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 #define pr_fmt(fmt)     "trace_kprobe: " fmt
  20
  21 #include <linux/module.h>
  22 #include <linux/uaccess.h>
  23 #include <linux/rculist.h>
  24
  25 #include "trace_probe.h"
  26
  27 #define KPROBE_EVENT_SYSTEM "kprobes"
  28
  29 /**
  30  * Kprobe event core functions
  31  */
  32 struct trace_kprobe {
  33         struct list_head        list;
  34         struct kretprobe        rp;     /* Use rp.kp for kprobe use */
  35         unsigned long __percpu *nhit;
  36         const char              *symbol;        /* symbol name */
  37         struct trace_probe      tp;
  38 };
  39
  40 #define SIZEOF_TRACE_KPROBE(n)                          \
  41         (offsetof(struct trace_kprobe, tp.args) +       \
  42         (sizeof(struct probe_arg) * (n)))
  43
  44
  45 static nokprobe_inline bool trace_kprobe_is_return(struct trace_kprobe *tk)
  46 {
  47         return tk->rp.handler != NULL;
  48 }
  49
  50 static nokprobe_inline const char *trace_kprobe_symbol(struct trace_kprobe *tk)
  51 {
  52         return tk->symbol ? tk->symbol : "unknown";
  53 }
  54
  55 static nokprobe_inline unsigned long trace_kprobe_offset(struct trace_kprobe *tk)
  56 {
  57         return tk->rp.kp.offset;
  58 }
  59
  60 static nokprobe_inline bool trace_kprobe_has_gone(struct trace_kprobe *tk)
  61 {
  62         return !!(kprobe_gone(&tk->rp.kp));
  63 }
  64
  65 static nokprobe_inline bool trace_kprobe_within_module(struct trace_kprobe *tk,
  66                                                  struct module *mod)
  67 {
  68         int len = strlen(mod->name);
  69         const char *name = trace_kprobe_symbol(tk);
  70         return strncmp(mod->name, name, len) == 0 && name[len] == ':';
  71 }
  72
  73 static nokprobe_inline bool trace_kprobe_is_on_module(struct trace_kprobe *tk)
  74 {
  75         return !!strchr(trace_kprobe_symbol(tk), ':');
  76 }
  77
  78 static nokprobe_inline unsigned long trace_kprobe_nhit(struct trace_kprobe *tk)
  79 {
  80         unsigned long nhit = 0;
  81         int cpu;
  82
  83         for_each_possible_cpu(cpu)
  84                 nhit += *per_cpu_ptr(tk->nhit, cpu);
  85
  86         return nhit;
  87 }
  88
  89 static int register_kprobe_event(struct trace_kprobe *tk);
  90 static int unregister_kprobe_event(struct trace_kprobe *tk);
  91
  92 static DEFINE_MUTEX(probe_lock);
  93 static LIST_HEAD(probe_list);
  94
  95 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
  96 static int kretprobe_dispatcher(struct kretprobe_instance *ri,
  97                                 struct pt_regs *regs);
  98
  99 /* Memory fetching by symbol */
 100 struct symbol_cache {
 101         char            *symbol;
 102         long            offset;
 103         unsigned long   addr;
 104 };
 105
 106 unsigned long update_symbol_cache(struct symbol_cache *sc)
 107 {
 108         sc->addr = (unsigned long)kallsyms_lookup_name(sc->symbol);
 109
 110         if (sc->addr)
 111                 sc->addr += sc->offset;
 112
 113         return sc->addr;
 114 }
 115
 116 void free_symbol_cache(struct symbol_cache *sc)
 117 {
 118         kfree(sc->symbol);
 119         kfree(sc);
 120 }
 121
 122 struct symbol_cache *alloc_symbol_cache(const char *sym, long offset)
 123 {
 124         struct symbol_cache *sc;
 125
 126         if (!sym || strlen(sym) == 0)
 127                 return NULL;
 128
 129         sc = kzalloc(sizeof(struct symbol_cache), GFP_KERNEL);
 130         if (!sc)
 131                 return NULL;
 132
 133         sc->symbol = kstrdup(sym, GFP_KERNEL);
 134         if (!sc->symbol) {
 135                 kfree(sc);
 136                 return NULL;
 137         }
 138         sc->offset = offset;
 139         update_symbol_cache(sc);
 140
 141         return sc;
 142 }
 143
 144 /*
 145  * Kprobes-specific fetch functions
 146  */
 147 #define DEFINE_FETCH_stack(type)                                        \
 148 static void FETCH_FUNC_NAME(stack, type)(struct pt_regs *regs,          \
 149                                           void *offset, void *dest)     \
 150 {                                                                       \
 151         *(type *)dest = (type)regs_get_kernel_stack_nth(regs,           \
 152                                 (unsigned int)((unsigned long)offset)); \
 153 }                                                                       \
 154 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(stack, type));
 155
 156 DEFINE_BASIC_FETCH_FUNCS(stack)
 157 /* No string on the stack entry */
 158 #define fetch_stack_string      NULL
 159 #define fetch_stack_string_size NULL
 160
 161 #define DEFINE_FETCH_memory(type)                                       \
 162 static void FETCH_FUNC_NAME(memory, type)(struct pt_regs *regs,         \
 163                                           void *addr, void *dest)       \
 164 {                                                                       \
 165         type retval;                                                    \
 166         if (probe_kernel_address(addr, retval))                         \
 167                 *(type *)dest = 0;                                      \
 168         else                                                            \
 169                 *(type *)dest = retval;                                 \
 170 }                                                                       \
 171 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(memory, type));
 172
 173 DEFINE_BASIC_FETCH_FUNCS(memory)
 174 /*
 175  * Fetch a null-terminated string. Caller MUST set *(u32 *)dest with max
 176  * length and relative data location.
 177  */
 178 static void FETCH_FUNC_NAME(memory, string)(struct pt_regs *regs,
 179                                             void *addr, void *dest)
 180 {
 181         int maxlen = get_rloc_len(*(u32 *)dest);
 182         u8 *dst = get_rloc_data(dest);
 183         long ret;
 184
 185         if (!maxlen)
 186                 return;
 187
 188         /*
 189          * Try to get string again, since the string can be changed while
 190          * probing.
 191          */
 192         ret = strncpy_from_unsafe(dst, addr, maxlen);
 193
 194         if (ret < 0) {  /* Failed to fetch string */
 195                 dst[0] = '\0';
 196                 *(u32 *)dest = make_data_rloc(0, get_rloc_offs(*(u32 *)dest));
 197         } else {
 198                 *(u32 *)dest = make_data_rloc(ret, get_rloc_offs(*(u32 *)dest));
 199         }
 200 }
 201 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(memory, string));
 202
 203 /* Return the length of string -- including null terminal byte */
 204 static void FETCH_FUNC_NAME(memory, string_size)(struct pt_regs *regs,
 205                                                  void *addr, void *dest)
 206 {
 207         mm_segment_t old_fs;
 208         int ret, len = 0;
 209         u8 c;
 210
 211         old_fs = get_fs();
 212         set_fs(KERNEL_DS);
 213         pagefault_disable();
 214
 215         do {
 216                 ret = __copy_from_user_inatomic(&c, (u8 *)addr + len, 1);
 217                 len++;
 218         } while (c && ret == 0 && len < MAX_STRING_SIZE);
 219
 220         pagefault_enable();
 221         set_fs(old_fs);
 222
 223         if (ret < 0)    /* Failed to check the length */
 224                 *(u32 *)dest = 0;
 225         else
 226                 *(u32 *)dest = len;
 227 }
 228 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(memory, string_size));
 229
 230 #define DEFINE_FETCH_symbol(type)                                       \
 231 void FETCH_FUNC_NAME(symbol, type)(struct pt_regs *regs, void *data, void *dest)\
 232 {                                                                       \
 233         struct symbol_cache *sc = data;                                 \
 234         if (sc->addr)                                                   \
 235                 fetch_memory_##type(regs, (void *)sc->addr, dest);      \
 236         else                                                            \
 237                 *(type *)dest = 0;                                      \
 238 }                                                                       \
 239 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(symbol, type));
 240
 241 DEFINE_BASIC_FETCH_FUNCS(symbol)
 242 DEFINE_FETCH_symbol(string)
 243 DEFINE_FETCH_symbol(string_size)
 244
 245 /* kprobes don't support file_offset fetch methods */
 246 #define fetch_file_offset_u8            NULL
 247 #define fetch_file_offset_u16           NULL
 248 #define fetch_file_offset_u32           NULL
 249 #define fetch_file_offset_u64           NULL
 250 #define fetch_file_offset_string        NULL
 251 #define fetch_file_offset_string_size   NULL
 252
 253 /* Fetch type information table */
 254 static const struct fetch_type kprobes_fetch_type_table[] = {
 255         /* Special types */
 256         [FETCH_TYPE_STRING] = __ASSIGN_FETCH_TYPE("string", string, string,
 257                                         sizeof(u32), 1, "__data_loc char[]"),
 258         [FETCH_TYPE_STRSIZE] = __ASSIGN_FETCH_TYPE("string_size", u32,
 259                                         string_size, sizeof(u32), 0, "u32"),
 260         /* Basic types */
 261         ASSIGN_FETCH_TYPE(u8,  u8,  0),
 262         ASSIGN_FETCH_TYPE(u16, u16, 0),
 263         ASSIGN_FETCH_TYPE(u32, u32, 0),
 264         ASSIGN_FETCH_TYPE(u64, u64, 0),
 265         ASSIGN_FETCH_TYPE(s8,  u8,  1),
 266         ASSIGN_FETCH_TYPE(s16, u16, 1),
 267         ASSIGN_FETCH_TYPE(s32, u32, 1),
 268         ASSIGN_FETCH_TYPE(s64, u64, 1),
 269         ASSIGN_FETCH_TYPE_ALIAS(x8,  u8,  u8,  0),
 270         ASSIGN_FETCH_TYPE_ALIAS(x16, u16, u16, 0),
 271         ASSIGN_FETCH_TYPE_ALIAS(x32, u32, u32, 0),
 272         ASSIGN_FETCH_TYPE_ALIAS(x64, u64, u64, 0),
 273
 274         ASSIGN_FETCH_TYPE_END
 275 };
 276
 277 /*
 278  * Allocate new trace_probe and initialize it (including kprobes).
 279  */
 280 static struct trace_kprobe *alloc_trace_kprobe(const char *group,
 281                                              const char *event,
 282                                              void *addr,
 283                                              const char *symbol,
 284                                              unsigned long offs,
 285                                              int nargs, bool is_return)
 286 {
 287         struct trace_kprobe *tk;
 288         int ret = -ENOMEM;
 289
 290         tk = kzalloc(SIZEOF_TRACE_KPROBE(nargs), GFP_KERNEL);
 291         if (!tk)
 292                 return ERR_PTR(ret);
 293
 294         tk->nhit = alloc_percpu(unsigned long);
 295         if (!tk->nhit)
 296                 goto error;
 297
 298         if (symbol) {
 299                 tk->symbol = kstrdup(symbol, GFP_KERNEL);
 300                 if (!tk->symbol)
 301                         goto error;
 302                 tk->rp.kp.symbol_name = tk->symbol;
 303                 tk->rp.kp.offset = offs;
 304         } else
 305                 tk->rp.kp.addr = addr;
 306
 307         if (is_return)
 308                 tk->rp.handler = kretprobe_dispatcher;
 309         else
 310                 tk->rp.kp.pre_handler = kprobe_dispatcher;
 311
 312         if (!event || !is_good_name(event)) {
 313                 ret = -EINVAL;
 314                 goto error;
 315         }
 316
 317         tk->tp.call.class = &tk->tp.class;
 318         tk->tp.call.name = kstrdup(event, GFP_KERNEL);
 319         if (!tk->tp.call.name)
 320                 goto error;
 321
 322         if (!group || !is_good_name(group)) {
 323                 ret = -EINVAL;
 324                 goto error;
 325         }
 326
 327         tk->tp.class.system = kstrdup(group, GFP_KERNEL);
 328         if (!tk->tp.class.system)
 329                 goto error;
 330
 331         INIT_LIST_HEAD(&tk->list);
 332         INIT_LIST_HEAD(&tk->tp.files);
 333         return tk;
 334 error:
 335         kfree(tk->tp.call.name);
 336         kfree(tk->symbol);
 337         free_percpu(tk->nhit);
 338         kfree(tk);
 339         return ERR_PTR(ret);
 340 }
 341
 342 static void free_trace_kprobe(struct trace_kprobe *tk)
 343 {
 344         int i;
 345
 346         for (i = 0; i < tk->tp.nr_args; i++)
 347                 traceprobe_free_probe_arg(&tk->tp.args[i]);
 348
 349         kfree(tk->tp.call.class->system);
 350         kfree(tk->tp.call.name);
 351         kfree(tk->symbol);
 352         free_percpu(tk->nhit);
 353         kfree(tk);
 354 }
 355
 356 static struct trace_kprobe *find_trace_kprobe(const char *event,
 357                                               const char *group)
 358 {
 359         struct trace_kprobe *tk;
 360
 361         list_for_each_entry(tk, &probe_list, list)
 362                 if (strcmp(trace_event_name(&tk->tp.call), event) == 0 &&
 363                     strcmp(tk->tp.call.class->system, group) == 0)
 364                         return tk;
 365         return NULL;
 366 }
 367
 368 /*
 369  * Enable trace_probe
 370  * if the file is NULL, enable "perf" handler, or enable "trace" handler.
 371  */
 372 static int
 373 enable_trace_kprobe(struct trace_kprobe *tk, struct trace_event_file *file)
 374 {
 375         int ret = 0;
 376
 377         if (file) {
 378                 struct event_file_link *link;
 379
 380                 link = kmalloc(sizeof(*link), GFP_KERNEL);
 381                 if (!link) {
 382                         ret = -ENOMEM;
 383                         goto out;
 384                 }
 385
 386                 link->file = file;
 387                 list_add_tail_rcu(&link->list, &tk->tp.files);
 388
 389                 tk->tp.flags |= TP_FLAG_TRACE;
 390         } else
 391                 tk->tp.flags |= TP_FLAG_PROFILE;
 392
 393         if (trace_probe_is_registered(&tk->tp) && !trace_kprobe_has_gone(tk)) {
 394                 if (trace_kprobe_is_return(tk))
 395                         ret = enable_kretprobe(&tk->rp);
 396                 else
 397                         ret = enable_kprobe(&tk->rp.kp);
 398         }
 399  out:
 400         return ret;
 401 }
 402
 403 /*
 404  * Disable trace_probe
 405  * if the file is NULL, disable "perf" handler, or disable "trace" handler.
 406  */
 407 static int
 408 disable_trace_kprobe(struct trace_kprobe *tk, struct trace_event_file *file)
 409 {
 410         struct event_file_link *link = NULL;
 411         int wait = 0;
 412         int ret = 0;
 413
 414         if (file) {
 415                 link = find_event_file_link(&tk->tp, file);
 416                 if (!link) {
 417                         ret = -EINVAL;
 418                         goto out;
 419                 }
 420
 421                 list_del_rcu(&link->list);
 422                 wait = 1;
 423                 if (!list_empty(&tk->tp.files))
 424                         goto out;
 425
 426                 tk->tp.flags &= ~TP_FLAG_TRACE;
 427         } else
 428                 tk->tp.flags &= ~TP_FLAG_PROFILE;
 429
 430         if (!trace_probe_is_enabled(&tk->tp) && trace_probe_is_registered(&tk->tp)) {
 431                 if (trace_kprobe_is_return(tk))
 432                         disable_kretprobe(&tk->rp);
 433                 else
 434                         disable_kprobe(&tk->rp.kp);
 435                 wait = 1;
 436         }
 437  out:
 438         if (wait) {
 439                 /*
 440                  * Synchronize with kprobe_trace_func/kretprobe_trace_func
 441                  * to ensure disabled (all running handlers are finished).
 442                  * This is not only for kfree(), but also the caller,
 443                  * trace_remove_event_call() supposes it for releasing
 444                  * event_call related objects, which will be accessed in
 445                  * the kprobe_trace_func/kretprobe_trace_func.
 446                  */
 447                 synchronize_sched();
 448                 kfree(link);    /* Ignored if link == NULL */
 449         }
 450
 451         return ret;
 452 }
 453
 454 /* Internal register function - just handle k*probes and flags */
 455 static int __register_trace_kprobe(struct trace_kprobe *tk)
 456 {
 457         int i, ret;
 458
 459         if (trace_probe_is_registered(&tk->tp))
 460                 return -EINVAL;
 461
 462         for (i = 0; i < tk->tp.nr_args; i++)
 463                 traceprobe_update_arg(&tk->tp.args[i]);
 464
 465         /* Set/clear disabled flag according to tp->flag */
 466         if (trace_probe_is_enabled(&tk->tp))
 467                 tk->rp.kp.flags &= ~KPROBE_FLAG_DISABLED;
 468         else
 469                 tk->rp.kp.flags |= KPROBE_FLAG_DISABLED;
 470
 471         if (trace_kprobe_is_return(tk))
 472                 ret = register_kretprobe(&tk->rp);
 473         else
 474                 ret = register_kprobe(&tk->rp.kp);
 475
 476         if (ret == 0)
 477                 tk->tp.flags |= TP_FLAG_REGISTERED;
 478         else {
 479                 pr_warn("Could not insert probe at %s+%lu: %d\n",
 480                         trace_kprobe_symbol(tk), trace_kprobe_offset(tk), ret);
 481                 if (ret == -ENOENT && trace_kprobe_is_on_module(tk)) {
 482                         pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
 483                         ret = 0;
 484                 } else if (ret == -EILSEQ) {
 485                         pr_warn("Probing address(0x%p) is not an instruction boundary.\n",
 486                                 tk->rp.kp.addr);
 487                         ret = -EINVAL;
 488                 }
 489         }
 490
 491         return ret;
 492 }
 493
 494 /* Internal unregister function - just handle k*probes and flags */
 495 static void __unregister_trace_kprobe(struct trace_kprobe *tk)
 496 {
 497         if (trace_probe_is_registered(&tk->tp)) {
 498                 if (trace_kprobe_is_return(tk))
 499                         unregister_kretprobe(&tk->rp);
 500                 else
 501                         unregister_kprobe(&tk->rp.kp);
 502                 tk->tp.flags &= ~TP_FLAG_REGISTERED;
 503                 /* Cleanup kprobe for reuse */
 504                 if (tk->rp.kp.symbol_name)
 505                         tk->rp.kp.addr = NULL;
 506         }
 507 }
 508
 509 /* Unregister a trace_probe and probe_event: call with locking probe_lock */
 510 static int unregister_trace_kprobe(struct trace_kprobe *tk)
 511 {
 512         /* Enabled event can not be unregistered */
 513         if (trace_probe_is_enabled(&tk->tp))
 514                 return -EBUSY;
 515
 516         /* Will fail if probe is being used by ftrace or perf */
 517         if (unregister_kprobe_event(tk))
 518                 return -EBUSY;
 519
 520         __unregister_trace_kprobe(tk);
 521         list_del(&tk->list);
 522
 523         return 0;
 524 }
 525
 526 /* Register a trace_probe and probe_event */
 527 static int register_trace_kprobe(struct trace_kprobe *tk)
 528 {
 529         struct trace_kprobe *old_tk;
 530         int ret;
 531
 532         mutex_lock(&probe_lock);
 533
 534         /* Delete old (same name) event if exist */
 535         old_tk = find_trace_kprobe(trace_event_name(&tk->tp.call),
 536                         tk->tp.call.class->system);
 537         if (old_tk) {
 538                 ret = unregister_trace_kprobe(old_tk);
 539                 if (ret < 0)
 540                         goto end;
 541                 free_trace_kprobe(old_tk);
 542         }
 543
 544         /* Register new event */
 545         ret = register_kprobe_event(tk);
 546         if (ret) {
 547                 pr_warn("Failed to register probe event(%d)\n", ret);
 548                 goto end;
 549         }
 550
 551         /* Register k*probe */
 552         ret = __register_trace_kprobe(tk);
 553         if (ret < 0)
 554                 unregister_kprobe_event(tk);
 555         else
 556                 list_add_tail(&tk->list, &probe_list);
 557
 558 end:
 559         mutex_unlock(&probe_lock);
 560         return ret;
 561 }
 562
 563 /* Module notifier call back, checking event on the module */
 564 static int trace_kprobe_module_callback(struct notifier_block *nb,
 565                                        unsigned long val, void *data)
 566 {
 567         struct module *mod = data;
 568         struct trace_kprobe *tk;
 569         int ret;
 570
 571         if (val != MODULE_STATE_COMING)
 572                 return NOTIFY_DONE;
 573
 574         /* Update probes on coming module */
 575         mutex_lock(&probe_lock);
 576         list_for_each_entry(tk, &probe_list, list) {
 577                 if (trace_kprobe_within_module(tk, mod)) {
 578                         /* Don't need to check busy - this should have gone. */
 579                         __unregister_trace_kprobe(tk);
 580                         ret = __register_trace_kprobe(tk);
 581                         if (ret)
 582                                 pr_warn("Failed to re-register probe %s on %s: %d\n",
 583                                         trace_event_name(&tk->tp.call),
 584                                         mod->name, ret);
 585                 }
 586         }
 587         mutex_unlock(&probe_lock);
 588
 589         return NOTIFY_DONE;
 590 }
 591
 592 static struct notifier_block trace_kprobe_module_nb = {
 593         .notifier_call = trace_kprobe_module_callback,
 594         .priority = 1   /* Invoked after kprobe module callback */
 595 };
 596
 597 static int create_trace_kprobe(int argc, char **argv)
 598 {
 599         /*
 600          * Argument syntax:
 601          *  - Add kprobe: p[:[GRP/]EVENT] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
 602          *  - Add kretprobe: r[:[GRP/]EVENT] [MOD:]KSYM[+0] [FETCHARGS]
 603          * Fetch args:
 604          *  $retval     : fetch return value
 605          *  $stack      : fetch stack address
 606          *  $stackN     : fetch Nth of stack (N:0-)
 607          *  $comm       : fetch current task comm
 608          *  @ADDR       : fetch memory at ADDR (ADDR should be in kernel)
 609          *  @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
 610          *  %REG        : fetch register REG
 611          * Dereferencing memory fetch:
 612          *  +|-offs(ARG) : fetch memory at ARG +|- offs address.
 613          * Alias name of args:
 614          *  NAME=FETCHARG : set NAME as alias of FETCHARG.
 615          * Type of args:
 616          *  FETCHARG:TYPE : use TYPE instead of unsigned long.
 617          */
 618         struct trace_kprobe *tk;
 619         int i, ret = 0;
 620         bool is_return = false, is_delete = false;
 621         char *symbol = NULL, *event = NULL, *group = NULL;
 622         char *arg;
 623         unsigned long offset = 0;
 624         void *addr = NULL;
 625         char buf[MAX_EVENT_NAME_LEN];
 626
 627         /* argc must be >= 1 */
 628         if (argv[0][0] == 'p')
 629                 is_return = false;
 630         else if (argv[0][0] == 'r')
 631                 is_return = true;
 632         else if (argv[0][0] == '-')
 633                 is_delete = true;
 634         else {
 635                 pr_info("Probe definition must be started with 'p', 'r' or"
 636                         " '-'.\n");
 637                 return -EINVAL;
 638         }
 639
 640         if (argv[0][1] == ':') {
 641                 event = &argv[0][2];
 642                 if (strchr(event, '/')) {
 643                         group = event;
 644                         event = strchr(group, '/') + 1;
 645                         event[-1] = '\0';
 646                         if (strlen(group) == 0) {
 647                                 pr_info("Group name is not specified\n");
 648                                 return -EINVAL;
 649                         }
 650                 }
 651                 if (strlen(event) == 0) {
 652                         pr_info("Event name is not specified\n");
 653                         return -EINVAL;
 654                 }
 655         }
 656         if (!group)
 657                 group = KPROBE_EVENT_SYSTEM;
 658
 659         if (is_delete) {
 660                 if (!event) {
 661                         pr_info("Delete command needs an event name.\n");
 662                         return -EINVAL;
 663                 }
 664                 mutex_lock(&probe_lock);
 665                 tk = find_trace_kprobe(event, group);
 666                 if (!tk) {
 667                         mutex_unlock(&probe_lock);
 668                         pr_info("Event %s/%s doesn't exist.\n", group, event);
 669                         return -ENOENT;
 670                 }
 671                 /* delete an event */
 672                 ret = unregister_trace_kprobe(tk);
 673                 if (ret == 0)
 674                         free_trace_kprobe(tk);
 675                 mutex_unlock(&probe_lock);
 676                 return ret;
 677         }
 678
 679         if (argc < 2) {
 680                 pr_info("Probe point is not specified.\n");
 681                 return -EINVAL;
 682         }
 683         if (isdigit(argv[1][0])) {
 684                 /* an address specified */
 685                 ret = kstrtoul(&argv[1][0], 0, (unsigned long *)&addr);
 686                 if (ret) {
 687                         pr_info("Failed to parse address.\n");
 688                         return ret;
 689                 }
 690         } else {
 691                 /* a symbol specified */
 692                 symbol = argv[1];
 693                 /* TODO: support .init module functions */
 694                 ret = traceprobe_split_symbol_offset(symbol, &offset);
 695                 if (ret) {
 696                         pr_info("Failed to parse symbol.\n");
 697                         return ret;
 698                 }
 699                 if (offset && is_return &&
 700                     !function_offset_within_entry(NULL, symbol, offset)) {
 701                         pr_info("Given offset is not valid for return probe.\n");
 702                         return -EINVAL;
 703                 }
 704         }
 705         argc -= 2; argv += 2;
 706
 707         /* setup a probe */
 708         if (!event) {
 709                 /* Make a new event name */
 710                 if (symbol)
 711                         snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
 712                                  is_return ? 'r' : 'p', symbol, offset);
 713                 else
 714                         snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
 715                                  is_return ? 'r' : 'p', addr);
 716                 event = buf;
 717         }
 718         tk = alloc_trace_kprobe(group, event, addr, symbol, offset, argc,
 719                                is_return);
 720         if (IS_ERR(tk)) {
 721                 pr_info("Failed to allocate trace_probe.(%d)\n",
 722                         (int)PTR_ERR(tk));
 723                 return PTR_ERR(tk);
 724         }
 725
 726         /* parse arguments */
 727         ret = 0;
 728         for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
 729                 struct probe_arg *parg = &tk->tp.args[i];
 730
 731                 /* Increment count for freeing args in error case */
 732                 tk->tp.nr_args++;
 733
 734                 /* Parse argument name */
 735                 arg = strchr(argv[i], '=');
 736                 if (arg) {
 737                         *arg++ = '\0';
 738                         parg->name = kstrdup(argv[i], GFP_KERNEL);
 739                 } else {
 740                         arg = argv[i];
 741                         /* If argument name is omitted, set "argN" */
 742                         snprintf(buf, MAX_EVENT_NAME_LEN, "arg%d", i + 1);
 743                         parg->name = kstrdup(buf, GFP_KERNEL);
 744                 }
 745
 746                 if (!parg->name) {
 747                         pr_info("Failed to allocate argument[%d] name.\n", i);
 748                         ret = -ENOMEM;
 749                         goto error;
 750                 }
 751
 752                 if (!is_good_name(parg->name)) {
 753                         pr_info("Invalid argument[%d] name: %s\n",
 754                                 i, parg->name);
 755                         ret = -EINVAL;
 756                         goto error;
 757                 }
 758
 759                 if (traceprobe_conflict_field_name(parg->name,
 760                                                         tk->tp.args, i)) {
 761                         pr_info("Argument[%d] name '%s' conflicts with "
 762                                 "another field.\n", i, argv[i]);
 763                         ret = -EINVAL;
 764                         goto error;
 765                 }
 766
 767                 /* Parse fetch argument */
 768                 ret = traceprobe_parse_probe_arg(arg, &tk->tp.size, parg,
 769                                                 is_return, true,
 770                                                 kprobes_fetch_type_table);
 771                 if (ret) {
 772                         pr_info("Parse error at argument[%d]. (%d)\n", i, ret);
 773                         goto error;
 774                 }
 775         }
 776
 777         ret = register_trace_kprobe(tk);
 778         if (ret)
 779                 goto error;
 780         return 0;
 781
 782 error:
 783         free_trace_kprobe(tk);
 784         return ret;
 785 }
 786
 787 static int release_all_trace_kprobes(void)
 788 {
 789         struct trace_kprobe *tk;
 790         int ret = 0;
 791
 792         mutex_lock(&probe_lock);
 793         /* Ensure no probe is in use. */
 794         list_for_each_entry(tk, &probe_list, list)
 795                 if (trace_probe_is_enabled(&tk->tp)) {
 796                         ret = -EBUSY;
 797                         goto end;
 798                 }
 799         /* TODO: Use batch unregistration */
 800         while (!list_empty(&probe_list)) {
 801                 tk = list_entry(probe_list.next, struct trace_kprobe, list);
 802                 ret = unregister_trace_kprobe(tk);
 803                 if (ret)
 804                         goto end;
 805                 free_trace_kprobe(tk);
 806         }
 807
 808 end:
 809         mutex_unlock(&probe_lock);
 810
 811         return ret;
 812 }
 813
 814 /* Probes listing interfaces */
 815 static void *probes_seq_start(struct seq_file *m, loff_t *pos)
 816 {
 817         mutex_lock(&probe_lock);
 818         return seq_list_start(&probe_list, *pos);
 819 }
 820
 821 static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
 822 {
 823         return seq_list_next(v, &probe_list, pos);
 824 }
 825
 826 static void probes_seq_stop(struct seq_file *m, void *v)
 827 {
 828         mutex_unlock(&probe_lock);
 829 }
 830
 831 static int probes_seq_show(struct seq_file *m, void *v)
 832 {
 833         struct trace_kprobe *tk = v;
 834         int i;
 835
 836         seq_putc(m, trace_kprobe_is_return(tk) ? 'r' : 'p');
 837         seq_printf(m, ":%s/%s", tk->tp.call.class->system,
 838                         trace_event_name(&tk->tp.call));
 839
 840         if (!tk->symbol)
 841                 seq_printf(m, " 0x%p", tk->rp.kp.addr);
 842         else if (tk->rp.kp.offset)
 843                 seq_printf(m, " %s+%u", trace_kprobe_symbol(tk),
 844                            tk->rp.kp.offset);
 845         else
 846                 seq_printf(m, " %s", trace_kprobe_symbol(tk));
 847
 848         for (i = 0; i < tk->tp.nr_args; i++)
 849                 seq_printf(m, " %s=%s", tk->tp.args[i].name, tk->tp.args[i].comm);
 850         seq_putc(m, '\n');
 851
 852         return 0;
 853 }
 854
 855 static const struct seq_operations probes_seq_op = {
 856         .start  = probes_seq_start,
 857         .next   = probes_seq_next,
 858         .stop   = probes_seq_stop,
 859         .show   = probes_seq_show
 860 };
 861
 862 static int probes_open(struct inode *inode, struct file *file)
 863 {
 864         int ret;
 865
 866         if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
 867                 ret = release_all_trace_kprobes();
 868                 if (ret < 0)
 869                         return ret;
 870         }
 871
 872         return seq_open(file, &probes_seq_op);
 873 }
 874
 875 static ssize_t probes_write(struct file *file, const char __user *buffer,
 876                             size_t count, loff_t *ppos)
 877 {
 878         return traceprobe_probes_write(file, buffer, count, ppos,
 879                         create_trace_kprobe);
 880 }
 881
 882 static const struct file_operations kprobe_events_ops = {
 883         .owner          = THIS_MODULE,
 884         .open           = probes_open,
 885         .read           = seq_read,
 886         .llseek         = seq_lseek,
 887         .release        = seq_release,
 888         .write          = probes_write,
 889 };
 890
 891 /* Probes profiling interfaces */
 892 static int probes_profile_seq_show(struct seq_file *m, void *v)
 893 {
 894         struct trace_kprobe *tk = v;
 895
 896         seq_printf(m, "  %-44s %15lu %15lu\n",
 897                    trace_event_name(&tk->tp.call),
 898                    trace_kprobe_nhit(tk),
 899                    tk->rp.kp.nmissed);
 900
 901         return 0;
 902 }
 903
 904 static const struct seq_operations profile_seq_op = {
 905         .start  = probes_seq_start,
 906         .next   = probes_seq_next,
 907         .stop   = probes_seq_stop,
 908         .show   = probes_profile_seq_show
 909 };
 910
 911 static int profile_open(struct inode *inode, struct file *file)
 912 {
 913         return seq_open(file, &profile_seq_op);
 914 }
 915
 916 static const struct file_operations kprobe_profile_ops = {
 917         .owner          = THIS_MODULE,
 918         .open           = profile_open,
 919         .read           = seq_read,
 920         .llseek         = seq_lseek,
 921         .release        = seq_release,
 922 };
 923
 924 /* Kprobe handler */
 925 static nokprobe_inline void
 926 __kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs,
 927                     struct trace_event_file *trace_file)
 928 {
 929         struct kprobe_trace_entry_head *entry;
 930         struct ring_buffer_event *event;
 931         struct ring_buffer *buffer;
 932         int size, dsize, pc;
 933         unsigned long irq_flags;
 934         struct trace_event_call *call = &tk->tp.call;
 935
 936         WARN_ON(call != trace_file->event_call);
 937
 938         if (trace_trigger_soft_disabled(trace_file))
 939                 return;
 940
 941         local_save_flags(irq_flags);
 942         pc = preempt_count();
 943
 944         dsize = __get_data_size(&tk->tp, regs);
 945         size = sizeof(*entry) + tk->tp.size + dsize;
 946
 947         event = trace_event_buffer_lock_reserve(&buffer, trace_file,
 948                                                 call->event.type,
 949                                                 size, irq_flags, pc);
 950         if (!event)
 951                 return;
 952
 953         entry = ring_buffer_event_data(event);
 954         entry->ip = (unsigned long)tk->rp.kp.addr;
 955         store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
 956
 957         event_trigger_unlock_commit_regs(trace_file, buffer, event,
 958                                          entry, irq_flags, pc, regs);
 959 }
 960
 961 static void
 962 kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs)
 963 {
 964         struct event_file_link *link;
 965
 966         list_for_each_entry_rcu(link, &tk->tp.files, list)
 967                 __kprobe_trace_func(tk, regs, link->file);
 968 }
 969 NOKPROBE_SYMBOL(kprobe_trace_func);
 970
 971 /* Kretprobe handler */
 972 static nokprobe_inline void
 973 __kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
 974                        struct pt_regs *regs,
 975                        struct trace_event_file *trace_file)
 976 {
 977         struct kretprobe_trace_entry_head *entry;
 978         struct ring_buffer_event *event;
 979         struct ring_buffer *buffer;
 980         int size, pc, dsize;
 981         unsigned long irq_flags;
 982         struct trace_event_call *call = &tk->tp.call;
 983
 984         WARN_ON(call != trace_file->event_call);
 985
 986         if (trace_trigger_soft_disabled(trace_file))
 987                 return;
 988
 989         local_save_flags(irq_flags);
 990         pc = preempt_count();
 991
 992         dsize = __get_data_size(&tk->tp, regs);
 993         size = sizeof(*entry) + tk->tp.size + dsize;
 994
 995         event = trace_event_buffer_lock_reserve(&buffer, trace_file,
 996                                                 call->event.type,
 997                                                 size, irq_flags, pc);
 998         if (!event)
 999                 return;
1000
1001         entry = ring_buffer_event_data(event);
1002         entry->func = (unsigned long)tk->rp.kp.addr;
1003         entry->ret_ip = (unsigned long)ri->ret_addr;
1004         store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1005
1006         event_trigger_unlock_commit_regs(trace_file, buffer, event,
1007                                          entry, irq_flags, pc, regs);
1008 }
1009
1010 static void
1011 kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1012                      struct pt_regs *regs)
1013 {
1014         struct event_file_link *link;
1015
1016         list_for_each_entry_rcu(link, &tk->tp.files, list)
1017                 __kretprobe_trace_func(tk, ri, regs, link->file);
1018 }
1019 NOKPROBE_SYMBOL(kretprobe_trace_func);
1020
1021 /* Event entry printers */
1022 static enum print_line_t
1023 print_kprobe_event(struct trace_iterator *iter, int flags,
1024                    struct trace_event *event)
1025 {
1026         struct kprobe_trace_entry_head *field;
1027         struct trace_seq *s = &iter->seq;
1028         struct trace_probe *tp;
1029         u8 *data;
1030         int i;
1031
1032         field = (struct kprobe_trace_entry_head *)iter->ent;
1033         tp = container_of(event, struct trace_probe, call.event);
1034
1035         trace_seq_printf(s, "%s: (", trace_event_name(&tp->call));
1036
1037         if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
1038                 goto out;
1039
1040         trace_seq_putc(s, ')');
1041
1042         data = (u8 *)&field[1];
1043         for (i = 0; i < tp->nr_args; i++)
1044                 if (!tp->args[i].type->print(s, tp->args[i].name,
1045                                              data + tp->args[i].offset, field))
1046                         goto out;
1047
1048         trace_seq_putc(s, '\n');
1049  out:
1050         return trace_handle_return(s);
1051 }
1052
1053 static enum print_line_t
1054 print_kretprobe_event(struct trace_iterator *iter, int flags,
1055                       struct trace_event *event)
1056 {
1057         struct kretprobe_trace_entry_head *field;
1058         struct trace_seq *s = &iter->seq;
1059         struct trace_probe *tp;
1060         u8 *data;
1061         int i;
1062
1063         field = (struct kretprobe_trace_entry_head *)iter->ent;
1064         tp = container_of(event, struct trace_probe, call.event);
1065
1066         trace_seq_printf(s, "%s: (", trace_event_name(&tp->call));
1067
1068         if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1069                 goto out;
1070
1071         trace_seq_puts(s, " <- ");
1072
1073         if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1074                 goto out;
1075
1076         trace_seq_putc(s, ')');
1077
1078         data = (u8 *)&field[1];
1079         for (i = 0; i < tp->nr_args; i++)
1080                 if (!tp->args[i].type->print(s, tp->args[i].name,
1081                                              data + tp->args[i].offset, field))
1082                         goto out;
1083
1084         trace_seq_putc(s, '\n');
1085
1086  out:
1087         return trace_handle_return(s);
1088 }
1089
1090
1091 static int kprobe_event_define_fields(struct trace_event_call *event_call)
1092 {
1093         int ret, i;
1094         struct kprobe_trace_entry_head field;
1095         struct trace_kprobe *tk = (struct trace_kprobe *)event_call->data;
1096
1097         DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1098         /* Set argument names as fields */
1099         for (i = 0; i < tk->tp.nr_args; i++) {
1100                 struct probe_arg *parg = &tk->tp.args[i];
1101
1102                 ret = trace_define_field(event_call, parg->type->fmttype,
1103                                          parg->name,
1104                                          sizeof(field) + parg->offset,
1105                                          parg->type->size,
1106                                          parg->type->is_signed,
1107                                          FILTER_OTHER);
1108                 if (ret)
1109                         return ret;
1110         }
1111         return 0;
1112 }
1113
1114 static int kretprobe_event_define_fields(struct trace_event_call *event_call)
1115 {
1116         int ret, i;
1117         struct kretprobe_trace_entry_head field;
1118         struct trace_kprobe *tk = (struct trace_kprobe *)event_call->data;
1119
1120         DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1121         DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1122         /* Set argument names as fields */
1123         for (i = 0; i < tk->tp.nr_args; i++) {
1124                 struct probe_arg *parg = &tk->tp.args[i];
1125
1126                 ret = trace_define_field(event_call, parg->type->fmttype,
1127                                          parg->name,
1128                                          sizeof(field) + parg->offset,
1129                                          parg->type->size,
1130                                          parg->type->is_signed,
1131                                          FILTER_OTHER);
1132                 if (ret)
1133                         return ret;
1134         }
1135         return 0;
1136 }
1137
1138 #ifdef CONFIG_PERF_EVENTS
1139
1140 /* Kprobe profile handler */
1141 static void
1142 kprobe_perf_func(struct trace_kprobe *tk, struct pt_regs *regs)
1143 {
1144         struct trace_event_call *call = &tk->tp.call;
1145         struct bpf_prog *prog = call->prog;
1146         struct kprobe_trace_entry_head *entry;
1147         struct hlist_head *head;
1148         int size, __size, dsize;
1149         int rctx;
1150
1151         if (prog && !trace_call_bpf(prog, regs))
1152                 return;
1153
1154         head = this_cpu_ptr(call->perf_events);
1155         if (hlist_empty(head))
1156                 return;
1157
1158         dsize = __get_data_size(&tk->tp, regs);
1159         __size = sizeof(*entry) + tk->tp.size + dsize;
1160         size = ALIGN(__size + sizeof(u32), sizeof(u64));
1161         size -= sizeof(u32);
1162
1163         entry = perf_trace_buf_alloc(size, NULL, &rctx);
1164         if (!entry)
1165                 return;
1166
1167         entry->ip = (unsigned long)tk->rp.kp.addr;
1168         memset(&entry[1], 0, dsize);
1169         store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1170         perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1171                               head, NULL);
1172 }
1173 NOKPROBE_SYMBOL(kprobe_perf_func);
1174
1175 /* Kretprobe profile handler */
1176 static void
1177 kretprobe_perf_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1178                     struct pt_regs *regs)
1179 {
1180         struct trace_event_call *call = &tk->tp.call;
1181         struct bpf_prog *prog = call->prog;
1182         struct kretprobe_trace_entry_head *entry;
1183         struct hlist_head *head;
1184         int size, __size, dsize;
1185         int rctx;
1186
1187         if (prog && !trace_call_bpf(prog, regs))
1188                 return;
1189
1190         head = this_cpu_ptr(call->perf_events);
1191         if (hlist_empty(head))
1192                 return;
1193
1194         dsize = __get_data_size(&tk->tp, regs);
1195         __size = sizeof(*entry) + tk->tp.size + dsize;
1196         size = ALIGN(__size + sizeof(u32), sizeof(u64));
1197         size -= sizeof(u32);
1198
1199         entry = perf_trace_buf_alloc(size, NULL, &rctx);
1200         if (!entry)
1201                 return;
1202
1203         entry->func = (unsigned long)tk->rp.kp.addr;
1204         entry->ret_ip = (unsigned long)ri->ret_addr;
1205         store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1206         perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1207                               head, NULL);
1208 }
1209 NOKPROBE_SYMBOL(kretprobe_perf_func);
1210 #endif  /* CONFIG_PERF_EVENTS */
1211
1212 /*
1213  * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex.
1214  *
1215  * kprobe_trace_self_tests_init() does enable_trace_probe/disable_trace_probe
1216  * lockless, but we can't race with this __init function.
1217  */
1218 static int kprobe_register(struct trace_event_call *event,
1219                            enum trace_reg type, void *data)
1220 {
1221         struct trace_kprobe *tk = (struct trace_kprobe *)event->data;
1222         struct trace_event_file *file = data;
1223
1224         switch (type) {
1225         case TRACE_REG_REGISTER:
1226                 return enable_trace_kprobe(tk, file);
1227         case TRACE_REG_UNREGISTER:
1228                 return disable_trace_kprobe(tk, file);
1229
1230 #ifdef CONFIG_PERF_EVENTS
1231         case TRACE_REG_PERF_REGISTER:
1232                 return enable_trace_kprobe(tk, NULL);
1233         case TRACE_REG_PERF_UNREGISTER:
1234                 return disable_trace_kprobe(tk, NULL);
1235         case TRACE_REG_PERF_OPEN:
1236         case TRACE_REG_PERF_CLOSE:
1237         case TRACE_REG_PERF_ADD:
1238         case TRACE_REG_PERF_DEL:
1239                 return 0;
1240 #endif
1241         }
1242         return 0;
1243 }
1244
1245 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1246 {
1247         struct trace_kprobe *tk = container_of(kp, struct trace_kprobe, rp.kp);
1248
1249         raw_cpu_inc(*tk->nhit);
1250
1251         if (tk->tp.flags & TP_FLAG_TRACE)
1252                 kprobe_trace_func(tk, regs);
1253 #ifdef CONFIG_PERF_EVENTS
1254         if (tk->tp.flags & TP_FLAG_PROFILE)
1255                 kprobe_perf_func(tk, regs);
1256 #endif
1257         return 0;       /* We don't tweek kernel, so just return 0 */
1258 }
1259 NOKPROBE_SYMBOL(kprobe_dispatcher);
1260
1261 static int
1262 kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1263 {
1264         struct trace_kprobe *tk = container_of(ri->rp, struct trace_kprobe, rp);
1265
1266         raw_cpu_inc(*tk->nhit);
1267
1268         if (tk->tp.flags & TP_FLAG_TRACE)
1269                 kretprobe_trace_func(tk, ri, regs);
1270 #ifdef CONFIG_PERF_EVENTS
1271         if (tk->tp.flags & TP_FLAG_PROFILE)
1272                 kretprobe_perf_func(tk, ri, regs);
1273 #endif
1274         return 0;       /* We don't tweek kernel, so just return 0 */
1275 }
1276 NOKPROBE_SYMBOL(kretprobe_dispatcher);
1277
1278 static struct trace_event_functions kretprobe_funcs = {
1279         .trace          = print_kretprobe_event
1280 };
1281
1282 static struct trace_event_functions kprobe_funcs = {
1283         .trace          = print_kprobe_event
1284 };
1285
1286 static int register_kprobe_event(struct trace_kprobe *tk)
1287 {
1288         struct trace_event_call *call = &tk->tp.call;
1289         int ret;
1290
1291         /* Initialize trace_event_call */
1292         INIT_LIST_HEAD(&call->class->fields);
1293         if (trace_kprobe_is_return(tk)) {
1294                 call->event.funcs = &kretprobe_funcs;
1295                 call->class->define_fields = kretprobe_event_define_fields;
1296         } else {
1297                 call->event.funcs = &kprobe_funcs;
1298                 call->class->define_fields = kprobe_event_define_fields;
1299         }
1300         if (set_print_fmt(&tk->tp, trace_kprobe_is_return(tk)) < 0)
1301                 return -ENOMEM;
1302         ret = register_trace_event(&call->event);
1303         if (!ret) {
1304                 kfree(call->print_fmt);
1305                 return -ENODEV;
1306         }
1307         call->flags = TRACE_EVENT_FL_KPROBE;
1308         call->class->reg = kprobe_register;
1309         call->data = tk;
1310         ret = trace_add_event_call(call);
1311         if (ret) {
1312                 pr_info("Failed to register kprobe event: %s\n",
1313                         trace_event_name(call));
1314                 kfree(call->print_fmt);
1315                 unregister_trace_event(&call->event);
1316         }
1317         return ret;
1318 }
1319
1320 static int unregister_kprobe_event(struct trace_kprobe *tk)
1321 {
1322         int ret;
1323
1324         /* tp->event is unregistered in trace_remove_event_call() */
1325         ret = trace_remove_event_call(&tk->tp.call);
1326         if (!ret)
1327                 kfree(tk->tp.call.print_fmt);
1328         return ret;
1329 }
1330
1331 /* Make a tracefs interface for controlling probe points */
1332 static __init int init_kprobe_trace(void)
1333 {
1334         struct dentry *d_tracer;
1335         struct dentry *entry;
1336
1337         if (register_module_notifier(&trace_kprobe_module_nb))
1338                 return -EINVAL;
1339
1340         d_tracer = tracing_init_dentry();
1341         if (IS_ERR(d_tracer))
1342                 return 0;
1343
1344         entry = tracefs_create_file("kprobe_events", 0644, d_tracer,
1345                                     NULL, &kprobe_events_ops);
1346
1347         /* Event list interface */
1348         if (!entry)
1349                 pr_warn("Could not create tracefs 'kprobe_events' entry\n");
1350
1351         /* Profile interface */
1352         entry = tracefs_create_file("kprobe_profile", 0444, d_tracer,
1353                                     NULL, &kprobe_profile_ops);
1354
1355         if (!entry)
1356                 pr_warn("Could not create tracefs 'kprobe_profile' entry\n");
1357         return 0;
1358 }
1359 fs_initcall(init_kprobe_trace);
1360
1361
1362 #ifdef CONFIG_FTRACE_STARTUP_TEST
1363 /*
1364  * The "__used" keeps gcc from removing the function symbol
1365  * from the kallsyms table. 'noinline' makes sure that there
1366  * isn't an inlined version used by the test method below
1367  */
1368 static __used __init noinline int
1369 kprobe_trace_selftest_target(int a1, int a2, int a3, int a4, int a5, int a6)
1370 {
1371         return a1 + a2 + a3 + a4 + a5 + a6;
1372 }
1373
1374 static __init struct trace_event_file *
1375 find_trace_probe_file(struct trace_kprobe *tk, struct trace_array *tr)
1376 {
1377         struct trace_event_file *file;
1378
1379         list_for_each_entry(file, &tr->events, list)
1380                 if (file->event_call == &tk->tp.call)
1381                         return file;
1382
1383         return NULL;
1384 }
1385
1386 /*
1387  * Nobody but us can call enable_trace_kprobe/disable_trace_kprobe at this
1388  * stage, we can do this lockless.
1389  */
1390 static __init int kprobe_trace_self_tests_init(void)
1391 {
1392         int ret, warn = 0;
1393         int (*target)(int, int, int, int, int, int);
1394         struct trace_kprobe *tk;
1395         struct trace_event_file *file;
1396
1397         if (tracing_is_disabled())
1398                 return -ENODEV;
1399
1400         target = kprobe_trace_selftest_target;
1401
1402         pr_info("Testing kprobe tracing: ");
1403
1404         ret = traceprobe_command("p:testprobe kprobe_trace_selftest_target "
1405                                   "$stack $stack0 +0($stack)",
1406                                   create_trace_kprobe);
1407         if (WARN_ON_ONCE(ret)) {
1408                 pr_warn("error on probing function entry.\n");
1409                 warn++;
1410         } else {
1411                 /* Enable trace point */
1412                 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1413                 if (WARN_ON_ONCE(tk == NULL)) {
1414                         pr_warn("error on getting new probe.\n");
1415                         warn++;
1416                 } else {
1417                         file = find_trace_probe_file(tk, top_trace_array());
1418                         if (WARN_ON_ONCE(file == NULL)) {
1419                                 pr_warn("error on getting probe file.\n");
1420                                 warn++;
1421                         } else
1422                                 enable_trace_kprobe(tk, file);
1423                 }
1424         }
1425
1426         ret = traceprobe_command("r:testprobe2 kprobe_trace_selftest_target "
1427                                   "$retval", create_trace_kprobe);
1428         if (WARN_ON_ONCE(ret)) {
1429                 pr_warn("error on probing function return.\n");
1430                 warn++;
1431         } else {
1432                 /* Enable trace point */
1433                 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1434                 if (WARN_ON_ONCE(tk == NULL)) {
1435                         pr_warn("error on getting 2nd new probe.\n");
1436                         warn++;
1437                 } else {
1438                         file = find_trace_probe_file(tk, top_trace_array());
1439                         if (WARN_ON_ONCE(file == NULL)) {
1440                                 pr_warn("error on getting probe file.\n");
1441                                 warn++;
1442                         } else
1443                                 enable_trace_kprobe(tk, file);
1444                 }
1445         }
1446
1447         if (warn)
1448                 goto end;
1449
1450         ret = target(1, 2, 3, 4, 5, 6);
1451
1452         /*
1453          * Not expecting an error here, the check is only to prevent the
1454          * optimizer from removing the call to target() as otherwise there
1455          * are no side-effects and the call is never performed.
1456          */
1457         if (ret != 21)
1458                 warn++;
1459
1460         /* Disable trace points before removing it */
1461         tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1462         if (WARN_ON_ONCE(tk == NULL)) {
1463                 pr_warn("error on getting test probe.\n");
1464                 warn++;
1465         } else {
1466                 if (trace_kprobe_nhit(tk) != 1) {
1467                         pr_warn("incorrect number of testprobe hits\n");
1468                         warn++;
1469                 }
1470
1471                 file = find_trace_probe_file(tk, top_trace_array());
1472                 if (WARN_ON_ONCE(file == NULL)) {
1473                         pr_warn("error on getting probe file.\n");
1474                         warn++;
1475                 } else
1476                         disable_trace_kprobe(tk, file);
1477         }
1478
1479         tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1480         if (WARN_ON_ONCE(tk == NULL)) {
1481                 pr_warn("error on getting 2nd test probe.\n");
1482                 warn++;
1483         } else {
1484                 if (trace_kprobe_nhit(tk) != 1) {
1485                         pr_warn("incorrect number of testprobe2 hits\n");
1486                         warn++;
1487                 }
1488
1489                 file = find_trace_probe_file(tk, top_trace_array());
1490                 if (WARN_ON_ONCE(file == NULL)) {
1491                         pr_warn("error on getting probe file.\n");
1492                         warn++;
1493                 } else
1494                         disable_trace_kprobe(tk, file);
1495         }
1496
1497         ret = traceprobe_command("-:testprobe", create_trace_kprobe);
1498         if (WARN_ON_ONCE(ret)) {
1499                 pr_warn("error on deleting a probe.\n");
1500                 warn++;
1501         }
1502
1503         ret = traceprobe_command("-:testprobe2", create_trace_kprobe);
1504         if (WARN_ON_ONCE(ret)) {
1505                 pr_warn("error on deleting a probe.\n");
1506                 warn++;
1507         }
1508
1509 end:
1510         release_all_trace_kprobes();
1511         if (warn)
1512                 pr_cont("NG: Some tests are failed. Please check them.\n");
1513         else
1514                 pr_cont("OK\n");
1515         return 0;
1516 }
1517
1518 late_initcall(kprobe_trace_self_tests_init);
1519
1520 #endif