kernel/trace/trace_kprobe.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Kprobes-based tracing events
   4  *
   5  * Created by Masami Hiramatsu <mhiramat@redhat.com>
   6  *
   7  */
   8 #define pr_fmt(fmt)     "trace_kprobe: " fmt
   9
  10 #include <linux/module.h>
  11 #include <linux/uaccess.h>
  12 #include <linux/rculist.h>
  13 #include <linux/error-injection.h>
  14
  15 #include "trace_dynevent.h"
  16 #include "trace_kprobe_selftest.h"
  17 #include "trace_probe.h"
  18 #include "trace_probe_tmpl.h"
  19
  20 #define KPROBE_EVENT_SYSTEM "kprobes"
  21 #define KRETPROBE_MAXACTIVE_MAX 4096
  22
  23 static int trace_kprobe_create(int argc, const char **argv);
  24 static int trace_kprobe_show(struct seq_file *m, struct dyn_event *ev);
  25 static int trace_kprobe_release(struct dyn_event *ev);
  26 static bool trace_kprobe_is_busy(struct dyn_event *ev);
  27 static bool trace_kprobe_match(const char *system, const char *event,
  28                                struct dyn_event *ev);
  29
  30 static struct dyn_event_operations trace_kprobe_ops = {
  31         .create = trace_kprobe_create,
  32         .show = trace_kprobe_show,
  33         .is_busy = trace_kprobe_is_busy,
  34         .free = trace_kprobe_release,
  35         .match = trace_kprobe_match,
  36 };
  37
  38 /**
  39  * Kprobe event core functions
  40  */
  41 struct trace_kprobe {
  42         struct dyn_event        devent;
  43         struct kretprobe        rp;     /* Use rp.kp for kprobe use */
  44         unsigned long __percpu *nhit;
  45         const char              *symbol;        /* symbol name */
  46         struct trace_probe      tp;
  47 };
  48
  49 static bool is_trace_kprobe(struct dyn_event *ev)
  50 {
  51         return ev->ops == &trace_kprobe_ops;
  52 }
  53
  54 static struct trace_kprobe *to_trace_kprobe(struct dyn_event *ev)
  55 {
  56         return container_of(ev, struct trace_kprobe, devent);
  57 }
  58
  59 /**
  60  * for_each_trace_kprobe - iterate over the trace_kprobe list
  61  * @pos:        the struct trace_kprobe * for each entry
  62  * @dpos:       the struct dyn_event * to use as a loop cursor
  63  */
  64 #define for_each_trace_kprobe(pos, dpos)        \
  65         for_each_dyn_event(dpos)                \
  66                 if (is_trace_kprobe(dpos) && (pos = to_trace_kprobe(dpos)))
  67
  68 #define SIZEOF_TRACE_KPROBE(n)                          \
  69         (offsetof(struct trace_kprobe, tp.args) +       \
  70         (sizeof(struct probe_arg) * (n)))
  71
  72 static nokprobe_inline bool trace_kprobe_is_return(struct trace_kprobe *tk)
  73 {
  74         return tk->rp.handler != NULL;
  75 }
  76
  77 static nokprobe_inline const char *trace_kprobe_symbol(struct trace_kprobe *tk)
  78 {
  79         return tk->symbol ? tk->symbol : "unknown";
  80 }
  81
  82 static nokprobe_inline unsigned long trace_kprobe_offset(struct trace_kprobe *tk)
  83 {
  84         return tk->rp.kp.offset;
  85 }
  86
  87 static nokprobe_inline bool trace_kprobe_has_gone(struct trace_kprobe *tk)
  88 {
  89         return !!(kprobe_gone(&tk->rp.kp));
  90 }
  91
  92 static nokprobe_inline bool trace_kprobe_within_module(struct trace_kprobe *tk,
  93                                                  struct module *mod)
  94 {
  95         int len = strlen(mod->name);
  96         const char *name = trace_kprobe_symbol(tk);
  97         return strncmp(mod->name, name, len) == 0 && name[len] == ':';
  98 }
  99
 100 static nokprobe_inline bool trace_kprobe_module_exist(struct trace_kprobe *tk)
 101 {
 102         char *p;
 103         bool ret;
 104
 105         if (!tk->symbol)
 106                 return false;
 107         p = strchr(tk->symbol, ':');
 108         if (!p)
 109                 return true;
 110         *p = '\0';
 111         mutex_lock(&module_mutex);
 112         ret = !!find_module(tk->symbol);
 113         mutex_unlock(&module_mutex);
 114         *p = ':';
 115
 116         return ret;
 117 }
 118
 119 static bool trace_kprobe_is_busy(struct dyn_event *ev)
 120 {
 121         struct trace_kprobe *tk = to_trace_kprobe(ev);
 122
 123         return trace_probe_is_enabled(&tk->tp);
 124 }
 125
 126 static bool trace_kprobe_match(const char *system, const char *event,
 127                                struct dyn_event *ev)
 128 {
 129         struct trace_kprobe *tk = to_trace_kprobe(ev);
 130
 131         return strcmp(trace_event_name(&tk->tp.call), event) == 0 &&
 132             (!system || strcmp(tk->tp.call.class->system, system) == 0);
 133 }
 134
 135 static nokprobe_inline unsigned long trace_kprobe_nhit(struct trace_kprobe *tk)
 136 {
 137         unsigned long nhit = 0;
 138         int cpu;
 139
 140         for_each_possible_cpu(cpu)
 141                 nhit += *per_cpu_ptr(tk->nhit, cpu);
 142
 143         return nhit;
 144 }
 145
 146 /* Return 0 if it fails to find the symbol address */
 147 static nokprobe_inline
 148 unsigned long trace_kprobe_address(struct trace_kprobe *tk)
 149 {
 150         unsigned long addr;
 151
 152         if (tk->symbol) {
 153                 addr = (unsigned long)
 154                         kallsyms_lookup_name(trace_kprobe_symbol(tk));
 155                 if (addr)
 156                         addr += tk->rp.kp.offset;
 157         } else {
 158                 addr = (unsigned long)tk->rp.kp.addr;
 159         }
 160         return addr;
 161 }
 162
 163 bool trace_kprobe_on_func_entry(struct trace_event_call *call)
 164 {
 165         struct trace_kprobe *tk = (struct trace_kprobe *)call->data;
 166
 167         return kprobe_on_func_entry(tk->rp.kp.addr,
 168                         tk->rp.kp.addr ? NULL : tk->rp.kp.symbol_name,
 169                         tk->rp.kp.addr ? 0 : tk->rp.kp.offset);
 170 }
 171
 172 bool trace_kprobe_error_injectable(struct trace_event_call *call)
 173 {
 174         struct trace_kprobe *tk = (struct trace_kprobe *)call->data;
 175
 176         return within_error_injection_list(trace_kprobe_address(tk));
 177 }
 178
 179 static int register_kprobe_event(struct trace_kprobe *tk);
 180 static int unregister_kprobe_event(struct trace_kprobe *tk);
 181
 182 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
 183 static int kretprobe_dispatcher(struct kretprobe_instance *ri,
 184                                 struct pt_regs *regs);
 185
 186 /*
 187  * Allocate new trace_probe and initialize it (including kprobes).
 188  */
 189 static struct trace_kprobe *alloc_trace_kprobe(const char *group,
 190                                              const char *event,
 191                                              void *addr,
 192                                              const char *symbol,
 193                                              unsigned long offs,
 194                                              int maxactive,
 195                                              int nargs, bool is_return)
 196 {
 197         struct trace_kprobe *tk;
 198         int ret = -ENOMEM;
 199
 200         tk = kzalloc(SIZEOF_TRACE_KPROBE(nargs), GFP_KERNEL);
 201         if (!tk)
 202                 return ERR_PTR(ret);
 203
 204         tk->nhit = alloc_percpu(unsigned long);
 205         if (!tk->nhit)
 206                 goto error;
 207
 208         if (symbol) {
 209                 tk->symbol = kstrdup(symbol, GFP_KERNEL);
 210                 if (!tk->symbol)
 211                         goto error;
 212                 tk->rp.kp.symbol_name = tk->symbol;
 213                 tk->rp.kp.offset = offs;
 214         } else
 215                 tk->rp.kp.addr = addr;
 216
 217         if (is_return)
 218                 tk->rp.handler = kretprobe_dispatcher;
 219         else
 220                 tk->rp.kp.pre_handler = kprobe_dispatcher;
 221
 222         tk->rp.maxactive = maxactive;
 223
 224         if (!event || !is_good_name(event)) {
 225                 ret = -EINVAL;
 226                 goto error;
 227         }
 228
 229         tk->tp.call.class = &tk->tp.class;
 230         tk->tp.call.name = kstrdup(event, GFP_KERNEL);
 231         if (!tk->tp.call.name)
 232                 goto error;
 233
 234         if (!group || !is_good_name(group)) {
 235                 ret = -EINVAL;
 236                 goto error;
 237         }
 238
 239         tk->tp.class.system = kstrdup(group, GFP_KERNEL);
 240         if (!tk->tp.class.system)
 241                 goto error;
 242
 243         dyn_event_init(&tk->devent, &trace_kprobe_ops);
 244         INIT_LIST_HEAD(&tk->tp.files);
 245         return tk;
 246 error:
 247         kfree(tk->tp.call.name);
 248         kfree(tk->symbol);
 249         free_percpu(tk->nhit);
 250         kfree(tk);
 251         return ERR_PTR(ret);
 252 }
 253
 254 static void free_trace_kprobe(struct trace_kprobe *tk)
 255 {
 256         int i;
 257
 258         if (!tk)
 259                 return;
 260
 261         for (i = 0; i < tk->tp.nr_args; i++)
 262                 traceprobe_free_probe_arg(&tk->tp.args[i]);
 263
 264         kfree(tk->tp.call.class->system);
 265         kfree(tk->tp.call.name);
 266         kfree(tk->symbol);
 267         free_percpu(tk->nhit);
 268         kfree(tk);
 269 }
 270
 271 static struct trace_kprobe *find_trace_kprobe(const char *event,
 272                                               const char *group)
 273 {
 274         struct dyn_event *pos;
 275         struct trace_kprobe *tk;
 276
 277         for_each_trace_kprobe(tk, pos)
 278                 if (strcmp(trace_event_name(&tk->tp.call), event) == 0 &&
 279                     strcmp(tk->tp.call.class->system, group) == 0)
 280                         return tk;
 281         return NULL;
 282 }
 283
 284 static inline int __enable_trace_kprobe(struct trace_kprobe *tk)
 285 {
 286         int ret = 0;
 287
 288         if (trace_probe_is_registered(&tk->tp) && !trace_kprobe_has_gone(tk)) {
 289                 if (trace_kprobe_is_return(tk))
 290                         ret = enable_kretprobe(&tk->rp);
 291                 else
 292                         ret = enable_kprobe(&tk->rp.kp);
 293         }
 294
 295         return ret;
 296 }
 297
 298 /*
 299  * Enable trace_probe
 300  * if the file is NULL, enable "perf" handler, or enable "trace" handler.
 301  */
 302 static int
 303 enable_trace_kprobe(struct trace_kprobe *tk, struct trace_event_file *file)
 304 {
 305         struct event_file_link *link;
 306         int ret = 0;
 307
 308         if (file) {
 309                 link = kmalloc(sizeof(*link), GFP_KERNEL);
 310                 if (!link) {
 311                         ret = -ENOMEM;
 312                         goto out;
 313                 }
 314
 315                 link->file = file;
 316                 list_add_tail_rcu(&link->list, &tk->tp.files);
 317
 318                 tk->tp.flags |= TP_FLAG_TRACE;
 319                 ret = __enable_trace_kprobe(tk);
 320                 if (ret) {
 321                         list_del_rcu(&link->list);
 322                         kfree(link);
 323                         tk->tp.flags &= ~TP_FLAG_TRACE;
 324                 }
 325
 326         } else {
 327                 tk->tp.flags |= TP_FLAG_PROFILE;
 328                 ret = __enable_trace_kprobe(tk);
 329                 if (ret)
 330                         tk->tp.flags &= ~TP_FLAG_PROFILE;
 331         }
 332  out:
 333         return ret;
 334 }
 335
 336 /*
 337  * Disable trace_probe
 338  * if the file is NULL, disable "perf" handler, or disable "trace" handler.
 339  */
 340 static int
 341 disable_trace_kprobe(struct trace_kprobe *tk, struct trace_event_file *file)
 342 {
 343         struct event_file_link *link = NULL;
 344         int wait = 0;
 345         int ret = 0;
 346
 347         if (file) {
 348                 link = find_event_file_link(&tk->tp, file);
 349                 if (!link) {
 350                         ret = -EINVAL;
 351                         goto out;
 352                 }
 353
 354                 list_del_rcu(&link->list);
 355                 wait = 1;
 356                 if (!list_empty(&tk->tp.files))
 357                         goto out;
 358
 359                 tk->tp.flags &= ~TP_FLAG_TRACE;
 360         } else
 361                 tk->tp.flags &= ~TP_FLAG_PROFILE;
 362
 363         if (!trace_probe_is_enabled(&tk->tp) && trace_probe_is_registered(&tk->tp)) {
 364                 if (trace_kprobe_is_return(tk))
 365                         disable_kretprobe(&tk->rp);
 366                 else
 367                         disable_kprobe(&tk->rp.kp);
 368                 wait = 1;
 369         }
 370
 371         /*
 372          * if tk is not added to any list, it must be a local trace_kprobe
 373          * created with perf_event_open. We don't need to wait for these
 374          * trace_kprobes
 375          */
 376         if (list_empty(&tk->devent.list))
 377                 wait = 0;
 378  out:
 379         if (wait) {
 380                 /*
 381                  * Synchronize with kprobe_trace_func/kretprobe_trace_func
 382                  * to ensure disabled (all running handlers are finished).
 383                  * This is not only for kfree(), but also the caller,
 384                  * trace_remove_event_call() supposes it for releasing
 385                  * event_call related objects, which will be accessed in
 386                  * the kprobe_trace_func/kretprobe_trace_func.
 387                  */
 388                 synchronize_rcu();
 389                 kfree(link);    /* Ignored if link == NULL */
 390         }
 391
 392         return ret;
 393 }
 394
 395 #if defined(CONFIG_KPROBES_ON_FTRACE) && \
 396         !defined(CONFIG_KPROBE_EVENTS_ON_NOTRACE)
 397 static bool within_notrace_func(struct trace_kprobe *tk)
 398 {
 399         unsigned long offset, size, addr;
 400
 401         addr = trace_kprobe_address(tk);
 402         if (!addr || !kallsyms_lookup_size_offset(addr, &size, &offset))
 403                 return false;
 404
 405         /* Get the entry address of the target function */
 406         addr -= offset;
 407
 408         /*
 409          * Since ftrace_location_range() does inclusive range check, we need
 410          * to subtract 1 byte from the end address.
 411          */
 412         return !ftrace_location_range(addr, addr + size - 1);
 413 }
 414 #else
 415 #define within_notrace_func(tk) (false)
 416 #endif
 417
 418 /* Internal register function - just handle k*probes and flags */
 419 static int __register_trace_kprobe(struct trace_kprobe *tk)
 420 {
 421         int i, ret;
 422
 423         if (trace_probe_is_registered(&tk->tp))
 424                 return -EINVAL;
 425
 426         if (within_notrace_func(tk)) {
 427                 pr_warn("Could not probe notrace function %s\n",
 428                         trace_kprobe_symbol(tk));
 429                 return -EINVAL;
 430         }
 431
 432         for (i = 0; i < tk->tp.nr_args; i++) {
 433                 ret = traceprobe_update_arg(&tk->tp.args[i]);
 434                 if (ret)
 435                         return ret;
 436         }
 437
 438         /* Set/clear disabled flag according to tp->flag */
 439         if (trace_probe_is_enabled(&tk->tp))
 440                 tk->rp.kp.flags &= ~KPROBE_FLAG_DISABLED;
 441         else
 442                 tk->rp.kp.flags |= KPROBE_FLAG_DISABLED;
 443
 444         if (trace_kprobe_is_return(tk))
 445                 ret = register_kretprobe(&tk->rp);
 446         else
 447                 ret = register_kprobe(&tk->rp.kp);
 448
 449         if (ret == 0) {
 450                 tk->tp.flags |= TP_FLAG_REGISTERED;
 451         } else if (ret == -EILSEQ) {
 452                 pr_warn("Probing address(0x%p) is not an instruction boundary.\n",
 453                         tk->rp.kp.addr);
 454                 ret = -EINVAL;
 455         }
 456         return ret;
 457 }
 458
 459 /* Internal unregister function - just handle k*probes and flags */
 460 static void __unregister_trace_kprobe(struct trace_kprobe *tk)
 461 {
 462         if (trace_probe_is_registered(&tk->tp)) {
 463                 if (trace_kprobe_is_return(tk))
 464                         unregister_kretprobe(&tk->rp);
 465                 else
 466                         unregister_kprobe(&tk->rp.kp);
 467                 tk->tp.flags &= ~TP_FLAG_REGISTERED;
 468                 /* Cleanup kprobe for reuse */
 469                 if (tk->rp.kp.symbol_name)
 470                         tk->rp.kp.addr = NULL;
 471         }
 472 }
 473
 474 /* Unregister a trace_probe and probe_event */
 475 static int unregister_trace_kprobe(struct trace_kprobe *tk)
 476 {
 477         /* Enabled event can not be unregistered */
 478         if (trace_probe_is_enabled(&tk->tp))
 479                 return -EBUSY;
 480
 481         /* Will fail if probe is being used by ftrace or perf */
 482         if (unregister_kprobe_event(tk))
 483                 return -EBUSY;
 484
 485         __unregister_trace_kprobe(tk);
 486         dyn_event_remove(&tk->devent);
 487
 488         return 0;
 489 }
 490
 491 /* Register a trace_probe and probe_event */
 492 static int register_trace_kprobe(struct trace_kprobe *tk)
 493 {
 494         struct trace_kprobe *old_tk;
 495         int ret;
 496
 497         mutex_lock(&event_mutex);
 498
 499         /* Delete old (same name) event if exist */
 500         old_tk = find_trace_kprobe(trace_event_name(&tk->tp.call),
 501                         tk->tp.call.class->system);
 502         if (old_tk) {
 503                 ret = unregister_trace_kprobe(old_tk);
 504                 if (ret < 0)
 505                         goto end;
 506                 free_trace_kprobe(old_tk);
 507         }
 508
 509         /* Register new event */
 510         ret = register_kprobe_event(tk);
 511         if (ret) {
 512                 pr_warn("Failed to register probe event(%d)\n", ret);
 513                 goto end;
 514         }
 515
 516         /* Register k*probe */
 517         ret = __register_trace_kprobe(tk);
 518         if (ret == -ENOENT && !trace_kprobe_module_exist(tk)) {
 519                 pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
 520                 ret = 0;
 521         }
 522
 523         if (ret < 0)
 524                 unregister_kprobe_event(tk);
 525         else
 526                 dyn_event_add(&tk->devent);
 527
 528 end:
 529         mutex_unlock(&event_mutex);
 530         return ret;
 531 }
 532
 533 /* Module notifier call back, checking event on the module */
 534 static int trace_kprobe_module_callback(struct notifier_block *nb,
 535                                        unsigned long val, void *data)
 536 {
 537         struct module *mod = data;
 538         struct dyn_event *pos;
 539         struct trace_kprobe *tk;
 540         int ret;
 541
 542         if (val != MODULE_STATE_COMING)
 543                 return NOTIFY_DONE;
 544
 545         /* Update probes on coming module */
 546         mutex_lock(&event_mutex);
 547         for_each_trace_kprobe(tk, pos) {
 548                 if (trace_kprobe_within_module(tk, mod)) {
 549                         /* Don't need to check busy - this should have gone. */
 550                         __unregister_trace_kprobe(tk);
 551                         ret = __register_trace_kprobe(tk);
 552                         if (ret)
 553                                 pr_warn("Failed to re-register probe %s on %s: %d\n",
 554                                         trace_event_name(&tk->tp.call),
 555                                         mod->name, ret);
 556                 }
 557         }
 558         mutex_unlock(&event_mutex);
 559
 560         return NOTIFY_DONE;
 561 }
 562
 563 static struct notifier_block trace_kprobe_module_nb = {
 564         .notifier_call = trace_kprobe_module_callback,
 565         .priority = 1   /* Invoked after kprobe module callback */
 566 };
 567
 568 /* Convert certain expected symbols into '_' when generating event names */
 569 static inline void sanitize_event_name(char *name)
 570 {
 571         while (*name++ != '\0')
 572                 if (*name == ':' || *name == '.')
 573                         *name = '_';
 574 }
 575
 576 static int trace_kprobe_create(int argc, const char *argv[])
 577 {
 578         /*
 579          * Argument syntax:
 580          *  - Add kprobe:
 581          *      p[:[GRP/]EVENT] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
 582          *  - Add kretprobe:
 583          *      r[MAXACTIVE][:[GRP/]EVENT] [MOD:]KSYM[+0] [FETCHARGS]
 584          * Fetch args:
 585          *  $retval     : fetch return value
 586          *  $stack      : fetch stack address
 587          *  $stackN     : fetch Nth of stack (N:0-)
 588          *  $comm       : fetch current task comm
 589          *  @ADDR       : fetch memory at ADDR (ADDR should be in kernel)
 590          *  @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
 591          *  %REG        : fetch register REG
 592          * Dereferencing memory fetch:
 593          *  +|-offs(ARG) : fetch memory at ARG +|- offs address.
 594          * Alias name of args:
 595          *  NAME=FETCHARG : set NAME as alias of FETCHARG.
 596          * Type of args:
 597          *  FETCHARG:TYPE : use TYPE instead of unsigned long.
 598          */
 599         struct trace_kprobe *tk;
 600         int i, len, ret = 0;
 601         bool is_return = false;
 602         char *symbol = NULL, *tmp = NULL;
 603         const char *event = NULL, *group = KPROBE_EVENT_SYSTEM;
 604         int maxactive = 0;
 605         long offset = 0;
 606         void *addr = NULL;
 607         char buf[MAX_EVENT_NAME_LEN];
 608         unsigned int flags = TPARG_FL_KERNEL;
 609
 610         /* argc must be >= 1 */
 611         if (argv[0][0] == 'r') {
 612                 is_return = true;
 613                 flags |= TPARG_FL_RETURN;
 614         } else if (argv[0][0] != 'p' || argc < 2)
 615                 return -ECANCELED;
 616
 617         event = strchr(&argv[0][1], ':');
 618         if (event)
 619                 event++;
 620
 621         if (is_return && isdigit(argv[0][1])) {
 622                 if (event)
 623                         len = event - &argv[0][1] - 1;
 624                 else
 625                         len = strlen(&argv[0][1]);
 626                 if (len > MAX_EVENT_NAME_LEN - 1)
 627                         return -E2BIG;
 628                 memcpy(buf, &argv[0][1], len);
 629                 buf[len] = '\0';
 630                 ret = kstrtouint(buf, 0, &maxactive);
 631                 if (ret) {
 632                         pr_info("Failed to parse maxactive.\n");
 633                         return ret;
 634                 }
 635                 /* kretprobes instances are iterated over via a list. The
 636                  * maximum should stay reasonable.
 637                  */
 638                 if (maxactive > KRETPROBE_MAXACTIVE_MAX) {
 639                         pr_info("Maxactive is too big (%d > %d).\n",
 640                                 maxactive, KRETPROBE_MAXACTIVE_MAX);
 641                         return -E2BIG;
 642                 }
 643         }
 644
 645         /* try to parse an address. if that fails, try to read the
 646          * input as a symbol. */
 647         if (kstrtoul(argv[1], 0, (unsigned long *)&addr)) {
 648                 /* Check whether uprobe event specified */
 649                 if (strchr(argv[1], '/') && strchr(argv[1], ':'))
 650                         return -ECANCELED;
 651                 /* a symbol specified */
 652                 symbol = kstrdup(argv[1], GFP_KERNEL);
 653                 if (!symbol)
 654                         return -ENOMEM;
 655                 /* TODO: support .init module functions */
 656                 ret = traceprobe_split_symbol_offset(symbol, &offset);
 657                 if (ret || offset < 0 || offset > UINT_MAX) {
 658                         pr_info("Failed to parse either an address or a symbol.\n");
 659                         goto out;
 660                 }
 661                 if (kprobe_on_func_entry(NULL, symbol, offset))
 662                         flags |= TPARG_FL_FENTRY;
 663                 if (offset && is_return && !(flags & TPARG_FL_FENTRY)) {
 664                         pr_info("Given offset is not valid for return probe.\n");
 665                         ret = -EINVAL;
 666                         goto out;
 667                 }
 668         }
 669         argc -= 2; argv += 2;
 670
 671         if (event) {
 672                 ret = traceprobe_parse_event_name(&event, &group, buf);
 673                 if (ret)
 674                         goto out;
 675         } else {
 676                 /* Make a new event name */
 677                 if (symbol)
 678                         snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
 679                                  is_return ? 'r' : 'p', symbol, offset);
 680                 else
 681                         snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
 682                                  is_return ? 'r' : 'p', addr);
 683                 sanitize_event_name(buf);
 684                 event = buf;
 685         }
 686
 687         /* setup a probe */
 688         tk = alloc_trace_kprobe(group, event, addr, symbol, offset, maxactive,
 689                                argc, is_return);
 690         if (IS_ERR(tk)) {
 691                 pr_info("Failed to allocate trace_probe.(%d)\n",
 692                         (int)PTR_ERR(tk));
 693                 ret = PTR_ERR(tk);
 694                 goto out;
 695         }
 696
 697         /* parse arguments */
 698         for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
 699                 tmp = kstrdup(argv[i], GFP_KERNEL);
 700                 if (!tmp) {
 701                         ret = -ENOMEM;
 702                         goto error;
 703                 }
 704
 705                 ret = traceprobe_parse_probe_arg(&tk->tp, i, tmp, flags);
 706                 kfree(tmp);
 707                 if (ret)
 708                         goto error;
 709         }
 710
 711         ret = register_trace_kprobe(tk);
 712         if (ret)
 713                 goto error;
 714 out:
 715         kfree(symbol);
 716         return ret;
 717
 718 error:
 719         free_trace_kprobe(tk);
 720         goto out;
 721 }
 722
 723 static int create_or_delete_trace_kprobe(int argc, char **argv)
 724 {
 725         int ret;
 726
 727         if (argv[0][0] == '-')
 728                 return dyn_event_release(argc, argv, &trace_kprobe_ops);
 729
 730         ret = trace_kprobe_create(argc, (const char **)argv);
 731         return ret == -ECANCELED ? -EINVAL : ret;
 732 }
 733
 734 static int trace_kprobe_release(struct dyn_event *ev)
 735 {
 736         struct trace_kprobe *tk = to_trace_kprobe(ev);
 737         int ret = unregister_trace_kprobe(tk);
 738
 739         if (!ret)
 740                 free_trace_kprobe(tk);
 741         return ret;
 742 }
 743
 744 static int trace_kprobe_show(struct seq_file *m, struct dyn_event *ev)
 745 {
 746         struct trace_kprobe *tk = to_trace_kprobe(ev);
 747         int i;
 748
 749         seq_putc(m, trace_kprobe_is_return(tk) ? 'r' : 'p');
 750         seq_printf(m, ":%s/%s", tk->tp.call.class->system,
 751                         trace_event_name(&tk->tp.call));
 752
 753         if (!tk->symbol)
 754                 seq_printf(m, " 0x%p", tk->rp.kp.addr);
 755         else if (tk->rp.kp.offset)
 756                 seq_printf(m, " %s+%u", trace_kprobe_symbol(tk),
 757                            tk->rp.kp.offset);
 758         else
 759                 seq_printf(m, " %s", trace_kprobe_symbol(tk));
 760
 761         for (i = 0; i < tk->tp.nr_args; i++)
 762                 seq_printf(m, " %s=%s", tk->tp.args[i].name, tk->tp.args[i].comm);
 763         seq_putc(m, '\n');
 764
 765         return 0;
 766 }
 767
 768 static int probes_seq_show(struct seq_file *m, void *v)
 769 {
 770         struct dyn_event *ev = v;
 771
 772         if (!is_trace_kprobe(ev))
 773                 return 0;
 774
 775         return trace_kprobe_show(m, ev);
 776 }
 777
 778 static const struct seq_operations probes_seq_op = {
 779         .start  = dyn_event_seq_start,
 780         .next   = dyn_event_seq_next,
 781         .stop   = dyn_event_seq_stop,
 782         .show   = probes_seq_show
 783 };
 784
 785 static int probes_open(struct inode *inode, struct file *file)
 786 {
 787         int ret;
 788
 789         if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
 790                 ret = dyn_events_release_all(&trace_kprobe_ops);
 791                 if (ret < 0)
 792                         return ret;
 793         }
 794
 795         return seq_open(file, &probes_seq_op);
 796 }
 797
 798 static ssize_t probes_write(struct file *file, const char __user *buffer,
 799                             size_t count, loff_t *ppos)
 800 {
 801         return trace_parse_run_command(file, buffer, count, ppos,
 802                                        create_or_delete_trace_kprobe);
 803 }
 804
 805 static const struct file_operations kprobe_events_ops = {
 806         .owner          = THIS_MODULE,
 807         .open           = probes_open,
 808         .read           = seq_read,
 809         .llseek         = seq_lseek,
 810         .release        = seq_release,
 811         .write          = probes_write,
 812 };
 813
 814 /* Probes profiling interfaces */
 815 static int probes_profile_seq_show(struct seq_file *m, void *v)
 816 {
 817         struct dyn_event *ev = v;
 818         struct trace_kprobe *tk;
 819
 820         if (!is_trace_kprobe(ev))
 821                 return 0;
 822
 823         tk = to_trace_kprobe(ev);
 824         seq_printf(m, "  %-44s %15lu %15lu\n",
 825                    trace_event_name(&tk->tp.call),
 826                    trace_kprobe_nhit(tk),
 827                    tk->rp.kp.nmissed);
 828
 829         return 0;
 830 }
 831
 832 static const struct seq_operations profile_seq_op = {
 833         .start  = dyn_event_seq_start,
 834         .next   = dyn_event_seq_next,
 835         .stop   = dyn_event_seq_stop,
 836         .show   = probes_profile_seq_show
 837 };
 838
 839 static int profile_open(struct inode *inode, struct file *file)
 840 {
 841         return seq_open(file, &profile_seq_op);
 842 }
 843
 844 static const struct file_operations kprobe_profile_ops = {
 845         .owner          = THIS_MODULE,
 846         .open           = profile_open,
 847         .read           = seq_read,
 848         .llseek         = seq_lseek,
 849         .release        = seq_release,
 850 };
 851
 852 /* Kprobe specific fetch functions */
 853
 854 /* Return the length of string -- including null terminal byte */
 855 static nokprobe_inline int
 856 fetch_store_strlen(unsigned long addr)
 857 {
 858         mm_segment_t old_fs;
 859         int ret, len = 0;
 860         u8 c;
 861
 862         old_fs = get_fs();
 863         set_fs(KERNEL_DS);
 864         pagefault_disable();
 865
 866         do {
 867                 ret = __copy_from_user_inatomic(&c, (u8 *)addr + len, 1);
 868                 len++;
 869         } while (c && ret == 0 && len < MAX_STRING_SIZE);
 870
 871         pagefault_enable();
 872         set_fs(old_fs);
 873
 874         return (ret < 0) ? ret : len;
 875 }
 876
 877 /*
 878  * Fetch a null-terminated string. Caller MUST set *(u32 *)buf with max
 879  * length and relative data location.
 880  */
 881 static nokprobe_inline int
 882 fetch_store_string(unsigned long addr, void *dest, void *base)
 883 {
 884         int maxlen = get_loc_len(*(u32 *)dest);
 885         u8 *dst = get_loc_data(dest, base);
 886         long ret;
 887
 888         if (unlikely(!maxlen))
 889                 return -ENOMEM;
 890         /*
 891          * Try to get string again, since the string can be changed while
 892          * probing.
 893          */
 894         ret = strncpy_from_unsafe(dst, (void *)addr, maxlen);
 895
 896         if (ret >= 0)
 897                 *(u32 *)dest = make_data_loc(ret, (void *)dst - base);
 898         return ret;
 899 }
 900
 901 static nokprobe_inline int
 902 probe_mem_read(void *dest, void *src, size_t size)
 903 {
 904         return probe_kernel_read(dest, src, size);
 905 }
 906
 907 /* Note that we don't verify it, since the code does not come from user space */
 908 static int
 909 process_fetch_insn(struct fetch_insn *code, struct pt_regs *regs, void *dest,
 910                    void *base)
 911 {
 912         unsigned long val;
 913
 914 retry:
 915         /* 1st stage: get value from context */
 916         switch (code->op) {
 917         case FETCH_OP_REG:
 918                 val = regs_get_register(regs, code->param);
 919                 break;
 920         case FETCH_OP_STACK:
 921                 val = regs_get_kernel_stack_nth(regs, code->param);
 922                 break;
 923         case FETCH_OP_STACKP:
 924                 val = kernel_stack_pointer(regs);
 925                 break;
 926         case FETCH_OP_RETVAL:
 927                 val = regs_return_value(regs);
 928                 break;
 929         case FETCH_OP_IMM:
 930                 val = code->immediate;
 931                 break;
 932         case FETCH_OP_COMM:
 933                 val = (unsigned long)current->comm;
 934                 break;
 935 #ifdef CONFIG_HAVE_FUNCTION_ARG_ACCESS_API
 936         case FETCH_OP_ARG:
 937                 val = regs_get_kernel_argument(regs, code->param);
 938                 break;
 939 #endif
 940         case FETCH_NOP_SYMBOL:  /* Ignore a place holder */
 941                 code++;
 942                 goto retry;
 943         default:
 944                 return -EILSEQ;
 945         }
 946         code++;
 947
 948         return process_fetch_insn_bottom(code, val, dest, base);
 949 }
 950 NOKPROBE_SYMBOL(process_fetch_insn)
 951
 952 /* Kprobe handler */
 953 static nokprobe_inline void
 954 __kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs,
 955                     struct trace_event_file *trace_file)
 956 {
 957         struct kprobe_trace_entry_head *entry;
 958         struct ring_buffer_event *event;
 959         struct ring_buffer *buffer;
 960         int size, dsize, pc;
 961         unsigned long irq_flags;
 962         struct trace_event_call *call = &tk->tp.call;
 963
 964         WARN_ON(call != trace_file->event_call);
 965
 966         if (trace_trigger_soft_disabled(trace_file))
 967                 return;
 968
 969         local_save_flags(irq_flags);
 970         pc = preempt_count();
 971
 972         dsize = __get_data_size(&tk->tp, regs);
 973         size = sizeof(*entry) + tk->tp.size + dsize;
 974
 975         event = trace_event_buffer_lock_reserve(&buffer, trace_file,
 976                                                 call->event.type,
 977                                                 size, irq_flags, pc);
 978         if (!event)
 979                 return;
 980
 981         entry = ring_buffer_event_data(event);
 982         entry->ip = (unsigned long)tk->rp.kp.addr;
 983         store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
 984
 985         event_trigger_unlock_commit_regs(trace_file, buffer, event,
 986                                          entry, irq_flags, pc, regs);
 987 }
 988
 989 static void
 990 kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs)
 991 {
 992         struct event_file_link *link;
 993
 994         list_for_each_entry_rcu(link, &tk->tp.files, list)
 995                 __kprobe_trace_func(tk, regs, link->file);
 996 }
 997 NOKPROBE_SYMBOL(kprobe_trace_func);
 998
 999 /* Kretprobe handler */
1000 static nokprobe_inline void
1001 __kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1002                        struct pt_regs *regs,
1003                        struct trace_event_file *trace_file)
1004 {
1005         struct kretprobe_trace_entry_head *entry;
1006         struct ring_buffer_event *event;
1007         struct ring_buffer *buffer;
1008         int size, pc, dsize;
1009         unsigned long irq_flags;
1010         struct trace_event_call *call = &tk->tp.call;
1011
1012         WARN_ON(call != trace_file->event_call);
1013
1014         if (trace_trigger_soft_disabled(trace_file))
1015                 return;
1016
1017         local_save_flags(irq_flags);
1018         pc = preempt_count();
1019
1020         dsize = __get_data_size(&tk->tp, regs);
1021         size = sizeof(*entry) + tk->tp.size + dsize;
1022
1023         event = trace_event_buffer_lock_reserve(&buffer, trace_file,
1024                                                 call->event.type,
1025                                                 size, irq_flags, pc);
1026         if (!event)
1027                 return;
1028
1029         entry = ring_buffer_event_data(event);
1030         entry->func = (unsigned long)tk->rp.kp.addr;
1031         entry->ret_ip = (unsigned long)ri->ret_addr;
1032         store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1033
1034         event_trigger_unlock_commit_regs(trace_file, buffer, event,
1035                                          entry, irq_flags, pc, regs);
1036 }
1037
1038 static void
1039 kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1040                      struct pt_regs *regs)
1041 {
1042         struct event_file_link *link;
1043
1044         list_for_each_entry_rcu(link, &tk->tp.files, list)
1045                 __kretprobe_trace_func(tk, ri, regs, link->file);
1046 }
1047 NOKPROBE_SYMBOL(kretprobe_trace_func);
1048
1049 /* Event entry printers */
1050 static enum print_line_t
1051 print_kprobe_event(struct trace_iterator *iter, int flags,
1052                    struct trace_event *event)
1053 {
1054         struct kprobe_trace_entry_head *field;
1055         struct trace_seq *s = &iter->seq;
1056         struct trace_probe *tp;
1057
1058         field = (struct kprobe_trace_entry_head *)iter->ent;
1059         tp = container_of(event, struct trace_probe, call.event);
1060
1061         trace_seq_printf(s, "%s: (", trace_event_name(&tp->call));
1062
1063         if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
1064                 goto out;
1065
1066         trace_seq_putc(s, ')');
1067
1068         if (print_probe_args(s, tp->args, tp->nr_args,
1069                              (u8 *)&field[1], field) < 0)
1070                 goto out;
1071
1072         trace_seq_putc(s, '\n');
1073  out:
1074         return trace_handle_return(s);
1075 }
1076
1077 static enum print_line_t
1078 print_kretprobe_event(struct trace_iterator *iter, int flags,
1079                       struct trace_event *event)
1080 {
1081         struct kretprobe_trace_entry_head *field;
1082         struct trace_seq *s = &iter->seq;
1083         struct trace_probe *tp;
1084
1085         field = (struct kretprobe_trace_entry_head *)iter->ent;
1086         tp = container_of(event, struct trace_probe, call.event);
1087
1088         trace_seq_printf(s, "%s: (", trace_event_name(&tp->call));
1089
1090         if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1091                 goto out;
1092
1093         trace_seq_puts(s, " <- ");
1094
1095         if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1096                 goto out;
1097
1098         trace_seq_putc(s, ')');
1099
1100         if (print_probe_args(s, tp->args, tp->nr_args,
1101                              (u8 *)&field[1], field) < 0)
1102                 goto out;
1103
1104         trace_seq_putc(s, '\n');
1105
1106  out:
1107         return trace_handle_return(s);
1108 }
1109
1110
1111 static int kprobe_event_define_fields(struct trace_event_call *event_call)
1112 {
1113         int ret;
1114         struct kprobe_trace_entry_head field;
1115         struct trace_kprobe *tk = (struct trace_kprobe *)event_call->data;
1116
1117         DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1118
1119         return traceprobe_define_arg_fields(event_call, sizeof(field), &tk->tp);
1120 }
1121
1122 static int kretprobe_event_define_fields(struct trace_event_call *event_call)
1123 {
1124         int ret;
1125         struct kretprobe_trace_entry_head field;
1126         struct trace_kprobe *tk = (struct trace_kprobe *)event_call->data;
1127
1128         DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1129         DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1130
1131         return traceprobe_define_arg_fields(event_call, sizeof(field), &tk->tp);
1132 }
1133
1134 #ifdef CONFIG_PERF_EVENTS
1135
1136 /* Kprobe profile handler */
1137 static int
1138 kprobe_perf_func(struct trace_kprobe *tk, struct pt_regs *regs)
1139 {
1140         struct trace_event_call *call = &tk->tp.call;
1141         struct kprobe_trace_entry_head *entry;
1142         struct hlist_head *head;
1143         int size, __size, dsize;
1144         int rctx;
1145
1146         if (bpf_prog_array_valid(call)) {
1147                 unsigned long orig_ip = instruction_pointer(regs);
1148                 int ret;
1149
1150                 ret = trace_call_bpf(call, regs);
1151
1152                 /*
1153                  * We need to check and see if we modified the pc of the
1154                  * pt_regs, and if so return 1 so that we don't do the
1155                  * single stepping.
1156                  */
1157                 if (orig_ip != instruction_pointer(regs))
1158                         return 1;
1159                 if (!ret)
1160                         return 0;
1161         }
1162
1163         head = this_cpu_ptr(call->perf_events);
1164         if (hlist_empty(head))
1165                 return 0;
1166
1167         dsize = __get_data_size(&tk->tp, regs);
1168         __size = sizeof(*entry) + tk->tp.size + dsize;
1169         size = ALIGN(__size + sizeof(u32), sizeof(u64));
1170         size -= sizeof(u32);
1171
1172         entry = perf_trace_buf_alloc(size, NULL, &rctx);
1173         if (!entry)
1174                 return 0;
1175
1176         entry->ip = (unsigned long)tk->rp.kp.addr;
1177         memset(&entry[1], 0, dsize);
1178         store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1179         perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1180                               head, NULL);
1181         return 0;
1182 }
1183 NOKPROBE_SYMBOL(kprobe_perf_func);
1184
1185 /* Kretprobe profile handler */
1186 static void
1187 kretprobe_perf_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1188                     struct pt_regs *regs)
1189 {
1190         struct trace_event_call *call = &tk->tp.call;
1191         struct kretprobe_trace_entry_head *entry;
1192         struct hlist_head *head;
1193         int size, __size, dsize;
1194         int rctx;
1195
1196         if (bpf_prog_array_valid(call) && !trace_call_bpf(call, regs))
1197                 return;
1198
1199         head = this_cpu_ptr(call->perf_events);
1200         if (hlist_empty(head))
1201                 return;
1202
1203         dsize = __get_data_size(&tk->tp, regs);
1204         __size = sizeof(*entry) + tk->tp.size + dsize;
1205         size = ALIGN(__size + sizeof(u32), sizeof(u64));
1206         size -= sizeof(u32);
1207
1208         entry = perf_trace_buf_alloc(size, NULL, &rctx);
1209         if (!entry)
1210                 return;
1211
1212         entry->func = (unsigned long)tk->rp.kp.addr;
1213         entry->ret_ip = (unsigned long)ri->ret_addr;
1214         store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1215         perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1216                               head, NULL);
1217 }
1218 NOKPROBE_SYMBOL(kretprobe_perf_func);
1219
1220 int bpf_get_kprobe_info(const struct perf_event *event, u32 *fd_type,
1221                         const char **symbol, u64 *probe_offset,
1222                         u64 *probe_addr, bool perf_type_tracepoint)
1223 {
1224         const char *pevent = trace_event_name(event->tp_event);
1225         const char *group = event->tp_event->class->system;
1226         struct trace_kprobe *tk;
1227
1228         if (perf_type_tracepoint)
1229                 tk = find_trace_kprobe(pevent, group);
1230         else
1231                 tk = event->tp_event->data;
1232         if (!tk)
1233                 return -EINVAL;
1234
1235         *fd_type = trace_kprobe_is_return(tk) ? BPF_FD_TYPE_KRETPROBE
1236                                               : BPF_FD_TYPE_KPROBE;
1237         if (tk->symbol) {
1238                 *symbol = tk->symbol;
1239                 *probe_offset = tk->rp.kp.offset;
1240                 *probe_addr = 0;
1241         } else {
1242                 *symbol = NULL;
1243                 *probe_offset = 0;
1244                 *probe_addr = (unsigned long)tk->rp.kp.addr;
1245         }
1246         return 0;
1247 }
1248 #endif  /* CONFIG_PERF_EVENTS */
1249
1250 /*
1251  * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex.
1252  *
1253  * kprobe_trace_self_tests_init() does enable_trace_probe/disable_trace_probe
1254  * lockless, but we can't race with this __init function.
1255  */
1256 static int kprobe_register(struct trace_event_call *event,
1257                            enum trace_reg type, void *data)
1258 {
1259         struct trace_kprobe *tk = (struct trace_kprobe *)event->data;
1260         struct trace_event_file *file = data;
1261
1262         switch (type) {
1263         case TRACE_REG_REGISTER:
1264                 return enable_trace_kprobe(tk, file);
1265         case TRACE_REG_UNREGISTER:
1266                 return disable_trace_kprobe(tk, file);
1267
1268 #ifdef CONFIG_PERF_EVENTS
1269         case TRACE_REG_PERF_REGISTER:
1270                 return enable_trace_kprobe(tk, NULL);
1271         case TRACE_REG_PERF_UNREGISTER:
1272                 return disable_trace_kprobe(tk, NULL);
1273         case TRACE_REG_PERF_OPEN:
1274         case TRACE_REG_PERF_CLOSE:
1275         case TRACE_REG_PERF_ADD:
1276         case TRACE_REG_PERF_DEL:
1277                 return 0;
1278 #endif
1279         }
1280         return 0;
1281 }
1282
1283 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1284 {
1285         struct trace_kprobe *tk = container_of(kp, struct trace_kprobe, rp.kp);
1286         int ret = 0;
1287
1288         raw_cpu_inc(*tk->nhit);
1289
1290         if (tk->tp.flags & TP_FLAG_TRACE)
1291                 kprobe_trace_func(tk, regs);
1292 #ifdef CONFIG_PERF_EVENTS
1293         if (tk->tp.flags & TP_FLAG_PROFILE)
1294                 ret = kprobe_perf_func(tk, regs);
1295 #endif
1296         return ret;
1297 }
1298 NOKPROBE_SYMBOL(kprobe_dispatcher);
1299
1300 static int
1301 kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1302 {
1303         struct trace_kprobe *tk = container_of(ri->rp, struct trace_kprobe, rp);
1304
1305         raw_cpu_inc(*tk->nhit);
1306
1307         if (tk->tp.flags & TP_FLAG_TRACE)
1308                 kretprobe_trace_func(tk, ri, regs);
1309 #ifdef CONFIG_PERF_EVENTS
1310         if (tk->tp.flags & TP_FLAG_PROFILE)
1311                 kretprobe_perf_func(tk, ri, regs);
1312 #endif
1313         return 0;       /* We don't tweek kernel, so just return 0 */
1314 }
1315 NOKPROBE_SYMBOL(kretprobe_dispatcher);
1316
1317 static struct trace_event_functions kretprobe_funcs = {
1318         .trace          = print_kretprobe_event
1319 };
1320
1321 static struct trace_event_functions kprobe_funcs = {
1322         .trace          = print_kprobe_event
1323 };
1324
1325 static inline void init_trace_event_call(struct trace_kprobe *tk,
1326                                          struct trace_event_call *call)
1327 {
1328         INIT_LIST_HEAD(&call->class->fields);
1329         if (trace_kprobe_is_return(tk)) {
1330                 call->event.funcs = &kretprobe_funcs;
1331                 call->class->define_fields = kretprobe_event_define_fields;
1332         } else {
1333                 call->event.funcs = &kprobe_funcs;
1334                 call->class->define_fields = kprobe_event_define_fields;
1335         }
1336
1337         call->flags = TRACE_EVENT_FL_KPROBE;
1338         call->class->reg = kprobe_register;
1339         call->data = tk;
1340 }
1341
1342 static int register_kprobe_event(struct trace_kprobe *tk)
1343 {
1344         struct trace_event_call *call = &tk->tp.call;
1345         int ret = 0;
1346
1347         init_trace_event_call(tk, call);
1348
1349         if (traceprobe_set_print_fmt(&tk->tp, trace_kprobe_is_return(tk)) < 0)
1350                 return -ENOMEM;
1351         ret = register_trace_event(&call->event);
1352         if (!ret) {
1353                 kfree(call->print_fmt);
1354                 return -ENODEV;
1355         }
1356         ret = trace_add_event_call(call);
1357         if (ret) {
1358                 pr_info("Failed to register kprobe event: %s\n",
1359                         trace_event_name(call));
1360                 kfree(call->print_fmt);
1361                 unregister_trace_event(&call->event);
1362         }
1363         return ret;
1364 }
1365
1366 static int unregister_kprobe_event(struct trace_kprobe *tk)
1367 {
1368         int ret;
1369
1370         /* tp->event is unregistered in trace_remove_event_call() */
1371         ret = trace_remove_event_call(&tk->tp.call);
1372         if (!ret)
1373                 kfree(tk->tp.call.print_fmt);
1374         return ret;
1375 }
1376
1377 #ifdef CONFIG_PERF_EVENTS
1378 /* create a trace_kprobe, but don't add it to global lists */
1379 struct trace_event_call *
1380 create_local_trace_kprobe(char *func, void *addr, unsigned long offs,
1381                           bool is_return)
1382 {
1383         struct trace_kprobe *tk;
1384         int ret;
1385         char *event;
1386
1387         /*
1388          * local trace_kprobes are not added to dyn_event, so they are never
1389          * searched in find_trace_kprobe(). Therefore, there is no concern of
1390          * duplicated name here.
1391          */
1392         event = func ? func : "DUMMY_EVENT";
1393
1394         tk = alloc_trace_kprobe(KPROBE_EVENT_SYSTEM, event, (void *)addr, func,
1395                                 offs, 0 /* maxactive */, 0 /* nargs */,
1396                                 is_return);
1397
1398         if (IS_ERR(tk)) {
1399                 pr_info("Failed to allocate trace_probe.(%d)\n",
1400                         (int)PTR_ERR(tk));
1401                 return ERR_CAST(tk);
1402         }
1403
1404         init_trace_event_call(tk, &tk->tp.call);
1405
1406         if (traceprobe_set_print_fmt(&tk->tp, trace_kprobe_is_return(tk)) < 0) {
1407                 ret = -ENOMEM;
1408                 goto error;
1409         }
1410
1411         ret = __register_trace_kprobe(tk);
1412         if (ret < 0) {
1413                 kfree(tk->tp.call.print_fmt);
1414                 goto error;
1415         }
1416
1417         return &tk->tp.call;
1418 error:
1419         free_trace_kprobe(tk);
1420         return ERR_PTR(ret);
1421 }
1422
1423 void destroy_local_trace_kprobe(struct trace_event_call *event_call)
1424 {
1425         struct trace_kprobe *tk;
1426
1427         tk = container_of(event_call, struct trace_kprobe, tp.call);
1428
1429         if (trace_probe_is_enabled(&tk->tp)) {
1430                 WARN_ON(1);
1431                 return;
1432         }
1433
1434         __unregister_trace_kprobe(tk);
1435
1436         kfree(tk->tp.call.print_fmt);
1437         free_trace_kprobe(tk);
1438 }
1439 #endif /* CONFIG_PERF_EVENTS */
1440
1441 /* Make a tracefs interface for controlling probe points */
1442 static __init int init_kprobe_trace(void)
1443 {
1444         struct dentry *d_tracer;
1445         struct dentry *entry;
1446         int ret;
1447
1448         ret = dyn_event_register(&trace_kprobe_ops);
1449         if (ret)
1450                 return ret;
1451
1452         if (register_module_notifier(&trace_kprobe_module_nb))
1453                 return -EINVAL;
1454
1455         d_tracer = tracing_init_dentry();
1456         if (IS_ERR(d_tracer))
1457                 return 0;
1458
1459         entry = tracefs_create_file("kprobe_events", 0644, d_tracer,
1460                                     NULL, &kprobe_events_ops);
1461
1462         /* Event list interface */
1463         if (!entry)
1464                 pr_warn("Could not create tracefs 'kprobe_events' entry\n");
1465
1466         /* Profile interface */
1467         entry = tracefs_create_file("kprobe_profile", 0444, d_tracer,
1468                                     NULL, &kprobe_profile_ops);
1469
1470         if (!entry)
1471                 pr_warn("Could not create tracefs 'kprobe_profile' entry\n");
1472         return 0;
1473 }
1474 fs_initcall(init_kprobe_trace);
1475
1476
1477 #ifdef CONFIG_FTRACE_STARTUP_TEST
1478 static __init struct trace_event_file *
1479 find_trace_probe_file(struct trace_kprobe *tk, struct trace_array *tr)
1480 {
1481         struct trace_event_file *file;
1482
1483         list_for_each_entry(file, &tr->events, list)
1484                 if (file->event_call == &tk->tp.call)
1485                         return file;
1486
1487         return NULL;
1488 }
1489
1490 /*
1491  * Nobody but us can call enable_trace_kprobe/disable_trace_kprobe at this
1492  * stage, we can do this lockless.
1493  */
1494 static __init int kprobe_trace_self_tests_init(void)
1495 {
1496         int ret, warn = 0;
1497         int (*target)(int, int, int, int, int, int);
1498         struct trace_kprobe *tk;
1499         struct trace_event_file *file;
1500
1501         if (tracing_is_disabled())
1502                 return -ENODEV;
1503
1504         target = kprobe_trace_selftest_target;
1505
1506         pr_info("Testing kprobe tracing: ");
1507
1508         ret = trace_run_command("p:testprobe kprobe_trace_selftest_target $stack $stack0 +0($stack)",
1509                                 create_or_delete_trace_kprobe);
1510         if (WARN_ON_ONCE(ret)) {
1511                 pr_warn("error on probing function entry.\n");
1512                 warn++;
1513         } else {
1514                 /* Enable trace point */
1515                 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1516                 if (WARN_ON_ONCE(tk == NULL)) {
1517                         pr_warn("error on getting new probe.\n");
1518                         warn++;
1519                 } else {
1520                         file = find_trace_probe_file(tk, top_trace_array());
1521                         if (WARN_ON_ONCE(file == NULL)) {
1522                                 pr_warn("error on getting probe file.\n");
1523                                 warn++;
1524                         } else
1525                                 enable_trace_kprobe(tk, file);
1526                 }
1527         }
1528
1529         ret = trace_run_command("r:testprobe2 kprobe_trace_selftest_target $retval",
1530                                 create_or_delete_trace_kprobe);
1531         if (WARN_ON_ONCE(ret)) {
1532                 pr_warn("error on probing function return.\n");
1533                 warn++;
1534         } else {
1535                 /* Enable trace point */
1536                 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1537                 if (WARN_ON_ONCE(tk == NULL)) {
1538                         pr_warn("error on getting 2nd new probe.\n");
1539                         warn++;
1540                 } else {
1541                         file = find_trace_probe_file(tk, top_trace_array());
1542                         if (WARN_ON_ONCE(file == NULL)) {
1543                                 pr_warn("error on getting probe file.\n");
1544                                 warn++;
1545                         } else
1546                                 enable_trace_kprobe(tk, file);
1547                 }
1548         }
1549
1550         if (warn)
1551                 goto end;
1552
1553         ret = target(1, 2, 3, 4, 5, 6);
1554
1555         /*
1556          * Not expecting an error here, the check is only to prevent the
1557          * optimizer from removing the call to target() as otherwise there
1558          * are no side-effects and the call is never performed.
1559          */
1560         if (ret != 21)
1561                 warn++;
1562
1563         /* Disable trace points before removing it */
1564         tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1565         if (WARN_ON_ONCE(tk == NULL)) {
1566                 pr_warn("error on getting test probe.\n");
1567                 warn++;
1568         } else {
1569                 if (trace_kprobe_nhit(tk) != 1) {
1570                         pr_warn("incorrect number of testprobe hits\n");
1571                         warn++;
1572                 }
1573
1574                 file = find_trace_probe_file(tk, top_trace_array());
1575                 if (WARN_ON_ONCE(file == NULL)) {
1576                         pr_warn("error on getting probe file.\n");
1577                         warn++;
1578                 } else
1579                         disable_trace_kprobe(tk, file);
1580         }
1581
1582         tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1583         if (WARN_ON_ONCE(tk == NULL)) {
1584                 pr_warn("error on getting 2nd test probe.\n");
1585                 warn++;
1586         } else {
1587                 if (trace_kprobe_nhit(tk) != 1) {
1588                         pr_warn("incorrect number of testprobe2 hits\n");
1589                         warn++;
1590                 }
1591
1592                 file = find_trace_probe_file(tk, top_trace_array());
1593                 if (WARN_ON_ONCE(file == NULL)) {
1594                         pr_warn("error on getting probe file.\n");
1595                         warn++;
1596                 } else
1597                         disable_trace_kprobe(tk, file);
1598         }
1599
1600         ret = trace_run_command("-:testprobe", create_or_delete_trace_kprobe);
1601         if (WARN_ON_ONCE(ret)) {
1602                 pr_warn("error on deleting a probe.\n");
1603                 warn++;
1604         }
1605
1606         ret = trace_run_command("-:testprobe2", create_or_delete_trace_kprobe);
1607         if (WARN_ON_ONCE(ret)) {
1608                 pr_warn("error on deleting a probe.\n");
1609                 warn++;
1610         }
1611
1612 end:
1613         ret = dyn_events_release_all(&trace_kprobe_ops);
1614         if (WARN_ON_ONCE(ret)) {
1615                 pr_warn("error on cleaning up probes.\n");
1616                 warn++;
1617         }
1618         /*
1619          * Wait for the optimizer work to finish. Otherwise it might fiddle
1620          * with probes in already freed __init text.
1621          */
1622         wait_for_kprobe_optimizer();
1623         if (warn)
1624                 pr_cont("NG: Some tests are failed. Please check them.\n");
1625         else
1626                 pr_cont("OK\n");
1627         return 0;
1628 }
1629
1630 late_initcall(kprobe_trace_self_tests_init);
1631
1632 #endif