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

   1 /*
   2  * kprobe based kernel tracer
   3  *
   4  * Created by Masami Hiramatsu <mhiramat@redhat.com>
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 as
   8  * published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful,
  11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  * GNU General Public License for more details.
  14  *
  15  * You should have received a copy of the GNU General Public License
  16  * along with this program; if not, write to the Free Software
  17  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
  18  */
  19
  20 #include <linux/module.h>
  21 #include <linux/uaccess.h>
  22 #include <linux/kprobes.h>
  23 #include <linux/seq_file.h>
  24 #include <linux/slab.h>
  25 #include <linux/smp.h>
  26 #include <linux/debugfs.h>
  27 #include <linux/types.h>
  28 #include <linux/string.h>
  29 #include <linux/ctype.h>
  30 #include <linux/ptrace.h>
  31 #include <linux/perf_counter.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 /* currently, trace_kprobe only supports X86. */
  42
  43 struct fetch_func {
  44         unsigned long (*func)(struct pt_regs *, void *);
  45         void *data;
  46 };
  47
  48 static __kprobes unsigned long call_fetch(struct fetch_func *f,
  49                                           struct pt_regs *regs)
  50 {
  51         return f->func(regs, f->data);
  52 }
  53
  54 /* fetch handlers */
  55 static __kprobes unsigned long fetch_register(struct pt_regs *regs,
  56                                               void *offset)
  57 {
  58         return regs_get_register(regs, (unsigned int)((unsigned long)offset));
  59 }
  60
  61 static __kprobes unsigned long fetch_stack(struct pt_regs *regs,
  62                                            void *num)
  63 {
  64         return regs_get_kernel_stack_nth(regs,
  65                                          (unsigned int)((unsigned long)num));
  66 }
  67
  68 static __kprobes unsigned long fetch_memory(struct pt_regs *regs, void *addr)
  69 {
  70         unsigned long retval;
  71
  72         if (probe_kernel_address(addr, retval))
  73                 return 0;
  74         return retval;
  75 }
  76
  77 static __kprobes unsigned long fetch_argument(struct pt_regs *regs, void *num)
  78 {
  79         return regs_get_argument_nth(regs, (unsigned int)((unsigned long)num));
  80 }
  81
  82 static __kprobes unsigned long fetch_retvalue(struct pt_regs *regs,
  83                                               void *dummy)
  84 {
  85         return regs_return_value(regs);
  86 }
  87
  88 static __kprobes unsigned long fetch_ip(struct pt_regs *regs, void *dummy)
  89 {
  90         return instruction_pointer(regs);
  91 }
  92
  93 static __kprobes unsigned long fetch_stack_address(struct pt_regs *regs,
  94                                                    void *dummy)
  95 {
  96         return kernel_stack_pointer(regs);
  97 }
  98
  99 /* Memory fetching by symbol */
 100 struct symbol_cache {
 101         char *symbol;
 102         long offset;
 103         unsigned long addr;
 104 };
 105
 106 static unsigned long update_symbol_cache(struct symbol_cache *sc)
 107 {
 108         sc->addr = (unsigned long)kallsyms_lookup_name(sc->symbol);
 109         if (sc->addr)
 110                 sc->addr += sc->offset;
 111         return sc->addr;
 112 }
 113
 114 static void free_symbol_cache(struct symbol_cache *sc)
 115 {
 116         kfree(sc->symbol);
 117         kfree(sc);
 118 }
 119
 120 static struct symbol_cache *alloc_symbol_cache(const char *sym, long offset)
 121 {
 122         struct symbol_cache *sc;
 123
 124         if (!sym || strlen(sym) == 0)
 125                 return NULL;
 126         sc = kzalloc(sizeof(struct symbol_cache), GFP_KERNEL);
 127         if (!sc)
 128                 return NULL;
 129
 130         sc->symbol = kstrdup(sym, GFP_KERNEL);
 131         if (!sc->symbol) {
 132                 kfree(sc);
 133                 return NULL;
 134         }
 135         sc->offset = offset;
 136
 137         update_symbol_cache(sc);
 138         return sc;
 139 }
 140
 141 static __kprobes unsigned long fetch_symbol(struct pt_regs *regs, void *data)
 142 {
 143         struct symbol_cache *sc = data;
 144
 145         if (sc->addr)
 146                 return fetch_memory(regs, (void *)sc->addr);
 147         else
 148                 return 0;
 149 }
 150
 151 /* Special indirect memory access interface */
 152 struct indirect_fetch_data {
 153         struct fetch_func orig;
 154         long offset;
 155 };
 156
 157 static __kprobes unsigned long fetch_indirect(struct pt_regs *regs, void *data)
 158 {
 159         struct indirect_fetch_data *ind = data;
 160         unsigned long addr;
 161
 162         addr = call_fetch(&ind->orig, regs);
 163         if (addr) {
 164                 addr += ind->offset;
 165                 return fetch_memory(regs, (void *)addr);
 166         } else
 167                 return 0;
 168 }
 169
 170 static __kprobes void free_indirect_fetch_data(struct indirect_fetch_data *data)
 171 {
 172         if (data->orig.func == fetch_indirect)
 173                 free_indirect_fetch_data(data->orig.data);
 174         else if (data->orig.func == fetch_symbol)
 175                 free_symbol_cache(data->orig.data);
 176         kfree(data);
 177 }
 178
 179 /**
 180  * Kprobe tracer core functions
 181  */
 182
 183 struct probe_arg {
 184         struct fetch_func       fetch;
 185         const char              *name;
 186 };
 187
 188 struct trace_probe {
 189         struct list_head        list;
 190         struct kretprobe        rp;     /* Use rp.kp for kprobe use */
 191         unsigned long           nhit;
 192         const char              *symbol;        /* symbol name */
 193         struct ftrace_event_call        call;
 194         struct trace_event              event;
 195         unsigned int            nr_args;
 196         struct probe_arg        args[];
 197 };
 198
 199 #define SIZEOF_TRACE_PROBE(n)                   \
 200         (offsetof(struct trace_probe, args) +   \
 201         (sizeof(struct probe_arg) * (n)))
 202
 203 static int kprobe_trace_func(struct kprobe *kp, struct pt_regs *regs);
 204 static int kretprobe_trace_func(struct kretprobe_instance *ri,
 205                                 struct pt_regs *regs);
 206
 207 static __kprobes int probe_is_return(struct trace_probe *tp)
 208 {
 209         return tp->rp.handler != NULL;
 210 }
 211
 212 static __kprobes const char *probe_symbol(struct trace_probe *tp)
 213 {
 214         return tp->symbol ? tp->symbol : "unknown";
 215 }
 216
 217 static int probe_arg_string(char *buf, size_t n, struct fetch_func *ff)
 218 {
 219         int ret = -EINVAL;
 220
 221         if (ff->func == fetch_argument)
 222                 ret = snprintf(buf, n, "a%lu", (unsigned long)ff->data);
 223         else if (ff->func == fetch_register) {
 224                 const char *name;
 225                 name = regs_query_register_name((unsigned int)((long)ff->data));
 226                 ret = snprintf(buf, n, "%%%s", name);
 227         } else if (ff->func == fetch_stack)
 228                 ret = snprintf(buf, n, "s%lu", (unsigned long)ff->data);
 229         else if (ff->func == fetch_memory)
 230                 ret = snprintf(buf, n, "@0x%p", ff->data);
 231         else if (ff->func == fetch_symbol) {
 232                 struct symbol_cache *sc = ff->data;
 233                 ret = snprintf(buf, n, "@%s%+ld", sc->symbol, sc->offset);
 234         } else if (ff->func == fetch_retvalue)
 235                 ret = snprintf(buf, n, "rv");
 236         else if (ff->func == fetch_ip)
 237                 ret = snprintf(buf, n, "ra");
 238         else if (ff->func == fetch_stack_address)
 239                 ret = snprintf(buf, n, "sa");
 240         else if (ff->func == fetch_indirect) {
 241                 struct indirect_fetch_data *id = ff->data;
 242                 size_t l = 0;
 243                 ret = snprintf(buf, n, "%+ld(", id->offset);
 244                 if (ret >= n)
 245                         goto end;
 246                 l += ret;
 247                 ret = probe_arg_string(buf + l, n - l, &id->orig);
 248                 if (ret < 0)
 249                         goto end;
 250                 l += ret;
 251                 ret = snprintf(buf + l, n - l, ")");
 252                 ret += l;
 253         }
 254 end:
 255         if (ret >= n)
 256                 return -ENOSPC;
 257         return ret;
 258 }
 259
 260 static int register_probe_event(struct trace_probe *tp);
 261 static void unregister_probe_event(struct trace_probe *tp);
 262
 263 static DEFINE_MUTEX(probe_lock);
 264 static LIST_HEAD(probe_list);
 265
 266 /*
 267  * Allocate new trace_probe and initialize it (including kprobes).
 268  */
 269 static struct trace_probe *alloc_trace_probe(const char *group,
 270                                              const char *event,
 271                                              void *addr,
 272                                              const char *symbol,
 273                                              unsigned long offs,
 274                                              int nargs, int is_return)
 275 {
 276         struct trace_probe *tp;
 277
 278         tp = kzalloc(SIZEOF_TRACE_PROBE(nargs), GFP_KERNEL);
 279         if (!tp)
 280                 return ERR_PTR(-ENOMEM);
 281
 282         if (symbol) {
 283                 tp->symbol = kstrdup(symbol, GFP_KERNEL);
 284                 if (!tp->symbol)
 285                         goto error;
 286                 tp->rp.kp.symbol_name = tp->symbol;
 287                 tp->rp.kp.offset = offs;
 288         } else
 289                 tp->rp.kp.addr = addr;
 290
 291         /* Set handler here for checking whether this probe is return or not. */
 292         if (is_return)
 293                 tp->rp.handler = kretprobe_trace_func;
 294         else
 295                 tp->rp.kp.pre_handler = kprobe_trace_func;
 296
 297         if (!event)
 298                 goto error;
 299         tp->call.name = kstrdup(event, GFP_KERNEL);
 300         if (!tp->call.name)
 301                 goto error;
 302
 303         if (!group)
 304                 goto error;
 305         tp->call.system = kstrdup(group, GFP_KERNEL);
 306         if (!tp->call.system)
 307                 goto error;
 308
 309         INIT_LIST_HEAD(&tp->list);
 310         return tp;
 311 error:
 312         kfree(tp->call.name);
 313         kfree(tp->symbol);
 314         kfree(tp);
 315         return ERR_PTR(-ENOMEM);
 316 }
 317
 318 static void free_probe_arg(struct probe_arg *arg)
 319 {
 320         if (arg->fetch.func == fetch_symbol)
 321                 free_symbol_cache(arg->fetch.data);
 322         else if (arg->fetch.func == fetch_indirect)
 323                 free_indirect_fetch_data(arg->fetch.data);
 324         kfree(arg->name);
 325 }
 326
 327 static void free_trace_probe(struct trace_probe *tp)
 328 {
 329         int i;
 330
 331         for (i = 0; i < tp->nr_args; i++)
 332                 free_probe_arg(&tp->args[i]);
 333
 334         kfree(tp->call.system);
 335         kfree(tp->call.name);
 336         kfree(tp->symbol);
 337         kfree(tp);
 338 }
 339
 340 static struct trace_probe *find_probe_event(const char *event)
 341 {
 342         struct trace_probe *tp;
 343
 344         list_for_each_entry(tp, &probe_list, list)
 345                 if (!strcmp(tp->call.name, event))
 346                         return tp;
 347         return NULL;
 348 }
 349
 350 static void __unregister_trace_probe(struct trace_probe *tp)
 351 {
 352         if (probe_is_return(tp))
 353                 unregister_kretprobe(&tp->rp);
 354         else
 355                 unregister_kprobe(&tp->rp.kp);
 356 }
 357
 358 /* Unregister a trace_probe and probe_event: call with locking probe_lock */
 359 static void unregister_trace_probe(struct trace_probe *tp)
 360 {
 361         unregister_probe_event(tp);
 362         __unregister_trace_probe(tp);
 363         list_del(&tp->list);
 364 }
 365
 366 /* Register a trace_probe and probe_event */
 367 static int register_trace_probe(struct trace_probe *tp)
 368 {
 369         struct trace_probe *old_tp;
 370         int ret;
 371
 372         mutex_lock(&probe_lock);
 373
 374         if (probe_is_return(tp))
 375                 ret = register_kretprobe(&tp->rp);
 376         else
 377                 ret = register_kprobe(&tp->rp.kp);
 378
 379         if (ret) {
 380                 pr_warning("Could not insert probe(%d)\n", ret);
 381                 if (ret == -EILSEQ) {
 382                         pr_warning("Probing address(0x%p) is not an "
 383                                    "instruction boundary.\n",
 384                                    tp->rp.kp.addr);
 385                         ret = -EINVAL;
 386                 }
 387                 goto end;
 388         }
 389         /* register as an event */
 390         old_tp = find_probe_event(tp->call.name);
 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                 __unregister_trace_probe(tp);
 400         }
 401         list_add_tail(&tp->list, &probe_list);
 402 end:
 403         mutex_unlock(&probe_lock);
 404         return ret;
 405 }
 406
 407 /* Split symbol and offset. */
 408 static int split_symbol_offset(char *symbol, unsigned long *offset)
 409 {
 410         char *tmp;
 411         int ret;
 412
 413         if (!offset)
 414                 return -EINVAL;
 415
 416         tmp = strchr(symbol, '+');
 417         if (tmp) {
 418                 /* skip sign because strict_strtol doesn't accept '+' */
 419                 ret = strict_strtoul(tmp + 1, 0, offset);
 420                 if (ret)
 421                         return ret;
 422                 *tmp = '\0';
 423         } else
 424                 *offset = 0;
 425         return 0;
 426 }
 427
 428 #define PARAM_MAX_ARGS 16
 429 #define PARAM_MAX_STACK (THREAD_SIZE / sizeof(unsigned long))
 430
 431 static int parse_probe_arg(char *arg, struct fetch_func *ff, int is_return)
 432 {
 433         int ret = 0;
 434         unsigned long param;
 435         long offset;
 436         char *tmp;
 437
 438         switch (arg[0]) {
 439         case 'a':       /* argument */
 440                 ret = strict_strtoul(arg + 1, 10, &param);
 441                 if (ret || param > PARAM_MAX_ARGS)
 442                         ret = -EINVAL;
 443                 else {
 444                         ff->func = fetch_argument;
 445                         ff->data = (void *)param;
 446                 }
 447                 break;
 448         case 'r':       /* retval or retaddr */
 449                 if (is_return && arg[1] == 'v') {
 450                         ff->func = fetch_retvalue;
 451                         ff->data = NULL;
 452                 } else if (is_return && arg[1] == 'a') {
 453                         ff->func = fetch_ip;
 454                         ff->data = NULL;
 455                 } else
 456                         ret = -EINVAL;
 457                 break;
 458         case '%':       /* named register */
 459                 ret = regs_query_register_offset(arg + 1);
 460                 if (ret >= 0) {
 461                         ff->func = fetch_register;
 462                         ff->data = (void *)(unsigned long)ret;
 463                         ret = 0;
 464                 }
 465                 break;
 466         case 's':       /* stack */
 467                 if (arg[1] == 'a') {
 468                         ff->func = fetch_stack_address;
 469                         ff->data = NULL;
 470                 } else {
 471                         ret = strict_strtoul(arg + 1, 10, &param);
 472                         if (ret || param > PARAM_MAX_STACK)
 473                                 ret = -EINVAL;
 474                         else {
 475                                 ff->func = fetch_stack;
 476                                 ff->data = (void *)param;
 477                         }
 478                 }
 479                 break;
 480         case '@':       /* memory or symbol */
 481                 if (isdigit(arg[1])) {
 482                         ret = strict_strtoul(arg + 1, 0, &param);
 483                         if (ret)
 484                                 break;
 485                         ff->func = fetch_memory;
 486                         ff->data = (void *)param;
 487                 } else {
 488                         ret = split_symbol_offset(arg + 1, &offset);
 489                         if (ret)
 490                                 break;
 491                         ff->data = alloc_symbol_cache(arg + 1,
 492                                                               offset);
 493                         if (ff->data)
 494                                 ff->func = fetch_symbol;
 495                         else
 496                                 ret = -EINVAL;
 497                 }
 498                 break;
 499         case '+':       /* indirect memory */
 500         case '-':
 501                 tmp = strchr(arg, '(');
 502                 if (!tmp) {
 503                         ret = -EINVAL;
 504                         break;
 505                 }
 506                 *tmp = '\0';
 507                 ret = strict_strtol(arg + 1, 0, &offset);
 508                 if (ret)
 509                         break;
 510                 if (arg[0] == '-')
 511                         offset = -offset;
 512                 arg = tmp + 1;
 513                 tmp = strrchr(arg, ')');
 514                 if (tmp) {
 515                         struct indirect_fetch_data *id;
 516                         *tmp = '\0';
 517                         id = kzalloc(sizeof(struct indirect_fetch_data),
 518                                      GFP_KERNEL);
 519                         if (!id)
 520                                 return -ENOMEM;
 521                         id->offset = offset;
 522                         ret = parse_probe_arg(arg, &id->orig, is_return);
 523                         if (ret)
 524                                 kfree(id);
 525                         else {
 526                                 ff->func = fetch_indirect;
 527                                 ff->data = (void *)id;
 528                         }
 529                 } else
 530                         ret = -EINVAL;
 531                 break;
 532         default:
 533                 /* TODO: support custom handler */
 534                 ret = -EINVAL;
 535         }
 536         return ret;
 537 }
 538
 539 static int create_trace_probe(int argc, char **argv)
 540 {
 541         /*
 542          * Argument syntax:
 543          *  - Add kprobe: p[:[GRP/]EVENT] KSYM[+OFFS]|KADDR [FETCHARGS]
 544          *  - Add kretprobe: r[:[GRP/]EVENT] KSYM[+0] [FETCHARGS]
 545          * Fetch args:
 546          *  aN  : fetch Nth of function argument. (N:0-)
 547          *  rv  : fetch return value
 548          *  ra  : fetch return address
 549          *  sa  : fetch stack address
 550          *  sN  : fetch Nth of stack (N:0-)
 551          *  @ADDR       : fetch memory at ADDR (ADDR should be in kernel)
 552          *  @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
 553          *  %REG        : fetch register REG
 554          * Indirect memory fetch:
 555          *  +|-offs(ARG) : fetch memory at ARG +|- offs address.
 556          * Alias name of args:
 557          *  NAME=FETCHARG : set NAME as alias of FETCHARG.
 558          */
 559         struct trace_probe *tp;
 560         int i, ret = 0;
 561         int is_return = 0;
 562         char *symbol = NULL, *event = NULL, *arg = NULL, *group = NULL;
 563         unsigned long offset = 0;
 564         void *addr = NULL;
 565         char buf[MAX_EVENT_NAME_LEN];
 566
 567         if (argc < 2)
 568                 return -EINVAL;
 569
 570         if (argv[0][0] == 'p')
 571                 is_return = 0;
 572         else if (argv[0][0] == 'r')
 573                 is_return = 1;
 574         else
 575                 return -EINVAL;
 576
 577         if (argv[0][1] == ':') {
 578                 event = &argv[0][2];
 579                 if (strchr(event, '/')) {
 580                         group = event;
 581                         event = strchr(group, '/') + 1;
 582                         event[-1] = '\0';
 583                         if (strlen(group) == 0) {
 584                                 pr_info("Group name is not specifiled\n");
 585                                 return -EINVAL;
 586                         }
 587                 }
 588                 if (strlen(event) == 0) {
 589                         pr_info("Event name is not specifiled\n");
 590                         return -EINVAL;
 591                 }
 592         }
 593
 594         if (isdigit(argv[1][0])) {
 595                 if (is_return)
 596                         return -EINVAL;
 597                 /* an address specified */
 598                 ret = strict_strtoul(&argv[0][2], 0, (unsigned long *)&addr);
 599                 if (ret)
 600                         return ret;
 601         } else {
 602                 /* a symbol specified */
 603                 symbol = argv[1];
 604                 /* TODO: support .init module functions */
 605                 ret = split_symbol_offset(symbol, &offset);
 606                 if (ret)
 607                         return ret;
 608                 if (offset && is_return)
 609                         return -EINVAL;
 610         }
 611         argc -= 2; argv += 2;
 612
 613         /* setup a probe */
 614         if (!group)
 615                 group = KPROBE_EVENT_SYSTEM;
 616         if (!event) {
 617                 /* Make a new event name */
 618                 if (symbol)
 619                         snprintf(buf, MAX_EVENT_NAME_LEN, "%c@%s%+ld",
 620                                  is_return ? 'r' : 'p', symbol, offset);
 621                 else
 622                         snprintf(buf, MAX_EVENT_NAME_LEN, "%c@0x%p",
 623                                  is_return ? 'r' : 'p', addr);
 624                 event = buf;
 625         }
 626         tp = alloc_trace_probe(group, event, addr, symbol, offset, argc,
 627                                is_return);
 628         if (IS_ERR(tp))
 629                 return PTR_ERR(tp);
 630
 631         /* parse arguments */
 632         ret = 0;
 633         for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
 634                 /* Parse argument name */
 635                 arg = strchr(argv[i], '=');
 636                 if (arg)
 637                         *arg++ = '\0';
 638                 else
 639                         arg = argv[i];
 640                 tp->args[i].name = kstrdup(argv[i], GFP_KERNEL);
 641
 642                 /* Parse fetch argument */
 643                 if (strlen(arg) > MAX_ARGSTR_LEN) {
 644                         pr_info("Argument%d(%s) is too long.\n", i, arg);
 645                         ret = -ENOSPC;
 646                         goto error;
 647                 }
 648                 ret = parse_probe_arg(arg, &tp->args[i].fetch, is_return);
 649                 if (ret)
 650                         goto error;
 651         }
 652         tp->nr_args = i;
 653
 654         ret = register_trace_probe(tp);
 655         if (ret)
 656                 goto error;
 657         return 0;
 658
 659 error:
 660         free_trace_probe(tp);
 661         return ret;
 662 }
 663
 664 static void cleanup_all_probes(void)
 665 {
 666         struct trace_probe *tp;
 667
 668         mutex_lock(&probe_lock);
 669         /* TODO: Use batch unregistration */
 670         while (!list_empty(&probe_list)) {
 671                 tp = list_entry(probe_list.next, struct trace_probe, list);
 672                 unregister_trace_probe(tp);
 673                 free_trace_probe(tp);
 674         }
 675         mutex_unlock(&probe_lock);
 676 }
 677
 678
 679 /* Probes listing interfaces */
 680 static void *probes_seq_start(struct seq_file *m, loff_t *pos)
 681 {
 682         mutex_lock(&probe_lock);
 683         return seq_list_start(&probe_list, *pos);
 684 }
 685
 686 static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
 687 {
 688         return seq_list_next(v, &probe_list, pos);
 689 }
 690
 691 static void probes_seq_stop(struct seq_file *m, void *v)
 692 {
 693         mutex_unlock(&probe_lock);
 694 }
 695
 696 static int probes_seq_show(struct seq_file *m, void *v)
 697 {
 698         struct trace_probe *tp = v;
 699         int i, ret;
 700         char buf[MAX_ARGSTR_LEN + 1];
 701
 702         seq_printf(m, "%c", probe_is_return(tp) ? 'r' : 'p');
 703         seq_printf(m, ":%s", tp->call.name);
 704
 705         if (tp->symbol)
 706                 seq_printf(m, " %s+%u", probe_symbol(tp), tp->rp.kp.offset);
 707         else
 708                 seq_printf(m, " 0x%p", tp->rp.kp.addr);
 709
 710         for (i = 0; i < tp->nr_args; i++) {
 711                 ret = probe_arg_string(buf, MAX_ARGSTR_LEN, &tp->args[i].fetch);
 712                 if (ret < 0) {
 713                         pr_warning("Argument%d decoding error(%d).\n", i, ret);
 714                         return ret;
 715                 }
 716                 seq_printf(m, " %s=%s", tp->args[i].name, buf);
 717         }
 718         seq_printf(m, "\n");
 719         return 0;
 720 }
 721
 722 static const struct seq_operations probes_seq_op = {
 723         .start  = probes_seq_start,
 724         .next   = probes_seq_next,
 725         .stop   = probes_seq_stop,
 726         .show   = probes_seq_show
 727 };
 728
 729 static int probes_open(struct inode *inode, struct file *file)
 730 {
 731         if ((file->f_mode & FMODE_WRITE) &&
 732             (file->f_flags & O_TRUNC))
 733                 cleanup_all_probes();
 734
 735         return seq_open(file, &probes_seq_op);
 736 }
 737
 738 static int command_trace_probe(const char *buf)
 739 {
 740         char **argv;
 741         int argc = 0, ret = 0;
 742
 743         argv = argv_split(GFP_KERNEL, buf, &argc);
 744         if (!argv)
 745                 return -ENOMEM;
 746
 747         if (argc)
 748                 ret = create_trace_probe(argc, argv);
 749
 750         argv_free(argv);
 751         return ret;
 752 }
 753
 754 #define WRITE_BUFSIZE 128
 755
 756 static ssize_t probes_write(struct file *file, const char __user *buffer,
 757                             size_t count, loff_t *ppos)
 758 {
 759         char *kbuf, *tmp;
 760         int ret;
 761         size_t done;
 762         size_t size;
 763
 764         kbuf = kmalloc(WRITE_BUFSIZE, GFP_KERNEL);
 765         if (!kbuf)
 766                 return -ENOMEM;
 767
 768         ret = done = 0;
 769         while (done < count) {
 770                 size = count - done;
 771                 if (size >= WRITE_BUFSIZE)
 772                         size = WRITE_BUFSIZE - 1;
 773                 if (copy_from_user(kbuf, buffer + done, size)) {
 774                         ret = -EFAULT;
 775                         goto out;
 776                 }
 777                 kbuf[size] = '\0';
 778                 tmp = strchr(kbuf, '\n');
 779                 if (tmp) {
 780                         *tmp = '\0';
 781                         size = tmp - kbuf + 1;
 782                 } else if (done + size < count) {
 783                         pr_warning("Line length is too long: "
 784                                    "Should be less than %d.", WRITE_BUFSIZE);
 785                         ret = -EINVAL;
 786                         goto out;
 787                 }
 788                 done += size;
 789                 /* Remove comments */
 790                 tmp = strchr(kbuf, '#');
 791                 if (tmp)
 792                         *tmp = '\0';
 793
 794                 ret = command_trace_probe(kbuf);
 795                 if (ret)
 796                         goto out;
 797         }
 798         ret = done;
 799 out:
 800         kfree(kbuf);
 801         return ret;
 802 }
 803
 804 static const struct file_operations kprobe_events_ops = {
 805         .owner          = THIS_MODULE,
 806         .open           = probes_open,
 807         .read           = seq_read,
 808         .llseek         = seq_lseek,
 809         .release        = seq_release,
 810         .write          = probes_write,
 811 };
 812
 813 /* Probes profiling interfaces */
 814 static int probes_profile_seq_show(struct seq_file *m, void *v)
 815 {
 816         struct trace_probe *tp = v;
 817
 818         seq_printf(m, "  %-44s %15lu %15lu\n", tp->call.name, tp->nhit,
 819                    tp->rp.kp.nmissed);
 820
 821         return 0;
 822 }
 823
 824 static const struct seq_operations profile_seq_op = {
 825         .start  = probes_seq_start,
 826         .next   = probes_seq_next,
 827         .stop   = probes_seq_stop,
 828         .show   = probes_profile_seq_show
 829 };
 830
 831 static int profile_open(struct inode *inode, struct file *file)
 832 {
 833         return seq_open(file, &profile_seq_op);
 834 }
 835
 836 static const struct file_operations kprobe_profile_ops = {
 837         .owner          = THIS_MODULE,
 838         .open           = profile_open,
 839         .read           = seq_read,
 840         .llseek         = seq_lseek,
 841         .release        = seq_release,
 842 };
 843
 844 /* Kprobe handler */
 845 static __kprobes int kprobe_trace_func(struct kprobe *kp, struct pt_regs *regs)
 846 {
 847         struct trace_probe *tp = container_of(kp, struct trace_probe, rp.kp);
 848         struct kprobe_trace_entry *entry;
 849         struct ring_buffer_event *event;
 850         struct ring_buffer *buffer;
 851         int size, i, pc;
 852         unsigned long irq_flags;
 853         struct ftrace_event_call *call = &tp->call;
 854
 855         tp->nhit++;
 856
 857         local_save_flags(irq_flags);
 858         pc = preempt_count();
 859
 860         size = SIZEOF_KPROBE_TRACE_ENTRY(tp->nr_args);
 861
 862         event = trace_current_buffer_lock_reserve(&buffer, call->id, size,
 863                                                   irq_flags, pc);
 864         if (!event)
 865                 return 0;
 866
 867         entry = ring_buffer_event_data(event);
 868         entry->nargs = tp->nr_args;
 869         entry->ip = (unsigned long)kp->addr;
 870         for (i = 0; i < tp->nr_args; i++)
 871                 entry->args[i] = call_fetch(&tp->args[i].fetch, regs);
 872
 873         if (!filter_current_check_discard(buffer, call, entry, event))
 874                 trace_nowake_buffer_unlock_commit(buffer, event, irq_flags, pc);
 875         return 0;
 876 }
 877
 878 /* Kretprobe handler */
 879 static __kprobes int kretprobe_trace_func(struct kretprobe_instance *ri,
 880                                           struct pt_regs *regs)
 881 {
 882         struct trace_probe *tp = container_of(ri->rp, struct trace_probe, rp);
 883         struct kretprobe_trace_entry *entry;
 884         struct ring_buffer_event *event;
 885         struct ring_buffer *buffer;
 886         int size, i, pc;
 887         unsigned long irq_flags;
 888         struct ftrace_event_call *call = &tp->call;
 889
 890         local_save_flags(irq_flags);
 891         pc = preempt_count();
 892
 893         size = SIZEOF_KRETPROBE_TRACE_ENTRY(tp->nr_args);
 894
 895         event = trace_current_buffer_lock_reserve(&buffer, call->id, size,
 896                                                   irq_flags, pc);
 897         if (!event)
 898                 return 0;
 899
 900         entry = ring_buffer_event_data(event);
 901         entry->nargs = tp->nr_args;
 902         entry->func = (unsigned long)tp->rp.kp.addr;
 903         entry->ret_ip = (unsigned long)ri->ret_addr;
 904         for (i = 0; i < tp->nr_args; i++)
 905                 entry->args[i] = call_fetch(&tp->args[i].fetch, regs);
 906
 907         if (!filter_current_check_discard(buffer, call, entry, event))
 908                 trace_nowake_buffer_unlock_commit(buffer, event, irq_flags, pc);
 909
 910         return 0;
 911 }
 912
 913 /* Event entry printers */
 914 enum print_line_t
 915 print_kprobe_event(struct trace_iterator *iter, int flags)
 916 {
 917         struct kprobe_trace_entry *field;
 918         struct trace_seq *s = &iter->seq;
 919         struct trace_event *event;
 920         struct trace_probe *tp;
 921         int i;
 922
 923         field = (struct kprobe_trace_entry *)iter->ent;
 924         event = ftrace_find_event(field->ent.type);
 925         tp = container_of(event, struct trace_probe, event);
 926
 927         if (!trace_seq_printf(s, "%s: (", tp->call.name))
 928                 goto partial;
 929
 930         if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
 931                 goto partial;
 932
 933         if (!trace_seq_puts(s, ")"))
 934                 goto partial;
 935
 936         for (i = 0; i < field->nargs; i++)
 937                 if (!trace_seq_printf(s, " %s=%lx",
 938                                       tp->args[i].name, field->args[i]))
 939                         goto partial;
 940
 941         if (!trace_seq_puts(s, "\n"))
 942                 goto partial;
 943
 944         return TRACE_TYPE_HANDLED;
 945 partial:
 946         return TRACE_TYPE_PARTIAL_LINE;
 947 }
 948
 949 enum print_line_t
 950 print_kretprobe_event(struct trace_iterator *iter, int flags)
 951 {
 952         struct kretprobe_trace_entry *field;
 953         struct trace_seq *s = &iter->seq;
 954         struct trace_event *event;
 955         struct trace_probe *tp;
 956         int i;
 957
 958         field = (struct kretprobe_trace_entry *)iter->ent;
 959         event = ftrace_find_event(field->ent.type);
 960         tp = container_of(event, struct trace_probe, event);
 961
 962         if (!trace_seq_printf(s, "%s: (", tp->call.name))
 963                 goto partial;
 964
 965         if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
 966                 goto partial;
 967
 968         if (!trace_seq_puts(s, " <- "))
 969                 goto partial;
 970
 971         if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
 972                 goto partial;
 973
 974         if (!trace_seq_puts(s, ")"))
 975                 goto partial;
 976
 977         for (i = 0; i < field->nargs; i++)
 978                 if (!trace_seq_printf(s, " %s=%lx",
 979                                       tp->args[i].name, field->args[i]))
 980                         goto partial;
 981
 982         if (!trace_seq_puts(s, "\n"))
 983                 goto partial;
 984
 985         return TRACE_TYPE_HANDLED;
 986 partial:
 987         return TRACE_TYPE_PARTIAL_LINE;
 988 }
 989
 990 static int probe_event_enable(struct ftrace_event_call *call)
 991 {
 992         struct trace_probe *tp = (struct trace_probe *)call->data;
 993
 994         if (probe_is_return(tp)) {
 995                 tp->rp.handler = kretprobe_trace_func;
 996                 return enable_kretprobe(&tp->rp);
 997         } else {
 998                 tp->rp.kp.pre_handler = kprobe_trace_func;
 999                 return enable_kprobe(&tp->rp.kp);
1000         }
1001 }
1002
1003 static void probe_event_disable(struct ftrace_event_call *call)
1004 {
1005         struct trace_probe *tp = (struct trace_probe *)call->data;
1006
1007         if (probe_is_return(tp))
1008                 disable_kretprobe(&tp->rp);
1009         else
1010                 disable_kprobe(&tp->rp.kp);
1011 }
1012
1013 static int probe_event_raw_init(struct ftrace_event_call *event_call)
1014 {
1015         INIT_LIST_HEAD(&event_call->fields);
1016
1017         return 0;
1018 }
1019
1020 #undef DEFINE_FIELD
1021 #define DEFINE_FIELD(type, item, name, is_signed)                       \
1022         do {                                                            \
1023                 ret = trace_define_field(event_call, #type, name,       \
1024                                          offsetof(typeof(field), item), \
1025                                          sizeof(field.item), is_signed, \
1026                                          FILTER_OTHER);                 \
1027                 if (ret)                                                \
1028                         return ret;                                     \
1029         } while (0)
1030
1031 static int kprobe_event_define_fields(struct ftrace_event_call *event_call)
1032 {
1033         int ret, i;
1034         struct kprobe_trace_entry field;
1035         struct trace_probe *tp = (struct trace_probe *)event_call->data;
1036
1037         ret = trace_define_common_fields(event_call);
1038         if (!ret)
1039                 return ret;
1040
1041         DEFINE_FIELD(unsigned long, ip, "ip", 0);
1042         DEFINE_FIELD(int, nargs, "nargs", 1);
1043         /* Set argument names as fields */
1044         for (i = 0; i < tp->nr_args; i++)
1045                 DEFINE_FIELD(unsigned long, args[i], tp->args[i].name, 0);
1046         return 0;
1047 }
1048
1049 static int kretprobe_event_define_fields(struct ftrace_event_call *event_call)
1050 {
1051         int ret, i;
1052         struct kretprobe_trace_entry field;
1053         struct trace_probe *tp = (struct trace_probe *)event_call->data;
1054
1055         ret = trace_define_common_fields(event_call);
1056         if (!ret)
1057                 return ret;
1058
1059         DEFINE_FIELD(unsigned long, func, "func", 0);
1060         DEFINE_FIELD(unsigned long, ret_ip, "ret_ip", 0);
1061         DEFINE_FIELD(int, nargs, "nargs", 1);
1062         /* Set argument names as fields */
1063         for (i = 0; i < tp->nr_args; i++)
1064                 DEFINE_FIELD(unsigned long, args[i], tp->args[i].name, 0);
1065         return 0;
1066 }
1067
1068 static int __probe_event_show_format(struct trace_seq *s,
1069                                      struct trace_probe *tp, const char *fmt,
1070                                      const char *arg)
1071 {
1072         int i;
1073
1074         /* Show format */
1075         if (!trace_seq_printf(s, "\nprint fmt: \"%s", fmt))
1076                 return 0;
1077
1078         for (i = 0; i < tp->nr_args; i++)
1079                 if (!trace_seq_printf(s, " %s=%%lx", tp->args[i].name))
1080                         return 0;
1081
1082         if (!trace_seq_printf(s, "\", %s", arg))
1083                 return 0;
1084
1085         for (i = 0; i < tp->nr_args; i++)
1086                 if (!trace_seq_printf(s, ", REC->%s", tp->args[i].name))
1087                         return 0;
1088
1089         return trace_seq_puts(s, "\n");
1090 }
1091
1092 #undef SHOW_FIELD
1093 #define SHOW_FIELD(type, item, name)                                    \
1094         do {                                                            \
1095                 ret = trace_seq_printf(s, "\tfield: " #type " %s;\t"    \
1096                                 "offset:%u;\tsize:%u;\n", name,         \
1097                                 (unsigned int)offsetof(typeof(field), item),\
1098                                 (unsigned int)sizeof(type));            \
1099                 if (!ret)                                               \
1100                         return 0;                                       \
1101         } while (0)
1102
1103 static int kprobe_event_show_format(struct ftrace_event_call *call,
1104                                     struct trace_seq *s)
1105 {
1106         struct kprobe_trace_entry field __attribute__((unused));
1107         int ret, i;
1108         struct trace_probe *tp = (struct trace_probe *)call->data;
1109
1110         SHOW_FIELD(unsigned long, ip, "ip");
1111         SHOW_FIELD(int, nargs, "nargs");
1112
1113         /* Show fields */
1114         for (i = 0; i < tp->nr_args; i++)
1115                 SHOW_FIELD(unsigned long, args[i], tp->args[i].name);
1116         trace_seq_puts(s, "\n");
1117
1118         return __probe_event_show_format(s, tp, "(%lx)", "REC->ip");
1119 }
1120
1121 static int kretprobe_event_show_format(struct ftrace_event_call *call,
1122                                        struct trace_seq *s)
1123 {
1124         struct kretprobe_trace_entry field __attribute__((unused));
1125         int ret, i;
1126         struct trace_probe *tp = (struct trace_probe *)call->data;
1127
1128         SHOW_FIELD(unsigned long, func, "func");
1129         SHOW_FIELD(unsigned long, ret_ip, "ret_ip");
1130         SHOW_FIELD(int, nargs, "nargs");
1131
1132         /* Show fields */
1133         for (i = 0; i < tp->nr_args; i++)
1134                 SHOW_FIELD(unsigned long, args[i], tp->args[i].name);
1135         trace_seq_puts(s, "\n");
1136
1137         return __probe_event_show_format(s, tp, "(%lx <- %lx)",
1138                                           "REC->func, REC->ret_ip");
1139 }
1140
1141 #ifdef CONFIG_EVENT_PROFILE
1142
1143 /* Kprobe profile handler */
1144 static __kprobes int kprobe_profile_func(struct kprobe *kp,
1145                                          struct pt_regs *regs)
1146 {
1147         struct trace_probe *tp = container_of(kp, struct trace_probe, rp.kp);
1148         struct ftrace_event_call *call = &tp->call;
1149         struct kprobe_trace_entry *entry;
1150         int size, i, pc;
1151         unsigned long irq_flags;
1152
1153         local_save_flags(irq_flags);
1154         pc = preempt_count();
1155
1156         size = SIZEOF_KPROBE_TRACE_ENTRY(tp->nr_args);
1157
1158         do {
1159                 char raw_data[size];
1160                 struct trace_entry *ent;
1161
1162                 *(u64 *)(&raw_data[size - sizeof(u64)]) = 0ULL;
1163                 entry = (struct kprobe_trace_entry *)raw_data;
1164                 ent = &entry->ent;
1165
1166                 tracing_generic_entry_update(ent, irq_flags, pc);
1167                 ent->type = call->id;
1168                 entry->nargs = tp->nr_args;
1169                 entry->ip = (unsigned long)kp->addr;
1170                 for (i = 0; i < tp->nr_args; i++)
1171                         entry->args[i] = call_fetch(&tp->args[i].fetch, regs);
1172                 perf_tpcounter_event(call->id, entry->ip, 1, entry, size);
1173         } while (0);
1174         return 0;
1175 }
1176
1177 /* Kretprobe profile handler */
1178 static __kprobes int kretprobe_profile_func(struct kretprobe_instance *ri,
1179                                             struct pt_regs *regs)
1180 {
1181         struct trace_probe *tp = container_of(ri->rp, struct trace_probe, rp);
1182         struct ftrace_event_call *call = &tp->call;
1183         struct kretprobe_trace_entry *entry;
1184         int size, i, pc;
1185         unsigned long irq_flags;
1186
1187         local_save_flags(irq_flags);
1188         pc = preempt_count();
1189
1190         size = SIZEOF_KRETPROBE_TRACE_ENTRY(tp->nr_args);
1191
1192         do {
1193                 char raw_data[size];
1194                 struct trace_entry *ent;
1195
1196                 *(u64 *)(&raw_data[size - sizeof(u64)]) = 0ULL;
1197                 entry = (struct kretprobe_trace_entry *)raw_data;
1198                 ent = &entry->ent;
1199
1200                 tracing_generic_entry_update(ent, irq_flags, pc);
1201                 ent->type = call->id;
1202                 entry->nargs = tp->nr_args;
1203                 entry->func = (unsigned long)tp->rp.kp.addr;
1204                 entry->ret_ip = (unsigned long)ri->ret_addr;
1205                 for (i = 0; i < tp->nr_args; i++)
1206                         entry->args[i] = call_fetch(&tp->args[i].fetch, regs);
1207                 perf_tpcounter_event(call->id, entry->ret_ip, 1, entry, size);
1208         } while (0);
1209         return 0;
1210 }
1211
1212 static int probe_profile_enable(struct ftrace_event_call *call)
1213 {
1214         struct trace_probe *tp = (struct trace_probe *)call->data;
1215
1216         if (atomic_inc_return(&call->profile_count))
1217                 return 0;
1218
1219         if (probe_is_return(tp)) {
1220                 tp->rp.handler = kretprobe_profile_func;
1221                 return enable_kretprobe(&tp->rp);
1222         } else {
1223                 tp->rp.kp.pre_handler = kprobe_profile_func;
1224                 return enable_kprobe(&tp->rp.kp);
1225         }
1226 }
1227
1228 static void probe_profile_disable(struct ftrace_event_call *call)
1229 {
1230         if (atomic_add_negative(-1, &call->profile_count))
1231                 probe_event_disable(call);
1232 }
1233
1234 #endif  /* CONFIG_EVENT_PROFILE */
1235
1236 static int register_probe_event(struct trace_probe *tp)
1237 {
1238         struct ftrace_event_call *call = &tp->call;
1239         int ret;
1240
1241         /* Initialize ftrace_event_call */
1242         if (probe_is_return(tp)) {
1243                 tp->event.trace = print_kretprobe_event;
1244                 call->raw_init = probe_event_raw_init;
1245                 call->show_format = kretprobe_event_show_format;
1246                 call->define_fields = kretprobe_event_define_fields;
1247         } else {
1248                 tp->event.trace = print_kprobe_event;
1249                 call->raw_init = probe_event_raw_init;
1250                 call->show_format = kprobe_event_show_format;
1251                 call->define_fields = kprobe_event_define_fields;
1252         }
1253         call->event = &tp->event;
1254         call->id = register_ftrace_event(&tp->event);
1255         if (!call->id)
1256                 return -ENODEV;
1257         call->enabled = 1;
1258         call->regfunc = probe_event_enable;
1259         call->unregfunc = probe_event_disable;
1260
1261 #ifdef CONFIG_EVENT_PROFILE
1262         atomic_set(&call->profile_count, -1);
1263         call->profile_enable = probe_profile_enable;
1264         call->profile_disable = probe_profile_disable;
1265 #endif
1266         call->data = tp;
1267         ret = trace_add_event_call(call);
1268         if (ret) {
1269                 pr_info("Failed to register kprobe event: %s\n", call->name);
1270                 unregister_ftrace_event(&tp->event);
1271         }
1272         return ret;
1273 }
1274
1275 static void unregister_probe_event(struct trace_probe *tp)
1276 {
1277         /* tp->event is unregistered in trace_remove_event_call() */
1278         trace_remove_event_call(&tp->call);
1279 }
1280
1281 /* Make a debugfs interface for controling probe points */
1282 static __init int init_kprobe_trace(void)
1283 {
1284         struct dentry *d_tracer;
1285         struct dentry *entry;
1286
1287         d_tracer = tracing_init_dentry();
1288         if (!d_tracer)
1289                 return 0;
1290
1291         entry = debugfs_create_file("kprobe_events", 0644, d_tracer,
1292                                     NULL, &kprobe_events_ops);
1293
1294         /* Event list interface */
1295         if (!entry)
1296                 pr_warning("Could not create debugfs "
1297                            "'kprobe_events' entry\n");
1298
1299         /* Profile interface */
1300         entry = debugfs_create_file("kprobe_profile", 0444, d_tracer,
1301                                     NULL, &kprobe_profile_ops);
1302
1303         if (!entry)
1304                 pr_warning("Could not create debugfs "
1305                            "'kprobe_profile' entry\n");
1306         return 0;
1307 }
1308 fs_initcall(init_kprobe_trace);
1309
1310
1311 #ifdef CONFIG_FTRACE_STARTUP_TEST
1312
1313 static int kprobe_trace_selftest_target(int a1, int a2, int a3,
1314                                         int a4, int a5, int a6)
1315 {
1316         return a1 + a2 + a3 + a4 + a5 + a6;
1317 }
1318
1319 static __init int kprobe_trace_self_tests_init(void)
1320 {
1321         int ret;
1322         int (*target)(int, int, int, int, int, int);
1323
1324         target = kprobe_trace_selftest_target;
1325
1326         pr_info("Testing kprobe tracing: ");
1327
1328         ret = command_trace_probe("p:testprobe kprobe_trace_selftest_target "
1329                                   "a1 a2 a3 a4 a5 a6");
1330         if (WARN_ON_ONCE(ret))
1331                 pr_warning("error enabling function entry\n");
1332
1333         ret = command_trace_probe("r:testprobe2 kprobe_trace_selftest_target "
1334                                   "ra rv");
1335         if (WARN_ON_ONCE(ret))
1336                 pr_warning("error enabling function return\n");
1337
1338         ret = target(1, 2, 3, 4, 5, 6);
1339
1340         cleanup_all_probes();
1341
1342         pr_cont("OK\n");
1343         return 0;
1344 }
1345
1346 late_initcall(kprobe_trace_self_tests_init);
1347
1348 #endif