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 #include <linux/error-injection.h>
  25
  26 #include "trace_probe.h"
  27
  28 #define KPROBE_EVENT_SYSTEM "kprobes"
  29 #define KRETPROBE_MAXACTIVE_MAX 4096
  30
  31 /**
  32  * Kprobe event core functions
  33  */
  34 struct trace_kprobe {
  35         struct list_head        list;
  36         struct kretprobe        rp;     /* Use rp.kp for kprobe use */
  37         unsigned long __percpu *nhit;
  38         const char              *symbol;        /* symbol name */
  39         struct trace_probe      tp;
  40 };
  41
  42 #define SIZEOF_TRACE_KPROBE(n)                          \
  43         (offsetof(struct trace_kprobe, tp.args) +       \
  44         (sizeof(struct probe_arg) * (n)))
  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 bool trace_kprobe_on_func_entry(struct trace_event_call *call)
  91 {
  92         struct trace_kprobe *tk = (struct trace_kprobe *)call->data;
  93
  94         return kprobe_on_func_entry(tk->rp.kp.addr,
  95                         tk->rp.kp.addr ? NULL : tk->rp.kp.symbol_name,
  96                         tk->rp.kp.addr ? 0 : tk->rp.kp.offset);
  97 }
  98
  99 bool trace_kprobe_error_injectable(struct trace_event_call *call)
 100 {
 101         struct trace_kprobe *tk = (struct trace_kprobe *)call->data;
 102         unsigned long addr;
 103
 104         if (tk->symbol) {
 105                 addr = (unsigned long)
 106                         kallsyms_lookup_name(trace_kprobe_symbol(tk));
 107                 addr += tk->rp.kp.offset;
 108         } else {
 109                 addr = (unsigned long)tk->rp.kp.addr;
 110         }
 111         return within_error_injection_list(addr);
 112 }
 113
 114 static int register_kprobe_event(struct trace_kprobe *tk);
 115 static int unregister_kprobe_event(struct trace_kprobe *tk);
 116
 117 static DEFINE_MUTEX(probe_lock);
 118 static LIST_HEAD(probe_list);
 119
 120 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
 121 static int kretprobe_dispatcher(struct kretprobe_instance *ri,
 122                                 struct pt_regs *regs);
 123
 124 /* Memory fetching by symbol */
 125 struct symbol_cache {
 126         char            *symbol;
 127         long            offset;
 128         unsigned long   addr;
 129 };
 130
 131 unsigned long update_symbol_cache(struct symbol_cache *sc)
 132 {
 133         sc->addr = (unsigned long)kallsyms_lookup_name(sc->symbol);
 134
 135         if (sc->addr)
 136                 sc->addr += sc->offset;
 137
 138         return sc->addr;
 139 }
 140
 141 void free_symbol_cache(struct symbol_cache *sc)
 142 {
 143         kfree(sc->symbol);
 144         kfree(sc);
 145 }
 146
 147 struct symbol_cache *alloc_symbol_cache(const char *sym, long offset)
 148 {
 149         struct symbol_cache *sc;
 150
 151         if (!sym || strlen(sym) == 0)
 152                 return NULL;
 153
 154         sc = kzalloc(sizeof(struct symbol_cache), GFP_KERNEL);
 155         if (!sc)
 156                 return NULL;
 157
 158         sc->symbol = kstrdup(sym, GFP_KERNEL);
 159         if (!sc->symbol) {
 160                 kfree(sc);
 161                 return NULL;
 162         }
 163         sc->offset = offset;
 164         update_symbol_cache(sc);
 165
 166         return sc;
 167 }
 168
 169 /*
 170  * Kprobes-specific fetch functions
 171  */
 172 #define DEFINE_FETCH_stack(type)                                        \
 173 static void FETCH_FUNC_NAME(stack, type)(struct pt_regs *regs,          \
 174                                           void *offset, void *dest)     \
 175 {                                                                       \
 176         *(type *)dest = (type)regs_get_kernel_stack_nth(regs,           \
 177                                 (unsigned int)((unsigned long)offset)); \
 178 }                                                                       \
 179 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(stack, type));
 180
 181 DEFINE_BASIC_FETCH_FUNCS(stack)
 182 /* No string on the stack entry */
 183 #define fetch_stack_string      NULL
 184 #define fetch_stack_string_size NULL
 185
 186 #define DEFINE_FETCH_memory(type)                                       \
 187 static void FETCH_FUNC_NAME(memory, type)(struct pt_regs *regs,         \
 188                                           void *addr, void *dest)       \
 189 {                                                                       \
 190         type retval;                                                    \
 191         if (probe_kernel_address(addr, retval))                         \
 192                 *(type *)dest = 0;                                      \
 193         else                                                            \
 194                 *(type *)dest = retval;                                 \
 195 }                                                                       \
 196 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(memory, type));
 197
 198 DEFINE_BASIC_FETCH_FUNCS(memory)
 199 /*
 200  * Fetch a null-terminated string. Caller MUST set *(u32 *)dest with max
 201  * length and relative data location.
 202  */
 203 static void FETCH_FUNC_NAME(memory, string)(struct pt_regs *regs,
 204                                             void *addr, void *dest)
 205 {
 206         int maxlen = get_rloc_len(*(u32 *)dest);
 207         u8 *dst = get_rloc_data(dest);
 208         long ret;
 209
 210         if (!maxlen)
 211                 return;
 212
 213         /*
 214          * Try to get string again, since the string can be changed while
 215          * probing.
 216          */
 217         ret = strncpy_from_unsafe(dst, addr, maxlen);
 218
 219         if (ret < 0) {  /* Failed to fetch string */
 220                 dst[0] = '\0';
 221                 *(u32 *)dest = make_data_rloc(0, get_rloc_offs(*(u32 *)dest));
 222         } else {
 223                 *(u32 *)dest = make_data_rloc(ret, get_rloc_offs(*(u32 *)dest));
 224         }
 225 }
 226 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(memory, string));
 227
 228 /* Return the length of string -- including null terminal byte */
 229 static void FETCH_FUNC_NAME(memory, string_size)(struct pt_regs *regs,
 230                                                  void *addr, void *dest)
 231 {
 232         mm_segment_t old_fs;
 233         int ret, len = 0;
 234         u8 c;
 235
 236         old_fs = get_fs();
 237         set_fs(KERNEL_DS);
 238         pagefault_disable();
 239
 240         do {
 241                 ret = __copy_from_user_inatomic(&c, (u8 *)addr + len, 1);
 242                 len++;
 243         } while (c && ret == 0 && len < MAX_STRING_SIZE);
 244
 245         pagefault_enable();
 246         set_fs(old_fs);
 247
 248         if (ret < 0)    /* Failed to check the length */
 249                 *(u32 *)dest = 0;
 250         else
 251                 *(u32 *)dest = len;
 252 }
 253 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(memory, string_size));
 254
 255 #define DEFINE_FETCH_symbol(type)                                       \
 256 void FETCH_FUNC_NAME(symbol, type)(struct pt_regs *regs, void *data, void *dest)\
 257 {                                                                       \
 258         struct symbol_cache *sc = data;                                 \
 259         if (sc->addr)                                                   \
 260                 fetch_memory_##type(regs, (void *)sc->addr, dest);      \
 261         else                                                            \
 262                 *(type *)dest = 0;                                      \
 263 }                                                                       \
 264 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(symbol, type));
 265
 266 DEFINE_BASIC_FETCH_FUNCS(symbol)
 267 DEFINE_FETCH_symbol(string)
 268 DEFINE_FETCH_symbol(string_size)
 269
 270 /* kprobes don't support file_offset fetch methods */
 271 #define fetch_file_offset_u8            NULL
 272 #define fetch_file_offset_u16           NULL
 273 #define fetch_file_offset_u32           NULL
 274 #define fetch_file_offset_u64           NULL
 275 #define fetch_file_offset_string        NULL
 276 #define fetch_file_offset_string_size   NULL
 277
 278 /* Fetch type information table */
 279 static const struct fetch_type kprobes_fetch_type_table[] = {
 280         /* Special types */
 281         [FETCH_TYPE_STRING] = __ASSIGN_FETCH_TYPE("string", string, string,
 282                                         sizeof(u32), 1, "__data_loc char[]"),
 283         [FETCH_TYPE_STRSIZE] = __ASSIGN_FETCH_TYPE("string_size", u32,
 284                                         string_size, sizeof(u32), 0, "u32"),
 285         /* Basic types */
 286         ASSIGN_FETCH_TYPE(u8,  u8,  0),
 287         ASSIGN_FETCH_TYPE(u16, u16, 0),
 288         ASSIGN_FETCH_TYPE(u32, u32, 0),
 289         ASSIGN_FETCH_TYPE(u64, u64, 0),
 290         ASSIGN_FETCH_TYPE(s8,  u8,  1),
 291         ASSIGN_FETCH_TYPE(s16, u16, 1),
 292         ASSIGN_FETCH_TYPE(s32, u32, 1),
 293         ASSIGN_FETCH_TYPE(s64, u64, 1),
 294         ASSIGN_FETCH_TYPE_ALIAS(x8,  u8,  u8,  0),
 295         ASSIGN_FETCH_TYPE_ALIAS(x16, u16, u16, 0),
 296         ASSIGN_FETCH_TYPE_ALIAS(x32, u32, u32, 0),
 297         ASSIGN_FETCH_TYPE_ALIAS(x64, u64, u64, 0),
 298
 299         ASSIGN_FETCH_TYPE_END
 300 };
 301
 302 /*
 303  * Allocate new trace_probe and initialize it (including kprobes).
 304  */
 305 static struct trace_kprobe *alloc_trace_kprobe(const char *group,
 306                                              const char *event,
 307                                              void *addr,
 308                                              const char *symbol,
 309                                              unsigned long offs,
 310                                              int maxactive,
 311                                              int nargs, bool is_return)
 312 {
 313         struct trace_kprobe *tk;
 314         int ret = -ENOMEM;
 315
 316         tk = kzalloc(SIZEOF_TRACE_KPROBE(nargs), GFP_KERNEL);
 317         if (!tk)
 318                 return ERR_PTR(ret);
 319
 320         tk->nhit = alloc_percpu(unsigned long);
 321         if (!tk->nhit)
 322                 goto error;
 323
 324         if (symbol) {
 325                 tk->symbol = kstrdup(symbol, GFP_KERNEL);
 326                 if (!tk->symbol)
 327                         goto error;
 328                 tk->rp.kp.symbol_name = tk->symbol;
 329                 tk->rp.kp.offset = offs;
 330         } else
 331                 tk->rp.kp.addr = addr;
 332
 333         if (is_return)
 334                 tk->rp.handler = kretprobe_dispatcher;
 335         else
 336                 tk->rp.kp.pre_handler = kprobe_dispatcher;
 337
 338         tk->rp.maxactive = maxactive;
 339
 340         if (!event || !is_good_name(event)) {
 341                 ret = -EINVAL;
 342                 goto error;
 343         }
 344
 345         tk->tp.call.class = &tk->tp.class;
 346         tk->tp.call.name = kstrdup(event, GFP_KERNEL);
 347         if (!tk->tp.call.name)
 348                 goto error;
 349
 350         if (!group || !is_good_name(group)) {
 351                 ret = -EINVAL;
 352                 goto error;
 353         }
 354
 355         tk->tp.class.system = kstrdup(group, GFP_KERNEL);
 356         if (!tk->tp.class.system)
 357                 goto error;
 358
 359         INIT_LIST_HEAD(&tk->list);
 360         INIT_LIST_HEAD(&tk->tp.files);
 361         return tk;
 362 error:
 363         kfree(tk->tp.call.name);
 364         kfree(tk->symbol);
 365         free_percpu(tk->nhit);
 366         kfree(tk);
 367         return ERR_PTR(ret);
 368 }
 369
 370 static void free_trace_kprobe(struct trace_kprobe *tk)
 371 {
 372         int i;
 373
 374         for (i = 0; i < tk->tp.nr_args; i++)
 375                 traceprobe_free_probe_arg(&tk->tp.args[i]);
 376
 377         kfree(tk->tp.call.class->system);
 378         kfree(tk->tp.call.name);
 379         kfree(tk->symbol);
 380         free_percpu(tk->nhit);
 381         kfree(tk);
 382 }
 383
 384 static struct trace_kprobe *find_trace_kprobe(const char *event,
 385                                               const char *group)
 386 {
 387         struct trace_kprobe *tk;
 388
 389         list_for_each_entry(tk, &probe_list, list)
 390                 if (strcmp(trace_event_name(&tk->tp.call), event) == 0 &&
 391                     strcmp(tk->tp.call.class->system, group) == 0)
 392                         return tk;
 393         return NULL;
 394 }
 395
 396 /*
 397  * Enable trace_probe
 398  * if the file is NULL, enable "perf" handler, or enable "trace" handler.
 399  */
 400 static int
 401 enable_trace_kprobe(struct trace_kprobe *tk, struct trace_event_file *file)
 402 {
 403         int ret = 0;
 404
 405         if (file) {
 406                 struct event_file_link *link;
 407
 408                 link = kmalloc(sizeof(*link), GFP_KERNEL);
 409                 if (!link) {
 410                         ret = -ENOMEM;
 411                         goto out;
 412                 }
 413
 414                 link->file = file;
 415                 list_add_tail_rcu(&link->list, &tk->tp.files);
 416
 417                 tk->tp.flags |= TP_FLAG_TRACE;
 418         } else
 419                 tk->tp.flags |= TP_FLAG_PROFILE;
 420
 421         if (trace_probe_is_registered(&tk->tp) && !trace_kprobe_has_gone(tk)) {
 422                 if (trace_kprobe_is_return(tk))
 423                         ret = enable_kretprobe(&tk->rp);
 424                 else
 425                         ret = enable_kprobe(&tk->rp.kp);
 426         }
 427  out:
 428         return ret;
 429 }
 430
 431 /*
 432  * Disable trace_probe
 433  * if the file is NULL, disable "perf" handler, or disable "trace" handler.
 434  */
 435 static int
 436 disable_trace_kprobe(struct trace_kprobe *tk, struct trace_event_file *file)
 437 {
 438         struct event_file_link *link = NULL;
 439         int wait = 0;
 440         int ret = 0;
 441
 442         if (file) {
 443                 link = find_event_file_link(&tk->tp, file);
 444                 if (!link) {
 445                         ret = -EINVAL;
 446                         goto out;
 447                 }
 448
 449                 list_del_rcu(&link->list);
 450                 wait = 1;
 451                 if (!list_empty(&tk->tp.files))
 452                         goto out;
 453
 454                 tk->tp.flags &= ~TP_FLAG_TRACE;
 455         } else
 456                 tk->tp.flags &= ~TP_FLAG_PROFILE;
 457
 458         if (!trace_probe_is_enabled(&tk->tp) && trace_probe_is_registered(&tk->tp)) {
 459                 if (trace_kprobe_is_return(tk))
 460                         disable_kretprobe(&tk->rp);
 461                 else
 462                         disable_kprobe(&tk->rp.kp);
 463                 wait = 1;
 464         }
 465  out:
 466         if (wait) {
 467                 /*
 468                  * Synchronize with kprobe_trace_func/kretprobe_trace_func
 469                  * to ensure disabled (all running handlers are finished).
 470                  * This is not only for kfree(), but also the caller,
 471                  * trace_remove_event_call() supposes it for releasing
 472                  * event_call related objects, which will be accessed in
 473                  * the kprobe_trace_func/kretprobe_trace_func.
 474                  */
 475                 synchronize_sched();
 476                 kfree(link);    /* Ignored if link == NULL */
 477         }
 478
 479         return ret;
 480 }
 481
 482 /* Internal register function - just handle k*probes and flags */
 483 static int __register_trace_kprobe(struct trace_kprobe *tk)
 484 {
 485         int i, ret;
 486
 487         if (trace_probe_is_registered(&tk->tp))
 488                 return -EINVAL;
 489
 490         for (i = 0; i < tk->tp.nr_args; i++)
 491                 traceprobe_update_arg(&tk->tp.args[i]);
 492
 493         /* Set/clear disabled flag according to tp->flag */
 494         if (trace_probe_is_enabled(&tk->tp))
 495                 tk->rp.kp.flags &= ~KPROBE_FLAG_DISABLED;
 496         else
 497                 tk->rp.kp.flags |= KPROBE_FLAG_DISABLED;
 498
 499         if (trace_kprobe_is_return(tk))
 500                 ret = register_kretprobe(&tk->rp);
 501         else
 502                 ret = register_kprobe(&tk->rp.kp);
 503
 504         if (ret == 0)
 505                 tk->tp.flags |= TP_FLAG_REGISTERED;
 506         else {
 507                 pr_warn("Could not insert probe at %s+%lu: %d\n",
 508                         trace_kprobe_symbol(tk), trace_kprobe_offset(tk), ret);
 509                 if (ret == -ENOENT && trace_kprobe_is_on_module(tk)) {
 510                         pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
 511                         ret = 0;
 512                 } else if (ret == -EILSEQ) {
 513                         pr_warn("Probing address(0x%p) is not an instruction boundary.\n",
 514                                 tk->rp.kp.addr);
 515                         ret = -EINVAL;
 516                 }
 517         }
 518
 519         return ret;
 520 }
 521
 522 /* Internal unregister function - just handle k*probes and flags */
 523 static void __unregister_trace_kprobe(struct trace_kprobe *tk)
 524 {
 525         if (trace_probe_is_registered(&tk->tp)) {
 526                 if (trace_kprobe_is_return(tk))
 527                         unregister_kretprobe(&tk->rp);
 528                 else
 529                         unregister_kprobe(&tk->rp.kp);
 530                 tk->tp.flags &= ~TP_FLAG_REGISTERED;
 531                 /* Cleanup kprobe for reuse */
 532                 if (tk->rp.kp.symbol_name)
 533                         tk->rp.kp.addr = NULL;
 534         }
 535 }
 536
 537 /* Unregister a trace_probe and probe_event: call with locking probe_lock */
 538 static int unregister_trace_kprobe(struct trace_kprobe *tk)
 539 {
 540         /* Enabled event can not be unregistered */
 541         if (trace_probe_is_enabled(&tk->tp))
 542                 return -EBUSY;
 543
 544         /* Will fail if probe is being used by ftrace or perf */
 545         if (unregister_kprobe_event(tk))
 546                 return -EBUSY;
 547
 548         __unregister_trace_kprobe(tk);
 549         list_del(&tk->list);
 550
 551         return 0;
 552 }
 553
 554 /* Register a trace_probe and probe_event */
 555 static int register_trace_kprobe(struct trace_kprobe *tk)
 556 {
 557         struct trace_kprobe *old_tk;
 558         int ret;
 559
 560         mutex_lock(&probe_lock);
 561
 562         /* Delete old (same name) event if exist */
 563         old_tk = find_trace_kprobe(trace_event_name(&tk->tp.call),
 564                         tk->tp.call.class->system);
 565         if (old_tk) {
 566                 ret = unregister_trace_kprobe(old_tk);
 567                 if (ret < 0)
 568                         goto end;
 569                 free_trace_kprobe(old_tk);
 570         }
 571
 572         /* Register new event */
 573         ret = register_kprobe_event(tk);
 574         if (ret) {
 575                 pr_warn("Failed to register probe event(%d)\n", ret);
 576                 goto end;
 577         }
 578
 579         /* Register k*probe */
 580         ret = __register_trace_kprobe(tk);
 581         if (ret < 0)
 582                 unregister_kprobe_event(tk);
 583         else
 584                 list_add_tail(&tk->list, &probe_list);
 585
 586 end:
 587         mutex_unlock(&probe_lock);
 588         return ret;
 589 }
 590
 591 /* Module notifier call back, checking event on the module */
 592 static int trace_kprobe_module_callback(struct notifier_block *nb,
 593                                        unsigned long val, void *data)
 594 {
 595         struct module *mod = data;
 596         struct trace_kprobe *tk;
 597         int ret;
 598
 599         if (val != MODULE_STATE_COMING)
 600                 return NOTIFY_DONE;
 601
 602         /* Update probes on coming module */
 603         mutex_lock(&probe_lock);
 604         list_for_each_entry(tk, &probe_list, list) {
 605                 if (trace_kprobe_within_module(tk, mod)) {
 606                         /* Don't need to check busy - this should have gone. */
 607                         __unregister_trace_kprobe(tk);
 608                         ret = __register_trace_kprobe(tk);
 609                         if (ret)
 610                                 pr_warn("Failed to re-register probe %s on %s: %d\n",
 611                                         trace_event_name(&tk->tp.call),
 612                                         mod->name, ret);
 613                 }
 614         }
 615         mutex_unlock(&probe_lock);
 616
 617         return NOTIFY_DONE;
 618 }
 619
 620 static struct notifier_block trace_kprobe_module_nb = {
 621         .notifier_call = trace_kprobe_module_callback,
 622         .priority = 1   /* Invoked after kprobe module callback */
 623 };
 624
 625 /* Convert certain expected symbols into '_' when generating event names */
 626 static inline void sanitize_event_name(char *name)
 627 {
 628         while (*name++ != '\0')
 629                 if (*name == ':' || *name == '.')
 630                         *name = '_';
 631 }
 632
 633 static int create_trace_kprobe(int argc, char **argv)
 634 {
 635         /*
 636          * Argument syntax:
 637          *  - Add kprobe:
 638          *      p[:[GRP/]EVENT] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
 639          *  - Add kretprobe:
 640          *      r[MAXACTIVE][:[GRP/]EVENT] [MOD:]KSYM[+0] [FETCHARGS]
 641          * Fetch args:
 642          *  $retval     : fetch return value
 643          *  $stack      : fetch stack address
 644          *  $stackN     : fetch Nth of stack (N:0-)
 645          *  $comm       : fetch current task comm
 646          *  @ADDR       : fetch memory at ADDR (ADDR should be in kernel)
 647          *  @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
 648          *  %REG        : fetch register REG
 649          * Dereferencing memory fetch:
 650          *  +|-offs(ARG) : fetch memory at ARG +|- offs address.
 651          * Alias name of args:
 652          *  NAME=FETCHARG : set NAME as alias of FETCHARG.
 653          * Type of args:
 654          *  FETCHARG:TYPE : use TYPE instead of unsigned long.
 655          */
 656         struct trace_kprobe *tk;
 657         int i, ret = 0;
 658         bool is_return = false, is_delete = false;
 659         char *symbol = NULL, *event = NULL, *group = NULL;
 660         int maxactive = 0;
 661         char *arg;
 662         long offset = 0;
 663         void *addr = NULL;
 664         char buf[MAX_EVENT_NAME_LEN];
 665
 666         /* argc must be >= 1 */
 667         if (argv[0][0] == 'p')
 668                 is_return = false;
 669         else if (argv[0][0] == 'r')
 670                 is_return = true;
 671         else if (argv[0][0] == '-')
 672                 is_delete = true;
 673         else {
 674                 pr_info("Probe definition must be started with 'p', 'r' or"
 675                         " '-'.\n");
 676                 return -EINVAL;
 677         }
 678
 679         event = strchr(&argv[0][1], ':');
 680         if (event) {
 681                 event[0] = '\0';
 682                 event++;
 683         }
 684         if (is_return && isdigit(argv[0][1])) {
 685                 ret = kstrtouint(&argv[0][1], 0, &maxactive);
 686                 if (ret) {
 687                         pr_info("Failed to parse maxactive.\n");
 688                         return ret;
 689                 }
 690                 /* kretprobes instances are iterated over via a list. The
 691                  * maximum should stay reasonable.
 692                  */
 693                 if (maxactive > KRETPROBE_MAXACTIVE_MAX) {
 694                         pr_info("Maxactive is too big (%d > %d).\n",
 695                                 maxactive, KRETPROBE_MAXACTIVE_MAX);
 696                         return -E2BIG;
 697                 }
 698         }
 699
 700         if (event) {
 701                 if (strchr(event, '/')) {
 702                         group = event;
 703                         event = strchr(group, '/') + 1;
 704                         event[-1] = '\0';
 705                         if (strlen(group) == 0) {
 706                                 pr_info("Group name is not specified\n");
 707                                 return -EINVAL;
 708                         }
 709                 }
 710                 if (strlen(event) == 0) {
 711                         pr_info("Event name is not specified\n");
 712                         return -EINVAL;
 713                 }
 714         }
 715         if (!group)
 716                 group = KPROBE_EVENT_SYSTEM;
 717
 718         if (is_delete) {
 719                 if (!event) {
 720                         pr_info("Delete command needs an event name.\n");
 721                         return -EINVAL;
 722                 }
 723                 mutex_lock(&probe_lock);
 724                 tk = find_trace_kprobe(event, group);
 725                 if (!tk) {
 726                         mutex_unlock(&probe_lock);
 727                         pr_info("Event %s/%s doesn't exist.\n", group, event);
 728                         return -ENOENT;
 729                 }
 730                 /* delete an event */
 731                 ret = unregister_trace_kprobe(tk);
 732                 if (ret == 0)
 733                         free_trace_kprobe(tk);
 734                 mutex_unlock(&probe_lock);
 735                 return ret;
 736         }
 737
 738         if (argc < 2) {
 739                 pr_info("Probe point is not specified.\n");
 740                 return -EINVAL;
 741         }
 742
 743         /* try to parse an address. if that fails, try to read the
 744          * input as a symbol. */
 745         if (kstrtoul(argv[1], 0, (unsigned long *)&addr)) {
 746                 /* a symbol specified */
 747                 symbol = argv[1];
 748                 /* TODO: support .init module functions */
 749                 ret = traceprobe_split_symbol_offset(symbol, &offset);
 750                 if (ret || offset < 0 || offset > UINT_MAX) {
 751                         pr_info("Failed to parse either an address or a symbol.\n");
 752                         return ret;
 753                 }
 754                 if (offset && is_return &&
 755                     !kprobe_on_func_entry(NULL, symbol, offset)) {
 756                         pr_info("Given offset is not valid for return probe.\n");
 757                         return -EINVAL;
 758                 }
 759         }
 760         argc -= 2; argv += 2;
 761
 762         /* setup a probe */
 763         if (!event) {
 764                 /* Make a new event name */
 765                 if (symbol)
 766                         snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
 767                                  is_return ? 'r' : 'p', symbol, offset);
 768                 else
 769                         snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
 770                                  is_return ? 'r' : 'p', addr);
 771                 sanitize_event_name(buf);
 772                 event = buf;
 773         }
 774         tk = alloc_trace_kprobe(group, event, addr, symbol, offset, maxactive,
 775                                argc, is_return);
 776         if (IS_ERR(tk)) {
 777                 pr_info("Failed to allocate trace_probe.(%d)\n",
 778                         (int)PTR_ERR(tk));
 779                 return PTR_ERR(tk);
 780         }
 781
 782         /* parse arguments */
 783         ret = 0;
 784         for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
 785                 struct probe_arg *parg = &tk->tp.args[i];
 786
 787                 /* Increment count for freeing args in error case */
 788                 tk->tp.nr_args++;
 789
 790                 /* Parse argument name */
 791                 arg = strchr(argv[i], '=');
 792                 if (arg) {
 793                         *arg++ = '\0';
 794                         parg->name = kstrdup(argv[i], GFP_KERNEL);
 795                 } else {
 796                         arg = argv[i];
 797                         /* If argument name is omitted, set "argN" */
 798                         snprintf(buf, MAX_EVENT_NAME_LEN, "arg%d", i + 1);
 799                         parg->name = kstrdup(buf, GFP_KERNEL);
 800                 }
 801
 802                 if (!parg->name) {
 803                         pr_info("Failed to allocate argument[%d] name.\n", i);
 804                         ret = -ENOMEM;
 805                         goto error;
 806                 }
 807
 808                 if (!is_good_name(parg->name)) {
 809                         pr_info("Invalid argument[%d] name: %s\n",
 810                                 i, parg->name);
 811                         ret = -EINVAL;
 812                         goto error;
 813                 }
 814
 815                 if (traceprobe_conflict_field_name(parg->name,
 816                                                         tk->tp.args, i)) {
 817                         pr_info("Argument[%d] name '%s' conflicts with "
 818                                 "another field.\n", i, argv[i]);
 819                         ret = -EINVAL;
 820                         goto error;
 821                 }
 822
 823                 /* Parse fetch argument */
 824                 ret = traceprobe_parse_probe_arg(arg, &tk->tp.size, parg,
 825                                                 is_return, true,
 826                                                 kprobes_fetch_type_table);
 827                 if (ret) {
 828                         pr_info("Parse error at argument[%d]. (%d)\n", i, ret);
 829                         goto error;
 830                 }
 831         }
 832
 833         ret = register_trace_kprobe(tk);
 834         if (ret)
 835                 goto error;
 836         return 0;
 837
 838 error:
 839         free_trace_kprobe(tk);
 840         return ret;
 841 }
 842
 843 static int release_all_trace_kprobes(void)
 844 {
 845         struct trace_kprobe *tk;
 846         int ret = 0;
 847
 848         mutex_lock(&probe_lock);
 849         /* Ensure no probe is in use. */
 850         list_for_each_entry(tk, &probe_list, list)
 851                 if (trace_probe_is_enabled(&tk->tp)) {
 852                         ret = -EBUSY;
 853                         goto end;
 854                 }
 855         /* TODO: Use batch unregistration */
 856         while (!list_empty(&probe_list)) {
 857                 tk = list_entry(probe_list.next, struct trace_kprobe, list);
 858                 ret = unregister_trace_kprobe(tk);
 859                 if (ret)
 860                         goto end;
 861                 free_trace_kprobe(tk);
 862         }
 863
 864 end:
 865         mutex_unlock(&probe_lock);
 866
 867         return ret;
 868 }
 869
 870 /* Probes listing interfaces */
 871 static void *probes_seq_start(struct seq_file *m, loff_t *pos)
 872 {
 873         mutex_lock(&probe_lock);
 874         return seq_list_start(&probe_list, *pos);
 875 }
 876
 877 static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
 878 {
 879         return seq_list_next(v, &probe_list, pos);
 880 }
 881
 882 static void probes_seq_stop(struct seq_file *m, void *v)
 883 {
 884         mutex_unlock(&probe_lock);
 885 }
 886
 887 static int probes_seq_show(struct seq_file *m, void *v)
 888 {
 889         struct trace_kprobe *tk = v;
 890         int i;
 891
 892         seq_putc(m, trace_kprobe_is_return(tk) ? 'r' : 'p');
 893         seq_printf(m, ":%s/%s", tk->tp.call.class->system,
 894                         trace_event_name(&tk->tp.call));
 895
 896         if (!tk->symbol)
 897                 seq_printf(m, " 0x%p", tk->rp.kp.addr);
 898         else if (tk->rp.kp.offset)
 899                 seq_printf(m, " %s+%u", trace_kprobe_symbol(tk),
 900                            tk->rp.kp.offset);
 901         else
 902                 seq_printf(m, " %s", trace_kprobe_symbol(tk));
 903
 904         for (i = 0; i < tk->tp.nr_args; i++)
 905                 seq_printf(m, " %s=%s", tk->tp.args[i].name, tk->tp.args[i].comm);
 906         seq_putc(m, '\n');
 907
 908         return 0;
 909 }
 910
 911 static const struct seq_operations probes_seq_op = {
 912         .start  = probes_seq_start,
 913         .next   = probes_seq_next,
 914         .stop   = probes_seq_stop,
 915         .show   = probes_seq_show
 916 };
 917
 918 static int probes_open(struct inode *inode, struct file *file)
 919 {
 920         int ret;
 921
 922         if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
 923                 ret = release_all_trace_kprobes();
 924                 if (ret < 0)
 925                         return ret;
 926         }
 927
 928         return seq_open(file, &probes_seq_op);
 929 }
 930
 931 static ssize_t probes_write(struct file *file, const char __user *buffer,
 932                             size_t count, loff_t *ppos)
 933 {
 934         return trace_parse_run_command(file, buffer, count, ppos,
 935                                        create_trace_kprobe);
 936 }
 937
 938 static const struct file_operations kprobe_events_ops = {
 939         .owner          = THIS_MODULE,
 940         .open           = probes_open,
 941         .read           = seq_read,
 942         .llseek         = seq_lseek,
 943         .release        = seq_release,
 944         .write          = probes_write,
 945 };
 946
 947 /* Probes profiling interfaces */
 948 static int probes_profile_seq_show(struct seq_file *m, void *v)
 949 {
 950         struct trace_kprobe *tk = v;
 951
 952         seq_printf(m, "  %-44s %15lu %15lu\n",
 953                    trace_event_name(&tk->tp.call),
 954                    trace_kprobe_nhit(tk),
 955                    tk->rp.kp.nmissed);
 956
 957         return 0;
 958 }
 959
 960 static const struct seq_operations profile_seq_op = {
 961         .start  = probes_seq_start,
 962         .next   = probes_seq_next,
 963         .stop   = probes_seq_stop,
 964         .show   = probes_profile_seq_show
 965 };
 966
 967 static int profile_open(struct inode *inode, struct file *file)
 968 {
 969         return seq_open(file, &profile_seq_op);
 970 }
 971
 972 static const struct file_operations kprobe_profile_ops = {
 973         .owner          = THIS_MODULE,
 974         .open           = profile_open,
 975         .read           = seq_read,
 976         .llseek         = seq_lseek,
 977         .release        = seq_release,
 978 };
 979
 980 /* Kprobe handler */
 981 static nokprobe_inline void
 982 __kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs,
 983                     struct trace_event_file *trace_file)
 984 {
 985         struct kprobe_trace_entry_head *entry;
 986         struct ring_buffer_event *event;
 987         struct ring_buffer *buffer;
 988         int size, dsize, pc;
 989         unsigned long irq_flags;
 990         struct trace_event_call *call = &tk->tp.call;
 991
 992         WARN_ON(call != trace_file->event_call);
 993
 994         if (trace_trigger_soft_disabled(trace_file))
 995                 return;
 996
 997         local_save_flags(irq_flags);
 998         pc = preempt_count();
 999
1000         dsize = __get_data_size(&tk->tp, regs);
1001         size = sizeof(*entry) + tk->tp.size + dsize;
1002
1003         event = trace_event_buffer_lock_reserve(&buffer, trace_file,
1004                                                 call->event.type,
1005                                                 size, irq_flags, pc);
1006         if (!event)
1007                 return;
1008
1009         entry = ring_buffer_event_data(event);
1010         entry->ip = (unsigned long)tk->rp.kp.addr;
1011         store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1012
1013         event_trigger_unlock_commit_regs(trace_file, buffer, event,
1014                                          entry, irq_flags, pc, regs);
1015 }
1016
1017 static void
1018 kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs)
1019 {
1020         struct event_file_link *link;
1021
1022         list_for_each_entry_rcu(link, &tk->tp.files, list)
1023                 __kprobe_trace_func(tk, regs, link->file);
1024 }
1025 NOKPROBE_SYMBOL(kprobe_trace_func);
1026
1027 /* Kretprobe handler */
1028 static nokprobe_inline void
1029 __kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1030                        struct pt_regs *regs,
1031                        struct trace_event_file *trace_file)
1032 {
1033         struct kretprobe_trace_entry_head *entry;
1034         struct ring_buffer_event *event;
1035         struct ring_buffer *buffer;
1036         int size, pc, dsize;
1037         unsigned long irq_flags;
1038         struct trace_event_call *call = &tk->tp.call;
1039
1040         WARN_ON(call != trace_file->event_call);
1041
1042         if (trace_trigger_soft_disabled(trace_file))
1043                 return;
1044
1045         local_save_flags(irq_flags);
1046         pc = preempt_count();
1047
1048         dsize = __get_data_size(&tk->tp, regs);
1049         size = sizeof(*entry) + tk->tp.size + dsize;
1050
1051         event = trace_event_buffer_lock_reserve(&buffer, trace_file,
1052                                                 call->event.type,
1053                                                 size, irq_flags, pc);
1054         if (!event)
1055                 return;
1056
1057         entry = ring_buffer_event_data(event);
1058         entry->func = (unsigned long)tk->rp.kp.addr;
1059         entry->ret_ip = (unsigned long)ri->ret_addr;
1060         store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1061
1062         event_trigger_unlock_commit_regs(trace_file, buffer, event,
1063                                          entry, irq_flags, pc, regs);
1064 }
1065
1066 static void
1067 kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1068                      struct pt_regs *regs)
1069 {
1070         struct event_file_link *link;
1071
1072         list_for_each_entry_rcu(link, &tk->tp.files, list)
1073                 __kretprobe_trace_func(tk, ri, regs, link->file);
1074 }
1075 NOKPROBE_SYMBOL(kretprobe_trace_func);
1076
1077 /* Event entry printers */
1078 static enum print_line_t
1079 print_kprobe_event(struct trace_iterator *iter, int flags,
1080                    struct trace_event *event)
1081 {
1082         struct kprobe_trace_entry_head *field;
1083         struct trace_seq *s = &iter->seq;
1084         struct trace_probe *tp;
1085         u8 *data;
1086         int i;
1087
1088         field = (struct kprobe_trace_entry_head *)iter->ent;
1089         tp = container_of(event, struct trace_probe, call.event);
1090
1091         trace_seq_printf(s, "%s: (", trace_event_name(&tp->call));
1092
1093         if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
1094                 goto out;
1095
1096         trace_seq_putc(s, ')');
1097
1098         data = (u8 *)&field[1];
1099         for (i = 0; i < tp->nr_args; i++)
1100                 if (!tp->args[i].type->print(s, tp->args[i].name,
1101                                              data + tp->args[i].offset, field))
1102                         goto out;
1103
1104         trace_seq_putc(s, '\n');
1105  out:
1106         return trace_handle_return(s);
1107 }
1108
1109 static enum print_line_t
1110 print_kretprobe_event(struct trace_iterator *iter, int flags,
1111                       struct trace_event *event)
1112 {
1113         struct kretprobe_trace_entry_head *field;
1114         struct trace_seq *s = &iter->seq;
1115         struct trace_probe *tp;
1116         u8 *data;
1117         int i;
1118
1119         field = (struct kretprobe_trace_entry_head *)iter->ent;
1120         tp = container_of(event, struct trace_probe, call.event);
1121
1122         trace_seq_printf(s, "%s: (", trace_event_name(&tp->call));
1123
1124         if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1125                 goto out;
1126
1127         trace_seq_puts(s, " <- ");
1128
1129         if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1130                 goto out;
1131
1132         trace_seq_putc(s, ')');
1133
1134         data = (u8 *)&field[1];
1135         for (i = 0; i < tp->nr_args; i++)
1136                 if (!tp->args[i].type->print(s, tp->args[i].name,
1137                                              data + tp->args[i].offset, field))
1138                         goto out;
1139
1140         trace_seq_putc(s, '\n');
1141
1142  out:
1143         return trace_handle_return(s);
1144 }
1145
1146
1147 static int kprobe_event_define_fields(struct trace_event_call *event_call)
1148 {
1149         int ret, i;
1150         struct kprobe_trace_entry_head field;
1151         struct trace_kprobe *tk = (struct trace_kprobe *)event_call->data;
1152
1153         DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1154         /* Set argument names as fields */
1155         for (i = 0; i < tk->tp.nr_args; i++) {
1156                 struct probe_arg *parg = &tk->tp.args[i];
1157
1158                 ret = trace_define_field(event_call, parg->type->fmttype,
1159                                          parg->name,
1160                                          sizeof(field) + parg->offset,
1161                                          parg->type->size,
1162                                          parg->type->is_signed,
1163                                          FILTER_OTHER);
1164                 if (ret)
1165                         return ret;
1166         }
1167         return 0;
1168 }
1169
1170 static int kretprobe_event_define_fields(struct trace_event_call *event_call)
1171 {
1172         int ret, i;
1173         struct kretprobe_trace_entry_head field;
1174         struct trace_kprobe *tk = (struct trace_kprobe *)event_call->data;
1175
1176         DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1177         DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1178         /* Set argument names as fields */
1179         for (i = 0; i < tk->tp.nr_args; i++) {
1180                 struct probe_arg *parg = &tk->tp.args[i];
1181
1182                 ret = trace_define_field(event_call, parg->type->fmttype,
1183                                          parg->name,
1184                                          sizeof(field) + parg->offset,
1185                                          parg->type->size,
1186                                          parg->type->is_signed,
1187                                          FILTER_OTHER);
1188                 if (ret)
1189                         return ret;
1190         }
1191         return 0;
1192 }
1193
1194 #ifdef CONFIG_PERF_EVENTS
1195
1196 /* Kprobe profile handler */
1197 static int
1198 kprobe_perf_func(struct trace_kprobe *tk, struct pt_regs *regs)
1199 {
1200         struct trace_event_call *call = &tk->tp.call;
1201         struct kprobe_trace_entry_head *entry;
1202         struct hlist_head *head;
1203         int size, __size, dsize;
1204         int rctx;
1205
1206         if (bpf_prog_array_valid(call)) {
1207                 unsigned long orig_ip = instruction_pointer(regs);
1208                 int ret;
1209
1210                 ret = trace_call_bpf(call, regs);
1211
1212                 /*
1213                  * We need to check and see if we modified the pc of the
1214                  * pt_regs, and if so clear the kprobe and return 1 so that we
1215                  * don't do the single stepping.
1216                  * The ftrace kprobe handler leaves it up to us to re-enable
1217                  * preemption here before returning if we've modified the ip.
1218                  */
1219                 if (orig_ip != instruction_pointer(regs)) {
1220                         reset_current_kprobe();
1221                         preempt_enable_no_resched();
1222                         return 1;
1223                 }
1224                 if (!ret)
1225                         return 0;
1226         }
1227
1228         head = this_cpu_ptr(call->perf_events);
1229         if (hlist_empty(head))
1230                 return 0;
1231
1232         dsize = __get_data_size(&tk->tp, regs);
1233         __size = sizeof(*entry) + tk->tp.size + dsize;
1234         size = ALIGN(__size + sizeof(u32), sizeof(u64));
1235         size -= sizeof(u32);
1236
1237         entry = perf_trace_buf_alloc(size, NULL, &rctx);
1238         if (!entry)
1239                 return 0;
1240
1241         entry->ip = (unsigned long)tk->rp.kp.addr;
1242         memset(&entry[1], 0, dsize);
1243         store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1244         perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1245                               head, NULL);
1246         return 0;
1247 }
1248 NOKPROBE_SYMBOL(kprobe_perf_func);
1249
1250 /* Kretprobe profile handler */
1251 static void
1252 kretprobe_perf_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1253                     struct pt_regs *regs)
1254 {
1255         struct trace_event_call *call = &tk->tp.call;
1256         struct kretprobe_trace_entry_head *entry;
1257         struct hlist_head *head;
1258         int size, __size, dsize;
1259         int rctx;
1260
1261         if (bpf_prog_array_valid(call) && !trace_call_bpf(call, regs))
1262                 return;
1263
1264         head = this_cpu_ptr(call->perf_events);
1265         if (hlist_empty(head))
1266                 return;
1267
1268         dsize = __get_data_size(&tk->tp, regs);
1269         __size = sizeof(*entry) + tk->tp.size + dsize;
1270         size = ALIGN(__size + sizeof(u32), sizeof(u64));
1271         size -= sizeof(u32);
1272
1273         entry = perf_trace_buf_alloc(size, NULL, &rctx);
1274         if (!entry)
1275                 return;
1276
1277         entry->func = (unsigned long)tk->rp.kp.addr;
1278         entry->ret_ip = (unsigned long)ri->ret_addr;
1279         store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1280         perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1281                               head, NULL);
1282 }
1283 NOKPROBE_SYMBOL(kretprobe_perf_func);
1284 #endif  /* CONFIG_PERF_EVENTS */
1285
1286 /*
1287  * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex.
1288  *
1289  * kprobe_trace_self_tests_init() does enable_trace_probe/disable_trace_probe
1290  * lockless, but we can't race with this __init function.
1291  */
1292 static int kprobe_register(struct trace_event_call *event,
1293                            enum trace_reg type, void *data)
1294 {
1295         struct trace_kprobe *tk = (struct trace_kprobe *)event->data;
1296         struct trace_event_file *file = data;
1297
1298         switch (type) {
1299         case TRACE_REG_REGISTER:
1300                 return enable_trace_kprobe(tk, file);
1301         case TRACE_REG_UNREGISTER:
1302                 return disable_trace_kprobe(tk, file);
1303
1304 #ifdef CONFIG_PERF_EVENTS
1305         case TRACE_REG_PERF_REGISTER:
1306                 return enable_trace_kprobe(tk, NULL);
1307         case TRACE_REG_PERF_UNREGISTER:
1308                 return disable_trace_kprobe(tk, NULL);
1309         case TRACE_REG_PERF_OPEN:
1310         case TRACE_REG_PERF_CLOSE:
1311         case TRACE_REG_PERF_ADD:
1312         case TRACE_REG_PERF_DEL:
1313                 return 0;
1314 #endif
1315         }
1316         return 0;
1317 }
1318
1319 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1320 {
1321         struct trace_kprobe *tk = container_of(kp, struct trace_kprobe, rp.kp);
1322         int ret = 0;
1323
1324         raw_cpu_inc(*tk->nhit);
1325
1326         if (tk->tp.flags & TP_FLAG_TRACE)
1327                 kprobe_trace_func(tk, regs);
1328 #ifdef CONFIG_PERF_EVENTS
1329         if (tk->tp.flags & TP_FLAG_PROFILE)
1330                 ret = kprobe_perf_func(tk, regs);
1331 #endif
1332         return ret;
1333 }
1334 NOKPROBE_SYMBOL(kprobe_dispatcher);
1335
1336 static int
1337 kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1338 {
1339         struct trace_kprobe *tk = container_of(ri->rp, struct trace_kprobe, rp);
1340
1341         raw_cpu_inc(*tk->nhit);
1342
1343         if (tk->tp.flags & TP_FLAG_TRACE)
1344                 kretprobe_trace_func(tk, ri, regs);
1345 #ifdef CONFIG_PERF_EVENTS
1346         if (tk->tp.flags & TP_FLAG_PROFILE)
1347                 kretprobe_perf_func(tk, ri, regs);
1348 #endif
1349         return 0;       /* We don't tweek kernel, so just return 0 */
1350 }
1351 NOKPROBE_SYMBOL(kretprobe_dispatcher);
1352
1353 static struct trace_event_functions kretprobe_funcs = {
1354         .trace          = print_kretprobe_event
1355 };
1356
1357 static struct trace_event_functions kprobe_funcs = {
1358         .trace          = print_kprobe_event
1359 };
1360
1361 static int register_kprobe_event(struct trace_kprobe *tk)
1362 {
1363         struct trace_event_call *call = &tk->tp.call;
1364         int ret;
1365
1366         /* Initialize trace_event_call */
1367         INIT_LIST_HEAD(&call->class->fields);
1368         if (trace_kprobe_is_return(tk)) {
1369                 call->event.funcs = &kretprobe_funcs;
1370                 call->class->define_fields = kretprobe_event_define_fields;
1371         } else {
1372                 call->event.funcs = &kprobe_funcs;
1373                 call->class->define_fields = kprobe_event_define_fields;
1374         }
1375         if (set_print_fmt(&tk->tp, trace_kprobe_is_return(tk)) < 0)
1376                 return -ENOMEM;
1377         ret = register_trace_event(&call->event);
1378         if (!ret) {
1379                 kfree(call->print_fmt);
1380                 return -ENODEV;
1381         }
1382         call->flags = TRACE_EVENT_FL_KPROBE;
1383         call->class->reg = kprobe_register;
1384         call->data = tk;
1385         ret = trace_add_event_call(call);
1386         if (ret) {
1387                 pr_info("Failed to register kprobe event: %s\n",
1388                         trace_event_name(call));
1389                 kfree(call->print_fmt);
1390                 unregister_trace_event(&call->event);
1391         }
1392         return ret;
1393 }
1394
1395 static int unregister_kprobe_event(struct trace_kprobe *tk)
1396 {
1397         int ret;
1398
1399         /* tp->event is unregistered in trace_remove_event_call() */
1400         ret = trace_remove_event_call(&tk->tp.call);
1401         if (!ret)
1402                 kfree(tk->tp.call.print_fmt);
1403         return ret;
1404 }
1405
1406 /* Make a tracefs interface for controlling probe points */
1407 static __init int init_kprobe_trace(void)
1408 {
1409         struct dentry *d_tracer;
1410         struct dentry *entry;
1411
1412         if (register_module_notifier(&trace_kprobe_module_nb))
1413                 return -EINVAL;
1414
1415         d_tracer = tracing_init_dentry();
1416         if (IS_ERR(d_tracer))
1417                 return 0;
1418
1419         entry = tracefs_create_file("kprobe_events", 0644, d_tracer,
1420                                     NULL, &kprobe_events_ops);
1421
1422         /* Event list interface */
1423         if (!entry)
1424                 pr_warn("Could not create tracefs 'kprobe_events' entry\n");
1425
1426         /* Profile interface */
1427         entry = tracefs_create_file("kprobe_profile", 0444, d_tracer,
1428                                     NULL, &kprobe_profile_ops);
1429
1430         if (!entry)
1431                 pr_warn("Could not create tracefs 'kprobe_profile' entry\n");
1432         return 0;
1433 }
1434 fs_initcall(init_kprobe_trace);
1435
1436
1437 #ifdef CONFIG_FTRACE_STARTUP_TEST
1438 /*
1439  * The "__used" keeps gcc from removing the function symbol
1440  * from the kallsyms table. 'noinline' makes sure that there
1441  * isn't an inlined version used by the test method below
1442  */
1443 static __used __init noinline int
1444 kprobe_trace_selftest_target(int a1, int a2, int a3, int a4, int a5, int a6)
1445 {
1446         return a1 + a2 + a3 + a4 + a5 + a6;
1447 }
1448
1449 static __init struct trace_event_file *
1450 find_trace_probe_file(struct trace_kprobe *tk, struct trace_array *tr)
1451 {
1452         struct trace_event_file *file;
1453
1454         list_for_each_entry(file, &tr->events, list)
1455                 if (file->event_call == &tk->tp.call)
1456                         return file;
1457
1458         return NULL;
1459 }
1460
1461 /*
1462  * Nobody but us can call enable_trace_kprobe/disable_trace_kprobe at this
1463  * stage, we can do this lockless.
1464  */
1465 static __init int kprobe_trace_self_tests_init(void)
1466 {
1467         int ret, warn = 0;
1468         int (*target)(int, int, int, int, int, int);
1469         struct trace_kprobe *tk;
1470         struct trace_event_file *file;
1471
1472         if (tracing_is_disabled())
1473                 return -ENODEV;
1474
1475         target = kprobe_trace_selftest_target;
1476
1477         pr_info("Testing kprobe tracing: ");
1478
1479         ret = trace_run_command("p:testprobe kprobe_trace_selftest_target "
1480                                 "$stack $stack0 +0($stack)",
1481                                 create_trace_kprobe);
1482         if (WARN_ON_ONCE(ret)) {
1483                 pr_warn("error on probing function entry.\n");
1484                 warn++;
1485         } else {
1486                 /* Enable trace point */
1487                 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1488                 if (WARN_ON_ONCE(tk == NULL)) {
1489                         pr_warn("error on getting new probe.\n");
1490                         warn++;
1491                 } else {
1492                         file = find_trace_probe_file(tk, top_trace_array());
1493                         if (WARN_ON_ONCE(file == NULL)) {
1494                                 pr_warn("error on getting probe file.\n");
1495                                 warn++;
1496                         } else
1497                                 enable_trace_kprobe(tk, file);
1498                 }
1499         }
1500
1501         ret = trace_run_command("r:testprobe2 kprobe_trace_selftest_target "
1502                                 "$retval", create_trace_kprobe);
1503         if (WARN_ON_ONCE(ret)) {
1504                 pr_warn("error on probing function return.\n");
1505                 warn++;
1506         } else {
1507                 /* Enable trace point */
1508                 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1509                 if (WARN_ON_ONCE(tk == NULL)) {
1510                         pr_warn("error on getting 2nd new probe.\n");
1511                         warn++;
1512                 } else {
1513                         file = find_trace_probe_file(tk, top_trace_array());
1514                         if (WARN_ON_ONCE(file == NULL)) {
1515                                 pr_warn("error on getting probe file.\n");
1516                                 warn++;
1517                         } else
1518                                 enable_trace_kprobe(tk, file);
1519                 }
1520         }
1521
1522         if (warn)
1523                 goto end;
1524
1525         ret = target(1, 2, 3, 4, 5, 6);
1526
1527         /*
1528          * Not expecting an error here, the check is only to prevent the
1529          * optimizer from removing the call to target() as otherwise there
1530          * are no side-effects and the call is never performed.
1531          */
1532         if (ret != 21)
1533                 warn++;
1534
1535         /* Disable trace points before removing it */
1536         tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1537         if (WARN_ON_ONCE(tk == NULL)) {
1538                 pr_warn("error on getting test probe.\n");
1539                 warn++;
1540         } else {
1541                 if (trace_kprobe_nhit(tk) != 1) {
1542                         pr_warn("incorrect number of testprobe hits\n");
1543                         warn++;
1544                 }
1545
1546                 file = find_trace_probe_file(tk, top_trace_array());
1547                 if (WARN_ON_ONCE(file == NULL)) {
1548                         pr_warn("error on getting probe file.\n");
1549                         warn++;
1550                 } else
1551                         disable_trace_kprobe(tk, file);
1552         }
1553
1554         tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1555         if (WARN_ON_ONCE(tk == NULL)) {
1556                 pr_warn("error on getting 2nd test probe.\n");
1557                 warn++;
1558         } else {
1559                 if (trace_kprobe_nhit(tk) != 1) {
1560                         pr_warn("incorrect number of testprobe2 hits\n");
1561                         warn++;
1562                 }
1563
1564                 file = find_trace_probe_file(tk, top_trace_array());
1565                 if (WARN_ON_ONCE(file == NULL)) {
1566                         pr_warn("error on getting probe file.\n");
1567                         warn++;
1568                 } else
1569                         disable_trace_kprobe(tk, file);
1570         }
1571
1572         ret = trace_run_command("-:testprobe", create_trace_kprobe);
1573         if (WARN_ON_ONCE(ret)) {
1574                 pr_warn("error on deleting a probe.\n");
1575                 warn++;
1576         }
1577
1578         ret = trace_run_command("-:testprobe2", create_trace_kprobe);
1579         if (WARN_ON_ONCE(ret)) {
1580                 pr_warn("error on deleting a probe.\n");
1581                 warn++;
1582         }
1583
1584 end:
1585         release_all_trace_kprobes();
1586         /*
1587          * Wait for the optimizer work to finish. Otherwise it might fiddle
1588          * with probes in already freed __init text.
1589          */
1590         wait_for_kprobe_optimizer();
1591         if (warn)
1592                 pr_cont("NG: Some tests are failed. Please check them.\n");
1593         else
1594                 pr_cont("OK\n");
1595         return 0;
1596 }
1597
1598 late_initcall(kprobe_trace_self_tests_init);
1599
1600 #endif