char *name,
 561                                struct probe_arg *args, int narg)
 562 {
 563         int i;
 564         for (i = 0; i < ARRAY_SIZE(reserved_field_names); i++)
 565                 if (strcmp(reserved_field_names[i], name) == 0)
 566                         return 1;
 567         for (i = 0; i < narg; i++)
 568                 if (strcmp(args[i].name, name) == 0)
 569                         return 1;
 570         return 0;
 571 }
 572
 573 static int create_trace_probe(int argc, char **argv)
 574 {
 575         /*
 576          * Argument syntax:
 577          *  - Add kprobe: p[:[GRP/]EVENT] KSYM[+OFFS]|KADDR [FETCHARGS]
 578          *  - Add kretprobe: r[:[GRP/]EVENT] KSYM[+0] [FETCHARGS]
 579          * Fetch args:
 580          *  $argN       : fetch Nth of function argument. (N:0-)
 581          *  $retval     : fetch return value
 582          *  $stack      : fetch stack address
 583          *  $stackN     : fetch Nth of stack (N:0-)
 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          * Indirect 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          */
 592         struct trace_probe *tp;
 593         int i, ret = 0;
 594         int is_return = 0;
 595         char *symbol = NULL, *event = NULL, *arg = NULL, *group = NULL;
 596         unsigned long offset = 0;
 597         void *addr = NULL;
 598         char buf[MAX_EVENT_NAME_LEN];
 599
 600         if (argc < 2) {
 601                 pr_info("Probe point is not specified.\n");
 602                 return -EINVAL;
 603         }
 604
 605         if (argv[0][0] == 'p')
 606                 is_return = 0;
 607         else if (argv[0][0] == 'r')
 608                 is_return = 1;
 609         else {
 610                 pr_info("Probe definition must be started with 'p' or 'r'.\n");
 611                 return -EINVAL;
 612         }
 613
 614         if (argv[0][1] == ':') {
 615                 event = &argv[0][2];
 616                 if (strchr(event, '/')) {
 617                         group = event;
 618                         event = strchr(group, '/') + 1;
 619                         event[-1] = '\0';
 620                         if (strlen(group) == 0) {
 621                                 pr_info("Group name is not specifiled\n");
 622                                 return -EINVAL;
 623                         }
 624                 }
 625                 if (strlen(event) == 0) {
 626                         pr_info("Event name is not specifiled\n");
 627                         return -EINVAL;
 628                 }
 629         }
 630
 631         if (isdigit(argv[1][0])) {
 632                 if (is_return) {
 633                         pr_info("Return probe point must be a symbol.\n");
 634                         return -EINVAL;
 635                 }
 636                 /* an address specified */
 637                 ret = strict_strtoul(&argv[0][2], 0, (unsigned long *)&addr);
 638                 if (ret) {
 639                         pr_info("Failed to parse address.\n");
 640                         return ret;
 641                 }
 642         } else {
 643                 /* a symbol specified */
 644                 symbol = argv[1];
 645                 /* TODO: support .init module functions */
 646                 ret = split_symbol_offset(symbol, &offset);
 647                 if (ret) {
 648                         pr_info("Failed to parse symbol.\n");
 649                         return ret;
 650                 }
 651                 if (offset && is_return) {
 652                         pr_info("Return probe must be used without offset.\n");
 653                         return -EINVAL;
 654                 }
 655         }
 656         argc -= 2; argv += 2;
 657
 658         /* setup a probe */
 659         if (!group)
 660                 group = KPROBE_EVENT_SYSTEM;
 661         if (!event) {
 662                 /* Make a new event name */
 663                 if (symbol)
 664                         snprintf(buf, MAX_EVENT_NAME_LEN, "%c@%s%+ld",
 665                                  is_return ? 'r' : 'p', symbol, offset);
 666                 else
 667                         snprintf(buf, MAX_EVENT_NAME_LEN, "%c@0x%p",
 668                                  is_return ? 'r' : 'p', addr);
 669                 event = buf;
 670         }
 671         tp = alloc_trace_probe(group, event, addr, symbol, offset, argc,
 672                                is_return);
 673         if (IS_ERR(tp)) {
 674                 pr_info("Failed to allocate trace_probe.(%d)\n",
 675                         (int)PTR_ERR(tp));
 676                 return PTR_ERR(tp);
 677         }
 678
 679         /* parse arguments */
 680         ret = 0;
 681         for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
 682                 /* Parse argument name */
 683                 arg = strchr(argv[i], '=');
 684                 if (arg)
 685                         *arg++ = '\0';
 686                 else
 687                         arg = argv[i];
 688
 689                 if (conflict_field_name(argv[i], tp->args, i)) {
 690                         pr_info("Argument%d name '%s' conflicts with "
 691                                 "another field.\n", i, argv[i]);
 692                         ret = -EINVAL;
 693                         goto error;
 694                 }
 695
 696                 tp->args[i].name = kstrdup(argv[i], GFP_KERNEL);
 697
 698                 /* Parse fetch argument */
 699                 if (strlen(arg) > MAX_ARGSTR_LEN) {
 700                         pr_info("Argument%d(%s) is too long.\n", i, arg);
 701                         ret = -ENOSPC;
 702                         goto error;
 703                 }
 704                 ret = parse_probe_arg(arg, &tp->args[i].fetch, is_return);
 705                 if (ret) {
 706                         pr_info("Parse error at argument%d. (%d)\n", i, ret);
 707                         goto error;
 708                 }
 709         }
 710         tp->nr_args = i;
 711
 712         ret = register_trace_probe(tp);
 713         if (ret)
 714                 goto error;
 715         return 0;
 716
 717 error:
 718         free_trace_probe(tp);
 719         return ret;
 720 }
 721
 722 static void cleanup_all_probes(void)
 723 {
 724         struct trace_probe *tp;
 725
 726         mutex_lock(&probe_lock);
 727         /* TODO: Use batch unregistration */
 728         while (!list_empty(&probe_list)) {
 729                 tp = list_entry(probe_list.next, struct trace_probe, list);
 730                 unregister_trace_probe(tp);
 731                 free_trace_probe(tp);
 732         }
 733         mutex_unlock(&probe_lock);
 734 }
 735
 736
 737 /* Probes listing interfaces */
 738 static void *probes_seq_start(struct seq_file *m, loff_t *pos)
 739 {
 740         mutex_lock(&probe_lock);
 741         return seq_list_start(&probe_list, *pos);
 742 }
 743
 744 static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
 745 {
 746         return seq_list_next(v, &probe_list, pos);
 747 }
 748
 749 static void probes_seq_stop(struct seq_file *m, void *v)
 750 {
 751         mutex_unlock(&probe_lock);
 752 }
 753
 754 static int probes_seq_show(struct seq_file *m, void *v)
 755 {
 756         struct trace_probe *tp = v;
 757         int i, ret;
 758         char buf[MAX_ARGSTR_LEN + 1];
 759
 760         seq_printf(m, "%c", probe_is_return(tp) ? 'r' : 'p');
 761         seq_printf(m, ":%s", tp->call.name);
 762
 763         if (tp->symbol)
 764                 seq_printf(m, " %s+%u", probe_symbol(tp), tp->rp.kp.offset);
 765         else
 766                 seq_printf(m, " 0x%p", tp->rp.kp.addr);
 767
 768         for (i = 0; i < tp->nr_args; i++) {
 769                 ret = probe_arg_string(buf, MAX_ARGSTR_LEN, &tp->args[i].fetch);
 770                 if (ret < 0) {
 771                         pr_warning("Argument%d decoding error(%d).\n", i, ret);
 772                         return ret;
 773                 }
 774                 seq_printf(m, " %s=%s", tp->args[i].name, buf);
 775         }
 776         seq_printf(m, "\n");
 777         return 0;
 778 }
 779
 780 static const struct seq_operations probes_seq_op = {
 781         .start  = probes_seq_start,
 782         .next   = probes_seq_next,
 783         .stop   = probes_seq_stop,
 784         .show   = probes_seq_show
 785 };
 786
 787 static int probes_open(struct inode *inode, struct file *file)
 788 {
 789         if ((file->f_mode & FMODE_WRITE) &&
 790             (file->f_flags & O_TRUNC))
 791                 cleanup_all_probes();
 792
 793         return seq_open(file, &probes_seq_op);
 794 }
 795
 796 static int command_trace_probe(const char *buf)
 797 {
 798         char **argv;
 799         int argc = 0, ret = 0;
 800
 801         argv = argv_split(GFP_KERNEL, buf, &argc);
 802         if (!argv)
 803                 return -ENOMEM;
 804
 805         if (argc)
 806                 ret = create_trace_probe(argc, argv);
 807
 808         argv_free(argv);
 809         return ret;
 810 }
 811
 812 #define WRITE_BUFSIZE 128
 813
 814 static ssize_t probes_write(struct file *file, const char __user *buffer,
 815                             size_t count, loff_t *ppos)
 816 {
 817         char *kbuf, *tmp;
 818         int ret;
 819         size_t done;
 820         size_t size;
 821
 822         kbuf = kmalloc(WRITE_BUFSIZE, GFP_KERNEL);
 823         if (!kbuf)
 824                 return -ENOMEM;
 825
 826         ret = done = 0;
 827         while (done < count) {
 828                 size = count - done;
 829                 if (size >= WRITE_BUFSIZE)
 830                         size = WRITE_BUFSIZE - 1;
 831                 if (copy_from_user(kbuf, buffer + done, size)) {
 832                         ret = -EFAULT;
 833                         goto out;
 834                 }
 835                 kbuf[size] = '\0';
 836                 tmp = strchr(kbuf, '\n');
 837                 if (tmp) {
 838                         *tmp = '\0';
 839                         size = tmp - kbuf + 1;
 840                 } else if (done + size < count) {
 841                         pr_warning("Line length is too long: "
 842                                    "Should be less than %d.", WRITE_BUFSIZE);
 843                         ret = -EINVAL;
 844                         goto out;
 845                 }
 846                 done += size;
 847                 /* Remove comments */
 848                 tmp = strchr(kbuf, '#');
 849                 if (tmp)
 850                         *tmp = '\0';
 851
 852                 ret = command_trace_probe(kbuf);
 853                 if (ret)
 854                         goto out;
 855         }
 856         ret = done;
 857 out:
 858         kfree(kbuf);
 859         return ret;
 860 }
 861
 862 static const struct file_operations kprobe_events_ops = {
 863         .owner          = THIS_MODULE,
 864         .open           = probes_open,
 865         .read           = seq_read,
 866         .llseek         = seq_lseek,
 867         .release        = seq_release,
 868         .write          = probes_write,
 869 };
 870
 871 /* Probes profiling interfaces */
 872 static int probes_profile_seq_show(struct seq_file *m, void *v)
 873 {
 874         struct trace_probe *tp = v;
 875
 876         seq_printf(m, "  %-44s %15lu %15lu\n", tp->call.name, tp->nhit,
 877                    tp->rp.kp.nmissed);
 878
 879         return 0;
 880 }
 881
 882 static const struct seq_operations profile_seq_op = {
 883         .start  = probes_seq_start,
 884         .next   = probes_seq_next,
 885         .stop   = probes_seq_stop,
 886         .show   = probes_profile_seq_show
 887 };
 888
 889 static int profile_open(struct inode *inode, struct file *file)
 890 {
 891         return seq_open(file, &profile_seq_op);
 892 }
 893
 894 static const struct file_operations kprobe_profile_ops = {
 895         .owner          = THIS_MODULE,
 896         .open           = profile_open,
 897         .read           = seq_read,
 898         .llseek         = seq_lseek,
 899         .release        = seq_release,
 900 };
 901
 902 /* Kprobe handler */
 903 static __kprobes int kprobe_trace_func(struct kprobe *kp, struct pt_regs *regs)
 904 {
 905         struct trace_probe *tp = container_of(kp, struct trace_probe, rp.kp);
 906         struct kprobe_trace_entry *entry;
 907         struct ring_buffer_event *event;
 908         struct ring_buffer *buffer;
 909         int size, i, pc;
 910         unsigned long irq_flags;
 911         struct ftrace_event_call *call = &tp->call;
 912
 913         tp->nhit++;
 914
 915         local_save_flags(irq_flags);
 916         pc = preempt_count();
 917
 918         size = SIZEOF_KPROBE_TRACE_ENTRY(tp->nr_args);
 919
 920         event = trace_current_buffer_lock_reserve(&buffer, call->id, size,
 921                                                   irq_flags, pc);
 922         if (!event)
 923                 return 0;
 924
 925         entry = ring_buffer_event_data(event);
 926         entry->nargs = tp->nr_args;
 927         entry->ip = (unsigned long)kp->addr;
 928         for (i = 0; i < tp->nr_args; i++)
 929                 entry->args[i] = call_fetch(&tp->args[i].fetch, regs);
 930
 931         if (!filter_current_check_discard(buffer, call, entry, event))
 932                 trace_nowake_buffer_unlock_commit(buffer, event, irq_flags, pc);
 933         return 0;
 934 }
 935
 936 /* Kretprobe handler */
 937 static __kprobes int kretprobe_trace_func(struct kretprobe_instance *ri,
 938                                           struct pt_regs *regs)
 939 {
 940         struct trace_probe *tp = container_of(ri->rp, struct trace_probe, rp);
 941         struct kretprobe_trace_entry *entry;
 942         struct ring_buffer_event *event;
 943         struct ring_buffer *buffer;
 944         int size, i, pc;
 945         unsigned long irq_flags;
 946         struct ftrace_event_call *call = &tp->call;
 947
 948         local_save_flags(irq_flags);
 949         pc = preempt_count();
 950
 951         size = SIZEOF_KRETPROBE_TRACE_ENTRY(tp->nr_args);
 952
 953         event = trace_current_buffer_lock_reserve(&buffer, call->id, size,
 954                                                   irq_flags, pc);
 955         if (!event)
 956                 return 0;
 957
 958         entry = ring_buffer_event_data(event);
 959         entry->nargs = tp->nr_args;
 960         entry->func = (unsigned long)tp->rp.kp.addr;
 961         entry->ret_ip = (unsigned long)ri->ret_addr;
 962         for (i = 0; i < tp->nr_args; i++)
 963                 entry->args[i] = call_fetch(&tp->args[i].fetch, regs);
 964
 965         if (!filter_current_check_discard(buffer, call, entry, event))
 966                 trace_nowake_buffer_unlock_commit(buffer, event, irq_flags, pc);
 967
 968         return 0;
 969 }
 970
 971 /* Event entry printers */
 972 enum print_line_t
 973 print_kprobe_event(struct trace_iterator *iter, int flags)
 974 {
 975         struct kprobe_trace_entry *field;
 976         struct trace_seq *s = &iter->seq;
 977         struct trace_event *event;
 978         struct trace_probe *tp;
 979         int i;
 980
 981         field = (struct kprobe_trace_entry *)iter->ent;
 982         event = ftrace_find_event(field->ent.type);
 983         tp = container_of(event, struct trace_probe, event);
 984
 985         if (!trace_seq_printf(s, "%s: (", tp->call.name))
 986                 goto partial;
 987
 988         if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
 989                 goto partial;
 990
 991         if (!trace_seq_puts(s, ")"))
 992                 goto partial;
 993
 994         for (i = 0; i < field->nargs; i++)
 995                 if (!trace_seq_printf(s, " %s=%lx",
 996                                       tp->args[i].name, field->args[i]))
 997                         goto partial;
 998
 999         if (!trace_seq_puts(s, "\n"))
1000                 goto partial;
1001
1002         return TRACE_TYPE_HANDLED;
1003 partial:
1004         return TRACE_TYPE_PARTIAL_LINE;
1005 }
1006
1007 enum print_line_t
1008 print_kretprobe_event(struct trace_iterator *iter, int flags)
1009 {
1010         struct kretprobe_trace_entry *field;
1011         struct trace_seq *s = &iter->seq;
1012         struct trace_event *event;
1013         struct trace_probe *tp;
1014         int i;
1015
1016         field = (struct kretprobe_trace_entry *)iter->ent;
1017         event = ftrace_find_event(field->ent.type);
1018         tp = container_of(event, struct trace_probe, event);
1019
1020         if (!trace_seq_printf(s, "%s: (", tp->call.name))
1021                 goto partial;
1022
1023         if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1024                 goto partial;
1025
1026         if (!trace_seq_puts(s, " <- "))
1027                 goto partial;
1028
1029         if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1030                 goto partial;
1031
1032         if (!trace_seq_puts(s, ")"))
1033                 goto partial;
1034
1035         for (i = 0; i < field->nargs; i++)
1036                 if (!trace_seq_printf(s, " %s=%lx",
1037                                       tp->args[i].name, field->args[i]))
1038                         goto partial;
1039
1040         if (!trace_seq_puts(s, "\n"))
1041                 goto partial;
1042
1043         return TRACE_TYPE_HANDLED;
1044 partial:
1045         return TRACE_TYPE_PARTIAL_LINE;
1046 }
1047
1048 static int probe_event_enable(struct ftrace_event_call *call)
1049 {
1050         struct trace_probe *tp = (struct trace_probe *)call->data;
1051
1052         tp->flags |= TP_FLAG_TRACE;
1053         if (probe_is_return(tp))
1054                 return enable_kretprobe(&tp->rp);
1055         else
1056                 return enable_kprobe(&tp->rp.kp);
1057 }
1058
1059 static void probe_event_disable(struct ftrace_event_call *call)
1060 {
1061         struct trace_probe *tp = (struct trace_probe *)call->data;
1062
1063         tp->flags &= ~TP_FLAG_TRACE;
1064         if (!(tp->flags & (TP_FLAG_TRACE | TP_FLAG_PROFILE))) {
1065                 if (probe_is_return(tp))
1066                         disable_kretprobe(&tp->rp);
1067                 else
1068                         disable_kprobe(&tp->rp.kp);
1069         }
1070 }
1071
1072 static int probe_event_raw_init(struct ftrace_event_call *event_call)
1073 {
1074         INIT_LIST_HEAD(&event_call->fields);
1075
1076         return 0;
1077 }
1078
1079 #undef DEFINE_FIELD
1080 #define DEFINE_FIELD(type, item, name, is_signed)                       \
1081         do {                                                            \
1082                 ret = trace_define_field(event_call, #type, name,       \
1083                                          offsetof(typeof(field), item), \
1084                                          sizeof(field.item), is_signed, \
1085                                          FILTER_OTHER);                 \
1086                 if (ret)                                                \
1087                         return ret;                                     \
1088         } while (0)
1089
1090 static int kprobe_event_define_fields(struct ftrace_event_call *event_call)
1091 {
1092         int ret, i;
1093         struct kprobe_trace_entry field;
1094         struct trace_probe *tp = (struct trace_probe *)event_call->data;
1095
1096         ret = trace_define_common_fields(event_call);
1097         if (!ret)
1098                 return ret;
1099
1100         DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1101         DEFINE_FIELD(int, nargs, FIELD_STRING_NARGS, 1);
1102         /* Set argument names as fields */
1103         for (i = 0; i < tp->nr_args; i++)
1104                 DEFINE_FIELD(unsigned long, args[i], tp->args[i].name, 0);
1105         return 0;
1106 }
1107
1108 static int kretprobe_event_define_fields(struct ftrace_event_call *event_call)
1109 {
1110         int ret, i;
1111         struct kretprobe_trace_entry field;
1112         struct trace_probe *tp = (struct trace_probe *)event_call->data;
1113
1114         ret = trace_define_common_fields(event_call);
1115         if (!ret)
1116                 return ret;
1117
1118         DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1119         DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1120         DEFINE_FIELD(int, nargs, FIELD_STRING_NARGS, 1);
1121         /* Set argument names as fields */
1122         for (i = 0; i < tp->nr_args; i++)
1123                 DEFINE_FIELD(unsigned long, args[i], tp->args[i].name, 0);
1124         return 0;
1125 }
1126
1127 static int __probe_event_show_format(struct trace_seq *s,
1128                                      struct trace_probe *tp, const char *fmt,
1129                                      const char *arg)
1130 {
1131         int i;
1132
1133         /* Show format */
1134         if (!trace_seq_printf(s, "\nprint fmt: \"%s", fmt))
1135                 return 0;
1136
1137         for (i = 0; i < tp->nr_args; i++)
1138                 if (!trace_seq_printf(s, " %s=%%lx", tp->args[i].name))
1139                         return 0;
1140
1141         if (!trace_seq_printf(s, "\", %s", arg))
1142                 return 0;
1143
1144         for (i = 0; i < tp->nr_args; i++)
1145                 if (!trace_seq_printf(s, ", REC->%s", tp->args[i].name))
1146                         return 0;
1147
1148         return trace_seq_puts(s, "\n");
1149 }
1150
1151 #undef SHOW_FIELD
1152 #define SHOW_FIELD(type, item, name)                                    \
1153         do {                                                            \
1154                 ret = trace_seq_printf(s, "\tfield: " #type " %s;\t"    \
1155                                 "offset:%u;\tsize:%u;\n", name,         \
1156                                 (unsigned int)offsetof(typeof(field), item),\
1157                                 (unsigned int)sizeof(type));            \
1158                 if (!ret)                                               \
1159                         return 0;                                       \
1160         } while (0)
1161
1162 static int kprobe_event_show_format(struct ftrace_event_call *call,
1163                                     struct trace_seq *s)
1164 {
1165         struct kprobe_trace_entry field __attribute__((unused));
1166         int ret, i;
1167         struct trace_probe *tp = (struct trace_probe *)call->data;
1168
1169         SHOW_FIELD(unsigned long, ip, FIELD_STRING_IP);
1170         SHOW_FIELD(int, nargs, FIELD_STRING_NARGS);
1171
1172         /* Show fields */
1173         for (i = 0; i < tp->nr_args; i++)
1174                 SHOW_FIELD(unsigned long, args[i], tp->args[i].name);
1175         trace_seq_puts(s, "\n");
1176
1177         return __probe_event_show_format(s, tp, "(%lx)",
1178                                          "REC->" FIELD_STRING_IP);
1179 }
1180
1181 static int kretprobe_event_show_format(struct ftrace_event_call *call,
1182                                        struct trace_seq *s)
1183 {
1184         struct kretprobe_trace_entry field __attribute__((unused));
1185         int ret, i;
1186         struct trace_probe *tp = (struct trace_probe *)call->data;
1187
1188         SHOW_FIELD(unsigned long, func, FIELD_STRING_FUNC);
1189         SHOW_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP);
1190         SHOW_FIELD(int, nargs, FIELD_STRING_NARGS);
1191
1192         /* Show fields */
1193         for (i = 0; i < tp->nr_args; i++)
1194                 SHOW_FIELD(unsigned long, args[i], tp->args[i].name);
1195         trace_seq_puts(s, "\n");
1196
1197         return __probe_event_show_format(s, tp, "(%lx <- %lx)",
1198                                          "REC->" FIELD_STRING_FUNC
1199                                          ", REC->" FIELD_STRING_RETIP);
1200 }
1201
1202 #ifdef CONFIG_EVENT_PROFILE
1203
1204 /* Kprobe profile handler */
1205 static __kprobes int kprobe_profile_func(struct kprobe *kp,
1206                                          struct pt_regs *regs)
1207 {
1208         struct trace_probe *tp = container_of(kp, struct trace_probe, rp.kp);
1209         struct ftrace_event_call *call = &tp->call;
1210         struct kprobe_trace_entry *entry;
1211         struct trace_entry *ent;
1212         int size, __size, i, pc, __cpu;
1213         unsigned long irq_flags;
1214         char *raw_data;
1215
1216         pc = preempt_count();
1217         __size = SIZEOF_KPROBE_TRACE_ENTRY(tp->nr_args);
1218         size = ALIGN(__size + sizeof(u32), sizeof(u64));
1219         size -= sizeof(u32);
1220         if (WARN_ONCE(size > FTRACE_MAX_PROFILE_SIZE,
1221                      "profile buffer not large enough"))
1222                 return 0;
1223
1224         /*
1225          * Protect the non nmi buffer
1226          * This also protects the rcu read side
1227          */
1228         local_irq_save(irq_flags);
1229         __cpu = smp_processor_id();
1230
1231         if (in_nmi())
1232                 raw_data = rcu_dereference(trace_profile_buf_nmi);
1233         else
1234                 raw_data = rcu_dereference(trace_profile_buf);
1235
1236         if (!raw_data)
1237                 goto end;
1238
1239         raw_data = per_cpu_ptr(raw_data, __cpu);
1240         /* Zero dead bytes from alignment to avoid buffer leak to userspace */
1241         *(u64 *)(&raw_data[size - sizeof(u64)]) = 0ULL;
1242         entry = (struct kprobe_trace_entry *)raw_data;
1243         ent = &entry->ent;
1244
1245         tracing_generic_entry_update(ent, irq_flags, pc);
1246         ent->type = call->id;
1247         entry->nargs = tp->nr_args;
1248         entry->ip = (unsigned long)kp->addr;
1249         for (i = 0; i < tp->nr_args; i++)
1250                 entry->args[i] = call_fetch(&tp->args[i].fetch, regs);
1251         perf_tp_event(call->id, entry->ip, 1, entry, size);
1252 end:
1253         local_irq_restore(irq_flags);
1254         return 0;
1255 }
1256
1257 /* Kretprobe profile handler */
1258 static __kprobes int kretprobe_profile_func(struct kretprobe_instance *ri,
1259                                             struct pt_regs *regs)
1260 {
1261         struct trace_probe *tp = container_of(ri->rp, struct trace_probe, rp);
1262         struct ftrace_event_call *call = &tp->call;
1263         struct kretprobe_trace_entry *entry;
1264         struct trace_entry *ent;
1265         int size, __size, i, pc, __cpu;
1266         unsigned long irq_flags;
1267         char *raw_data;
1268
1269         pc = preempt_count();
1270         __size = SIZEOF_KRETPROBE_TRACE_ENTRY(tp->nr_args);
1271         size = ALIGN(__size + sizeof(u32), sizeof(u64));
1272         size -= sizeof(u32);
1273         if (WARN_ONCE(size > FTRACE_MAX_PROFILE_SIZE,
1274                      "profile buffer not large enough"))
1275                 return 0;
1276
1277         /*
1278          * Protect the non nmi buffer
1279          * This also protects the rcu read side
1280          */
1281         local_irq_save(irq_flags);
1282         __cpu = smp_processor_id();
1283
1284         if (in_nmi())
1285                 raw_data = rcu_dereference(trace_profile_buf_nmi);
1286         else
1287                 raw_data = rcu_dereference(trace_profile_buf);
1288
1289         if (!raw_data)
1290                 goto end;
1291
1292         raw_data = per_cpu_ptr(raw_data, __cpu);
1293         /* Zero dead bytes from alignment to avoid buffer leak to userspace */
1294         *(u64 *)(&raw_data[size - sizeof(u64)]) = 0ULL;
1295         entry = (struct kretprobe_trace_entry *)raw_data;
1296         ent = &entry->ent;
1297
1298         tracing_generic_entry_update(ent, irq_flags, pc);
1299         ent->type = call->id;
1300         entry->nargs = tp->nr_args;
1301         entry->func = (unsigned long)tp->rp.kp.addr;
1302         entry->ret_ip = (unsigned long)ri->ret_addr;
1303         for (i = 0; i < tp->nr_args; i++)
1304                 entry->args[i] = call_fetch(&tp->args[i].fetch, regs);
1305         perf_tp_event(call->id, entry->ret_ip, 1, entry, size);
1306 end:
1307         local_irq_restore(irq_flags);
1308         return 0;
1309 }
1310
1311 static int probe_profile_enable(struct ftrace_event_call *call)
1312 {
1313         struct trace_probe *tp = (struct trace_probe *)call->data;
1314
1315         tp->flags |= TP_FLAG_PROFILE;
1316
1317         if (probe_is_return(tp))
1318                 return enable_kretprobe(&tp->rp);
1319         else
1320                 return enable_kprobe(&tp->rp.kp);
1321 }
1322
1323 static void probe_profile_disable(struct ftrace_event_call *call)
1324 {
1325         struct trace_probe *tp = (struct trace_probe *)call->data;
1326
1327         tp->flags &= ~TP_FLAG_PROFILE;
1328
1329         if (!(tp->flags & TP_FLAG_TRACE)) {
1330                 if (probe_is_return(tp))
1331                         disable_kretprobe(&tp->rp);
1332                 else
1333                         disable_kprobe(&tp->rp.kp);
1334         }
1335 }
1336 #endif  /* CONFIG_EVENT_PROFILE */
1337
1338
1339 static __kprobes
1340 int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1341 {
1342         struct trace_probe *tp = container_of(kp, struct trace_probe, rp.kp);
1343
1344         if (tp->flags & TP_FLAG_TRACE)
1345                 kprobe_trace_func(kp, regs);
1346 #ifdef CONFIG_EVENT_PROFILE
1347         if (tp->flags & TP_FLAG_PROFILE)
1348                 kprobe_profile_func(kp, regs);
1349 #endif  /* CONFIG_EVENT_PROFILE */
1350         return 0;       /* We don't tweek kernel, so just return 0 */
1351 }
1352
1353 static __kprobes
1354 int kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1355 {
1356         struct trace_probe *tp = container_of(ri->rp, struct trace_probe, rp);
1357
1358         if (tp->flags & TP_FLAG_TRACE)
1359                 kretprobe_trace_func(ri, regs);
1360 #ifdef CONFIG_EVENT_PROFILE
1361         if (tp->flags & TP_FLAG_PROFILE)
1362                 kretprobe_profile_func(ri, regs);
1363 #endif  /* CONFIG_EVENT_PROFILE */
1364         return 0;       /* We don't tweek kernel, so just return 0 */
1365 }
1366
1367 static int register_probe_event(struct trace_probe *tp)
1368 {
1369         struct ftrace_event_call *call = &tp->call;
1370         int ret;
1371
1372         /* Initialize ftrace_event_call */
1373         if (probe_is_return(tp)) {
1374                 tp->event.trace = print_kretprobe_event;
1375                 call->raw_init = probe_event_raw_init;
1376                 call->show_format = kretprobe_event_show_format;
1377                 call->define_fields = kretprobe_event_define_fields;
1378         } else {
1379                 tp->event.trace = print_kprobe_event;
1380                 call->raw_init = probe_event_raw_init;
1381                 call->show_format = kprobe_event_show_format;
1382                 call->define_fields = kprobe_event_define_fields;
1383         }
1384         call->event = &tp->event;
1385         call->id = register_ftrace_event(&tp->event);
1386         if (!call->id)
1387                 return -ENODEV;
1388         call->enabled = 0;
1389         call->regfunc = probe_event_enable;
1390         call->unregfunc = probe_event_disable;
1391
1392 #ifdef CONFIG_EVENT_PROFILE
1393         atomic_set(&call->profile_count, -1);
1394         call->profile_enable = probe_profile_enable;
1395         call->profile_disable = probe_profile_disable;
1396 #endif
1397         call->data = tp;
1398         ret = trace_add_event_call(call);
1399         if (ret) {
1400                 pr_info("Failed to register kprobe event: %s\n", call->name);
1401                 unregister_ftrace_event(&tp->event);
1402         }
1403         return ret;
1404 }
1405
1406 static void unregister_probe_event(struct trace_probe *tp)
1407 {
1408         /* tp->event is unregistered in trace_remove_event_call() */
1409         trace_remove_event_call(&tp->call);
1410 }
1411
1412 /* Make a debugfs interface for controling probe points */
1413 static __init int init_kprobe_trace(void)
1414 {
1415         struct dentry *d_tracer;
1416         struct dentry *entry;
1417
1418         d_tracer = tracing_init_dentry();
1419         if (!d_tracer)
1420                 return 0;
1421
1422         entry = debugfs_create_file("kprobe_events", 0644, d_tracer,
1423                                     NULL, &kprobe_events_ops);
1424
1425         /* Event list interface */
1426         if (!entry)
1427                 pr_warning("Could not create debugfs "
1428                            "'kprobe_events' entry\n");
1429
1430         /* Profile interface */
1431         entry = debugfs_create_file("kprobe_profile", 0444, d_tracer,
1432                                     NULL, &kprobe_profile_ops);
1433
1434         if (!entry)
1435                 pr_warning("Could not create debugfs "
1436                            "'kprobe_profile' entry\n");
1437         return 0;
1438 }
1439 fs_initcall(init_kprobe_trace);
1440
1441
1442 #ifdef CONFIG_FTRACE_STARTUP_TEST
1443
1444 static int kprobe_trace_selftest_target(int a1, int a2, int a3,
1445                                         int a4, int a5, int a6)
1446 {
1447         return a1 + a2 + a3 + a4 + a5 + a6;
1448 }
1449
1450 static __init int kprobe_trace_self_tests_init(void)
1451 {
1452         int ret;
1453         int (*target)(int, int, int, int, int, int);
1454
1455         target = kprobe_trace_selftest_target;
1456
1457         pr_info("Testing kprobe tracing: ");
1458
1459         ret = command_trace_probe("p:testprobe kprobe_trace_selftest_target "
1460                                   "$arg1 $arg2 $arg3 $arg4 $stack $stack0");
1461         if (WARN_ON_ONCE(ret))
1462                 pr_warning("error enabling function entry\n");
1463
1464         ret = command_trace_probe("r:testprobe2 kprobe_trace_selftest_target "
1465                                   "$retval");
1466         if (WARN_ON_ONCE(ret))
1467                 pr_warning("error enabling function return\n");
1468
1469         ret = target(1, 2, 3, 4, 5, 6);
1470
1471         cleanup_all_probes();
1472
1473         pr_cont("OK\n");
1474         return 0;
1475 }
1476
1477 late_initcall(kprobe_trace_self_tests_init);
1478
1479 #endif