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 #define KRETPROBE_MAXACTIVE_MAX 4096
  29
  30 /**
  31  * Kprobe event core functions
  32  */
  33 struct trace_kprobe {
  34         struct list_head        list;
  35         struct kretprobe        rp;     /* Use rp.kp for kprobe use */
  36         unsigned long __percpu *nhit;
  37         const char              *symbol;        /* symbol name */
  38         struct trace_probe      tp;
  39 };
  40
  41 #define SIZEOF_TRACE_KPROBE(n)                          \
  42         (offsetof(struct trace_kprobe, tp.args) +       \
  43         (sizeof(struct probe_arg) * (n)))
  44
  45
  46 static nokprobe_inline bool trace_kprobe_is_return(struct trace_kprobe *tk)
  47 {
  48         return tk->rp.handler != NULL;
  49 }
  50
  51 static nokprobe_inline const char *trace_kprobe_symbol(struct trace_kprobe *tk)
  52 {
  53         return tk->symbol ? tk->symbol : "unknown";
  54 }
  55
  56 static nokprobe_inline unsigned long trace_kprobe_offset(struct trace_kprobe *tk)
  57 {
  58         return tk->rp.kp.offset;
  59 }
  60
  61 static nokprobe_inline bool trace_kprobe_has_gone(struct trace_kprobe *tk)
  62 {
  63         return !!(kprobe_gone(&tk->rp.kp));
  64 }
  65
  66 static nokprobe_inline bool trace_kprobe_within_module(struct trace_kprobe *tk,
  67                                                  struct module *mod)
  68 {
  69         int len = strlen(mod->name);
  70         const char *name = trace_kprobe_symbol(tk);
  71         return strncmp(mod->name, name, len) == 0 && name[len] == ':';
  72 }
  73
  74 static nokprobe_inline bool trace_kprobe_is_on_module(struct trace_kprobe *tk)
  75 {
  76         return !!strchr(trace_kprobe_symbol(tk), ':');
  77 }
  78
  79 static nokprobe_inline unsigned long trace_kprobe_nhit(struct trace_kprobe *tk)
  80 {
  81         unsigned long nhit = 0;
  82         int cpu;
  83
  84         for_each_possible_cpu(cpu)
  85                 nhit += *per_cpu_ptr(tk->nhit, cpu);
  86
  87         return nhit;
  88 }
  89
  90 static int register_kprobe_event(struct trace_kprobe *tk);
  91 static int unregister_kprobe_event(struct trace_kprobe *tk);
  92
  93 static DEFINE_MUTEX(probe_lock);
  94 static LIST_HEAD(probe_list);
  95
  96 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
  97 static int kretprobe_dispatcher(struct kretprobe_instance *ri,
  98                                 struct pt_regs *regs);
  99
 100 /* Memory fetching by symbol */
 101 struct symbol_cache {
 102         char            *symbol;
 103         long            offset;
 104         unsigned long   addr;
 105 };
 106
 107 unsigned long update_symbol_cache(struct symbol_cache *sc)
 108 {
 109         sc->addr = (unsigned long)kallsyms_lookup_name(sc->symbol);
 110
 111         if (sc->addr)
 112                 sc->addr += sc->offset;
 113
 114         return sc->addr;
 115 }
 116
 117 void free_symbol_cache(struct symbol_cache *sc)
 118 {
 119         kfree(sc->symbol);
 120         kfree(sc);
 121 }
 122
 123 struct symbol_cache *alloc_symbol_cache(const char *sym, long offset)
 124 {
 125         struct symbol_cache *sc;
 126
 127         if (!sym || strlen(sym) == 0)
 128                 return NULL;
 129
 130         sc = kzalloc(sizeof(struct symbol_cache), GFP_KERNEL);
 131         if (!sc)
 132                 return NULL;
 133
 134         sc->symbol = kstrdup(sym, GFP_KERNEL);
 135         if (!sc->symbol) {
 136                 kfree(sc);
 137                 return NULL;
 138         }
 139         sc->offset = offset;
 140         update_symbol_cache(sc);
 141
 142         return sc;
 143 }
 144
 145 /*
 146  * Kprobes-specific fetch functions
 147  */
 148 #define DEFINE_FETCH_stack(type)                                        \
 149 static void FETCH_FUNC_NAME(stack, type)(struct pt_regs *regs,          \
 150                                           void *offset, void *dest)     \
 151 {                                                                       \
 152         *(type *)dest = (type)regs_get_kernel_stack_nth(regs,           \
 153                                 (unsigned int)((unsigned long)offset)); \
 154 }                                                                       \
 155 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(stack, type));
 156
 157 DEFINE_BASIC_FETCH_FUNCS(stack)
 158 /* No string on the stack entry */
 159 #define fetch_stack_string      NULL
 160 #define fetch_stack_string_size NULL
 161
 162 #define DEFINE_FETCH_memory(type)                                       \
 163 static void FETCH_FUNC_NAME(memory, type)(struct pt_regs *regs,         \
 164                                           void *addr, void *dest)       \
 165 {                                                                       \
 166         type retval;                                                    \
 167         if (probe_kernel_address(addr, retval))                         \
 168                 *(type *)dest = 0;                                      \
 169         else                                                            \
 170                 *(type *)dest = retval;                                 \
 171 }                                                                       \
 172 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(memory, type));
 173
 174 DEFINE_BASIC_FETCH_FUNCS(memory)
 175 /*
 176  * Fetch a null-terminated string. Caller MUST set *(u32 *)dest with max
 177  * length and relative data location.
 178  */
 179 static void FETCH_FUNC_NAME(memory, string)(struct pt_regs *regs,
 180                                             void *addr, void *dest)
 181 {
 182         int maxlen = get_rloc_len(*(u32 *)dest);
 183         u8 *dst = get_rloc_data(dest);
 184         long ret;
 185
 186         if (!maxlen)
 187                 return;
 188
 189         /*
 190          * Try to get string again, since the string can be changed while
 191          * probing.
 192          */
 193         ret = strncpy_from_unsafe(dst, addr, maxlen);
 194
 195         if (ret < 0) {  /* Failed to fetch string */
 196                 dst[0] = '\0';
 197                 *(u32 *)dest = make_data_rloc(0, get_rloc_offs(*(u32 *)dest));
 198         } else {
 199                 *(u32 *)dest = make_data_rloc(ret, get_rloc_offs(*(u32 *)dest));
 200         }
 201 }
 202 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(memory, string));
 203
 204 /* Return the length of string -- including null terminal byte */
 205 static void FETCH_FUNC_NAME(memory, string_size)(struct pt_regs *regs,
 206                                                  void *addr, void *dest)
 207 {
 208         mm_segment_t old_fs;
 209         int ret, len = 0;
 210         u8 c;
 211
 212         old_fs = get_fs();
 213         set_fs(KERNEL_DS);
 214         pagefault_disable();
 215
 216         do {
 217                 ret = __copy_from_user_inatomic(&c, (u8 *)addr + len, 1);
 218                 len++;
 219         } while (c && ret == 0 && len < MAX_STRING_SIZE);
 220
 221         pagefault_enable();
 222         set_fs(old_fs);
 223
 224         if (ret < 0)    /* Failed to check the length */
 225                 *(u32 *)dest = 0;
 226         else
 227                 *(u32 *)dest = len;
 228 }
 229 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(memory, string_size));
 230
 231 #define DEFINE_FETCH_symbol(type)                                       \
 232 void FETCH_FUNC_NAME(symbol, type)(struct pt_regs *regs, void *data, void *dest)\
 233 {                                                                       \
 234         struct symbol_cache *sc = data;                                 \
 235         if (sc->addr)                                                   \
 236                 fetch_memory_##type(regs, (void *)sc->addr, dest);      \
 237         else                                                            \
 238                 *(type *)dest = 0;                                      \
 239 }                                                                       \
 240 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(symbol, type));
 241
 242 DEFINE_BASIC_FETCH_FUNCS(symbol)
 243 DEFINE_FETCH_symbol(string)
 244 DEFINE_FETCH_symbol(string_size)
 245
 246 /* kprobes don't support file_offset fetch methods */
 247 #define fetch_file_offset_u8            NULL
 248 #define fetch_file_offset_u16           NULL
 249 #define fetch_file_offset_u32           NULL
 250 #define fetch_file_offset_u64           NULL
 251 #define fetch_file_offset_string        NULL
 252 #define fetch_file_offset_string_size   NULL
 253
 254 /* Fetch type information table */
 255 static const struct fetch_type kprobes_fetch_type_table[] = {
 256         /* Special types */
 257         [FETCH_TYPE_STRING] = __ASSIGN_FETCH_TYPE("string", string, string,
 258                                         sizeof(u32), 1, "__data_loc char[]"),
 259         [FETCH_TYPE_STRSIZE] = __ASSIGN_FETCH_TYPE("string_size", u32,
 260                                         string_size, sizeof(u32), 0, "u32"),
 261         /* Basic types */
 262         ASSIGN_FETCH_TYPE(u8,  u8,  0),
 263         ASSIGN_FETCH_TYPE(u16, u16, 0),
 264         ASSIGN_FETCH_TYPE(u32, u32, 0),
 265         ASSIGN_FETCH_TYPE(u64, u64, 0),
 266         ASSIGN_FETCH_TYPE(s8,  u8,  1),
 267         ASSIGN_FETCH_TYPE(s16, u16, 1),
 268         ASSIGN_FETCH_TYPE(s32, u32, 1),
 269         ASSIGN_FETCH_TYPE(s64, u64, 1),
 270         ASSIGN_FETCH_TYPE_ALIAS(x8,  u8,  u8,  0),
 271         ASSIGN_FETCH_TYPE_ALIAS(x16, u16, u16, 0),
 272         ASSIGN_FETCH_TYPE_ALIAS(x32, u32, u32, 0),
 273         ASSIGN_FETCH_TYPE_ALIAS(x64, u64, u64, 0),
 274
 275         ASSIGN_FETCH_TYPE_END
 276 };
 277
 278 /*
 279  * Allocate new trace_probe and initialize it (including kprobes).
 280  */
 281 static struct trace_kprobe *alloc_trace_kprobe(const char *group,
 282                                              const char *event,
 283                                              void *addr,
 284                                              const char *symbol,
 285                                              unsigned long offs,
 286                                              int maxactive,
 287                                              int nargs, bool is_return)
 288 {
 289         struct trace_kprobe *tk;
 290         int ret = -ENOMEM;
 291
 292         tk = kzalloc(SIZEOF_TRACE_KPROBE(nargs), GFP_KERNEL);
 293         if (!tk)
 294                 return ERR_PTR(ret);
 295
 296         tk->nhit = alloc_percpu(unsigned long);
 297         if (!tk->nhit)
 298                 goto error;
 299
 300         if (symbol) {
 301                 tk->symbol = kstrdup(symbol, GFP_KERNEL);
 302                 if (!tk->symbol)
 303                         goto error;
 304                 tk->rp.kp.symbol_name = tk->symbol;
 305                 tk->rp.kp.offset = offs;
 306         } else
 307                 tk->rp.kp.addr = addr;
 308
 309         if (is_return)
 310                 tk->rp.handler = kretprobe_dispatcher;
 311         else
 312                 tk->rp.kp.pre_handler = kprobe_dispatcher;
 313
 314         tk->rp.maxactive = maxactive;
 315
 316         if (!event || !is_good_name(event)) {
 317                 ret = -EINVAL;
 318                 goto error;
 319         }
 320
 321         tk->tp.call.class = &tk->tp.class;
 322         tk->tp.call.name = kstrdup(event, GFP_KERNEL);
 323         if (!tk->tp.call.name)
 324                 goto error;
 325
 326         if (!group || !is_good_name(group)) {
 327                 ret = -EINVAL;
 328                 goto error;
 329         }
 330
 331         tk->tp.class.system = kstrdup(group, GFP_KERNEL);
 332         if (!tk->tp.class.system)
 333                 goto error;
 334
 335         INIT_LIST_HEAD(&tk->list);
 336         INIT_LIST_HEAD(&tk->tp.files);
 337         return tk;
 338 error:
 339         kfree(tk->tp.call.name);
 340         kfree(tk->symbol);
 341         free_percpu(tk->nhit);
 342         kfree(tk);
 343         return ERR_PTR(ret);
 344 }
 345
 346 static void free_trace_kprobe(struct trace_kprobe *tk)
 347 {
 348         int i;
 349
 350         for (i = 0; i < tk->tp.nr_args; i++)
 351                 traceprobe_free_probe_arg(&tk->tp.args[i]);
 352
 353         kfree(tk->tp.call.class->system);
 354         kfree(tk->tp.call.name);
 355         kfree(tk->symbol);
 356         free_percpu(tk->nhit);
 357         kfree(tk);
 358 }
 359
 360 static struct trace_kprobe *find_trace_kprobe(const char *event,
 361                                               const char *group)
 362 {
 363         struct trace_kprobe *tk;
 364
 365         list_for_each_entry(tk, &probe_list, list)
 366                 if (strcmp(trace_event_name(&tk->tp.call), event) == 0 &&
 367                     strcmp(tk->tp.call.class->system, group) == 0)
 368                         return tk;
 369         return NULL;
 370 }
 371
 372 /*
 373  * Enable trace_probe
 374  * if the file is NULL, enable "perf" handler, or enable "trace" handler.
 375  */
 376 static int
 377 enable_trace_kprobe(struct trace_kprobe *tk, struct trace_event_file *file)
 378 {
 379         struct event_file_link *link = NULL;
 380         int ret = 0;
 381
 382         if (file) {
 383                 link = kmalloc(sizeof(*link), GFP_KERNEL);
 384                 if (!link) {
 385                         ret = -ENOMEM;
 386                         goto out;
 387                 }
 388
 389                 link->file = file;
 390                 list_add_tail_rcu(&link->list, &tk->tp.files);
 391
 392                 tk->tp.flags |= TP_FLAG_TRACE;
 393         } else
 394                 tk->tp.flags |= TP_FLAG_PROFILE;
 395
 396         if (trace_probe_is_registered(&tk->tp) && !trace_kprobe_has_gone(tk)) {
 397                 if (trace_kprobe_is_return(tk))
 398                         ret = enable_kretprobe(&tk->rp);
 399                 else
 400                         ret = enable_kprobe(&tk->rp.kp);
 401         }
 402
 403         if (ret) {
 404                 if (file) {
 405                         /* Notice the if is true on not WARN() */
 406                         if (!WARN_ON_ONCE(!link))
 407                                 list_del_rcu(&link->list);
 408                         kfree(link);
 409                         tk->tp.flags &= ~TP_FLAG_TRACE;
 410                 } else {
 411                         tk->tp.flags &= ~TP_FLAG_PROFILE;
 412                 }
 413         }
 414  out:
 415         return ret;
 416 }
 417
 418 /*
 419  * Disable trace_probe
 420  * if the file is NULL, disable "perf" handler, or disable "trace" handler.
 421  */
 422 static int
 423 disable_trace_kprobe(struct trace_kprobe *tk, struct trace_event_file *file)
 424 {
 425         struct event_file_link *link = NULL;
 426         int wait = 0;
 427         int ret = 0;
 428
 429         if (file) {
 430                 link = find_event_file_link(&tk->tp, file);
 431                 if (!link) {
 432                         ret = -EINVAL;
 433                         goto out;
 434                 }
 435
 436                 list_del_rcu(&link->list);
 437                 wait = 1;
 438                 if (!list_empty(&tk->tp.files))
 439                         goto out;
 440
 441                 tk->tp.flags &= ~TP_FLAG_TRACE;
 442         } else
 443                 tk->tp.flags &= ~TP_FLAG_PROFILE;
 444
 445         if (!trace_probe_is_enabled(&tk->tp) && trace_probe_is_registered(&tk->tp)) {
 446                 if (trace_kprobe_is_return(tk))
 447                         disable_kretprobe(&tk->rp);
 448                 else
 449                         disable_kprobe(&tk->rp.kp);
 450                 wait = 1;
 451         }
 452  out:
 453         if (wait) {
 454                 /*
 455                  * Synchronize with kprobe_trace_func/kretprobe_trace_func
 456                  * to ensure disabled (all running handlers are finished).
 457                  * This is not only for kfree(), but also the caller,
 458                  * trace_remove_event_call() supposes it for releasing
 459                  * event_call related objects, which will be accessed in
 460                  * the kprobe_trace_func/kretprobe_trace_func.
 461                  */
 462                 synchronize_sched();
 463                 kfree(link);    /* Ignored if link == NULL */
 464         }
 465
 466         return ret;
 467 }
 468
 469 /* Internal register function - just handle k*probes and flags */
 470 static int __register_trace_kprobe(struct trace_kprobe *tk)
 471 {
 472         int i, ret;
 473
 474         if (trace_probe_is_registered(&tk->tp))
 475                 return -EINVAL;
 476
 477         for (i = 0; i < tk->tp.nr_args; i++)
 478                 traceprobe_update_arg(&tk->tp.args[i]);
 479
 480         /* Set/clear disabled flag according to tp->flag */
 481         if (trace_probe_is_enabled(&tk->tp))
 482                 tk->rp.kp.flags &= ~KPROBE_FLAG_DISABLED;
 483         else
 484                 tk->rp.kp.flags |= KPROBE_FLAG_DISABLED;
 485
 486         if (trace_kprobe_is_return(tk))
 487                 ret = register_kretprobe(&tk->rp);
 488         else
 489                 ret = register_kprobe(&tk->rp.kp);
 490
 491         if (ret == 0)
 492                 tk->tp.flags |= TP_FLAG_REGISTERED;
 493         else {
 494                 pr_warn("Could not insert probe at %s+%lu: %d\n",
 495                         trace_kprobe_symbol(tk), trace_kprobe_offset(tk), ret);
 496                 if (ret == -ENOENT && trace_kprobe_is_on_module(tk)) {
 497                         pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
 498                         ret = 0;
 499                 } else if (ret == -EILSEQ) {
 500                         pr_warn("Probing address(0x%p) is not an instruction boundary.\n",
 501                                 tk->rp.kp.addr);
 502                         ret = -EINVAL;
 503                 }
 504         }
 505
 506         return ret;
 507 }
 508
 509 /* Internal unregister function - just handle k*probes and flags */
 510 static void __unregister_trace_kprobe(struct trace_kprobe *tk)
 511 {
 512         if (trace_probe_is_registered(&tk->tp)) {
 513                 if (trace_kprobe_is_return(tk))
 514                         unregister_kretprobe(&tk->rp);
 515                 else
 516                         unregister_kprobe(&tk->rp.kp);
 517                 tk->tp.flags &= ~TP_FLAG_REGISTERED;
 518                 /* Cleanup kprobe for reuse */
 519                 if (tk->rp.kp.symbol_name)
 520                         tk->rp.kp.addr = NULL;
 521         }
 522 }
 523
 524 /* Unregister a trace_probe and probe_event: call with locking probe_lock */
 525 static int unregister_trace_kprobe(struct trace_kprobe *tk)
 526 {
 527         /* Enabled event can not be unregistered */
 528         if (trace_probe_is_enabled(&tk->tp))
 529                 return -EBUSY;
 530
 531         /* Will fail if probe is being used by ftrace or perf */
 532         if (unregister_kprobe_event(tk))
 533                 return -EBUSY;
 534
 535         __unregister_trace_kprobe(tk);
 536         list_del(&tk->list);
 537
 538         return 0;
 539 }
 540
 541 /* Register a trace_probe and probe_event */
 542 static int register_trace_kprobe(struct trace_kprobe *tk)
 543 {
 544         struct trace_kprobe *old_tk;
 545         int ret;
 546
 547         mutex_lock(&probe_lock);
 548
 549         /* Delete old (same name) event if exist */
 550         old_tk = find_trace_kprobe(trace_event_name(&tk->tp.call),
 551                         tk->tp.call.class->system);
 552         if (old_tk) {
 553                 ret = unregister_trace_kprobe(old_tk);
 554                 if (ret < 0)
 555                         goto end;
 556                 free_trace_kprobe(old_tk);
 557         }
 558
 559         /* Register new event */
 560         ret = register_kprobe_event(tk);
 561         if (ret) {
 562                 pr_warn("Failed to register probe event(%d)\n", ret);
 563                 goto end;
 564         }
 565
 566         /* Register k*probe */
 567         ret = __register_trace_kprobe(tk);
 568         if (ret < 0)
 569                 unregister_kprobe_event(tk);
 570         else
 571                 list_add_tail(&tk->list, &probe_list);
 572
 573 end:
 574         mutex_unlock(&probe_lock);
 575         return ret;
 576 }
 577
 578 /* Module notifier call back, checking event on the module */
 579 static int trace_kprobe_module_callback(struct notifier_block *nb,
 580                                        unsigned long val, void *data)
 581 {
 582         struct module *mod = data;
 583         struct trace_kprobe *tk;
 584         int ret;
 585
 586         if (val != MODULE_STATE_COMING)
 587                 return NOTIFY_DONE;
 588
 589         /* Update probes on coming module */
 590         mutex_lock(&probe_lock);
 591         list_for_each_entry(tk, &probe_list, list) {
 592                 if (trace_kprobe_within_module(tk, mod)) {
 593                         /* Don't need to check busy - this should have gone. */
 594                         __unregister_trace_kprobe(tk);
 595                         ret = __register_trace_kprobe(tk);
 596                         if (ret)
 597                                 pr_warn("Failed to re-register probe %s on %s: %d\n",
 598                                         trace_event_name(&tk->tp.call),
 599                                         mod->name, ret);
 600                 }
 601         }
 602         mutex_unlock(&probe_lock);
 603
 604         return NOTIFY_DONE;
 605 }
 606
 607 static struct notifier_block trace_kprobe_module_nb = {
 608         .notifier_call = trace_kprobe_module_callback,
 609         .priority = 1   /* Invoked after kprobe module callback */
 610 };
 611
 612 /* Convert certain expected symbols into '_' when generating event names */
 613 static inline void sanitize_event_name(char *name)
 614 {
 615         while (*name++ != '\0')
 616                 if (*name == ':' || *name == '.')
 617                         *name = '_';
 618 }
 619
 620 static int create_trace_kprobe(int argc, char **argv)
 621 {
 622         /*
 623          * Argument syntax:
 624          *  - Add kprobe:
 625          *      p[:[GRP/]EVENT] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
 626          *  - Add kretprobe:
 627          *      r[MAXACTIVE][:[GRP/]EVENT] [MOD:]KSYM[+0] [FETCHARGS]
 628          * Fetch args:
 629          *  $retval     : fetch return value
 630          *  $stack      : fetch stack address
 631          *  $stackN     : fetch Nth of stack (N:0-)
 632          *  $comm       : fetch current task comm
 633          *  @ADDR       : fetch memory at ADDR (ADDR should be in kernel)
 634          *  @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
 635          *  %REG        : fetch register REG
 636          * Dereferencing memory fetch:
 637          *  +|-offs(ARG) : fetch memory at ARG +|- offs address.
 638          * Alias name of args:
 639          *  NAME=FETCHARG : set NAME as alias of FETCHARG.
 640          * Type of args:
 641          *  FETCHARG:TYPE : use TYPE instead of unsigned long.
 642          */
 643         struct trace_kprobe *tk;
 644         int i, ret = 0;
 645         bool is_return = false, is_delete = false;
 646         char *symbol = NULL, *event = NULL, *group = NULL;
 647         int maxactive = 0;
 648         char *arg;
 649         long offset = 0;
 650         void *addr = NULL;
 651         char buf[MAX_EVENT_NAME_LEN];
 652
 653         /* argc must be >= 1 */
 654         if (argv[0][0] == 'p')
 655                 is_return = false;
 656         else if (argv[0][0] == 'r')
 657                 is_return = true;
 658         else if (argv[0][0] == '-')
 659                 is_delete = true;
 660         else {
 661                 pr_info("Probe definition must be started with 'p', 'r' or"
 662                         " '-'.\n");
 663                 return -EINVAL;
 664         }
 665
 666         event = strchr(&argv[0][1], ':');
 667         if (event) {
 668                 event[0] = '\0';
 669                 event++;
 670         }
 671         if (is_return && isdigit(argv[0][1])) {
 672                 ret = kstrtouint(&argv[0][1], 0, &maxactive);
 673                 if (ret) {
 674                         pr_info("Failed to parse maxactive.\n");
 675                         return ret;
 676                 }
 677                 /* kretprobes instances are iterated over via a list. The
 678                  * maximum should stay reasonable.
 679                  */
 680                 if (maxactive > KRETPROBE_MAXACTIVE_MAX) {
 681                         pr_info("Maxactive is too big (%d > %d).\n",
 682                                 maxactive, KRETPROBE_MAXACTIVE_MAX);
 683                         return -E2BIG;
 684                 }
 685         }
 686
 687         if (event) {
 688                 if (strchr(event, '/')) {
 689                         group = event;
 690                         event = strchr(group, '/') + 1;
 691                         event[-1] = '\0';
 692                         if (strlen(group) == 0) {
 693                                 pr_info("Group name is not specified\n");
 694                                 return -EINVAL;
 695                         }
 696                 }
 697                 if (strlen(event) == 0) {
 698                         pr_info("Event name is not specified\n");
 699                         return -EINVAL;
 700                 }
 701         }
 702         if (!group)
 703                 group = KPROBE_EVENT_SYSTEM;
 704
 705         if (is_delete) {
 706                 if (!event) {
 707                         pr_info("Delete command needs an event name.\n");
 708                         return -EINVAL;
 709                 }
 710                 mutex_lock(&probe_lock);
 711                 tk = find_trace_kprobe(event, group);
 712                 if (!tk) {
 713                         mutex_unlock(&probe_lock);
 714                         pr_info("Event %s/%s doesn't exist.\n", group, event);
 715                         return -ENOENT;
 716                 }
 717                 /* delete an event */
 718                 ret = unregister_trace_kprobe(tk);
 719                 if (ret == 0)
 720                         free_trace_kprobe(tk);
 721                 mutex_unlock(&probe_lock);
 722                 return ret;
 723         }
 724
 725         if (argc < 2) {
 726                 pr_info("Probe point is not specified.\n");
 727                 return -EINVAL;
 728         }
 729
 730         /* try to parse an address. if that fails, try to read the
 731          * input as a symbol. */
 732         if (kstrtoul(argv[1], 0, (unsigned long *)&addr)) {
 733                 /* a symbol specified */
 734                 symbol = argv[1];
 735                 /* TODO: support .init module functions */
 736                 ret = traceprobe_split_symbol_offset(symbol, &offset);
 737                 if (ret || offset < 0 || offset > UINT_MAX) {
 738                         pr_info("Failed to parse either an address or a symbol.\n");
 739                         return ret;
 740                 }
 741                 if (offset && is_return &&
 742                     !kprobe_on_func_entry(NULL, symbol, offset)) {
 743                         pr_info("Given offset is not valid for return probe.\n");
 744                         return -EINVAL;
 745                 }
 746         }
 747         argc -= 2; argv += 2;
 748
 749         /* setup a probe */
 750         if (!event) {
 751                 /* Make a new event name */
 752                 if (symbol)
 753                         snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
 754                                  is_return ? 'r' : 'p', symbol, offset);
 755                 else
 756                         snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
 757                                  is_return ? 'r' : 'p', addr);
 758                 sanitize_event_name(buf);
 759                 event = buf;
 760         }
 761         tk = alloc_trace_kprobe(group, event, addr, symbol, offset, maxactive,
 762                                argc, is_return);
 763         if (IS_ERR(tk)) {
 764                 pr_info("Failed to allocate trace_probe.(%d)\n",
 765                         (int)PTR_ERR(tk));
 766                 return PTR_ERR(tk);
 767         }
 768
 769         /* parse arguments */
 770         ret = 0;
 771         for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
 772                 struct probe_arg *parg = &tk->tp.args[i];
 773
 774                 /* Increment count for freeing args in error case */
 775                 tk->tp.nr_args++;
 776
 777                 /* Parse argument name */
 778                 arg = strchr(argv[i], '=');
 779                 if (arg) {
 780                         *arg++ = '\0';
 781                         parg->name = kstrdup(argv[i], GFP_KERNEL);
 782                 } else {
 783                         arg = argv[i];
 784                         /* If argument name is omitted, set "argN" */
 785                         snprintf(buf, MAX_EVENT_NAME_LEN, "arg%d", i + 1);
 786                         parg->name = kstrdup(buf, GFP_KERNEL);
 787                 }
 788
 789                 if (!parg->name) {
 790                         pr_info("Failed to allocate argument[%d] name.\n", i);
 791                         ret = -ENOMEM;
 792                         goto error;
 793                 }
 794
 795                 if (!is_good_name(parg->name)) {
 796                         pr_info("Invalid argument[%d] name: %s\n",
 797                                 i, parg->name);
 798                         ret = -EINVAL;
 799                         goto error;
 800                 }
 801
 802                 if (traceprobe_conflict_field_name(parg->name,
 803                                                         tk->tp.args, i)) {
 804                         pr_info("Argument[%d] name '%s' conflicts with "
 805                                 "another field.\n", i, argv[i]);
 806                         ret = -EINVAL;
 807                         goto error;
 808                 }
 809
 810                 /* Parse fetch argument */
 811                 ret = traceprobe_parse_probe_arg(arg, &tk->tp.size, parg,
 812                                                 is_return, true,
 813                                                 kprobes_fetch_type_table);
 814                 if (ret) {
 815                         pr_info("Parse error at argument[%d]. (%d)\n", i, ret);
 816                         goto error;
 817                 }
 818         }
 819
 820         ret = register_trace_kprobe(tk);
 821         if (ret)
 822                 goto error;
 823         return 0;
 824
 825 error:
 826         free_trace_kprobe(tk);
 827         return ret;
 828 }
 829
 830 static int release_all_trace_kprobes(void)
 831 {
 832         struct trace_kprobe *tk;
 833         int ret = 0;
 834
 835         mutex_lock(&probe_lock);
 836         /* Ensure no probe is in use. */
 837         list_for_each_entry(tk, &probe_list, list)
 838                 if (trace_probe_is_enabled(&tk->tp)) {
 839                         ret = -EBUSY;
 840                         goto end;
 841                 }
 842         /* TODO: Use batch unregistration */
 843         while (!list_empty(&probe_list)) {
 844                 tk = list_entry(probe_list.next, struct trace_kprobe, list);
 845                 ret = unregister_trace_kprobe(tk);
 846                 if (ret)
 847                         goto end;
 848                 free_trace_kprobe(tk);
 849         }
 850
 851 end:
 852         mutex_unlock(&probe_lock);
 853
 854         return ret;
 855 }
 856
 857 /* Probes listing interfaces */
 858 static void *probes_seq_start(struct seq_file *m, loff_t *pos)
 859 {
 860         mutex_lock(&probe_lock);
 861         return seq_list_start(&probe_list, *pos);
 862 }
 863
 864 static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
 865 {
 866         return seq_list_next(v, &probe_list, pos);
 867 }
 868
 869 static void probes_seq_stop(struct seq_file *m, void *v)
 870 {
 871         mutex_unlock(&probe_lock);
 872 }
 873
 874 static int probes_seq_show(struct seq_file *m, void *v)
 875 {
 876         struct trace_kprobe *tk = v;
 877         int i;
 878
 879         seq_putc(m, trace_kprobe_is_return(tk) ? 'r' : 'p');
 880         if (trace_kprobe_is_return(tk) && tk->rp.maxactive)
 881                 seq_printf(m, "%d", tk->rp.maxactive);
 882         seq_printf(m, ":%s/%s", tk->tp.call.class->system,
 883                         trace_event_name(&tk->tp.call));
 884
 885         if (!tk->symbol)
 886                 seq_printf(m, " 0x%p", tk->rp.kp.addr);
 887         else if (tk->rp.kp.offset)
 888                 seq_printf(m, " %s+%u", trace_kprobe_symbol(tk),
 889                            tk->rp.kp.offset);
 890         else
 891                 seq_printf(m, " %s", trace_kprobe_symbol(tk));
 892
 893         for (i = 0; i < tk->tp.nr_args; i++)
 894                 seq_printf(m, " %s=%s", tk->tp.args[i].name, tk->tp.args[i].comm);
 895         seq_putc(m, '\n');
 896
 897         return 0;
 898 }
 899
 900 static const struct seq_operations probes_seq_op = {
 901         .start  = probes_seq_start,
 902         .next   = probes_seq_next,
 903         .stop   = probes_seq_stop,
 904         .show   = probes_seq_show
 905 };
 906
 907 static int probes_open(struct inode *inode, struct file *file)
 908 {
 909         int ret;
 910
 911         if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
 912                 ret = release_all_trace_kprobes();
 913                 if (ret < 0)
 914                         return ret;
 915         }
 916
 917         return seq_open(file, &probes_seq_op);
 918 }
 919
 920 static ssize_t probes_write(struct file *file, const char __user *buffer,
 921                             size_t count, loff_t *ppos)
 922 {
 923         return traceprobe_probes_write(file, buffer, count, ppos,
 924                         create_trace_kprobe);
 925 }
 926
 927 static const struct file_operations kprobe_events_ops = {
 928         .owner          = THIS_MODULE,
 929         .open           = probes_open,
 930         .read           = seq_read,
 931         .llseek         = seq_lseek,
 932         .release        = seq_release,
 933         .write          = probes_write,
 934 };
 935
 936 /* Probes profiling interfaces */
 937 static int probes_profile_seq_show(struct seq_file *m, void *v)
 938 {
 939         struct trace_kprobe *tk = v;
 940
 941         seq_printf(m, "  %-44s %15lu %15lu\n",
 942                    trace_event_name(&tk->tp.call),
 943                    trace_kprobe_nhit(tk),
 944                    tk->rp.kp.nmissed);
 945
 946         return 0;
 947 }
 948
 949 static const struct seq_operations profile_seq_op = {
 950         .start  = probes_seq_start,
 951         .next   = probes_seq_next,
 952         .stop   = probes_seq_stop,
 953         .show   = probes_profile_seq_show
 954 };
 955
 956 static int profile_open(struct inode *inode, struct file *file)
 957 {
 958         return seq_open(file, &profile_seq_op);
 959 }
 960
 961 static const struct file_operations kprobe_profile_ops = {
 962         .owner          = THIS_MODULE,
 963         .open           = profile_open,
 964         .read           = seq_read,
 965         .llseek         = seq_lseek,
 966         .release        = seq_release,
 967 };
 968
 969 /* Kprobe handler */
 970 static nokprobe_inline void
 971 __kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs,
 972                     struct trace_event_file *trace_file)
 973 {
 974         struct kprobe_trace_entry_head *entry;
 975         struct ring_buffer_event *event;
 976         struct ring_buffer *buffer;
 977         int size, dsize, pc;
 978         unsigned long irq_flags;
 979         struct trace_event_call *call = &tk->tp.call;
 980
 981         WARN_ON(call != trace_file->event_call);
 982
 983         if (trace_trigger_soft_disabled(trace_file))
 984                 return;
 985
 986         local_save_flags(irq_flags);
 987         pc = preempt_count();
 988
 989         dsize = __get_data_size(&tk->tp, regs);
 990         size = sizeof(*entry) + tk->tp.size + dsize;
 991
 992         event = trace_event_buffer_lock_reserve(&buffer, trace_file,
 993                                                 call->event.type,
 994                                                 size, irq_flags, pc);
 995         if (!event)
 996                 return;
 997
 998         entry = ring_buffer_event_data(event);
 999         entry->ip = (unsigned long)tk->rp.kp.addr;
1000         store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1001
1002         event_trigger_unlock_commit_regs(trace_file, buffer, event,
1003                                          entry, irq_flags, pc, regs);
1004 }
1005
1006 static void
1007 kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs)
1008 {
1009         struct event_file_link *link;
1010
1011         list_for_each_entry_rcu(link, &tk->tp.files, list)
1012                 __kprobe_trace_func(tk, regs, link->file);
1013 }
1014 NOKPROBE_SYMBOL(kprobe_trace_func);
1015
1016 /* Kretprobe handler */
1017 static nokprobe_inline void
1018 __kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1019                        struct pt_regs *regs,
1020                        struct trace_event_file *trace_file)
1021 {
1022         struct kretprobe_trace_entry_head *entry;
1023         struct ring_buffer_event *event;
1024         struct ring_buffer *buffer;
1025         int size, pc, dsize;
1026         unsigned long irq_flags;
1027         struct trace_event_call *call = &tk->tp.call;
1028
1029         WARN_ON(call != trace_file->event_call);
1030
1031         if (trace_trigger_soft_disabled(trace_file))
1032                 return;
1033
1034         local_save_flags(irq_flags);
1035         pc = preempt_count();
1036
1037         dsize = __get_data_size(&tk->tp, regs);
1038         size = sizeof(*entry) + tk->tp.size + dsize;
1039
1040         event = trace_event_buffer_lock_reserve(&buffer, trace_file,
1041                                                 call->event.type,
1042                                                 size, irq_flags, pc);
1043         if (!event)
1044                 return;
1045
1046         entry = ring_buffer_event_data(event);
1047         entry->func = (unsigned long)tk->rp.kp.addr;
1048         entry->ret_ip = (unsigned long)ri->ret_addr;
1049         store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1050
1051         event_trigger_unlock_commit_regs(trace_file, buffer, event,
1052                                          entry, irq_flags, pc, regs);
1053 }
1054
1055 static void
1056 kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1057                      struct pt_regs *regs)
1058 {
1059         struct event_file_link *link;
1060
1061         list_for_each_entry_rcu(link, &tk->tp.files, list)
1062                 __kretprobe_trace_func(tk, ri, regs, link->file);
1063 }
1064 NOKPROBE_SYMBOL(kretprobe_trace_func);
1065
1066 /* Event entry printers */
1067 static enum print_line_t
1068 print_kprobe_event(struct trace_iterator *iter, int flags,
1069                    struct trace_event *event)
1070 {
1071         struct kprobe_trace_entry_head *field;
1072         struct trace_seq *s = &iter->seq;
1073         struct trace_probe *tp;
1074         u8 *data;
1075         int i;
1076
1077         field = (struct kprobe_trace_entry_head *)iter->ent;
1078         tp = container_of(event, struct trace_probe, call.event);
1079
1080         trace_seq_printf(s, "%s: (", trace_event_name(&tp->call));
1081
1082         if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
1083                 goto out;
1084
1085         trace_seq_putc(s, ')');
1086
1087         data = (u8 *)&field[1];
1088         for (i = 0; i < tp->nr_args; i++)
1089                 if (!tp->args[i].type->print(s, tp->args[i].name,
1090                                              data + tp->args[i].offset, field))
1091                         goto out;
1092
1093         trace_seq_putc(s, '\n');
1094  out:
1095         return trace_handle_return(s);
1096 }
1097
1098 static enum print_line_t
1099 print_kretprobe_event(struct trace_iterator *iter, int flags,
1100                       struct trace_event *event)
1101 {
1102         struct kretprobe_trace_entry_head *field;
1103         struct trace_seq *s = &iter->seq;
1104         struct trace_probe *tp;
1105         u8 *data;
1106         int i;
1107
1108         field = (struct kretprobe_trace_entry_head *)iter->ent;
1109         tp = container_of(event, struct trace_probe, call.event);
1110
1111         trace_seq_printf(s, "%s: (", trace_event_name(&tp->call));
1112
1113         if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1114                 goto out;
1115
1116         trace_seq_puts(s, " <- ");
1117
1118         if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1119                 goto out;
1120
1121         trace_seq_putc(s, ')');
1122
1123         data = (u8 *)&field[1];
1124         for (i = 0; i < tp->nr_args; i++)
1125                 if (!tp->args[i].type->print(s, tp->args[i].name,
1126                                              data + tp->args[i].offset, field))
1127                         goto out;
1128
1129         trace_seq_putc(s, '\n');
1130
1131  out:
1132         return trace_handle_return(s);
1133 }
1134
1135
1136 static int kprobe_event_define_fields(struct trace_event_call *event_call)
1137 {
1138         int ret, i;
1139         struct kprobe_trace_entry_head field;
1140         struct trace_kprobe *tk = (struct trace_kprobe *)event_call->data;
1141
1142         DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1143         /* Set argument names as fields */
1144         for (i = 0; i < tk->tp.nr_args; i++) {
1145                 struct probe_arg *parg = &tk->tp.args[i];
1146
1147                 ret = trace_define_field(event_call, parg->type->fmttype,
1148                                          parg->name,
1149                                          sizeof(field) + parg->offset,
1150                                          parg->type->size,
1151                                          parg->type->is_signed,
1152                                          FILTER_OTHER);
1153                 if (ret)
1154                         return ret;
1155         }
1156         return 0;
1157 }
1158
1159 static int kretprobe_event_define_fields(struct trace_event_call *event_call)
1160 {
1161         int ret, i;
1162         struct kretprobe_trace_entry_head field;
1163         struct trace_kprobe *tk = (struct trace_kprobe *)event_call->data;
1164
1165         DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1166         DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1167         /* Set argument names as fields */
1168         for (i = 0; i < tk->tp.nr_args; i++) {
1169                 struct probe_arg *parg = &tk->tp.args[i];
1170
1171                 ret = trace_define_field(event_call, parg->type->fmttype,
1172                                          parg->name,
1173                                          sizeof(field) + parg->offset,
1174                                          parg->type->size,
1175                                          parg->type->is_signed,
1176                                          FILTER_OTHER);
1177                 if (ret)
1178                         return ret;
1179         }
1180         return 0;
1181 }
1182
1183 #ifdef CONFIG_PERF_EVENTS
1184
1185 /* Kprobe profile handler */
1186 static void
1187 kprobe_perf_func(struct trace_kprobe *tk, struct pt_regs *regs)
1188 {
1189         struct trace_event_call *call = &tk->tp.call;
1190         struct bpf_prog *prog = call->prog;
1191         struct kprobe_trace_entry_head *entry;
1192         struct hlist_head *head;
1193         int size, __size, dsize;
1194         int rctx;
1195
1196         if (prog && !trace_call_bpf(prog, regs))
1197                 return;
1198
1199         head = this_cpu_ptr(call->perf_events);
1200         if (hlist_empty(head))
1201                 return;
1202
1203         dsize = __get_data_size(&tk->tp, regs);
1204         __size = sizeof(*entry) + tk->tp.size + dsize;
1205         size = ALIGN(__size + sizeof(u32), sizeof(u64));
1206         size -= sizeof(u32);
1207
1208         entry = perf_trace_buf_alloc(size, NULL, &rctx);
1209         if (!entry)
1210                 return;
1211
1212         entry->ip = (unsigned long)tk->rp.kp.addr;
1213         memset(&entry[1], 0, dsize);
1214         store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1215         perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1216                               head, NULL, NULL);
1217 }
1218 NOKPROBE_SYMBOL(kprobe_perf_func);
1219
1220 /* Kretprobe profile handler */
1221 static void
1222 kretprobe_perf_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1223                     struct pt_regs *regs)
1224 {
1225         struct trace_event_call *call = &tk->tp.call;
1226         struct bpf_prog *prog = call->prog;
1227         struct kretprobe_trace_entry_head *entry;
1228         struct hlist_head *head;
1229         int size, __size, dsize;
1230         int rctx;
1231
1232         if (prog && !trace_call_bpf(prog, regs))
1233                 return;
1234
1235         head = this_cpu_ptr(call->perf_events);
1236         if (hlist_empty(head))
1237                 return;
1238
1239         dsize = __get_data_size(&tk->tp, regs);
1240         __size = sizeof(*entry) + tk->tp.size + dsize;
1241         size = ALIGN(__size + sizeof(u32), sizeof(u64));
1242         size -= sizeof(u32);
1243
1244         entry = perf_trace_buf_alloc(size, NULL, &rctx);
1245         if (!entry)
1246                 return;
1247
1248         entry->func = (unsigned long)tk->rp.kp.addr;
1249         entry->ret_ip = (unsigned long)ri->ret_addr;
1250         store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1251         perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1252                               head, NULL, NULL);
1253 }
1254 NOKPROBE_SYMBOL(kretprobe_perf_func);
1255 #endif  /* CONFIG_PERF_EVENTS */
1256
1257 /*
1258  * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex.
1259  *
1260  * kprobe_trace_self_tests_init() does enable_trace_probe/disable_trace_probe
1261  * lockless, but we can't race with this __init function.
1262  */
1263 static int kprobe_register(struct trace_event_call *event,
1264                            enum trace_reg type, void *data)
1265 {
1266         struct trace_kprobe *tk = (struct trace_kprobe *)event->data;
1267         struct trace_event_file *file = data;
1268
1269         switch (type) {
1270         case TRACE_REG_REGISTER:
1271                 return enable_trace_kprobe(tk, file);
1272         case TRACE_REG_UNREGISTER:
1273                 return disable_trace_kprobe(tk, file);
1274
1275 #ifdef CONFIG_PERF_EVENTS
1276         case TRACE_REG_PERF_REGISTER:
1277                 return enable_trace_kprobe(tk, NULL);
1278         case TRACE_REG_PERF_UNREGISTER:
1279                 return disable_trace_kprobe(tk, NULL);
1280         case TRACE_REG_PERF_OPEN:
1281         case TRACE_REG_PERF_CLOSE:
1282         case TRACE_REG_PERF_ADD:
1283         case TRACE_REG_PERF_DEL:
1284                 return 0;
1285 #endif
1286         }
1287         return 0;
1288 }
1289
1290 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1291 {
1292         struct trace_kprobe *tk = container_of(kp, struct trace_kprobe, rp.kp);
1293
1294         raw_cpu_inc(*tk->nhit);
1295
1296         if (tk->tp.flags & TP_FLAG_TRACE)
1297                 kprobe_trace_func(tk, regs);
1298 #ifdef CONFIG_PERF_EVENTS
1299         if (tk->tp.flags & TP_FLAG_PROFILE)
1300                 kprobe_perf_func(tk, regs);
1301 #endif
1302         return 0;       /* We don't tweek kernel, so just return 0 */
1303 }
1304 NOKPROBE_SYMBOL(kprobe_dispatcher);
1305
1306 static int
1307 kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1308 {
1309         struct trace_kprobe *tk = container_of(ri->rp, struct trace_kprobe, rp);
1310
1311         raw_cpu_inc(*tk->nhit);
1312
1313         if (tk->tp.flags & TP_FLAG_TRACE)
1314                 kretprobe_trace_func(tk, ri, regs);
1315 #ifdef CONFIG_PERF_EVENTS
1316         if (tk->tp.flags & TP_FLAG_PROFILE)
1317                 kretprobe_perf_func(tk, ri, regs);
1318 #endif
1319         return 0;       /* We don't tweek kernel, so just return 0 */
1320 }
1321 NOKPROBE_SYMBOL(kretprobe_dispatcher);
1322
1323 static struct trace_event_functions kretprobe_funcs = {
1324         .trace          = print_kretprobe_event
1325 };
1326
1327 static struct trace_event_functions kprobe_funcs = {
1328         .trace          = print_kprobe_event
1329 };
1330
1331 static int register_kprobe_event(struct trace_kprobe *tk)
1332 {
1333         struct trace_event_call *call = &tk->tp.call;
1334         int ret;
1335
1336         /* Initialize trace_event_call */
1337         INIT_LIST_HEAD(&call->class->fields);
1338         if (trace_kprobe_is_return(tk)) {
1339                 call->event.funcs = &kretprobe_funcs;
1340                 call->class->define_fields = kretprobe_event_define_fields;
1341         } else {
1342                 call->event.funcs = &kprobe_funcs;
1343                 call->class->define_fields = kprobe_event_define_fields;
1344         }
1345         if (set_print_fmt(&tk->tp, trace_kprobe_is_return(tk)) < 0)
1346                 return -ENOMEM;
1347         ret = register_trace_event(&call->event);
1348         if (!ret) {
1349                 kfree(call->print_fmt);
1350                 return -ENODEV;
1351         }
1352         call->flags = TRACE_EVENT_FL_KPROBE;
1353         call->class->reg = kprobe_register;
1354         call->data = tk;
1355         ret = trace_add_event_call(call);
1356         if (ret) {
1357                 pr_info("Failed to register kprobe event: %s\n",
1358                         trace_event_name(call));
1359                 kfree(call->print_fmt);
1360                 unregister_trace_event(&call->event);
1361         }
1362         return ret;
1363 }
1364
1365 static int unregister_kprobe_event(struct trace_kprobe *tk)
1366 {
1367         int ret;
1368
1369         /* tp->event is unregistered in trace_remove_event_call() */
1370         ret = trace_remove_event_call(&tk->tp.call);
1371         if (!ret)
1372                 kfree(tk->tp.call.print_fmt);
1373         return ret;
1374 }
1375
1376 /* Make a tracefs interface for controlling probe points */
1377 static __init int init_kprobe_trace(void)
1378 {
1379         struct dentry *d_tracer;
1380         struct dentry *entry;
1381
1382         if (register_module_notifier(&trace_kprobe_module_nb))
1383                 return -EINVAL;
1384
1385         d_tracer = tracing_init_dentry();
1386         if (IS_ERR(d_tracer))
1387                 return 0;
1388
1389         entry = tracefs_create_file("kprobe_events", 0644, d_tracer,
1390                                     NULL, &kprobe_events_ops);
1391
1392         /* Event list interface */
1393         if (!entry)
1394                 pr_warn("Could not create tracefs 'kprobe_events' entry\n");
1395
1396         /* Profile interface */
1397         entry = tracefs_create_file("kprobe_profile", 0444, d_tracer,
1398                                     NULL, &kprobe_profile_ops);
1399
1400         if (!entry)
1401                 pr_warn("Could not create tracefs 'kprobe_profile' entry\n");
1402         return 0;
1403 }
1404 fs_initcall(init_kprobe_trace);
1405
1406
1407 #ifdef CONFIG_FTRACE_STARTUP_TEST
1408 /*
1409  * The "__used" keeps gcc from removing the function symbol
1410  * from the kallsyms table. 'noinline' makes sure that there
1411  * isn't an inlined version used by the test method below
1412  */
1413 static __used __init noinline int
1414 kprobe_trace_selftest_target(int a1, int a2, int a3, int a4, int a5, int a6)
1415 {
1416         return a1 + a2 + a3 + a4 + a5 + a6;
1417 }
1418
1419 static __init struct trace_event_file *
1420 find_trace_probe_file(struct trace_kprobe *tk, struct trace_array *tr)
1421 {
1422         struct trace_event_file *file;
1423
1424         list_for_each_entry(file, &tr->events, list)
1425                 if (file->event_call == &tk->tp.call)
1426                         return file;
1427
1428         return NULL;
1429 }
1430
1431 /*
1432  * Nobody but us can call enable_trace_kprobe/disable_trace_kprobe at this
1433  * stage, we can do this lockless.
1434  */
1435 static __init int kprobe_trace_self_tests_init(void)
1436 {
1437         int ret, warn = 0;
1438         int (*target)(int, int, int, int, int, int);
1439         struct trace_kprobe *tk;
1440         struct trace_event_file *file;
1441
1442         if (tracing_is_disabled())
1443                 return -ENODEV;
1444
1445         target = kprobe_trace_selftest_target;
1446
1447         pr_info("Testing kprobe tracing: ");
1448
1449         ret = traceprobe_command("p:testprobe kprobe_trace_selftest_target "
1450                                   "$stack $stack0 +0($stack)",
1451                                   create_trace_kprobe);
1452         if (WARN_ON_ONCE(ret)) {
1453                 pr_warn("error on probing function entry.\n");
1454                 warn++;
1455         } else {
1456                 /* Enable trace point */
1457                 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1458                 if (WARN_ON_ONCE(tk == NULL)) {
1459                         pr_warn("error on getting new probe.\n");
1460                         warn++;
1461                 } else {
1462                         file = find_trace_probe_file(tk, top_trace_array());
1463                         if (WARN_ON_ONCE(file == NULL)) {
1464                                 pr_warn("error on getting probe file.\n");
1465                                 warn++;
1466                         } else
1467                                 enable_trace_kprobe(tk, file);
1468                 }
1469         }
1470
1471         ret = traceprobe_command("r:testprobe2 kprobe_trace_selftest_target "
1472                                   "$retval", create_trace_kprobe);
1473         if (WARN_ON_ONCE(ret)) {
1474                 pr_warn("error on probing function return.\n");
1475                 warn++;
1476         } else {
1477                 /* Enable trace point */
1478                 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1479                 if (WARN_ON_ONCE(tk == NULL)) {
1480                         pr_warn("error on getting 2nd new probe.\n");
1481                         warn++;
1482                 } else {
1483                         file = find_trace_probe_file(tk, top_trace_array());
1484                         if (WARN_ON_ONCE(file == NULL)) {
1485                                 pr_warn("error on getting probe file.\n");
1486                                 warn++;
1487                         } else
1488                                 enable_trace_kprobe(tk, file);
1489                 }
1490         }
1491
1492         if (warn)
1493                 goto end;
1494
1495         ret = target(1, 2, 3, 4, 5, 6);
1496
1497         /*
1498          * Not expecting an error here, the check is only to prevent the
1499          * optimizer from removing the call to target() as otherwise there
1500          * are no side-effects and the call is never performed.
1501          */
1502         if (ret != 21)
1503                 warn++;
1504
1505         /* Disable trace points before removing it */
1506         tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1507         if (WARN_ON_ONCE(tk == NULL)) {
1508                 pr_warn("error on getting test probe.\n");
1509                 warn++;
1510         } else {
1511                 if (trace_kprobe_nhit(tk) != 1) {
1512                         pr_warn("incorrect number of testprobe hits\n");
1513                         warn++;
1514                 }
1515
1516                 file = find_trace_probe_file(tk, top_trace_array());
1517                 if (WARN_ON_ONCE(file == NULL)) {
1518                         pr_warn("error on getting probe file.\n");
1519                         warn++;
1520                 } else
1521                         disable_trace_kprobe(tk, file);
1522         }
1523
1524         tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1525         if (WARN_ON_ONCE(tk == NULL)) {
1526                 pr_warn("error on getting 2nd test probe.\n");
1527                 warn++;
1528         } else {
1529                 if (trace_kprobe_nhit(tk) != 1) {
1530                         pr_warn("incorrect number of testprobe2 hits\n");
1531                         warn++;
1532                 }
1533
1534                 file = find_trace_probe_file(tk, top_trace_array());
1535                 if (WARN_ON_ONCE(file == NULL)) {
1536                         pr_warn("error on getting probe file.\n");
1537                         warn++;
1538                 } else
1539                         disable_trace_kprobe(tk, file);
1540         }
1541
1542         ret = traceprobe_command("-:testprobe", create_trace_kprobe);
1543         if (WARN_ON_ONCE(ret)) {
1544                 pr_warn("error on deleting a probe.\n");
1545                 warn++;
1546         }
1547
1548         ret = traceprobe_command("-:testprobe2", create_trace_kprobe);
1549         if (WARN_ON_ONCE(ret)) {
1550                 pr_warn("error on deleting a probe.\n");
1551                 warn++;
1552         }
1553
1554 end:
1555         release_all_trace_kprobes();
1556         /*
1557          * Wait for the optimizer work to finish. Otherwise it might fiddle
1558          * with probes in already freed __init text.
1559          */
1560         wait_for_kprobe_optimizer();
1561         if (warn)
1562                 pr_cont("NG: Some tests are failed. Please check them.\n");
1563         else
1564                 pr_cont("OK\n");
1565         return 0;
1566 }
1567
1568 late_initcall(kprobe_trace_self_tests_init);
1569
1570 #endif