Pileus Git - ~andy/linux/blob - 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 struct trace_kprobe {
  31         struct list_head        list;
  32         struct kretprobe        rp;     /* Use rp.kp for kprobe use */
  33         unsigned long           nhit;
  34         const char              *symbol;        /* symbol name */
  35         struct trace_probe      tp;
  36 };
  37
  38 struct event_file_link {
  39         struct ftrace_event_file        *file;
  40         struct list_head                list;
  41 };
  42
  43 #define SIZEOF_TRACE_KPROBE(n)                          \
  44         (offsetof(struct trace_kprobe, tp.args) +       \
  45         (sizeof(struct probe_arg) * (n)))
  46
  47
  48 static __kprobes bool trace_kprobe_is_return(struct trace_kprobe *tk)
  49 {
  50         return tk->rp.handler != NULL;
  51 }
  52
  53 static __kprobes const char *trace_kprobe_symbol(struct trace_kprobe *tk)
  54 {
  55         return tk->symbol ? tk->symbol : "unknown";
  56 }
  57
  58 static __kprobes unsigned long trace_kprobe_offset(struct trace_kprobe *tk)
  59 {
  60         return tk->rp.kp.offset;
  61 }
  62
  63 static __kprobes bool trace_kprobe_has_gone(struct trace_kprobe *tk)
  64 {
  65         return !!(kprobe_gone(&tk->rp.kp));
  66 }
  67
  68 static __kprobes bool trace_kprobe_within_module(struct trace_kprobe *tk,
  69                                                  struct module *mod)
  70 {
  71         int len = strlen(mod->name);
  72         const char *name = trace_kprobe_symbol(tk);
  73         return strncmp(mod->name, name, len) == 0 && name[len] == ':';
  74 }
  75
  76 static __kprobes bool trace_kprobe_is_on_module(struct trace_kprobe *tk)
  77 {
  78         return !!strchr(trace_kprobe_symbol(tk), ':');
  79 }
  80
  81 static int register_kprobe_event(struct trace_kprobe *tk);
  82 static int unregister_kprobe_event(struct trace_kprobe *tk);
  83
  84 static DEFINE_MUTEX(probe_lock);
  85 static LIST_HEAD(probe_list);
  86
  87 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
  88 static int kretprobe_dispatcher(struct kretprobe_instance *ri,
  89                                 struct pt_regs *regs);
  90
  91 /*
  92  * Allocate new trace_probe and initialize it (including kprobes).
  93  */
  94 static struct trace_kprobe *alloc_trace_kprobe(const char *group,
  95                                              const char *event,
  96                                              void *addr,
  97                                              const char *symbol,
  98                                              unsigned long offs,
  99                                              int nargs, bool is_return)
 100 {
 101         struct trace_kprobe *tk;
 102         int ret = -ENOMEM;
 103
 104         tk = kzalloc(SIZEOF_TRACE_KPROBE(nargs), GFP_KERNEL);
 105         if (!tk)
 106                 return ERR_PTR(ret);
 107
 108         if (symbol) {
 109                 tk->symbol = kstrdup(symbol, GFP_KERNEL);
 110                 if (!tk->symbol)
 111                         goto error;
 112                 tk->rp.kp.symbol_name = tk->symbol;
 113                 tk->rp.kp.offset = offs;
 114         } else
 115                 tk->rp.kp.addr = addr;
 116
 117         if (is_return)
 118                 tk->rp.handler = kretprobe_dispatcher;
 119         else
 120                 tk->rp.kp.pre_handler = kprobe_dispatcher;
 121
 122         if (!event || !is_good_name(event)) {
 123                 ret = -EINVAL;
 124                 goto error;
 125         }
 126
 127         tk->tp.call.class = &tk->tp.class;
 128         tk->tp.call.name = kstrdup(event, GFP_KERNEL);
 129         if (!tk->tp.call.name)
 130                 goto error;
 131
 132         if (!group || !is_good_name(group)) {
 133                 ret = -EINVAL;
 134                 goto error;
 135         }
 136
 137         tk->tp.class.system = kstrdup(group, GFP_KERNEL);
 138         if (!tk->tp.class.system)
 139                 goto error;
 140
 141         INIT_LIST_HEAD(&tk->list);
 142         INIT_LIST_HEAD(&tk->tp.files);
 143         return tk;
 144 error:
 145         kfree(tk->tp.call.name);
 146         kfree(tk->symbol);
 147         kfree(tk);
 148         return ERR_PTR(ret);
 149 }
 150
 151 static void free_trace_kprobe(struct trace_kprobe *tk)
 152 {
 153         int i;
 154
 155         for (i = 0; i < tk->tp.nr_args; i++)
 156                 traceprobe_free_probe_arg(&tk->tp.args[i]);
 157
 158         kfree(tk->tp.call.class->system);
 159         kfree(tk->tp.call.name);
 160         kfree(tk->symbol);
 161         kfree(tk);
 162 }
 163
 164 static struct trace_kprobe *find_trace_kprobe(const char *event,
 165                                               const char *group)
 166 {
 167         struct trace_kprobe *tk;
 168
 169         list_for_each_entry(tk, &probe_list, list)
 170                 if (strcmp(tk->tp.call.name, event) == 0 &&
 171                     strcmp(tk->tp.call.class->system, group) == 0)
 172                         return tk;
 173         return NULL;
 174 }
 175
 176 /*
 177  * Enable trace_probe
 178  * if the file is NULL, enable "perf" handler, or enable "trace" handler.
 179  */
 180 static int
 181 enable_trace_kprobe(struct trace_kprobe *tk, struct ftrace_event_file *file)
 182 {
 183         int ret = 0;
 184
 185         if (file) {
 186                 struct event_file_link *link;
 187
 188                 link = kmalloc(sizeof(*link), GFP_KERNEL);
 189                 if (!link) {
 190                         ret = -ENOMEM;
 191                         goto out;
 192                 }
 193
 194                 link->file = file;
 195                 list_add_tail_rcu(&link->list, &tk->tp.files);
 196
 197                 tk->tp.flags |= TP_FLAG_TRACE;
 198         } else
 199                 tk->tp.flags |= TP_FLAG_PROFILE;
 200
 201         if (trace_probe_is_registered(&tk->tp) && !trace_kprobe_has_gone(tk)) {
 202                 if (trace_kprobe_is_return(tk))
 203                         ret = enable_kretprobe(&tk->rp);
 204                 else
 205                         ret = enable_kprobe(&tk->rp.kp);
 206         }
 207  out:
 208         return ret;
 209 }
 210
 211 static struct event_file_link *
 212 find_event_file_link(struct trace_probe *tp, struct ftrace_event_file *file)
 213 {
 214         struct event_file_link *link;
 215
 216         list_for_each_entry(link, &tp->files, list)
 217                 if (link->file == file)
 218                         return link;
 219
 220         return NULL;
 221 }
 222
 223 /*
 224  * Disable trace_probe
 225  * if the file is NULL, disable "perf" handler, or disable "trace" handler.
 226  */
 227 static int
 228 disable_trace_kprobe(struct trace_kprobe *tk, struct ftrace_event_file *file)
 229 {
 230         struct event_file_link *link = NULL;
 231         int wait = 0;
 232         int ret = 0;
 233
 234         if (file) {
 235                 link = find_event_file_link(&tk->tp, file);
 236                 if (!link) {
 237                         ret = -EINVAL;
 238                         goto out;
 239                 }
 240
 241                 list_del_rcu(&link->list);
 242                 wait = 1;
 243                 if (!list_empty(&tk->tp.files))
 244                         goto out;
 245
 246                 tk->tp.flags &= ~TP_FLAG_TRACE;
 247         } else
 248                 tk->tp.flags &= ~TP_FLAG_PROFILE;
 249
 250         if (!trace_probe_is_enabled(&tk->tp) && trace_probe_is_registered(&tk->tp)) {
 251                 if (trace_kprobe_is_return(tk))
 252                         disable_kretprobe(&tk->rp);
 253                 else
 254                         disable_kprobe(&tk->rp.kp);
 255                 wait = 1;
 256         }
 257  out:
 258         if (wait) {
 259                 /*
 260                  * Synchronize with kprobe_trace_func/kretprobe_trace_func
 261                  * to ensure disabled (all running handlers are finished).
 262                  * This is not only for kfree(), but also the caller,
 263                  * trace_remove_event_call() supposes it for releasing
 264                  * event_call related objects, which will be accessed in
 265                  * the kprobe_trace_func/kretprobe_trace_func.
 266                  */
 267                 synchronize_sched();
 268                 kfree(link);    /* Ignored if link == NULL */
 269         }
 270
 271         return ret;
 272 }
 273
 274 /* Internal register function - just handle k*probes and flags */
 275 static int __register_trace_kprobe(struct trace_kprobe *tk)
 276 {
 277         int i, ret;
 278
 279         if (trace_probe_is_registered(&tk->tp))
 280                 return -EINVAL;
 281
 282         for (i = 0; i < tk->tp.nr_args; i++)
 283                 traceprobe_update_arg(&tk->tp.args[i]);
 284
 285         /* Set/clear disabled flag according to tp->flag */
 286         if (trace_probe_is_enabled(&tk->tp))
 287                 tk->rp.kp.flags &= ~KPROBE_FLAG_DISABLED;
 288         else
 289                 tk->rp.kp.flags |= KPROBE_FLAG_DISABLED;
 290
 291         if (trace_kprobe_is_return(tk))
 292                 ret = register_kretprobe(&tk->rp);
 293         else
 294                 ret = register_kprobe(&tk->rp.kp);
 295
 296         if (ret == 0)
 297                 tk->tp.flags |= TP_FLAG_REGISTERED;
 298         else {
 299                 pr_warning("Could not insert probe at %s+%lu: %d\n",
 300                            trace_kprobe_symbol(tk), trace_kprobe_offset(tk), ret);
 301                 if (ret == -ENOENT && trace_kprobe_is_on_module(tk)) {
 302                         pr_warning("This probe might be able to register after"
 303                                    "target module is loaded. Continue.\n");
 304                         ret = 0;
 305                 } else if (ret == -EILSEQ) {
 306                         pr_warning("Probing address(0x%p) is not an "
 307                                    "instruction boundary.\n",
 308                                    tk->rp.kp.addr);
 309                         ret = -EINVAL;
 310                 }
 311         }
 312
 313         return ret;
 314 }
 315
 316 /* Internal unregister function - just handle k*probes and flags */
 317 static void __unregister_trace_kprobe(struct trace_kprobe *tk)
 318 {
 319         if (trace_probe_is_registered(&tk->tp)) {
 320                 if (trace_kprobe_is_return(tk))
 321                         unregister_kretprobe(&tk->rp);
 322                 else
 323                         unregister_kprobe(&tk->rp.kp);
 324                 tk->tp.flags &= ~TP_FLAG_REGISTERED;
 325                 /* Cleanup kprobe for reuse */
 326                 if (tk->rp.kp.symbol_name)
 327                         tk->rp.kp.addr = NULL;
 328         }
 329 }
 330
 331 /* Unregister a trace_probe and probe_event: call with locking probe_lock */
 332 static int unregister_trace_kprobe(struct trace_kprobe *tk)
 333 {
 334         /* Enabled event can not be unregistered */
 335         if (trace_probe_is_enabled(&tk->tp))
 336                 return -EBUSY;
 337
 338         /* Will fail if probe is being used by ftrace or perf */
 339         if (unregister_kprobe_event(tk))
 340                 return -EBUSY;
 341
 342         __unregister_trace_kprobe(tk);
 343         list_del(&tk->list);
 344
 345         return 0;
 346 }
 347
 348 /* Register a trace_probe and probe_event */
 349 static int register_trace_kprobe(struct trace_kprobe *tk)
 350 {
 351         struct trace_kprobe *old_tk;
 352         int ret;
 353
 354         mutex_lock(&probe_lock);
 355
 356         /* Delete old (same name) event if exist */
 357         old_tk = find_trace_kprobe(tk->tp.call.name, tk->tp.call.class->system);
 358         if (old_tk) {
 359                 ret = unregister_trace_kprobe(old_tk);
 360                 if (ret < 0)
 361                         goto end;
 362                 free_trace_kprobe(old_tk);
 363         }
 364
 365         /* Register new event */
 366         ret = register_kprobe_event(tk);
 367         if (ret) {
 368                 pr_warning("Failed to register probe event(%d)\n", ret);
 369                 goto end;
 370         }
 371
 372         /* Register k*probe */
 373         ret = __register_trace_kprobe(tk);
 374         if (ret < 0)
 375                 unregister_kprobe_event(tk);
 376         else
 377                 list_add_tail(&tk->list, &probe_list);
 378
 379 end:
 380         mutex_unlock(&probe_lock);
 381         return ret;
 382 }
 383
 384 /* Module notifier call back, checking event on the module */
 385 static int trace_kprobe_module_callback(struct notifier_block *nb,
 386                                        unsigned long val, void *data)
 387 {
 388         struct module *mod = data;
 389         struct trace_kprobe *tk;
 390         int ret;
 391
 392         if (val != MODULE_STATE_COMING)
 393                 return NOTIFY_DONE;
 394
 395         /* Update probes on coming module */
 396         mutex_lock(&probe_lock);
 397         list_for_each_entry(tk, &probe_list, list) {
 398                 if (trace_kprobe_within_module(tk, mod)) {
 399                         /* Don't need to check busy - this should have gone. */
 400                         __unregister_trace_kprobe(tk);
 401                         ret = __register_trace_kprobe(tk);
 402                         if (ret)
 403                                 pr_warning("Failed to re-register probe %s on"
 404                                            "%s: %d\n",
 405                                            tk->tp.call.name, mod->name, ret);
 406                 }
 407         }
 408         mutex_unlock(&probe_lock);
 409
 410         return NOTIFY_DONE;
 411 }
 412
 413 static struct notifier_block trace_kprobe_module_nb = {
 414         .notifier_call = trace_kprobe_module_callback,
 415         .priority = 1   /* Invoked after kprobe module callback */
 416 };
 417
 418 static int create_trace_kprobe(int argc, char **argv)
 419 {
 420         /*
 421          * Argument syntax:
 422          *  - Add kprobe: p[:[GRP/]EVENT] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
 423          *  - Add kretprobe: r[:[GRP/]EVENT] [MOD:]KSYM[+0] [FETCHARGS]
 424          * Fetch args:
 425          *  $retval     : fetch return value
 426          *  $stack      : fetch stack address
 427          *  $stackN     : fetch Nth of stack (N:0-)
 428          *  @ADDR       : fetch memory at ADDR (ADDR should be in kernel)
 429          *  @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
 430          *  %REG        : fetch register REG
 431          * Dereferencing memory fetch:
 432          *  +|-offs(ARG) : fetch memory at ARG +|- offs address.
 433          * Alias name of args:
 434          *  NAME=FETCHARG : set NAME as alias of FETCHARG.
 435          * Type of args:
 436          *  FETCHARG:TYPE : use TYPE instead of unsigned long.
 437          */
 438         struct trace_kprobe *tk;
 439         int i, ret = 0;
 440         bool is_return = false, is_delete = false;
 441         char *symbol = NULL, *event = NULL, *group = NULL;
 442         char *arg;
 443         unsigned long offset = 0;
 444         void *addr = NULL;
 445         char buf[MAX_EVENT_NAME_LEN];
 446
 447         /* argc must be >= 1 */
 448         if (argv[0][0] == 'p')
 449                 is_return = false;
 450         else if (argv[0][0] == 'r')
 451                 is_return = true;
 452         else if (argv[0][0] == '-')
 453                 is_delete = true;
 454         else {
 455                 pr_info("Probe definition must be started with 'p', 'r' or"
 456                         " '-'.\n");
 457                 return -EINVAL;
 458         }
 459
 460         if (argv[0][1] == ':') {
 461                 event = &argv[0][2];
 462                 if (strchr(event, '/')) {
 463                         group = event;
 464                         event = strchr(group, '/') + 1;
 465                         event[-1] = '\0';
 466                         if (strlen(group) == 0) {
 467                                 pr_info("Group name is not specified\n");
 468                                 return -EINVAL;
 469                         }
 470                 }
 471                 if (strlen(event) == 0) {
 472                         pr_info("Event name is not specified\n");
 473                         return -EINVAL;
 474                 }
 475         }
 476         if (!group)
 477                 group = KPROBE_EVENT_SYSTEM;
 478
 479         if (is_delete) {
 480                 if (!event) {
 481                         pr_info("Delete command needs an event name.\n");
 482                         return -EINVAL;
 483                 }
 484                 mutex_lock(&probe_lock);
 485                 tk = find_trace_kprobe(event, group);
 486                 if (!tk) {
 487                         mutex_unlock(&probe_lock);
 488                         pr_info("Event %s/%s doesn't exist.\n", group, event);
 489                         return -ENOENT;
 490                 }
 491                 /* delete an event */
 492                 ret = unregister_trace_kprobe(tk);
 493                 if (ret == 0)
 494                         free_trace_kprobe(tk);
 495                 mutex_unlock(&probe_lock);
 496                 return ret;
 497         }
 498
 499         if (argc < 2) {
 500                 pr_info("Probe point is not specified.\n");
 501                 return -EINVAL;
 502         }
 503         if (isdigit(argv[1][0])) {
 504                 if (is_return) {
 505                         pr_info("Return probe point must be a symbol.\n");
 506                         return -EINVAL;
 507                 }
 508                 /* an address specified */
 509                 ret = kstrtoul(&argv[1][0], 0, (unsigned long *)&addr);
 510                 if (ret) {
 511                         pr_info("Failed to parse address.\n");
 512                         return ret;
 513                 }
 514         } else {
 515                 /* a symbol specified */
 516                 symbol = argv[1];
 517                 /* TODO: support .init module functions */
 518                 ret = traceprobe_split_symbol_offset(symbol, &offset);
 519                 if (ret) {
 520                         pr_info("Failed to parse symbol.\n");
 521                         return ret;
 522                 }
 523                 if (offset && is_return) {
 524                         pr_info("Return probe must be used without offset.\n");
 525                         return -EINVAL;
 526                 }
 527         }
 528         argc -= 2; argv += 2;
 529
 530         /* setup a probe */
 531         if (!event) {
 532                 /* Make a new event name */
 533                 if (symbol)
 534                         snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
 535                                  is_return ? 'r' : 'p', symbol, offset);
 536                 else
 537                         snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
 538                                  is_return ? 'r' : 'p', addr);
 539                 event = buf;
 540         }
 541         tk = alloc_trace_kprobe(group, event, addr, symbol, offset, argc,
 542                                is_return);
 543         if (IS_ERR(tk)) {
 544                 pr_info("Failed to allocate trace_probe.(%d)\n",
 545                         (int)PTR_ERR(tk));
 546                 return PTR_ERR(tk);
 547         }
 548
 549         /* parse arguments */
 550         ret = 0;
 551         for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
 552                 struct probe_arg *parg = &tk->tp.args[i];
 553
 554                 /* Increment count for freeing args in error case */
 555                 tk->tp.nr_args++;
 556
 557                 /* Parse argument name */
 558                 arg = strchr(argv[i], '=');
 559                 if (arg) {
 560                         *arg++ = '\0';
 561                         parg->name = kstrdup(argv[i], GFP_KERNEL);
 562                 } else {
 563                         arg = argv[i];
 564                         /* If argument name is omitted, set "argN" */
 565                         snprintf(buf, MAX_EVENT_NAME_LEN, "arg%d", i + 1);
 566                         parg->name = kstrdup(buf, GFP_KERNEL);
 567                 }
 568
 569                 if (!parg->name) {
 570                         pr_info("Failed to allocate argument[%d] name.\n", i);
 571                         ret = -ENOMEM;
 572                         goto error;
 573                 }
 574
 575                 if (!is_good_name(parg->name)) {
 576                         pr_info("Invalid argument[%d] name: %s\n",
 577                                 i, parg->name);
 578                         ret = -EINVAL;
 579                         goto error;
 580                 }
 581
 582                 if (traceprobe_conflict_field_name(parg->name,
 583                                                         tk->tp.args, i)) {
 584                         pr_info("Argument[%d] name '%s' conflicts with "
 585                                 "another field.\n", i, argv[i]);
 586                         ret = -EINVAL;
 587                         goto error;
 588                 }
 589
 590                 /* Parse fetch argument */
 591                 ret = traceprobe_parse_probe_arg(arg, &tk->tp.size, parg,
 592                                                 is_return, true);
 593                 if (ret) {
 594                         pr_info("Parse error at argument[%d]. (%d)\n", i, ret);
 595                         goto error;
 596                 }
 597         }
 598
 599         ret = register_trace_kprobe(tk);
 600         if (ret)
 601                 goto error;
 602         return 0;
 603
 604 error:
 605         free_trace_kprobe(tk);
 606         return ret;
 607 }
 608
 609 static int release_all_trace_kprobes(void)
 610 {
 611         struct trace_kprobe *tk;
 612         int ret = 0;
 613
 614         mutex_lock(&probe_lock);
 615         /* Ensure no probe is in use. */
 616         list_for_each_entry(tk, &probe_list, list)
 617                 if (trace_probe_is_enabled(&tk->tp)) {
 618                         ret = -EBUSY;
 619                         goto end;
 620                 }
 621         /* TODO: Use batch unregistration */
 622         while (!list_empty(&probe_list)) {
 623                 tk = list_entry(probe_list.next, struct trace_kprobe, list);
 624                 ret = unregister_trace_kprobe(tk);
 625                 if (ret)
 626                         goto end;
 627                 free_trace_kprobe(tk);
 628         }
 629
 630 end:
 631         mutex_unlock(&probe_lock);
 632
 633         return ret;
 634 }
 635
 636 /* Probes listing interfaces */
 637 static void *probes_seq_start(struct seq_file *m, loff_t *pos)
 638 {
 639         mutex_lock(&probe_lock);
 640         return seq_list_start(&probe_list, *pos);
 641 }
 642
 643 static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
 644 {
 645         return seq_list_next(v, &probe_list, pos);
 646 }
 647
 648 static void probes_seq_stop(struct seq_file *m, void *v)
 649 {
 650         mutex_unlock(&probe_lock);
 651 }
 652
 653 static int probes_seq_show(struct seq_file *m, void *v)
 654 {
 655         struct trace_kprobe *tk = v;
 656         int i;
 657
 658         seq_printf(m, "%c", trace_kprobe_is_return(tk) ? 'r' : 'p');
 659         seq_printf(m, ":%s/%s", tk->tp.call.class->system, tk->tp.call.name);
 660
 661         if (!tk->symbol)
 662                 seq_printf(m, " 0x%p", tk->rp.kp.addr);
 663         else if (tk->rp.kp.offset)
 664                 seq_printf(m, " %s+%u", trace_kprobe_symbol(tk),
 665                            tk->rp.kp.offset);
 666         else
 667                 seq_printf(m, " %s", trace_kprobe_symbol(tk));
 668
 669         for (i = 0; i < tk->tp.nr_args; i++)
 670                 seq_printf(m, " %s=%s", tk->tp.args[i].name, tk->tp.args[i].comm);
 671         seq_printf(m, "\n");
 672
 673         return 0;
 674 }
 675
 676 static const struct seq_operations probes_seq_op = {
 677         .start  = probes_seq_start,
 678         .next   = probes_seq_next,
 679         .stop   = probes_seq_stop,
 680         .show   = probes_seq_show
 681 };
 682
 683 static int probes_open(struct inode *inode, struct file *file)
 684 {
 685         int ret;
 686
 687         if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
 688                 ret = release_all_trace_kprobes();
 689                 if (ret < 0)
 690                         return ret;
 691         }
 692
 693         return seq_open(file, &probes_seq_op);
 694 }
 695
 696 static ssize_t probes_write(struct file *file, const char __user *buffer,
 697                             size_t count, loff_t *ppos)
 698 {
 699         return traceprobe_probes_write(file, buffer, count, ppos,
 700                         create_trace_kprobe);
 701 }
 702
 703 static const struct file_operations kprobe_events_ops = {
 704         .owner          = THIS_MODULE,
 705         .open           = probes_open,
 706         .read           = seq_read,
 707         .llseek         = seq_lseek,
 708         .release        = seq_release,
 709         .write          = probes_write,
 710 };
 711
 712 /* Probes profiling interfaces */
 713 static int probes_profile_seq_show(struct seq_file *m, void *v)
 714 {
 715         struct trace_kprobe *tk = v;
 716
 717         seq_printf(m, "  %-44s %15lu %15lu\n", tk->tp.call.name, tk->nhit,
 718                    tk->rp.kp.nmissed);
 719
 720         return 0;
 721 }
 722
 723 static const struct seq_operations profile_seq_op = {
 724         .start  = probes_seq_start,
 725         .next   = probes_seq_next,
 726         .stop   = probes_seq_stop,
 727         .show   = probes_profile_seq_show
 728 };
 729
 730 static int profile_open(struct inode *inode, struct file *file)
 731 {
 732         return seq_open(file, &profile_seq_op);
 733 }
 734
 735 static const struct file_operations kprobe_profile_ops = {
 736         .owner          = THIS_MODULE,
 737         .open           = profile_open,
 738         .read           = seq_read,
 739         .llseek         = seq_lseek,
 740         .release        = seq_release,
 741 };
 742
 743 /* Kprobe handler */
 744 static __kprobes void
 745 __kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs,
 746                     struct ftrace_event_file *ftrace_file)
 747 {
 748         struct kprobe_trace_entry_head *entry;
 749         struct ring_buffer_event *event;
 750         struct ring_buffer *buffer;
 751         int size, dsize, pc;
 752         unsigned long irq_flags;
 753         struct ftrace_event_call *call = &tk->tp.call;
 754
 755         WARN_ON(call != ftrace_file->event_call);
 756
 757         if (test_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT, &ftrace_file->flags))
 758                 return;
 759
 760         local_save_flags(irq_flags);
 761         pc = preempt_count();
 762
 763         dsize = __get_data_size(&tk->tp, regs);
 764         size = sizeof(*entry) + tk->tp.size + dsize;
 765
 766         event = trace_event_buffer_lock_reserve(&buffer, ftrace_file,
 767                                                 call->event.type,
 768                                                 size, irq_flags, pc);
 769         if (!event)
 770                 return;
 771
 772         entry = ring_buffer_event_data(event);
 773         entry->ip = (unsigned long)tk->rp.kp.addr;
 774         store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
 775
 776         if (!filter_check_discard(ftrace_file, entry, buffer, event))
 777                 trace_buffer_unlock_commit_regs(buffer, event,
 778                                                 irq_flags, pc, regs);
 779 }
 780
 781 static __kprobes void
 782 kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs)
 783 {
 784         struct event_file_link *link;
 785
 786         list_for_each_entry_rcu(link, &tk->tp.files, list)
 787                 __kprobe_trace_func(tk, regs, link->file);
 788 }
 789
 790 /* Kretprobe handler */
 791 static __kprobes void
 792 __kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
 793                        struct pt_regs *regs,
 794                        struct ftrace_event_file *ftrace_file)
 795 {
 796         struct kretprobe_trace_entry_head *entry;
 797         struct ring_buffer_event *event;
 798         struct ring_buffer *buffer;
 799         int size, pc, dsize;
 800         unsigned long irq_flags;
 801         struct ftrace_event_call *call = &tk->tp.call;
 802
 803         WARN_ON(call != ftrace_file->event_call);
 804
 805         if (test_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT, &ftrace_file->flags))
 806                 return;
 807
 808         local_save_flags(irq_flags);
 809         pc = preempt_count();
 810
 811         dsize = __get_data_size(&tk->tp, regs);
 812         size = sizeof(*entry) + tk->tp.size + dsize;
 813
 814         event = trace_event_buffer_lock_reserve(&buffer, ftrace_file,
 815                                                 call->event.type,
 816                                                 size, irq_flags, pc);
 817         if (!event)
 818                 return;
 819
 820         entry = ring_buffer_event_data(event);
 821         entry->func = (unsigned long)tk->rp.kp.addr;
 822         entry->ret_ip = (unsigned long)ri->ret_addr;
 823         store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
 824
 825         if (!filter_check_discard(ftrace_file, entry, buffer, event))
 826                 trace_buffer_unlock_commit_regs(buffer, event,
 827                                                 irq_flags, pc, regs);
 828 }
 829
 830 static __kprobes void
 831 kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
 832                      struct pt_regs *regs)
 833 {
 834         struct event_file_link *link;
 835
 836         list_for_each_entry_rcu(link, &tk->tp.files, list)
 837                 __kretprobe_trace_func(tk, ri, regs, link->file);
 838 }
 839
 840 /* Event entry printers */
 841 static enum print_line_t
 842 print_kprobe_event(struct trace_iterator *iter, int flags,
 843                    struct trace_event *event)
 844 {
 845         struct kprobe_trace_entry_head *field;
 846         struct trace_seq *s = &iter->seq;
 847         struct trace_probe *tp;
 848         u8 *data;
 849         int i;
 850
 851         field = (struct kprobe_trace_entry_head *)iter->ent;
 852         tp = container_of(event, struct trace_probe, call.event);
 853
 854         if (!trace_seq_printf(s, "%s: (", tp->call.name))
 855                 goto partial;
 856
 857         if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
 858                 goto partial;
 859
 860         if (!trace_seq_puts(s, ")"))
 861                 goto partial;
 862
 863         data = (u8 *)&field[1];
 864         for (i = 0; i < tp->nr_args; i++)
 865                 if (!tp->args[i].type->print(s, tp->args[i].name,
 866                                              data + tp->args[i].offset, field))
 867                         goto partial;
 868
 869         if (!trace_seq_puts(s, "\n"))
 870                 goto partial;
 871
 872         return TRACE_TYPE_HANDLED;
 873 partial:
 874         return TRACE_TYPE_PARTIAL_LINE;
 875 }
 876
 877 static enum print_line_t
 878 print_kretprobe_event(struct trace_iterator *iter, int flags,
 879                       struct trace_event *event)
 880 {
 881         struct kretprobe_trace_entry_head *field;
 882         struct trace_seq *s = &iter->seq;
 883         struct trace_probe *tp;
 884         u8 *data;
 885         int i;
 886
 887         field = (struct kretprobe_trace_entry_head *)iter->ent;
 888         tp = container_of(event, struct trace_probe, call.event);
 889
 890         if (!trace_seq_printf(s, "%s: (", tp->call.name))
 891                 goto partial;
 892
 893         if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
 894                 goto partial;
 895
 896         if (!trace_seq_puts(s, " <- "))
 897                 goto partial;
 898
 899         if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
 900                 goto partial;
 901
 902         if (!trace_seq_puts(s, ")"))
 903                 goto partial;
 904
 905         data = (u8 *)&field[1];
 906         for (i = 0; i < tp->nr_args; i++)
 907                 if (!tp->args[i].type->print(s, tp->args[i].name,
 908                                              data + tp->args[i].offset, field))
 909                         goto partial;
 910
 911         if (!trace_seq_puts(s, "\n"))
 912                 goto partial;
 913
 914         return TRACE_TYPE_HANDLED;
 915 partial:
 916         return TRACE_TYPE_PARTIAL_LINE;
 917 }
 918
 919
 920 static int kprobe_event_define_fields(struct ftrace_event_call *event_call)
 921 {
 922         int ret, i;
 923         struct kprobe_trace_entry_head field;
 924         struct trace_kprobe *tk = (struct trace_kprobe *)event_call->data;
 925
 926         DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
 927         /* Set argument names as fields */
 928         for (i = 0; i < tk->tp.nr_args; i++) {
 929                 struct probe_arg *parg = &tk->tp.args[i];
 930
 931                 ret = trace_define_field(event_call, parg->type->fmttype,
 932                                          parg->name,
 933                                          sizeof(field) + parg->offset,
 934                                          parg->type->size,
 935                                          parg->type->is_signed,
 936                                          FILTER_OTHER);
 937                 if (ret)
 938                         return ret;
 939         }
 940         return 0;
 941 }
 942
 943 static int kretprobe_event_define_fields(struct ftrace_event_call *event_call)
 944 {
 945         int ret, i;
 946         struct kretprobe_trace_entry_head field;
 947         struct trace_kprobe *tk = (struct trace_kprobe *)event_call->data;
 948
 949         DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
 950         DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
 951         /* Set argument names as fields */
 952         for (i = 0; i < tk->tp.nr_args; i++) {
 953                 struct probe_arg *parg = &tk->tp.args[i];
 954
 955                 ret = trace_define_field(event_call, parg->type->fmttype,
 956                                          parg->name,
 957                                          sizeof(field) + parg->offset,
 958                                          parg->type->size,
 959                                          parg->type->is_signed,
 960                                          FILTER_OTHER);
 961                 if (ret)
 962                         return ret;
 963         }
 964         return 0;
 965 }
 966
 967 #ifdef CONFIG_PERF_EVENTS
 968
 969 /* Kprobe profile handler */
 970 static __kprobes void
 971 kprobe_perf_func(struct trace_kprobe *tk, struct pt_regs *regs)
 972 {
 973         struct ftrace_event_call *call = &tk->tp.call;
 974         struct kprobe_trace_entry_head *entry;
 975         struct hlist_head *head;
 976         int size, __size, dsize;
 977         int rctx;
 978
 979         head = this_cpu_ptr(call->perf_events);
 980         if (hlist_empty(head))
 981                 return;
 982
 983         dsize = __get_data_size(&tk->tp, regs);
 984         __size = sizeof(*entry) + tk->tp.size + dsize;
 985         size = ALIGN(__size + sizeof(u32), sizeof(u64));
 986         size -= sizeof(u32);
 987
 988         entry = perf_trace_buf_prepare(size, call->event.type, regs, &rctx);
 989         if (!entry)
 990                 return;
 991
 992         entry->ip = (unsigned long)tk->rp.kp.addr;
 993         memset(&entry[1], 0, dsize);
 994         store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
 995         perf_trace_buf_submit(entry, size, rctx, 0, 1, regs, head, NULL);
 996 }
 997
 998 /* Kretprobe profile handler */
 999 static __kprobes void
1000 kretprobe_perf_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1001                     struct pt_regs *regs)
1002 {
1003         struct ftrace_event_call *call = &tk->tp.call;
1004         struct kretprobe_trace_entry_head *entry;
1005         struct hlist_head *head;
1006         int size, __size, dsize;
1007         int rctx;
1008
1009         head = this_cpu_ptr(call->perf_events);
1010         if (hlist_empty(head))
1011                 return;
1012
1013         dsize = __get_data_size(&tk->tp, regs);
1014         __size = sizeof(*entry) + tk->tp.size + dsize;
1015         size = ALIGN(__size + sizeof(u32), sizeof(u64));
1016         size -= sizeof(u32);
1017
1018         entry = perf_trace_buf_prepare(size, call->event.type, regs, &rctx);
1019         if (!entry)
1020                 return;
1021
1022         entry->func = (unsigned long)tk->rp.kp.addr;
1023         entry->ret_ip = (unsigned long)ri->ret_addr;
1024         store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1025         perf_trace_buf_submit(entry, size, rctx, 0, 1, regs, head, NULL);
1026 }
1027 #endif  /* CONFIG_PERF_EVENTS */
1028
1029 /*
1030  * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex.
1031  *
1032  * kprobe_trace_self_tests_init() does enable_trace_probe/disable_trace_probe
1033  * lockless, but we can't race with this __init function.
1034  */
1035 static __kprobes
1036 int kprobe_register(struct ftrace_event_call *event,
1037                     enum trace_reg type, void *data)
1038 {
1039         struct trace_kprobe *tk = (struct trace_kprobe *)event->data;
1040         struct ftrace_event_file *file = data;
1041
1042         switch (type) {
1043         case TRACE_REG_REGISTER:
1044                 return enable_trace_kprobe(tk, file);
1045         case TRACE_REG_UNREGISTER:
1046                 return disable_trace_kprobe(tk, file);
1047
1048 #ifdef CONFIG_PERF_EVENTS
1049         case TRACE_REG_PERF_REGISTER:
1050                 return enable_trace_kprobe(tk, NULL);
1051         case TRACE_REG_PERF_UNREGISTER:
1052                 return disable_trace_kprobe(tk, NULL);
1053         case TRACE_REG_PERF_OPEN:
1054         case TRACE_REG_PERF_CLOSE:
1055         case TRACE_REG_PERF_ADD:
1056         case TRACE_REG_PERF_DEL:
1057                 return 0;
1058 #endif
1059         }
1060         return 0;
1061 }
1062
1063 static __kprobes
1064 int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1065 {
1066         struct trace_kprobe *tk = container_of(kp, struct trace_kprobe, rp.kp);
1067
1068         tk->nhit++;
1069
1070         if (tk->tp.flags & TP_FLAG_TRACE)
1071                 kprobe_trace_func(tk, regs);
1072 #ifdef CONFIG_PERF_EVENTS
1073         if (tk->tp.flags & TP_FLAG_PROFILE)
1074                 kprobe_perf_func(tk, regs);
1075 #endif
1076         return 0;       /* We don't tweek kernel, so just return 0 */
1077 }
1078
1079 static __kprobes
1080 int kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1081 {
1082         struct trace_kprobe *tk = container_of(ri->rp, struct trace_kprobe, rp);
1083
1084         tk->nhit++;
1085
1086         if (tk->tp.flags & TP_FLAG_TRACE)
1087                 kretprobe_trace_func(tk, ri, regs);
1088 #ifdef CONFIG_PERF_EVENTS
1089         if (tk->tp.flags & TP_FLAG_PROFILE)
1090                 kretprobe_perf_func(tk, ri, regs);
1091 #endif
1092         return 0;       /* We don't tweek kernel, so just return 0 */
1093 }
1094
1095 static struct trace_event_functions kretprobe_funcs = {
1096         .trace          = print_kretprobe_event
1097 };
1098
1099 static struct trace_event_functions kprobe_funcs = {
1100         .trace          = print_kprobe_event
1101 };
1102
1103 static int register_kprobe_event(struct trace_kprobe *tk)
1104 {
1105         struct ftrace_event_call *call = &tk->tp.call;
1106         int ret;
1107
1108         /* Initialize ftrace_event_call */
1109         INIT_LIST_HEAD(&call->class->fields);
1110         if (trace_kprobe_is_return(tk)) {
1111                 call->event.funcs = &kretprobe_funcs;
1112                 call->class->define_fields = kretprobe_event_define_fields;
1113         } else {
1114                 call->event.funcs = &kprobe_funcs;
1115                 call->class->define_fields = kprobe_event_define_fields;
1116         }
1117         if (set_print_fmt(&tk->tp, trace_kprobe_is_return(tk)) < 0)
1118                 return -ENOMEM;
1119         ret = register_ftrace_event(&call->event);
1120         if (!ret) {
1121                 kfree(call->print_fmt);
1122                 return -ENODEV;
1123         }
1124         call->flags = 0;
1125         call->class->reg = kprobe_register;
1126         call->data = tk;
1127         ret = trace_add_event_call(call);
1128         if (ret) {
1129                 pr_info("Failed to register kprobe event: %s\n", call->name);
1130                 kfree(call->print_fmt);
1131                 unregister_ftrace_event(&call->event);
1132         }
1133         return ret;
1134 }
1135
1136 static int unregister_kprobe_event(struct trace_kprobe *tk)
1137 {
1138         int ret;
1139
1140         /* tp->event is unregistered in trace_remove_event_call() */
1141         ret = trace_remove_event_call(&tk->tp.call);
1142         if (!ret)
1143                 kfree(tk->tp.call.print_fmt);
1144         return ret;
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_kprobe_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 struct ftrace_event_file *
1193 find_trace_probe_file(struct trace_kprobe *tk, struct trace_array *tr)
1194 {
1195         struct ftrace_event_file *file;
1196
1197         list_for_each_entry(file, &tr->events, list)
1198                 if (file->event_call == &tk->tp.call)
1199                         return file;
1200
1201         return NULL;
1202 }
1203
1204 /*
1205  * Nobody but us can call enable_trace_kprobe/disable_trace_kprobe at this
1206  * stage, we can do this lockless.
1207  */
1208 static __init int kprobe_trace_self_tests_init(void)
1209 {
1210         int ret, warn = 0;
1211         int (*target)(int, int, int, int, int, int);
1212         struct trace_kprobe *tk;
1213         struct ftrace_event_file *file;
1214
1215         target = kprobe_trace_selftest_target;
1216
1217         pr_info("Testing kprobe tracing: ");
1218
1219         ret = traceprobe_command("p:testprobe kprobe_trace_selftest_target "
1220                                   "$stack $stack0 +0($stack)",
1221                                   create_trace_kprobe);
1222         if (WARN_ON_ONCE(ret)) {
1223                 pr_warn("error on probing function entry.\n");
1224                 warn++;
1225         } else {
1226                 /* Enable trace point */
1227                 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1228                 if (WARN_ON_ONCE(tk == NULL)) {
1229                         pr_warn("error on getting new probe.\n");
1230                         warn++;
1231                 } else {
1232                         file = find_trace_probe_file(tk, top_trace_array());
1233                         if (WARN_ON_ONCE(file == NULL)) {
1234                                 pr_warn("error on getting probe file.\n");
1235                                 warn++;
1236                         } else
1237                                 enable_trace_kprobe(tk, file);
1238                 }
1239         }
1240
1241         ret = traceprobe_command("r:testprobe2 kprobe_trace_selftest_target "
1242                                   "$retval", create_trace_kprobe);
1243         if (WARN_ON_ONCE(ret)) {
1244                 pr_warn("error on probing function return.\n");
1245                 warn++;
1246         } else {
1247                 /* Enable trace point */
1248                 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1249                 if (WARN_ON_ONCE(tk == NULL)) {
1250                         pr_warn("error on getting 2nd new probe.\n");
1251                         warn++;
1252                 } else {
1253                         file = find_trace_probe_file(tk, top_trace_array());
1254                         if (WARN_ON_ONCE(file == NULL)) {
1255                                 pr_warn("error on getting probe file.\n");
1256                                 warn++;
1257                         } else
1258                                 enable_trace_kprobe(tk, file);
1259                 }
1260         }
1261
1262         if (warn)
1263                 goto end;
1264
1265         ret = target(1, 2, 3, 4, 5, 6);
1266
1267         /* Disable trace points before removing it */
1268         tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1269         if (WARN_ON_ONCE(tk == NULL)) {
1270                 pr_warn("error on getting test probe.\n");
1271                 warn++;
1272         } else {
1273                 file = find_trace_probe_file(tk, top_trace_array());
1274                 if (WARN_ON_ONCE(file == NULL)) {
1275                         pr_warn("error on getting probe file.\n");
1276                         warn++;
1277                 } else
1278                         disable_trace_kprobe(tk, file);
1279         }
1280
1281         tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1282         if (WARN_ON_ONCE(tk == NULL)) {
1283                 pr_warn("error on getting 2nd test probe.\n");
1284                 warn++;
1285         } else {
1286                 file = find_trace_probe_file(tk, top_trace_array());
1287                 if (WARN_ON_ONCE(file == NULL)) {
1288                         pr_warn("error on getting probe file.\n");
1289                         warn++;
1290                 } else
1291                         disable_trace_kprobe(tk, file);
1292         }
1293
1294         ret = traceprobe_command("-:testprobe", create_trace_kprobe);
1295         if (WARN_ON_ONCE(ret)) {
1296                 pr_warn("error on deleting a probe.\n");
1297                 warn++;
1298         }
1299
1300         ret = traceprobe_command("-:testprobe2", create_trace_kprobe);
1301         if (WARN_ON_ONCE(ret)) {
1302                 pr_warn("error on deleting a probe.\n");
1303                 warn++;
1304         }
1305
1306 end:
1307         release_all_trace_kprobes();
1308         if (warn)
1309                 pr_cont("NG: Some tests are failed. Please check them.\n");
1310         else
1311                 pr_cont("OK\n");
1312         return 0;
1313 }
1314
1315 late_initcall(kprobe_trace_self_tests_init);
1316
1317 #endif