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         int ret = 0;
 380
 381         if (file) {
 382                 struct event_file_link *link;
 383
 384                 link = kmalloc(sizeof(*link), GFP_KERNEL);
 385                 if (!link) {
 386                         ret = -ENOMEM;
 387                         goto out;
 388                 }
 389
 390                 link->file = file;
 391                 list_add_tail_rcu(&link->list, &tk->tp.files);
 392
 393                 tk->tp.flags |= TP_FLAG_TRACE;
 394         } else
 395                 tk->tp.flags |= TP_FLAG_PROFILE;
 396
 397         if (trace_probe_is_registered(&tk->tp) && !trace_kprobe_has_gone(tk)) {
 398                 if (trace_kprobe_is_return(tk))
 399                         ret = enable_kretprobe(&tk->rp);
 400                 else
 401                         ret = enable_kprobe(&tk->rp.kp);
 402         }
 403  out:
 404         return ret;
 405 }
 406
 407 /*
 408  * Disable trace_probe
 409  * if the file is NULL, disable "perf" handler, or disable "trace" handler.
 410  */
 411 static int
 412 disable_trace_kprobe(struct trace_kprobe *tk, struct trace_event_file *file)
 413 {
 414         struct event_file_link *link = NULL;
 415         int wait = 0;
 416         int ret = 0;
 417
 418         if (file) {
 419                 link = find_event_file_link(&tk->tp, file);
 420                 if (!link) {
 421                         ret = -EINVAL;
 422                         goto out;
 423                 }
 424
 425                 list_del_rcu(&link->list);
 426                 wait = 1;
 427                 if (!list_empty(&tk->tp.files))
 428                         goto out;
 429
 430                 tk->tp.flags &= ~TP_FLAG_TRACE;
 431         } else
 432                 tk->tp.flags &= ~TP_FLAG_PROFILE;
 433
 434         if (!trace_probe_is_enabled(&tk->tp) && trace_probe_is_registered(&tk->tp)) {
 435                 if (trace_kprobe_is_return(tk))
 436                         disable_kretprobe(&tk->rp);
 437                 else
 438                         disable_kprobe(&tk->rp.kp);
 439                 wait = 1;
 440         }
 441  out:
 442         if (wait) {
 443                 /*
 444                  * Synchronize with kprobe_trace_func/kretprobe_trace_func
 445                  * to ensure disabled (all running handlers are finished).
 446                  * This is not only for kfree(), but also the caller,
 447                  * trace_remove_event_call() supposes it for releasing
 448                  * event_call related objects, which will be accessed in
 449                  * the kprobe_trace_func/kretprobe_trace_func.
 450                  */
 451                 synchronize_sched();
 452                 kfree(link);    /* Ignored if link == NULL */
 453         }
 454
 455         return ret;
 456 }
 457
 458 /* Internal register function - just handle k*probes and flags */
 459 static int __register_trace_kprobe(struct trace_kprobe *tk)
 460 {
 461         int i, ret;
 462
 463         if (trace_probe_is_registered(&tk->tp))
 464                 return -EINVAL;
 465
 466         for (i = 0; i < tk->tp.nr_args; i++)
 467                 traceprobe_update_arg(&tk->tp.args[i]);
 468
 469         /* Set/clear disabled flag according to tp->flag */
 470         if (trace_probe_is_enabled(&tk->tp))
 471                 tk->rp.kp.flags &= ~KPROBE_FLAG_DISABLED;
 472         else
 473                 tk->rp.kp.flags |= KPROBE_FLAG_DISABLED;
 474
 475         if (trace_kprobe_is_return(tk))
 476                 ret = register_kretprobe(&tk->rp);
 477         else
 478                 ret = register_kprobe(&tk->rp.kp);
 479
 480         if (ret == 0)
 481                 tk->tp.flags |= TP_FLAG_REGISTERED;
 482         else {
 483                 pr_warn("Could not insert probe at %s+%lu: %d\n",
 484                         trace_kprobe_symbol(tk), trace_kprobe_offset(tk), ret);
 485                 if (ret == -ENOENT && trace_kprobe_is_on_module(tk)) {
 486                         pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
 487                         ret = 0;
 488                 } else if (ret == -EILSEQ) {
 489                         pr_warn("Probing address(0x%p) is not an instruction boundary.\n",
 490                                 tk->rp.kp.addr);
 491                         ret = -EINVAL;
 492                 }
 493         }
 494
 495         return ret;
 496 }
 497
 498 /* Internal unregister function - just handle k*probes and flags */
 499 static void __unregister_trace_kprobe(struct trace_kprobe *tk)
 500 {
 501         if (trace_probe_is_registered(&tk->tp)) {
 502                 if (trace_kprobe_is_return(tk))
 503                         unregister_kretprobe(&tk->rp);
 504                 else
 505                         unregister_kprobe(&tk->rp.kp);
 506                 tk->tp.flags &= ~TP_FLAG_REGISTERED;
 507                 /* Cleanup kprobe for reuse */
 508                 if (tk->rp.kp.symbol_name)
 509                         tk->rp.kp.addr = NULL;
 510         }
 511 }
 512
 513 /* Unregister a trace_probe and probe_event: call with locking probe_lock */
 514 static int unregister_trace_kprobe(struct trace_kprobe *tk)
 515 {
 516         /* Enabled event can not be unregistered */
 517         if (trace_probe_is_enabled(&tk->tp))
 518                 return -EBUSY;
 519
 520         /* Will fail if probe is being used by ftrace or perf */
 521         if (unregister_kprobe_event(tk))
 522                 return -EBUSY;
 523
 524         __unregister_trace_kprobe(tk);
 525         list_del(&tk->list);
 526
 527         return 0;
 528 }
 529
 530 /* Register a trace_probe and probe_event */
 531 static int register_trace_kprobe(struct trace_kprobe *tk)
 532 {
 533         struct trace_kprobe *old_tk;
 534         int ret;
 535
 536         mutex_lock(&probe_lock);
 537
 538         /* Delete old (same name) event if exist */
 539         old_tk = find_trace_kprobe(trace_event_name(&tk->tp.call),
 540                         tk->tp.call.class->system);
 541         if (old_tk) {
 542                 ret = unregister_trace_kprobe(old_tk);
 543                 if (ret < 0)
 544                         goto end;
 545                 free_trace_kprobe(old_tk);
 546         }
 547
 548         /* Register new event */
 549         ret = register_kprobe_event(tk);
 550         if (ret) {
 551                 pr_warn("Failed to register probe event(%d)\n", ret);
 552                 goto end;
 553         }
 554
 555         /* Register k*probe */
 556         ret = __register_trace_kprobe(tk);
 557         if (ret < 0)
 558                 unregister_kprobe_event(tk);
 559         else
 560                 list_add_tail(&tk->list, &probe_list);
 561
 562 end:
 563         mutex_unlock(&probe_lock);
 564         return ret;
 565 }
 566
 567 /* Module notifier call back, checking event on the module */
 568 static int trace_kprobe_module_callback(struct notifier_block *nb,
 569                                        unsigned long val, void *data)
 570 {
 571         struct module *mod = data;
 572         struct trace_kprobe *tk;
 573         int ret;
 574
 575         if (val != MODULE_STATE_COMING)
 576                 return NOTIFY_DONE;
 577
 578         /* Update probes on coming module */
 579         mutex_lock(&probe_lock);
 580         list_for_each_entry(tk, &probe_list, list) {
 581                 if (trace_kprobe_within_module(tk, mod)) {
 582                         /* Don't need to check busy - this should have gone. */
 583                         __unregister_trace_kprobe(tk);
 584                         ret = __register_trace_kprobe(tk);
 585                         if (ret)
 586                                 pr_warn("Failed to re-register probe %s on %s: %d\n",
 587                                         trace_event_name(&tk->tp.call),
 588                                         mod->name, ret);
 589                 }
 590         }
 591         mutex_unlock(&probe_lock);
 592
 593         return NOTIFY_DONE;
 594 }
 595
 596 static struct notifier_block trace_kprobe_module_nb = {
 597         .notifier_call = trace_kprobe_module_callback,
 598         .priority = 1   /* Invoked after kprobe module callback */
 599 };
 600
 601 static int create_trace_kprobe(int argc, char **argv)
 602 {
 603         /*
 604          * Argument syntax:
 605          *  - Add kprobe:
 606          *      p[:[GRP/]EVENT] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
 607          *  - Add kretprobe:
 608          *      r[MAXACTIVE][:[GRP/]EVENT] [MOD:]KSYM[+0] [FETCHARGS]
 609          * Fetch args:
 610          *  $retval     : fetch return value
 611          *  $stack      : fetch stack address
 612          *  $stackN     : fetch Nth of stack (N:0-)
 613          *  $comm       : fetch current task comm
 614          *  @ADDR       : fetch memory at ADDR (ADDR should be in kernel)
 615          *  @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
 616          *  %REG        : fetch register REG
 617          * Dereferencing memory fetch:
 618          *  +|-offs(ARG) : fetch memory at ARG +|- offs address.
 619          * Alias name of args:
 620          *  NAME=FETCHARG : set NAME as alias of FETCHARG.
 621          * Type of args:
 622          *  FETCHARG:TYPE : use TYPE instead of unsigned long.
 623          */
 624         struct trace_kprobe *tk;
 625         int i, ret = 0;
 626         bool is_return = false, is_delete = false;
 627         char *symbol = NULL, *event = NULL, *group = NULL;
 628         int maxactive = 0;
 629         char *arg;
 630         unsigned long offset = 0;
 631         void *addr = NULL;
 632         char buf[MAX_EVENT_NAME_LEN];
 633
 634         /* argc must be >= 1 */
 635         if (argv[0][0] == 'p')
 636                 is_return = false;
 637         else if (argv[0][0] == 'r')
 638                 is_return = true;
 639         else if (argv[0][0] == '-')
 640                 is_delete = true;
 641         else {
 642                 pr_info("Probe definition must be started with 'p', 'r' or"
 643                         " '-'.\n");
 644                 return -EINVAL;
 645         }
 646
 647         event = strchr(&argv[0][1], ':');
 648         if (event) {
 649                 event[0] = '\0';
 650                 event++;
 651         }
 652         if (is_return && isdigit(argv[0][1])) {
 653                 ret = kstrtouint(&argv[0][1], 0, &maxactive);
 654                 if (ret) {
 655                         pr_info("Failed to parse maxactive.\n");
 656                         return ret;
 657                 }
 658                 /* kretprobes instances are iterated over via a list. The
 659                  * maximum should stay reasonable.
 660                  */
 661                 if (maxactive > KRETPROBE_MAXACTIVE_MAX) {
 662                         pr_info("Maxactive is too big (%d > %d).\n",
 663                                 maxactive, KRETPROBE_MAXACTIVE_MAX);
 664                         return -E2BIG;
 665                 }
 666         }
 667
 668         if (event) {
 669                 if (strchr(event, '/')) {
 670                         group = event;
 671                         event = strchr(group, '/') + 1;
 672                         event[-1] = '\0';
 673                         if (strlen(group) == 0) {
 674                                 pr_info("Group name is not specified\n");
 675                                 return -EINVAL;
 676                         }
 677                 }
 678                 if (strlen(event) == 0) {
 679                         pr_info("Event name is not specified\n");
 680                         return -EINVAL;
 681                 }
 682         }
 683         if (!group)
 684                 group = KPROBE_EVENT_SYSTEM;
 685
 686         if (is_delete) {
 687                 if (!event) {
 688                         pr_info("Delete command needs an event name.\n");
 689                         return -EINVAL;
 690                 }
 691                 mutex_lock(&probe_lock);
 692                 tk = find_trace_kprobe(event, group);
 693                 if (!tk) {
 694                         mutex_unlock(&probe_lock);
 695                         pr_info("Event %s/%s doesn't exist.\n", group, event);
 696                         return -ENOENT;
 697                 }
 698                 /* delete an event */
 699                 ret = unregister_trace_kprobe(tk);
 700                 if (ret == 0)
 701                         free_trace_kprobe(tk);
 702                 mutex_unlock(&probe_lock);
 703                 return ret;
 704         }
 705
 706         if (argc < 2) {
 707                 pr_info("Probe point is not specified.\n");
 708                 return -EINVAL;
 709         }
 710
 711         /* try to parse an address. if that fails, try to read the
 712          * input as a symbol. */
 713         if (kstrtoul(argv[1], 0, (unsigned long *)&addr)) {
 714                 /* a symbol specified */
 715                 symbol = argv[1];
 716                 /* TODO: support .init module functions */
 717                 ret = traceprobe_split_symbol_offset(symbol, &offset);
 718                 if (ret) {
 719                         pr_info("Failed to parse either an address or a symbol.\n");
 720                         return ret;
 721                 }
 722                 if (offset && is_return &&
 723                     !kprobe_on_func_entry(NULL, symbol, offset)) {
 724                         pr_info("Given offset is not valid for return probe.\n");
 725                         return -EINVAL;
 726                 }
 727         }
 728         argc -= 2; argv += 2;
 729
 730         /* setup a probe */
 731         if (!event) {
 732                 /* Make a new event name */
 733                 if (symbol)
 734                         snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
 735                                  is_return ? 'r' : 'p', symbol, offset);
 736                 else
 737                         snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
 738                                  is_return ? 'r' : 'p', addr);
 739                 event = buf;
 740         }
 741         tk = alloc_trace_kprobe(group, event, addr, symbol, offset, maxactive,
 742                                argc, is_return);
 743         if (IS_ERR(tk)) {
 744                 pr_info("Failed to allocate trace_probe.(%d)\n",
 745                         (int)PTR_ERR(tk));
 746                 return PTR_ERR(tk);
 747         }
 748
 749         /* parse arguments */
 750         ret = 0;
 751         for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
 752                 struct probe_arg *parg = &tk->tp.args[i];
 753
 754                 /* Increment count for freeing args in error case */
 755                 tk->tp.nr_args++;
 756
 757                 /* Parse argument name */
 758                 arg = strchr(argv[i], '=');
 759                 if (arg) {
 760                         *arg++ = '\0';
 761                         parg->name = kstrdup(argv[i], GFP_KERNEL);
 762                 } else {
 763                         arg = argv[i];
 764                         /* If argument name is omitted, set "argN" */
 765                         snprintf(buf, MAX_EVENT_NAME_LEN, "arg%d", i + 1);
 766                         parg->name = kstrdup(buf, GFP_KERNEL);
 767                 }
 768
 769                 if (!parg->name) {
 770                         pr_info("Failed to allocate argument[%d] name.\n", i);
 771                         ret = -ENOMEM;
 772                         goto error;
 773                 }
 774
 775                 if (!is_good_name(parg->name)) {
 776                         pr_info("Invalid argument[%d] name: %s\n",
 777                                 i, parg->name);
 778                         ret = -EINVAL;
 779                         goto error;
 780                 }
 781
 782                 if (traceprobe_conflict_field_name(parg->name,
 783                                                         tk->tp.args, i)) {
 784                         pr_info("Argument[%d] name '%s' conflicts with "
 785                                 "another field.\n", i, argv[i]);
 786                         ret = -EINVAL;
 787                         goto error;
 788                 }
 789
 790                 /* Parse fetch argument */
 791                 ret = traceprobe_parse_probe_arg(arg, &tk->tp.size, parg,
 792                                                 is_return, true,
 793                                                 kprobes_fetch_type_table);
 794                 if (ret) {
 795                         pr_info("Parse error at argument[%d]. (%d)\n", i, ret);
 796                         goto error;
 797                 }
 798         }
 799
 800         ret = register_trace_kprobe(tk);
 801         if (ret)
 802                 goto error;
 803         return 0;
 804
 805 error:
 806         free_trace_kprobe(tk);
 807         return ret;
 808 }
 809
 810 static int release_all_trace_kprobes(void)
 811 {
 812         struct trace_kprobe *tk;
 813         int ret = 0;
 814
 815         mutex_lock(&probe_lock);
 816         /* Ensure no probe is in use. */
 817         list_for_each_entry(tk, &probe_list, list)
 818                 if (trace_probe_is_enabled(&tk->tp)) {
 819                         ret = -EBUSY;
 820                         goto end;
 821                 }
 822         /* TODO: Use batch unregistration */
 823         while (!list_empty(&probe_list)) {
 824                 tk = list_entry(probe_list.next, struct trace_kprobe, list);
 825                 ret = unregister_trace_kprobe(tk);
 826                 if (ret)
 827                         goto end;
 828                 free_trace_kprobe(tk);
 829         }
 830
 831 end:
 832         mutex_unlock(&probe_lock);
 833
 834         return ret;
 835 }
 836
 837 /* Probes listing interfaces */
 838 static void *probes_seq_start(struct seq_file *m, loff_t *pos)
 839 {
 840         mutex_lock(&probe_lock);
 841         return seq_list_start(&probe_list, *pos);
 842 }
 843
 844 static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
 845 {
 846         return seq_list_next(v, &probe_list, pos);
 847 }
 848
 849 static void probes_seq_stop(struct seq_file *m, void *v)
 850 {
 851         mutex_unlock(&probe_lock);
 852 }
 853
 854 static int probes_seq_show(struct seq_file *m, void *v)
 855 {
 856         struct trace_kprobe *tk = v;
 857         int i;
 858
 859         seq_putc(m, trace_kprobe_is_return(tk) ? 'r' : 'p');
 860         seq_printf(m, ":%s/%s", tk->tp.call.class->system,
 861                         trace_event_name(&tk->tp.call));
 862
 863         if (!tk->symbol)
 864                 seq_printf(m, " 0x%p", tk->rp.kp.addr);
 865         else if (tk->rp.kp.offset)
 866                 seq_printf(m, " %s+%u", trace_kprobe_symbol(tk),
 867                            tk->rp.kp.offset);
 868         else
 869                 seq_printf(m, " %s", trace_kprobe_symbol(tk));
 870
 871         for (i = 0; i < tk->tp.nr_args; i++)
 872                 seq_printf(m, " %s=%s", tk->tp.args[i].name, tk->tp.args[i].comm);
 873         seq_putc(m, '\n');
 874
 875         return 0;
 876 }
 877
 878 static const struct seq_operations probes_seq_op = {
 879         .start  = probes_seq_start,
 880         .next   = probes_seq_next,
 881         .stop   = probes_seq_stop,
 882         .show   = probes_seq_show
 883 };
 884
 885 static int probes_open(struct inode *inode, struct file *file)
 886 {
 887         int ret;
 888
 889         if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
 890                 ret = release_all_trace_kprobes();
 891                 if (ret < 0)
 892                         return ret;
 893         }
 894
 895         return seq_open(file, &probes_seq_op);
 896 }
 897
 898 static ssize_t probes_write(struct file *file, const char __user *buffer,
 899                             size_t count, loff_t *ppos)
 900 {
 901         return traceprobe_probes_write(file, buffer, count, ppos,
 902                         create_trace_kprobe);
 903 }
 904
 905 static const struct file_operations kprobe_events_ops = {
 906         .owner          = THIS_MODULE,
 907         .open           = probes_open,
 908         .read           = seq_read,
 909         .llseek         = seq_lseek,
 910         .release        = seq_release,
 911         .write          = probes_write,
 912 };
 913
 914 /* Probes profiling interfaces */
 915 static int probes_profile_seq_show(struct seq_file *m, void *v)
 916 {
 917         struct trace_kprobe *tk = v;
 918
 919         seq_printf(m, "  %-44s %15lu %15lu\n",
 920                    trace_event_name(&tk->tp.call),
 921                    trace_kprobe_nhit(tk),
 922                    tk->rp.kp.nmissed);
 923
 924         return 0;
 925 }
 926
 927 static const struct seq_operations profile_seq_op = {
 928         .start  = probes_seq_start,
 929         .next   = probes_seq_next,
 930         .stop   = probes_seq_stop,
 931         .show   = probes_profile_seq_show
 932 };
 933
 934 static int profile_open(struct inode *inode, struct file *file)
 935 {
 936         return seq_open(file, &profile_seq_op);
 937 }
 938
 939 static const struct file_operations kprobe_profile_ops = {
 940         .owner          = THIS_MODULE,
 941         .open           = profile_open,
 942         .read           = seq_read,
 943         .llseek         = seq_lseek,
 944         .release        = seq_release,
 945 };
 946
 947 /* Kprobe handler */
 948 static nokprobe_inline void
 949 __kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs,
 950                     struct trace_event_file *trace_file)
 951 {
 952         struct kprobe_trace_entry_head *entry;
 953         struct ring_buffer_event *event;
 954         struct ring_buffer *buffer;
 955         int size, dsize, pc;
 956         unsigned long irq_flags;
 957         struct trace_event_call *call = &tk->tp.call;
 958
 959         WARN_ON(call != trace_file->event_call);
 960
 961         if (trace_trigger_soft_disabled(trace_file))
 962                 return;
 963
 964         local_save_flags(irq_flags);
 965         pc = preempt_count();
 966
 967         dsize = __get_data_size(&tk->tp, regs);
 968         size = sizeof(*entry) + tk->tp.size + dsize;
 969
 970         event = trace_event_buffer_lock_reserve(&buffer, trace_file,
 971                                                 call->event.type,
 972                                                 size, irq_flags, pc);
 973         if (!event)
 974                 return;
 975
 976         entry = ring_buffer_event_data(event);
 977         entry->ip = (unsigned long)tk->rp.kp.addr;
 978         store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
 979
 980         event_trigger_unlock_commit_regs(trace_file, buffer, event,
 981                                          entry, irq_flags, pc, regs);
 982 }
 983
 984 static void
 985 kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs)
 986 {
 987         struct event_file_link *link;
 988
 989         list_for_each_entry_rcu(link, &tk->tp.files, list)
 990                 __kprobe_trace_func(tk, regs, link->file);
 991 }
 992 NOKPROBE_SYMBOL(kprobe_trace_func);
 993
 994 /* Kretprobe handler */
 995 static nokprobe_inline void
 996 __kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
 997                        struct pt_regs *regs,
 998                        struct trace_event_file *trace_file)
 999 {
1000         struct kretprobe_trace_entry_head *entry;
1001         struct ring_buffer_event *event;
1002         struct ring_buffer *buffer;
1003         int size, pc, dsize;
1004         unsigned long irq_flags;
1005         struct trace_event_call *call = &tk->tp.call;
1006
1007         WARN_ON(call != trace_file->event_call);
1008
1009         if (trace_trigger_soft_disabled(trace_file))
1010                 return;
1011
1012         local_save_flags(irq_flags);
1013         pc = preempt_count();
1014
1015         dsize = __get_data_size(&tk->tp, regs);
1016         size = sizeof(*entry) + tk->tp.size + dsize;
1017
1018         event = trace_event_buffer_lock_reserve(&buffer, trace_file,
1019                                                 call->event.type,
1020                                                 size, irq_flags, pc);
1021         if (!event)
1022                 return;
1023
1024         entry = ring_buffer_event_data(event);
1025         entry->func = (unsigned long)tk->rp.kp.addr;
1026         entry->ret_ip = (unsigned long)ri->ret_addr;
1027         store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1028
1029         event_trigger_unlock_commit_regs(trace_file, buffer, event,
1030                                          entry, irq_flags, pc, regs);
1031 }
1032
1033 static void
1034 kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1035                      struct pt_regs *regs)
1036 {
1037         struct event_file_link *link;
1038
1039         list_for_each_entry_rcu(link, &tk->tp.files, list)
1040                 __kretprobe_trace_func(tk, ri, regs, link->file);
1041 }
1042 NOKPROBE_SYMBOL(kretprobe_trace_func);
1043
1044 /* Event entry printers */
1045 static enum print_line_t
1046 print_kprobe_event(struct trace_iterator *iter, int flags,
1047                    struct trace_event *event)
1048 {
1049         struct kprobe_trace_entry_head *field;
1050         struct trace_seq *s = &iter->seq;
1051         struct trace_probe *tp;
1052         u8 *data;
1053         int i;
1054
1055         field = (struct kprobe_trace_entry_head *)iter->ent;
1056         tp = container_of(event, struct trace_probe, call.event);
1057
1058         trace_seq_printf(s, "%s: (", trace_event_name(&tp->call));
1059
1060         if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
1061                 goto out;
1062
1063         trace_seq_putc(s, ')');
1064
1065         data = (u8 *)&field[1];
1066         for (i = 0; i < tp->nr_args; i++)
1067                 if (!tp->args[i].type->print(s, tp->args[i].name,
1068                                              data + tp->args[i].offset, field))
1069                         goto out;
1070
1071         trace_seq_putc(s, '\n');
1072  out:
1073         return trace_handle_return(s);
1074 }
1075
1076 static enum print_line_t
1077 print_kretprobe_event(struct trace_iterator *iter, int flags,
1078                       struct trace_event *event)
1079 {
1080         struct kretprobe_trace_entry_head *field;
1081         struct trace_seq *s = &iter->seq;
1082         struct trace_probe *tp;
1083         u8 *data;
1084         int i;
1085
1086         field = (struct kretprobe_trace_entry_head *)iter->ent;
1087         tp = container_of(event, struct trace_probe, call.event);
1088
1089         trace_seq_printf(s, "%s: (", trace_event_name(&tp->call));
1090
1091         if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1092                 goto out;
1093
1094         trace_seq_puts(s, " <- ");
1095
1096         if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1097                 goto out;
1098
1099         trace_seq_putc(s, ')');
1100
1101         data = (u8 *)&field[1];
1102         for (i = 0; i < tp->nr_args; i++)
1103                 if (!tp->args[i].type->print(s, tp->args[i].name,
1104                                              data + tp->args[i].offset, field))
1105                         goto out;
1106
1107         trace_seq_putc(s, '\n');
1108
1109  out:
1110         return trace_handle_return(s);
1111 }
1112
1113
1114 static int kprobe_event_define_fields(struct trace_event_call *event_call)
1115 {
1116         int ret, i;
1117         struct kprobe_trace_entry_head field;
1118         struct trace_kprobe *tk = (struct trace_kprobe *)event_call->data;
1119
1120         DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1121         /* Set argument names as fields */
1122         for (i = 0; i < tk->tp.nr_args; i++) {
1123                 struct probe_arg *parg = &tk->tp.args[i];
1124
1125                 ret = trace_define_field(event_call, parg->type->fmttype,
1126                                          parg->name,
1127                                          sizeof(field) + parg->offset,
1128                                          parg->type->size,
1129                                          parg->type->is_signed,
1130                                          FILTER_OTHER);
1131                 if (ret)
1132                         return ret;
1133         }
1134         return 0;
1135 }
1136
1137 static int kretprobe_event_define_fields(struct trace_event_call *event_call)
1138 {
1139         int ret, i;
1140         struct kretprobe_trace_entry_head field;
1141         struct trace_kprobe *tk = (struct trace_kprobe *)event_call->data;
1142
1143         DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1144         DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1145         /* Set argument names as fields */
1146         for (i = 0; i < tk->tp.nr_args; i++) {
1147                 struct probe_arg *parg = &tk->tp.args[i];
1148
1149                 ret = trace_define_field(event_call, parg->type->fmttype,
1150                                          parg->name,
1151                                          sizeof(field) + parg->offset,
1152                                          parg->type->size,
1153                                          parg->type->is_signed,
1154                                          FILTER_OTHER);
1155                 if (ret)
1156                         return ret;
1157         }
1158         return 0;
1159 }
1160
1161 #ifdef CONFIG_PERF_EVENTS
1162
1163 /* Kprobe profile handler */
1164 static void
1165 kprobe_perf_func(struct trace_kprobe *tk, struct pt_regs *regs)
1166 {
1167         struct trace_event_call *call = &tk->tp.call;
1168         struct bpf_prog *prog = call->prog;
1169         struct kprobe_trace_entry_head *entry;
1170         struct hlist_head *head;
1171         int size, __size, dsize;
1172         int rctx;
1173
1174         if (prog && !trace_call_bpf(prog, regs))
1175                 return;
1176
1177         head = this_cpu_ptr(call->perf_events);
1178         if (hlist_empty(head))
1179                 return;
1180
1181         dsize = __get_data_size(&tk->tp, regs);
1182         __size = sizeof(*entry) + tk->tp.size + dsize;
1183         size = ALIGN(__size + sizeof(u32), sizeof(u64));
1184         size -= sizeof(u32);
1185
1186         entry = perf_trace_buf_alloc(size, NULL, &rctx);
1187         if (!entry)
1188                 return;
1189
1190         entry->ip = (unsigned long)tk->rp.kp.addr;
1191         memset(&entry[1], 0, dsize);
1192         store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1193         perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1194                               head, NULL);
1195 }
1196 NOKPROBE_SYMBOL(kprobe_perf_func);
1197
1198 /* Kretprobe profile handler */
1199 static void
1200 kretprobe_perf_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1201                     struct pt_regs *regs)
1202 {
1203         struct trace_event_call *call = &tk->tp.call;
1204         struct bpf_prog *prog = call->prog;
1205         struct kretprobe_trace_entry_head *entry;
1206         struct hlist_head *head;
1207         int size, __size, dsize;
1208         int rctx;
1209
1210         if (prog && !trace_call_bpf(prog, regs))
1211                 return;
1212
1213         head = this_cpu_ptr(call->perf_events);
1214         if (hlist_empty(head))
1215                 return;
1216
1217         dsize = __get_data_size(&tk->tp, regs);
1218         __size = sizeof(*entry) + tk->tp.size + dsize;
1219         size = ALIGN(__size + sizeof(u32), sizeof(u64));
1220         size -= sizeof(u32);
1221
1222         entry = perf_trace_buf_alloc(size, NULL, &rctx);
1223         if (!entry)
1224                 return;
1225
1226         entry->func = (unsigned long)tk->rp.kp.addr;
1227         entry->ret_ip = (unsigned long)ri->ret_addr;
1228         store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1229         perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1230                               head, NULL);
1231 }
1232 NOKPROBE_SYMBOL(kretprobe_perf_func);
1233 #endif  /* CONFIG_PERF_EVENTS */
1234
1235 /*
1236  * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex.
1237  *
1238  * kprobe_trace_self_tests_init() does enable_trace_probe/disable_trace_probe
1239  * lockless, but we can't race with this __init function.
1240  */
1241 static int kprobe_register(struct trace_event_call *event,
1242                            enum trace_reg type, void *data)
1243 {
1244         struct trace_kprobe *tk = (struct trace_kprobe *)event->data;
1245         struct trace_event_file *file = data;
1246
1247         switch (type) {
1248         case TRACE_REG_REGISTER:
1249                 return enable_trace_kprobe(tk, file);
1250         case TRACE_REG_UNREGISTER:
1251                 return disable_trace_kprobe(tk, file);
1252
1253 #ifdef CONFIG_PERF_EVENTS
1254         case TRACE_REG_PERF_REGISTER:
1255                 return enable_trace_kprobe(tk, NULL);
1256         case TRACE_REG_PERF_UNREGISTER:
1257                 return disable_trace_kprobe(tk, NULL);
1258         case TRACE_REG_PERF_OPEN:
1259         case TRACE_REG_PERF_CLOSE:
1260         case TRACE_REG_PERF_ADD:
1261         case TRACE_REG_PERF_DEL:
1262                 return 0;
1263 #endif
1264         }
1265         return 0;
1266 }
1267
1268 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1269 {
1270         struct trace_kprobe *tk = container_of(kp, struct trace_kprobe, rp.kp);
1271
1272         raw_cpu_inc(*tk->nhit);
1273
1274         if (tk->tp.flags & TP_FLAG_TRACE)
1275                 kprobe_trace_func(tk, regs);
1276 #ifdef CONFIG_PERF_EVENTS
1277         if (tk->tp.flags & TP_FLAG_PROFILE)
1278                 kprobe_perf_func(tk, regs);
1279 #endif
1280         return 0;       /* We don't tweek kernel, so just return 0 */
1281 }
1282 NOKPROBE_SYMBOL(kprobe_dispatcher);
1283
1284 static int
1285 kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1286 {
1287         struct trace_kprobe *tk = container_of(ri->rp, struct trace_kprobe, rp);
1288
1289         raw_cpu_inc(*tk->nhit);
1290
1291         if (tk->tp.flags & TP_FLAG_TRACE)
1292                 kretprobe_trace_func(tk, ri, regs);
1293 #ifdef CONFIG_PERF_EVENTS
1294         if (tk->tp.flags & TP_FLAG_PROFILE)
1295                 kretprobe_perf_func(tk, ri, regs);
1296 #endif
1297         return 0;       /* We don't tweek kernel, so just return 0 */
1298 }
1299 NOKPROBE_SYMBOL(kretprobe_dispatcher);
1300
1301 static struct trace_event_functions kretprobe_funcs = {
1302         .trace          = print_kretprobe_event
1303 };
1304
1305 static struct trace_event_functions kprobe_funcs = {
1306         .trace          = print_kprobe_event
1307 };
1308
1309 static int register_kprobe_event(struct trace_kprobe *tk)
1310 {
1311         struct trace_event_call *call = &tk->tp.call;
1312         int ret;
1313
1314         /* Initialize trace_event_call */
1315         INIT_LIST_HEAD(&call->class->fields);
1316         if (trace_kprobe_is_return(tk)) {
1317                 call->event.funcs = &kretprobe_funcs;
1318                 call->class->define_fields = kretprobe_event_define_fields;
1319         } else {
1320                 call->event.funcs = &kprobe_funcs;
1321                 call->class->define_fields = kprobe_event_define_fields;
1322         }
1323         if (set_print_fmt(&tk->tp, trace_kprobe_is_return(tk)) < 0)
1324                 return -ENOMEM;
1325         ret = register_trace_event(&call->event);
1326         if (!ret) {
1327                 kfree(call->print_fmt);
1328                 return -ENODEV;
1329         }
1330         call->flags = TRACE_EVENT_FL_KPROBE;
1331         call->class->reg = kprobe_register;
1332         call->data = tk;
1333         ret = trace_add_event_call(call);
1334         if (ret) {
1335                 pr_info("Failed to register kprobe event: %s\n",
1336                         trace_event_name(call));
1337                 kfree(call->print_fmt);
1338                 unregister_trace_event(&call->event);
1339         }
1340         return ret;
1341 }
1342
1343 static int unregister_kprobe_event(struct trace_kprobe *tk)
1344 {
1345         int ret;
1346
1347         /* tp->event is unregistered in trace_remove_event_call() */
1348         ret = trace_remove_event_call(&tk->tp.call);
1349         if (!ret)
1350                 kfree(tk->tp.call.print_fmt);
1351         return ret;
1352 }
1353
1354 /* Make a tracefs interface for controlling probe points */
1355 static __init int init_kprobe_trace(void)
1356 {
1357         struct dentry *d_tracer;
1358         struct dentry *entry;
1359
1360         if (register_module_notifier(&trace_kprobe_module_nb))
1361                 return -EINVAL;
1362
1363         d_tracer = tracing_init_dentry();
1364         if (IS_ERR(d_tracer))
1365                 return 0;
1366
1367         entry = tracefs_create_file("kprobe_events", 0644, d_tracer,
1368                                     NULL, &kprobe_events_ops);
1369
1370         /* Event list interface */
1371         if (!entry)
1372                 pr_warn("Could not create tracefs 'kprobe_events' entry\n");
1373
1374         /* Profile interface */
1375         entry = tracefs_create_file("kprobe_profile", 0444, d_tracer,
1376                                     NULL, &kprobe_profile_ops);
1377
1378         if (!entry)
1379                 pr_warn("Could not create tracefs 'kprobe_profile' entry\n");
1380         return 0;
1381 }
1382 fs_initcall(init_kprobe_trace);
1383
1384
1385 #ifdef CONFIG_FTRACE_STARTUP_TEST
1386 /*
1387  * The "__used" keeps gcc from removing the function symbol
1388  * from the kallsyms table. 'noinline' makes sure that there
1389  * isn't an inlined version used by the test method below
1390  */
1391 static __used __init noinline int
1392 kprobe_trace_selftest_target(int a1, int a2, int a3, int a4, int a5, int a6)
1393 {
1394         return a1 + a2 + a3 + a4 + a5 + a6;
1395 }
1396
1397 static __init struct trace_event_file *
1398 find_trace_probe_file(struct trace_kprobe *tk, struct trace_array *tr)
1399 {
1400         struct trace_event_file *file;
1401
1402         list_for_each_entry(file, &tr->events, list)
1403                 if (file->event_call == &tk->tp.call)
1404                         return file;
1405
1406         return NULL;
1407 }
1408
1409 /*
1410  * Nobody but us can call enable_trace_kprobe/disable_trace_kprobe at this
1411  * stage, we can do this lockless.
1412  */
1413 static __init int kprobe_trace_self_tests_init(void)
1414 {
1415         int ret, warn = 0;
1416         int (*target)(int, int, int, int, int, int);
1417         struct trace_kprobe *tk;
1418         struct trace_event_file *file;
1419
1420         if (tracing_is_disabled())
1421                 return -ENODEV;
1422
1423         target = kprobe_trace_selftest_target;
1424
1425         pr_info("Testing kprobe tracing: ");
1426
1427         ret = traceprobe_command("p:testprobe kprobe_trace_selftest_target "
1428                                   "$stack $stack0 +0($stack)",
1429                                   create_trace_kprobe);
1430         if (WARN_ON_ONCE(ret)) {
1431                 pr_warn("error on probing function entry.\n");
1432                 warn++;
1433         } else {
1434                 /* Enable trace point */
1435                 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1436                 if (WARN_ON_ONCE(tk == NULL)) {
1437                         pr_warn("error on getting new probe.\n");
1438                         warn++;
1439                 } else {
1440                         file = find_trace_probe_file(tk, top_trace_array());
1441                         if (WARN_ON_ONCE(file == NULL)) {
1442                                 pr_warn("error on getting probe file.\n");
1443                                 warn++;
1444                         } else
1445                                 enable_trace_kprobe(tk, file);
1446                 }
1447         }
1448
1449         ret = traceprobe_command("r:testprobe2 kprobe_trace_selftest_target "
1450                                   "$retval", create_trace_kprobe);
1451         if (WARN_ON_ONCE(ret)) {
1452                 pr_warn("error on probing function return.\n");
1453                 warn++;
1454         } else {
1455                 /* Enable trace point */
1456                 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1457                 if (WARN_ON_ONCE(tk == NULL)) {
1458                         pr_warn("error on getting 2nd new probe.\n");
1459                         warn++;
1460                 } else {
1461                         file = find_trace_probe_file(tk, top_trace_array());
1462                         if (WARN_ON_ONCE(file == NULL)) {
1463                                 pr_warn("error on getting probe file.\n");
1464                                 warn++;
1465                         } else
1466                                 enable_trace_kprobe(tk, file);
1467                 }
1468         }
1469
1470         if (warn)
1471                 goto end;
1472
1473         ret = target(1, 2, 3, 4, 5, 6);
1474
1475         /*
1476          * Not expecting an error here, the check is only to prevent the
1477          * optimizer from removing the call to target() as otherwise there
1478          * are no side-effects and the call is never performed.
1479          */
1480         if (ret != 21)
1481                 warn++;
1482
1483         /* Disable trace points before removing it */
1484         tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1485         if (WARN_ON_ONCE(tk == NULL)) {
1486                 pr_warn("error on getting test probe.\n");
1487                 warn++;
1488         } else {
1489                 if (trace_kprobe_nhit(tk) != 1) {
1490                         pr_warn("incorrect number of testprobe hits\n");
1491                         warn++;
1492                 }
1493
1494                 file = find_trace_probe_file(tk, top_trace_array());
1495                 if (WARN_ON_ONCE(file == NULL)) {
1496                         pr_warn("error on getting probe file.\n");
1497                         warn++;
1498                 } else
1499                         disable_trace_kprobe(tk, file);
1500         }
1501
1502         tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1503         if (WARN_ON_ONCE(tk == NULL)) {
1504                 pr_warn("error on getting 2nd test probe.\n");
1505                 warn++;
1506         } else {
1507                 if (trace_kprobe_nhit(tk) != 1) {
1508                         pr_warn("incorrect number of testprobe2 hits\n");
1509                         warn++;
1510                 }
1511
1512                 file = find_trace_probe_file(tk, top_trace_array());
1513                 if (WARN_ON_ONCE(file == NULL)) {
1514                         pr_warn("error on getting probe file.\n");
1515                         warn++;
1516                 } else
1517                         disable_trace_kprobe(tk, file);
1518         }
1519
1520         ret = traceprobe_command("-:testprobe", create_trace_kprobe);
1521         if (WARN_ON_ONCE(ret)) {
1522                 pr_warn("error on deleting a probe.\n");
1523                 warn++;
1524         }
1525
1526         ret = traceprobe_command("-:testprobe2", create_trace_kprobe);
1527         if (WARN_ON_ONCE(ret)) {
1528                 pr_warn("error on deleting a probe.\n");
1529                 warn++;
1530         }
1531
1532 end:
1533         release_all_trace_kprobes();
1534         /*
1535          * Wait for the optimizer work to finish. Otherwise it might fiddle
1536          * with probes in already freed __init text.
1537          */
1538         wait_for_kprobe_optimizer();
1539         if (warn)
1540                 pr_cont("NG: Some tests are failed. Please check them.\n");
1541         else
1542                 pr_cont("OK\n");
1543         return 0;
1544 }
1545
1546 late_initcall(kprobe_trace_self_tests_init);
1547
1548 #endif