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 /* Check the name is good for event/group */
 286 static int check_event_name(const char *name)
 287 {
 288         if (!isalpha(*name) && *name != '_')
 289                 return 0;
 290         while (*++name != '\0') {
 291                 if (!isalpha(*name) && !isdigit(*name) && *name != '_')
 292                         return 0;
 293         }
 294         return 1;
 295 }
 296
 297 /*
 298  * Allocate new trace_probe and initialize it (including kprobes).
 299  */
 300 static struct trace_probe *alloc_trace_probe(const char *group,
 301                                              const char *event,
 302                                              void *addr,
 303                                              const char *symbol,
 304                                              unsigned long offs,
 305                                              int nargs, int is_return)
 306 {
 307         struct trace_probe *tp;
 308         int ret = -ENOMEM;
 309
 310         tp = kzalloc(SIZEOF_TRACE_PROBE(nargs), GFP_KERNEL);
 311         if (!tp)
 312                 return ERR_PTR(ret);
 313
 314         if (symbol) {
 315                 tp->symbol = kstrdup(symbol, GFP_KERNEL);
 316                 if (!tp->symbol)
 317                         goto error;
 318                 tp->rp.kp.symbol_name = tp->symbol;
 319                 tp->rp.kp.offset = offs;
 320         } else
 321                 tp->rp.kp.addr = addr;
 322
 323         if (is_return)
 324                 tp->rp.handler = kretprobe_dispatcher;
 325         else
 326                 tp->rp.kp.pre_handler = kprobe_dispatcher;
 327
 328         if (!event || !check_event_name(event)) {
 329                 ret = -EINVAL;
 330                 goto error;
 331         }
 332
 333         tp->call.name = kstrdup(event, GFP_KERNEL);
 334         if (!tp->call.name)
 335                 goto error;
 336
 337         if (!group || !check_event_name(group)) {
 338                 ret = -EINVAL;
 339                 goto error;
 340         }
 341
 342         tp->call.system = kstrdup(group, GFP_KERNEL);
 343         if (!tp->call.system)
 344                 goto error;
 345
 346         INIT_LIST_HEAD(&tp->list);
 347         return tp;
 348 error:
 349         kfree(tp->call.name);
 350         kfree(tp->symbol);
 351         kfree(tp);
 352         return ERR_PTR(ret);
 353 }
 354
 355 static void free_probe_arg(struct probe_arg *arg)
 356 {
 357         if (arg->fetch.func == fetch_symbol)
 358                 free_symbol_cache(arg->fetch.data);
 359         else if (arg->fetch.func == fetch_indirect)
 360                 free_indirect_fetch_data(arg->fetch.data);
 361         kfree(arg->name);
 362 }
 363
 364 static void free_trace_probe(struct trace_probe *tp)
 365 {
 366         int i;
 367
 368         for (i = 0; i < tp->nr_args; i++)
 369                 free_probe_arg(&tp->args[i]);
 370
 371         kfree(tp->call.system);
 372         kfree(tp->call.name);
 373         kfree(tp->symbol);
 374         kfree(tp);
 375 }
 376
 377 static struct trace_probe *find_probe_event(const char *event,
 378                                             const char *group)
 379 {
 380         struct trace_probe *tp;
 381
 382         list_for_each_entry(tp, &probe_list, list)
 383                 if (strcmp(tp->call.name, event) == 0 &&
 384                     strcmp(tp->call.system, group) == 0)
 385                         return tp;
 386         return NULL;
 387 }
 388
 389 /* Unregister a trace_probe and probe_event: call with locking probe_lock */
 390 static void unregister_trace_probe(struct trace_probe *tp)
 391 {
 392         if (probe_is_return(tp))
 393                 unregister_kretprobe(&tp->rp);
 394         else
 395                 unregister_kprobe(&tp->rp.kp);
 396         list_del(&tp->list);
 397         unregister_probe_event(tp);
 398 }
 399
 400 /* Register a trace_probe and probe_event */
 401 static int register_trace_probe(struct trace_probe *tp)
 402 {
 403         struct trace_probe *old_tp;
 404         int ret;
 405
 406         mutex_lock(&probe_lock);
 407
 408         /* register as an event */
 409         old_tp = find_probe_event(tp->call.name, tp->call.system);
 410         if (old_tp) {
 411                 /* delete old event */
 412                 unregister_trace_probe(old_tp);
 413                 free_trace_probe(old_tp);
 414         }
 415         ret = register_probe_event(tp);
 416         if (ret) {
 417                 pr_warning("Faild to register probe event(%d)\n", ret);
 418                 goto end;
 419         }
 420
 421         tp->rp.kp.flags |= KPROBE_FLAG_DISABLED;
 422         if (probe_is_return(tp))
 423                 ret = register_kretprobe(&tp->rp);
 424         else
 425                 ret = register_kprobe(&tp->rp.kp);
 426
 427         if (ret) {
 428                 pr_warning("Could not insert probe(%d)\n", ret);
 429                 if (ret == -EILSEQ) {
 430                         pr_warning("Probing address(0x%p) is not an "
 431                                    "instruction boundary.\n",
 432                                    tp->rp.kp.addr);
 433                         ret = -EINVAL;
 434                 }
 435                 unregister_probe_event(tp);
 436         } else
 437                 list_add_tail(&tp->list, &probe_list);
 438 end:
 439         mutex_unlock(&probe_lock);
 440         return ret;
 441 }
 442
 443 /* Split symbol and offset. */
 444 static int split_symbol_offset(char *symbol, unsigned long *offset)
 445 {
 446         char *tmp;
 447         int ret;
 448
 449         if (!offset)
 450                 return -EINVAL;
 451
 452         tmp = strchr(symbol, '+');
 453         if (tmp) {
 454                 /* skip sign because strict_strtol doesn't accept '+' */
 455                 ret = strict_strtoul(tmp + 1, 0, offset);
 456                 if (ret)
 457                         return ret;
 458                 *tmp = '\0';
 459         } else
 460                 *offset = 0;
 461         return 0;
 462 }
 463
 464 #define PARAM_MAX_ARGS 16
 465 #define PARAM_MAX_STACK (THREAD_SIZE / sizeof(unsigned long))
 466
 467 static int parse_probe_vars(char *arg, struct fetch_func *ff, int is_return)
 468 {
 469         int ret = 0;
 470         unsigned long param;
 471
 472         if (strcmp(arg, "retval") == 0) {
 473                 if (is_return) {
 474                         ff->func = fetch_retvalue;
 475                         ff->data = NULL;
 476                 } else
 477                         ret = -EINVAL;
 478         } else if (strncmp(arg, "stack", 5) == 0) {
 479                 if (arg[5] == '\0') {
 480                         ff->func = fetch_stack_address;
 481                         ff->data = NULL;
 482                 } else if (isdigit(arg[5])) {
 483                         ret = strict_strtoul(arg + 5, 10, &param);
 484                         if (ret || param > PARAM_MAX_STACK)
 485                                 ret = -EINVAL;
 486                         else {
 487                                 ff->func = fetch_stack;
 488                                 ff->data = (void *)param;
 489                         }
 490                 } else
 491                         ret = -EINVAL;
 492         } else if (strncmp(arg, "arg", 3) == 0 && isdigit(arg[3])) {
 493                 ret = strict_strtoul(arg + 3, 10, &param);
 494                 if (ret || param > PARAM_MAX_ARGS)
 495                         ret = -EINVAL;
 496                 else {
 497                         ff->func = fetch_argument;
 498                         ff->data = (void *)param;
 499                 }
 500         } else
 501                 ret = -EINVAL;
 502         return ret;
 503 }
 504
 505 /* Recursive argument parser */
 506 static int __parse_probe_arg(char *arg, struct fetch_func *ff, int is_return)
 507 {
 508         int ret = 0;
 509         unsigned long param;
 510         long offset;
 511         char *tmp;
 512
 513         switch (arg[0]) {
 514         case '$':
 515                 ret = parse_probe_vars(arg + 1, ff, is_return);
 516                 break;
 517         case '%':       /* named register */
 518                 ret = regs_query_register_offset(arg + 1);
 519                 if (ret >= 0) {
 520                         ff->func = fetch_register;
 521                         ff->data = (void *)(unsigned long)ret;
 522                         ret = 0;
 523                 }
 524                 break;
 525         case '@':       /* memory or symbol */
 526                 if (isdigit(arg[1])) {
 527                         ret = strict_strtoul(arg + 1, 0, &param);
 528                         if (ret)
 529                                 break;
 530                         ff->func = fetch_memory;
 531                         ff->data = (void *)param;
 532                 } else {
 533                         ret = split_symbol_offset(arg + 1, &offset);
 534                         if (ret)
 535                                 break;
 536                         ff->data = alloc_symbol_cache(arg + 1, offset);
 537                         if (ff->data)
 538                                 ff->func = fetch_symbol;
 539                         else
 540                                 ret = -EINVAL;
 541                 }
 542                 break;
 543         case '+':       /* indirect memory */
 544         case '-':
 545                 tmp = strchr(arg, '(');
 546                 if (!tmp) {
 547                         ret = -EINVAL;
 548                         break;
 549                 }
 550                 *tmp = '\0';
 551                 ret = strict_strtol(arg + 1, 0, &offset);
 552                 if (ret)
 553                         break;
 554                 if (arg[0] == '-')
 555                         offset = -offset;
 556                 arg = tmp + 1;
 557                 tmp = strrchr(arg, ')');
 558                 if (tmp) {
 559                         struct indirect_fetch_data *id;
 560                         *tmp = '\0';
 561                         id = kzalloc(sizeof(struct indirect_fetch_data),
 562                                      GFP_KERNEL);
 563                         if (!id)
 564                                 return -ENOMEM;
 565                         id->offset = offset;
 566                         ret = __parse_probe_arg(arg, &id->orig, is_return);
 567                         if (ret)
 568                                 kfree(id);
 569                         else {
 570                                 ff->func = fetch_indirect;
 571                                 ff->data = (void *)id;
 572                         }
 573                 } else
 574                         ret = -EINVAL;
 575                 break;
 576         default:
 577                 /* TODO: support custom handler */
 578                 ret = -EINVAL;
 579         }
 580         return ret;
 581 }
 582
 583 /* String length checking wrapper */
 584 static int parse_probe_arg(char *arg, struct fetch_func *ff, int is_return)
 585 {
 586         if (strlen(arg) > MAX_ARGSTR_LEN) {
 587                 pr_info("Argument is too long.: %s\n",  arg);
 588                 return -ENOSPC;
 589         }
 590         return __parse_probe_arg(arg, ff, is_return);
 591 }
 592
 593 /* Return 1 if name is reserved or already used by another argument */
 594 static int conflict_field_name(const char *name,
 595                                struct probe_arg *args, int narg)
 596 {
 597         int i;
 598         for (i = 0; i < ARRAY_SIZE(reserved_field_names); i++)
 599                 if (strcmp(reserved_field_names[i], name) == 0)
 600                         return 1;
 601         for (i = 0; i < narg; i++)
 602                 if (strcmp(args[i].name, name) == 0)
 603                         return 1;
 604         return 0;
 605 }
 606
 607 static int create_trace_probe(int argc, char **argv)
 608 {
 609         /*
 610          * Argument syntax:
 611          *  - Add kprobe: p[:[GRP/]EVENT] KSYM[+OFFS]|KADDR [FETCHARGS]
 612          *  - Add kretprobe: r[:[GRP/]EVENT] KSYM[+0] [FETCHARGS]
 613          * Fetch args:
 614          *  $argN       : fetch Nth of function argument. (N:0-)
 615          *  $retval     : fetch return value
 616          *  $stack      : fetch stack address
 617          *  $stackN     : fetch Nth of stack (N:0-)
 618          *  @ADDR       : fetch memory at ADDR (ADDR should be in kernel)
 619          *  @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
 620          *  %REG        : fetch register REG
 621          * Indirect memory fetch:
 622          *  +|-offs(ARG) : fetch memory at ARG +|- offs address.
 623          * Alias name of args:
 624          *  NAME=FETCHARG : set NAME as alias of FETCHARG.
 625          */
 626         struct trace_probe *tp;
 627         int i, ret = 0;
 628         int is_return = 0, is_delete = 0;
 629         char *symbol = NULL, *event = NULL, *arg = NULL, *group = NULL;
 630         unsigned long offset = 0;
 631         void *addr = NULL;
 632         char buf[MAX_EVENT_NAME_LEN];
 633
 634         /* argc must be >= 1 */
 635         if (argv[0][0] == 'p')
 636                 is_return = 0;
 637         else if (argv[0][0] == 'r')
 638                 is_return = 1;
 639         else if (argv[0][0] == '-')
 640                 is_delete = 1;
 641         else {
 642                 pr_info("Probe definition must be started with 'p', 'r' or"
 643                         " '-'.\n");
 644                 return -EINVAL;
 645         }
 646
 647         if (argv[0][1] == ':') {
 648                 event = &argv[0][2];
 649                 if (strchr(event, '/')) {
 650                         group = event;
 651                         event = strchr(group, '/') + 1;
 652                         event[-1] = '\0';
 653                         if (strlen(group) == 0) {
 654                                 pr_info("Group name is not specifiled\n");
 655                                 return -EINVAL;
 656                         }
 657                 }
 658                 if (strlen(event) == 0) {
 659                         pr_info("Event name is not specifiled\n");
 660                         return -EINVAL;
 661                 }
 662         }
 663         if (!group)
 664                 group = KPROBE_EVENT_SYSTEM;
 665
 666         if (is_delete) {
 667                 if (!event) {
 668                         pr_info("Delete command needs an event name.\n");
 669                         return -EINVAL;
 670                 }
 671                 tp = find_probe_event(event, group);
 672                 if (!tp) {
 673                         pr_info("Event %s/%s doesn't exist.\n", group, event);
 674                         return -ENOENT;
 675                 }
 676                 /* delete an event */
 677                 unregister_trace_probe(tp);
 678                 free_trace_probe(tp);
 679                 return 0;
 680         }
 681
 682         if (argc < 2) {
 683                 pr_info("Probe point is not specified.\n");
 684                 return -EINVAL;
 685         }
 686         if (isdigit(argv[1][0])) {
 687                 if (is_return) {
 688                         pr_info("Return probe point must be a symbol.\n");
 689                         return -EINVAL;
 690                 }
 691                 /* an address specified */
 692                 ret = strict_strtoul(&argv[0][2], 0, (unsigned long *)&addr);
 693                 if (ret) {
 694                         pr_info("Failed to parse address.\n");
 695                         return ret;
 696                 }
 697         } else {
 698                 /* a symbol specified */
 699                 symbol = argv[1];
 700                 /* TODO: support .init module functions */
 701                 ret = split_symbol_offset(symbol, &offset);
 702                 if (ret) {
 703                         pr_info("Failed to parse symbol.\n");
 704                         return ret;
 705                 }
 706                 if (offset && is_return) {
 707                         pr_info("Return probe must be used without offset.\n");
 708                         return -EINVAL;
 709                 }
 710         }
 711         argc -= 2; argv += 2;
 712
 713         /* setup a probe */
 714         if (!event) {
 715                 /* Make a new event name */
 716                 if (symbol)
 717                         snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
 718                                  is_return ? 'r' : 'p', symbol, offset);
 719                 else
 720                         snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
 721                                  is_return ? 'r' : 'p', addr);
 722                 event = buf;
 723         }
 724         tp = alloc_trace_probe(group, event, addr, symbol, offset, argc,
 725                                is_return);
 726         if (IS_ERR(tp)) {
 727                 pr_info("Failed to allocate trace_probe.(%d)\n",
 728                         (int)PTR_ERR(tp));
 729                 return PTR_ERR(tp);
 730         }
 731
 732         /* parse arguments */
 733         ret = 0;
 734         for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
 735                 /* Parse argument name */
 736                 arg = strchr(argv[i], '=');
 737                 if (arg)
 738                         *arg++ = '\0';
 739                 else
 740                         arg = argv[i];
 741
 742                 if (conflict_field_name(argv[i], tp->args, i)) {
 743                         pr_info("Argument%d name '%s' conflicts with "
 744                                 "another field.\n", i, argv[i]);
 745                         ret = -EINVAL;
 746                         goto error;
 747                 }
 748
 749                 tp->args[i].name = kstrdup(argv[i], GFP_KERNEL);
 750                 if (!tp->args[i].name) {
 751                         pr_info("Failed to allocate argument%d name '%s'.\n",
 752                                 i, argv[i]);
 753                         ret = -ENOMEM;
 754                         goto error;
 755                 }
 756
 757                 /* Parse fetch argument */
 758                 ret = parse_probe_arg(arg, &tp->args[i].fetch, is_return);
 759                 if (ret) {
 760                         pr_info("Parse error at argument%d. (%d)\n", i, ret);
 761                         kfree(tp->args[i].name);
 762                         goto error;
 763                 }
 764
 765                 tp->nr_args++;
 766         }
 767
 768         ret = register_trace_probe(tp);
 769         if (ret)
 770                 goto error;
 771         return 0;
 772
 773 error:
 774         free_trace_probe(tp);
 775         return ret;
 776 }
 777
 778 static void cleanup_all_probes(void)
 779 {
 780         struct trace_probe *tp;
 781
 782         mutex_lock(&probe_lock);
 783         /* TODO: Use batch unregistration */
 784         while (!list_empty(&probe_list)) {
 785                 tp = list_entry(probe_list.next, struct trace_probe, list);
 786                 unregister_trace_probe(tp);
 787                 free_trace_probe(tp);
 788         }
 789         mutex_unlock(&probe_lock);
 790 }
 791
 792
 793 /* Probes listing interfaces */
 794 static void *probes_seq_start(struct seq_file *m, loff_t *pos)
 795 {
 796         mutex_lock(&probe_lock);
 797         return seq_list_start(&probe_list, *pos);
 798 }
 799
 800 static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
 801 {
 802         return seq_list_next(v, &probe_list, pos);
 803 }
 804
 805 static void probes_seq_stop(struct seq_file *m, void *v)
 806 {
 807         mutex_unlock(&probe_lock);
 808 }
 809
 810 static int probes_seq_show(struct seq_file *m, void *v)
 811 {
 812         struct trace_probe *tp = v;
 813         int i, ret;
 814         char buf[MAX_ARGSTR_LEN + 1];
 815
 816         seq_printf(m, "%c", probe_is_return(tp) ? 'r' : 'p');
 817         seq_printf(m, ":%s/%s", tp->call.system, tp->call.name);
 818
 819         if (!tp->symbol)
 820                 seq_printf(m, " 0x%p", tp->rp.kp.addr);
 821         else if (tp->rp.kp.offset)
 822                 seq_printf(m, " %s+%u", probe_symbol(tp), tp->rp.kp.offset);
 823         else
 824                 seq_printf(m, " %s", probe_symbol(tp));
 825
 826         for (i = 0; i < tp->nr_args; i++) {
 827                 ret = probe_arg_string(buf, MAX_ARGSTR_LEN, &tp->args[i].fetch);
 828                 if (ret < 0) {
 829                         pr_warning("Argument%d decoding error(%d).\n", i, ret);
 830                         return ret;
 831                 }
 832                 seq_printf(m, " %s=%s", tp->args[i].name, buf);
 833         }
 834         seq_printf(m, "\n");
 835         return 0;
 836 }
 837
 838 static const struct seq_operations probes_seq_op = {
 839         .start  = probes_seq_start,
 840         .next   = probes_seq_next,
 841         .stop   = probes_seq_stop,
 842         .show   = probes_seq_show
 843 };
 844
 845 static int probes_open(struct inode *inode, struct file *file)
 846 {
 847         if ((file->f_mode & FMODE_WRITE) &&
 848             (file->f_flags & O_TRUNC))
 849                 cleanup_all_probes();
 850
 851         return seq_open(file, &probes_seq_op);
 852 }
 853
 854 static int command_trace_probe(const char *buf)
 855 {
 856         char **argv;
 857         int argc = 0, ret = 0;
 858
 859         argv = argv_split(GFP_KERNEL, buf, &argc);
 860         if (!argv)
 861                 return -ENOMEM;
 862
 863         if (argc)
 864                 ret = create_trace_probe(argc, argv);
 865
 866         argv_free(argv);
 867         return ret;
 868 }
 869
 870 #define WRITE_BUFSIZE 128
 871
 872 static ssize_t probes_write(struct file *file, const char __user *buffer,
 873                             size_t count, loff_t *ppos)
 874 {
 875         char *kbuf, *tmp;
 876         int ret;
 877         size_t done;
 878         size_t size;
 879
 880         kbuf = kmalloc(WRITE_BUFSIZE, GFP_KERNEL);
 881         if (!kbuf)
 882                 return -ENOMEM;
 883
 884         ret = done = 0;
 885         while (done < count) {
 886                 size = count - done;
 887                 if (size >= WRITE_BUFSIZE)
 888                         size = WRITE_BUFSIZE - 1;
 889                 if (copy_from_user(kbuf, buffer + done, size)) {
 890                         ret = -EFAULT;
 891                         goto out;
 892                 }
 893                 kbuf[size] = '\0';
 894                 tmp = strchr(kbuf, '\n');
 895                 if (tmp) {
 896                         *tmp = '\0';
 897                         size = tmp - kbuf + 1;
 898                 } else if (done + size < count) {
 899                         pr_warning("Line length is too long: "
 900                                    "Should be less than %d.", WRITE_BUFSIZE);
 901                         ret = -EINVAL;
 902                         goto out;
 903                 }
 904                 done += size;
 905                 /* Remove comments */
 906                 tmp = strchr(kbuf, '#');
 907                 if (tmp)
 908                         *tmp = '\0';
 909
 910                 ret = command_trace_probe(kbuf);
 911                 if (ret)
 912                         goto out;
 913         }
 914         ret = done;
 915 out:
 916         kfree(kbuf);
 917         return ret;
 918 }
 919
 920 static const struct file_operations kprobe_events_ops = {
 921         .owner          = THIS_MODULE,
 922         .open           = probes_open,
 923         .read           = seq_read,
 924         .llseek         = seq_lseek,
 925         .release        = seq_release,
 926         .write          = probes_write,
 927 };
 928
 929 /* Probes profiling interfaces */
 930 static int probes_profile_seq_show(struct seq_file *m, void *v)
 931 {
 932         struct trace_probe *tp = v;
 933
 934         seq_printf(m, "  %-44s %15lu %15lu\n", tp->call.name, tp->nhit,
 935                    tp->rp.kp.nmissed);
 936
 937         return 0;
 938 }
 939
 940 static const struct seq_operations profile_seq_op = {
 941         .start  = probes_seq_start,
 942         .next   = probes_seq_next,
 943         .stop   = probes_seq_stop,
 944         .show   = probes_profile_seq_show
 945 };
 946
 947 static int profile_open(struct inode *inode, struct file *file)
 948 {
 949         return seq_open(file, &profile_seq_op);
 950 }
 951
 952 static const struct file_operations kprobe_profile_ops = {
 953         .owner          = THIS_MODULE,
 954         .open           = profile_open,
 955         .read           = seq_read,
 956         .llseek         = seq_lseek,
 957         .release        = seq_release,
 958 };
 959
 960 /* Kprobe handler */
 961 static __kprobes int kprobe_trace_func(struct kprobe *kp, struct pt_regs *regs)
 962 {
 963         struct trace_probe *tp = container_of(kp, struct trace_probe, rp.kp);
 964         struct kprobe_trace_entry *entry;
 965         struct ring_buffer_event *event;
 966         struct ring_buffer *buffer;
 967         int size, i, pc;
 968         unsigned long irq_flags;
 969         struct ftrace_event_call *call = &tp->call;
 970
 971         tp->nhit++;
 972
 973         local_save_flags(irq_flags);
 974         pc = preempt_count();
 975
 976         size = SIZEOF_KPROBE_TRACE_ENTRY(tp->nr_args);
 977
 978         event = trace_current_buffer_lock_reserve(&buffer, call->id, size,
 979                                                   irq_flags, pc);
 980         if (!event)
 981                 return 0;
 982
 983         entry = ring_buffer_event_data(event);
 984         entry->nargs = tp->nr_args;
 985         entry->ip = (unsigned long)kp->addr;
 986         for (i = 0; i < tp->nr_args; i++)
 987                 entry->args[i] = call_fetch(&tp->args[i].fetch, regs);
 988
 989         if (!filter_current_check_discard(buffer, call, entry, event))
 990                 trace_nowake_buffer_unlock_commit(buffer, event, irq_flags, pc);
 991         return 0;
 992 }
 993
 994 /* Kretprobe handler */
 995 static __kprobes int kretprobe_trace_func(struct kretprobe_instance *ri,
 996                                           struct pt_regs *regs)
 997 {
 998         struct trace_probe *tp = container_of(ri->rp, struct trace_probe, rp);
 999         struct kretprobe_trace_entry *entry;
1000         struct ring_buffer_event *event;
1001         struct ring_buffer *buffer;
1002         int size, i, pc;
1003         unsigned long irq_flags;
1004         struct ftrace_event_call *call = &tp->call;
1005
1006         local_save_flags(irq_flags);
1007         pc = preempt_count();
1008
1009         size = SIZEOF_KRETPROBE_TRACE_ENTRY(tp->nr_args);
1010
1011         event = trace_current_buffer_lock_reserve(&buffer, call->id, size,
1012                                                   irq_flags, pc);
1013         if (!event)
1014                 return 0;
1015
1016         entry = ring_buffer_event_data(event);
1017         entry->nargs = tp->nr_args;
1018         entry->func = (unsigned long)tp->rp.kp.addr;
1019         entry->ret_ip = (unsigned long)ri->ret_addr;
1020         for (i = 0; i < tp->nr_args; i++)
1021                 entry->args[i] = call_fetch(&tp->args[i].fetch, regs);
1022
1023         if (!filter_current_check_discard(buffer, call, entry, event))
1024                 trace_nowake_buffer_unlock_commit(buffer, event, irq_flags, pc);
1025
1026         return 0;
1027 }
1028
1029 /* Event entry printers */
1030 enum print_line_t
1031 print_kprobe_event(struct trace_iterator *iter, int flags)
1032 {
1033         struct kprobe_trace_entry *field;
1034         struct trace_seq *s = &iter->seq;
1035         struct trace_event *event;
1036         struct trace_probe *tp;
1037         int i;
1038
1039         field = (struct kprobe_trace_entry *)iter->ent;
1040         event = ftrace_find_event(field->ent.type);
1041         tp = container_of(event, struct trace_probe, event);
1042
1043         if (!trace_seq_printf(s, "%s: (", tp->call.name))
1044                 goto partial;
1045
1046         if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
1047                 goto partial;
1048
1049         if (!trace_seq_puts(s, ")"))
1050                 goto partial;
1051
1052         for (i = 0; i < field->nargs; i++)
1053                 if (!trace_seq_printf(s, " %s=%lx",
1054                                       tp->args[i].name, field->args[i]))
1055                         goto partial;
1056
1057         if (!trace_seq_puts(s, "\n"))
1058                 goto partial;
1059
1060         return TRACE_TYPE_HANDLED;
1061 partial:
1062         return TRACE_TYPE_PARTIAL_LINE;
1063 }
1064
1065 enum print_line_t
1066 print_kretprobe_event(struct trace_iterator *iter, int flags)
1067 {
1068         struct kretprobe_trace_entry *field;
1069         struct trace_seq *s = &iter->seq;
1070         struct trace_event *event;
1071         struct trace_probe *tp;
1072         int i;
1073
1074         field = (struct kretprobe_trace_entry *)iter->ent;
1075         event = ftrace_find_event(field->ent.type);
1076         tp = container_of(event, struct trace_probe, event);
1077
1078         if (!trace_seq_printf(s, "%s: (", tp->call.name))
1079                 goto partial;
1080
1081         if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1082                 goto partial;
1083
1084         if (!trace_seq_puts(s, " <- "))
1085                 goto partial;
1086
1087         if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1088                 goto partial;
1089
1090         if (!trace_seq_puts(s, ")"))
1091                 goto partial;
1092
1093         for (i = 0; i < field->nargs; i++)
1094                 if (!trace_seq_printf(s, " %s=%lx",
1095                                       tp->args[i].name, field->args[i]))
1096                         goto partial;
1097
1098         if (!trace_seq_puts(s, "\n"))
1099                 goto partial;
1100
1101         return TRACE_TYPE_HANDLED;
1102 partial:
1103         return TRACE_TYPE_PARTIAL_LINE;
1104 }
1105
1106 static int probe_event_enable(struct ftrace_event_call *call)
1107 {
1108         struct trace_probe *tp = (struct trace_probe *)call->data;
1109
1110         tp->flags |= TP_FLAG_TRACE;
1111         if (probe_is_return(tp))
1112                 return enable_kretprobe(&tp->rp);
1113         else
1114                 return enable_kprobe(&tp->rp.kp);
1115 }
1116
1117 static void probe_event_disable(struct ftrace_event_call *call)
1118 {
1119         struct trace_probe *tp = (struct trace_probe *)call->data;
1120
1121         tp->flags &= ~TP_FLAG_TRACE;
1122         if (!(tp->flags & (TP_FLAG_TRACE | TP_FLAG_PROFILE))) {
1123                 if (probe_is_return(tp))
1124                         disable_kretprobe(&tp->rp);
1125                 else
1126                         disable_kprobe(&tp->rp.kp);
1127         }
1128 }
1129
1130 static int probe_event_raw_init(struct ftrace_event_call *event_call)
1131 {
1132         INIT_LIST_HEAD(&event_call->fields);
1133
1134         return 0;
1135 }
1136
1137 #undef DEFINE_FIELD
1138 #define DEFINE_FIELD(type, item, name, is_signed)                       \
1139         do {                                                            \
1140                 ret = trace_define_field(event_call, #type, name,       \
1141                                          offsetof(typeof(field), item), \
1142                                          sizeof(field.item), is_signed, \
1143                                          FILTER_OTHER);                 \
1144                 if (ret)                                                \
1145                         return ret;                                     \
1146         } while (0)
1147
1148 static int kprobe_event_define_fields(struct ftrace_event_call *event_call)
1149 {
1150         int ret, i;
1151         struct kprobe_trace_entry field;
1152         struct trace_probe *tp = (struct trace_probe *)event_call->data;
1153
1154         DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1155         DEFINE_FIELD(int, nargs, FIELD_STRING_NARGS, 1);
1156         /* Set argument names as fields */
1157         for (i = 0; i < tp->nr_args; i++)
1158                 DEFINE_FIELD(unsigned long, args[i], tp->args[i].name, 0);
1159         return 0;
1160 }
1161
1162 static int kretprobe_event_define_fields(struct ftrace_event_call *event_call)
1163 {
1164         int ret, i;
1165         struct kretprobe_trace_entry field;
1166         struct trace_probe *tp = (struct trace_probe *)event_call->data;
1167
1168         DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1169         DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1170         DEFINE_FIELD(int, nargs, FIELD_STRING_NARGS, 1);
1171         /* Set argument names as fields */
1172         for (i = 0; i < tp->nr_args; i++)
1173                 DEFINE_FIELD(unsigned long, args[i], tp->args[i].name, 0);
1174         return 0;
1175 }
1176
1177 static int __probe_event_show_format(struct trace_seq *s,
1178                                      struct trace_probe *tp, const char *fmt,
1179                                      const char *arg)
1180 {
1181         int i;
1182
1183         /* Show format */
1184         if (!trace_seq_printf(s, "\nprint fmt: \"%s", fmt))
1185                 return 0;
1186
1187         for (i = 0; i < tp->nr_args; i++)
1188                 if (!trace_seq_printf(s, " %s=%%lx", tp->args[i].name))
1189                         return 0;
1190
1191         if (!trace_seq_printf(s, "\", %s", arg))
1192                 return 0;
1193
1194         for (i = 0; i < tp->nr_args; i++)
1195                 if (!trace_seq_printf(s, ", REC->%s", tp->args[i].name))
1196                         return 0;
1197
1198         return trace_seq_puts(s, "\n");
1199 }
1200
1201 #undef SHOW_FIELD
1202 #define SHOW_FIELD(type, item, name)                                    \
1203         do {                                                            \
1204                 ret = trace_seq_printf(s, "\tfield: " #type " %s;\t"    \
1205                                 "offset:%u;\tsize:%u;\n", name,         \
1206                                 (unsigned int)offsetof(typeof(field), item),\
1207                                 (unsigned int)sizeof(type));            \
1208                 if (!ret)                                               \
1209                         return 0;                                       \
1210         } while (0)
1211
1212 static int kprobe_event_show_format(struct ftrace_event_call *call,
1213                                     struct trace_seq *s)
1214 {
1215         struct kprobe_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, ip, FIELD_STRING_IP);
1220         SHOW_FIELD(int, nargs, FIELD_STRING_NARGS);
1221
1222         /* Show fields */
1223         for (i = 0; i < tp->nr_args; i++)
1224                 SHOW_FIELD(unsigned long, args[i], tp->args[i].name);
1225         trace_seq_puts(s, "\n");
1226
1227         return __probe_event_show_format(s, tp, "(%lx)",
1228                                          "REC->" FIELD_STRING_IP);
1229 }
1230
1231 static int kretprobe_event_show_format(struct ftrace_event_call *call,
1232                                        struct trace_seq *s)
1233 {
1234         struct kretprobe_trace_entry field __attribute__((unused));
1235         int ret, i;
1236         struct trace_probe *tp = (struct trace_probe *)call->data;
1237
1238         SHOW_FIELD(unsigned long, func, FIELD_STRING_FUNC);
1239         SHOW_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP);
1240         SHOW_FIELD(int, nargs, FIELD_STRING_NARGS);
1241
1242         /* Show fields */
1243         for (i = 0; i < tp->nr_args; i++)
1244                 SHOW_FIELD(unsigned long, args[i], tp->args[i].name);
1245         trace_seq_puts(s, "\n");
1246
1247         return __probe_event_show_format(s, tp, "(%lx <- %lx)",
1248                                          "REC->" FIELD_STRING_FUNC
1249                                          ", REC->" FIELD_STRING_RETIP);
1250 }
1251
1252 #ifdef CONFIG_EVENT_PROFILE
1253
1254 /* Kprobe profile handler */
1255 static __kprobes int kprobe_profile_func(struct kprobe *kp,
1256                                          struct pt_regs *regs)
1257 {
1258         struct trace_probe *tp = container_of(kp, struct trace_probe, rp.kp);
1259         struct ftrace_event_call *call = &tp->call;
1260         struct kprobe_trace_entry *entry;
1261         struct trace_entry *ent;
1262         int size, __size, i, pc, __cpu;
1263         unsigned long irq_flags;
1264         char *trace_buf;
1265         char *raw_data;
1266         int rctx;
1267
1268         pc = preempt_count();
1269         __size = SIZEOF_KPROBE_TRACE_ENTRY(tp->nr_args);
1270         size = ALIGN(__size + sizeof(u32), sizeof(u64));
1271         size -= sizeof(u32);
1272         if (WARN_ONCE(size > FTRACE_MAX_PROFILE_SIZE,
1273                      "profile buffer not large enough"))
1274                 return 0;
1275
1276         /*
1277          * Protect the non nmi buffer
1278          * This also protects the rcu read side
1279          */
1280         local_irq_save(irq_flags);
1281
1282         rctx = perf_swevent_get_recursion_context();
1283         if (rctx < 0)
1284                 goto end_recursion;
1285
1286         __cpu = smp_processor_id();
1287
1288         if (in_nmi())
1289                 trace_buf = rcu_dereference(perf_trace_buf_nmi);
1290         else
1291                 trace_buf = rcu_dereference(perf_trace_buf);
1292
1293         if (!trace_buf)
1294                 goto end;
1295
1296         raw_data = per_cpu_ptr(trace_buf, __cpu);
1297
1298         /* Zero dead bytes from alignment to avoid buffer leak to userspace */
1299         *(u64 *)(&raw_data[size - sizeof(u64)]) = 0ULL;
1300         entry = (struct kprobe_trace_entry *)raw_data;
1301         ent = &entry->ent;
1302
1303         tracing_generic_entry_update(ent, irq_flags, pc);
1304         ent->type = call->id;
1305         entry->nargs = tp->nr_args;
1306         entry->ip = (unsigned long)kp->addr;
1307         for (i = 0; i < tp->nr_args; i++)
1308                 entry->args[i] = call_fetch(&tp->args[i].fetch, regs);
1309         perf_tp_event(call->id, entry->ip, 1, entry, size);
1310
1311 end:
1312         perf_swevent_put_recursion_context(rctx);
1313 end_recursion:
1314         local_irq_restore(irq_flags);
1315
1316         return 0;
1317 }
1318
1319 /* Kretprobe profile handler */
1320 static __kprobes int kretprobe_profile_func(struct kretprobe_instance *ri,
1321                                             struct pt_regs *regs)
1322 {
1323         struct trace_probe *tp = container_of(ri->rp, struct trace_probe, rp);
1324         struct ftrace_event_call *call = &tp->call;
1325         struct kretprobe_trace_entry *entry;
1326         struct trace_entry *ent;
1327         int size, __size, i, pc, __cpu;
1328         unsigned long irq_flags;
1329         char *trace_buf;
1330         char *raw_data;
1331         int rctx;
1332
1333         pc = preempt_count();
1334         __size = SIZEOF_KRETPROBE_TRACE_ENTRY(tp->nr_args);
1335         size = ALIGN(__size + sizeof(u32), sizeof(u64));
1336         size -= sizeof(u32);
1337         if (WARN_ONCE(size > FTRACE_MAX_PROFILE_SIZE,
1338                      "profile buffer not large enough"))
1339                 return 0;
1340
1341         /*
1342          * Protect the non nmi buffer
1343          * This also protects the rcu read side
1344          */
1345         local_irq_save(irq_flags);
1346
1347         rctx = perf_swevent_get_recursion_context();
1348         if (rctx < 0)
1349                 goto end_recursion;
1350
1351         __cpu = smp_processor_id();
1352
1353         if (in_nmi())
1354                 trace_buf = rcu_dereference(perf_trace_buf_nmi);
1355         else
1356                 trace_buf = rcu_dereference(perf_trace_buf);
1357
1358         if (!trace_buf)
1359                 goto end;
1360
1361         raw_data = per_cpu_ptr(trace_buf, __cpu);
1362
1363         /* Zero dead bytes from alignment to avoid buffer leak to userspace */
1364         *(u64 *)(&raw_data[size - sizeof(u64)]) = 0ULL;
1365         entry = (struct kretprobe_trace_entry *)raw_data;
1366         ent = &entry->ent;
1367
1368         tracing_generic_entry_update(ent, irq_flags, pc);
1369         ent->type = call->id;
1370         entry->nargs = tp->nr_args;
1371         entry->func = (unsigned long)tp->rp.kp.addr;
1372         entry->ret_ip = (unsigned long)ri->ret_addr;
1373         for (i = 0; i < tp->nr_args; i++)
1374                 entry->args[i] = call_fetch(&tp->args[i].fetch, regs);
1375         perf_tp_event(call->id, entry->ret_ip, 1, entry, size);
1376
1377 end:
1378         perf_swevent_put_recursion_context(rctx);
1379 end_recursion:
1380         local_irq_restore(irq_flags);
1381
1382         return 0;
1383 }
1384
1385 static int probe_profile_enable(struct ftrace_event_call *call)
1386 {
1387         struct trace_probe *tp = (struct trace_probe *)call->data;
1388
1389         tp->flags |= TP_FLAG_PROFILE;
1390
1391         if (probe_is_return(tp))
1392                 return enable_kretprobe(&tp->rp);
1393         else
1394                 return enable_kprobe(&tp->rp.kp);
1395 }
1396
1397 static void probe_profile_disable(struct ftrace_event_call *call)
1398 {
1399         struct trace_probe *tp = (struct trace_probe *)call->data;
1400
1401         tp->flags &= ~TP_FLAG_PROFILE;
1402
1403         if (!(tp->flags & TP_FLAG_TRACE)) {
1404                 if (probe_is_return(tp))
1405                         disable_kretprobe(&tp->rp);
1406                 else
1407                         disable_kprobe(&tp->rp.kp);
1408         }
1409 }
1410 #endif  /* CONFIG_EVENT_PROFILE */
1411
1412
1413 static __kprobes
1414 int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1415 {
1416         struct trace_probe *tp = container_of(kp, struct trace_probe, rp.kp);
1417
1418         if (tp->flags & TP_FLAG_TRACE)
1419                 kprobe_trace_func(kp, regs);
1420 #ifdef CONFIG_EVENT_PROFILE
1421         if (tp->flags & TP_FLAG_PROFILE)
1422                 kprobe_profile_func(kp, regs);
1423 #endif  /* CONFIG_EVENT_PROFILE */
1424         return 0;       /* We don't tweek kernel, so just return 0 */
1425 }
1426
1427 static __kprobes
1428 int kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1429 {
1430         struct trace_probe *tp = container_of(ri->rp, struct trace_probe, rp);
1431
1432         if (tp->flags & TP_FLAG_TRACE)
1433                 kretprobe_trace_func(ri, regs);
1434 #ifdef CONFIG_EVENT_PROFILE
1435         if (tp->flags & TP_FLAG_PROFILE)
1436                 kretprobe_profile_func(ri, regs);
1437 #endif  /* CONFIG_EVENT_PROFILE */
1438         return 0;       /* We don't tweek kernel, so just return 0 */
1439 }
1440
1441 static int register_probe_event(struct trace_probe *tp)
1442 {
1443         struct ftrace_event_call *call = &tp->call;
1444         int ret;
1445
1446         /* Initialize ftrace_event_call */
1447         if (probe_is_return(tp)) {
1448                 tp->event.trace = print_kretprobe_event;
1449                 call->raw_init = probe_event_raw_init;
1450                 call->show_format = kretprobe_event_show_format;
1451                 call->define_fields = kretprobe_event_define_fields;
1452         } else {
1453                 tp->event.trace = print_kprobe_event;
1454                 call->raw_init = probe_event_raw_init;
1455                 call->show_format = kprobe_event_show_format;
1456                 call->define_fields = kprobe_event_define_fields;
1457         }
1458         call->event = &tp->event;
1459         call->id = register_ftrace_event(&tp->event);
1460         if (!call->id)
1461                 return -ENODEV;
1462         call->enabled = 0;
1463         call->regfunc = probe_event_enable;
1464         call->unregfunc = probe_event_disable;
1465
1466 #ifdef CONFIG_EVENT_PROFILE
1467         call->profile_enable = probe_profile_enable;
1468         call->profile_disable = probe_profile_disable;
1469 #endif
1470         call->data = tp;
1471         ret = trace_add_event_call(call);
1472         if (ret) {
1473                 pr_info("Failed to register kprobe event: %s\n", call->name);
1474                 unregister_ftrace_event(&tp->event);
1475         }
1476         return ret;
1477 }
1478
1479 static void unregister_probe_event(struct trace_probe *tp)
1480 {
1481         /* tp->event is unregistered in trace_remove_event_call() */
1482         trace_remove_event_call(&tp->call);
1483 }
1484
1485 /* Make a debugfs interface for controling probe points */
1486 static __init int init_kprobe_trace(void)
1487 {
1488         struct dentry *d_tracer;
1489         struct dentry *entry;
1490
1491         d_tracer = tracing_init_dentry();
1492         if (!d_tracer)
1493                 return 0;
1494
1495         entry = debugfs_create_file("kprobe_events", 0644, d_tracer,
1496                                     NULL, &kprobe_events_ops);
1497
1498         /* Event list interface */
1499         if (!entry)
1500                 pr_warning("Could not create debugfs "
1501                            "'kprobe_events' entry\n");
1502
1503         /* Profile interface */
1504         entry = debugfs_create_file("kprobe_profile", 0444, d_tracer,
1505                                     NULL, &kprobe_profile_ops);
1506
1507         if (!entry)
1508                 pr_warning("Could not create debugfs "
1509                            "'kprobe_profile' entry\n");
1510         return 0;
1511 }
1512 fs_initcall(init_kprobe_trace);
1513
1514
1515 #ifdef CONFIG_FTRACE_STARTUP_TEST
1516
1517 static int kprobe_trace_selftest_target(int a1, int a2, int a3,
1518                                         int a4, int a5, int a6)
1519 {
1520         return a1 + a2 + a3 + a4 + a5 + a6;
1521 }
1522
1523 static __init int kprobe_trace_self_tests_init(void)
1524 {
1525         int ret;
1526         int (*target)(int, int, int, int, int, int);
1527
1528         target = kprobe_trace_selftest_target;
1529
1530         pr_info("Testing kprobe tracing: ");
1531
1532         ret = command_trace_probe("p:testprobe kprobe_trace_selftest_target "
1533                                   "$arg1 $arg2 $arg3 $arg4 $stack $stack0");
1534         if (WARN_ON_ONCE(ret))
1535                 pr_warning("error enabling function entry\n");
1536
1537         ret = command_trace_probe("r:testprobe2 kprobe_trace_selftest_target "
1538                                   "$retval");
1539         if (WARN_ON_ONCE(ret))
1540                 pr_warning("error enabling function return\n");
1541
1542         ret = target(1, 2, 3, 4, 5, 6);
1543
1544         cleanup_all_probes();
1545
1546         pr_cont("OK\n");
1547         return 0;
1548 }
1549
1550 late_initcall(kprobe_trace_self_tests_init);
1551
1552 #endif