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_trace_core
 180  */
 181
 182 struct trace_probe {
 183         struct list_head        list;
 184         struct kretprobe        rp;     /* Use rp.kp for kprobe use */
 185         unsigned long           nhit;
 186         const char              *symbol;        /* symbol name */
 187         struct ftrace_event_call        call;
 188         struct trace_event              event;
 189         unsigned int            nr_args;
 190         struct fetch_func       args[];
 191 };
 192
 193 #define SIZEOF_TRACE_PROBE(n)                   \
 194         (offsetof(struct trace_probe, args) +   \
 195         (sizeof(struct fetch_func) * (n)))
 196
 197 static int kprobe_trace_func(struct kprobe *kp, struct pt_regs *regs);
 198 static int kretprobe_trace_func(struct kretprobe_instance *ri,
 199                                 struct pt_regs *regs);
 200
 201 static __kprobes int probe_is_return(struct trace_probe *tp)
 202 {
 203         return tp->rp.handler != NULL;
 204 }
 205
 206 static __kprobes const char *probe_symbol(struct trace_probe *tp)
 207 {
 208         return tp->symbol ? tp->symbol : "unknown";
 209 }
 210
 211 static int probe_arg_string(char *buf, size_t n, struct fetch_func *ff)
 212 {
 213         int ret = -EINVAL;
 214
 215         if (ff->func == fetch_argument)
 216                 ret = snprintf(buf, n, "a%lu", (unsigned long)ff->data);
 217         else if (ff->func == fetch_register) {
 218                 const char *name;
 219                 name = regs_query_register_name((unsigned int)((long)ff->data));
 220                 ret = snprintf(buf, n, "%%%s", name);
 221         } else if (ff->func == fetch_stack)
 222                 ret = snprintf(buf, n, "s%lu", (unsigned long)ff->data);
 223         else if (ff->func == fetch_memory)
 224                 ret = snprintf(buf, n, "@0x%p", ff->data);
 225         else if (ff->func == fetch_symbol) {
 226                 struct symbol_cache *sc = ff->data;
 227                 ret = snprintf(buf, n, "@%s%+ld", sc->symbol, sc->offset);
 228         } else if (ff->func == fetch_retvalue)
 229                 ret = snprintf(buf, n, "rv");
 230         else if (ff->func == fetch_ip)
 231                 ret = snprintf(buf, n, "ra");
 232         else if (ff->func == fetch_stack_address)
 233                 ret = snprintf(buf, n, "sa");
 234         else if (ff->func == fetch_indirect) {
 235                 struct indirect_fetch_data *id = ff->data;
 236                 size_t l = 0;
 237                 ret = snprintf(buf, n, "%+ld(", id->offset);
 238                 if (ret >= n)
 239                         goto end;
 240                 l += ret;
 241                 ret = probe_arg_string(buf + l, n - l, &id->orig);
 242                 if (ret < 0)
 243                         goto end;
 244                 l += ret;
 245                 ret = snprintf(buf + l, n - l, ")");
 246                 ret += l;
 247         }
 248 end:
 249         if (ret >= n)
 250                 return -ENOSPC;
 251         return ret;
 252 }
 253
 254 static int register_probe_event(struct trace_probe *tp);
 255 static void unregister_probe_event(struct trace_probe *tp);
 256
 257 static DEFINE_MUTEX(probe_lock);
 258 static LIST_HEAD(probe_list);
 259
 260 /*
 261  * Allocate new trace_probe and initialize it (including kprobes).
 262  */
 263 static struct trace_probe *alloc_trace_probe(const char *event,
 264                                              void *addr,
 265                                              const char *symbol,
 266                                              unsigned long offs,
 267                                              int nargs, int is_return)
 268 {
 269         struct trace_probe *tp;
 270
 271         tp = kzalloc(SIZEOF_TRACE_PROBE(nargs), GFP_KERNEL);
 272         if (!tp)
 273                 return ERR_PTR(-ENOMEM);
 274
 275         if (symbol) {
 276                 tp->symbol = kstrdup(symbol, GFP_KERNEL);
 277                 if (!tp->symbol)
 278                         goto error;
 279                 tp->rp.kp.symbol_name = tp->symbol;
 280                 tp->rp.kp.offset = offs;
 281         } else
 282                 tp->rp.kp.addr = addr;
 283
 284         /* Set handler here for checking whether this probe is return or not. */
 285         if (is_return)
 286                 tp->rp.handler = kretprobe_trace_func;
 287         else
 288                 tp->rp.kp.pre_handler = kprobe_trace_func;
 289
 290         if (!event)
 291                 goto error;
 292         tp->call.name = kstrdup(event, GFP_KERNEL);
 293         if (!tp->call.name)
 294                 goto error;
 295
 296         INIT_LIST_HEAD(&tp->list);
 297         return tp;
 298 error:
 299         kfree(tp->symbol);
 300         kfree(tp);
 301         return ERR_PTR(-ENOMEM);
 302 }
 303
 304 static void free_trace_probe(struct trace_probe *tp)
 305 {
 306         int i;
 307
 308         for (i = 0; i < tp->nr_args; i++)
 309                 if (tp->args[i].func == fetch_symbol)
 310                         free_symbol_cache(tp->args[i].data);
 311                 else if (tp->args[i].func == fetch_indirect)
 312                         free_indirect_fetch_data(tp->args[i].data);
 313
 314         kfree(tp->call.name);
 315         kfree(tp->symbol);
 316         kfree(tp);
 317 }
 318
 319 static struct trace_probe *find_probe_event(const char *event)
 320 {
 321         struct trace_probe *tp;
 322
 323         list_for_each_entry(tp, &probe_list, list)
 324                 if (!strcmp(tp->call.name, event))
 325                         return tp;
 326         return NULL;
 327 }
 328
 329 static void __unregister_trace_probe(struct trace_probe *tp)
 330 {
 331         if (probe_is_return(tp))
 332                 unregister_kretprobe(&tp->rp);
 333         else
 334                 unregister_kprobe(&tp->rp.kp);
 335 }
 336
 337 /* Unregister a trace_probe and probe_event: call with locking probe_lock */
 338 static void unregister_trace_probe(struct trace_probe *tp)
 339 {
 340         unregister_probe_event(tp);
 341         __unregister_trace_probe(tp);
 342         list_del(&tp->list);
 343 }
 344
 345 /* Register a trace_probe and probe_event */
 346 static int register_trace_probe(struct trace_probe *tp)
 347 {
 348         struct trace_probe *old_tp;
 349         int ret;
 350
 351         mutex_lock(&probe_lock);
 352
 353         if (probe_is_return(tp))
 354                 ret = register_kretprobe(&tp->rp);
 355         else
 356                 ret = register_kprobe(&tp->rp.kp);
 357
 358         if (ret) {
 359                 pr_warning("Could not insert probe(%d)\n", ret);
 360                 if (ret == -EILSEQ) {
 361                         pr_warning("Probing address(0x%p) is not an "
 362                                    "instruction boundary.\n",
 363                                    tp->rp.kp.addr);
 364                         ret = -EINVAL;
 365                 }
 366                 goto end;
 367         }
 368         /* register as an event */
 369         old_tp = find_probe_event(tp->call.name);
 370         if (old_tp) {
 371                 /* delete old event */
 372                 unregister_trace_probe(old_tp);
 373                 free_trace_probe(old_tp);
 374         }
 375         ret = register_probe_event(tp);
 376         if (ret) {
 377                 pr_warning("Faild to register probe event(%d)\n", ret);
 378                 __unregister_trace_probe(tp);
 379         }
 380         list_add_tail(&tp->list, &probe_list);
 381 end:
 382         mutex_unlock(&probe_lock);
 383         return ret;
 384 }
 385
 386 /* Split symbol and offset. */
 387 static int split_symbol_offset(char *symbol, unsigned long *offset)
 388 {
 389         char *tmp;
 390         int ret;
 391
 392         if (!offset)
 393                 return -EINVAL;
 394
 395         tmp = strchr(symbol, '+');
 396         if (tmp) {
 397                 /* skip sign because strict_strtol doesn't accept '+' */
 398                 ret = strict_strtoul(tmp + 1, 0, offset);
 399                 if (ret)
 400                         return ret;
 401                 *tmp = '\0';
 402         } else
 403                 *offset = 0;
 404         return 0;
 405 }
 406
 407 #define PARAM_MAX_ARGS 16
 408 #define PARAM_MAX_STACK (THREAD_SIZE / sizeof(unsigned long))
 409
 410 static int parse_probe_arg(char *arg, struct fetch_func *ff, int is_return)
 411 {
 412         int ret = 0;
 413         unsigned long param;
 414         long offset;
 415         char *tmp;
 416
 417         switch (arg[0]) {
 418         case 'a':       /* argument */
 419                 ret = strict_strtoul(arg + 1, 10, &param);
 420                 if (ret || param > PARAM_MAX_ARGS)
 421                         ret = -EINVAL;
 422                 else {
 423                         ff->func = fetch_argument;
 424                         ff->data = (void *)param;
 425                 }
 426                 break;
 427         case 'r':       /* retval or retaddr */
 428                 if (is_return && arg[1] == 'v') {
 429                         ff->func = fetch_retvalue;
 430                         ff->data = NULL;
 431                 } else if (is_return && arg[1] == 'a') {
 432                         ff->func = fetch_ip;
 433                         ff->data = NULL;
 434                 } else
 435                         ret = -EINVAL;
 436                 break;
 437         case '%':       /* named register */
 438                 ret = regs_query_register_offset(arg + 1);
 439                 if (ret >= 0) {
 440                         ff->func = fetch_register;
 441                         ff->data = (void *)(unsigned long)ret;
 442                         ret = 0;
 443                 }
 444                 break;
 445         case 's':       /* stack */
 446                 if (arg[1] == 'a') {
 447                         ff->func = fetch_stack_address;
 448                         ff->data = NULL;
 449                 } else {
 450                         ret = strict_strtoul(arg + 1, 10, &param);
 451                         if (ret || param > PARAM_MAX_STACK)
 452                                 ret = -EINVAL;
 453                         else {
 454                                 ff->func = fetch_stack;
 455                                 ff->data = (void *)param;
 456                         }
 457                 }
 458                 break;
 459         case '@':       /* memory or symbol */
 460                 if (isdigit(arg[1])) {
 461                         ret = strict_strtoul(arg + 1, 0, &param);
 462                         if (ret)
 463                                 break;
 464                         ff->func = fetch_memory;
 465                         ff->data = (void *)param;
 466                 } else {
 467                         ret = split_symbol_offset(arg + 1, &offset);
 468                         if (ret)
 469                                 break;
 470                         ff->data = alloc_symbol_cache(arg + 1,
 471                                                               offset);
 472                         if (ff->data)
 473                                 ff->func = fetch_symbol;
 474                         else
 475                                 ret = -EINVAL;
 476                 }
 477                 break;
 478         case '+':       /* indirect memory */
 479         case '-':
 480                 tmp = strchr(arg, '(');
 481                 if (!tmp) {
 482                         ret = -EINVAL;
 483                         break;
 484                 }
 485                 *tmp = '\0';
 486                 ret = strict_strtol(arg + 1, 0, &offset);
 487                 if (ret)
 488                         break;
 489                 if (arg[0] == '-')
 490                         offset = -offset;
 491                 arg = tmp + 1;
 492                 tmp = strrchr(arg, ')');
 493                 if (tmp) {
 494                         struct indirect_fetch_data *id;
 495                         *tmp = '\0';
 496                         id = kzalloc(sizeof(struct indirect_fetch_data),
 497                                      GFP_KERNEL);
 498                         if (!id)
 499                                 return -ENOMEM;
 500                         id->offset = offset;
 501                         ret = parse_probe_arg(arg, &id->orig, is_return);
 502                         if (ret)
 503                                 kfree(id);
 504                         else {
 505                                 ff->func = fetch_indirect;
 506                                 ff->data = (void *)id;
 507                         }
 508                 } else
 509                         ret = -EINVAL;
 510                 break;
 511         default:
 512                 /* TODO: support custom handler */
 513                 ret = -EINVAL;
 514         }
 515         return ret;
 516 }
 517
 518 static int create_trace_probe(int argc, char **argv)
 519 {
 520         /*
 521          * Argument syntax:
 522          *  - Add kprobe: p[:EVENT] SYMBOL[+OFFS]|ADDRESS [FETCHARGS]
 523          *  - Add kretprobe: r[:EVENT] SYMBOL[+0] [FETCHARGS]
 524          * Fetch args:
 525          *  aN  : fetch Nth of function argument. (N:0-)
 526          *  rv  : fetch return value
 527          *  ra  : fetch return address
 528          *  sa  : fetch stack address
 529          *  sN  : fetch Nth of stack (N:0-)
 530          *  @ADDR       : fetch memory at ADDR (ADDR should be in kernel)
 531          *  @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
 532          *  %REG        : fetch register REG
 533          * Indirect memory fetch:
 534          *  +|-offs(ARG) : fetch memory at ARG +|- offs address.
 535          */
 536         struct trace_probe *tp;
 537         int i, ret = 0;
 538         int is_return = 0;
 539         char *symbol = NULL, *event = NULL;
 540         unsigned long offset = 0;
 541         void *addr = NULL;
 542         char buf[MAX_EVENT_NAME_LEN];
 543
 544         if (argc < 2)
 545                 return -EINVAL;
 546
 547         if (argv[0][0] == 'p')
 548                 is_return = 0;
 549         else if (argv[0][0] == 'r')
 550                 is_return = 1;
 551         else
 552                 return -EINVAL;
 553
 554         if (argv[0][1] == ':') {
 555                 event = &argv[0][2];
 556                 if (strlen(event) == 0) {
 557                         pr_info("Event name is not specifiled\n");
 558                         return -EINVAL;
 559                 }
 560         }
 561
 562         if (isdigit(argv[1][0])) {
 563                 if (is_return)
 564                         return -EINVAL;
 565                 /* an address specified */
 566                 ret = strict_strtoul(&argv[0][2], 0, (unsigned long *)&addr);
 567                 if (ret)
 568                         return ret;
 569         } else {
 570                 /* a symbol specified */
 571                 symbol = argv[1];
 572                 /* TODO: support .init module functions */
 573                 ret = split_symbol_offset(symbol, &offset);
 574                 if (ret)
 575                         return ret;
 576                 if (offset && is_return)
 577                         return -EINVAL;
 578         }
 579         argc -= 2; argv += 2;
 580
 581         /* setup a probe */
 582         if (!event) {
 583                 /* Make a new event name */
 584                 if (symbol)
 585                         snprintf(buf, MAX_EVENT_NAME_LEN, "%c@%s%+ld",
 586                                  is_return ? 'r' : 'p', symbol, offset);
 587                 else
 588                         snprintf(buf, MAX_EVENT_NAME_LEN, "%c@0x%p",
 589                                  is_return ? 'r' : 'p', addr);
 590                 event = buf;
 591         }
 592         tp = alloc_trace_probe(event, addr, symbol, offset, argc, is_return);
 593         if (IS_ERR(tp))
 594                 return PTR_ERR(tp);
 595
 596         /* parse arguments */
 597         ret = 0;
 598         for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
 599                 if (strlen(argv[i]) > MAX_ARGSTR_LEN) {
 600                         pr_info("Argument%d(%s) is too long.\n", i, argv[i]);
 601                         ret = -ENOSPC;
 602                         goto error;
 603                 }
 604                 ret = parse_probe_arg(argv[i], &tp->args[i], is_return);
 605                 if (ret)
 606                         goto error;
 607         }
 608         tp->nr_args = i;
 609
 610         ret = register_trace_probe(tp);
 611         if (ret)
 612                 goto error;
 613         return 0;
 614
 615 error:
 616         free_trace_probe(tp);
 617         return ret;
 618 }
 619
 620 static void cleanup_all_probes(void)
 621 {
 622         struct trace_probe *tp;
 623
 624         mutex_lock(&probe_lock);
 625         /* TODO: Use batch unregistration */
 626         while (!list_empty(&probe_list)) {
 627                 tp = list_entry(probe_list.next, struct trace_probe, list);
 628                 unregister_trace_probe(tp);
 629                 free_trace_probe(tp);
 630         }
 631         mutex_unlock(&probe_lock);
 632 }
 633
 634
 635 /* Probes listing interfaces */
 636 static void *probes_seq_start(struct seq_file *m, loff_t *pos)
 637 {
 638         mutex_lock(&probe_lock);
 639         return seq_list_start(&probe_list, *pos);
 640 }
 641
 642 static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
 643 {
 644         return seq_list_next(v, &probe_list, pos);
 645 }
 646
 647 static void probes_seq_stop(struct seq_file *m, void *v)
 648 {
 649         mutex_unlock(&probe_lock);
 650 }
 651
 652 static int probes_seq_show(struct seq_file *m, void *v)
 653 {
 654         struct trace_probe *tp = v;
 655         int i, ret;
 656         char buf[MAX_ARGSTR_LEN + 1];
 657
 658         seq_printf(m, "%c", probe_is_return(tp) ? 'r' : 'p');
 659         seq_printf(m, ":%s", tp->call.name);
 660
 661         if (tp->symbol)
 662                 seq_printf(m, " %s+%u", probe_symbol(tp), tp->rp.kp.offset);
 663         else
 664                 seq_printf(m, " 0x%p", tp->rp.kp.addr);
 665
 666         for (i = 0; i < tp->nr_args; i++) {
 667                 ret = probe_arg_string(buf, MAX_ARGSTR_LEN, &tp->args[i]);
 668                 if (ret < 0) {
 669                         pr_warning("Argument%d decoding error(%d).\n", i, ret);
 670                         return ret;
 671                 }
 672                 seq_printf(m, " %s", buf);
 673         }
 674         seq_printf(m, "\n");
 675         return 0;
 676 }
 677
 678 static const struct seq_operations probes_seq_op = {
 679         .start  = probes_seq_start,
 680         .next   = probes_seq_next,
 681         .stop   = probes_seq_stop,
 682         .show   = probes_seq_show
 683 };
 684
 685 static int probes_open(struct inode *inode, struct file *file)
 686 {
 687         if ((file->f_mode & FMODE_WRITE) &&
 688             (file->f_flags & O_TRUNC))
 689                 cleanup_all_probes();
 690
 691         return seq_open(file, &probes_seq_op);
 692 }
 693
 694 static int command_trace_probe(const char *buf)
 695 {
 696         char **argv;
 697         int argc = 0, ret = 0;
 698
 699         argv = argv_split(GFP_KERNEL, buf, &argc);
 700         if (!argv)
 701                 return -ENOMEM;
 702
 703         if (argc)
 704                 ret = create_trace_probe(argc, argv);
 705
 706         argv_free(argv);
 707         return ret;
 708 }
 709
 710 #define WRITE_BUFSIZE 128
 711
 712 static ssize_t probes_write(struct file *file, const char __user *buffer,
 713                             size_t count, loff_t *ppos)
 714 {
 715         char *kbuf, *tmp;
 716         int ret;
 717         size_t done;
 718         size_t size;
 719
 720         kbuf = kmalloc(WRITE_BUFSIZE, GFP_KERNEL);
 721         if (!kbuf)
 722                 return -ENOMEM;
 723
 724         ret = done = 0;
 725         while (done < count) {
 726                 size = count - done;
 727                 if (size >= WRITE_BUFSIZE)
 728                         size = WRITE_BUFSIZE - 1;
 729                 if (copy_from_user(kbuf, buffer + done, size)) {
 730                         ret = -EFAULT;
 731                         goto out;
 732                 }
 733                 kbuf[size] = '\0';
 734                 tmp = strchr(kbuf, '\n');
 735                 if (tmp) {
 736                         *tmp = '\0';
 737                         size = tmp - kbuf + 1;
 738                 } else if (done + size < count) {
 739                         pr_warning("Line length is too long: "
 740                                    "Should be less than %d.", WRITE_BUFSIZE);
 741                         ret = -EINVAL;
 742                         goto out;
 743                 }
 744                 done += size;
 745                 /* Remove comments */
 746                 tmp = strchr(kbuf, '#');
 747                 if (tmp)
 748                         *tmp = '\0';
 749
 750                 ret = command_trace_probe(kbuf);
 751                 if (ret)
 752                         goto out;
 753         }
 754         ret = done;
 755 out:
 756         kfree(kbuf);
 757         return ret;
 758 }
 759
 760 static const struct file_operations kprobe_events_ops = {
 761         .owner          = THIS_MODULE,
 762         .open           = probes_open,
 763         .read           = seq_read,
 764         .llseek         = seq_lseek,
 765         .release        = seq_release,
 766         .write          = probes_write,
 767 };
 768
 769 /* Probes profiling interfaces */
 770 static int probes_profile_seq_show(struct seq_file *m, void *v)
 771 {
 772         struct trace_probe *tp = v;
 773
 774         seq_printf(m, "  %-44s %15lu %15lu\n", tp->call.name, tp->nhit,
 775                    tp->rp.kp.nmissed);
 776
 777         return 0;
 778 }
 779
 780 static const struct seq_operations profile_seq_op = {
 781         .start  = probes_seq_start,
 782         .next   = probes_seq_next,
 783         .stop   = probes_seq_stop,
 784         .show   = probes_profile_seq_show
 785 };
 786
 787 static int profile_open(struct inode *inode, struct file *file)
 788 {
 789         return seq_open(file, &profile_seq_op);
 790 }
 791
 792 static const struct file_operations kprobe_profile_ops = {
 793         .owner          = THIS_MODULE,
 794         .open           = profile_open,
 795         .read           = seq_read,
 796         .llseek         = seq_lseek,
 797         .release        = seq_release,
 798 };
 799
 800 /* Kprobe handler */
 801 static __kprobes int kprobe_trace_func(struct kprobe *kp, struct pt_regs *regs)
 802 {
 803         struct trace_probe *tp = container_of(kp, struct trace_probe, rp.kp);
 804         struct kprobe_trace_entry *entry;
 805         struct ring_buffer_event *event;
 806         struct ring_buffer *buffer;
 807         int size, i, pc;
 808         unsigned long irq_flags;
 809         struct ftrace_event_call *call = &tp->call;
 810
 811         tp->nhit++;
 812
 813         local_save_flags(irq_flags);
 814         pc = preempt_count();
 815
 816         size = SIZEOF_KPROBE_TRACE_ENTRY(tp->nr_args);
 817
 818         event = trace_current_buffer_lock_reserve(&buffer, call->id, size,
 819                                                   irq_flags, pc);
 820         if (!event)
 821                 return 0;
 822
 823         entry = ring_buffer_event_data(event);
 824         entry->nargs = tp->nr_args;
 825         entry->ip = (unsigned long)kp->addr;
 826         for (i = 0; i < tp->nr_args; i++)
 827                 entry->args[i] = call_fetch(&tp->args[i], regs);
 828
 829         if (!filter_current_check_discard(buffer, call, entry, event))
 830                 trace_nowake_buffer_unlock_commit(buffer, event, irq_flags, pc);
 831         return 0;
 832 }
 833
 834 /* Kretprobe handler */
 835 static __kprobes int kretprobe_trace_func(struct kretprobe_instance *ri,
 836                                           struct pt_regs *regs)
 837 {
 838         struct trace_probe *tp = container_of(ri->rp, struct trace_probe, rp);
 839         struct kretprobe_trace_entry *entry;
 840         struct ring_buffer_event *event;
 841         struct ring_buffer *buffer;
 842         int size, i, pc;
 843         unsigned long irq_flags;
 844         struct ftrace_event_call *call = &tp->call;
 845
 846         local_save_flags(irq_flags);
 847         pc = preempt_count();
 848
 849         size = SIZEOF_KRETPROBE_TRACE_ENTRY(tp->nr_args);
 850
 851         event = trace_current_buffer_lock_reserve(&buffer, call->id, size,
 852                                                   irq_flags, pc);
 853         if (!event)
 854                 return 0;
 855
 856         entry = ring_buffer_event_data(event);
 857         entry->nargs = tp->nr_args;
 858         entry->func = (unsigned long)tp->rp.kp.addr;
 859         entry->ret_ip = (unsigned long)ri->ret_addr;
 860         for (i = 0; i < tp->nr_args; i++)
 861                 entry->args[i] = call_fetch(&tp->args[i], regs);
 862
 863         if (!filter_current_check_discard(buffer, call, entry, event))
 864                 trace_nowake_buffer_unlock_commit(buffer, event, irq_flags, pc);
 865
 866         return 0;
 867 }
 868
 869 /* Event entry printers */
 870 enum print_line_t
 871 print_kprobe_event(struct trace_iterator *iter, int flags)
 872 {
 873         struct kprobe_trace_entry *field;
 874         struct trace_seq *s = &iter->seq;
 875         int i;
 876
 877         field = (struct kprobe_trace_entry *)iter->ent;
 878
 879         if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
 880                 goto partial;
 881
 882         if (!trace_seq_puts(s, ":"))
 883                 goto partial;
 884
 885         for (i = 0; i < field->nargs; i++)
 886                 if (!trace_seq_printf(s, " 0x%lx", field->args[i]))
 887                         goto partial;
 888
 889         if (!trace_seq_puts(s, "\n"))
 890                 goto partial;
 891
 892         return TRACE_TYPE_HANDLED;
 893 partial:
 894         return TRACE_TYPE_PARTIAL_LINE;
 895 }
 896
 897 enum print_line_t
 898 print_kretprobe_event(struct trace_iterator *iter, int flags)
 899 {
 900         struct kretprobe_trace_entry *field;
 901         struct trace_seq *s = &iter->seq;
 902         int i;
 903
 904         field = (struct kretprobe_trace_entry *)iter->ent;
 905
 906         if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
 907                 goto partial;
 908
 909         if (!trace_seq_puts(s, " <- "))
 910                 goto partial;
 911
 912         if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
 913                 goto partial;
 914
 915         if (!trace_seq_puts(s, ":"))
 916                 goto partial;
 917
 918         for (i = 0; i < field->nargs; i++)
 919                 if (!trace_seq_printf(s, " 0x%lx", field->args[i]))
 920                         goto partial;
 921
 922         if (!trace_seq_puts(s, "\n"))
 923                 goto partial;
 924
 925         return TRACE_TYPE_HANDLED;
 926 partial:
 927         return TRACE_TYPE_PARTIAL_LINE;
 928 }
 929
 930 static int probe_event_enable(struct ftrace_event_call *call)
 931 {
 932         struct trace_probe *tp = (struct trace_probe *)call->data;
 933
 934         if (probe_is_return(tp)) {
 935                 tp->rp.handler = kretprobe_trace_func;
 936                 return enable_kretprobe(&tp->rp);
 937         } else {
 938                 tp->rp.kp.pre_handler = kprobe_trace_func;
 939                 return enable_kprobe(&tp->rp.kp);
 940         }
 941 }
 942
 943 static void probe_event_disable(struct ftrace_event_call *call)
 944 {
 945         struct trace_probe *tp = (struct trace_probe *)call->data;
 946
 947         if (probe_is_return(tp))
 948                 disable_kretprobe(&tp->rp);
 949         else
 950                 disable_kprobe(&tp->rp.kp);
 951 }
 952
 953 static int probe_event_raw_init(struct ftrace_event_call *event_call)
 954 {
 955         INIT_LIST_HEAD(&event_call->fields);
 956
 957         return 0;
 958 }
 959
 960 #undef DEFINE_FIELD
 961 #define DEFINE_FIELD(type, item, name, is_signed)                       \
 962         do {                                                            \
 963                 ret = trace_define_field(event_call, #type, name,       \
 964                                          offsetof(typeof(field), item), \
 965                                          sizeof(field.item), is_signed, \
 966                                          FILTER_OTHER);                 \
 967                 if (ret)                                                \
 968                         return ret;                                     \
 969         } while (0)
 970
 971 static int kprobe_event_define_fields(struct ftrace_event_call *event_call)
 972 {
 973         int ret, i;
 974         struct kprobe_trace_entry field;
 975         char buf[MAX_ARGSTR_LEN + 1];
 976         struct trace_probe *tp = (struct trace_probe *)event_call->data;
 977
 978         ret = trace_define_common_fields(event_call);
 979         if (!ret)
 980                 return ret;
 981
 982         DEFINE_FIELD(unsigned long, ip, "ip", 0);
 983         DEFINE_FIELD(int, nargs, "nargs", 1);
 984         for (i = 0; i < tp->nr_args; i++) {
 985                 /* Set argN as a field */
 986                 sprintf(buf, "arg%d", i);
 987                 DEFINE_FIELD(unsigned long, args[i], buf, 0);
 988                 /* Set argument string as an alias field */
 989                 ret = probe_arg_string(buf, MAX_ARGSTR_LEN, &tp->args[i]);
 990                 if (ret < 0)
 991                         return ret;
 992                 DEFINE_FIELD(unsigned long, args[i], buf, 0);
 993         }
 994         return 0;
 995 }
 996
 997 static int kretprobe_event_define_fields(struct ftrace_event_call *event_call)
 998 {
 999         int ret, i;
1000         struct kretprobe_trace_entry field;
1001         char buf[MAX_ARGSTR_LEN + 1];
1002         struct trace_probe *tp = (struct trace_probe *)event_call->data;
1003
1004         ret = trace_define_common_fields(event_call);
1005         if (!ret)
1006                 return ret;
1007
1008         DEFINE_FIELD(unsigned long, func, "func", 0);
1009         DEFINE_FIELD(unsigned long, ret_ip, "ret_ip", 0);
1010         DEFINE_FIELD(int, nargs, "nargs", 1);
1011         for (i = 0; i < tp->nr_args; i++) {
1012                 /* Set argN as a field */
1013                 sprintf(buf, "arg%d", i);
1014                 DEFINE_FIELD(unsigned long, args[i], buf, 0);
1015                 /* Set argument string as an alias field */
1016                 ret = probe_arg_string(buf, MAX_ARGSTR_LEN, &tp->args[i]);
1017                 if (ret < 0)
1018                         return ret;
1019                 DEFINE_FIELD(unsigned long, args[i], buf, 0);
1020         }
1021         return 0;
1022 }
1023
1024 static int __probe_event_show_format(struct trace_seq *s,
1025                                      struct trace_probe *tp, const char *fmt,
1026                                      const char *arg)
1027 {
1028         int i, ret;
1029         char buf[MAX_ARGSTR_LEN + 1];
1030
1031         /* Show aliases */
1032         for (i = 0; i < tp->nr_args; i++) {
1033                 ret = probe_arg_string(buf, MAX_ARGSTR_LEN, &tp->args[i]);
1034                 if (ret < 0)
1035                         return ret;
1036                 if (!trace_seq_printf(s, "\talias: %s;\toriginal: arg%d;\n",
1037                                       buf, i))
1038                         return 0;
1039         }
1040         /* Show format */
1041         if (!trace_seq_printf(s, "\nprint fmt: \"%s", fmt))
1042                 return 0;
1043
1044         for (i = 0; i < tp->nr_args; i++)
1045                 if (!trace_seq_puts(s, " 0x%lx"))
1046                         return 0;
1047
1048         if (!trace_seq_printf(s, "\", %s", arg))
1049                 return 0;
1050
1051         for (i = 0; i < tp->nr_args; i++)
1052                 if (!trace_seq_printf(s, ", arg%d", i))
1053                         return 0;
1054
1055         return trace_seq_puts(s, "\n");
1056 }
1057
1058 #undef SHOW_FIELD
1059 #define SHOW_FIELD(type, item, name)                                    \
1060         do {                                                            \
1061                 ret = trace_seq_printf(s, "\tfield: " #type " %s;\t"    \
1062                                 "offset:%u;\tsize:%u;\n", name,         \
1063                                 (unsigned int)offsetof(typeof(field), item),\
1064                                 (unsigned int)sizeof(type));            \
1065                 if (!ret)                                               \
1066                         return 0;                                       \
1067         } while (0)
1068
1069 static int kprobe_event_show_format(struct ftrace_event_call *call,
1070                                     struct trace_seq *s)
1071 {
1072         struct kprobe_trace_entry field __attribute__((unused));
1073         int ret, i;
1074         char buf[8];
1075         struct trace_probe *tp = (struct trace_probe *)call->data;
1076
1077         SHOW_FIELD(unsigned long, ip, "ip");
1078         SHOW_FIELD(int, nargs, "nargs");
1079
1080         /* Show fields */
1081         for (i = 0; i < tp->nr_args; i++) {
1082                 sprintf(buf, "arg%d", i);
1083                 SHOW_FIELD(unsigned long, args[i], buf);
1084         }
1085         trace_seq_puts(s, "\n");
1086
1087         return __probe_event_show_format(s, tp, "%lx:", "ip");
1088 }
1089
1090 static int kretprobe_event_show_format(struct ftrace_event_call *call,
1091                                        struct trace_seq *s)
1092 {
1093         struct kretprobe_trace_entry field __attribute__((unused));
1094         int ret, i;
1095         char buf[8];
1096         struct trace_probe *tp = (struct trace_probe *)call->data;
1097
1098         SHOW_FIELD(unsigned long, func, "func");
1099         SHOW_FIELD(unsigned long, ret_ip, "ret_ip");
1100         SHOW_FIELD(int, nargs, "nargs");
1101
1102         /* Show fields */
1103         for (i = 0; i < tp->nr_args; i++) {
1104                 sprintf(buf, "arg%d", i);
1105                 SHOW_FIELD(unsigned long, args[i], buf);
1106         }
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], 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], 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