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 __percpu *nhit;
  34         const char              *symbol;        /* symbol name */
  35         struct trace_probe      tp;
  36 };
  37
  38 #define SIZEOF_TRACE_KPROBE(n)                          \
  39         (offsetof(struct trace_kprobe, tp.args) +       \
  40         (sizeof(struct probe_arg) * (n)))
  41
  42
  43 static nokprobe_inline bool trace_kprobe_is_return(struct trace_kprobe *tk)
  44 {
  45         return tk->rp.handler != NULL;
  46 }
  47
  48 static nokprobe_inline const char *trace_kprobe_symbol(struct trace_kprobe *tk)
  49 {
  50         return tk->symbol ? tk->symbol : "unknown";
  51 }
  52
  53 static nokprobe_inline unsigned long trace_kprobe_offset(struct trace_kprobe *tk)
  54 {
  55         return tk->rp.kp.offset;
  56 }
  57
  58 static nokprobe_inline bool trace_kprobe_has_gone(struct trace_kprobe *tk)
  59 {
  60         return !!(kprobe_gone(&tk->rp.kp));
  61 }
  62
  63 static nokprobe_inline bool trace_kprobe_within_module(struct trace_kprobe *tk,
  64                                                  struct module *mod)
  65 {
  66         int len = strlen(mod->name);
  67         const char *name = trace_kprobe_symbol(tk);
  68         return strncmp(mod->name, name, len) == 0 && name[len] == ':';
  69 }
  70
  71 static nokprobe_inline bool trace_kprobe_is_on_module(struct trace_kprobe *tk)
  72 {
  73         return !!strchr(trace_kprobe_symbol(tk), ':');
  74 }
  75
  76 static int register_kprobe_event(struct trace_kprobe *tk);
  77 static int unregister_kprobe_event(struct trace_kprobe *tk);
  78
  79 static DEFINE_MUTEX(probe_lock);
  80 static LIST_HEAD(probe_list);
  81
  82 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
  83 static int kretprobe_dispatcher(struct kretprobe_instance *ri,
  84                                 struct pt_regs *regs);
  85
  86 /* Memory fetching by symbol */
  87 struct symbol_cache {
  88         char            *symbol;
  89         long            offset;
  90         unsigned long   addr;
  91 };
  92
  93 unsigned long update_symbol_cache(struct symbol_cache *sc)
  94 {
  95         sc->addr = (unsigned long)kallsyms_lookup_name(sc->symbol);
  96
  97         if (sc->addr)
  98                 sc->addr += sc->offset;
  99
 100         return sc->addr;
 101 }
 102
 103 void free_symbol_cache(struct symbol_cache *sc)
 104 {
 105         kfree(sc->symbol);
 106         kfree(sc);
 107 }
 108
 109 struct symbol_cache *alloc_symbol_cache(const char *sym, long offset)
 110 {
 111         struct symbol_cache *sc;
 112
 113         if (!sym || strlen(sym) == 0)
 114                 return NULL;
 115
 116         sc = kzalloc(sizeof(struct symbol_cache), GFP_KERNEL);
 117         if (!sc)
 118                 return NULL;
 119
 120         sc->symbol = kstrdup(sym, GFP_KERNEL);
 121         if (!sc->symbol) {
 122                 kfree(sc);
 123                 return NULL;
 124         }
 125         sc->offset = offset;
 126         update_symbol_cache(sc);
 127
 128         return sc;
 129 }
 130
 131 /*
 132  * Kprobes-specific fetch functions
 133  */
 134 #define DEFINE_FETCH_stack(type)                                        \
 135 static void FETCH_FUNC_NAME(stack, type)(struct pt_regs *regs,          \
 136                                           void *offset, void *dest)     \
 137 {                                                                       \
 138         *(type *)dest = (type)regs_get_kernel_stack_nth(regs,           \
 139                                 (unsigned int)((unsigned long)offset)); \
 140 }                                                                       \
 141 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(stack, type));
 142
 143 DEFINE_BASIC_FETCH_FUNCS(stack)
 144 /* No string on the stack entry */
 145 #define fetch_stack_string      NULL
 146 #define fetch_stack_string_size NULL
 147
 148 #define DEFINE_FETCH_memory(type)                                       \
 149 static void FETCH_FUNC_NAME(memory, type)(struct pt_regs *regs,         \
 150                                           void *addr, void *dest)       \
 151 {                                                                       \
 152         type retval;                                                    \
 153         if (probe_kernel_address(addr, retval))                         \
 154                 *(type *)dest = 0;                                      \
 155         else                                                            \
 156                 *(type *)dest = retval;                                 \
 157 }                                                                       \
 158 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(memory, type));
 159
 160 DEFINE_BASIC_FETCH_FUNCS(memory)
 161 /*
 162  * Fetch a null-terminated string. Caller MUST set *(u32 *)dest with max
 163  * length and relative data location.
 164  */
 165 static void FETCH_FUNC_NAME(memory, string)(struct pt_regs *regs,
 166                                             void *addr, void *dest)
 167 {
 168         int maxlen = get_rloc_len(*(u32 *)dest);
 169         u8 *dst = get_rloc_data(dest);
 170         long ret;
 171
 172         if (!maxlen)
 173                 return;
 174
 175         /*
 176          * Try to get string again, since the string can be changed while
 177          * probing.
 178          */
 179         ret = strncpy_from_unsafe(dst, addr, maxlen);
 180
 181         if (ret < 0) {  /* Failed to fetch string */
 182                 dst[0] = '\0';
 183                 *(u32 *)dest = make_data_rloc(0, get_rloc_offs(*(u32 *)dest));
 184         } else {
 185                 *(u32 *)dest = make_data_rloc(ret, get_rloc_offs(*(u32 *)dest));
 186         }
 187 }
 188 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(memory, string));
 189
 190 /* Return the length of string -- including null terminal byte */
 191 static void FETCH_FUNC_NAME(memory, string_size)(struct pt_regs *regs,
 192                                                  void *addr, void *dest)
 193 {
 194         mm_segment_t old_fs;
 195         int ret, len = 0;
 196         u8 c;
 197
 198         old_fs = get_fs();
 199         set_fs(KERNEL_DS);
 200         pagefault_disable();
 201
 202         do {
 203                 ret = __copy_from_user_inatomic(&c, (u8 *)addr + len, 1);
 204                 len++;
 205         } while (c && ret == 0 && len < MAX_STRING_SIZE);
 206
 207         pagefault_enable();
 208         set_fs(old_fs);
 209
 210         if (ret < 0)    /* Failed to check the length */
 211                 *(u32 *)dest = 0;
 212         else
 213                 *(u32 *)dest = len;
 214 }
 215 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(memory, string_size));
 216
 217 #define DEFINE_FETCH_symbol(type)                                       \
 218 void FETCH_FUNC_NAME(symbol, type)(struct pt_regs *regs, void *data, void *dest)\
 219 {                                                                       \
 220         struct symbol_cache *sc = data;                                 \
 221         if (sc->addr)                                                   \
 222                 fetch_memory_##type(regs, (void *)sc->addr, dest);      \
 223         else                                                            \
 224                 *(type *)dest = 0;                                      \
 225 }                                                                       \
 226 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(symbol, type));
 227
 228 DEFINE_BASIC_FETCH_FUNCS(symbol)
 229 DEFINE_FETCH_symbol(string)
 230 DEFINE_FETCH_symbol(string_size)
 231
 232 /* kprobes don't support file_offset fetch methods */
 233 #define fetch_file_offset_u8            NULL
 234 #define fetch_file_offset_u16           NULL
 235 #define fetch_file_offset_u32           NULL
 236 #define fetch_file_offset_u64           NULL
 237 #define fetch_file_offset_string        NULL
 238 #define fetch_file_offset_string_size   NULL
 239
 240 /* Fetch type information table */
 241 static const struct fetch_type kprobes_fetch_type_table[] = {
 242         /* Special types */
 243         [FETCH_TYPE_STRING] = __ASSIGN_FETCH_TYPE("string", string, string,
 244                                         sizeof(u32), 1, "__data_loc char[]"),
 245         [FETCH_TYPE_STRSIZE] = __ASSIGN_FETCH_TYPE("string_size", u32,
 246                                         string_size, sizeof(u32), 0, "u32"),
 247         /* Basic types */
 248         ASSIGN_FETCH_TYPE(u8,  u8,  0),
 249         ASSIGN_FETCH_TYPE(u16, u16, 0),
 250         ASSIGN_FETCH_TYPE(u32, u32, 0),
 251         ASSIGN_FETCH_TYPE(u64, u64, 0),
 252         ASSIGN_FETCH_TYPE(s8,  u8,  1),
 253         ASSIGN_FETCH_TYPE(s16, u16, 1),
 254         ASSIGN_FETCH_TYPE(s32, u32, 1),
 255         ASSIGN_FETCH_TYPE(s64, u64, 1),
 256         ASSIGN_FETCH_TYPE_ALIAS(x8,  u8,  u8,  0),
 257         ASSIGN_FETCH_TYPE_ALIAS(x16, u16, u16, 0),
 258         ASSIGN_FETCH_TYPE_ALIAS(x32, u32, u32, 0),
 259         ASSIGN_FETCH_TYPE_ALIAS(x64, u64, u64, 0),
 260
 261         ASSIGN_FETCH_TYPE_END
 262 };
 263
 264 /*
 265  * Allocate new trace_probe and initialize it (including kprobes).
 266  */
 267 static struct trace_kprobe *alloc_trace_kprobe(const char *group,
 268                                              const char *event,
 269                                              void *addr,
 270                                              const char *symbol,
 271                                              unsigned long offs,
 272                                              int nargs, bool is_return)
 273 {
 274         struct trace_kprobe *tk;
 275         int ret = -ENOMEM;
 276
 277         tk = kzalloc(SIZEOF_TRACE_KPROBE(nargs), GFP_KERNEL);
 278         if (!tk)
 279                 return ERR_PTR(ret);
 280
 281         tk->nhit = alloc_percpu(unsigned long);
 282         if (!tk->nhit)
 283                 goto error;
 284
 285         if (symbol) {
 286                 tk->symbol = kstrdup(symbol, GFP_KERNEL);
 287                 if (!tk->symbol)
 288                         goto error;
 289                 tk->rp.kp.symbol_name = tk->symbol;
 290                 tk->rp.kp.offset = offs;
 291         } else
 292                 tk->rp.kp.addr = addr;
 293
 294         if (is_return)
 295                 tk->rp.handler = kretprobe_dispatcher;
 296         else
 297                 tk->rp.kp.pre_handler = kprobe_dispatcher;
 298
 299         if (!event || !is_good_name(event)) {
 300                 ret = -EINVAL;
 301                 goto error;
 302         }
 303
 304         tk->tp.call.class = &tk->tp.class;
 305         tk->tp.call.name = kstrdup(event, GFP_KERNEL);
 306         if (!tk->tp.call.name)
 307                 goto error;
 308
 309         if (!group || !is_good_name(group)) {
 310                 ret = -EINVAL;
 311                 goto error;
 312         }
 313
 314         tk->tp.class.system = kstrdup(group, GFP_KERNEL);
 315         if (!tk->tp.class.system)
 316                 goto error;
 317
 318         INIT_LIST_HEAD(&tk->list);
 319         INIT_LIST_HEAD(&tk->tp.files);
 320         return tk;
 321 error:
 322         kfree(tk->tp.call.name);
 323         kfree(tk->symbol);
 324         free_percpu(tk->nhit);
 325         kfree(tk);
 326         return ERR_PTR(ret);
 327 }
 328
 329 static void free_trace_kprobe(struct trace_kprobe *tk)
 330 {
 331         int i;
 332
 333         for (i = 0; i < tk->tp.nr_args; i++)
 334                 traceprobe_free_probe_arg(&tk->tp.args[i]);
 335
 336         kfree(tk->tp.call.class->system);
 337         kfree(tk->tp.call.name);
 338         kfree(tk->symbol);
 339         free_percpu(tk->nhit);
 340         kfree(tk);
 341 }
 342
 343 static struct trace_kprobe *find_trace_kprobe(const char *event,
 344                                               const char *group)
 345 {
 346         struct trace_kprobe *tk;
 347
 348         list_for_each_entry(tk, &probe_list, list)
 349                 if (strcmp(trace_event_name(&tk->tp.call), event) == 0 &&
 350                     strcmp(tk->tp.call.class->system, group) == 0)
 351                         return tk;
 352         return NULL;
 353 }
 354
 355 /*
 356  * Enable trace_probe
 357  * if the file is NULL, enable "perf" handler, or enable "trace" handler.
 358  */
 359 static int
 360 enable_trace_kprobe(struct trace_kprobe *tk, struct trace_event_file *file)
 361 {
 362         struct event_file_link *link = NULL;
 363         int ret = 0;
 364
 365         if (file) {
 366                 link = kmalloc(sizeof(*link), GFP_KERNEL);
 367                 if (!link) {
 368                         ret = -ENOMEM;
 369                         goto out;
 370                 }
 371
 372                 link->file = file;
 373                 list_add_tail_rcu(&link->list, &tk->tp.files);
 374
 375                 tk->tp.flags |= TP_FLAG_TRACE;
 376         } else
 377                 tk->tp.flags |= TP_FLAG_PROFILE;
 378
 379         if (trace_probe_is_registered(&tk->tp) && !trace_kprobe_has_gone(tk)) {
 380                 if (trace_kprobe_is_return(tk))
 381                         ret = enable_kretprobe(&tk->rp);
 382                 else
 383                         ret = enable_kprobe(&tk->rp.kp);
 384         }
 385
 386         if (ret) {
 387                 if (file) {
 388                         /* Notice the if is true on not WARN() */
 389                         if (!WARN_ON_ONCE(!link))
 390                                 list_del_rcu(&link->list);
 391                         kfree(link);
 392                         tk->tp.flags &= ~TP_FLAG_TRACE;
 393                 } else {
 394                         tk->tp.flags &= ~TP_FLAG_PROFILE;
 395                 }
 396         }
 397  out:
 398         return ret;
 399 }
 400
 401 /*
 402  * Disable trace_probe
 403  * if the file is NULL, disable "perf" handler, or disable "trace" handler.
 404  */
 405 static int
 406 disable_trace_kprobe(struct trace_kprobe *tk, struct trace_event_file *file)
 407 {
 408         struct event_file_link *link = NULL;
 409         int wait = 0;
 410         int ret = 0;
 411
 412         if (file) {
 413                 link = find_event_file_link(&tk->tp, file);
 414                 if (!link) {
 415                         ret = -EINVAL;
 416                         goto out;
 417                 }
 418
 419                 list_del_rcu(&link->list);
 420                 wait = 1;
 421                 if (!list_empty(&tk->tp.files))
 422                         goto out;
 423
 424                 tk->tp.flags &= ~TP_FLAG_TRACE;
 425         } else
 426                 tk->tp.flags &= ~TP_FLAG_PROFILE;
 427
 428         if (!trace_probe_is_enabled(&tk->tp) && trace_probe_is_registered(&tk->tp)) {
 429                 if (trace_kprobe_is_return(tk))
 430                         disable_kretprobe(&tk->rp);
 431                 else
 432                         disable_kprobe(&tk->rp.kp);
 433                 wait = 1;
 434         }
 435  out:
 436         if (wait) {
 437                 /*
 438                  * Synchronize with kprobe_trace_func/kretprobe_trace_func
 439                  * to ensure disabled (all running handlers are finished).
 440                  * This is not only for kfree(), but also the caller,
 441                  * trace_remove_event_call() supposes it for releasing
 442                  * event_call related objects, which will be accessed in
 443                  * the kprobe_trace_func/kretprobe_trace_func.
 444                  */
 445                 synchronize_sched();
 446                 kfree(link);    /* Ignored if link == NULL */
 447         }
 448
 449         return ret;
 450 }
 451
 452 /* Internal register function - just handle k*probes and flags */
 453 static int __register_trace_kprobe(struct trace_kprobe *tk)
 454 {
 455         int i, ret;
 456
 457         if (trace_probe_is_registered(&tk->tp))
 458                 return -EINVAL;
 459
 460         for (i = 0; i < tk->tp.nr_args; i++)
 461                 traceprobe_update_arg(&tk->tp.args[i]);
 462
 463         /* Set/clear disabled flag according to tp->flag */
 464         if (trace_probe_is_enabled(&tk->tp))
 465                 tk->rp.kp.flags &= ~KPROBE_FLAG_DISABLED;
 466         else
 467                 tk->rp.kp.flags |= KPROBE_FLAG_DISABLED;
 468
 469         if (trace_kprobe_is_return(tk))
 470                 ret = register_kretprobe(&tk->rp);
 471         else
 472                 ret = register_kprobe(&tk->rp.kp);
 473
 474         if (ret == 0)
 475                 tk->tp.flags |= TP_FLAG_REGISTERED;
 476         else {
 477                 pr_warn("Could not insert probe at %s+%lu: %d\n",
 478                         trace_kprobe_symbol(tk), trace_kprobe_offset(tk), ret);
 479                 if (ret == -ENOENT && trace_kprobe_is_on_module(tk)) {
 480                         pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
 481                         ret = 0;
 482                 } else if (ret == -EILSEQ) {
 483                         pr_warn("Probing address(0x%p) is not an instruction boundary.\n",
 484                                 tk->rp.kp.addr);
 485                         ret = -EINVAL;
 486                 }
 487         }
 488
 489         return ret;
 490 }
 491
 492 /* Internal unregister function - just handle k*probes and flags */
 493 static void __unregister_trace_kprobe(struct trace_kprobe *tk)
 494 {
 495         if (trace_probe_is_registered(&tk->tp)) {
 496                 if (trace_kprobe_is_return(tk))
 497                         unregister_kretprobe(&tk->rp);
 498                 else
 499                         unregister_kprobe(&tk->rp.kp);
 500                 tk->tp.flags &= ~TP_FLAG_REGISTERED;
 501                 /* Cleanup kprobe for reuse */
 502                 if (tk->rp.kp.symbol_name)
 503                         tk->rp.kp.addr = NULL;
 504         }
 505 }
 506
 507 /* Unregister a trace_probe and probe_event: call with locking probe_lock */
 508 static int unregister_trace_kprobe(struct trace_kprobe *tk)
 509 {
 510         /* Enabled event can not be unregistered */
 511         if (trace_probe_is_enabled(&tk->tp))
 512                 return -EBUSY;
 513
 514         /* Will fail if probe is being used by ftrace or perf */
 515         if (unregister_kprobe_event(tk))
 516                 return -EBUSY;
 517
 518         __unregister_trace_kprobe(tk);
 519         list_del(&tk->list);
 520
 521         return 0;
 522 }
 523
 524 /* Register a trace_probe and probe_event */
 525 static int register_trace_kprobe(struct trace_kprobe *tk)
 526 {
 527         struct trace_kprobe *old_tk;
 528         int ret;
 529
 530         mutex_lock(&probe_lock);
 531
 532         /* Delete old (same name) event if exist */
 533         old_tk = find_trace_kprobe(trace_event_name(&tk->tp.call),
 534                         tk->tp.call.class->system);
 535         if (old_tk) {
 536                 ret = unregister_trace_kprobe(old_tk);
 537                 if (ret < 0)
 538                         goto end;
 539                 free_trace_kprobe(old_tk);
 540         }
 541
 542         /* Register new event */
 543         ret = register_kprobe_event(tk);
 544         if (ret) {
 545                 pr_warn("Failed to register probe event(%d)\n", ret);
 546                 goto end;
 547         }
 548
 549         /* Register k*probe */
 550         ret = __register_trace_kprobe(tk);
 551         if (ret < 0)
 552                 unregister_kprobe_event(tk);
 553         else
 554                 list_add_tail(&tk->list, &probe_list);
 555
 556 end:
 557         mutex_unlock(&probe_lock);
 558         return ret;
 559 }
 560
 561 /* Module notifier call back, checking event on the module */
 562 static int trace_kprobe_module_callback(struct notifier_block *nb,
 563                                        unsigned long val, void *data)
 564 {
 565         struct module *mod = data;
 566         struct trace_kprobe *tk;
 567         int ret;
 568
 569         if (val != MODULE_STATE_COMING)
 570                 return NOTIFY_DONE;
 571
 572         /* Update probes on coming module */
 573         mutex_lock(&probe_lock);
 574         list_for_each_entry(tk, &probe_list, list) {
 575                 if (trace_kprobe_within_module(tk, mod)) {
 576                         /* Don't need to check busy - this should have gone. */
 577                         __unregister_trace_kprobe(tk);
 578                         ret = __register_trace_kprobe(tk);
 579                         if (ret)
 580                                 pr_warn("Failed to re-register probe %s on %s: %d\n",
 581                                         trace_event_name(&tk->tp.call),
 582                                         mod->name, ret);
 583                 }
 584         }
 585         mutex_unlock(&probe_lock);
 586
 587         return NOTIFY_DONE;
 588 }
 589
 590 static struct notifier_block trace_kprobe_module_nb = {
 591         .notifier_call = trace_kprobe_module_callback,
 592         .priority = 1   /* Invoked after kprobe module callback */
 593 };
 594
 595 static int create_trace_kprobe(int argc, char **argv)
 596 {
 597         /*
 598          * Argument syntax:
 599          *  - Add kprobe: p[:[GRP/]EVENT] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
 600          *  - Add kretprobe: r[:[GRP/]EVENT] [MOD:]KSYM[+0] [FETCHARGS]
 601          * Fetch args:
 602          *  $retval     : fetch return value
 603          *  $stack      : fetch stack address
 604          *  $stackN     : fetch Nth of stack (N:0-)
 605          *  $comm       : fetch current task comm
 606          *  @ADDR       : fetch memory at ADDR (ADDR should be in kernel)
 607          *  @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
 608          *  %REG        : fetch register REG
 609          * Dereferencing memory fetch:
 610          *  +|-offs(ARG) : fetch memory at ARG +|- offs address.
 611          * Alias name of args:
 612          *  NAME=FETCHARG : set NAME as alias of FETCHARG.
 613          * Type of args:
 614          *  FETCHARG:TYPE : use TYPE instead of unsigned long.
 615          */
 616         struct trace_kprobe *tk;
 617         int i, ret = 0;
 618         bool is_return = false, is_delete = false;
 619         char *symbol = NULL, *event = NULL, *group = NULL;
 620         char *arg;
 621         long offset = 0;
 622         void *addr = NULL;
 623         char buf[MAX_EVENT_NAME_LEN];
 624
 625         /* argc must be >= 1 */
 626         if (argv[0][0] == 'p')
 627                 is_return = false;
 628         else if (argv[0][0] == 'r')
 629                 is_return = true;
 630         else if (argv[0][0] == '-')
 631                 is_delete = true;
 632         else {
 633                 pr_info("Probe definition must be started with 'p', 'r' or"
 634                         " '-'.\n");
 635                 return -EINVAL;
 636         }
 637
 638         if (argv[0][1] == ':') {
 639                 event = &argv[0][2];
 640                 if (strchr(event, '/')) {
 641                         group = event;
 642                         event = strchr(group, '/') + 1;
 643                         event[-1] = '\0';
 644                         if (strlen(group) == 0) {
 645                                 pr_info("Group name is not specified\n");
 646                                 return -EINVAL;
 647                         }
 648                 }
 649                 if (strlen(event) == 0) {
 650                         pr_info("Event name is not specified\n");
 651                         return -EINVAL;
 652                 }
 653         }
 654         if (!group)
 655                 group = KPROBE_EVENT_SYSTEM;
 656
 657         if (is_delete) {
 658                 if (!event) {
 659                         pr_info("Delete command needs an event name.\n");
 660                         return -EINVAL;
 661                 }
 662                 mutex_lock(&probe_lock);
 663                 tk = find_trace_kprobe(event, group);
 664                 if (!tk) {
 665                         mutex_unlock(&probe_lock);
 666                         pr_info("Event %s/%s doesn't exist.\n", group, event);
 667                         return -ENOENT;
 668                 }
 669                 /* delete an event */
 670                 ret = unregister_trace_kprobe(tk);
 671                 if (ret == 0)
 672                         free_trace_kprobe(tk);
 673                 mutex_unlock(&probe_lock);
 674                 return ret;
 675         }
 676
 677         if (argc < 2) {
 678                 pr_info("Probe point is not specified.\n");
 679                 return -EINVAL;
 680         }
 681
 682         /* try to parse an address. if that fails, try to read the
 683          * input as a symbol. */
 684         if (kstrtoul(argv[1], 0, (unsigned long *)&addr)) {
 685                 /* a symbol specified */
 686                 symbol = argv[1];
 687                 /* TODO: support .init module functions */
 688                 ret = traceprobe_split_symbol_offset(symbol, &offset);
 689                 if (ret || offset < 0 || offset > UINT_MAX) {
 690                         pr_info("Failed to parse either an address or a symbol.\n");
 691                         return ret;
 692                 }
 693                 if (offset && is_return) {
 694                         pr_info("Return probe must be used without offset.\n");
 695                         return -EINVAL;
 696                 }
 697         } else if (is_return) {
 698                 pr_info("Return probe point must be a symbol.\n");
 699                 return -EINVAL;
 700         }
 701         argc -= 2; argv += 2;
 702
 703         /* setup a probe */
 704         if (!event) {
 705                 /* Make a new event name */
 706                 if (symbol)
 707                         snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
 708                                  is_return ? 'r' : 'p', symbol, offset);
 709                 else
 710                         snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
 711                                  is_return ? 'r' : 'p', addr);
 712                 event = buf;
 713         }
 714         tk = alloc_trace_kprobe(group, event, addr, symbol, offset, argc,
 715                                is_return);
 716         if (IS_ERR(tk)) {
 717                 pr_info("Failed to allocate trace_probe.(%d)\n",
 718                         (int)PTR_ERR(tk));
 719                 return PTR_ERR(tk);
 720         }
 721
 722         /* parse arguments */
 723         ret = 0;
 724         for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
 725                 struct probe_arg *parg = &tk->tp.args[i];
 726
 727                 /* Increment count for freeing args in error case */
 728                 tk->tp.nr_args++;
 729
 730                 /* Parse argument name */
 731                 arg = strchr(argv[i], '=');
 732                 if (arg) {
 733                         *arg++ = '\0';
 734                         parg->name = kstrdup(argv[i], GFP_KERNEL);
 735                 } else {
 736                         arg = argv[i];
 737                         /* If argument name is omitted, set "argN" */
 738                         snprintf(buf, MAX_EVENT_NAME_LEN, "arg%d", i + 1);
 739                         parg->name = kstrdup(buf, GFP_KERNEL);
 740                 }
 741
 742                 if (!parg->name) {
 743                         pr_info("Failed to allocate argument[%d] name.\n", i);
 744                         ret = -ENOMEM;
 745                         goto error;
 746                 }
 747
 748                 if (!is_good_name(parg->name)) {
 749                         pr_info("Invalid argument[%d] name: %s\n",
 750                                 i, parg->name);
 751                         ret = -EINVAL;
 752                         goto error;
 753                 }
 754
 755                 if (traceprobe_conflict_field_name(parg->name,
 756                                                         tk->tp.args, i)) {
 757                         pr_info("Argument[%d] name '%s' conflicts with "
 758                                 "another field.\n", i, argv[i]);
 759                         ret = -EINVAL;
 760                         goto error;
 761                 }
 762
 763                 /* Parse fetch argument */
 764                 ret = traceprobe_parse_probe_arg(arg, &tk->tp.size, parg,
 765                                                 is_return, true,
 766                                                 kprobes_fetch_type_table);
 767                 if (ret) {
 768                         pr_info("Parse error at argument[%d]. (%d)\n", i, ret);
 769                         goto error;
 770                 }
 771         }
 772
 773         ret = register_trace_kprobe(tk);
 774         if (ret)
 775                 goto error;
 776         return 0;
 777
 778 error:
 779         free_trace_kprobe(tk);
 780         return ret;
 781 }
 782
 783 static int release_all_trace_kprobes(void)
 784 {
 785         struct trace_kprobe *tk;
 786         int ret = 0;
 787
 788         mutex_lock(&probe_lock);
 789         /* Ensure no probe is in use. */
 790         list_for_each_entry(tk, &probe_list, list)
 791                 if (trace_probe_is_enabled(&tk->tp)) {
 792                         ret = -EBUSY;
 793                         goto end;
 794                 }
 795         /* TODO: Use batch unregistration */
 796         while (!list_empty(&probe_list)) {
 797                 tk = list_entry(probe_list.next, struct trace_kprobe, list);
 798                 ret = unregister_trace_kprobe(tk);
 799                 if (ret)
 800                         goto end;
 801                 free_trace_kprobe(tk);
 802         }
 803
 804 end:
 805         mutex_unlock(&probe_lock);
 806
 807         return ret;
 808 }
 809
 810 /* Probes listing interfaces */
 811 static void *probes_seq_start(struct seq_file *m, loff_t *pos)
 812 {
 813         mutex_lock(&probe_lock);
 814         return seq_list_start(&probe_list, *pos);
 815 }
 816
 817 static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
 818 {
 819         return seq_list_next(v, &probe_list, pos);
 820 }
 821
 822 static void probes_seq_stop(struct seq_file *m, void *v)
 823 {
 824         mutex_unlock(&probe_lock);
 825 }
 826
 827 static int probes_seq_show(struct seq_file *m, void *v)
 828 {
 829         struct trace_kprobe *tk = v;
 830         int i;
 831
 832         seq_putc(m, trace_kprobe_is_return(tk) ? 'r' : 'p');
 833         seq_printf(m, ":%s/%s", tk->tp.call.class->system,
 834                         trace_event_name(&tk->tp.call));
 835
 836         if (!tk->symbol)
 837                 seq_printf(m, " 0x%p", tk->rp.kp.addr);
 838         else if (tk->rp.kp.offset)
 839                 seq_printf(m, " %s+%u", trace_kprobe_symbol(tk),
 840                            tk->rp.kp.offset);
 841         else
 842                 seq_printf(m, " %s", trace_kprobe_symbol(tk));
 843
 844         for (i = 0; i < tk->tp.nr_args; i++)
 845                 seq_printf(m, " %s=%s", tk->tp.args[i].name, tk->tp.args[i].comm);
 846         seq_putc(m, '\n');
 847
 848         return 0;
 849 }
 850
 851 static const struct seq_operations probes_seq_op = {
 852         .start  = probes_seq_start,
 853         .next   = probes_seq_next,
 854         .stop   = probes_seq_stop,
 855         .show   = probes_seq_show
 856 };
 857
 858 static int probes_open(struct inode *inode, struct file *file)
 859 {
 860         int ret;
 861
 862         if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
 863                 ret = release_all_trace_kprobes();
 864                 if (ret < 0)
 865                         return ret;
 866         }
 867
 868         return seq_open(file, &probes_seq_op);
 869 }
 870
 871 static ssize_t probes_write(struct file *file, const char __user *buffer,
 872                             size_t count, loff_t *ppos)
 873 {
 874         return traceprobe_probes_write(file, buffer, count, ppos,
 875                         create_trace_kprobe);
 876 }
 877
 878 static const struct file_operations kprobe_events_ops = {
 879         .owner          = THIS_MODULE,
 880         .open           = probes_open,
 881         .read           = seq_read,
 882         .llseek         = seq_lseek,
 883         .release        = seq_release,
 884         .write          = probes_write,
 885 };
 886
 887 /* Probes profiling interfaces */
 888 static int probes_profile_seq_show(struct seq_file *m, void *v)
 889 {
 890         struct trace_kprobe *tk = v;
 891         unsigned long nhit = 0;
 892         int cpu;
 893
 894         for_each_possible_cpu(cpu)
 895                 nhit += *per_cpu_ptr(tk->nhit, cpu);
 896
 897         seq_printf(m, "  %-44s %15lu %15lu\n",
 898                    trace_event_name(&tk->tp.call), nhit,
 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 /*
1365  * The "__used" keeps gcc from removing the function symbol
1366  * from the kallsyms table.
1367  */
1368 static __used int kprobe_trace_selftest_target(int a1, int a2, int a3,
1369                                                int a4, int a5, int a6)
1370 {
1371         return a1 + a2 + a3 + a4 + a5 + a6;
1372 }
1373
1374 static 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         /* Disable trace points before removing it */
1453         tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1454         if (WARN_ON_ONCE(tk == NULL)) {
1455                 pr_warn("error on getting test probe.\n");
1456                 warn++;
1457         } else {
1458                 file = find_trace_probe_file(tk, top_trace_array());
1459                 if (WARN_ON_ONCE(file == NULL)) {
1460                         pr_warn("error on getting probe file.\n");
1461                         warn++;
1462                 } else
1463                         disable_trace_kprobe(tk, file);
1464         }
1465
1466         tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1467         if (WARN_ON_ONCE(tk == NULL)) {
1468                 pr_warn("error on getting 2nd test probe.\n");
1469                 warn++;
1470         } else {
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         ret = traceprobe_command("-:testprobe", create_trace_kprobe);
1480         if (WARN_ON_ONCE(ret)) {
1481                 pr_warn("error on deleting a probe.\n");
1482                 warn++;
1483         }
1484
1485         ret = traceprobe_command("-:testprobe2", create_trace_kprobe);
1486         if (WARN_ON_ONCE(ret)) {
1487                 pr_warn("error on deleting a probe.\n");
1488                 warn++;
1489         }
1490
1491 end:
1492         release_all_trace_kprobes();
1493         /*
1494          * Wait for the optimizer work to finish. Otherwise it might fiddle
1495          * with probes in already freed __init text.
1496          */
1497         wait_for_kprobe_optimizer();
1498         if (warn)
1499                 pr_cont("NG: Some tests are failed. Please check them.\n");
1500         else
1501                 pr_cont("OK\n");
1502         return 0;
1503 }
1504
1505 late_initcall(kprobe_trace_self_tests_init);
1506
1507 #endif