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 #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 struct fetch_func {
  61         unsigned long (*func)(struct pt_regs *, void *);
  62         void *data;
  63 };
  64
  65 static __kprobes unsigned long call_fetch(struct fetch_func *f,
  66                                           struct pt_regs *regs)
  67 {
  68         return f->func(regs, f->data);
  69 }
  70
  71 /* fetch handlers */
  72 static __kprobes unsigned long fetch_register(struct pt_regs *regs,
  73                                               void *offset)
  74 {
  75         return regs_get_register(regs, (unsigned int)((unsigned long)offset));
  76 }
  77
  78 static __kprobes unsigned long fetch_stack(struct pt_regs *regs,
  79                                            void *num)
  80 {
  81         return regs_get_kernel_stack_nth(regs,
  82                                          (unsigned int)((unsigned long)num));
  83 }
  84
  85 static __kprobes unsigned long fetch_memory(struct pt_regs *regs, void *addr)
  86 {
  87         unsigned long retval;
  88
  89         if (probe_kernel_address(addr, retval))
  90                 return 0;
  91         return retval;
  92 }
  93
  94 static __kprobes unsigned long fetch_argument(struct pt_regs *regs, void *num)
  95 {
  96         return regs_get_argument_nth(regs, (unsigned int)((unsigned long)num));
  97 }
  98
  99 static __kprobes unsigned long fetch_retvalue(struct pt_regs *regs,
 100                                               void *dummy)
 101 {
 102         return regs_return_value(regs);
 103 }
 104
 105 static __kprobes unsigned long fetch_stack_address(struct pt_regs *regs,
 106                                                    void *dummy)
 107 {
 108         return kernel_stack_pointer(regs);
 109 }
 110
 111 /* Memory fetching by symbol */
 112 struct symbol_cache {
 113         char *symbol;
 114         long offset;
 115         unsigned long addr;
 116 };
 117
 118 static unsigned long update_symbol_cache(struct symbol_cache *sc)
 119 {
 120         sc->addr = (unsigned long)kallsyms_lookup_name(sc->symbol);
 121         if (sc->addr)
 122                 sc->addr += sc->offset;
 123         return sc->addr;
 124 }
 125
 126 static void free_symbol_cache(struct symbol_cache *sc)
 127 {
 128         kfree(sc->symbol);
 129         kfree(sc);
 130 }
 131
 132 static struct symbol_cache *alloc_symbol_cache(const char *sym, long offset)
 133 {
 134         struct symbol_cache *sc;
 135
 136         if (!sym || strlen(sym) == 0)
 137                 return NULL;
 138         sc = kzalloc(sizeof(struct symbol_cache), GFP_KERNEL);
 139         if (!sc)
 140                 return NULL;
 141
 142         sc->symbol = kstrdup(sym, GFP_KERNEL);
 143         if (!sc->symbol) {
 144                 kfree(sc);
 145                 return NULL;
 146         }
 147         sc->offset = offset;
 148
 149         update_symbol_cache(sc);
 150         return sc;
 151 }
 152
 153 static __kprobes unsigned long fetch_symbol(struct pt_regs *regs, void *data)
 154 {
 155         struct symbol_cache *sc = data;
 156
 157         if (sc->addr)
 158                 return fetch_memory(regs, (void *)sc->addr);
 159         else
 160                 return 0;
 161 }
 162
 163 /* Special indirect memory access interface */
 164 struct indirect_fetch_data {
 165         struct fetch_func orig;
 166         long offset;
 167 };
 168
 169 static __kprobes unsigned long fetch_indirect(struct pt_regs *regs, void *data)
 170 {
 171         struct indirect_fetch_data *ind = data;
 172         unsigned long addr;
 173
 174         addr = call_fetch(&ind->orig, regs);
 175         if (addr) {
 176                 addr += ind->offset;
 177                 return fetch_memory(regs, (void *)addr);
 178         } else
 179                 return 0;
 180 }
 181
 182 static __kprobes void free_indirect_fetch_data(struct indirect_fetch_data *data)
 183 {
 184         if (data->orig.func == fetch_indirect)
 185                 free_indirect_fetch_data(data->orig.data);
 186         else if (data->orig.func == fetch_symbol)
 187                 free_symbol_cache(data->orig.data);
 188         kfree(data);
 189 }
 190
 191 /**
 192  * Kprobe event core functions
 193  */
 194
 195 struct probe_arg {
 196         struct fetch_func       fetch;
 197         const char              *name;
 198 };
 199
 200 /* Flags for trace_probe */
 201 #define TP_FLAG_TRACE   1
 202 #define TP_FLAG_PROFILE 2
 203
 204 struct trace_probe {
 205         struct list_head        list;
 206         struct kretprobe        rp;     /* Use rp.kp for kprobe use */
 207         unsigned long           nhit;
 208         unsigned int            flags;  /* For TP_FLAG_* */
 209         const char              *symbol;        /* symbol name */
 210         struct ftrace_event_call        call;
 211         struct trace_event              event;
 212         unsigned int            nr_args;
 213         struct probe_arg        args[];
 214 };
 215
 216 #define SIZEOF_TRACE_PROBE(n)                   \
 217         (offsetof(struct trace_probe, args) +   \
 218         (sizeof(struct probe_arg) * (n)))
 219
 220 static __kprobes int probe_is_return(struct trace_probe *tp)
 221 {
 222         return tp->rp.handler != NULL;
 223 }
 224
 225 static __kprobes const char *probe_symbol(struct trace_probe *tp)
 226 {
 227         return tp->symbol ? tp->symbol : "unknown";
 228 }
 229
 230 static int probe_arg_string(char *buf, size_t n, struct fetch_func *ff)
 231 {
 232         int ret = -EINVAL;
 233
 234         if (ff->func == fetch_argument)
 235                 ret = snprintf(buf, n, "$arg%lu", (unsigned long)ff->data);
 236         else if (ff->func == fetch_register) {
 237                 const char *name;
 238                 name = regs_query_register_name((unsigned int)((long)ff->data));
 239                 ret = snprintf(buf, n, "%%%s", name);
 240         } else if (ff->func == fetch_stack)
 241                 ret = snprintf(buf, n, "$stack%lu", (unsigned long)ff->data);
 242         else if (ff->func == fetch_memory)
 243                 ret = snprintf(buf, n, "@0x%p", ff->data);
 244         else if (ff->func == fetch_symbol) {
 245                 struct symbol_cache *sc = ff->data;
 246                 if (sc->offset)
 247                         ret = snprintf(buf, n, "@%s%+ld", sc->symbol,
 248                                         sc->offset);
 249                 else
 250                         ret = snprintf(buf, n, "@%s", sc->symbol);
 251         } else if (ff->func == fetch_retvalue)
 252                 ret = snprintf(buf, n, "$retval");
 253         else if (ff->func == fetch_stack_address)
 254                 ret = snprintf(buf, n, "$stack");
 255         else if (ff->func == fetch_indirect) {
 256                 struct indirect_fetch_data *id = ff->data;
 257                 size_t l = 0;
 258                 ret = snprintf(buf, n, "%+ld(", id->offset);
 259                 if (ret >= n)
 260                         goto end;
 261                 l += ret;
 262                 ret = probe_arg_string(buf + l, n - l, &id->orig);
 263                 if (ret < 0)
 264                         goto end;
 265                 l += ret;
 266                 ret = snprintf(buf + l, n - l, ")");
 267                 ret += l;
 268         }
 269 end:
 270         if (ret >= n)
 271                 return -ENOSPC;
 272         return ret;
 273 }
 274
 275 static int register_probe_event(struct trace_probe *tp);
 276 static void unregister_probe_event(struct trace_probe *tp);
 277
 278 static DEFINE_MUTEX(probe_lock);
 279 static LIST_HEAD(probe_list);
 280
 281 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
 282 static int kretprobe_dispatcher(struct kretprobe_instance *ri,
 283                                 struct pt_regs *regs);
 284
 285 /*
 286  * Allocate new trace_probe and initialize it (including kprobes).
 287  */
 288 static struct trace_probe *alloc_trace_probe(const char *group,
 289                                              const char *event,
 290                                              void *addr,
 291                                              const char *symbol,
 292                                              unsigned long offs,
 293                                              int nargs, int is_return)
 294 {
 295         struct trace_probe *tp;
 296
 297         tp = kzalloc(SIZEOF_TRACE_PROBE(nargs), GFP_KERNEL);
 298         if (!tp)
 299                 return ERR_PTR(-ENOMEM);
 300
 301         if (symbol) {
 302                 tp->symbol = kstrdup(symbol, GFP_KERNEL);
 303                 if (!tp->symbol)
 304                         goto error;
 305                 tp->rp.kp.symbol_name = tp->symbol;
 306                 tp->rp.kp.offset = offs;
 307         } else
 308                 tp->rp.kp.addr = addr;
 309
 310         if (is_return)
 311                 tp->rp.handler = kretprobe_dispatcher;
 312         else
 313                 tp->rp.kp.pre_handler = kprobe_dispatcher;
 314
 315         if (!event)
 316                 goto error;
 317         tp->call.name = kstrdup(event, GFP_KERNEL);
 318         if (!tp->call.name)
 319                 goto error;
 320
 321         if (!group)
 322                 goto error;
 323         tp->call.system = kstrdup(group, GFP_KERNEL);
 324         if (!tp->call.system)
 325                 goto error;
 326
 327         INIT_LIST_HEAD(&tp->list);
 328         return tp;
 329 error:
 330         kfree(tp->call.name);
 331         kfree(tp->symbol);
 332         kfree(tp);
 333         return ERR_PTR(-ENOMEM);
 334 }
 335
 336 static void free_probe_arg(struct probe_arg *arg)
 337 {
 338         if (arg->fetch.func == fetch_symbol)
 339                 free_symbol_cache(arg->fetch.data);
 340         else if (arg->fetch.func == fetch_indirect)
 341                 free_indirect_fetch_data(arg->fetch.data);
 342         kfree(arg->name);
 343 }
 344
 345 static void free_trace_probe(struct trace_probe *tp)
 346 {
 347         int i;
 348
 349         for (i = 0; i < tp->nr_args; i++)
 350                 free_probe_arg(&tp->args[i]);
 351
 352         kfree(tp->call.system);
 353         kfree(tp->call.name);
 354         kfree(tp->symbol);
 355         kfree(tp);
 356 }
 357
 358 static struct trace_probe *find_probe_event(const char *event,
 359                                             const char *group)
 360 {
 361         struct trace_probe *tp;
 362
 363         list_for_each_entry(tp, &probe_list, list)
 364                 if (strcmp(tp->call.name, event) == 0 &&
 365                     strcmp(tp->call.system, group) == 0)
 366                         return tp;
 367         return NULL;
 368 }
 369
 370 /* Unregister a trace_probe and probe_event: call with locking probe_lock */
 371 static void unregister_trace_probe(struct trace_probe *tp)
 372 {
 373         if (probe_is_return(tp))
 374                 unregister_kretprobe(&tp->rp);
 375         else
 376                 unregister_kprobe(&tp->rp.kp);
 377         list_del(&tp->list);
 378         unregister_probe_event(tp);
 379 }
 380
 381 /* Register a trace_probe and probe_event */
 382 static int register_trace_probe(struct trace_probe *tp)
 383 {
 384         struct trace_probe *old_tp;
 385         int ret;
 386
 387         mutex_lock(&probe_lock);
 388
 389         /* register as an event */
 390         old_tp = find_probe_event(tp->call.name, tp->call.system);
 391         if (old_tp) {
 392                 /* delete old event */
 393                 unregister_trace_probe(old_tp);
 394                 free_trace_probe(old_tp);
 395         }
 396         ret = register_probe_event(tp);
 397         if (ret) {
 398                 pr_warning("Faild to register probe event(%d)\n", ret);
 399                 goto end;
 400         }
 401
 402         tp->rp.kp.flags |= KPROBE_FLAG_DISABLED;
 403         if (probe_is_return(tp))
 404                 ret = register_kretprobe(&tp->rp);
 405         else
 406                 ret = register_kprobe(&tp->rp.kp);
 407
 408         if (ret) {
 409                 pr_warning("Could not insert probe(%d)\n", ret);
 410                 if (ret == -EILSEQ) {
 411                         pr_warning("Probing address(0x%p) is not an "
 412                                    "instruction boundary.\n",
 413                                    tp->rp.kp.addr);
 414                         ret = -EINVAL;
 415                 }
 416                 unregister_probe_event(tp);
 417         } else
 418                 list_add_tail(&tp->list, &probe_list);
 419 end:
 420         mutex_unlock(&probe_lock);
 421         return ret;
 422 }
 423
 424 /* Split symbol and offset. */
 425 static int split_symbol_offset(char *symbol, unsigned long *offset)
 426 {
 427         char *tmp;
 428         int ret;
 429
 430         if (!offset)
 431                 return -EINVAL;
 432
 433         tmp = strchr(symbol, '+');
 434         if (tmp) {
 435                 /* skip sign because strict_strtol doesn't accept '+' */
 436                 ret = strict_strtoul(tmp + 1, 0, offset);
 437                 if (ret)
 438                         return ret;
 439                 *tmp = '\0';
 440         } else
 441                 *offset = 0;
 442         return 0;
 443 }
 444
 445 #define PARAM_MAX_ARGS 16
 446 #define PARAM_MAX_STACK (THREAD_SIZE / sizeof(unsigned long))
 447
 448 static int parse_probe_vars(char *arg, struct fetch_func *ff, int is_return)
 449 {
 450         int ret = 0;
 451         unsigned long param;
 452
 453         if (strcmp(arg, "retval") == 0) {
 454                 if (is_return) {
 455                         ff->func = fetch_retvalue;
 456                         ff->data = NULL;
 457                 } else
 458                         ret = -EINVAL;
 459         } else if (strncmp(arg, "stack", 5) == 0) {
 460                 if (arg[5] == '\0') {
 461                         ff->func = fetch_stack_address;
 462                         ff->data = NULL;
 463                 } else if (isdigit(arg[5])) {
 464                         ret = strict_strtoul(arg + 5, 10, &param);
 465                         if (ret || param > PARAM_MAX_STACK)
 466                                 ret = -EINVAL;
 467                         else {
 468                                 ff->func = fetch_stack;
 469                                 ff->data = (void *)param;
 470                         }
 471                 } else
 472                         ret = -EINVAL;
 473         } else if (strncmp(arg, "arg", 3) == 0 && isdigit(arg[3])) {
 474                 ret = strict_strtoul(arg + 3, 10, &param);
 475                 if (ret || param > PARAM_MAX_ARGS)
 476                         ret = -EINVAL;
 477                 else {
 478                         ff->func = fetch_argument;
 479                         ff->data = (void *)param;
 480                 }
 481         } else
 482                 ret = -EINVAL;
 483         return ret;
 484 }
 485
 486 /* Recursive argument parser */
 487 static int __parse_probe_arg(char *arg, struct fetch_func *ff, int is_return)
 488 {
 489         int ret = 0;
 490         unsigned long param;
 491         long offset;
 492         char *tmp;
 493
 494         switch (arg[0]) {
 495         case '$':
 496                 ret = parse_probe_vars(arg + 1, ff, is_return);
 497                 break;
 498         case '%':       /* named register */
 499                 ret = regs_query_register_offset(arg + 1);
 500                 if (ret >= 0) {
 501                         ff->func = fetch_register;
 502                         ff->data = (void *)(unsigned long)ret;
 503                         ret = 0;
 504                 }
 505                 break;
 506         case '@':       /* memory or symbol */
 507                 if (isdigit(arg[1])) {
 508                         ret = strict_strtoul(arg + 1, 0, &param);
 509                         if (ret)
 510                                 break;
 511                         ff->func = fetch_memory;
 512                         ff->data = (void *)param;
 513                 } else {
 514                         ret = split_symbol_offset(arg + 1, &offset);
 515                         if (ret)
 516                                 break;
 517                         ff->data = alloc_symbol_cache(arg + 1, offset);
 518                         if (ff->data)
 519                                 ff->func = fetch_symbol;
 520                         else
 521                                 ret = -EINVAL;
 522                 }
 523                 break;
 524         case '+':       /* indirect memory */
 525         case '-':
 526                 tmp = strchr(arg, '(');
 527                 if (!tmp) {
 528                         ret = -EINVAL;
 529                         break;
 530                 }
 531                 *tmp = '\0';
 532                 ret = strict_strtol(arg + 1, 0, &offset);
 533                 if (ret)
 534                         break;
 535                 if (arg[0] == '-')
 536                         offset = -offset;
 537                 arg = tmp + 1;
 538                 tmp = strrchr(arg, ')');
 539                 if (tmp) {
 540                         struct indirect_fetch_data *id;
 541                         *tmp = '\0';
 542                         id = kzalloc(sizeof(struct indirect_fetch_data),
 543                                      GFP_KERNEL);
 544                         if (!id)
 545                                 return -ENOMEM;
 546                         id->offset = offset;
 547                         ret = __parse_probe_arg(arg, &id->orig, is_return);
 548                         if (ret)
 549                                 kfree(id);
 550                         else {
 551                                 ff->func = fetch_indirect;
 552                                 ff->data = (void *)id;
 553                         }
 554                 } else
 555                         ret = -EINVAL;
 556                 break;
 557         default:
 558                 /* TODO: support custom handler */
 559                 ret = -EINVAL;
 560         }
 561         return ret;
 562 }
 563
 564 /* String length checking wrapper */
 565 static int parse_probe_arg(char *arg, struct fetch_func *ff, int is_return)
 566 {
 567         if (strlen(arg) > MAX_ARGSTR_LEN) {
 568                 pr_info("Argument is too long.: %s\n",  arg);
 569                 return -ENOSPC;
 570         }
 571         return __parse_probe_arg(arg, ff, is_return);
 572 }
 573
 574 /* Return 1 if name is reserved or already used by another argument */
 575 static int conflict_field_name(const char *name,
 576                                struct probe_arg *args, int narg)
 577 {
 578         int i;
 579         for (i = 0; i < ARRAY_SIZE(reserved_field_names); i++)
 580                 if (strcmp(reserved_field_names[i], name) == 0)
 581                         return 1;
 582         for (i = 0; i < narg; i++)
 583                 if (strcmp(args[i].name, name) == 0)
 584                         return 1;
 585         return 0;
 586 }
 587
 588 static int create_trace_probe(int argc, char **argv)
 589 {
 590         /*
 591          * Argument syntax:
 592          *  - Add kprobe: p[:[GRP/]EVENT] KSYM[+OFFS]|KADDR [FETCHARGS]
 593          *  - Add kretprobe: r[:[GRP/]EVENT] KSYM[+0] [FETCHARGS]
 594          * Fetch args:
 595          *  $argN       : fetch Nth of function argument. (N:0-)
 596          *  $retval     : fetch return value
 597          *  $stack      : fetch stack address
 598          *  $stackN     : fetch Nth of stack (N:0-)
 599          *  @ADDR       : fetch memory at ADDR (ADDR should be in kernel)
 600          *  @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
 601          *  %REG        : fetch register REG
 602          * Indirect memory fetch:
 603          *  +|-offs(ARG) : fetch memory at ARG +|- offs address.
 604          * Alias name of args:
 605          *  NAME=FETCHARG : set NAME as alias of FETCHARG.
 606          */
 607         struct trace_probe *tp;
 608         int i, ret = 0;
 609         int is_return = 0, is_delete = 0;
 610         char *symbol = NULL, *event = NULL, *arg = NULL, *group = NULL;
 611         unsigned long offset = 0;
 612         void *addr = NULL;
 613         char buf[MAX_EVENT_NAME_LEN];
 614
 615         /* argc must be >= 1 */
 616         if (argv[0][0] == 'p')
 617                 is_return = 0;
 618         else if (argv[0][0] == 'r')
 619                 is_return = 1;
 620         else if (argv[0][0] == '-')
 621                 is_delete = 1;
 622         else {
 623                 pr_info("Probe definition must be started with 'p', 'r' or"
 624                         " '-'.\n");
 625                 return -EINVAL;
 626         }
 627
 628         if (argv[0][1] == ':') {
 629                 event = &argv[0][2];
 630                 if (strchr(event, '/')) {
 631                         group = event;
 632                         event = strchr(group, '/') + 1;
 633                         event[-1] = '\0';
 634                         if (strlen(group) == 0) {
 635                                 pr_info("Group name is not specifiled\n");
 636                                 return -EINVAL;
 637                         }
 638                 }
 639                 if (strlen(event) == 0) {
 640                         pr_info("Event name is not specifiled\n");
 641                         return -EINVAL;
 642                 }
 643         }
 644         if (!group)
 645                 group = KPROBE_EVENT_SYSTEM;
 646
 647         if (is_delete) {
 648                 if (!event) {
 649                         pr_info("Delete command needs an event name.\n");
 650                         return -EINVAL;
 651                 }
 652                 tp = find_probe_event(event, group);
 653                 if (!tp) {
 654                         pr_info("Event %s/%s doesn't exist.\n", group, event);
 655                         return -ENOENT;
 656                 }
 657                 /* delete an event */
 658                 unregister_trace_probe(tp);
 659                 free_trace_probe(tp);
 660                 return 0;
 661         }
 662
 663         if (argc < 2) {
 664                 pr_info("Probe point is not specified.\n");
 665                 return -EINVAL;
 666         }
 667         if (isdigit(argv[1][0])) {
 668                 if (is_return) {
 669                         pr_info("Return probe point must be a symbol.\n");
 670                         return -EINVAL;
 671                 }
 672                 /* an address specified */
 673                 ret = strict_strtoul(&argv[0][2], 0, (unsigned long *)&addr);
 674                 if (ret) {
 675                         pr_info("Failed to parse address.\n");
 676                         return ret;
 677                 }
 678         } else {
 679                 /* a symbol specified */
 680                 symbol = argv[1];
 681                 /* TODO: support .init module functions */
 682                 ret = split_symbol_offset(symbol, &offset);
 683                 if (ret) {
 684                         pr_info("Failed to parse symbol.\n");
 685                         return ret;
 686                 }
 687                 if (offset && is_return) {
 688                         pr_info("Return probe must be used without offset.\n");
 689                         return -EINVAL;
 690                 }
 691         }
 692         argc -= 2; argv += 2;
 693
 694         /* setup a probe */
 695         if (!event) {
 696                 /* Make a new event name */
 697                 if (symbol)
 698                         snprintf(buf, MAX_EVENT_NAME_LEN, "%c@%s%+ld",
 699                                  is_return ? 'r' : 'p', symbol, offset);
 700                 else
 701                         snprintf(buf, MAX_EVENT_NAME_LEN, "%c@0x%p",
 702                                  is_return ? 'r' : 'p', addr);
 703                 event = buf;
 704         }
 705         tp = alloc_trace_probe(group, event, addr, symbol, offset, argc,
 706                                is_return);
 707         if (IS_ERR(tp)) {
 708                 pr_info("Failed to allocate trace_probe.(%d)\n",
 709                         (int)PTR_ERR(tp));
 710                 return PTR_ERR(tp);
 711         }
 712
 713         /* parse arguments */
 714         ret = 0;
 715         for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
 716                 /* Parse argument name */
 717                 arg = strchr(argv[i], '=');
 718                 if (arg)
 719                         *arg++ = '\0';
 720                 else
 721                         arg = argv[i];
 722
 723                 if (conflict_field_name(argv[i], tp->args, i)) {
 724                         pr_info("Argument%d name '%s' conflicts with "
 725                                 "another field.\n", i, argv[i]);
 726                         ret = -EINVAL;
 727                         goto error;
 728                 }
 729
 730                 tp->args[i].name = kstrdup(argv[i], GFP_KERNEL);
 731                 if (!tp->args[i].name) {
 732                         pr_info("Failed to allocate argument%d name '%s'.\n",
 733                                 i, argv[i]);
 734                         ret = -ENOMEM;
 735                         goto error;
 736                 }
 737
 738                 /* Parse fetch argument */
 739                 ret = parse_probe_arg(arg, &tp->args[i].fetch, is_return);
 740                 if (ret) {
 741                         pr_info("Parse error at argument%d. (%d)\n", i, ret);
 742                         kfree(tp->args[i].name);
 743                         goto error;
 744                 }
 745
 746                 tp->nr_args++;
 747         }
 748
 749         ret = register_trace_probe(tp);
 750         if (ret)
 751                 goto error;
 752         return 0;
 753
 754 error:
 755         free_trace_probe(tp);
 756         return ret;
 757 }
 758
 759 static void cleanup_all_probes(void)
 760 {
 761         struct trace_probe *tp;
 762
 763         mutex_lock(&probe_lock);
 764         /* TODO: Use batch unregistration */
 765         while (!list_empty(&probe_list)) {
 766                 tp = list_entry(probe_list.next, struct trace_probe, list);
 767                 unregister_trace_probe(tp);
 768                 free_trace_probe(tp);
 769         }
 770         mutex_unlock(&probe_lock);
 771 }
 772
 773
 774 /* Probes listing interfaces */
 775 static void *probes_seq_start(struct seq_file *m, loff_t *pos)
 776 {
 777         mutex_lock(&probe_lock);
 778         return seq_list_start(&probe_list, *pos);
 779 }
 780
 781 static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
 782 {
 783         return seq_list_next(v, &probe_list, pos);
 784 }
 785
 786 static void probes_seq_stop(struct seq_file *m, void *v)
 787 {
 788         mutex_unlock(&probe_lock);
 789 }
 790
 791 static int probes_seq_show(struct seq_file *m, void *v)
 792 {
 793         struct trace_probe *tp = v;
 794         int i, ret;
 795         char buf[MAX_ARGSTR_LEN + 1];
 796
 797         seq_printf(m, "%c", probe_is_return(tp) ? 'r' : 'p');
 798         seq_printf(m, ":%s/%s", tp->call.system, tp->call.name);
 799
 800         if (!tp->symbol)
 801                 seq_printf(m, " 0x%p", tp->rp.kp.addr);
 802         else if (tp->rp.kp.offset)
 803                 seq_printf(m, " %s+%u", probe_symbol(tp), tp->rp.kp.offset);
 804         else
 805                 seq_printf(m, " %s", probe_symbol(tp));
 806
 807         for (i = 0; i < tp->nr_args; i++) {
 808                 ret = probe_arg_string(buf, MAX_ARGSTR_LEN, &tp->args[i].fetch);
 809                 if (ret < 0) {
 810                         pr_warning("Argument%d decoding error(%d).\n", i, ret);
 811                         return ret;
 812                 }
 813                 seq_printf(m, " %s=%s", tp->args[i].name, buf);
 814         }
 815         seq_printf(m, "\n");
 816         return 0;
 817 }
 818
 819 static const struct seq_operations probes_seq_op = {
 820         .start  = probes_seq_start,
 821         .next   = probes_seq_next,
 822         .stop   = probes_seq_stop,
 823         .show   = probes_seq_show
 824 };
 825
 826 static int probes_open(struct inode *inode, struct file *file)
 827 {
 828         if ((file->f_mode & FMODE_WRITE) &&
 829             (file->f_flags & O_TRUNC))
 830                 cleanup_all_probes();
 831
 832         return seq_open(file, &probes_seq_op);
 833 }
 834
 835 static int command_trace_probe(const char *buf)
 836 {
 837         char **argv;
 838         int argc = 0, ret = 0;
 839
 840         argv = argv_split(GFP_KERNEL, buf, &argc);
 841         if (!argv)
 842                 return -ENOMEM;
 843
 844         if (argc)
 845                 ret = create_trace_probe(argc, argv);
 846
 847         argv_free(argv);
 848         return ret;
 849 }
 850
 851 #define WRITE_BUFSIZE 128
 852
 853 static ssize_t probes_write(struct file *file, const char __user *buffer,
 854                             size_t count, loff_t *ppos)
 855 {
 856         char *kbuf, *tmp;
 857         int ret;
 858         size_t done;
 859         size_t size;
 860
 861         kbuf = kmalloc(WRITE_BUFSIZE, GFP_KERNEL);
 862         if (!kbuf)
 863                 return -ENOMEM;
 864
 865         ret = done = 0;
 866         while (done < count) {
 867                 size = count - done;
 868                 if (size >= WRITE_BUFSIZE)
 869                         size = WRITE_BUFSIZE - 1;
 870                 if (copy_from_user(kbuf, buffer + done, size)) {
 871                         ret = -EFAULT;
 872                         goto out;
 873                 }
 874                 kbuf[size] = '\0';
 875                 tmp = strchr(kbuf, '\n');
 876                 if (tmp) {
 877                         *tmp = '\0';
 878                         size = tmp - kbuf + 1;
 879                 } else if (done + size < count) {
 880                         pr_warning("Line length is too long: "
 881                                    "Should be less than %d.", WRITE_BUFSIZE);
 882                         ret = -EINVAL;
 883                         goto out;
 884                 }
 885                 done += size;
 886                 /* Remove comments */
 887                 tmp = strchr(kbuf, '#');
 888                 if (tmp)
 889                         *tmp = '\0';
 890
 891                 ret = command_trace_probe(kbuf);
 892                 if (ret)
 893                         goto out;
 894         }
 895         ret = done;
 896 out:
 897         kfree(kbuf);
 898         return ret;
 899 }
 900
 901 static const struct file_operations kprobe_events_ops = {
 902         .owner          = THIS_MODULE,
 903         .open           = probes_open,
 904         .read           = seq_read,
 905         .llseek         = seq_lseek,
 906         .release        = seq_release,
 907         .write          = probes_write,
 908 };
 909
 910 /* Probes profiling interfaces */
 911 static int probes_profile_seq_show(struct seq_file *m, void *v)
 912 {
 913         struct trace_probe *tp = v;
 914
 915         seq_printf(m, "  %-44s %15lu %15lu\n", tp->call.name, tp->nhit,
 916                    tp->rp.kp.nmissed);
 917
 918         return 0;
 919 }
 920
 921 static const struct seq_operations profile_seq_op = {
 922         .start  = probes_seq_start,
 923         .next   = probes_seq_next,
 924         .stop   = probes_seq_stop,
 925         .show   = probes_profile_seq_show
 926 };
 927
 928 static int profile_open(struct inode *inode, struct file *file)
 929 {
 930         return seq_open(file, &profile_seq_op);
 931 }
 932
 933 static const struct file_operations kprobe_profile_ops = {
 934         .owner          = THIS_MODULE,
 935         .open           = profile_open,
 936         .read           = seq_read,
 937         .llseek         = seq_lseek,
 938         .release        = seq_release,
 939 };
 940
 941 /* Kprobe handler */
 942 static __kprobes int kprobe_trace_func(struct kprobe *kp, struct pt_regs *regs)
 943 {
 944         struct trace_probe *tp = container_of(kp, struct trace_probe, rp.kp);
 945         struct kprobe_trace_entry *entry;
 946         struct ring_buffer_event *event;
 947         struct ring_buffer *buffer;
 948         int size, i, pc;
 949         unsigned long irq_flags;
 950         struct ftrace_event_call *call = &tp->call;
 951
 952         tp->nhit++;
 953
 954         local_save_flags(irq_flags);
 955         pc = preempt_count();
 956
 957         size = SIZEOF_KPROBE_TRACE_ENTRY(tp->nr_args);
 958
 959         event = trace_current_buffer_lock_reserve(&buffer, call->id, size,
 960                                                   irq_flags, pc);
 961         if (!event)
 962                 return 0;
 963
 964         entry = ring_buffer_event_data(event);
 965         entry->nargs = tp->nr_args;
 966         entry->ip = (unsigned long)kp->addr;
 967         for (i = 0; i < tp->nr_args; i++)
 968                 entry->args[i] = call_fetch(&tp->args[i].fetch, regs);
 969
 970         if (!filter_current_check_discard(buffer, call, entry, event))
 971                 trace_nowake_buffer_unlock_commit(buffer, event, irq_flags, pc);
 972         return 0;
 973 }
 974
 975 /* Kretprobe handler */
 976 static __kprobes int kretprobe_trace_func(struct kretprobe_instance *ri,
 977                                           struct pt_regs *regs)
 978 {
 979         struct trace_probe *tp = container_of(ri->rp, struct trace_probe, rp);
 980         struct kretprobe_trace_entry *entry;
 981         struct ring_buffer_event *event;
 982         struct ring_buffer *buffer;
 983         int size, i, pc;
 984         unsigned long irq_flags;
 985         struct ftrace_event_call *call = &tp->call;
 986
 987         local_save_flags(irq_flags);
 988         pc = preempt_count();
 989
 990         size = SIZEOF_KRETPROBE_TRACE_ENTRY(tp->nr_args);
 991
 992         event = trace_current_buffer_lock_reserve(&buffer, call->id, size,
 993                                                   irq_flags, pc);
 994         if (!event)
 995                 return 0;
 996
 997         entry = ring_buffer_event_data(event);
 998         entry->nargs = tp->nr_args;
 999         entry->func = (unsigned long)tp->rp.kp.addr;
1000         entry->ret_ip = (unsigned long)ri->ret_addr;
1001         for (i = 0; i < tp->nr_args; i++)
1002                 entry->args[i] = call_fetch(&tp->args[i].fetch, regs);
1003
1004         if (!filter_current_check_discard(buffer, call, entry, event))
1005                 trace_nowake_buffer_unlock_commit(buffer, event, irq_flags, pc);
1006
1007         return 0;
1008 }
1009
1010 /* Event entry printers */
1011 enum print_line_t
1012 print_kprobe_event(struct trace_iterator *iter, int flags)
1013 {
1014         struct kprobe_trace_entry *field;
1015         struct trace_seq *s = &iter->seq;
1016         struct trace_event *event;
1017         struct trace_probe *tp;
1018         int i;
1019
1020         field = (struct kprobe_trace_entry *)iter->ent;
1021         event = ftrace_find_event(field->ent.type);
1022         tp = container_of(event, struct trace_probe, event);
1023
1024         if (!trace_seq_printf(s, "%s: (", tp->call.name))
1025                 goto partial;
1026
1027         if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
1028                 goto partial;
1029
1030         if (!trace_seq_puts(s, ")"))
1031                 goto partial;
1032
1033         for (i = 0; i < field->nargs; i++)
1034                 if (!trace_seq_printf(s, " %s=%lx",
1035                                       tp->args[i].name, field->args[i]))
1036                         goto partial;
1037
1038         if (!trace_seq_puts(s, "\n"))
1039                 goto partial;
1040
1041         return TRACE_TYPE_HANDLED;
1042 partial:
1043         return TRACE_TYPE_PARTIAL_LINE;
1044 }
1045
1046 enum print_line_t
1047 print_kretprobe_event(struct trace_iterator *iter, int flags)
1048 {
1049         struct kretprobe_trace_entry *field;
1050         struct trace_seq *s = &iter->seq;
1051         struct trace_event *event;
1052         struct trace_probe *tp;
1053         int i;
1054
1055         field = (struct kretprobe_trace_entry *)iter->ent;
1056         event = ftrace_find_event(field->ent.type);
1057         tp = container_of(event, struct trace_probe, event);
1058
1059         if (!trace_seq_printf(s, "%s: (", tp->call.name))
1060                 goto partial;
1061
1062         if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1063                 goto partial;
1064
1065         if (!trace_seq_puts(s, " <- "))
1066                 goto partial;
1067
1068         if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1069                 goto partial;
1070
1071         if (!trace_seq_puts(s, ")"))
1072                 goto partial;
1073
1074         for (i = 0; i < field->nargs; i++)
1075                 if (!trace_seq_printf(s, " %s=%lx",
1076                                       tp->args[i].name, field->args[i]))
1077                         goto partial;
1078
1079         if (!trace_seq_puts(s, "\n"))
1080                 goto partial;
1081
1082         return TRACE_TYPE_HANDLED;
1083 partial:
1084         return TRACE_TYPE_PARTIAL_LINE;
1085 }
1086
1087 static int probe_event_enable(struct ftrace_event_call *call)
1088 {
1089         struct trace_probe *tp = (struct trace_probe *)call->data;
1090
1091         tp->flags |= TP_FLAG_TRACE;
1092         if (probe_is_return(tp))
1093                 return enable_kretprobe(&tp->rp);
1094         else
1095                 return enable_kprobe(&tp->rp.kp);
1096 }
1097
1098 static void probe_event_disable(struct ftrace_event_call *call)
1099 {
1100         struct trace_probe *tp = (struct trace_probe *)call->data;
1101
1102         tp->flags &= ~TP_FLAG_TRACE;
1103         if (!(tp->flags & (TP_FLAG_TRACE | TP_FLAG_PROFILE))) {
1104                 if (probe_is_return(tp))
1105                         disable_kretprobe(&tp->rp);
1106                 else
1107                         disable_kprobe(&tp->rp.kp);
1108         }
1109 }
1110
1111 static int probe_event_raw_init(struct ftrace_event_call *event_call)
1112 {
1113         INIT_LIST_HEAD(&event_call->fields);
1114
1115         return 0;
1116 }
1117
1118 #undef DEFINE_FIELD
1119 #define DEFINE_FIELD(type, item, name, is_signed)                       \
1120         do {                                                            \
1121                 ret = trace_define_field(event_call, #type, name,       \
1122                                          offsetof(typeof(field), item), \
1123                                          sizeof(field.item), is_signed, \
1124                                          FILTER_OTHER);                 \
1125                 if (ret)                                                \
1126                         return ret;                                     \
1127         } while (0)
1128
1129 static int kprobe_event_define_fields(struct ftrace_event_call *event_call)
1130 {
1131         int ret, i;
1132         struct kprobe_trace_entry field;
1133         struct trace_probe *tp = (struct trace_probe *)event_call->data;
1134
1135         DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1136         DEFINE_FIELD(int, nargs, FIELD_STRING_NARGS, 1);
1137         /* Set argument names as fields */
1138         for (i = 0; i < tp->nr_args; i++)
1139                 DEFINE_FIELD(unsigned long, args[i], tp->args[i].name, 0);
1140         return 0;
1141 }
1142
1143 static int kretprobe_event_define_fields(struct ftrace_event_call *event_call)
1144 {
1145         int ret, i;
1146         struct kretprobe_trace_entry field;
1147         struct trace_probe *tp = (struct trace_probe *)event_call->data;
1148
1149         DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1150         DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1151         DEFINE_FIELD(int, nargs, FIELD_STRING_NARGS, 1);
1152         /* Set argument names as fields */
1153         for (i = 0; i < tp->nr_args; i++)
1154                 DEFINE_FIELD(unsigned long, args[i], tp->args[i].name, 0);
1155         return 0;
1156 }
1157
1158 static int __probe_event_show_format(struct trace_seq *s,
1159                                      struct trace_probe *tp, const char *fmt,
1160                                      const char *arg)
1161 {
1162         int i;
1163
1164         /* Show format */
1165         if (!trace_seq_printf(s, "\nprint fmt: \"%s", fmt))
1166                 return 0;
1167
1168         for (i = 0; i < tp->nr_args; i++)
1169                 if (!trace_seq_printf(s, " %s=%%lx", tp->args[i].name))
1170                         return 0;
1171
1172         if (!trace_seq_printf(s, "\", %s", arg))
1173                 return 0;
1174
1175         for (i = 0; i < tp->nr_args; i++)
1176                 if (!trace_seq_printf(s, ", REC->%s", tp->args[i].name))
1177                         return 0;
1178
1179         return trace_seq_puts(s, "\n");
1180 }
1181
1182 #undef SHOW_FIELD
1183 #define SHOW_FIELD(type, item, name)                                    \
1184         do {                                                            \
1185                 ret = trace_seq_printf(s, "\tfield: " #type " %s;\t"    \
1186                                 "offset:%u;\tsize:%u;\n", name,         \
1187                                 (unsigned int)offsetof(typeof(field), item),\
1188                                 (unsigned int)sizeof(type));            \
1189                 if (!ret)                                               \
1190                         return 0;                                       \
1191         } while (0)
1192
1193 static int kprobe_event_show_format(struct ftrace_event_call *call,
1194                                     struct trace_seq *s)
1195 {
1196         struct kprobe_trace_entry field __attribute__((unused));
1197         int ret, i;
1198         struct trace_probe *tp = (struct trace_probe *)call->data;
1199
1200         SHOW_FIELD(unsigned long, ip, FIELD_STRING_IP);
1201         SHOW_FIELD(int, nargs, FIELD_STRING_NARGS);
1202
1203         /* Show fields */
1204         for (i = 0; i < tp->nr_args; i++)
1205                 SHOW_FIELD(unsigned long, args[i], tp->args[i].name);
1206         trace_seq_puts(s, "\n");
1207
1208         return __probe_event_show_format(s, tp, "(%lx)",
1209                                          "REC->" FIELD_STRING_IP);
1210 }
1211
1212 static int kretprobe_event_show_format(struct ftrace_event_call *call,
1213                                        struct trace_seq *s)
1214 {
1215         struct kretprobe_trace_entry field __attribute__((unused));
1216         int ret, i;
1217         struct trace_probe *tp = (struct trace_probe *)call->data;
1218
1219         SHOW_FIELD(unsigned long, func, FIELD_STRING_FUNC);
1220         SHOW_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP);
1221         SHOW_FIELD(int, nargs, FIELD_STRING_NARGS);
1222
1223         /* Show fields */
1224         for (i = 0; i < tp->nr_args; i++)
1225                 SHOW_FIELD(unsigned long, args[i], tp->args[i].name);
1226         trace_seq_puts(s, "\n");
1227
1228         return __probe_event_show_format(s, tp, "(%lx <- %lx)",
1229                                          "REC->" FIELD_STRING_FUNC
1230                                          ", REC->" FIELD_STRING_RETIP);
1231 }
1232
1233 #ifdef CONFIG_EVENT_PROFILE
1234
1235 /* Kprobe profile handler */
1236 static __kprobes int kprobe_profile_func(struct kprobe *kp,
1237                                          struct pt_regs *regs)
1238 {
1239         struct trace_probe *tp = container_of(kp, struct trace_probe, rp.kp);
1240         struct ftrace_event_call *call = &tp->call;
1241         struct kprobe_trace_entry *entry;
1242         struct trace_entry *ent;
1243         int size, __size, i, pc, __cpu;
1244         unsigned long irq_flags;
1245         char *trace_buf;
1246         char *raw_data;
1247         int rctx;
1248
1249         pc = preempt_count();
1250         __size = SIZEOF_KPROBE_TRACE_ENTRY(tp->nr_args);
1251         size = ALIGN(__size + sizeof(u32), sizeof(u64));
1252         size -= sizeof(u32);
1253         if (WARN_ONCE(size > FTRACE_MAX_PROFILE_SIZE,
1254                      "profile buffer not large enough"))
1255                 return 0;
1256
1257         /*
1258          * Protect the non nmi buffer
1259          * This also protects the rcu read side
1260          */
1261         local_irq_save(irq_flags);
1262
1263         rctx = perf_swevent_get_recursion_context();
1264         if (rctx < 0)
1265                 goto end_recursion;
1266
1267         __cpu = smp_processor_id();
1268
1269         if (in_nmi())
1270                 trace_buf = rcu_dereference(perf_trace_buf_nmi);
1271         else
1272                 trace_buf = rcu_dereference(perf_trace_buf);
1273
1274         if (!trace_buf)
1275                 goto end;
1276
1277         raw_data = per_cpu_ptr(trace_buf, __cpu);
1278
1279         /* Zero dead bytes from alignment to avoid buffer leak to userspace */
1280         *(u64 *)(&raw_data[size - sizeof(u64)]) = 0ULL;
1281         entry = (struct kprobe_trace_entry *)raw_data;
1282         ent = &entry->ent;
1283
1284         tracing_generic_entry_update(ent, irq_flags, pc);
1285         ent->type = call->id;
1286         entry->nargs = tp->nr_args;
1287         entry->ip = (unsigned long)kp->addr;
1288         for (i = 0; i < tp->nr_args; i++)
1289                 entry->args[i] = call_fetch(&tp->args[i].fetch, regs);
1290         perf_tp_event(call->id, entry->ip, 1, entry, size);
1291
1292 end:
1293         perf_swevent_put_recursion_context(rctx);
1294 end_recursion:
1295         local_irq_restore(irq_flags);
1296
1297         return 0;
1298 }
1299
1300 /* Kretprobe profile handler */
1301 static __kprobes int kretprobe_profile_func(struct kretprobe_instance *ri,
1302                                             struct pt_regs *regs)
1303 {
1304         struct trace_probe *tp = container_of(ri->rp, struct trace_probe, rp);
1305         struct ftrace_event_call *call = &tp->call;
1306         struct kretprobe_trace_entry *entry;
1307         struct trace_entry *ent;
1308         int size, __size, i, pc, __cpu;
1309         unsigned long irq_flags;
1310         char *trace_buf;
1311         char *raw_data;
1312         int rctx;
1313
1314         pc = preempt_count();
1315         __size = SIZEOF_KRETPROBE_TRACE_ENTRY(tp->nr_args);
1316         size = ALIGN(__size + sizeof(u32), sizeof(u64));
1317         size -= sizeof(u32);
1318         if (WARN_ONCE(size > FTRACE_MAX_PROFILE_SIZE,
1319                      "profile buffer not large enough"))
1320                 return 0;
1321
1322         /*
1323          * Protect the non nmi buffer
1324          * This also protects the rcu read side
1325          */
1326         local_irq_save(irq_flags);
1327
1328         rctx = perf_swevent_get_recursion_context();
1329         if (rctx < 0)
1330                 goto end_recursion;
1331
1332         __cpu = smp_processor_id();
1333
1334         if (in_nmi())
1335                 trace_buf = rcu_dereference(perf_trace_buf_nmi);
1336         else
1337                 trace_buf = rcu_dereference(perf_trace_buf);
1338
1339         if (!trace_buf)
1340                 goto end;
1341
1342         raw_data = per_cpu_ptr(trace_buf, __cpu);
1343
1344         /* Zero dead bytes from alignment to avoid buffer leak to userspace */
1345         *(u64 *)(&raw_data[size - sizeof(u64)]) = 0ULL;
1346         entry = (struct kretprobe_trace_entry *)raw_data;
1347         ent = &entry->ent;
1348
1349         tracing_generic_entry_update(ent, irq_flags, pc);
1350         ent->type = call->id;
1351         entry->nargs = tp->nr_args;
1352         entry->func = (unsigned long)tp->rp.kp.addr;
1353         entry->ret_ip = (unsigned long)ri->ret_addr;
1354         for (i = 0; i < tp->nr_args; i++)
1355                 entry->args[i] = call_fetch(&tp->args[i].fetch, regs);
1356         perf_tp_event(call->id, entry->ret_ip, 1, entry, size);
1357
1358 end:
1359         perf_swevent_put_recursion_context(rctx);
1360 end_recursion:
1361         local_irq_restore(irq_flags);
1362
1363         return 0;
1364 }
1365
1366 static int probe_profile_enable(struct ftrace_event_call *call)
1367 {
1368         struct trace_probe *tp = (struct trace_probe *)call->data;
1369
1370         tp->flags |= TP_FLAG_PROFILE;
1371
1372         if (probe_is_return(tp))
1373                 return enable_kretprobe(&tp->rp);
1374         else
1375                 return enable_kprobe(&tp->rp.kp);
1376 }
1377
1378 static void probe_profile_disable(struct ftrace_event_call *call)
1379 {
1380         struct trace_probe *tp = (struct trace_probe *)call->data;
1381
1382         tp->flags &= ~TP_FLAG_PROFILE;
1383
1384         if (!(tp->flags & TP_FLAG_TRACE)) {
1385                 if (probe_is_return(tp))
1386                         disable_kretprobe(&tp->rp);
1387                 else
1388                         disable_kprobe(&tp->rp.kp);
1389         }
1390 }
1391 #endif  /* CONFIG_EVENT_PROFILE */
1392
1393
1394 static __kprobes
1395 int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1396 {
1397         struct trace_probe *tp = container_of(kp, struct trace_probe, rp.kp);
1398
1399         if (tp->flags & TP_FLAG_TRACE)
1400                 kprobe_trace_func(kp, regs);
1401 #ifdef CONFIG_EVENT_PROFILE
1402         if (tp->flags & TP_FLAG_PROFILE)
1403                 kprobe_profile_func(kp, regs);
1404 #endif  /* CONFIG_EVENT_PROFILE */
1405         return 0;       /* We don't tweek kernel, so just return 0 */
1406 }
1407
1408 static __kprobes
1409 int kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1410 {
1411         struct trace_probe *tp = container_of(ri->rp, struct trace_probe, rp);
1412
1413         if (tp->flags & TP_FLAG_TRACE)
1414                 kretprobe_trace_func(ri, regs);
1415 #ifdef CONFIG_EVENT_PROFILE
1416         if (tp->flags & TP_FLAG_PROFILE)
1417                 kretprobe_profile_func(ri, regs);
1418 #endif  /* CONFIG_EVENT_PROFILE */
1419         return 0;       /* We don't tweek kernel, so just return 0 */
1420 }
1421
1422 static int register_probe_event(struct trace_probe *tp)
1423 {
1424         struct ftrace_event_call *call = &tp->call;
1425         int ret;
1426
1427         /* Initialize ftrace_event_call */
1428         if (probe_is_return(tp)) {
1429                 tp->event.trace = print_kretprobe_event;
1430                 call->raw_init = probe_event_raw_init;
1431                 call->show_format = kretprobe_event_show_format;
1432                 call->define_fields = kretprobe_event_define_fields;
1433         } else {
1434                 tp->event.trace = print_kprobe_event;
1435                 call->raw_init = probe_event_raw_init;
1436                 call->show_format = kprobe_event_show_format;
1437                 call->define_fields = kprobe_event_define_fields;
1438         }
1439         call->event = &tp->event;
1440         call->id = register_ftrace_event(&tp->event);
1441         if (!call->id)
1442                 return -ENODEV;
1443         call->enabled = 0;
1444         call->regfunc = probe_event_enable;
1445         call->unregfunc = probe_event_disable;
1446
1447 #ifdef CONFIG_EVENT_PROFILE
1448         call->profile_enable = probe_profile_enable;
1449         call->profile_disable = probe_profile_disable;
1450 #endif
1451         call->data = tp;
1452         ret = trace_add_event_call(call);
1453         if (ret) {
1454                 pr_info("Failed to register kprobe event: %s\n", call->name);
1455                 unregister_ftrace_event(&tp->event);
1456         }
1457         return ret;
1458 }
1459
1460 static void unregister_probe_event(struct trace_probe *tp)
1461 {
1462         /* tp->event is unregistered in trace_remove_event_call() */
1463         trace_remove_event_call(&tp->call);
1464 }
1465
1466 /* Make a debugfs interface for controling probe points */
1467 static __init int init_kprobe_trace(void)
1468 {
1469         struct dentry *d_tracer;
1470         struct dentry *entry;
1471
1472         d_tracer = tracing_init_dentry();
1473         if (!d_tracer)
1474                 return 0;
1475
1476         entry = debugfs_create_file("kprobe_events", 0644, d_tracer,
1477                                     NULL, &kprobe_events_ops);
1478
1479         /* Event list interface */
1480         if (!entry)
1481                 pr_warning("Could not create debugfs "
1482                            "'kprobe_events' entry\n");
1483
1484         /* Profile interface */
1485         entry = debugfs_create_file("kprobe_profile", 0444, d_tracer,
1486                                     NULL, &kprobe_profile_ops);
1487
1488         if (!entry)
1489                 pr_warning("Could not create debugfs "
1490                            "'kprobe_profile' entry\n");
1491         return 0;
1492 }
1493 fs_initcall(init_kprobe_trace);
1494
1495
1496 #ifdef CONFIG_FTRACE_STARTUP_TEST
1497
1498 static int kprobe_trace_selftest_target(int a1, int a2, int a3,
1499                                         int a4, int a5, int a6)
1500 {
1501         return a1 + a2 + a3 + a4 + a5 + a6;
1502 }
1503
1504 static __init int kprobe_trace_self_tests_init(void)
1505 {
1506         int ret;
1507         int (*target)(int, int, int, int, int, int);
1508
1509         target = kprobe_trace_selftest_target;
1510
1511         pr_info("Testing kprobe tracing: ");
1512
1513         ret = command_trace_probe("p:testprobe kprobe_trace_selftest_target "
1514                                   "$arg1 $arg2 $arg3 $arg4 $stack $stack0");
1515         if (WARN_ON_ONCE(ret))
1516                 pr_warning("error enabling function entry\n");
1517
1518         ret = command_trace_probe("r:testprobe2 kprobe_trace_selftest_target "
1519                                   "$retval");
1520         if (WARN_ON_ONCE(ret))
1521                 pr_warning("error enabling function return\n");
1522
1523         ret = target(1, 2, 3, 4, 5, 6);
1524
1525         cleanup_all_probes();
1526
1527         pr_cont("OK\n");
1528         return 0;
1529 }
1530
1531 late_initcall(kprobe_trace_self_tests_init);
1532
1533 #endif