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
  40 /* currently, trace_kprobe only supports X86. */
  41
  42 struct fetch_func {
  43         unsigned long (*func)(struct pt_regs *, void *);
  44         void *data;
  45 };
  46
  47 static __kprobes unsigned long call_fetch(struct fetch_func *f,
  48                                           struct pt_regs *regs)
  49 {
  50         return f->func(regs, f->data);
  51 }
  52
  53 /* fetch handlers */
  54 static __kprobes unsigned long fetch_register(struct pt_regs *regs,
  55                                               void *offset)
  56 {
  57         return regs_get_register(regs, (unsigned int)((unsigned long)offset));
  58 }
  59
  60 static __kprobes unsigned long fetch_stack(struct pt_regs *regs,
  61                                            void *num)
  62 {
  63         return regs_get_kernel_stack_nth(regs,
  64                                          (unsigned int)((unsigned long)num));
  65 }
  66
  67 static __kprobes unsigned long fetch_memory(struct pt_regs *regs, void *addr)
  68 {
  69         unsigned long retval;
  70
  71         if (probe_kernel_address(addr, retval))
  72                 return 0;
  73         return retval;
  74 }
  75
  76 static __kprobes unsigned long fetch_argument(struct pt_regs *regs, void *num)
  77 {
  78         return regs_get_argument_nth(regs, (unsigned int)((unsigned long)num));
  79 }
  80
  81 static __kprobes unsigned long fetch_retvalue(struct pt_regs *regs,
  82                                               void *dummy)
  83 {
  84         return regs_return_value(regs);
  85 }
  86
  87 static __kprobes unsigned long fetch_ip(struct pt_regs *regs, void *dummy)
  88 {
  89         return instruction_pointer(regs);
  90 }
  91
  92 static __kprobes unsigned long fetch_stack_address(struct pt_regs *regs,
  93                                                    void *dummy)
  94 {
  95         return kernel_stack_pointer(regs);
  96 }
  97
  98 /* Memory fetching by symbol */
  99 struct symbol_cache {
 100         char *symbol;
 101         long offset;
 102         unsigned long addr;
 103 };
 104
 105 static unsigned long update_symbol_cache(struct symbol_cache *sc)
 106 {
 107         sc->addr = (unsigned long)kallsyms_lookup_name(sc->symbol);
 108         if (sc->addr)
 109                 sc->addr += sc->offset;
 110         return sc->addr;
 111 }
 112
 113 static void free_symbol_cache(struct symbol_cache *sc)
 114 {
 115         kfree(sc->symbol);
 116         kfree(sc);
 117 }
 118
 119 static struct symbol_cache *alloc_symbol_cache(const char *sym, long offset)
 120 {
 121         struct symbol_cache *sc;
 122
 123         if (!sym || strlen(sym) == 0)
 124                 return NULL;
 125         sc = kzalloc(sizeof(struct symbol_cache), GFP_KERNEL);
 126         if (!sc)
 127                 return NULL;
 128
 129         sc->symbol = kstrdup(sym, GFP_KERNEL);
 130         if (!sc->symbol) {
 131                 kfree(sc);
 132                 return NULL;
 133         }
 134         sc->offset = offset;
 135
 136         update_symbol_cache(sc);
 137         return sc;
 138 }
 139
 140 static __kprobes unsigned long fetch_symbol(struct pt_regs *regs, void *data)
 141 {
 142         struct symbol_cache *sc = data;
 143
 144         if (sc->addr)
 145                 return fetch_memory(regs, (void *)sc->addr);
 146         else
 147                 return 0;
 148 }
 149
 150 /* Special indirect memory access interface */
 151 struct indirect_fetch_data {
 152         struct fetch_func orig;
 153         long offset;
 154 };
 155
 156 static __kprobes unsigned long fetch_indirect(struct pt_regs *regs, void *data)
 157 {
 158         struct indirect_fetch_data *ind = data;
 159         unsigned long addr;
 160
 161         addr = call_fetch(&ind->orig, regs);
 162         if (addr) {
 163                 addr += ind->offset;
 164                 return fetch_memory(regs, (void *)addr);
 165         } else
 166                 return 0;
 167 }
 168
 169 static __kprobes void free_indirect_fetch_data(struct indirect_fetch_data *data)
 170 {
 171         if (data->orig.func == fetch_indirect)
 172                 free_indirect_fetch_data(data->orig.data);
 173         else if (data->orig.func == fetch_symbol)
 174                 free_symbol_cache(data->orig.data);
 175         kfree(data);
 176 }
 177
 178 /**
 179  * Kprobe tracer core functions
 180  */
 181
 182 struct probe_arg {
 183         struct fetch_func       fetch;
 184         const char              *name;
 185 };
 186
 187 struct trace_probe {
 188         struct list_head        list;
 189         struct kretprobe        rp;     /* Use rp.kp for kprobe use */
 190         unsigned long           nhit;
 191         const char              *symbol;        /* symbol name */
 192         struct ftrace_event_call        call;
 193         struct trace_event              event;
 194         unsigned int            nr_args;
 195         struct probe_arg        args[];
 196 };
 197
 198 #define SIZEOF_TRACE_PROBE(n)                   \
 199         (offsetof(struct trace_probe, args) +   \
 200         (sizeof(struct probe_arg) * (n)))
 201
 202 static int kprobe_trace_func(struct kprobe *kp, struct pt_regs *regs);
 203 static int kretprobe_trace_func(struct kretprobe_instance *ri,
 204                                 struct pt_regs *regs);
 205
 206 static __kprobes int probe_is_return(struct trace_probe *tp)
 207 {
 208         return tp->rp.handler != NULL;
 209 }
 210
 211 static __kprobes const char *probe_symbol(struct trace_probe *tp)
 212 {
 213         return tp->symbol ? tp->symbol : "unknown";
 214 }
 215
 216 static int probe_arg_string(char *buf, size_t n, struct fetch_func *ff)
 217 {
 218         int ret = -EINVAL;
 219
 220         if (ff->func == fetch_argument)
 221                 ret = snprintf(buf, n, "a%lu", (unsigned long)ff->data);
 222         else if (ff->func == fetch_register) {
 223                 const char *name;
 224                 name = regs_query_register_name((unsigned int)((long)ff->data));
 225                 ret = snprintf(buf, n, "%%%s", name);
 226         } else if (ff->func == fetch_stack)
 227                 ret = snprintf(buf, n, "s%lu", (unsigned long)ff->data);
 228         else if (ff->func == fetch_memory)
 229                 ret = snprintf(buf, n, "@0x%p", ff->data);
 230         else if (ff->func == fetch_symbol) {
 231                 struct symbol_cache *sc = ff->data;
 232                 ret = snprintf(buf, n, "@%s%+ld", sc->symbol, sc->offset);
 233         } else if (ff->func == fetch_retvalue)
 234                 ret = snprintf(buf, n, "rv");
 235         else if (ff->func == fetch_ip)
 236                 ret = snprintf(buf, n, "ra");
 237         else if (ff->func == fetch_stack_address)
 238                 ret = snprintf(buf, n, "sa");
 239         else if (ff->func == fetch_indirect) {
 240                 struct indirect_fetch_data *id = ff->data;
 241                 size_t l = 0;
 242                 ret = snprintf(buf, n, "%+ld(", id->offset);
 243                 if (ret >= n)
 244                         goto end;
 245                 l += ret;
 246                 ret = probe_arg_string(buf + l, n - l, &id->orig);
 247                 if (ret < 0)
 248                         goto end;
 249                 l += ret;
 250                 ret = snprintf(buf + l, n - l, ")");
 251                 ret += l;
 252         }
 253 end:
 254         if (ret >= n)
 255                 return -ENOSPC;
 256         return ret;
 257 }
 258
 259 static int register_probe_event(struct trace_probe *tp);
 260 static void unregister_probe_event(struct trace_probe *tp);
 261
 262 static DEFINE_MUTEX(probe_lock);
 263 static LIST_HEAD(probe_list);
 264
 265 /*
 266  * Allocate new trace_probe and initialize it (including kprobes).
 267  */
 268 static struct trace_probe *alloc_trace_probe(const char *event,
 269                                              void *addr,
 270                                              const char *symbol,
 271                                              unsigned long offs,
 272                                              int nargs, int is_return)
 273 {
 274         struct trace_probe *tp;
 275
 276         tp = kzalloc(SIZEOF_TRACE_PROBE(nargs), GFP_KERNEL);
 277         if (!tp)
 278                 return ERR_PTR(-ENOMEM);
 279
 280         if (symbol) {
 281                 tp->symbol = kstrdup(symbol, GFP_KERNEL);
 282                 if (!tp->symbol)
 283                         goto error;
 284                 tp->rp.kp.symbol_name = tp->symbol;
 285                 tp->rp.kp.offset = offs;
 286         } else
 287                 tp->rp.kp.addr = addr;
 288
 289         /* Set handler here for checking whether this probe is return or not. */
 290         if (is_return)
 291                 tp->rp.handler = kretprobe_trace_func;
 292         else
 293                 tp->rp.kp.pre_handler = kprobe_trace_func;
 294
 295         if (!event)
 296                 goto error;
 297         tp->call.name = kstrdup(event, GFP_KERNEL);
 298         if (!tp->call.name)
 299                 goto error;
 300
 301         INIT_LIST_HEAD(&tp->list);
 302         return tp;
 303 error:
 304         kfree(tp->symbol);
 305         kfree(tp);
 306         return ERR_PTR(-ENOMEM);
 307 }
 308
 309 static void free_probe_arg(struct probe_arg *arg)
 310 {
 311         if (arg->fetch.func == fetch_symbol)
 312                 free_symbol_cache(arg->fetch.data);
 313         else if (arg->fetch.func == fetch_indirect)
 314                 free_indirect_fetch_data(arg->fetch.data);
 315         kfree(arg->name);
 316 }
 317
 318 static void free_trace_probe(struct trace_probe *tp)
 319 {
 320         int i;
 321
 322         for (i = 0; i < tp->nr_args; i++)
 323                 free_probe_arg(&tp->args[i]);
 324
 325         kfree(tp->call.name);
 326         kfree(tp->symbol);
 327         kfree(tp);
 328 }
 329
 330 static struct trace_probe *find_probe_event(const char *event)
 331 {
 332         struct trace_probe *tp;
 333
 334         list_for_each_entry(tp, &probe_list, list)
 335                 if (!strcmp(tp->call.name, event))
 336                         return tp;
 337         return NULL;
 338 }
 339
 340 static void __unregister_trace_probe(struct trace_probe *tp)
 341 {
 342         if (probe_is_return(tp))
 343                 unregister_kretprobe(&tp->rp);
 344         else
 345                 unregister_kprobe(&tp->rp.kp);
 346 }
 347
 348 /* Unregister a trace_probe and probe_event: call with locking probe_lock */
 349 static void unregister_trace_probe(struct trace_probe *tp)
 350 {
 351         unregister_probe_event(tp);
 352         __unregister_trace_probe(tp);
 353         list_del(&tp->list);
 354 }
 355
 356 /* Register a trace_probe and probe_event */
 357 static int register_trace_probe(struct trace_probe *tp)
 358 {
 359         struct trace_probe *old_tp;
 360         int ret;
 361
 362         mutex_lock(&probe_lock);
 363
 364         if (probe_is_return(tp))
 365                 ret = register_kretprobe(&tp->rp);
 366         else
 367                 ret = register_kprobe(&tp->rp.kp);
 368
 369         if (ret) {
 370                 pr_warning("Could not insert probe(%d)\n", ret);
 371                 if (ret == -EILSEQ) {
 372                         pr_warning("Probing address(0x%p) is not an "
 373                                    "instruction boundary.\n",
 374                                    tp->rp.kp.addr);
 375                         ret = -EINVAL;
 376                 }
 377                 goto end;
 378         }
 379         /* register as an event */
 380         old_tp = find_probe_event(tp->call.name);
 381         if (old_tp) {
 382                 /* delete old event */
 383                 unregister_trace_probe(old_tp);
 384                 free_trace_probe(old_tp);
 385         }
 386         ret = register_probe_event(tp);
 387         if (ret) {
 388                 pr_warning("Faild to register probe event(%d)\n", ret);
 389                 __unregister_trace_probe(tp);
 390         }
 391         list_add_tail(&tp->list, &probe_list);
 392 end:
 393         mutex_unlock(&probe_lock);
 394         return ret;
 395 }
 396
 397 /* Split symbol and offset. */
 398 static int split_symbol_offset(char *symbol, unsigned long *offset)
 399 {
 400         char *tmp;
 401         int ret;
 402
 403         if (!offset)
 404                 return -EINVAL;
 405
 406         tmp = strchr(symbol, '+');
 407         if (tmp) {
 408                 /* skip sign because strict_strtol doesn't accept '+' */
 409                 ret = strict_strtoul(tmp + 1, 0, offset);
 410                 if (ret)
 411                         return ret;
 412                 *tmp = '\0';
 413         } else
 414                 *offset = 0;
 415         return 0;
 416 }
 417
 418 #define PARAM_MAX_ARGS 16
 419 #define PARAM_MAX_STACK (THREAD_SIZE / sizeof(unsigned long))
 420
 421 static int parse_probe_arg(char *arg, struct fetch_func *ff, int is_return)
 422 {
 423         int ret = 0;
 424         unsigned long param;
 425         long offset;
 426         char *tmp;
 427
 428         switch (arg[0]) {
 429         case 'a':       /* argument */
 430                 ret = strict_strtoul(arg + 1, 10, &param);
 431                 if (ret || param > PARAM_MAX_ARGS)
 432                         ret = -EINVAL;
 433                 else {
 434                         ff->func = fetch_argument;
 435                         ff->data = (void *)param;
 436                 }
 437                 break;
 438         case 'r':       /* retval or retaddr */
 439                 if (is_return && arg[1] == 'v') {
 440                         ff->func = fetch_retvalue;
 441                         ff->data = NULL;
 442                 } else if (is_return && arg[1] == 'a') {
 443                         ff->func = fetch_ip;
 444                         ff->data = NULL;
 445                 } else
 446                         ret = -EINVAL;
 447                 break;
 448         case '%':       /* named register */
 449                 ret = regs_query_register_offset(arg + 1);
 450                 if (ret >= 0) {
 451                         ff->func = fetch_register;
 452                         ff->data = (void *)(unsigned long)ret;
 453                         ret = 0;
 454                 }
 455                 break;
 456         case 's':       /* stack */
 457                 if (arg[1] == 'a') {
 458                         ff->func = fetch_stack_address;
 459                         ff->data = NULL;
 460                 } else {
 461                         ret = strict_strtoul(arg + 1, 10, &param);
 462                         if (ret || param > PARAM_MAX_STACK)
 463                                 ret = -EINVAL;
 464                         else {
 465                                 ff->func = fetch_stack;
 466                                 ff->data = (void *)param;
 467                         }
 468                 }
 469                 break;
 470         case '@':       /* memory or symbol */
 471                 if (isdigit(arg[1])) {
 472                         ret = strict_strtoul(arg + 1, 0, &param);
 473                         if (ret)
 474                                 break;
 475                         ff->func = fetch_memory;
 476                         ff->data = (void *)param;
 477                 } else {
 478                         ret = split_symbol_offset(arg + 1, &offset);
 479                         if (ret)
 480                                 break;
 481                         ff->data = alloc_symbol_cache(arg + 1,
 482                                                               offset);
 483                         if (ff->data)
 484                                 ff->func = fetch_symbol;
 485                         else
 486                                 ret = -EINVAL;
 487                 }
 488                 break;
 489         case '+':       /* indirect memory */
 490         case '-':
 491                 tmp = strchr(arg, '(');
 492                 if (!tmp) {
 493                         ret = -EINVAL;
 494                         break;
 495                 }
 496                 *tmp = '\0';
 497                 ret = strict_strtol(arg + 1, 0, &offset);
 498                 if (ret)
 499                         break;
 500                 if (arg[0] == '-')
 501                         offset = -offset;
 502                 arg = tmp + 1;
 503                 tmp = strrchr(arg, ')');
 504                 if (tmp) {
 505                         struct indirect_fetch_data *id;
 506                         *tmp = '\0';
 507                         id = kzalloc(sizeof(struct indirect_fetch_data),
 508                                      GFP_KERNEL);
 509                         if (!id)
 510                                 return -ENOMEM;
 511                         id->offset = offset;
 512                         ret = parse_probe_arg(arg, &id->orig, is_return);
 513                         if (ret)
 514                                 kfree(id);
 515                         else {
 516                                 ff->func = fetch_indirect;
 517                                 ff->data = (void *)id;
 518                         }
 519                 } else
 520                         ret = -EINVAL;
 521                 break;
 522         default:
 523                 /* TODO: support custom handler */
 524                 ret = -EINVAL;
 525         }
 526         return ret;
 527 }
 528
 529 static int create_trace_probe(int argc, char **argv)
 530 {
 531         /*
 532          * Argument syntax:
 533          *  - Add kprobe: p[:EVENT] SYMBOL[+OFFS]|ADDRESS [FETCHARGS]
 534          *  - Add kretprobe: r[:EVENT] SYMBOL[+0] [FETCHARGS]
 535          * Fetch args:
 536          *  aN  : fetch Nth of function argument. (N:0-)
 537          *  rv  : fetch return value
 538          *  ra  : fetch return address
 539          *  sa  : fetch stack address
 540          *  sN  : fetch Nth of stack (N:0-)
 541          *  @ADDR       : fetch memory at ADDR (ADDR should be in kernel)
 542          *  @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
 543          *  %REG        : fetch register REG
 544          * Indirect memory fetch:
 545          *  +|-offs(ARG) : fetch memory at ARG +|- offs address.
 546          * Alias name of args:
 547          *  NAME=FETCHARG : set NAME as alias of FETCHARG.
 548          */
 549         struct trace_probe *tp;
 550         int i, ret = 0;
 551         int is_return = 0;
 552         char *symbol = NULL, *event = NULL, *arg = NULL;
 553         unsigned long offset = 0;
 554         void *addr = NULL;
 555         char buf[MAX_EVENT_NAME_LEN];
 556
 557         if (argc < 2)
 558                 return -EINVAL;
 559
 560         if (argv[0][0] == 'p')
 561                 is_return = 0;
 562         else if (argv[0][0] == 'r')
 563                 is_return = 1;
 564         else
 565                 return -EINVAL;
 566
 567         if (argv[0][1] == ':') {
 568                 event = &argv[0][2];
 569                 if (strlen(event) == 0) {
 570                         pr_info("Event name is not specifiled\n");
 571                         return -EINVAL;
 572                 }
 573         }
 574
 575         if (isdigit(argv[1][0])) {
 576                 if (is_return)
 577                         return -EINVAL;
 578                 /* an address specified */
 579                 ret = strict_strtoul(&argv[0][2], 0, (unsigned long *)&addr);
 580                 if (ret)
 581                         return ret;
 582         } else {
 583                 /* a symbol specified */
 584                 symbol = argv[1];
 585                 /* TODO: support .init module functions */
 586                 ret = split_symbol_offset(symbol, &offset);
 587                 if (ret)
 588                         return ret;
 589                 if (offset && is_return)
 590                         return -EINVAL;
 591         }
 592         argc -= 2; argv += 2;
 593
 594         /* setup a probe */
 595         if (!event) {
 596                 /* Make a new event name */
 597                 if (symbol)
 598                         snprintf(buf, MAX_EVENT_NAME_LEN, "%c@%s%+ld",
 599                                  is_return ? 'r' : 'p', symbol, offset);
 600                 else
 601                         snprintf(buf, MAX_EVENT_NAME_LEN, "%c@0x%p",
 602                                  is_return ? 'r' : 'p', addr);
 603                 event = buf;
 604         }
 605         tp = alloc_trace_probe(event, addr, symbol, offset, argc, is_return);
 606         if (IS_ERR(tp))
 607                 return PTR_ERR(tp);
 608
 609         /* parse arguments */
 610         ret = 0;
 611         for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
 612                 /* Parse argument name */
 613                 arg = strchr(argv[i], '=');
 614                 if (arg)
 615                         *arg++ = '\0';
 616                 else
 617                         arg = argv[i];
 618                 tp->args[i].name = kstrdup(argv[i], GFP_KERNEL);
 619
 620                 /* Parse fetch argument */
 621                 if (strlen(arg) > MAX_ARGSTR_LEN) {
 622                         pr_info("Argument%d(%s) is too long.\n", i, arg);
 623                         ret = -ENOSPC;
 624                         goto error;
 625                 }
 626                 ret = parse_probe_arg(arg, &tp->args[i].fetch, is_return);
 627                 if (ret)
 628                         goto error;
 629         }
 630         tp->nr_args = i;
 631
 632         ret = register_trace_probe(tp);
 633         if (ret)
 634                 goto error;
 635         return 0;
 636
 637 error:
 638         free_trace_probe(tp);
 639         return ret;
 640 }
 641
 642 static void cleanup_all_probes(void)
 643 {
 644         struct trace_probe *tp;
 645
 646         mutex_lock(&probe_lock);
 647         /* TODO: Use batch unregistration */
 648         while (!list_empty(&probe_list)) {
 649                 tp = list_entry(probe_list.next, struct trace_probe, list);
 650                 unregister_trace_probe(tp);
 651                 free_trace_probe(tp);
 652         }
 653         mutex_unlock(&probe_lock);
 654 }
 655
 656
 657 /* Probes listing interfaces */
 658 static void *probes_seq_start(struct seq_file *m, loff_t *pos)
 659 {
 660         mutex_lock(&probe_lock);
 661         return seq_list_start(&probe_list, *pos);
 662 }
 663
 664 static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
 665 {
 666         return seq_list_next(v, &probe_list, pos);
 667 }
 668
 669 static void probes_seq_stop(struct seq_file *m, void *v)
 670 {
 671         mutex_unlock(&probe_lock);
 672 }
 673
 674 static int probes_seq_show(struct seq_file *m, void *v)
 675 {
 676         struct trace_probe *tp = v;
 677         int i, ret;
 678         char buf[MAX_ARGSTR_LEN + 1];
 679
 680         seq_printf(m, "%c", probe_is_return(tp) ? 'r' : 'p');
 681         seq_printf(m, ":%s", tp->call.name);
 682
 683         if (tp->symbol)
 684                 seq_printf(m, " %s+%u", probe_symbol(tp), tp->rp.kp.offset);
 685         else
 686                 seq_printf(m, " 0x%p", tp->rp.kp.addr);
 687
 688         for (i = 0; i < tp->nr_args; i++) {
 689                 ret = probe_arg_string(buf, MAX_ARGSTR_LEN, &tp->args[i].fetch);
 690                 if (ret < 0) {
 691                         pr_warning("Argument%d decoding error(%d).\n", i, ret);
 692                         return ret;
 693                 }
 694                 seq_printf(m, " %s=%s", tp->args[i].name, buf);
 695         }
 696         seq_printf(m, "\n");
 697         return 0;
 698 }
 699
 700 static const struct seq_operations probes_seq_op = {
 701         .start  = probes_seq_start,
 702         .next   = probes_seq_next,
 703         .stop   = probes_seq_stop,
 704         .show   = probes_seq_show
 705 };
 706
 707 static int probes_open(struct inode *inode, struct file *file)
 708 {
 709         if ((file->f_mode & FMODE_WRITE) &&
 710             (file->f_flags & O_TRUNC))
 711                 cleanup_all_probes();
 712
 713         return seq_open(file, &probes_seq_op);
 714 }
 715
 716 static int command_trace_probe(const char *buf)
 717 {
 718         char **argv;
 719         int argc = 0, ret = 0;
 720
 721         argv = argv_split(GFP_KERNEL, buf, &argc);
 722         if (!argv)
 723                 return -ENOMEM;
 724
 725         if (argc)
 726                 ret = create_trace_probe(argc, argv);
 727
 728         argv_free(argv);
 729         return ret;
 730 }
 731
 732 #define WRITE_BUFSIZE 128
 733
 734 static ssize_t probes_write(struct file *file, const char __user *buffer,
 735                             size_t count, loff_t *ppos)
 736 {
 737         char *kbuf, *tmp;
 738         int ret;
 739         size_t done;
 740         size_t size;
 741
 742         kbuf = kmalloc(WRITE_BUFSIZE, GFP_KERNEL);
 743         if (!kbuf)
 744                 return -ENOMEM;
 745
 746         ret = done = 0;
 747         while (done < count) {
 748                 size = count - done;
 749                 if (size >= WRITE_BUFSIZE)
 750                         size = WRITE_BUFSIZE - 1;
 751                 if (copy_from_user(kbuf, buffer + done, size)) {
 752                         ret = -EFAULT;
 753                         goto out;
 754                 }
 755                 kbuf[size] = '\0';
 756                 tmp = strchr(kbuf, '\n');
 757                 if (tmp) {
 758                         *tmp = '\0';
 759                         size = tmp - kbuf + 1;
 760                 } else if (done + size < count) {
 761                         pr_warning("Line length is too long: "
 762                                    "Should be less than %d.", WRITE_BUFSIZE);
 763                         ret = -EINVAL;
 764                         goto out;
 765                 }
 766                 done += size;
 767                 /* Remove comments */
 768                 tmp = strchr(kbuf, '#');
 769                 if (tmp)
 770                         *tmp = '\0';
 771
 772                 ret = command_trace_probe(kbuf);
 773                 if (ret)
 774                         goto out;
 775         }
 776         ret = done;
 777 out:
 778         kfree(kbuf);
 779         return ret;
 780 }
 781
 782 static const struct file_operations kprobe_events_ops = {
 783         .owner          = THIS_MODULE,
 784         .open           = probes_open,
 785         .read           = seq_read,
 786         .llseek         = seq_lseek,
 787         .release        = seq_release,
 788         .write          = probes_write,
 789 };
 790
 791 /* Probes profiling interfaces */
 792 static int probes_profile_seq_show(struct seq_file *m, void *v)
 793 {
 794         struct trace_probe *tp = v;
 795
 796         seq_printf(m, "  %-44s %15lu %15lu\n", tp->call.name, tp->nhit,
 797                    tp->rp.kp.nmissed);
 798
 799         return 0;
 800 }
 801
 802 static const struct seq_operations profile_seq_op = {
 803         .start  = probes_seq_start,
 804         .next   = probes_seq_next,
 805         .stop   = probes_seq_stop,
 806         .show   = probes_profile_seq_show
 807 };
 808
 809 static int profile_open(struct inode *inode, struct file *file)
 810 {
 811         return seq_open(file, &profile_seq_op);
 812 }
 813
 814 static const struct file_operations kprobe_profile_ops = {
 815         .owner          = THIS_MODULE,
 816         .open           = profile_open,
 817         .read           = seq_read,
 818         .llseek         = seq_lseek,
 819         .release        = seq_release,
 820 };
 821
 822 /* Kprobe handler */
 823 static __kprobes int kprobe_trace_func(struct kprobe *kp, struct pt_regs *regs)
 824 {
 825         struct trace_probe *tp = container_of(kp, struct trace_probe, rp.kp);
 826         struct kprobe_trace_entry *entry;
 827         struct ring_buffer_event *event;
 828         struct ring_buffer *buffer;
 829         int size, i, pc;
 830         unsigned long irq_flags;
 831         struct ftrace_event_call *call = &tp->call;
 832
 833         tp->nhit++;
 834
 835         local_save_flags(irq_flags);
 836         pc = preempt_count();
 837
 838         size = SIZEOF_KPROBE_TRACE_ENTRY(tp->nr_args);
 839
 840         event = trace_current_buffer_lock_reserve(&buffer, call->id, size,
 841                                                   irq_flags, pc);
 842         if (!event)
 843                 return 0;
 844
 845         entry = ring_buffer_event_data(event);
 846         entry->nargs = tp->nr_args;
 847         entry->ip = (unsigned long)kp->addr;
 848         for (i = 0; i < tp->nr_args; i++)
 849                 entry->args[i] = call_fetch(&tp->args[i].fetch, regs);
 850
 851         if (!filter_current_check_discard(buffer, call, entry, event))
 852                 trace_nowake_buffer_unlock_commit(buffer, event, irq_flags, pc);
 853         return 0;
 854 }
 855
 856 /* Kretprobe handler */
 857 static __kprobes int kretprobe_trace_func(struct kretprobe_instance *ri,
 858                                           struct pt_regs *regs)
 859 {
 860         struct trace_probe *tp = container_of(ri->rp, struct trace_probe, rp);
 861         struct kretprobe_trace_entry *entry;
 862         struct ring_buffer_event *event;
 863         struct ring_buffer *buffer;
 864         int size, i, pc;
 865         unsigned long irq_flags;
 866         struct ftrace_event_call *call = &tp->call;
 867
 868         local_save_flags(irq_flags);
 869         pc = preempt_count();
 870
 871         size = SIZEOF_KRETPROBE_TRACE_ENTRY(tp->nr_args);
 872
 873         event = trace_current_buffer_lock_reserve(&buffer, call->id, size,
 874                                                   irq_flags, pc);
 875         if (!event)
 876                 return 0;
 877
 878         entry = ring_buffer_event_data(event);
 879         entry->nargs = tp->nr_args;
 880         entry->func = (unsigned long)tp->rp.kp.addr;
 881         entry->ret_ip = (unsigned long)ri->ret_addr;
 882         for (i = 0; i < tp->nr_args; i++)
 883                 entry->args[i] = call_fetch(&tp->args[i].fetch, regs);
 884
 885         if (!filter_current_check_discard(buffer, call, entry, event))
 886                 trace_nowake_buffer_unlock_commit(buffer, event, irq_flags, pc);
 887
 888         return 0;
 889 }
 890
 891 /* Event entry printers */
 892 enum print_line_t
 893 print_kprobe_event(struct trace_iterator *iter, int flags)
 894 {
 895         struct kprobe_trace_entry *field;
 896         struct trace_seq *s = &iter->seq;
 897         struct trace_event *event;
 898         struct trace_probe *tp;
 899         int i;
 900
 901         field = (struct kprobe_trace_entry *)iter->ent;
 902         event = ftrace_find_event(field->ent.type);
 903         tp = container_of(event, struct trace_probe, event);
 904
 905         if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
 906                 goto partial;
 907
 908         if (!trace_seq_puts(s, ":"))
 909                 goto partial;
 910
 911         for (i = 0; i < field->nargs; i++)
 912                 if (!trace_seq_printf(s, " %s=%lx",
 913                                       tp->args[i].name, field->args[i]))
 914                         goto partial;
 915
 916         if (!trace_seq_puts(s, "\n"))
 917                 goto partial;
 918
 919         return TRACE_TYPE_HANDLED;
 920 partial:
 921         return TRACE_TYPE_PARTIAL_LINE;
 922 }
 923
 924 enum print_line_t
 925 print_kretprobe_event(struct trace_iterator *iter, int flags)
 926 {
 927         struct kretprobe_trace_entry *field;
 928         struct trace_seq *s = &iter->seq;
 929         struct trace_event *event;
 930         struct trace_probe *tp;
 931         int i;
 932
 933         field = (struct kretprobe_trace_entry *)iter->ent;
 934         event = ftrace_find_event(field->ent.type);
 935         tp = container_of(event, struct trace_probe, event);
 936
 937         if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
 938                 goto partial;
 939
 940         if (!trace_seq_puts(s, " <- "))
 941                 goto partial;
 942
 943         if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
 944                 goto partial;
 945
 946         if (!trace_seq_puts(s, ":"))
 947                 goto partial;
 948
 949         for (i = 0; i < field->nargs; i++)
 950                 if (!trace_seq_printf(s, " %s=%lx",
 951                                       tp->args[i].name, field->args[i]))
 952                         goto partial;
 953
 954         if (!trace_seq_puts(s, "\n"))
 955                 goto partial;
 956
 957         return TRACE_TYPE_HANDLED;
 958 partial:
 959         return TRACE_TYPE_PARTIAL_LINE;
 960 }
 961
 962 static int probe_event_enable(struct ftrace_event_call *call)
 963 {
 964         struct trace_probe *tp = (struct trace_probe *)call->data;
 965
 966         if (probe_is_return(tp)) {
 967                 tp->rp.handler = kretprobe_trace_func;
 968                 return enable_kretprobe(&tp->rp);
 969         } else {
 970                 tp->rp.kp.pre_handler = kprobe_trace_func;
 971                 return enable_kprobe(&tp->rp.kp);
 972         }
 973 }
 974
 975 static void probe_event_disable(struct ftrace_event_call *call)
 976 {
 977         struct trace_probe *tp = (struct trace_probe *)call->data;
 978
 979         if (probe_is_return(tp))
 980                 disable_kretprobe(&tp->rp);
 981         else
 982                 disable_kprobe(&tp->rp.kp);
 983 }
 984
 985 static int probe_event_raw_init(struct ftrace_event_call *event_call)
 986 {
 987         INIT_LIST_HEAD(&event_call->fields);
 988
 989         return 0;
 990 }
 991
 992 #undef DEFINE_FIELD
 993 #define DEFINE_FIELD(type, item, name, is_signed)                       \
 994         do {                                                            \
 995                 ret = trace_define_field(event_call, #type, name,       \
 996                                          offsetof(typeof(field), item), \
 997                                          sizeof(field.item), is_signed, \
 998                                          FILTER_OTHER);                 \
 999                 if (ret)                                                \
1000                         return ret;                                     \
1001         } while (0)
1002
1003 static int kprobe_event_define_fields(struct ftrace_event_call *event_call)
1004 {
1005         int ret, i;
1006         struct kprobe_trace_entry field;
1007         struct trace_probe *tp = (struct trace_probe *)event_call->data;
1008
1009         ret = trace_define_common_fields(event_call);
1010         if (!ret)
1011                 return ret;
1012
1013         DEFINE_FIELD(unsigned long, ip, "ip", 0);
1014         DEFINE_FIELD(int, nargs, "nargs", 1);
1015         /* Set argument names as fields */
1016         for (i = 0; i < tp->nr_args; i++)
1017                 DEFINE_FIELD(unsigned long, args[i], tp->args[i].name, 0);
1018         return 0;
1019 }
1020
1021 static int kretprobe_event_define_fields(struct ftrace_event_call *event_call)
1022 {
1023         int ret, i;
1024         struct kretprobe_trace_entry field;
1025         struct trace_probe *tp = (struct trace_probe *)event_call->data;
1026
1027         ret = trace_define_common_fields(event_call);
1028         if (!ret)
1029                 return ret;
1030
1031         DEFINE_FIELD(unsigned long, func, "func", 0);
1032         DEFINE_FIELD(unsigned long, ret_ip, "ret_ip", 0);
1033         DEFINE_FIELD(int, nargs, "nargs", 1);
1034         /* Set argument names as fields */
1035         for (i = 0; i < tp->nr_args; i++)
1036                 DEFINE_FIELD(unsigned long, args[i], tp->args[i].name, 0);
1037         return 0;
1038 }
1039
1040 static int __probe_event_show_format(struct trace_seq *s,
1041                                      struct trace_probe *tp, const char *fmt,
1042                                      const char *arg)
1043 {
1044         int i;
1045
1046         /* Show format */
1047         if (!trace_seq_printf(s, "\nprint fmt: \"%s", fmt))
1048                 return 0;
1049
1050         for (i = 0; i < tp->nr_args; i++)
1051                 if (!trace_seq_printf(s, " %s=%%lx", tp->args[i].name))
1052                         return 0;
1053
1054         if (!trace_seq_printf(s, "\", %s", arg))
1055                 return 0;
1056
1057         for (i = 0; i < tp->nr_args; i++)
1058                 if (!trace_seq_printf(s, ", REC->%s", tp->args[i].name))
1059                         return 0;
1060
1061         return trace_seq_puts(s, "\n");
1062 }
1063
1064 #undef SHOW_FIELD
1065 #define SHOW_FIELD(type, item, name)                                    \
1066         do {                                                            \
1067                 ret = trace_seq_printf(s, "\tfield: " #type " %s;\t"    \
1068                                 "offset:%u;\tsize:%u;\n", name,         \
1069                                 (unsigned int)offsetof(typeof(field), item),\
1070                                 (unsigned int)sizeof(type));            \
1071                 if (!ret)                                               \
1072                         return 0;                                       \
1073         } while (0)
1074
1075 static int kprobe_event_show_format(struct ftrace_event_call *call,
1076                                     struct trace_seq *s)
1077 {
1078         struct kprobe_trace_entry field __attribute__((unused));
1079         int ret, i;
1080         struct trace_probe *tp = (struct trace_probe *)call->data;
1081
1082         SHOW_FIELD(unsigned long, ip, "ip");
1083         SHOW_FIELD(int, nargs, "nargs");
1084
1085         /* Show fields */
1086         for (i = 0; i < tp->nr_args; i++)
1087                 SHOW_FIELD(unsigned long, args[i], tp->args[i].name);
1088         trace_seq_puts(s, "\n");
1089
1090         return __probe_event_show_format(s, tp, "%lx:", "ip");
1091 }
1092
1093 static int kretprobe_event_show_format(struct ftrace_event_call *call,
1094                                        struct trace_seq *s)
1095 {
1096         struct kretprobe_trace_entry field __attribute__((unused));
1097         int ret, i;
1098         struct trace_probe *tp = (struct trace_probe *)call->data;
1099
1100         SHOW_FIELD(unsigned long, func, "func");
1101         SHOW_FIELD(unsigned long, ret_ip, "ret_ip");
1102         SHOW_FIELD(int, nargs, "nargs");
1103
1104         /* Show fields */
1105         for (i = 0; i < tp->nr_args; i++)
1106                 SHOW_FIELD(unsigned long, args[i], tp->args[i].name);
1107         trace_seq_puts(s, "\n");
1108
1109         return __probe_event_show_format(s, tp, "%lx <- %lx:",
1110                                           "func, ret_ip");
1111 }
1112
1113 #ifdef CONFIG_EVENT_PROFILE
1114
1115 /* Kprobe profile handler */
1116 static __kprobes int kprobe_profile_func(struct kprobe *kp,
1117                                          struct pt_regs *regs)
1118 {
1119         struct trace_probe *tp = container_of(kp, struct trace_probe, rp.kp);
1120         struct ftrace_event_call *call = &tp->call;
1121         struct kprobe_trace_entry *entry;
1122         int size, i, pc;
1123         unsigned long irq_flags;
1124
1125         local_save_flags(irq_flags);
1126         pc = preempt_count();
1127
1128         size = SIZEOF_KPROBE_TRACE_ENTRY(tp->nr_args);
1129
1130         do {
1131                 char raw_data[size];
1132                 struct trace_entry *ent;
1133
1134                 *(u64 *)(&raw_data[size - sizeof(u64)]) = 0ULL;
1135                 entry = (struct kprobe_trace_entry *)raw_data;
1136                 ent = &entry->ent;
1137
1138                 tracing_generic_entry_update(ent, irq_flags, pc);
1139                 ent->type = call->id;
1140                 entry->nargs = tp->nr_args;
1141                 entry->ip = (unsigned long)kp->addr;
1142                 for (i = 0; i < tp->nr_args; i++)
1143                         entry->args[i] = call_fetch(&tp->args[i].fetch, regs);
1144                 perf_tpcounter_event(call->id, entry->ip, 1, entry, size);
1145         } while (0);
1146         return 0;
1147 }
1148
1149 /* Kretprobe profile handler */
1150 static __kprobes int kretprobe_profile_func(struct kretprobe_instance *ri,
1151                                             struct pt_regs *regs)
1152 {
1153         struct trace_probe *tp = container_of(ri->rp, struct trace_probe, rp);
1154         struct ftrace_event_call *call = &tp->call;
1155         struct kretprobe_trace_entry *entry;
1156         int size, i, pc;
1157         unsigned long irq_flags;
1158
1159         local_save_flags(irq_flags);
1160         pc = preempt_count();
1161
1162         size = SIZEOF_KRETPROBE_TRACE_ENTRY(tp->nr_args);
1163
1164         do {
1165                 char raw_data[size];
1166                 struct trace_entry *ent;
1167
1168                 *(u64 *)(&raw_data[size - sizeof(u64)]) = 0ULL;
1169                 entry = (struct kretprobe_trace_entry *)raw_data;
1170                 ent = &entry->ent;
1171
1172                 tracing_generic_entry_update(ent, irq_flags, pc);
1173                 ent->type = call->id;
1174                 entry->nargs = tp->nr_args;
1175                 entry->func = (unsigned long)tp->rp.kp.addr;
1176                 entry->ret_ip = (unsigned long)ri->ret_addr;
1177                 for (i = 0; i < tp->nr_args; i++)
1178                         entry->args[i] = call_fetch(&tp->args[i].fetch, regs);
1179                 perf_tpcounter_event(call->id, entry->ret_ip, 1, entry, size);
1180         } while (0);
1181         return 0;
1182 }
1183
1184 static int probe_profile_enable(struct ftrace_event_call *call)
1185 {
1186         struct trace_probe *tp = (struct trace_probe *)call->data;
1187
1188         if (atomic_inc_return(&call->profile_count))
1189                 return 0;
1190
1191         if (probe_is_return(tp)) {
1192                 tp->rp.handler = kretprobe_profile_func;
1193                 return enable_kretprobe(&tp->rp);
1194         } else {
1195                 tp->rp.kp.pre_handler = kprobe_profile_func;
1196                 return enable_kprobe(&tp->rp.kp);
1197         }
1198 }
1199
1200 static void probe_profile_disable(struct ftrace_event_call *call)
1201 {
1202         if (atomic_add_negative(-1, &call->profile_count))
1203                 probe_event_disable(call);
1204 }
1205
1206 #endif  /* CONFIG_EVENT_PROFILE */
1207
1208 static int register_probe_event(struct trace_probe *tp)
1209 {
1210         struct ftrace_event_call *call = &tp->call;
1211         int ret;
1212
1213         /* Initialize ftrace_event_call */
1214         call->system = "kprobes";
1215         if (probe_is_return(tp)) {
1216                 tp->event.trace = print_kretprobe_event;
1217                 call->raw_init = probe_event_raw_init;
1218                 call->show_format = kretprobe_event_show_format;
1219                 call->define_fields = kretprobe_event_define_fields;
1220         } else {
1221                 tp->event.trace = print_kprobe_event;
1222                 call->raw_init = probe_event_raw_init;
1223                 call->show_format = kprobe_event_show_format;
1224                 call->define_fields = kprobe_event_define_fields;
1225         }
1226         call->event = &tp->event;
1227         call->id = register_ftrace_event(&tp->event);
1228         if (!call->id)
1229                 return -ENODEV;
1230         call->enabled = 1;
1231         call->regfunc = probe_event_enable;
1232         call->unregfunc = probe_event_disable;
1233
1234 #ifdef CONFIG_EVENT_PROFILE
1235         atomic_set(&call->profile_count, -1);
1236         call->profile_enable = probe_profile_enable;
1237         call->profile_disable = probe_profile_disable;
1238 #endif
1239         call->data = tp;
1240         ret = trace_add_event_call(call);
1241         if (ret) {
1242                 pr_info("Failed to register kprobe event: %s\n", call->name);
1243                 unregister_ftrace_event(&tp->event);
1244         }
1245         return ret;
1246 }
1247
1248 static void unregister_probe_event(struct trace_probe *tp)
1249 {
1250         /* tp->event is unregistered in trace_remove_event_call() */
1251         trace_remove_event_call(&tp->call);
1252 }
1253
1254 /* Make a debugfs interface for controling probe points */
1255 static __init int init_kprobe_trace(void)
1256 {
1257         struct dentry *d_tracer;
1258         struct dentry *entry;
1259
1260         d_tracer = tracing_init_dentry();
1261         if (!d_tracer)
1262                 return 0;
1263
1264         entry = debugfs_create_file("kprobe_events", 0644, d_tracer,
1265                                     NULL, &kprobe_events_ops);
1266
1267         /* Event list interface */
1268         if (!entry)
1269                 pr_warning("Could not create debugfs "
1270                            "'kprobe_events' entry\n");
1271
1272         /* Profile interface */
1273         entry = debugfs_create_file("kprobe_profile", 0444, d_tracer,
1274                                     NULL, &kprobe_profile_ops);
1275
1276         if (!entry)
1277                 pr_warning("Could not create debugfs "
1278                            "'kprobe_profile' entry\n");
1279         return 0;
1280 }
1281 fs_initcall(init_kprobe_trace);
1282
1283
1284 #ifdef CONFIG_FTRACE_STARTUP_TEST
1285
1286 static int kprobe_trace_selftest_target(int a1, int a2, int a3,
1287                                         int a4, int a5, int a6)
1288 {
1289         return a1 + a2 + a3 + a4 + a5 + a6;
1290 }
1291
1292 static __init int kprobe_trace_self_tests_init(void)
1293 {
1294         int ret;
1295         int (*target)(int, int, int, int, int, int);
1296
1297         target = kprobe_trace_selftest_target;
1298
1299         pr_info("Testing kprobe tracing: ");
1300
1301         ret = command_trace_probe("p:testprobe kprobe_trace_selftest_target "
1302                                   "a1 a2 a3 a4 a5 a6");
1303         if (WARN_ON_ONCE(ret))
1304                 pr_warning("error enabling function entry\n");
1305
1306         ret = command_trace_probe("r:testprobe2 kprobe_trace_selftest_target "
1307                                   "ra rv");
1308         if (WARN_ON_ONCE(ret))
1309                 pr_warning("error enabling function return\n");
1310
1311         ret = target(1, 2, 3, 4, 5, 6);
1312
1313         cleanup_all_probes();
1314
1315         pr_cont("OK\n");
1316         return 0;
1317 }
1318
1319 late_initcall(kprobe_trace_self_tests_init);
1320
1321 #endif