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 <asm/setup.h>  /* for COMMAND_LINE_SIZE */
  16
  17 #include "trace_dynevent.h"
  18 #include "trace_kprobe_selftest.h"
  19 #include "trace_probe.h"
  20 #include "trace_probe_tmpl.h"
  21
  22 #define KPROBE_EVENT_SYSTEM "kprobes"
  23 #define KRETPROBE_MAXACTIVE_MAX 4096
  24 #define MAX_KPROBE_CMDLINE_SIZE 1024
  25
  26 /* Kprobe early definition from command line */
  27 static char kprobe_boot_events_buf[COMMAND_LINE_SIZE] __initdata;
  28 static bool kprobe_boot_events_enabled __initdata;
  29
  30 static int __init set_kprobe_boot_events(char *str)
  31 {
  32         strlcpy(kprobe_boot_events_buf, str, COMMAND_LINE_SIZE);
  33         return 0;
  34 }
  35 __setup("kprobe_event=", set_kprobe_boot_events);
  36
  37 static int trace_kprobe_create(int argc, const char **argv);
  38 static int trace_kprobe_show(struct seq_file *m, struct dyn_event *ev);
  39 static int trace_kprobe_release(struct dyn_event *ev);
  40 static bool trace_kprobe_is_busy(struct dyn_event *ev);
  41 static bool trace_kprobe_match(const char *system, const char *event,
  42                         int argc, const char **argv, struct dyn_event *ev);
  43
  44 static struct dyn_event_operations trace_kprobe_ops = {
  45         .create = trace_kprobe_create,
  46         .show = trace_kprobe_show,
  47         .is_busy = trace_kprobe_is_busy,
  48         .free = trace_kprobe_release,
  49         .match = trace_kprobe_match,
  50 };
  51
  52 /*
  53  * Kprobe event core functions
  54  */
  55 struct trace_kprobe {
  56         struct dyn_event        devent;
  57         struct kretprobe        rp;     /* Use rp.kp for kprobe use */
  58         unsigned long __percpu *nhit;
  59         const char              *symbol;        /* symbol name */
  60         struct trace_probe      tp;
  61 };
  62
  63 static bool is_trace_kprobe(struct dyn_event *ev)
  64 {
  65         return ev->ops == &trace_kprobe_ops;
  66 }
  67
  68 static struct trace_kprobe *to_trace_kprobe(struct dyn_event *ev)
  69 {
  70         return container_of(ev, struct trace_kprobe, devent);
  71 }
  72
  73 /**
  74  * for_each_trace_kprobe - iterate over the trace_kprobe list
  75  * @pos:        the struct trace_kprobe * for each entry
  76  * @dpos:       the struct dyn_event * to use as a loop cursor
  77  */
  78 #define for_each_trace_kprobe(pos, dpos)        \
  79         for_each_dyn_event(dpos)                \
  80                 if (is_trace_kprobe(dpos) && (pos = to_trace_kprobe(dpos)))
  81
  82 #define SIZEOF_TRACE_KPROBE(n)                          \
  83         (offsetof(struct trace_kprobe, tp.args) +       \
  84         (sizeof(struct probe_arg) * (n)))
  85
  86 static nokprobe_inline bool trace_kprobe_is_return(struct trace_kprobe *tk)
  87 {
  88         return tk->rp.handler != NULL;
  89 }
  90
  91 static nokprobe_inline const char *trace_kprobe_symbol(struct trace_kprobe *tk)
  92 {
  93         return tk->symbol ? tk->symbol : "unknown";
  94 }
  95
  96 static nokprobe_inline unsigned long trace_kprobe_offset(struct trace_kprobe *tk)
  97 {
  98         return tk->rp.kp.offset;
  99 }
 100
 101 static nokprobe_inline bool trace_kprobe_has_gone(struct trace_kprobe *tk)
 102 {
 103         return !!(kprobe_gone(&tk->rp.kp));
 104 }
 105
 106 static nokprobe_inline bool trace_kprobe_within_module(struct trace_kprobe *tk,
 107                                                  struct module *mod)
 108 {
 109         int len = strlen(mod->name);
 110         const char *name = trace_kprobe_symbol(tk);
 111         return strncmp(mod->name, name, len) == 0 && name[len] == ':';
 112 }
 113
 114 static nokprobe_inline bool trace_kprobe_module_exist(struct trace_kprobe *tk)
 115 {
 116         char *p;
 117         bool ret;
 118
 119         if (!tk->symbol)
 120                 return false;
 121         p = strchr(tk->symbol, ':');
 122         if (!p)
 123                 return true;
 124         *p = '\0';
 125         mutex_lock(&module_mutex);
 126         ret = !!find_module(tk->symbol);
 127         mutex_unlock(&module_mutex);
 128         *p = ':';
 129
 130         return ret;
 131 }
 132
 133 static bool trace_kprobe_is_busy(struct dyn_event *ev)
 134 {
 135         struct trace_kprobe *tk = to_trace_kprobe(ev);
 136
 137         return trace_probe_is_enabled(&tk->tp);
 138 }
 139
 140 static bool trace_kprobe_match(const char *system, const char *event,
 141                         int argc, const char **argv, struct dyn_event *ev)
 142 {
 143         struct trace_kprobe *tk = to_trace_kprobe(ev);
 144
 145         return strcmp(trace_probe_name(&tk->tp), event) == 0 &&
 146             (!system || strcmp(trace_probe_group_name(&tk->tp), system) == 0);
 147 }
 148
 149 static nokprobe_inline unsigned long trace_kprobe_nhit(struct trace_kprobe *tk)
 150 {
 151         unsigned long nhit = 0;
 152         int cpu;
 153
 154         for_each_possible_cpu(cpu)
 155                 nhit += *per_cpu_ptr(tk->nhit, cpu);
 156
 157         return nhit;
 158 }
 159
 160 static nokprobe_inline bool trace_kprobe_is_registered(struct trace_kprobe *tk)
 161 {
 162         return !(list_empty(&tk->rp.kp.list) &&
 163                  hlist_unhashed(&tk->rp.kp.hlist));
 164 }
 165
 166 /* Return 0 if it fails to find the symbol address */
 167 static nokprobe_inline
 168 unsigned long trace_kprobe_address(struct trace_kprobe *tk)
 169 {
 170         unsigned long addr;
 171
 172         if (tk->symbol) {
 173                 addr = (unsigned long)
 174                         kallsyms_lookup_name(trace_kprobe_symbol(tk));
 175                 if (addr)
 176                         addr += tk->rp.kp.offset;
 177         } else {
 178                 addr = (unsigned long)tk->rp.kp.addr;
 179         }
 180         return addr;
 181 }
 182
 183 static nokprobe_inline struct trace_kprobe *
 184 trace_kprobe_primary_from_call(struct trace_event_call *call)
 185 {
 186         struct trace_probe *tp;
 187
 188         tp = trace_probe_primary_from_call(call);
 189         if (WARN_ON_ONCE(!tp))
 190                 return NULL;
 191
 192         return container_of(tp, struct trace_kprobe, tp);
 193 }
 194
 195 bool trace_kprobe_on_func_entry(struct trace_event_call *call)
 196 {
 197         struct trace_kprobe *tk = trace_kprobe_primary_from_call(call);
 198
 199         return tk ? kprobe_on_func_entry(tk->rp.kp.addr,
 200                         tk->rp.kp.addr ? NULL : tk->rp.kp.symbol_name,
 201                         tk->rp.kp.addr ? 0 : tk->rp.kp.offset) : false;
 202 }
 203
 204 bool trace_kprobe_error_injectable(struct trace_event_call *call)
 205 {
 206         struct trace_kprobe *tk = trace_kprobe_primary_from_call(call);
 207
 208         return tk ? within_error_injection_list(trace_kprobe_address(tk)) :
 209                false;
 210 }
 211
 212 static int register_kprobe_event(struct trace_kprobe *tk);
 213 static int unregister_kprobe_event(struct trace_kprobe *tk);
 214
 215 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
 216 static int kretprobe_dispatcher(struct kretprobe_instance *ri,
 217                                 struct pt_regs *regs);
 218
 219 static void free_trace_kprobe(struct trace_kprobe *tk)
 220 {
 221         if (tk) {
 222                 trace_probe_cleanup(&tk->tp);
 223                 kfree(tk->symbol);
 224                 free_percpu(tk->nhit);
 225                 kfree(tk);
 226         }
 227 }
 228
 229 /*
 230  * Allocate new trace_probe and initialize it (including kprobes).
 231  */
 232 static struct trace_kprobe *alloc_trace_kprobe(const char *group,
 233                                              const char *event,
 234                                              void *addr,
 235                                              const char *symbol,
 236                                              unsigned long offs,
 237                                              int maxactive,
 238                                              int nargs, bool is_return)
 239 {
 240         struct trace_kprobe *tk;
 241         int ret = -ENOMEM;
 242
 243         tk = kzalloc(SIZEOF_TRACE_KPROBE(nargs), GFP_KERNEL);
 244         if (!tk)
 245                 return ERR_PTR(ret);
 246
 247         tk->nhit = alloc_percpu(unsigned long);
 248         if (!tk->nhit)
 249                 goto error;
 250
 251         if (symbol) {
 252                 tk->symbol = kstrdup(symbol, GFP_KERNEL);
 253                 if (!tk->symbol)
 254                         goto error;
 255                 tk->rp.kp.symbol_name = tk->symbol;
 256                 tk->rp.kp.offset = offs;
 257         } else
 258                 tk->rp.kp.addr = addr;
 259
 260         if (is_return)
 261                 tk->rp.handler = kretprobe_dispatcher;
 262         else
 263                 tk->rp.kp.pre_handler = kprobe_dispatcher;
 264
 265         tk->rp.maxactive = maxactive;
 266         INIT_HLIST_NODE(&tk->rp.kp.hlist);
 267         INIT_LIST_HEAD(&tk->rp.kp.list);
 268
 269         ret = trace_probe_init(&tk->tp, event, group);
 270         if (ret < 0)
 271                 goto error;
 272
 273         dyn_event_init(&tk->devent, &trace_kprobe_ops);
 274         return tk;
 275 error:
 276         free_trace_kprobe(tk);
 277         return ERR_PTR(ret);
 278 }
 279
 280 static struct trace_kprobe *find_trace_kprobe(const char *event,
 281                                               const char *group)
 282 {
 283         struct dyn_event *pos;
 284         struct trace_kprobe *tk;
 285
 286         for_each_trace_kprobe(tk, pos)
 287                 if (strcmp(trace_probe_name(&tk->tp), event) == 0 &&
 288                     strcmp(trace_probe_group_name(&tk->tp), group) == 0)
 289                         return tk;
 290         return NULL;
 291 }
 292
 293 static inline int __enable_trace_kprobe(struct trace_kprobe *tk)
 294 {
 295         int ret = 0;
 296
 297         if (trace_kprobe_is_registered(tk) && !trace_kprobe_has_gone(tk)) {
 298                 if (trace_kprobe_is_return(tk))
 299                         ret = enable_kretprobe(&tk->rp);
 300                 else
 301                         ret = enable_kprobe(&tk->rp.kp);
 302         }
 303
 304         return ret;
 305 }
 306
 307 static void __disable_trace_kprobe(struct trace_probe *tp)
 308 {
 309         struct trace_probe *pos;
 310         struct trace_kprobe *tk;
 311
 312         list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
 313                 tk = container_of(pos, struct trace_kprobe, tp);
 314                 if (!trace_kprobe_is_registered(tk))
 315                         continue;
 316                 if (trace_kprobe_is_return(tk))
 317                         disable_kretprobe(&tk->rp);
 318                 else
 319                         disable_kprobe(&tk->rp.kp);
 320         }
 321 }
 322
 323 /*
 324  * Enable trace_probe
 325  * if the file is NULL, enable "perf" handler, or enable "trace" handler.
 326  */
 327 static int enable_trace_kprobe(struct trace_event_call *call,
 328                                 struct trace_event_file *file)
 329 {
 330         struct trace_probe *pos, *tp;
 331         struct trace_kprobe *tk;
 332         bool enabled;
 333         int ret = 0;
 334
 335         tp = trace_probe_primary_from_call(call);
 336         if (WARN_ON_ONCE(!tp))
 337                 return -ENODEV;
 338         enabled = trace_probe_is_enabled(tp);
 339
 340         /* This also changes "enabled" state */
 341         if (file) {
 342                 ret = trace_probe_add_file(tp, file);
 343                 if (ret)
 344                         return ret;
 345         } else
 346                 trace_probe_set_flag(tp, TP_FLAG_PROFILE);
 347
 348         if (enabled)
 349                 return 0;
 350
 351         enabled = false;
 352         list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
 353                 tk = container_of(pos, struct trace_kprobe, tp);
 354                 if (trace_kprobe_has_gone(tk))
 355                         continue;
 356                 ret = __enable_trace_kprobe(tk);
 357                 if (ret) {
 358                         if (enabled) {
 359                                 __disable_trace_kprobe(tp);
 360                                 enabled = false;
 361                         }
 362                         break;
 363                 }
 364                 enabled = true;
 365         }
 366
 367         if (!enabled) {
 368                 /* No probe is enabled. Roll back */
 369                 if (file)
 370                         trace_probe_remove_file(tp, file);
 371                 else
 372                         trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
 373                 if (!ret)
 374                         /* Since all probes are gone, this is not available */
 375                         ret = -EADDRNOTAVAIL;
 376         }
 377
 378         return ret;
 379 }
 380
 381 /*
 382  * Disable trace_probe
 383  * if the file is NULL, disable "perf" handler, or disable "trace" handler.
 384  */
 385 static int disable_trace_kprobe(struct trace_event_call *call,
 386                                 struct trace_event_file *file)
 387 {
 388         struct trace_probe *tp;
 389
 390         tp = trace_probe_primary_from_call(call);
 391         if (WARN_ON_ONCE(!tp))
 392                 return -ENODEV;
 393
 394         if (file) {
 395                 if (!trace_probe_get_file_link(tp, file))
 396                         return -ENOENT;
 397                 if (!trace_probe_has_single_file(tp))
 398                         goto out;
 399                 trace_probe_clear_flag(tp, TP_FLAG_TRACE);
 400         } else
 401                 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
 402
 403         if (!trace_probe_is_enabled(tp))
 404                 __disable_trace_kprobe(tp);
 405
 406  out:
 407         if (file)
 408                 /*
 409                  * Synchronization is done in below function. For perf event,
 410                  * file == NULL and perf_trace_event_unreg() calls
 411                  * tracepoint_synchronize_unregister() to ensure synchronize
 412                  * event. We don't need to care about it.
 413                  */
 414                 trace_probe_remove_file(tp, file);
 415
 416         return 0;
 417 }
 418
 419 #if defined(CONFIG_KPROBES_ON_FTRACE) && \
 420         !defined(CONFIG_KPROBE_EVENTS_ON_NOTRACE)
 421 static bool within_notrace_func(struct trace_kprobe *tk)
 422 {
 423         unsigned long offset, size, addr;
 424
 425         addr = trace_kprobe_address(tk);
 426         if (!addr || !kallsyms_lookup_size_offset(addr, &size, &offset))
 427                 return false;
 428
 429         /* Get the entry address of the target function */
 430         addr -= offset;
 431
 432         /*
 433          * Since ftrace_location_range() does inclusive range check, we need
 434          * to subtract 1 byte from the end address.
 435          */
 436         return !ftrace_location_range(addr, addr + size - 1);
 437 }
 438 #else
 439 #define within_notrace_func(tk) (false)
 440 #endif
 441
 442 /* Internal register function - just handle k*probes and flags */
 443 static int __register_trace_kprobe(struct trace_kprobe *tk)
 444 {
 445         int i, ret;
 446
 447         if (trace_kprobe_is_registered(tk))
 448                 return -EINVAL;
 449
 450         if (within_notrace_func(tk)) {
 451                 pr_warn("Could not probe notrace function %s\n",
 452                         trace_kprobe_symbol(tk));
 453                 return -EINVAL;
 454         }
 455
 456         for (i = 0; i < tk->tp.nr_args; i++) {
 457                 ret = traceprobe_update_arg(&tk->tp.args[i]);
 458                 if (ret)
 459                         return ret;
 460         }
 461
 462         /* Set/clear disabled flag according to tp->flag */
 463         if (trace_probe_is_enabled(&tk->tp))
 464                 tk->rp.kp.flags &= ~KPROBE_FLAG_DISABLED;
 465         else
 466                 tk->rp.kp.flags |= KPROBE_FLAG_DISABLED;
 467
 468         if (trace_kprobe_is_return(tk))
 469                 ret = register_kretprobe(&tk->rp);
 470         else
 471                 ret = register_kprobe(&tk->rp.kp);
 472
 473         return ret;
 474 }
 475
 476 /* Internal unregister function - just handle k*probes and flags */
 477 static void __unregister_trace_kprobe(struct trace_kprobe *tk)
 478 {
 479         if (trace_kprobe_is_registered(tk)) {
 480                 if (trace_kprobe_is_return(tk))
 481                         unregister_kretprobe(&tk->rp);
 482                 else
 483                         unregister_kprobe(&tk->rp.kp);
 484                 /* Cleanup kprobe for reuse and mark it unregistered */
 485                 INIT_HLIST_NODE(&tk->rp.kp.hlist);
 486                 INIT_LIST_HEAD(&tk->rp.kp.list);
 487                 if (tk->rp.kp.symbol_name)
 488                         tk->rp.kp.addr = NULL;
 489         }
 490 }
 491
 492 /* Unregister a trace_probe and probe_event */
 493 static int unregister_trace_kprobe(struct trace_kprobe *tk)
 494 {
 495         /* If other probes are on the event, just unregister kprobe */
 496         if (trace_probe_has_sibling(&tk->tp))
 497                 goto unreg;
 498
 499         /* Enabled event can not be unregistered */
 500         if (trace_probe_is_enabled(&tk->tp))
 501                 return -EBUSY;
 502
 503         /* Will fail if probe is being used by ftrace or perf */
 504         if (unregister_kprobe_event(tk))
 505                 return -EBUSY;
 506
 507 unreg:
 508         __unregister_trace_kprobe(tk);
 509         dyn_event_remove(&tk->devent);
 510         trace_probe_unlink(&tk->tp);
 511
 512         return 0;
 513 }
 514
 515 static int append_trace_kprobe(struct trace_kprobe *tk, struct trace_kprobe *to)
 516 {
 517         int ret;
 518
 519         /* Append to existing event */
 520         ret = trace_probe_append(&tk->tp, &to->tp);
 521         if (ret)
 522                 return ret;
 523
 524         /* Register k*probe */
 525         ret = __register_trace_kprobe(tk);
 526         if (ret == -ENOENT && !trace_kprobe_module_exist(tk)) {
 527                 pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
 528                 ret = 0;
 529         }
 530
 531         if (ret)
 532                 trace_probe_unlink(&tk->tp);
 533         else
 534                 dyn_event_add(&tk->devent);
 535
 536         return ret;
 537 }
 538
 539 /* Register a trace_probe and probe_event */
 540 static int register_trace_kprobe(struct trace_kprobe *tk)
 541 {
 542         struct trace_kprobe *old_tk;
 543         int ret;
 544
 545         mutex_lock(&event_mutex);
 546
 547         old_tk = find_trace_kprobe(trace_probe_name(&tk->tp),
 548                                    trace_probe_group_name(&tk->tp));
 549         if (old_tk) {
 550                 if (trace_kprobe_is_return(tk) != trace_kprobe_is_return(old_tk)) {
 551                         trace_probe_log_set_index(0);
 552                         trace_probe_log_err(0, DIFF_PROBE_TYPE);
 553                         ret = -EEXIST;
 554                 } else {
 555                         ret = trace_probe_compare_arg_type(&tk->tp, &old_tk->tp);
 556                         if (ret) {
 557                                 /* Note that argument starts index = 2 */
 558                                 trace_probe_log_set_index(ret + 1);
 559                                 trace_probe_log_err(0, DIFF_ARG_TYPE);
 560                                 ret = -EEXIST;
 561                         } else
 562                                 ret = append_trace_kprobe(tk, old_tk);
 563                 }
 564                 goto end;
 565         }
 566
 567         /* Register new event */
 568         ret = register_kprobe_event(tk);
 569         if (ret) {
 570                 pr_warn("Failed to register probe event(%d)\n", ret);
 571                 goto end;
 572         }
 573
 574         /* Register k*probe */
 575         ret = __register_trace_kprobe(tk);
 576         if (ret == -ENOENT && !trace_kprobe_module_exist(tk)) {
 577                 pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
 578                 ret = 0;
 579         }
 580
 581         if (ret < 0)
 582                 unregister_kprobe_event(tk);
 583         else
 584                 dyn_event_add(&tk->devent);
 585
 586 end:
 587         mutex_unlock(&event_mutex);
 588         return ret;
 589 }
 590
 591 /* Module notifier call back, checking event on the module */
 592 static int trace_kprobe_module_callback(struct notifier_block *nb,
 593                                        unsigned long val, void *data)
 594 {
 595         struct module *mod = data;
 596         struct dyn_event *pos;
 597         struct trace_kprobe *tk;
 598         int ret;
 599
 600         if (val != MODULE_STATE_COMING)
 601                 return NOTIFY_DONE;
 602
 603         /* Update probes on coming module */
 604         mutex_lock(&event_mutex);
 605         for_each_trace_kprobe(tk, pos) {
 606                 if (trace_kprobe_within_module(tk, mod)) {
 607                         /* Don't need to check busy - this should have gone. */
 608                         __unregister_trace_kprobe(tk);
 609                         ret = __register_trace_kprobe(tk);
 610                         if (ret)
 611                                 pr_warn("Failed to re-register probe %s on %s: %d\n",
 612                                         trace_probe_name(&tk->tp),
 613                                         mod->name, ret);
 614                 }
 615         }
 616         mutex_unlock(&event_mutex);
 617
 618         return NOTIFY_DONE;
 619 }
 620
 621 static struct notifier_block trace_kprobe_module_nb = {
 622         .notifier_call = trace_kprobe_module_callback,
 623         .priority = 1   /* Invoked after kprobe module callback */
 624 };
 625
 626 /* Convert certain expected symbols into '_' when generating event names */
 627 static inline void sanitize_event_name(char *name)
 628 {
 629         while (*name++ != '\0')
 630                 if (*name == ':' || *name == '.')
 631                         *name = '_';
 632 }
 633
 634 static int trace_kprobe_create(int argc, const char *argv[])
 635 {
 636         /*
 637          * Argument syntax:
 638          *  - Add kprobe:
 639          *      p[:[GRP/]EVENT] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
 640          *  - Add kretprobe:
 641          *      r[MAXACTIVE][:[GRP/]EVENT] [MOD:]KSYM[+0] [FETCHARGS]
 642          * Fetch args:
 643          *  $retval     : fetch return value
 644          *  $stack      : fetch stack address
 645          *  $stackN     : fetch Nth of stack (N:0-)
 646          *  $comm       : fetch current task comm
 647          *  @ADDR       : fetch memory at ADDR (ADDR should be in kernel)
 648          *  @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
 649          *  %REG        : fetch register REG
 650          * Dereferencing memory fetch:
 651          *  +|-offs(ARG) : fetch memory at ARG +|- offs address.
 652          * Alias name of args:
 653          *  NAME=FETCHARG : set NAME as alias of FETCHARG.
 654          * Type of args:
 655          *  FETCHARG:TYPE : use TYPE instead of unsigned long.
 656          */
 657         struct trace_kprobe *tk = NULL;
 658         int i, len, ret = 0;
 659         bool is_return = false;
 660         char *symbol = NULL, *tmp = NULL;
 661         const char *event = NULL, *group = KPROBE_EVENT_SYSTEM;
 662         int maxactive = 0;
 663         long offset = 0;
 664         void *addr = NULL;
 665         char buf[MAX_EVENT_NAME_LEN];
 666         unsigned int flags = TPARG_FL_KERNEL;
 667
 668         switch (argv[0][0]) {
 669         case 'r':
 670                 is_return = true;
 671                 flags |= TPARG_FL_RETURN;
 672                 break;
 673         case 'p':
 674                 break;
 675         default:
 676                 return -ECANCELED;
 677         }
 678         if (argc < 2)
 679                 return -ECANCELED;
 680
 681         trace_probe_log_init("trace_kprobe", argc, argv);
 682
 683         event = strchr(&argv[0][1], ':');
 684         if (event)
 685                 event++;
 686
 687         if (isdigit(argv[0][1])) {
 688                 if (!is_return) {
 689                         trace_probe_log_err(1, MAXACT_NO_KPROBE);
 690                         goto parse_error;
 691                 }
 692                 if (event)
 693                         len = event - &argv[0][1] - 1;
 694                 else
 695                         len = strlen(&argv[0][1]);
 696                 if (len > MAX_EVENT_NAME_LEN - 1) {
 697                         trace_probe_log_err(1, BAD_MAXACT);
 698                         goto parse_error;
 699                 }
 700                 memcpy(buf, &argv[0][1], len);
 701                 buf[len] = '\0';
 702                 ret = kstrtouint(buf, 0, &maxactive);
 703                 if (ret || !maxactive) {
 704                         trace_probe_log_err(1, BAD_MAXACT);
 705                         goto parse_error;
 706                 }
 707                 /* kretprobes instances are iterated over via a list. The
 708                  * maximum should stay reasonable.
 709                  */
 710                 if (maxactive > KRETPROBE_MAXACTIVE_MAX) {
 711                         trace_probe_log_err(1, MAXACT_TOO_BIG);
 712                         goto parse_error;
 713                 }
 714         }
 715
 716         /* try to parse an address. if that fails, try to read the
 717          * input as a symbol. */
 718         if (kstrtoul(argv[1], 0, (unsigned long *)&addr)) {
 719                 trace_probe_log_set_index(1);
 720                 /* Check whether uprobe event specified */
 721                 if (strchr(argv[1], '/') && strchr(argv[1], ':')) {
 722                         ret = -ECANCELED;
 723                         goto error;
 724                 }
 725                 /* a symbol specified */
 726                 symbol = kstrdup(argv[1], GFP_KERNEL);
 727                 if (!symbol)
 728                         return -ENOMEM;
 729                 /* TODO: support .init module functions */
 730                 ret = traceprobe_split_symbol_offset(symbol, &offset);
 731                 if (ret || offset < 0 || offset > UINT_MAX) {
 732                         trace_probe_log_err(0, BAD_PROBE_ADDR);
 733                         goto parse_error;
 734                 }
 735                 if (kprobe_on_func_entry(NULL, symbol, offset))
 736                         flags |= TPARG_FL_FENTRY;
 737                 if (offset && is_return && !(flags & TPARG_FL_FENTRY)) {
 738                         trace_probe_log_err(0, BAD_RETPROBE);
 739                         goto parse_error;
 740                 }
 741         }
 742
 743         trace_probe_log_set_index(0);
 744         if (event) {
 745                 ret = traceprobe_parse_event_name(&event, &group, buf,
 746                                                   event - argv[0]);
 747                 if (ret)
 748                         goto parse_error;
 749         } else {
 750                 /* Make a new event name */
 751                 if (symbol)
 752                         snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
 753                                  is_return ? 'r' : 'p', symbol, offset);
 754                 else
 755                         snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
 756                                  is_return ? 'r' : 'p', addr);
 757                 sanitize_event_name(buf);
 758                 event = buf;
 759         }
 760
 761         /* setup a probe */
 762         tk = alloc_trace_kprobe(group, event, addr, symbol, offset, maxactive,
 763                                argc - 2, is_return);
 764         if (IS_ERR(tk)) {
 765                 ret = PTR_ERR(tk);
 766                 /* This must return -ENOMEM, else there is a bug */
 767                 WARN_ON_ONCE(ret != -ENOMEM);
 768                 goto out;       /* We know tk is not allocated */
 769         }
 770         argc -= 2; argv += 2;
 771
 772         /* parse arguments */
 773         for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
 774                 tmp = kstrdup(argv[i], GFP_KERNEL);
 775                 if (!tmp) {
 776                         ret = -ENOMEM;
 777                         goto error;
 778                 }
 779
 780                 trace_probe_log_set_index(i + 2);
 781                 ret = traceprobe_parse_probe_arg(&tk->tp, i, tmp, flags);
 782                 kfree(tmp);
 783                 if (ret)
 784                         goto error;     /* This can be -ENOMEM */
 785         }
 786
 787         ret = traceprobe_set_print_fmt(&tk->tp, is_return);
 788         if (ret < 0)
 789                 goto error;
 790
 791         ret = register_trace_kprobe(tk);
 792         if (ret) {
 793                 trace_probe_log_set_index(1);
 794                 if (ret == -EILSEQ)
 795                         trace_probe_log_err(0, BAD_INSN_BNDRY);
 796                 else if (ret == -ENOENT)
 797                         trace_probe_log_err(0, BAD_PROBE_ADDR);
 798                 else if (ret != -ENOMEM && ret != -EEXIST)
 799                         trace_probe_log_err(0, FAIL_REG_PROBE);
 800                 goto error;
 801         }
 802
 803 out:
 804         trace_probe_log_clear();
 805         kfree(symbol);
 806         return ret;
 807
 808 parse_error:
 809         ret = -EINVAL;
 810 error:
 811         free_trace_kprobe(tk);
 812         goto out;
 813 }
 814
 815 static int create_or_delete_trace_kprobe(int argc, char **argv)
 816 {
 817         int ret;
 818
 819         if (argv[0][0] == '-')
 820                 return dyn_event_release(argc, argv, &trace_kprobe_ops);
 821
 822         ret = trace_kprobe_create(argc, (const char **)argv);
 823         return ret == -ECANCELED ? -EINVAL : ret;
 824 }
 825
 826 static int trace_kprobe_release(struct dyn_event *ev)
 827 {
 828         struct trace_kprobe *tk = to_trace_kprobe(ev);
 829         int ret = unregister_trace_kprobe(tk);
 830
 831         if (!ret)
 832                 free_trace_kprobe(tk);
 833         return ret;
 834 }
 835
 836 static int trace_kprobe_show(struct seq_file *m, struct dyn_event *ev)
 837 {
 838         struct trace_kprobe *tk = to_trace_kprobe(ev);
 839         int i;
 840
 841         seq_putc(m, trace_kprobe_is_return(tk) ? 'r' : 'p');
 842         seq_printf(m, ":%s/%s", trace_probe_group_name(&tk->tp),
 843                                 trace_probe_name(&tk->tp));
 844
 845         if (!tk->symbol)
 846                 seq_printf(m, " 0x%p", tk->rp.kp.addr);
 847         else if (tk->rp.kp.offset)
 848                 seq_printf(m, " %s+%u", trace_kprobe_symbol(tk),
 849                            tk->rp.kp.offset);
 850         else
 851                 seq_printf(m, " %s", trace_kprobe_symbol(tk));
 852
 853         for (i = 0; i < tk->tp.nr_args; i++)
 854                 seq_printf(m, " %s=%s", tk->tp.args[i].name, tk->tp.args[i].comm);
 855         seq_putc(m, '\n');
 856
 857         return 0;
 858 }
 859
 860 static int probes_seq_show(struct seq_file *m, void *v)
 861 {
 862         struct dyn_event *ev = v;
 863
 864         if (!is_trace_kprobe(ev))
 865                 return 0;
 866
 867         return trace_kprobe_show(m, ev);
 868 }
 869
 870 static const struct seq_operations probes_seq_op = {
 871         .start  = dyn_event_seq_start,
 872         .next   = dyn_event_seq_next,
 873         .stop   = dyn_event_seq_stop,
 874         .show   = probes_seq_show
 875 };
 876
 877 static int probes_open(struct inode *inode, struct file *file)
 878 {
 879         int ret;
 880
 881         if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
 882                 ret = dyn_events_release_all(&trace_kprobe_ops);
 883                 if (ret < 0)
 884                         return ret;
 885         }
 886
 887         return seq_open(file, &probes_seq_op);
 888 }
 889
 890 static ssize_t probes_write(struct file *file, const char __user *buffer,
 891                             size_t count, loff_t *ppos)
 892 {
 893         return trace_parse_run_command(file, buffer, count, ppos,
 894                                        create_or_delete_trace_kprobe);
 895 }
 896
 897 static const struct file_operations kprobe_events_ops = {
 898         .owner          = THIS_MODULE,
 899         .open           = probes_open,
 900         .read           = seq_read,
 901         .llseek         = seq_lseek,
 902         .release        = seq_release,
 903         .write          = probes_write,
 904 };
 905
 906 /* Probes profiling interfaces */
 907 static int probes_profile_seq_show(struct seq_file *m, void *v)
 908 {
 909         struct dyn_event *ev = v;
 910         struct trace_kprobe *tk;
 911
 912         if (!is_trace_kprobe(ev))
 913                 return 0;
 914
 915         tk = to_trace_kprobe(ev);
 916         seq_printf(m, "  %-44s %15lu %15lu\n",
 917                    trace_probe_name(&tk->tp),
 918                    trace_kprobe_nhit(tk),
 919                    tk->rp.kp.nmissed);
 920
 921         return 0;
 922 }
 923
 924 static const struct seq_operations profile_seq_op = {
 925         .start  = dyn_event_seq_start,
 926         .next   = dyn_event_seq_next,
 927         .stop   = dyn_event_seq_stop,
 928         .show   = probes_profile_seq_show
 929 };
 930
 931 static int profile_open(struct inode *inode, struct file *file)
 932 {
 933         return seq_open(file, &profile_seq_op);
 934 }
 935
 936 static const struct file_operations kprobe_profile_ops = {
 937         .owner          = THIS_MODULE,
 938         .open           = profile_open,
 939         .read           = seq_read,
 940         .llseek         = seq_lseek,
 941         .release        = seq_release,
 942 };
 943
 944 /* Kprobe specific fetch functions */
 945
 946 /* Return the length of string -- including null terminal byte */
 947 static nokprobe_inline int
 948 fetch_store_strlen(unsigned long addr)
 949 {
 950         int ret, len = 0;
 951         u8 c;
 952
 953         do {
 954                 ret = probe_kernel_read(&c, (u8 *)addr + len, 1);
 955                 len++;
 956         } while (c && ret == 0 && len < MAX_STRING_SIZE);
 957
 958         return (ret < 0) ? ret : len;
 959 }
 960
 961 /* Return the length of string -- including null terminal byte */
 962 static nokprobe_inline int
 963 fetch_store_strlen_user(unsigned long addr)
 964 {
 965         const void __user *uaddr =  (__force const void __user *)addr;
 966
 967         return strnlen_unsafe_user(uaddr, MAX_STRING_SIZE);
 968 }
 969
 970 /*
 971  * Fetch a null-terminated string. Caller MUST set *(u32 *)buf with max
 972  * length and relative data location.
 973  */
 974 static nokprobe_inline int
 975 fetch_store_string(unsigned long addr, void *dest, void *base)
 976 {
 977         int maxlen = get_loc_len(*(u32 *)dest);
 978         void *__dest;
 979         long ret;
 980
 981         if (unlikely(!maxlen))
 982                 return -ENOMEM;
 983
 984         __dest = get_loc_data(dest, base);
 985
 986         /*
 987          * Try to get string again, since the string can be changed while
 988          * probing.
 989          */
 990         ret = strncpy_from_unsafe(__dest, (void *)addr, maxlen);
 991         if (ret >= 0)
 992                 *(u32 *)dest = make_data_loc(ret, __dest - base);
 993
 994         return ret;
 995 }
 996
 997 /*
 998  * Fetch a null-terminated string from user. Caller MUST set *(u32 *)buf
 999  * with max length and relative data location.
1000  */
1001 static nokprobe_inline int
1002 fetch_store_string_user(unsigned long addr, void *dest, void *base)
1003 {
1004         const void __user *uaddr =  (__force const void __user *)addr;
1005         int maxlen = get_loc_len(*(u32 *)dest);
1006         void *__dest;
1007         long ret;
1008
1009         if (unlikely(!maxlen))
1010                 return -ENOMEM;
1011
1012         __dest = get_loc_data(dest, base);
1013
1014         ret = strncpy_from_unsafe_user(__dest, uaddr, maxlen);
1015         if (ret >= 0)
1016                 *(u32 *)dest = make_data_loc(ret, __dest - base);
1017
1018         return ret;
1019 }
1020
1021 static nokprobe_inline int
1022 probe_mem_read(void *dest, void *src, size_t size)
1023 {
1024         return probe_kernel_read(dest, src, size);
1025 }
1026
1027 static nokprobe_inline int
1028 probe_mem_read_user(void *dest, void *src, size_t size)
1029 {
1030         const void __user *uaddr =  (__force const void __user *)src;
1031
1032         return probe_user_read(dest, uaddr, size);
1033 }
1034
1035 /* Note that we don't verify it, since the code does not come from user space */
1036 static int
1037 process_fetch_insn(struct fetch_insn *code, struct pt_regs *regs, void *dest,
1038                    void *base)
1039 {
1040         unsigned long val;
1041
1042 retry:
1043         /* 1st stage: get value from context */
1044         switch (code->op) {
1045         case FETCH_OP_REG:
1046                 val = regs_get_register(regs, code->param);
1047                 break;
1048         case FETCH_OP_STACK:
1049                 val = regs_get_kernel_stack_nth(regs, code->param);
1050                 break;
1051         case FETCH_OP_STACKP:
1052                 val = kernel_stack_pointer(regs);
1053                 break;
1054         case FETCH_OP_RETVAL:
1055                 val = regs_return_value(regs);
1056                 break;
1057         case FETCH_OP_IMM:
1058                 val = code->immediate;
1059                 break;
1060         case FETCH_OP_COMM:
1061                 val = (unsigned long)current->comm;
1062                 break;
1063 #ifdef CONFIG_HAVE_FUNCTION_ARG_ACCESS_API
1064         case FETCH_OP_ARG:
1065                 val = regs_get_kernel_argument(regs, code->param);
1066                 break;
1067 #endif
1068         case FETCH_NOP_SYMBOL:  /* Ignore a place holder */
1069                 code++;
1070                 goto retry;
1071         default:
1072                 return -EILSEQ;
1073         }
1074         code++;
1075
1076         return process_fetch_insn_bottom(code, val, dest, base);
1077 }
1078 NOKPROBE_SYMBOL(process_fetch_insn)
1079
1080 /* Kprobe handler */
1081 static nokprobe_inline void
1082 __kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs,
1083                     struct trace_event_file *trace_file)
1084 {
1085         struct kprobe_trace_entry_head *entry;
1086         struct ring_buffer_event *event;
1087         struct ring_buffer *buffer;
1088         int size, dsize, pc;
1089         unsigned long irq_flags;
1090         struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1091
1092         WARN_ON(call != trace_file->event_call);
1093
1094         if (trace_trigger_soft_disabled(trace_file))
1095                 return;
1096
1097         local_save_flags(irq_flags);
1098         pc = preempt_count();
1099
1100         dsize = __get_data_size(&tk->tp, regs);
1101         size = sizeof(*entry) + tk->tp.size + dsize;
1102
1103         event = trace_event_buffer_lock_reserve(&buffer, trace_file,
1104                                                 call->event.type,
1105                                                 size, irq_flags, pc);
1106         if (!event)
1107                 return;
1108
1109         entry = ring_buffer_event_data(event);
1110         entry->ip = (unsigned long)tk->rp.kp.addr;
1111         store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1112
1113         event_trigger_unlock_commit_regs(trace_file, buffer, event,
1114                                          entry, irq_flags, pc, regs);
1115 }
1116
1117 static void
1118 kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs)
1119 {
1120         struct event_file_link *link;
1121
1122         trace_probe_for_each_link_rcu(link, &tk->tp)
1123                 __kprobe_trace_func(tk, regs, link->file);
1124 }
1125 NOKPROBE_SYMBOL(kprobe_trace_func);
1126
1127 /* Kretprobe handler */
1128 static nokprobe_inline void
1129 __kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1130                        struct pt_regs *regs,
1131                        struct trace_event_file *trace_file)
1132 {
1133         struct kretprobe_trace_entry_head *entry;
1134         struct ring_buffer_event *event;
1135         struct ring_buffer *buffer;
1136         int size, pc, dsize;
1137         unsigned long irq_flags;
1138         struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1139
1140         WARN_ON(call != trace_file->event_call);
1141
1142         if (trace_trigger_soft_disabled(trace_file))
1143                 return;
1144
1145         local_save_flags(irq_flags);
1146         pc = preempt_count();
1147
1148         dsize = __get_data_size(&tk->tp, regs);
1149         size = sizeof(*entry) + tk->tp.size + dsize;
1150
1151         event = trace_event_buffer_lock_reserve(&buffer, trace_file,
1152                                                 call->event.type,
1153                                                 size, irq_flags, pc);
1154         if (!event)
1155                 return;
1156
1157         entry = ring_buffer_event_data(event);
1158         entry->func = (unsigned long)tk->rp.kp.addr;
1159         entry->ret_ip = (unsigned long)ri->ret_addr;
1160         store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1161
1162         event_trigger_unlock_commit_regs(trace_file, buffer, event,
1163                                          entry, irq_flags, pc, regs);
1164 }
1165
1166 static void
1167 kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1168                      struct pt_regs *regs)
1169 {
1170         struct event_file_link *link;
1171
1172         trace_probe_for_each_link_rcu(link, &tk->tp)
1173                 __kretprobe_trace_func(tk, ri, regs, link->file);
1174 }
1175 NOKPROBE_SYMBOL(kretprobe_trace_func);
1176
1177 /* Event entry printers */
1178 static enum print_line_t
1179 print_kprobe_event(struct trace_iterator *iter, int flags,
1180                    struct trace_event *event)
1181 {
1182         struct kprobe_trace_entry_head *field;
1183         struct trace_seq *s = &iter->seq;
1184         struct trace_probe *tp;
1185
1186         field = (struct kprobe_trace_entry_head *)iter->ent;
1187         tp = trace_probe_primary_from_call(
1188                 container_of(event, struct trace_event_call, event));
1189         if (WARN_ON_ONCE(!tp))
1190                 goto out;
1191
1192         trace_seq_printf(s, "%s: (", trace_probe_name(tp));
1193
1194         if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
1195                 goto out;
1196
1197         trace_seq_putc(s, ')');
1198
1199         if (print_probe_args(s, tp->args, tp->nr_args,
1200                              (u8 *)&field[1], field) < 0)
1201                 goto out;
1202
1203         trace_seq_putc(s, '\n');
1204  out:
1205         return trace_handle_return(s);
1206 }
1207
1208 static enum print_line_t
1209 print_kretprobe_event(struct trace_iterator *iter, int flags,
1210                       struct trace_event *event)
1211 {
1212         struct kretprobe_trace_entry_head *field;
1213         struct trace_seq *s = &iter->seq;
1214         struct trace_probe *tp;
1215
1216         field = (struct kretprobe_trace_entry_head *)iter->ent;
1217         tp = trace_probe_primary_from_call(
1218                 container_of(event, struct trace_event_call, event));
1219         if (WARN_ON_ONCE(!tp))
1220                 goto out;
1221
1222         trace_seq_printf(s, "%s: (", trace_probe_name(tp));
1223
1224         if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1225                 goto out;
1226
1227         trace_seq_puts(s, " <- ");
1228
1229         if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1230                 goto out;
1231
1232         trace_seq_putc(s, ')');
1233
1234         if (print_probe_args(s, tp->args, tp->nr_args,
1235                              (u8 *)&field[1], field) < 0)
1236                 goto out;
1237
1238         trace_seq_putc(s, '\n');
1239
1240  out:
1241         return trace_handle_return(s);
1242 }
1243
1244
1245 static int kprobe_event_define_fields(struct trace_event_call *event_call)
1246 {
1247         int ret;
1248         struct kprobe_trace_entry_head field;
1249         struct trace_probe *tp;
1250
1251         tp = trace_probe_primary_from_call(event_call);
1252         if (WARN_ON_ONCE(!tp))
1253                 return -ENOENT;
1254
1255         DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1256
1257         return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
1258 }
1259
1260 static int kretprobe_event_define_fields(struct trace_event_call *event_call)
1261 {
1262         int ret;
1263         struct kretprobe_trace_entry_head field;
1264         struct trace_probe *tp;
1265
1266         tp = trace_probe_primary_from_call(event_call);
1267         if (WARN_ON_ONCE(!tp))
1268                 return -ENOENT;
1269
1270         DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1271         DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1272
1273         return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
1274 }
1275
1276 #ifdef CONFIG_PERF_EVENTS
1277
1278 /* Kprobe profile handler */
1279 static int
1280 kprobe_perf_func(struct trace_kprobe *tk, struct pt_regs *regs)
1281 {
1282         struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1283         struct kprobe_trace_entry_head *entry;
1284         struct hlist_head *head;
1285         int size, __size, dsize;
1286         int rctx;
1287
1288         if (bpf_prog_array_valid(call)) {
1289                 unsigned long orig_ip = instruction_pointer(regs);
1290                 int ret;
1291
1292                 ret = trace_call_bpf(call, regs);
1293
1294                 /*
1295                  * We need to check and see if we modified the pc of the
1296                  * pt_regs, and if so return 1 so that we don't do the
1297                  * single stepping.
1298                  */
1299                 if (orig_ip != instruction_pointer(regs))
1300                         return 1;
1301                 if (!ret)
1302                         return 0;
1303         }
1304
1305         head = this_cpu_ptr(call->perf_events);
1306         if (hlist_empty(head))
1307                 return 0;
1308
1309         dsize = __get_data_size(&tk->tp, regs);
1310         __size = sizeof(*entry) + tk->tp.size + dsize;
1311         size = ALIGN(__size + sizeof(u32), sizeof(u64));
1312         size -= sizeof(u32);
1313
1314         entry = perf_trace_buf_alloc(size, NULL, &rctx);
1315         if (!entry)
1316                 return 0;
1317
1318         entry->ip = (unsigned long)tk->rp.kp.addr;
1319         memset(&entry[1], 0, dsize);
1320         store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1321         perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1322                               head, NULL);
1323         return 0;
1324 }
1325 NOKPROBE_SYMBOL(kprobe_perf_func);
1326
1327 /* Kretprobe profile handler */
1328 static void
1329 kretprobe_perf_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1330                     struct pt_regs *regs)
1331 {
1332         struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1333         struct kretprobe_trace_entry_head *entry;
1334         struct hlist_head *head;
1335         int size, __size, dsize;
1336         int rctx;
1337
1338         if (bpf_prog_array_valid(call) && !trace_call_bpf(call, regs))
1339                 return;
1340
1341         head = this_cpu_ptr(call->perf_events);
1342         if (hlist_empty(head))
1343                 return;
1344
1345         dsize = __get_data_size(&tk->tp, regs);
1346         __size = sizeof(*entry) + tk->tp.size + dsize;
1347         size = ALIGN(__size + sizeof(u32), sizeof(u64));
1348         size -= sizeof(u32);
1349
1350         entry = perf_trace_buf_alloc(size, NULL, &rctx);
1351         if (!entry)
1352                 return;
1353
1354         entry->func = (unsigned long)tk->rp.kp.addr;
1355         entry->ret_ip = (unsigned long)ri->ret_addr;
1356         store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1357         perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1358                               head, NULL);
1359 }
1360 NOKPROBE_SYMBOL(kretprobe_perf_func);
1361
1362 int bpf_get_kprobe_info(const struct perf_event *event, u32 *fd_type,
1363                         const char **symbol, u64 *probe_offset,
1364                         u64 *probe_addr, bool perf_type_tracepoint)
1365 {
1366         const char *pevent = trace_event_name(event->tp_event);
1367         const char *group = event->tp_event->class->system;
1368         struct trace_kprobe *tk;
1369
1370         if (perf_type_tracepoint)
1371                 tk = find_trace_kprobe(pevent, group);
1372         else
1373                 tk = event->tp_event->data;
1374         if (!tk)
1375                 return -EINVAL;
1376
1377         *fd_type = trace_kprobe_is_return(tk) ? BPF_FD_TYPE_KRETPROBE
1378                                               : BPF_FD_TYPE_KPROBE;
1379         if (tk->symbol) {
1380                 *symbol = tk->symbol;
1381                 *probe_offset = tk->rp.kp.offset;
1382                 *probe_addr = 0;
1383         } else {
1384                 *symbol = NULL;
1385                 *probe_offset = 0;
1386                 *probe_addr = (unsigned long)tk->rp.kp.addr;
1387         }
1388         return 0;
1389 }
1390 #endif  /* CONFIG_PERF_EVENTS */
1391
1392 /*
1393  * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex.
1394  *
1395  * kprobe_trace_self_tests_init() does enable_trace_probe/disable_trace_probe
1396  * lockless, but we can't race with this __init function.
1397  */
1398 static int kprobe_register(struct trace_event_call *event,
1399                            enum trace_reg type, void *data)
1400 {
1401         struct trace_event_file *file = data;
1402
1403         switch (type) {
1404         case TRACE_REG_REGISTER:
1405                 return enable_trace_kprobe(event, file);
1406         case TRACE_REG_UNREGISTER:
1407                 return disable_trace_kprobe(event, file);
1408
1409 #ifdef CONFIG_PERF_EVENTS
1410         case TRACE_REG_PERF_REGISTER:
1411                 return enable_trace_kprobe(event, NULL);
1412         case TRACE_REG_PERF_UNREGISTER:
1413                 return disable_trace_kprobe(event, NULL);
1414         case TRACE_REG_PERF_OPEN:
1415         case TRACE_REG_PERF_CLOSE:
1416         case TRACE_REG_PERF_ADD:
1417         case TRACE_REG_PERF_DEL:
1418                 return 0;
1419 #endif
1420         }
1421         return 0;
1422 }
1423
1424 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1425 {
1426         struct trace_kprobe *tk = container_of(kp, struct trace_kprobe, rp.kp);
1427         int ret = 0;
1428
1429         raw_cpu_inc(*tk->nhit);
1430
1431         if (trace_probe_test_flag(&tk->tp, TP_FLAG_TRACE))
1432                 kprobe_trace_func(tk, regs);
1433 #ifdef CONFIG_PERF_EVENTS
1434         if (trace_probe_test_flag(&tk->tp, TP_FLAG_PROFILE))
1435                 ret = kprobe_perf_func(tk, regs);
1436 #endif
1437         return ret;
1438 }
1439 NOKPROBE_SYMBOL(kprobe_dispatcher);
1440
1441 static int
1442 kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1443 {
1444         struct trace_kprobe *tk = container_of(ri->rp, struct trace_kprobe, rp);
1445
1446         raw_cpu_inc(*tk->nhit);
1447
1448         if (trace_probe_test_flag(&tk->tp, TP_FLAG_TRACE))
1449                 kretprobe_trace_func(tk, ri, regs);
1450 #ifdef CONFIG_PERF_EVENTS
1451         if (trace_probe_test_flag(&tk->tp, TP_FLAG_PROFILE))
1452                 kretprobe_perf_func(tk, ri, regs);
1453 #endif
1454         return 0;       /* We don't tweek kernel, so just return 0 */
1455 }
1456 NOKPROBE_SYMBOL(kretprobe_dispatcher);
1457
1458 static struct trace_event_functions kretprobe_funcs = {
1459         .trace          = print_kretprobe_event
1460 };
1461
1462 static struct trace_event_functions kprobe_funcs = {
1463         .trace          = print_kprobe_event
1464 };
1465
1466 static inline void init_trace_event_call(struct trace_kprobe *tk)
1467 {
1468         struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1469
1470         if (trace_kprobe_is_return(tk)) {
1471                 call->event.funcs = &kretprobe_funcs;
1472                 call->class->define_fields = kretprobe_event_define_fields;
1473         } else {
1474                 call->event.funcs = &kprobe_funcs;
1475                 call->class->define_fields = kprobe_event_define_fields;
1476         }
1477
1478         call->flags = TRACE_EVENT_FL_KPROBE;
1479         call->class->reg = kprobe_register;
1480 }
1481
1482 static int register_kprobe_event(struct trace_kprobe *tk)
1483 {
1484         init_trace_event_call(tk);
1485
1486         return trace_probe_register_event_call(&tk->tp);
1487 }
1488
1489 static int unregister_kprobe_event(struct trace_kprobe *tk)
1490 {
1491         return trace_probe_unregister_event_call(&tk->tp);
1492 }
1493
1494 #ifdef CONFIG_PERF_EVENTS
1495 /* create a trace_kprobe, but don't add it to global lists */
1496 struct trace_event_call *
1497 create_local_trace_kprobe(char *func, void *addr, unsigned long offs,
1498                           bool is_return)
1499 {
1500         struct trace_kprobe *tk;
1501         int ret;
1502         char *event;
1503
1504         /*
1505          * local trace_kprobes are not added to dyn_event, so they are never
1506          * searched in find_trace_kprobe(). Therefore, there is no concern of
1507          * duplicated name here.
1508          */
1509         event = func ? func : "DUMMY_EVENT";
1510
1511         tk = alloc_trace_kprobe(KPROBE_EVENT_SYSTEM, event, (void *)addr, func,
1512                                 offs, 0 /* maxactive */, 0 /* nargs */,
1513                                 is_return);
1514
1515         if (IS_ERR(tk)) {
1516                 pr_info("Failed to allocate trace_probe.(%d)\n",
1517                         (int)PTR_ERR(tk));
1518                 return ERR_CAST(tk);
1519         }
1520
1521         init_trace_event_call(tk);
1522
1523         if (traceprobe_set_print_fmt(&tk->tp, trace_kprobe_is_return(tk)) < 0) {
1524                 ret = -ENOMEM;
1525                 goto error;
1526         }
1527
1528         ret = __register_trace_kprobe(tk);
1529         if (ret < 0)
1530                 goto error;
1531
1532         return trace_probe_event_call(&tk->tp);
1533 error:
1534         free_trace_kprobe(tk);
1535         return ERR_PTR(ret);
1536 }
1537
1538 void destroy_local_trace_kprobe(struct trace_event_call *event_call)
1539 {
1540         struct trace_kprobe *tk;
1541
1542         tk = trace_kprobe_primary_from_call(event_call);
1543         if (unlikely(!tk))
1544                 return;
1545
1546         if (trace_probe_is_enabled(&tk->tp)) {
1547                 WARN_ON(1);
1548                 return;
1549         }
1550
1551         __unregister_trace_kprobe(tk);
1552
1553         free_trace_kprobe(tk);
1554 }
1555 #endif /* CONFIG_PERF_EVENTS */
1556
1557 static __init void enable_boot_kprobe_events(void)
1558 {
1559         struct trace_array *tr = top_trace_array();
1560         struct trace_event_file *file;
1561         struct trace_kprobe *tk;
1562         struct dyn_event *pos;
1563
1564         mutex_lock(&event_mutex);
1565         for_each_trace_kprobe(tk, pos) {
1566                 list_for_each_entry(file, &tr->events, list)
1567                         if (file->event_call == trace_probe_event_call(&tk->tp))
1568                                 trace_event_enable_disable(file, 1, 0);
1569         }
1570         mutex_unlock(&event_mutex);
1571 }
1572
1573 static __init void setup_boot_kprobe_events(void)
1574 {
1575         char *p, *cmd = kprobe_boot_events_buf;
1576         int ret;
1577
1578         strreplace(kprobe_boot_events_buf, ',', ' ');
1579
1580         while (cmd && *cmd != '\0') {
1581                 p = strchr(cmd, ';');
1582                 if (p)
1583                         *p++ = '\0';
1584
1585                 ret = trace_run_command(cmd, create_or_delete_trace_kprobe);
1586                 if (ret)
1587                         pr_warn("Failed to add event(%d): %s\n", ret, cmd);
1588                 else
1589                         kprobe_boot_events_enabled = true;
1590
1591                 cmd = p;
1592         }
1593
1594         enable_boot_kprobe_events();
1595 }
1596
1597 /* Make a tracefs interface for controlling probe points */
1598 static __init int init_kprobe_trace(void)
1599 {
1600         struct dentry *d_tracer;
1601         struct dentry *entry;
1602         int ret;
1603
1604         ret = dyn_event_register(&trace_kprobe_ops);
1605         if (ret)
1606                 return ret;
1607
1608         if (register_module_notifier(&trace_kprobe_module_nb))
1609                 return -EINVAL;
1610
1611         d_tracer = tracing_init_dentry();
1612         if (IS_ERR(d_tracer))
1613                 return 0;
1614
1615         entry = tracefs_create_file("kprobe_events", 0644, d_tracer,
1616                                     NULL, &kprobe_events_ops);
1617
1618         /* Event list interface */
1619         if (!entry)
1620                 pr_warn("Could not create tracefs 'kprobe_events' entry\n");
1621
1622         /* Profile interface */
1623         entry = tracefs_create_file("kprobe_profile", 0444, d_tracer,
1624                                     NULL, &kprobe_profile_ops);
1625
1626         if (!entry)
1627                 pr_warn("Could not create tracefs 'kprobe_profile' entry\n");
1628
1629         setup_boot_kprobe_events();
1630
1631         return 0;
1632 }
1633 fs_initcall(init_kprobe_trace);
1634
1635
1636 #ifdef CONFIG_FTRACE_STARTUP_TEST
1637 static __init struct trace_event_file *
1638 find_trace_probe_file(struct trace_kprobe *tk, struct trace_array *tr)
1639 {
1640         struct trace_event_file *file;
1641
1642         list_for_each_entry(file, &tr->events, list)
1643                 if (file->event_call == trace_probe_event_call(&tk->tp))
1644                         return file;
1645
1646         return NULL;
1647 }
1648
1649 /*
1650  * Nobody but us can call enable_trace_kprobe/disable_trace_kprobe at this
1651  * stage, we can do this lockless.
1652  */
1653 static __init int kprobe_trace_self_tests_init(void)
1654 {
1655         int ret, warn = 0;
1656         int (*target)(int, int, int, int, int, int);
1657         struct trace_kprobe *tk;
1658         struct trace_event_file *file;
1659
1660         if (tracing_is_disabled())
1661                 return -ENODEV;
1662
1663         if (kprobe_boot_events_enabled) {
1664                 pr_info("Skipping kprobe tests due to kprobe_event on cmdline\n");
1665                 return 0;
1666         }
1667
1668         target = kprobe_trace_selftest_target;
1669
1670         pr_info("Testing kprobe tracing: ");
1671
1672         ret = trace_run_command("p:testprobe kprobe_trace_selftest_target $stack $stack0 +0($stack)",
1673                                 create_or_delete_trace_kprobe);
1674         if (WARN_ON_ONCE(ret)) {
1675                 pr_warn("error on probing function entry.\n");
1676                 warn++;
1677         } else {
1678                 /* Enable trace point */
1679                 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1680                 if (WARN_ON_ONCE(tk == NULL)) {
1681                         pr_warn("error on getting new probe.\n");
1682                         warn++;
1683                 } else {
1684                         file = find_trace_probe_file(tk, top_trace_array());
1685                         if (WARN_ON_ONCE(file == NULL)) {
1686                                 pr_warn("error on getting probe file.\n");
1687                                 warn++;
1688                         } else
1689                                 enable_trace_kprobe(
1690                                         trace_probe_event_call(&tk->tp), file);
1691                 }
1692         }
1693
1694         ret = trace_run_command("r:testprobe2 kprobe_trace_selftest_target $retval",
1695                                 create_or_delete_trace_kprobe);
1696         if (WARN_ON_ONCE(ret)) {
1697                 pr_warn("error on probing function return.\n");
1698                 warn++;
1699         } else {
1700                 /* Enable trace point */
1701                 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1702                 if (WARN_ON_ONCE(tk == NULL)) {
1703                         pr_warn("error on getting 2nd new probe.\n");
1704                         warn++;
1705                 } else {
1706                         file = find_trace_probe_file(tk, top_trace_array());
1707                         if (WARN_ON_ONCE(file == NULL)) {
1708                                 pr_warn("error on getting probe file.\n");
1709                                 warn++;
1710                         } else
1711                                 enable_trace_kprobe(
1712                                         trace_probe_event_call(&tk->tp), file);
1713                 }
1714         }
1715
1716         if (warn)
1717                 goto end;
1718
1719         ret = target(1, 2, 3, 4, 5, 6);
1720
1721         /*
1722          * Not expecting an error here, the check is only to prevent the
1723          * optimizer from removing the call to target() as otherwise there
1724          * are no side-effects and the call is never performed.
1725          */
1726         if (ret != 21)
1727                 warn++;
1728
1729         /* Disable trace points before removing it */
1730         tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1731         if (WARN_ON_ONCE(tk == NULL)) {
1732                 pr_warn("error on getting test probe.\n");
1733                 warn++;
1734         } else {
1735                 if (trace_kprobe_nhit(tk) != 1) {
1736                         pr_warn("incorrect number of testprobe hits\n");
1737                         warn++;
1738                 }
1739
1740                 file = find_trace_probe_file(tk, top_trace_array());
1741                 if (WARN_ON_ONCE(file == NULL)) {
1742                         pr_warn("error on getting probe file.\n");
1743                         warn++;
1744                 } else
1745                         disable_trace_kprobe(
1746                                 trace_probe_event_call(&tk->tp), file);
1747         }
1748
1749         tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1750         if (WARN_ON_ONCE(tk == NULL)) {
1751                 pr_warn("error on getting 2nd test probe.\n");
1752                 warn++;
1753         } else {
1754                 if (trace_kprobe_nhit(tk) != 1) {
1755                         pr_warn("incorrect number of testprobe2 hits\n");
1756                         warn++;
1757                 }
1758
1759                 file = find_trace_probe_file(tk, top_trace_array());
1760                 if (WARN_ON_ONCE(file == NULL)) {
1761                         pr_warn("error on getting probe file.\n");
1762                         warn++;
1763                 } else
1764                         disable_trace_kprobe(
1765                                 trace_probe_event_call(&tk->tp), file);
1766         }
1767
1768         ret = trace_run_command("-:testprobe", create_or_delete_trace_kprobe);
1769         if (WARN_ON_ONCE(ret)) {
1770                 pr_warn("error on deleting a probe.\n");
1771                 warn++;
1772         }
1773
1774         ret = trace_run_command("-:testprobe2", create_or_delete_trace_kprobe);
1775         if (WARN_ON_ONCE(ret)) {
1776                 pr_warn("error on deleting a probe.\n");
1777                 warn++;
1778         }
1779
1780 end:
1781         ret = dyn_events_release_all(&trace_kprobe_ops);
1782         if (WARN_ON_ONCE(ret)) {
1783                 pr_warn("error on cleaning up probes.\n");
1784                 warn++;
1785         }
1786         /*
1787          * Wait for the optimizer work to finish. Otherwise it might fiddle
1788          * with probes in already freed __init text.
1789          */
1790         wait_for_kprobe_optimizer();
1791         if (warn)
1792                 pr_cont("NG: Some tests are failed. Please check them.\n");
1793         else
1794                 pr_cont("OK\n");
1795         return 0;
1796 }
1797
1798 late_initcall(kprobe_trace_self_tests_init);
1799
1800 #endif