kernel/trace/trace_kprobe.c

   1 /*
   2  * Kprobes-based tracing events
   3  *
   4  * Created by Masami Hiramatsu <mhiramat@redhat.com>
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 as
   8  * published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful,
  11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  * GNU General Public License for more details.
  14  *
  15  * You should have received a copy of the GNU General Public License
  16  * along with this program; if not, write to the Free Software
  17  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
  18  */
  19
  20 #include <linux/module.h>
  21 #include <linux/uaccess.h>
  22
  23 #include "trace_probe.h"
  24
  25 #define KPROBE_EVENT_SYSTEM "kprobes"
  26
  27 /**
  28  * Kprobe event core functions
  29  */
  30
  31 struct trace_probe {
  32         struct list_head        list;
  33         struct kretprobe        rp;     /* Use rp.kp for kprobe use */
  34         unsigned long           nhit;
  35         unsigned int            flags;  /* For TP_FLAG_* */
  36         const char              *symbol;        /* symbol name */
  37         struct ftrace_event_class       class;
  38         struct ftrace_event_call        call;
  39         ssize_t                 size;           /* trace entry size */
  40         unsigned int            nr_args;
  41         struct probe_arg        args[];
  42 };
  43
  44 #define SIZEOF_TRACE_PROBE(n)                   \
  45         (offsetof(struct trace_probe, args) +   \
  46         (sizeof(struct probe_arg) * (n)))
  47
  48
  49 static __kprobes int trace_probe_is_return(struct trace_probe *tp)
  50 {
  51         return tp->rp.handler != NULL;
  52 }
  53
  54 static __kprobes const char *trace_probe_symbol(struct trace_probe *tp)
  55 {
  56         return tp->symbol ? tp->symbol : "unknown";
  57 }
  58
  59 static __kprobes unsigned long trace_probe_offset(struct trace_probe *tp)
  60 {
  61         return tp->rp.kp.offset;
  62 }
  63
  64 static __kprobes bool trace_probe_is_enabled(struct trace_probe *tp)
  65 {
  66         return !!(tp->flags & (TP_FLAG_TRACE | TP_FLAG_PROFILE));
  67 }
  68
  69 static __kprobes bool trace_probe_is_registered(struct trace_probe *tp)
  70 {
  71         return !!(tp->flags & TP_FLAG_REGISTERED);
  72 }
  73
  74 static __kprobes bool trace_probe_has_gone(struct trace_probe *tp)
  75 {
  76         return !!(kprobe_gone(&tp->rp.kp));
  77 }
  78
  79 static __kprobes bool trace_probe_within_module(struct trace_probe *tp,
  80                                                 struct module *mod)
  81 {
  82         int len = strlen(mod->name);
  83         const char *name = trace_probe_symbol(tp);
  84         return strncmp(mod->name, name, len) == 0 && name[len] == ':';
  85 }
  86
  87 static __kprobes bool trace_probe_is_on_module(struct trace_probe *tp)
  88 {
  89         return !!strchr(trace_probe_symbol(tp), ':');
  90 }
  91
  92 static int register_probe_event(struct trace_probe *tp);
  93 static void unregister_probe_event(struct trace_probe *tp);
  94
  95 static DEFINE_MUTEX(probe_lock);
  96 static LIST_HEAD(probe_list);
  97
  98 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
  99 static int kretprobe_dispatcher(struct kretprobe_instance *ri,
 100                                 struct pt_regs *regs);
 101
 102 /*
 103  * Allocate new trace_probe and initialize it (including kprobes).
 104  */
 105 static struct trace_probe *alloc_trace_probe(const char *group,
 106                                              const char *event,
 107                                              void *addr,
 108                                              const char *symbol,
 109                                              unsigned long offs,
 110                                              int nargs, bool is_return)
 111 {
 112         struct trace_probe *tp;
 113         int ret = -ENOMEM;
 114
 115         tp = kzalloc(SIZEOF_TRACE_PROBE(nargs), GFP_KERNEL);
 116         if (!tp)
 117                 return ERR_PTR(ret);
 118
 119         if (symbol) {
 120                 tp->symbol = kstrdup(symbol, GFP_KERNEL);
 121                 if (!tp->symbol)
 122                         goto error;
 123                 tp->rp.kp.symbol_name = tp->symbol;
 124                 tp->rp.kp.offset = offs;
 125         } else
 126                 tp->rp.kp.addr = addr;
 127
 128         if (is_return)
 129                 tp->rp.handler = kretprobe_dispatcher;
 130         else
 131                 tp->rp.kp.pre_handler = kprobe_dispatcher;
 132
 133         if (!event || !is_good_name(event)) {
 134                 ret = -EINVAL;
 135                 goto error;
 136         }
 137
 138         tp->call.class = &tp->class;
 139         tp->call.name = kstrdup(event, GFP_KERNEL);
 140         if (!tp->call.name)
 141                 goto error;
 142
 143         if (!group || !is_good_name(group)) {
 144                 ret = -EINVAL;
 145                 goto error;
 146         }
 147
 148         tp->class.system = kstrdup(group, GFP_KERNEL);
 149         if (!tp->class.system)
 150                 goto error;
 151
 152         INIT_LIST_HEAD(&tp->list);
 153         return tp;
 154 error:
 155         kfree(tp->call.name);
 156         kfree(tp->symbol);
 157         kfree(tp);
 158         return ERR_PTR(ret);
 159 }
 160
 161 static void free_trace_probe(struct trace_probe *tp)
 162 {
 163         int i;
 164
 165         for (i = 0; i < tp->nr_args; i++)
 166                 traceprobe_free_probe_arg(&tp->args[i]);
 167
 168         kfree(tp->call.class->system);
 169         kfree(tp->call.name);
 170         kfree(tp->symbol);
 171         kfree(tp);
 172 }
 173
 174 static struct trace_probe *find_trace_probe(const char *event,
 175                                             const char *group)
 176 {
 177         struct trace_probe *tp;
 178
 179         list_for_each_entry(tp, &probe_list, list)
 180                 if (strcmp(tp->call.name, event) == 0 &&
 181                     strcmp(tp->call.class->system, group) == 0)
 182                         return tp;
 183         return NULL;
 184 }
 185
 186 /* Enable trace_probe - @flag must be TP_FLAG_TRACE or TP_FLAG_PROFILE */
 187 static int enable_trace_probe(struct trace_probe *tp, int flag)
 188 {
 189         int ret = 0;
 190
 191         tp->flags |= flag;
 192         if (trace_probe_is_enabled(tp) && trace_probe_is_registered(tp) &&
 193             !trace_probe_has_gone(tp)) {
 194                 if (trace_probe_is_return(tp))
 195                         ret = enable_kretprobe(&tp->rp);
 196                 else
 197                         ret = enable_kprobe(&tp->rp.kp);
 198         }
 199
 200         return ret;
 201 }
 202
 203 /* Disable trace_probe - @flag must be TP_FLAG_TRACE or TP_FLAG_PROFILE */
 204 static void disable_trace_probe(struct trace_probe *tp, int flag)
 205 {
 206         tp->flags &= ~flag;
 207         if (!trace_probe_is_enabled(tp) && trace_probe_is_registered(tp)) {
 208                 if (trace_probe_is_return(tp))
 209                         disable_kretprobe(&tp->rp);
 210                 else
 211                         disable_kprobe(&tp->rp.kp);
 212         }
 213 }
 214
 215 /* Internal register function - just handle k*probes and flags */
 216 static int __register_trace_probe(struct trace_probe *tp)
 217 {
 218         int i, ret;
 219
 220         if (trace_probe_is_registered(tp))
 221                 return -EINVAL;
 222
 223         for (i = 0; i < tp->nr_args; i++)
 224                 traceprobe_update_arg(&tp->args[i]);
 225
 226         /* Set/clear disabled flag according to tp->flag */
 227         if (trace_probe_is_enabled(tp))
 228                 tp->rp.kp.flags &= ~KPROBE_FLAG_DISABLED;
 229         else
 230                 tp->rp.kp.flags |= KPROBE_FLAG_DISABLED;
 231
 232         if (trace_probe_is_return(tp))
 233                 ret = register_kretprobe(&tp->rp);
 234         else
 235                 ret = register_kprobe(&tp->rp.kp);
 236
 237         if (ret == 0)
 238                 tp->flags |= TP_FLAG_REGISTERED;
 239         else {
 240                 pr_warning("Could not insert probe at %s+%lu: %d\n",
 241                            trace_probe_symbol(tp), trace_probe_offset(tp), ret);
 242                 if (ret == -ENOENT && trace_probe_is_on_module(tp)) {
 243                         pr_warning("This probe might be able to register after"
 244                                    "target module is loaded. Continue.\n");
 245                         ret = 0;
 246                 } else if (ret == -EILSEQ) {
 247                         pr_warning("Probing address(0x%p) is not an "
 248                                    "instruction boundary.\n",
 249                                    tp->rp.kp.addr);
 250                         ret = -EINVAL;
 251                 }
 252         }
 253
 254         return ret;
 255 }
 256
 257 /* Internal unregister function - just handle k*probes and flags */
 258 static void __unregister_trace_probe(struct trace_probe *tp)
 259 {
 260         if (trace_probe_is_registered(tp)) {
 261                 if (trace_probe_is_return(tp))
 262                         unregister_kretprobe(&tp->rp);
 263                 else
 264                         unregister_kprobe(&tp->rp.kp);
 265                 tp->flags &= ~TP_FLAG_REGISTERED;
 266                 /* Cleanup kprobe for reuse */
 267                 if (tp->rp.kp.symbol_name)
 268                         tp->rp.kp.addr = NULL;
 269         }
 270 }
 271
 272 /* Unregister a trace_probe and probe_event: call with locking probe_lock */
 273 static int unregister_trace_probe(struct trace_probe *tp)
 274 {
 275         /* Enabled event can not be unregistered */
 276         if (trace_probe_is_enabled(tp))
 277                 return -EBUSY;
 278
 279         __unregister_trace_probe(tp);
 280         list_del(&tp->list);
 281         unregister_probe_event(tp);
 282
 283         return 0;
 284 }
 285
 286 /* Register a trace_probe and probe_event */
 287 static int register_trace_probe(struct trace_probe *tp)
 288 {
 289         struct trace_probe *old_tp;
 290         int ret;
 291
 292         mutex_lock(&probe_lock);
 293
 294         /* Delete old (same name) event if exist */
 295         old_tp = find_trace_probe(tp->call.name, tp->call.class->system);
 296         if (old_tp) {
 297                 ret = unregister_trace_probe(old_tp);
 298                 if (ret < 0)
 299                         goto end;
 300                 free_trace_probe(old_tp);
 301         }
 302
 303         /* Register new event */
 304         ret = register_probe_event(tp);
 305         if (ret) {
 306                 pr_warning("Failed to register probe event(%d)\n", ret);
 307                 goto end;
 308         }
 309
 310         /* Register k*probe */
 311         ret = __register_trace_probe(tp);
 312         if (ret < 0)
 313                 unregister_probe_event(tp);
 314         else
 315                 list_add_tail(&tp->list, &probe_list);
 316
 317 end:
 318         mutex_unlock(&probe_lock);
 319         return ret;
 320 }
 321
 322 /* Module notifier call back, checking event on the module */
 323 static int trace_probe_module_callback(struct notifier_block *nb,
 324                                        unsigned long val, void *data)
 325 {
 326         struct module *mod = data;
 327         struct trace_probe *tp;
 328         int ret;
 329
 330         if (val != MODULE_STATE_COMING)
 331                 return NOTIFY_DONE;
 332
 333         /* Update probes on coming module */
 334         mutex_lock(&probe_lock);
 335         list_for_each_entry(tp, &probe_list, list) {
 336                 if (trace_probe_within_module(tp, mod)) {
 337                         /* Don't need to check busy - this should have gone. */
 338                         __unregister_trace_probe(tp);
 339                         ret = __register_trace_probe(tp);
 340                         if (ret)
 341                                 pr_warning("Failed to re-register probe %s on"
 342                                            "%s: %d\n",
 343                                            tp->call.name, mod->name, ret);
 344                 }
 345         }
 346         mutex_unlock(&probe_lock);
 347
 348         return NOTIFY_DONE;
 349 }
 350
 351 static struct notifier_block trace_probe_module_nb = {
 352         .notifier_call = trace_probe_module_callback,
 353         .priority = 1   /* Invoked after kprobe module callback */
 354 };
 355
 356 static int create_trace_probe(int argc, char **argv)
 357 {
 358         /*
 359          * Argument syntax:
 360          *  - Add kprobe: p[:[GRP/]EVENT] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
 361          *  - Add kretprobe: r[:[GRP/]EVENT] [MOD:]KSYM[+0] [FETCHARGS]
 362          * Fetch args:
 363          *  $retval     : fetch return value
 364          *  $stack      : fetch stack address
 365          *  $stackN     : fetch Nth of stack (N:0-)
 366          *  @ADDR       : fetch memory at ADDR (ADDR should be in kernel)
 367          *  @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
 368          *  %REG        : fetch register REG
 369          * Dereferencing memory fetch:
 370          *  +|-offs(ARG) : fetch memory at ARG +|- offs address.
 371          * Alias name of args:
 372          *  NAME=FETCHARG : set NAME as alias of FETCHARG.
 373          * Type of args:
 374          *  FETCHARG:TYPE : use TYPE instead of unsigned long.
 375          */
 376         struct trace_probe *tp;
 377         int i, ret = 0;
 378         bool is_return = false, is_delete = false;
 379         char *symbol = NULL, *event = NULL, *group = NULL;
 380         char *arg;
 381         unsigned long offset = 0;
 382         void *addr = NULL;
 383         char buf[MAX_EVENT_NAME_LEN];
 384
 385         /* argc must be >= 1 */
 386         if (argv[0][0] == 'p')
 387                 is_return = false;
 388         else if (argv[0][0] == 'r')
 389                 is_return = true;
 390         else if (argv[0][0] == '-')
 391                 is_delete = true;
 392         else {
 393                 pr_info("Probe definition must be started with 'p', 'r' or"
 394                         " '-'.\n");
 395                 return -EINVAL;
 396         }
 397
 398         if (argv[0][1] == ':') {
 399                 event = &argv[0][2];
 400                 if (strchr(event, '/')) {
 401                         group = event;
 402                         event = strchr(group, '/') + 1;
 403                         event[-1] = '\0';
 404                         if (strlen(group) == 0) {
 405                                 pr_info("Group name is not specified\n");
 406                                 return -EINVAL;
 407                         }
 408                 }
 409                 if (strlen(event) == 0) {
 410                         pr_info("Event name is not specified\n");
 411                         return -EINVAL;
 412                 }
 413         }
 414         if (!group)
 415                 group = KPROBE_EVENT_SYSTEM;
 416
 417         if (is_delete) {
 418                 if (!event) {
 419                         pr_info("Delete command needs an event name.\n");
 420                         return -EINVAL;
 421                 }
 422                 mutex_lock(&probe_lock);
 423                 tp = find_trace_probe(event, group);
 424                 if (!tp) {
 425                         mutex_unlock(&probe_lock);
 426                         pr_info("Event %s/%s doesn't exist.\n", group, event);
 427                         return -ENOENT;
 428                 }
 429                 /* delete an event */
 430                 ret = unregister_trace_probe(tp);
 431                 if (ret == 0)
 432                         free_trace_probe(tp);
 433                 mutex_unlock(&probe_lock);
 434                 return ret;
 435         }
 436
 437         if (argc < 2) {
 438                 pr_info("Probe point is not specified.\n");
 439                 return -EINVAL;
 440         }
 441         if (isdigit(argv[1][0])) {
 442                 if (is_return) {
 443                         pr_info("Return probe point must be a symbol.\n");
 444                         return -EINVAL;
 445                 }
 446                 /* an address specified */
 447                 ret = kstrtoul(&argv[1][0], 0, (unsigned long *)&addr);
 448                 if (ret) {
 449                         pr_info("Failed to parse address.\n");
 450                         return ret;
 451                 }
 452         } else {
 453                 /* a symbol specified */
 454                 symbol = argv[1];
 455                 /* TODO: support .init module functions */
 456                 ret = traceprobe_split_symbol_offset(symbol, &offset);
 457                 if (ret) {
 458                         pr_info("Failed to parse symbol.\n");
 459                         return ret;
 460                 }
 461                 if (offset && is_return) {
 462                         pr_info("Return probe must be used without offset.\n");
 463                         return -EINVAL;
 464                 }
 465         }
 466         argc -= 2; argv += 2;
 467
 468         /* setup a probe */
 469         if (!event) {
 470                 /* Make a new event name */
 471                 if (symbol)
 472                         snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
 473                                  is_return ? 'r' : 'p', symbol, offset);
 474                 else
 475                         snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
 476                                  is_return ? 'r' : 'p', addr);
 477                 event = buf;
 478         }
 479         tp = alloc_trace_probe(group, event, addr, symbol, offset, argc,
 480                                is_return);
 481         if (IS_ERR(tp)) {
 482                 pr_info("Failed to allocate trace_probe.(%d)\n",
 483                         (int)PTR_ERR(tp));
 484                 return PTR_ERR(tp);
 485         }
 486
 487         /* parse arguments */
 488         ret = 0;
 489         for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
 490                 /* Increment count for freeing args in error case */
 491                 tp->nr_args++;
 492
 493                 /* Parse argument name */
 494                 arg = strchr(argv[i], '=');
 495                 if (arg) {
 496                         *arg++ = '\0';
 497                         tp->args[i].name = kstrdup(argv[i], GFP_KERNEL);
 498                 } else {
 499                         arg = argv[i];
 500                         /* If argument name is omitted, set "argN" */
 501                         snprintf(buf, MAX_EVENT_NAME_LEN, "arg%d", i + 1);
 502                         tp->args[i].name = kstrdup(buf, GFP_KERNEL);
 503                 }
 504
 505                 if (!tp->args[i].name) {
 506                         pr_info("Failed to allocate argument[%d] name.\n", i);
 507                         ret = -ENOMEM;
 508                         goto error;
 509                 }
 510
 511                 if (!is_good_name(tp->args[i].name)) {
 512                         pr_info("Invalid argument[%d] name: %s\n",
 513                                 i, tp->args[i].name);
 514                         ret = -EINVAL;
 515                         goto error;
 516                 }
 517
 518                 if (traceprobe_conflict_field_name(tp->args[i].name,
 519                                                         tp->args, i)) {
 520                         pr_info("Argument[%d] name '%s' conflicts with "
 521                                 "another field.\n", i, argv[i]);
 522                         ret = -EINVAL;
 523                         goto error;
 524                 }
 525
 526                 /* Parse fetch argument */
 527                 ret = traceprobe_parse_probe_arg(arg, &tp->size, &tp->args[i],
 528                                                 is_return, true);
 529                 if (ret) {
 530                         pr_info("Parse error at argument[%d]. (%d)\n", i, ret);
 531                         goto error;
 532                 }
 533         }
 534
 535         ret = register_trace_probe(tp);
 536         if (ret)
 537                 goto error;
 538         return 0;
 539
 540 error:
 541         free_trace_probe(tp);
 542         return ret;
 543 }
 544
 545 static int release_all_trace_probes(void)
 546 {
 547         struct trace_probe *tp;
 548         int ret = 0;
 549
 550         mutex_lock(&probe_lock);
 551         /* Ensure no probe is in use. */
 552         list_for_each_entry(tp, &probe_list, list)
 553                 if (trace_probe_is_enabled(tp)) {
 554                         ret = -EBUSY;
 555                         goto end;
 556                 }
 557         /* TODO: Use batch unregistration */
 558         while (!list_empty(&probe_list)) {
 559                 tp = list_entry(probe_list.next, struct trace_probe, list);
 560                 unregister_trace_probe(tp);
 561                 free_trace_probe(tp);
 562         }
 563
 564 end:
 565         mutex_unlock(&probe_lock);
 566
 567         return ret;
 568 }
 569
 570 /* Probes listing interfaces */
 571 static void *probes_seq_start(struct seq_file *m, loff_t *pos)
 572 {
 573         mutex_lock(&probe_lock);
 574         return seq_list_start(&probe_list, *pos);
 575 }
 576
 577 static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
 578 {
 579         return seq_list_next(v, &probe_list, pos);
 580 }
 581
 582 static void probes_seq_stop(struct seq_file *m, void *v)
 583 {
 584         mutex_unlock(&probe_lock);
 585 }
 586
 587 static int probes_seq_show(struct seq_file *m, void *v)
 588 {
 589         struct trace_probe *tp = v;
 590         int i;
 591
 592         seq_printf(m, "%c", trace_probe_is_return(tp) ? 'r' : 'p');
 593         seq_printf(m, ":%s/%s", tp->call.class->system, tp->call.name);
 594
 595         if (!tp->symbol)
 596                 seq_printf(m, " 0x%p", tp->rp.kp.addr);
 597         else if (tp->rp.kp.offset)
 598                 seq_printf(m, " %s+%u", trace_probe_symbol(tp),
 599                            tp->rp.kp.offset);
 600         else
 601                 seq_printf(m, " %s", trace_probe_symbol(tp));
 602
 603         for (i = 0; i < tp->nr_args; i++)
 604                 seq_printf(m, " %s=%s", tp->args[i].name, tp->args[i].comm);
 605         seq_printf(m, "\n");
 606
 607         return 0;
 608 }
 609
 610 static const struct seq_operations probes_seq_op = {
 611         .start  = probes_seq_start,
 612         .next   = probes_seq_next,
 613         .stop   = probes_seq_stop,
 614         .show   = probes_seq_show
 615 };
 616
 617 static int probes_open(struct inode *inode, struct file *file)
 618 {
 619         int ret;
 620
 621         if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
 622                 ret = release_all_trace_probes();
 623                 if (ret < 0)
 624                         return ret;
 625         }
 626
 627         return seq_open(file, &probes_seq_op);
 628 }
 629
 630 static ssize_t probes_write(struct file *file, const char __user *buffer,
 631                             size_t count, loff_t *ppos)
 632 {
 633         return traceprobe_probes_write(file, buffer, count, ppos,
 634                         create_trace_probe);
 635 }
 636
 637 static const struct file_operations kprobe_events_ops = {
 638         .owner          = THIS_MODULE,
 639         .open           = probes_open,
 640         .read           = seq_read,
 641         .llseek         = seq_lseek,
 642         .release        = seq_release,
 643         .write          = probes_write,
 644 };
 645
 646 /* Probes profiling interfaces */
 647 static int probes_profile_seq_show(struct seq_file *m, void *v)
 648 {
 649         struct trace_probe *tp = v;
 650
 651         seq_printf(m, "  %-44s %15lu %15lu\n", tp->call.name, tp->nhit,
 652                    tp->rp.kp.nmissed);
 653
 654         return 0;
 655 }
 656
 657 static const struct seq_operations profile_seq_op = {
 658         .start  = probes_seq_start,
 659         .next   = probes_seq_next,
 660         .stop   = probes_seq_stop,
 661         .show   = probes_profile_seq_show
 662 };
 663
 664 static int profile_open(struct inode *inode, struct file *file)
 665 {
 666         return seq_open(file, &profile_seq_op);
 667 }
 668
 669 static const struct file_operations kprobe_profile_ops = {
 670         .owner          = THIS_MODULE,
 671         .open           = profile_open,
 672         .read           = seq_read,
 673         .llseek         = seq_lseek,
 674         .release        = seq_release,
 675 };
 676
 677 /* Sum up total data length for dynamic arraies (strings) */
 678 static __kprobes int __get_data_size(struct trace_probe *tp,
 679                                      struct pt_regs *regs)
 680 {
 681         int i, ret = 0;
 682         u32 len;
 683
 684         for (i = 0; i < tp->nr_args; i++)
 685                 if (unlikely(tp->args[i].fetch_size.fn)) {
 686                         call_fetch(&tp->args[i].fetch_size, regs, &len);
 687                         ret += len;
 688                 }
 689
 690         return ret;
 691 }
 692
 693 /* Store the value of each argument */
 694 static __kprobes void store_trace_args(int ent_size, struct trace_probe *tp,
 695                                        struct pt_regs *regs,
 696                                        u8 *data, int maxlen)
 697 {
 698         int i;
 699         u32 end = tp->size;
 700         u32 *dl;        /* Data (relative) location */
 701
 702         for (i = 0; i < tp->nr_args; i++) {
 703                 if (unlikely(tp->args[i].fetch_size.fn)) {
 704                         /*
 705                          * First, we set the relative location and
 706                          * maximum data length to *dl
 707                          */
 708                         dl = (u32 *)(data + tp->args[i].offset);
 709                         *dl = make_data_rloc(maxlen, end - tp->args[i].offset);
 710                         /* Then try to fetch string or dynamic array data */
 711                         call_fetch(&tp->args[i].fetch, regs, dl);
 712                         /* Reduce maximum length */
 713                         end += get_rloc_len(*dl);
 714                         maxlen -= get_rloc_len(*dl);
 715                         /* Trick here, convert data_rloc to data_loc */
 716                         *dl = convert_rloc_to_loc(*dl,
 717                                  ent_size + tp->args[i].offset);
 718                 } else
 719                         /* Just fetching data normally */
 720                         call_fetch(&tp->args[i].fetch, regs,
 721                                    data + tp->args[i].offset);
 722         }
 723 }
 724
 725 /* Kprobe handler */
 726 static __kprobes void kprobe_trace_func(struct kprobe *kp, struct pt_regs *regs)
 727 {
 728         struct trace_probe *tp = container_of(kp, struct trace_probe, rp.kp);
 729         struct kprobe_trace_entry_head *entry;
 730         struct ring_buffer_event *event;
 731         struct ring_buffer *buffer;
 732         int size, dsize, pc;
 733         unsigned long irq_flags;
 734         struct ftrace_event_call *call = &tp->call;
 735
 736         tp->nhit++;
 737
 738         local_save_flags(irq_flags);
 739         pc = preempt_count();
 740
 741         dsize = __get_data_size(tp, regs);
 742         size = sizeof(*entry) + tp->size + dsize;
 743
 744         event = trace_current_buffer_lock_reserve(&buffer, call->event.type,
 745                                                   size, irq_flags, pc);
 746         if (!event)
 747                 return;
 748
 749         entry = ring_buffer_event_data(event);
 750         entry->ip = (unsigned long)kp->addr;
 751         store_trace_args(sizeof(*entry), tp, regs, (u8 *)&entry[1], dsize);
 752
 753         if (!filter_current_check_discard(buffer, call, entry, event))
 754                 trace_buffer_unlock_commit_regs(buffer, event,
 755                                                 irq_flags, pc, regs);
 756 }
 757
 758 /* Kretprobe handler */
 759 static __kprobes void kretprobe_trace_func(struct kretprobe_instance *ri,
 760                                           struct pt_regs *regs)
 761 {
 762         struct trace_probe *tp = container_of(ri->rp, struct trace_probe, rp);
 763         struct kretprobe_trace_entry_head *entry;
 764         struct ring_buffer_event *event;
 765         struct ring_buffer *buffer;
 766         int size, pc, dsize;
 767         unsigned long irq_flags;
 768         struct ftrace_event_call *call = &tp->call;
 769
 770         local_save_flags(irq_flags);
 771         pc = preempt_count();
 772
 773         dsize = __get_data_size(tp, regs);
 774         size = sizeof(*entry) + tp->size + dsize;
 775
 776         event = trace_current_buffer_lock_reserve(&buffer, call->event.type,
 777                                                   size, irq_flags, pc);
 778         if (!event)
 779                 return;
 780
 781         entry = ring_buffer_event_data(event);
 782         entry->func = (unsigned long)tp->rp.kp.addr;
 783         entry->ret_ip = (unsigned long)ri->ret_addr;
 784         store_trace_args(sizeof(*entry), tp, regs, (u8 *)&entry[1], dsize);
 785
 786         if (!filter_current_check_discard(buffer, call, entry, event))
 787                 trace_buffer_unlock_commit_regs(buffer, event,
 788                                                 irq_flags, pc, regs);
 789 }
 790
 791 /* Event entry printers */
 792 enum print_line_t
 793 print_kprobe_event(struct trace_iterator *iter, int flags,
 794                    struct trace_event *event)
 795 {
 796         struct kprobe_trace_entry_head *field;
 797         struct trace_seq *s = &iter->seq;
 798         struct trace_probe *tp;
 799         u8 *data;
 800         int i;
 801
 802         field = (struct kprobe_trace_entry_head *)iter->ent;
 803         tp = container_of(event, struct trace_probe, call.event);
 804
 805         if (!trace_seq_printf(s, "%s: (", tp->call.name))
 806                 goto partial;
 807
 808         if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
 809                 goto partial;
 810
 811         if (!trace_seq_puts(s, ")"))
 812                 goto partial;
 813
 814         data = (u8 *)&field[1];
 815         for (i = 0; i < tp->nr_args; i++)
 816                 if (!tp->args[i].type->print(s, tp->args[i].name,
 817                                              data + tp->args[i].offset, field))
 818                         goto partial;
 819
 820         if (!trace_seq_puts(s, "\n"))
 821                 goto partial;
 822
 823         return TRACE_TYPE_HANDLED;
 824 partial:
 825         return TRACE_TYPE_PARTIAL_LINE;
 826 }
 827
 828 enum print_line_t
 829 print_kretprobe_event(struct trace_iterator *iter, int flags,
 830                       struct trace_event *event)
 831 {
 832         struct kretprobe_trace_entry_head *field;
 833         struct trace_seq *s = &iter->seq;
 834         struct trace_probe *tp;
 835         u8 *data;
 836         int i;
 837
 838         field = (struct kretprobe_trace_entry_head *)iter->ent;
 839         tp = container_of(event, struct trace_probe, call.event);
 840
 841         if (!trace_seq_printf(s, "%s: (", tp->call.name))
 842                 goto partial;
 843
 844         if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
 845                 goto partial;
 846
 847         if (!trace_seq_puts(s, " <- "))
 848                 goto partial;
 849
 850         if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
 851                 goto partial;
 852
 853         if (!trace_seq_puts(s, ")"))
 854                 goto partial;
 855
 856         data = (u8 *)&field[1];
 857         for (i = 0; i < tp->nr_args; i++)
 858                 if (!tp->args[i].type->print(s, tp->args[i].name,
 859                                              data + tp->args[i].offset, field))
 860                         goto partial;
 861
 862         if (!trace_seq_puts(s, "\n"))
 863                 goto partial;
 864
 865         return TRACE_TYPE_HANDLED;
 866 partial:
 867         return TRACE_TYPE_PARTIAL_LINE;
 868 }
 869
 870
 871 static int kprobe_event_define_fields(struct ftrace_event_call *event_call)
 872 {
 873         int ret, i;
 874         struct kprobe_trace_entry_head field;
 875         struct trace_probe *tp = (struct trace_probe *)event_call->data;
 876
 877         DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
 878         /* Set argument names as fields */
 879         for (i = 0; i < tp->nr_args; i++) {
 880                 ret = trace_define_field(event_call, tp->args[i].type->fmttype,
 881                                          tp->args[i].name,
 882                                          sizeof(field) + tp->args[i].offset,
 883                                          tp->args[i].type->size,
 884                                          tp->args[i].type->is_signed,
 885                                          FILTER_OTHER);
 886                 if (ret)
 887                         return ret;
 888         }
 889         return 0;
 890 }
 891
 892 static int kretprobe_event_define_fields(struct ftrace_event_call *event_call)
 893 {
 894         int ret, i;
 895         struct kretprobe_trace_entry_head field;
 896         struct trace_probe *tp = (struct trace_probe *)event_call->data;
 897
 898         DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
 899         DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
 900         /* Set argument names as fields */
 901         for (i = 0; i < tp->nr_args; i++) {
 902                 ret = trace_define_field(event_call, tp->args[i].type->fmttype,
 903                                          tp->args[i].name,
 904                                          sizeof(field) + tp->args[i].offset,
 905                                          tp->args[i].type->size,
 906                                          tp->args[i].type->is_signed,
 907                                          FILTER_OTHER);
 908                 if (ret)
 909                         return ret;
 910         }
 911         return 0;
 912 }
 913
 914 static int __set_print_fmt(struct trace_probe *tp, char *buf, int len)
 915 {
 916         int i;
 917         int pos = 0;
 918
 919         const char *fmt, *arg;
 920
 921         if (!trace_probe_is_return(tp)) {
 922                 fmt = "(%lx)";
 923                 arg = "REC->" FIELD_STRING_IP;
 924         } else {
 925                 fmt = "(%lx <- %lx)";
 926                 arg = "REC->" FIELD_STRING_FUNC ", REC->" FIELD_STRING_RETIP;
 927         }
 928
 929         /* When len=0, we just calculate the needed length */
 930 #define LEN_OR_ZERO (len ? len - pos : 0)
 931
 932         pos += snprintf(buf + pos, LEN_OR_ZERO, "\"%s", fmt);
 933
 934         for (i = 0; i < tp->nr_args; i++) {
 935                 pos += snprintf(buf + pos, LEN_OR_ZERO, " %s=%s",
 936                                 tp->args[i].name, tp->args[i].type->fmt);
 937         }
 938
 939         pos += snprintf(buf + pos, LEN_OR_ZERO, "\", %s", arg);
 940
 941         for (i = 0; i < tp->nr_args; i++) {
 942                 if (strcmp(tp->args[i].type->name, "string") == 0)
 943                         pos += snprintf(buf + pos, LEN_OR_ZERO,
 944                                         ", __get_str(%s)",
 945                                         tp->args[i].name);
 946                 else
 947                         pos += snprintf(buf + pos, LEN_OR_ZERO, ", REC->%s",
 948                                         tp->args[i].name);
 949         }
 950
 951 #undef LEN_OR_ZERO
 952
 953         /* return the length of print_fmt */
 954         return pos;
 955 }
 956
 957 static int set_print_fmt(struct trace_probe *tp)
 958 {
 959         int len;
 960         char *print_fmt;
 961
 962         /* First: called with 0 length to calculate the needed length */
 963         len = __set_print_fmt(tp, NULL, 0);
 964         print_fmt = kmalloc(len + 1, GFP_KERNEL);
 965         if (!print_fmt)
 966                 return -ENOMEM;
 967
 968         /* Second: actually write the @print_fmt */
 969         __set_print_fmt(tp, print_fmt, len + 1);
 970         tp->call.print_fmt = print_fmt;
 971
 972         return 0;
 973 }
 974
 975 #ifdef CONFIG_PERF_EVENTS
 976
 977 /* Kprobe profile handler */
 978 static __kprobes void kprobe_perf_func(struct kprobe *kp,
 979                                          struct pt_regs *regs)
 980 {
 981         struct trace_probe *tp = container_of(kp, struct trace_probe, rp.kp);
 982         struct ftrace_event_call *call = &tp->call;
 983         struct kprobe_trace_entry_head *entry;
 984         struct hlist_head *head;
 985         int size, __size, dsize;
 986         int rctx;
 987
 988         dsize = __get_data_size(tp, regs);
 989         __size = sizeof(*entry) + tp->size + dsize;
 990         size = ALIGN(__size + sizeof(u32), sizeof(u64));
 991         size -= sizeof(u32);
 992         if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE,
 993                      "profile buffer not large enough"))
 994                 return;
 995
 996         entry = perf_trace_buf_prepare(size, call->event.type, regs, &rctx);
 997         if (!entry)
 998                 return;
 999
1000         entry->ip = (unsigned long)kp->addr;
1001         memset(&entry[1], 0, dsize);
1002         store_trace_args(sizeof(*entry), tp, regs, (u8 *)&entry[1], dsize);
1003
1004         head = this_cpu_ptr(call->perf_events);
1005         perf_trace_buf_submit(entry, size, rctx,
1006                                         entry->ip, 1, regs, head, NULL);
1007 }
1008
1009 /* Kretprobe profile handler */
1010 static __kprobes void kretprobe_perf_func(struct kretprobe_instance *ri,
1011                                             struct pt_regs *regs)
1012 {
1013         struct trace_probe *tp = container_of(ri->rp, struct trace_probe, rp);
1014         struct ftrace_event_call *call = &tp->call;
1015         struct kretprobe_trace_entry_head *entry;
1016         struct hlist_head *head;
1017         int size, __size, dsize;
1018         int rctx;
1019
1020         dsize = __get_data_size(tp, regs);
1021         __size = sizeof(*entry) + tp->size + dsize;
1022         size = ALIGN(__size + sizeof(u32), sizeof(u64));
1023         size -= sizeof(u32);
1024         if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE,
1025                      "profile buffer not large enough"))
1026                 return;
1027
1028         entry = perf_trace_buf_prepare(size, call->event.type, regs, &rctx);
1029         if (!entry)
1030                 return;
1031
1032         entry->func = (unsigned long)tp->rp.kp.addr;
1033         entry->ret_ip = (unsigned long)ri->ret_addr;
1034         store_trace_args(sizeof(*entry), tp, regs, (u8 *)&entry[1], dsize);
1035
1036         head = this_cpu_ptr(call->perf_events);
1037         perf_trace_buf_submit(entry, size, rctx,
1038                                         entry->ret_ip, 1, regs, head, NULL);
1039 }
1040 #endif  /* CONFIG_PERF_EVENTS */
1041
1042 static __kprobes
1043 int kprobe_register(struct ftrace_event_call *event,
1044                     enum trace_reg type, void *data)
1045 {
1046         struct trace_probe *tp = (struct trace_probe *)event->data;
1047
1048         switch (type) {
1049         case TRACE_REG_REGISTER:
1050                 return enable_trace_probe(tp, TP_FLAG_TRACE);
1051         case TRACE_REG_UNREGISTER:
1052                 disable_trace_probe(tp, TP_FLAG_TRACE);
1053                 return 0;
1054
1055 #ifdef CONFIG_PERF_EVENTS
1056         case TRACE_REG_PERF_REGISTER:
1057                 return enable_trace_probe(tp, TP_FLAG_PROFILE);
1058         case TRACE_REG_PERF_UNREGISTER:
1059                 disable_trace_probe(tp, TP_FLAG_PROFILE);
1060                 return 0;
1061         case TRACE_REG_PERF_OPEN:
1062         case TRACE_REG_PERF_CLOSE:
1063         case TRACE_REG_PERF_ADD:
1064         case TRACE_REG_PERF_DEL:
1065                 return 0;
1066 #endif
1067         }
1068         return 0;
1069 }
1070
1071 static __kprobes
1072 int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1073 {
1074         struct trace_probe *tp = container_of(kp, struct trace_probe, rp.kp);
1075
1076         if (tp->flags & TP_FLAG_TRACE)
1077                 kprobe_trace_func(kp, regs);
1078 #ifdef CONFIG_PERF_EVENTS
1079         if (tp->flags & TP_FLAG_PROFILE)
1080                 kprobe_perf_func(kp, regs);
1081 #endif
1082         return 0;       /* We don't tweek kernel, so just return 0 */
1083 }
1084
1085 static __kprobes
1086 int kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1087 {
1088         struct trace_probe *tp = container_of(ri->rp, struct trace_probe, rp);
1089
1090         if (tp->flags & TP_FLAG_TRACE)
1091                 kretprobe_trace_func(ri, regs);
1092 #ifdef CONFIG_PERF_EVENTS
1093         if (tp->flags & TP_FLAG_PROFILE)
1094                 kretprobe_perf_func(ri, regs);
1095 #endif
1096         return 0;       /* We don't tweek kernel, so just return 0 */
1097 }
1098
1099 static struct trace_event_functions kretprobe_funcs = {
1100         .trace          = print_kretprobe_event
1101 };
1102
1103 static struct trace_event_functions kprobe_funcs = {
1104         .trace          = print_kprobe_event
1105 };
1106
1107 static int register_probe_event(struct trace_probe *tp)
1108 {
1109         struct ftrace_event_call *call = &tp->call;
1110         int ret;
1111
1112         /* Initialize ftrace_event_call */
1113         INIT_LIST_HEAD(&call->class->fields);
1114         if (trace_probe_is_return(tp)) {
1115                 call->event.funcs = &kretprobe_funcs;
1116                 call->class->define_fields = kretprobe_event_define_fields;
1117         } else {
1118                 call->event.funcs = &kprobe_funcs;
1119                 call->class->define_fields = kprobe_event_define_fields;
1120         }
1121         if (set_print_fmt(tp) < 0)
1122                 return -ENOMEM;
1123         ret = register_ftrace_event(&call->event);
1124         if (!ret) {
1125                 kfree(call->print_fmt);
1126                 return -ENODEV;
1127         }
1128         call->flags = 0;
1129         call->class->reg = kprobe_register;
1130         call->data = tp;
1131         ret = trace_add_event_call(call);
1132         if (ret) {
1133                 pr_info("Failed to register kprobe event: %s\n", call->name);
1134                 kfree(call->print_fmt);
1135                 unregister_ftrace_event(&call->event);
1136         }
1137         return ret;
1138 }
1139
1140 static void unregister_probe_event(struct trace_probe *tp)
1141 {
1142         /* tp->event is unregistered in trace_remove_event_call() */
1143         trace_remove_event_call(&tp->call);
1144         kfree(tp->call.print_fmt);
1145 }
1146
1147 /* Make a debugfs interface for controlling probe points */
1148 static __init int init_kprobe_trace(void)
1149 {
1150         struct dentry *d_tracer;
1151         struct dentry *entry;
1152
1153         if (register_module_notifier(&trace_probe_module_nb))
1154                 return -EINVAL;
1155
1156         d_tracer = tracing_init_dentry();
1157         if (!d_tracer)
1158                 return 0;
1159
1160         entry = debugfs_create_file("kprobe_events", 0644, d_tracer,
1161                                     NULL, &kprobe_events_ops);
1162
1163         /* Event list interface */
1164         if (!entry)
1165                 pr_warning("Could not create debugfs "
1166                            "'kprobe_events' entry\n");
1167
1168         /* Profile interface */
1169         entry = debugfs_create_file("kprobe_profile", 0444, d_tracer,
1170                                     NULL, &kprobe_profile_ops);
1171
1172         if (!entry)
1173                 pr_warning("Could not create debugfs "
1174                            "'kprobe_profile' entry\n");
1175         return 0;
1176 }
1177 fs_initcall(init_kprobe_trace);
1178
1179
1180 #ifdef CONFIG_FTRACE_STARTUP_TEST
1181
1182 /*
1183  * The "__used" keeps gcc from removing the function symbol
1184  * from the kallsyms table.
1185  */
1186 static __used int kprobe_trace_selftest_target(int a1, int a2, int a3,
1187                                                int a4, int a5, int a6)
1188 {
1189         return a1 + a2 + a3 + a4 + a5 + a6;
1190 }
1191
1192 static __init int kprobe_trace_self_tests_init(void)
1193 {
1194         int ret, warn = 0;
1195         int (*target)(int, int, int, int, int, int);
1196         struct trace_probe *tp;
1197
1198         target = kprobe_trace_selftest_target;
1199
1200         pr_info("Testing kprobe tracing: ");
1201
1202         ret = traceprobe_command("p:testprobe kprobe_trace_selftest_target "
1203                                   "$stack $stack0 +0($stack)",
1204                                   create_trace_probe);
1205         if (WARN_ON_ONCE(ret)) {
1206                 pr_warning("error on probing function entry.\n");
1207                 warn++;
1208         } else {
1209                 /* Enable trace point */
1210                 tp = find_trace_probe("testprobe", KPROBE_EVENT_SYSTEM);
1211                 if (WARN_ON_ONCE(tp == NULL)) {
1212                         pr_warning("error on getting new probe.\n");
1213                         warn++;
1214                 } else
1215                         enable_trace_probe(tp, TP_FLAG_TRACE);
1216         }
1217
1218         ret = traceprobe_command("r:testprobe2 kprobe_trace_selftest_target "
1219                                   "$retval", create_trace_probe);
1220         if (WARN_ON_ONCE(ret)) {
1221                 pr_warning("error on probing function return.\n");
1222                 warn++;
1223         } else {
1224                 /* Enable trace point */
1225                 tp = find_trace_probe("testprobe2", KPROBE_EVENT_SYSTEM);
1226                 if (WARN_ON_ONCE(tp == NULL)) {
1227                         pr_warning("error on getting new probe.\n");
1228                         warn++;
1229                 } else
1230                         enable_trace_probe(tp, TP_FLAG_TRACE);
1231         }
1232
1233         if (warn)
1234                 goto end;
1235
1236         ret = target(1, 2, 3, 4, 5, 6);
1237
1238         /* Disable trace points before removing it */
1239         tp = find_trace_probe("testprobe", KPROBE_EVENT_SYSTEM);
1240         if (WARN_ON_ONCE(tp == NULL)) {
1241                 pr_warning("error on getting test probe.\n");
1242                 warn++;
1243         } else
1244                 disable_trace_probe(tp, TP_FLAG_TRACE);
1245
1246         tp = find_trace_probe("testprobe2", KPROBE_EVENT_SYSTEM);
1247         if (WARN_ON_ONCE(tp == NULL)) {
1248                 pr_warning("error on getting 2nd test probe.\n");
1249                 warn++;
1250         } else
1251                 disable_trace_probe(tp, TP_FLAG_TRACE);
1252
1253         ret = traceprobe_command("-:testprobe", create_trace_probe);
1254         if (WARN_ON_ONCE(ret)) {
1255                 pr_warning("error on deleting a probe.\n");
1256                 warn++;
1257         }
1258
1259         ret = traceprobe_command("-:testprobe2", create_trace_probe);
1260         if (WARN_ON_ONCE(ret)) {
1261                 pr_warning("error on deleting a probe.\n");
1262                 warn++;
1263         }
1264
1265 end:
1266         release_all_trace_probes();
1267         if (warn)
1268                 pr_cont("NG: Some tests are failed. Please check them.\n");
1269         else
1270                 pr_cont("OK\n");
1271         return 0;
1272 }
1273
1274 late_initcall(kprobe_trace_self_tests_init);
1275
1276 #endif