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_command_head(struct trace_kprobe *tk,
 141                                             int argc, const char **argv)
 142 {
 143         char buf[MAX_ARGSTR_LEN + 1];
 144
 145         if (!argc)
 146                 return true;
 147
 148         if (!tk->symbol)
 149                 snprintf(buf, sizeof(buf), "0x%p", tk->rp.kp.addr);
 150         else if (tk->rp.kp.offset)
 151                 snprintf(buf, sizeof(buf), "%s+%u",
 152                          trace_kprobe_symbol(tk), tk->rp.kp.offset);
 153         else
 154                 snprintf(buf, sizeof(buf), "%s", trace_kprobe_symbol(tk));
 155         if (strcmp(buf, argv[0]))
 156                 return false;
 157         argc--; argv++;
 158
 159         return trace_probe_match_command_args(&tk->tp, argc, argv);
 160 }
 161
 162 static bool trace_kprobe_match(const char *system, const char *event,
 163                         int argc, const char **argv, struct dyn_event *ev)
 164 {
 165         struct trace_kprobe *tk = to_trace_kprobe(ev);
 166
 167         return strcmp(trace_probe_name(&tk->tp), event) == 0 &&
 168             (!system || strcmp(trace_probe_group_name(&tk->tp), system) == 0) &&
 169             trace_kprobe_match_command_head(tk, argc, argv);
 170 }
 171
 172 static nokprobe_inline unsigned long trace_kprobe_nhit(struct trace_kprobe *tk)
 173 {
 174         unsigned long nhit = 0;
 175         int cpu;
 176
 177         for_each_possible_cpu(cpu)
 178                 nhit += *per_cpu_ptr(tk->nhit, cpu);
 179
 180         return nhit;
 181 }
 182
 183 static nokprobe_inline bool trace_kprobe_is_registered(struct trace_kprobe *tk)
 184 {
 185         return !(list_empty(&tk->rp.kp.list) &&
 186                  hlist_unhashed(&tk->rp.kp.hlist));
 187 }
 188
 189 /* Return 0 if it fails to find the symbol address */
 190 static nokprobe_inline
 191 unsigned long trace_kprobe_address(struct trace_kprobe *tk)
 192 {
 193         unsigned long addr;
 194
 195         if (tk->symbol) {
 196                 addr = (unsigned long)
 197                         kallsyms_lookup_name(trace_kprobe_symbol(tk));
 198                 if (addr)
 199                         addr += tk->rp.kp.offset;
 200         } else {
 201                 addr = (unsigned long)tk->rp.kp.addr;
 202         }
 203         return addr;
 204 }
 205
 206 static nokprobe_inline struct trace_kprobe *
 207 trace_kprobe_primary_from_call(struct trace_event_call *call)
 208 {
 209         struct trace_probe *tp;
 210
 211         tp = trace_probe_primary_from_call(call);
 212         if (WARN_ON_ONCE(!tp))
 213                 return NULL;
 214
 215         return container_of(tp, struct trace_kprobe, tp);
 216 }
 217
 218 bool trace_kprobe_on_func_entry(struct trace_event_call *call)
 219 {
 220         struct trace_kprobe *tk = trace_kprobe_primary_from_call(call);
 221
 222         return tk ? kprobe_on_func_entry(tk->rp.kp.addr,
 223                         tk->rp.kp.addr ? NULL : tk->rp.kp.symbol_name,
 224                         tk->rp.kp.addr ? 0 : tk->rp.kp.offset) : false;
 225 }
 226
 227 bool trace_kprobe_error_injectable(struct trace_event_call *call)
 228 {
 229         struct trace_kprobe *tk = trace_kprobe_primary_from_call(call);
 230
 231         return tk ? within_error_injection_list(trace_kprobe_address(tk)) :
 232                false;
 233 }
 234
 235 static int register_kprobe_event(struct trace_kprobe *tk);
 236 static int unregister_kprobe_event(struct trace_kprobe *tk);
 237
 238 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
 239 static int kretprobe_dispatcher(struct kretprobe_instance *ri,
 240                                 struct pt_regs *regs);
 241
 242 static void free_trace_kprobe(struct trace_kprobe *tk)
 243 {
 244         if (tk) {
 245                 trace_probe_cleanup(&tk->tp);
 246                 kfree(tk->symbol);
 247                 free_percpu(tk->nhit);
 248                 kfree(tk);
 249         }
 250 }
 251
 252 /*
 253  * Allocate new trace_probe and initialize it (including kprobes).
 254  */
 255 static struct trace_kprobe *alloc_trace_kprobe(const char *group,
 256                                              const char *event,
 257                                              void *addr,
 258                                              const char *symbol,
 259                                              unsigned long offs,
 260                                              int maxactive,
 261                                              int nargs, bool is_return)
 262 {
 263         struct trace_kprobe *tk;
 264         int ret = -ENOMEM;
 265
 266         tk = kzalloc(SIZEOF_TRACE_KPROBE(nargs), GFP_KERNEL);
 267         if (!tk)
 268                 return ERR_PTR(ret);
 269
 270         tk->nhit = alloc_percpu(unsigned long);
 271         if (!tk->nhit)
 272                 goto error;
 273
 274         if (symbol) {
 275                 tk->symbol = kstrdup(symbol, GFP_KERNEL);
 276                 if (!tk->symbol)
 277                         goto error;
 278                 tk->rp.kp.symbol_name = tk->symbol;
 279                 tk->rp.kp.offset = offs;
 280         } else
 281                 tk->rp.kp.addr = addr;
 282
 283         if (is_return)
 284                 tk->rp.handler = kretprobe_dispatcher;
 285         else
 286                 tk->rp.kp.pre_handler = kprobe_dispatcher;
 287
 288         tk->rp.maxactive = maxactive;
 289         INIT_HLIST_NODE(&tk->rp.kp.hlist);
 290         INIT_LIST_HEAD(&tk->rp.kp.list);
 291
 292         ret = trace_probe_init(&tk->tp, event, group);
 293         if (ret < 0)
 294                 goto error;
 295
 296         dyn_event_init(&tk->devent, &trace_kprobe_ops);
 297         return tk;
 298 error:
 299         free_trace_kprobe(tk);
 300         return ERR_PTR(ret);
 301 }
 302
 303 static struct trace_kprobe *find_trace_kprobe(const char *event,
 304                                               const char *group)
 305 {
 306         struct dyn_event *pos;
 307         struct trace_kprobe *tk;
 308
 309         for_each_trace_kprobe(tk, pos)
 310                 if (strcmp(trace_probe_name(&tk->tp), event) == 0 &&
 311                     strcmp(trace_probe_group_name(&tk->tp), group) == 0)
 312                         return tk;
 313         return NULL;
 314 }
 315
 316 static inline int __enable_trace_kprobe(struct trace_kprobe *tk)
 317 {
 318         int ret = 0;
 319
 320         if (trace_kprobe_is_registered(tk) && !trace_kprobe_has_gone(tk)) {
 321                 if (trace_kprobe_is_return(tk))
 322                         ret = enable_kretprobe(&tk->rp);
 323                 else
 324                         ret = enable_kprobe(&tk->rp.kp);
 325         }
 326
 327         return ret;
 328 }
 329
 330 static void __disable_trace_kprobe(struct trace_probe *tp)
 331 {
 332         struct trace_probe *pos;
 333         struct trace_kprobe *tk;
 334
 335         list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
 336                 tk = container_of(pos, struct trace_kprobe, tp);
 337                 if (!trace_kprobe_is_registered(tk))
 338                         continue;
 339                 if (trace_kprobe_is_return(tk))
 340                         disable_kretprobe(&tk->rp);
 341                 else
 342                         disable_kprobe(&tk->rp.kp);
 343         }
 344 }
 345
 346 /*
 347  * Enable trace_probe
 348  * if the file is NULL, enable "perf" handler, or enable "trace" handler.
 349  */
 350 static int enable_trace_kprobe(struct trace_event_call *call,
 351                                 struct trace_event_file *file)
 352 {
 353         struct trace_probe *pos, *tp;
 354         struct trace_kprobe *tk;
 355         bool enabled;
 356         int ret = 0;
 357
 358         tp = trace_probe_primary_from_call(call);
 359         if (WARN_ON_ONCE(!tp))
 360                 return -ENODEV;
 361         enabled = trace_probe_is_enabled(tp);
 362
 363         /* This also changes "enabled" state */
 364         if (file) {
 365                 ret = trace_probe_add_file(tp, file);
 366                 if (ret)
 367                         return ret;
 368         } else
 369                 trace_probe_set_flag(tp, TP_FLAG_PROFILE);
 370
 371         if (enabled)
 372                 return 0;
 373
 374         enabled = false;
 375         list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
 376                 tk = container_of(pos, struct trace_kprobe, tp);
 377                 if (trace_kprobe_has_gone(tk))
 378                         continue;
 379                 ret = __enable_trace_kprobe(tk);
 380                 if (ret) {
 381                         if (enabled) {
 382                                 __disable_trace_kprobe(tp);
 383                                 enabled = false;
 384                         }
 385                         break;
 386                 }
 387                 enabled = true;
 388         }
 389
 390         if (!enabled) {
 391                 /* No probe is enabled. Roll back */
 392                 if (file)
 393                         trace_probe_remove_file(tp, file);
 394                 else
 395                         trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
 396                 if (!ret)
 397                         /* Since all probes are gone, this is not available */
 398                         ret = -EADDRNOTAVAIL;
 399         }
 400
 401         return ret;
 402 }
 403
 404 /*
 405  * Disable trace_probe
 406  * if the file is NULL, disable "perf" handler, or disable "trace" handler.
 407  */
 408 static int disable_trace_kprobe(struct trace_event_call *call,
 409                                 struct trace_event_file *file)
 410 {
 411         struct trace_probe *tp;
 412
 413         tp = trace_probe_primary_from_call(call);
 414         if (WARN_ON_ONCE(!tp))
 415                 return -ENODEV;
 416
 417         if (file) {
 418                 if (!trace_probe_get_file_link(tp, file))
 419                         return -ENOENT;
 420                 if (!trace_probe_has_single_file(tp))
 421                         goto out;
 422                 trace_probe_clear_flag(tp, TP_FLAG_TRACE);
 423         } else
 424                 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
 425
 426         if (!trace_probe_is_enabled(tp))
 427                 __disable_trace_kprobe(tp);
 428
 429  out:
 430         if (file)
 431                 /*
 432                  * Synchronization is done in below function. For perf event,
 433                  * file == NULL and perf_trace_event_unreg() calls
 434                  * tracepoint_synchronize_unregister() to ensure synchronize
 435                  * event. We don't need to care about it.
 436                  */
 437                 trace_probe_remove_file(tp, file);
 438
 439         return 0;
 440 }
 441
 442 #if defined(CONFIG_KPROBES_ON_FTRACE) && \
 443         !defined(CONFIG_KPROBE_EVENTS_ON_NOTRACE)
 444 static bool within_notrace_func(struct trace_kprobe *tk)
 445 {
 446         unsigned long offset, size, addr;
 447
 448         addr = trace_kprobe_address(tk);
 449         if (!addr || !kallsyms_lookup_size_offset(addr, &size, &offset))
 450                 return false;
 451
 452         /* Get the entry address of the target function */
 453         addr -= offset;
 454
 455         /*
 456          * Since ftrace_location_range() does inclusive range check, we need
 457          * to subtract 1 byte from the end address.
 458          */
 459         return !ftrace_location_range(addr, addr + size - 1);
 460 }
 461 #else
 462 #define within_notrace_func(tk) (false)
 463 #endif
 464
 465 /* Internal register function - just handle k*probes and flags */
 466 static int __register_trace_kprobe(struct trace_kprobe *tk)
 467 {
 468         int i, ret;
 469
 470         if (trace_kprobe_is_registered(tk))
 471                 return -EINVAL;
 472
 473         if (within_notrace_func(tk)) {
 474                 pr_warn("Could not probe notrace function %s\n",
 475                         trace_kprobe_symbol(tk));
 476                 return -EINVAL;
 477         }
 478
 479         for (i = 0; i < tk->tp.nr_args; i++) {
 480                 ret = traceprobe_update_arg(&tk->tp.args[i]);
 481                 if (ret)
 482                         return ret;
 483         }
 484
 485         /* Set/clear disabled flag according to tp->flag */
 486         if (trace_probe_is_enabled(&tk->tp))
 487                 tk->rp.kp.flags &= ~KPROBE_FLAG_DISABLED;
 488         else
 489                 tk->rp.kp.flags |= KPROBE_FLAG_DISABLED;
 490
 491         if (trace_kprobe_is_return(tk))
 492                 ret = register_kretprobe(&tk->rp);
 493         else
 494                 ret = register_kprobe(&tk->rp.kp);
 495
 496         return ret;
 497 }
 498
 499 /* Internal unregister function - just handle k*probes and flags */
 500 static void __unregister_trace_kprobe(struct trace_kprobe *tk)
 501 {
 502         if (trace_kprobe_is_registered(tk)) {
 503                 if (trace_kprobe_is_return(tk))
 504                         unregister_kretprobe(&tk->rp);
 505                 else
 506                         unregister_kprobe(&tk->rp.kp);
 507                 /* Cleanup kprobe for reuse and mark it unregistered */
 508                 INIT_HLIST_NODE(&tk->rp.kp.hlist);
 509                 INIT_LIST_HEAD(&tk->rp.kp.list);
 510                 if (tk->rp.kp.symbol_name)
 511                         tk->rp.kp.addr = NULL;
 512         }
 513 }
 514
 515 /* Unregister a trace_probe and probe_event */
 516 static int unregister_trace_kprobe(struct trace_kprobe *tk)
 517 {
 518         /* If other probes are on the event, just unregister kprobe */
 519         if (trace_probe_has_sibling(&tk->tp))
 520                 goto unreg;
 521
 522         /* Enabled event can not be unregistered */
 523         if (trace_probe_is_enabled(&tk->tp))
 524                 return -EBUSY;
 525
 526         /* Will fail if probe is being used by ftrace or perf */
 527         if (unregister_kprobe_event(tk))
 528                 return -EBUSY;
 529
 530 unreg:
 531         __unregister_trace_kprobe(tk);
 532         dyn_event_remove(&tk->devent);
 533         trace_probe_unlink(&tk->tp);
 534
 535         return 0;
 536 }
 537
 538 static int append_trace_kprobe(struct trace_kprobe *tk, struct trace_kprobe *to)
 539 {
 540         int ret;
 541
 542         /* Append to existing event */
 543         ret = trace_probe_append(&tk->tp, &to->tp);
 544         if (ret)
 545                 return ret;
 546
 547         /* Register k*probe */
 548         ret = __register_trace_kprobe(tk);
 549         if (ret == -ENOENT && !trace_kprobe_module_exist(tk)) {
 550                 pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
 551                 ret = 0;
 552         }
 553
 554         if (ret)
 555                 trace_probe_unlink(&tk->tp);
 556         else
 557                 dyn_event_add(&tk->devent);
 558
 559         return ret;
 560 }
 561
 562 /* Register a trace_probe and probe_event */
 563 static int register_trace_kprobe(struct trace_kprobe *tk)
 564 {
 565         struct trace_kprobe *old_tk;
 566         int ret;
 567
 568         mutex_lock(&event_mutex);
 569
 570         old_tk = find_trace_kprobe(trace_probe_name(&tk->tp),
 571                                    trace_probe_group_name(&tk->tp));
 572         if (old_tk) {
 573                 if (trace_kprobe_is_return(tk) != trace_kprobe_is_return(old_tk)) {
 574                         trace_probe_log_set_index(0);
 575                         trace_probe_log_err(0, DIFF_PROBE_TYPE);
 576                         ret = -EEXIST;
 577                 } else {
 578                         ret = trace_probe_compare_arg_type(&tk->tp, &old_tk->tp);
 579                         if (ret) {
 580                                 /* Note that argument starts index = 2 */
 581                                 trace_probe_log_set_index(ret + 1);
 582                                 trace_probe_log_err(0, DIFF_ARG_TYPE);
 583                                 ret = -EEXIST;
 584                         } else
 585                                 ret = append_trace_kprobe(tk, old_tk);
 586                 }
 587                 goto end;
 588         }
 589
 590         /* Register new event */
 591         ret = register_kprobe_event(tk);
 592         if (ret) {
 593                 pr_warn("Failed to register probe event(%d)\n", ret);
 594                 goto end;
 595         }
 596
 597         /* Register k*probe */
 598         ret = __register_trace_kprobe(tk);
 599         if (ret == -ENOENT && !trace_kprobe_module_exist(tk)) {
 600                 pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
 601                 ret = 0;
 602         }
 603
 604         if (ret < 0)
 605                 unregister_kprobe_event(tk);
 606         else
 607                 dyn_event_add(&tk->devent);
 608
 609 end:
 610         mutex_unlock(&event_mutex);
 611         return ret;
 612 }
 613
 614 /* Module notifier call back, checking event on the module */
 615 static int trace_kprobe_module_callback(struct notifier_block *nb,
 616                                        unsigned long val, void *data)
 617 {
 618         struct module *mod = data;
 619         struct dyn_event *pos;
 620         struct trace_kprobe *tk;
 621         int ret;
 622
 623         if (val != MODULE_STATE_COMING)
 624                 return NOTIFY_DONE;
 625
 626         /* Update probes on coming module */
 627         mutex_lock(&event_mutex);
 628         for_each_trace_kprobe(tk, pos) {
 629                 if (trace_kprobe_within_module(tk, mod)) {
 630                         /* Don't need to check busy - this should have gone. */
 631                         __unregister_trace_kprobe(tk);
 632                         ret = __register_trace_kprobe(tk);
 633                         if (ret)
 634                                 pr_warn("Failed to re-register probe %s on %s: %d\n",
 635                                         trace_probe_name(&tk->tp),
 636                                         mod->name, ret);
 637                 }
 638         }
 639         mutex_unlock(&event_mutex);
 640
 641         return NOTIFY_DONE;
 642 }
 643
 644 static struct notifier_block trace_kprobe_module_nb = {
 645         .notifier_call = trace_kprobe_module_callback,
 646         .priority = 1   /* Invoked after kprobe module callback */
 647 };
 648
 649 /* Convert certain expected symbols into '_' when generating event names */
 650 static inline void sanitize_event_name(char *name)
 651 {
 652         while (*name++ != '\0')
 653                 if (*name == ':' || *name == '.')
 654                         *name = '_';
 655 }
 656
 657 static int trace_kprobe_create(int argc, const char *argv[])
 658 {
 659         /*
 660          * Argument syntax:
 661          *  - Add kprobe:
 662          *      p[:[GRP/]EVENT] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
 663          *  - Add kretprobe:
 664          *      r[MAXACTIVE][:[GRP/]EVENT] [MOD:]KSYM[+0] [FETCHARGS]
 665          * Fetch args:
 666          *  $retval     : fetch return value
 667          *  $stack      : fetch stack address
 668          *  $stackN     : fetch Nth of stack (N:0-)
 669          *  $comm       : fetch current task comm
 670          *  @ADDR       : fetch memory at ADDR (ADDR should be in kernel)
 671          *  @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
 672          *  %REG        : fetch register REG
 673          * Dereferencing memory fetch:
 674          *  +|-offs(ARG) : fetch memory at ARG +|- offs address.
 675          * Alias name of args:
 676          *  NAME=FETCHARG : set NAME as alias of FETCHARG.
 677          * Type of args:
 678          *  FETCHARG:TYPE : use TYPE instead of unsigned long.
 679          */
 680         struct trace_kprobe *tk = NULL;
 681         int i, len, ret = 0;
 682         bool is_return = false;
 683         char *symbol = NULL, *tmp = NULL;
 684         const char *event = NULL, *group = KPROBE_EVENT_SYSTEM;
 685         int maxactive = 0;
 686         long offset = 0;
 687         void *addr = NULL;
 688         char buf[MAX_EVENT_NAME_LEN];
 689         unsigned int flags = TPARG_FL_KERNEL;
 690
 691         switch (argv[0][0]) {
 692         case 'r':
 693                 is_return = true;
 694                 flags |= TPARG_FL_RETURN;
 695                 break;
 696         case 'p':
 697                 break;
 698         default:
 699                 return -ECANCELED;
 700         }
 701         if (argc < 2)
 702                 return -ECANCELED;
 703
 704         trace_probe_log_init("trace_kprobe", argc, argv);
 705
 706         event = strchr(&argv[0][1], ':');
 707         if (event)
 708                 event++;
 709
 710         if (isdigit(argv[0][1])) {
 711                 if (!is_return) {
 712                         trace_probe_log_err(1, MAXACT_NO_KPROBE);
 713                         goto parse_error;
 714                 }
 715                 if (event)
 716                         len = event - &argv[0][1] - 1;
 717                 else
 718                         len = strlen(&argv[0][1]);
 719                 if (len > MAX_EVENT_NAME_LEN - 1) {
 720                         trace_probe_log_err(1, BAD_MAXACT);
 721                         goto parse_error;
 722                 }
 723                 memcpy(buf, &argv[0][1], len);
 724                 buf[len] = '\0';
 725                 ret = kstrtouint(buf, 0, &maxactive);
 726                 if (ret || !maxactive) {
 727                         trace_probe_log_err(1, BAD_MAXACT);
 728                         goto parse_error;
 729                 }
 730                 /* kretprobes instances are iterated over via a list. The
 731                  * maximum should stay reasonable.
 732                  */
 733                 if (maxactive > KRETPROBE_MAXACTIVE_MAX) {
 734                         trace_probe_log_err(1, MAXACT_TOO_BIG);
 735                         goto parse_error;
 736                 }
 737         }
 738
 739         /* try to parse an address. if that fails, try to read the
 740          * input as a symbol. */
 741         if (kstrtoul(argv[1], 0, (unsigned long *)&addr)) {
 742                 trace_probe_log_set_index(1);
 743                 /* Check whether uprobe event specified */
 744                 if (strchr(argv[1], '/') && strchr(argv[1], ':')) {
 745                         ret = -ECANCELED;
 746                         goto error;
 747                 }
 748                 /* a symbol specified */
 749                 symbol = kstrdup(argv[1], GFP_KERNEL);
 750                 if (!symbol)
 751                         return -ENOMEM;
 752                 /* TODO: support .init module functions */
 753                 ret = traceprobe_split_symbol_offset(symbol, &offset);
 754                 if (ret || offset < 0 || offset > UINT_MAX) {
 755                         trace_probe_log_err(0, BAD_PROBE_ADDR);
 756                         goto parse_error;
 757                 }
 758                 if (kprobe_on_func_entry(NULL, symbol, offset))
 759                         flags |= TPARG_FL_FENTRY;
 760                 if (offset && is_return && !(flags & TPARG_FL_FENTRY)) {
 761                         trace_probe_log_err(0, BAD_RETPROBE);
 762                         goto parse_error;
 763                 }
 764         }
 765
 766         trace_probe_log_set_index(0);
 767         if (event) {
 768                 ret = traceprobe_parse_event_name(&event, &group, buf,
 769                                                   event - argv[0]);
 770                 if (ret)
 771                         goto parse_error;
 772         } else {
 773                 /* Make a new event name */
 774                 if (symbol)
 775                         snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
 776                                  is_return ? 'r' : 'p', symbol, offset);
 777                 else
 778                         snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
 779                                  is_return ? 'r' : 'p', addr);
 780                 sanitize_event_name(buf);
 781                 event = buf;
 782         }
 783
 784         /* setup a probe */
 785         tk = alloc_trace_kprobe(group, event, addr, symbol, offset, maxactive,
 786                                argc - 2, is_return);
 787         if (IS_ERR(tk)) {
 788                 ret = PTR_ERR(tk);
 789                 /* This must return -ENOMEM, else there is a bug */
 790                 WARN_ON_ONCE(ret != -ENOMEM);
 791                 goto out;       /* We know tk is not allocated */
 792         }
 793         argc -= 2; argv += 2;
 794
 795         /* parse arguments */
 796         for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
 797                 tmp = kstrdup(argv[i], GFP_KERNEL);
 798                 if (!tmp) {
 799                         ret = -ENOMEM;
 800                         goto error;
 801                 }
 802
 803                 trace_probe_log_set_index(i + 2);
 804                 ret = traceprobe_parse_probe_arg(&tk->tp, i, tmp, flags);
 805                 kfree(tmp);
 806                 if (ret)
 807                         goto error;     /* This can be -ENOMEM */
 808         }
 809
 810         ret = traceprobe_set_print_fmt(&tk->tp, is_return);
 811         if (ret < 0)
 812                 goto error;
 813
 814         ret = register_trace_kprobe(tk);
 815         if (ret) {
 816                 trace_probe_log_set_index(1);
 817                 if (ret == -EILSEQ)
 818                         trace_probe_log_err(0, BAD_INSN_BNDRY);
 819                 else if (ret == -ENOENT)
 820                         trace_probe_log_err(0, BAD_PROBE_ADDR);
 821                 else if (ret != -ENOMEM && ret != -EEXIST)
 822                         trace_probe_log_err(0, FAIL_REG_PROBE);
 823                 goto error;
 824         }
 825
 826 out:
 827         trace_probe_log_clear();
 828         kfree(symbol);
 829         return ret;
 830
 831 parse_error:
 832         ret = -EINVAL;
 833 error:
 834         free_trace_kprobe(tk);
 835         goto out;
 836 }
 837
 838 static int create_or_delete_trace_kprobe(int argc, char **argv)
 839 {
 840         int ret;
 841
 842         if (argv[0][0] == '-')
 843                 return dyn_event_release(argc, argv, &trace_kprobe_ops);
 844
 845         ret = trace_kprobe_create(argc, (const char **)argv);
 846         return ret == -ECANCELED ? -EINVAL : ret;
 847 }
 848
 849 static int trace_kprobe_release(struct dyn_event *ev)
 850 {
 851         struct trace_kprobe *tk = to_trace_kprobe(ev);
 852         int ret = unregister_trace_kprobe(tk);
 853
 854         if (!ret)
 855                 free_trace_kprobe(tk);
 856         return ret;
 857 }
 858
 859 static int trace_kprobe_show(struct seq_file *m, struct dyn_event *ev)
 860 {
 861         struct trace_kprobe *tk = to_trace_kprobe(ev);
 862         int i;
 863
 864         seq_putc(m, trace_kprobe_is_return(tk) ? 'r' : 'p');
 865         seq_printf(m, ":%s/%s", trace_probe_group_name(&tk->tp),
 866                                 trace_probe_name(&tk->tp));
 867
 868         if (!tk->symbol)
 869                 seq_printf(m, " 0x%p", tk->rp.kp.addr);
 870         else if (tk->rp.kp.offset)
 871                 seq_printf(m, " %s+%u", trace_kprobe_symbol(tk),
 872                            tk->rp.kp.offset);
 873         else
 874                 seq_printf(m, " %s", trace_kprobe_symbol(tk));
 875
 876         for (i = 0; i < tk->tp.nr_args; i++)
 877                 seq_printf(m, " %s=%s", tk->tp.args[i].name, tk->tp.args[i].comm);
 878         seq_putc(m, '\n');
 879
 880         return 0;
 881 }
 882
 883 static int probes_seq_show(struct seq_file *m, void *v)
 884 {
 885         struct dyn_event *ev = v;
 886
 887         if (!is_trace_kprobe(ev))
 888                 return 0;
 889
 890         return trace_kprobe_show(m, ev);
 891 }
 892
 893 static const struct seq_operations probes_seq_op = {
 894         .start  = dyn_event_seq_start,
 895         .next   = dyn_event_seq_next,
 896         .stop   = dyn_event_seq_stop,
 897         .show   = probes_seq_show
 898 };
 899
 900 static int probes_open(struct inode *inode, struct file *file)
 901 {
 902         int ret;
 903
 904         if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
 905                 ret = dyn_events_release_all(&trace_kprobe_ops);
 906                 if (ret < 0)
 907                         return ret;
 908         }
 909
 910         return seq_open(file, &probes_seq_op);
 911 }
 912
 913 static ssize_t probes_write(struct file *file, const char __user *buffer,
 914                             size_t count, loff_t *ppos)
 915 {
 916         return trace_parse_run_command(file, buffer, count, ppos,
 917                                        create_or_delete_trace_kprobe);
 918 }
 919
 920 static const struct file_operations kprobe_events_ops = {
 921         .owner          = THIS_MODULE,
 922         .open           = probes_open,
 923         .read           = seq_read,
 924         .llseek         = seq_lseek,
 925         .release        = seq_release,
 926         .write          = probes_write,
 927 };
 928
 929 /* Probes profiling interfaces */
 930 static int probes_profile_seq_show(struct seq_file *m, void *v)
 931 {
 932         struct dyn_event *ev = v;
 933         struct trace_kprobe *tk;
 934
 935         if (!is_trace_kprobe(ev))
 936                 return 0;
 937
 938         tk = to_trace_kprobe(ev);
 939         seq_printf(m, "  %-44s %15lu %15lu\n",
 940                    trace_probe_name(&tk->tp),
 941                    trace_kprobe_nhit(tk),
 942                    tk->rp.kp.nmissed);
 943
 944         return 0;
 945 }
 946
 947 static const struct seq_operations profile_seq_op = {
 948         .start  = dyn_event_seq_start,
 949         .next   = dyn_event_seq_next,
 950         .stop   = dyn_event_seq_stop,
 951         .show   = probes_profile_seq_show
 952 };
 953
 954 static int profile_open(struct inode *inode, struct file *file)
 955 {
 956         return seq_open(file, &profile_seq_op);
 957 }
 958
 959 static const struct file_operations kprobe_profile_ops = {
 960         .owner          = THIS_MODULE,
 961         .open           = profile_open,
 962         .read           = seq_read,
 963         .llseek         = seq_lseek,
 964         .release        = seq_release,
 965 };
 966
 967 /* Kprobe specific fetch functions */
 968
 969 /* Return the length of string -- including null terminal byte */
 970 static nokprobe_inline int
 971 fetch_store_strlen(unsigned long addr)
 972 {
 973         int ret, len = 0;
 974         u8 c;
 975
 976         do {
 977                 ret = probe_kernel_read(&c, (u8 *)addr + len, 1);
 978                 len++;
 979         } while (c && ret == 0 && len < MAX_STRING_SIZE);
 980
 981         return (ret < 0) ? ret : len;
 982 }
 983
 984 /* Return the length of string -- including null terminal byte */
 985 static nokprobe_inline int
 986 fetch_store_strlen_user(unsigned long addr)
 987 {
 988         const void __user *uaddr =  (__force const void __user *)addr;
 989
 990         return strnlen_unsafe_user(uaddr, MAX_STRING_SIZE);
 991 }
 992
 993 /*
 994  * Fetch a null-terminated string. Caller MUST set *(u32 *)buf with max
 995  * length and relative data location.
 996  */
 997 static nokprobe_inline int
 998 fetch_store_string(unsigned long addr, void *dest, void *base)
 999 {
1000         int maxlen = get_loc_len(*(u32 *)dest);
1001         void *__dest;
1002         long ret;
1003
1004         if (unlikely(!maxlen))
1005                 return -ENOMEM;
1006
1007         __dest = get_loc_data(dest, base);
1008
1009         /*
1010          * Try to get string again, since the string can be changed while
1011          * probing.
1012          */
1013         ret = strncpy_from_unsafe(__dest, (void *)addr, maxlen);
1014         if (ret >= 0)
1015                 *(u32 *)dest = make_data_loc(ret, __dest - base);
1016
1017         return ret;
1018 }
1019
1020 /*
1021  * Fetch a null-terminated string from user. Caller MUST set *(u32 *)buf
1022  * with max length and relative data location.
1023  */
1024 static nokprobe_inline int
1025 fetch_store_string_user(unsigned long addr, void *dest, void *base)
1026 {
1027         const void __user *uaddr =  (__force const void __user *)addr;
1028         int maxlen = get_loc_len(*(u32 *)dest);
1029         void *__dest;
1030         long ret;
1031
1032         if (unlikely(!maxlen))
1033                 return -ENOMEM;
1034
1035         __dest = get_loc_data(dest, base);
1036
1037         ret = strncpy_from_unsafe_user(__dest, uaddr, maxlen);
1038         if (ret >= 0)
1039                 *(u32 *)dest = make_data_loc(ret, __dest - base);
1040
1041         return ret;
1042 }
1043
1044 static nokprobe_inline int
1045 probe_mem_read(void *dest, void *src, size_t size)
1046 {
1047         return probe_kernel_read(dest, src, size);
1048 }
1049
1050 static nokprobe_inline int
1051 probe_mem_read_user(void *dest, void *src, size_t size)
1052 {
1053         const void __user *uaddr =  (__force const void __user *)src;
1054
1055         return probe_user_read(dest, uaddr, size);
1056 }
1057
1058 /* Note that we don't verify it, since the code does not come from user space */
1059 static int
1060 process_fetch_insn(struct fetch_insn *code, struct pt_regs *regs, void *dest,
1061                    void *base)
1062 {
1063         unsigned long val;
1064
1065 retry:
1066         /* 1st stage: get value from context */
1067         switch (code->op) {
1068         case FETCH_OP_REG:
1069                 val = regs_get_register(regs, code->param);
1070                 break;
1071         case FETCH_OP_STACK:
1072                 val = regs_get_kernel_stack_nth(regs, code->param);
1073                 break;
1074         case FETCH_OP_STACKP:
1075                 val = kernel_stack_pointer(regs);
1076                 break;
1077         case FETCH_OP_RETVAL:
1078                 val = regs_return_value(regs);
1079                 break;
1080         case FETCH_OP_IMM:
1081                 val = code->immediate;
1082                 break;
1083         case FETCH_OP_COMM:
1084                 val = (unsigned long)current->comm;
1085                 break;
1086         case FETCH_OP_DATA:
1087                 val = (unsigned long)code->data;
1088                 break;
1089 #ifdef CONFIG_HAVE_FUNCTION_ARG_ACCESS_API
1090         case FETCH_OP_ARG:
1091                 val = regs_get_kernel_argument(regs, code->param);
1092                 break;
1093 #endif
1094         case FETCH_NOP_SYMBOL:  /* Ignore a place holder */
1095                 code++;
1096                 goto retry;
1097         default:
1098                 return -EILSEQ;
1099         }
1100         code++;
1101
1102         return process_fetch_insn_bottom(code, val, dest, base);
1103 }
1104 NOKPROBE_SYMBOL(process_fetch_insn)
1105
1106 /* Kprobe handler */
1107 static nokprobe_inline void
1108 __kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs,
1109                     struct trace_event_file *trace_file)
1110 {
1111         struct kprobe_trace_entry_head *entry;
1112         struct ring_buffer_event *event;
1113         struct ring_buffer *buffer;
1114         int size, dsize, pc;
1115         unsigned long irq_flags;
1116         struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1117
1118         WARN_ON(call != trace_file->event_call);
1119
1120         if (trace_trigger_soft_disabled(trace_file))
1121                 return;
1122
1123         local_save_flags(irq_flags);
1124         pc = preempt_count();
1125
1126         dsize = __get_data_size(&tk->tp, regs);
1127         size = sizeof(*entry) + tk->tp.size + dsize;
1128
1129         event = trace_event_buffer_lock_reserve(&buffer, trace_file,
1130                                                 call->event.type,
1131                                                 size, irq_flags, pc);
1132         if (!event)
1133                 return;
1134
1135         entry = ring_buffer_event_data(event);
1136         entry->ip = (unsigned long)tk->rp.kp.addr;
1137         store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1138
1139         event_trigger_unlock_commit_regs(trace_file, buffer, event,
1140                                          entry, irq_flags, pc, regs);
1141 }
1142
1143 static void
1144 kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs)
1145 {
1146         struct event_file_link *link;
1147
1148         trace_probe_for_each_link_rcu(link, &tk->tp)
1149                 __kprobe_trace_func(tk, regs, link->file);
1150 }
1151 NOKPROBE_SYMBOL(kprobe_trace_func);
1152
1153 /* Kretprobe handler */
1154 static nokprobe_inline void
1155 __kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1156                        struct pt_regs *regs,
1157                        struct trace_event_file *trace_file)
1158 {
1159         struct kretprobe_trace_entry_head *entry;
1160         struct ring_buffer_event *event;
1161         struct ring_buffer *buffer;
1162         int size, pc, dsize;
1163         unsigned long irq_flags;
1164         struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1165
1166         WARN_ON(call != trace_file->event_call);
1167
1168         if (trace_trigger_soft_disabled(trace_file))
1169                 return;
1170
1171         local_save_flags(irq_flags);
1172         pc = preempt_count();
1173
1174         dsize = __get_data_size(&tk->tp, regs);
1175         size = sizeof(*entry) + tk->tp.size + dsize;
1176
1177         event = trace_event_buffer_lock_reserve(&buffer, trace_file,
1178                                                 call->event.type,
1179                                                 size, irq_flags, pc);
1180         if (!event)
1181                 return;
1182
1183         entry = ring_buffer_event_data(event);
1184         entry->func = (unsigned long)tk->rp.kp.addr;
1185         entry->ret_ip = (unsigned long)ri->ret_addr;
1186         store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1187
1188         event_trigger_unlock_commit_regs(trace_file, buffer, event,
1189                                          entry, irq_flags, pc, regs);
1190 }
1191
1192 static void
1193 kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1194                      struct pt_regs *regs)
1195 {
1196         struct event_file_link *link;
1197
1198         trace_probe_for_each_link_rcu(link, &tk->tp)
1199                 __kretprobe_trace_func(tk, ri, regs, link->file);
1200 }
1201 NOKPROBE_SYMBOL(kretprobe_trace_func);
1202
1203 /* Event entry printers */
1204 static enum print_line_t
1205 print_kprobe_event(struct trace_iterator *iter, int flags,
1206                    struct trace_event *event)
1207 {
1208         struct kprobe_trace_entry_head *field;
1209         struct trace_seq *s = &iter->seq;
1210         struct trace_probe *tp;
1211
1212         field = (struct kprobe_trace_entry_head *)iter->ent;
1213         tp = trace_probe_primary_from_call(
1214                 container_of(event, struct trace_event_call, event));
1215         if (WARN_ON_ONCE(!tp))
1216                 goto out;
1217
1218         trace_seq_printf(s, "%s: (", trace_probe_name(tp));
1219
1220         if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
1221                 goto out;
1222
1223         trace_seq_putc(s, ')');
1224
1225         if (print_probe_args(s, tp->args, tp->nr_args,
1226                              (u8 *)&field[1], field) < 0)
1227                 goto out;
1228
1229         trace_seq_putc(s, '\n');
1230  out:
1231         return trace_handle_return(s);
1232 }
1233
1234 static enum print_line_t
1235 print_kretprobe_event(struct trace_iterator *iter, int flags,
1236                       struct trace_event *event)
1237 {
1238         struct kretprobe_trace_entry_head *field;
1239         struct trace_seq *s = &iter->seq;
1240         struct trace_probe *tp;
1241
1242         field = (struct kretprobe_trace_entry_head *)iter->ent;
1243         tp = trace_probe_primary_from_call(
1244                 container_of(event, struct trace_event_call, event));
1245         if (WARN_ON_ONCE(!tp))
1246                 goto out;
1247
1248         trace_seq_printf(s, "%s: (", trace_probe_name(tp));
1249
1250         if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1251                 goto out;
1252
1253         trace_seq_puts(s, " <- ");
1254
1255         if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1256                 goto out;
1257
1258         trace_seq_putc(s, ')');
1259
1260         if (print_probe_args(s, tp->args, tp->nr_args,
1261                              (u8 *)&field[1], field) < 0)
1262                 goto out;
1263
1264         trace_seq_putc(s, '\n');
1265
1266  out:
1267         return trace_handle_return(s);
1268 }
1269
1270
1271 static int kprobe_event_define_fields(struct trace_event_call *event_call)
1272 {
1273         int ret;
1274         struct kprobe_trace_entry_head field;
1275         struct trace_probe *tp;
1276
1277         tp = trace_probe_primary_from_call(event_call);
1278         if (WARN_ON_ONCE(!tp))
1279                 return -ENOENT;
1280
1281         DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1282
1283         return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
1284 }
1285
1286 static int kretprobe_event_define_fields(struct trace_event_call *event_call)
1287 {
1288         int ret;
1289         struct kretprobe_trace_entry_head field;
1290         struct trace_probe *tp;
1291
1292         tp = trace_probe_primary_from_call(event_call);
1293         if (WARN_ON_ONCE(!tp))
1294                 return -ENOENT;
1295
1296         DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1297         DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1298
1299         return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
1300 }
1301
1302 #ifdef CONFIG_PERF_EVENTS
1303
1304 /* Kprobe profile handler */
1305 static int
1306 kprobe_perf_func(struct trace_kprobe *tk, struct pt_regs *regs)
1307 {
1308         struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1309         struct kprobe_trace_entry_head *entry;
1310         struct hlist_head *head;
1311         int size, __size, dsize;
1312         int rctx;
1313
1314         if (bpf_prog_array_valid(call)) {
1315                 unsigned long orig_ip = instruction_pointer(regs);
1316                 int ret;
1317
1318                 ret = trace_call_bpf(call, regs);
1319
1320                 /*
1321                  * We need to check and see if we modified the pc of the
1322                  * pt_regs, and if so return 1 so that we don't do the
1323                  * single stepping.
1324                  */
1325                 if (orig_ip != instruction_pointer(regs))
1326                         return 1;
1327                 if (!ret)
1328                         return 0;
1329         }
1330
1331         head = this_cpu_ptr(call->perf_events);
1332         if (hlist_empty(head))
1333                 return 0;
1334
1335         dsize = __get_data_size(&tk->tp, regs);
1336         __size = sizeof(*entry) + tk->tp.size + dsize;
1337         size = ALIGN(__size + sizeof(u32), sizeof(u64));
1338         size -= sizeof(u32);
1339
1340         entry = perf_trace_buf_alloc(size, NULL, &rctx);
1341         if (!entry)
1342                 return 0;
1343
1344         entry->ip = (unsigned long)tk->rp.kp.addr;
1345         memset(&entry[1], 0, dsize);
1346         store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1347         perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1348                               head, NULL);
1349         return 0;
1350 }
1351 NOKPROBE_SYMBOL(kprobe_perf_func);
1352
1353 /* Kretprobe profile handler */
1354 static void
1355 kretprobe_perf_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1356                     struct pt_regs *regs)
1357 {
1358         struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1359         struct kretprobe_trace_entry_head *entry;
1360         struct hlist_head *head;
1361         int size, __size, dsize;
1362         int rctx;
1363
1364         if (bpf_prog_array_valid(call) && !trace_call_bpf(call, regs))
1365                 return;
1366
1367         head = this_cpu_ptr(call->perf_events);
1368         if (hlist_empty(head))
1369                 return;
1370
1371         dsize = __get_data_size(&tk->tp, regs);
1372         __size = sizeof(*entry) + tk->tp.size + dsize;
1373         size = ALIGN(__size + sizeof(u32), sizeof(u64));
1374         size -= sizeof(u32);
1375
1376         entry = perf_trace_buf_alloc(size, NULL, &rctx);
1377         if (!entry)
1378                 return;
1379
1380         entry->func = (unsigned long)tk->rp.kp.addr;
1381         entry->ret_ip = (unsigned long)ri->ret_addr;
1382         store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1383         perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1384                               head, NULL);
1385 }
1386 NOKPROBE_SYMBOL(kretprobe_perf_func);
1387
1388 int bpf_get_kprobe_info(const struct perf_event *event, u32 *fd_type,
1389                         const char **symbol, u64 *probe_offset,
1390                         u64 *probe_addr, bool perf_type_tracepoint)
1391 {
1392         const char *pevent = trace_event_name(event->tp_event);
1393         const char *group = event->tp_event->class->system;
1394         struct trace_kprobe *tk;
1395
1396         if (perf_type_tracepoint)
1397                 tk = find_trace_kprobe(pevent, group);
1398         else
1399                 tk = event->tp_event->data;
1400         if (!tk)
1401                 return -EINVAL;
1402
1403         *fd_type = trace_kprobe_is_return(tk) ? BPF_FD_TYPE_KRETPROBE
1404                                               : BPF_FD_TYPE_KPROBE;
1405         if (tk->symbol) {
1406                 *symbol = tk->symbol;
1407                 *probe_offset = tk->rp.kp.offset;
1408                 *probe_addr = 0;
1409         } else {
1410                 *symbol = NULL;
1411                 *probe_offset = 0;
1412                 *probe_addr = (unsigned long)tk->rp.kp.addr;
1413         }
1414         return 0;
1415 }
1416 #endif  /* CONFIG_PERF_EVENTS */
1417
1418 /*
1419  * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex.
1420  *
1421  * kprobe_trace_self_tests_init() does enable_trace_probe/disable_trace_probe
1422  * lockless, but we can't race with this __init function.
1423  */
1424 static int kprobe_register(struct trace_event_call *event,
1425                            enum trace_reg type, void *data)
1426 {
1427         struct trace_event_file *file = data;
1428
1429         switch (type) {
1430         case TRACE_REG_REGISTER:
1431                 return enable_trace_kprobe(event, file);
1432         case TRACE_REG_UNREGISTER:
1433                 return disable_trace_kprobe(event, file);
1434
1435 #ifdef CONFIG_PERF_EVENTS
1436         case TRACE_REG_PERF_REGISTER:
1437                 return enable_trace_kprobe(event, NULL);
1438         case TRACE_REG_PERF_UNREGISTER:
1439                 return disable_trace_kprobe(event, NULL);
1440         case TRACE_REG_PERF_OPEN:
1441         case TRACE_REG_PERF_CLOSE:
1442         case TRACE_REG_PERF_ADD:
1443         case TRACE_REG_PERF_DEL:
1444                 return 0;
1445 #endif
1446         }
1447         return 0;
1448 }
1449
1450 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1451 {
1452         struct trace_kprobe *tk = container_of(kp, struct trace_kprobe, rp.kp);
1453         int ret = 0;
1454
1455         raw_cpu_inc(*tk->nhit);
1456
1457         if (trace_probe_test_flag(&tk->tp, TP_FLAG_TRACE))
1458                 kprobe_trace_func(tk, regs);
1459 #ifdef CONFIG_PERF_EVENTS
1460         if (trace_probe_test_flag(&tk->tp, TP_FLAG_PROFILE))
1461                 ret = kprobe_perf_func(tk, regs);
1462 #endif
1463         return ret;
1464 }
1465 NOKPROBE_SYMBOL(kprobe_dispatcher);
1466
1467 static int
1468 kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1469 {
1470         struct trace_kprobe *tk = container_of(ri->rp, struct trace_kprobe, rp);
1471
1472         raw_cpu_inc(*tk->nhit);
1473
1474         if (trace_probe_test_flag(&tk->tp, TP_FLAG_TRACE))
1475                 kretprobe_trace_func(tk, ri, regs);
1476 #ifdef CONFIG_PERF_EVENTS
1477         if (trace_probe_test_flag(&tk->tp, TP_FLAG_PROFILE))
1478                 kretprobe_perf_func(tk, ri, regs);
1479 #endif
1480         return 0;       /* We don't tweek kernel, so just return 0 */
1481 }
1482 NOKPROBE_SYMBOL(kretprobe_dispatcher);
1483
1484 static struct trace_event_functions kretprobe_funcs = {
1485         .trace          = print_kretprobe_event
1486 };
1487
1488 static struct trace_event_functions kprobe_funcs = {
1489         .trace          = print_kprobe_event
1490 };
1491
1492 static inline void init_trace_event_call(struct trace_kprobe *tk)
1493 {
1494         struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1495
1496         if (trace_kprobe_is_return(tk)) {
1497                 call->event.funcs = &kretprobe_funcs;
1498                 call->class->define_fields = kretprobe_event_define_fields;
1499         } else {
1500                 call->event.funcs = &kprobe_funcs;
1501                 call->class->define_fields = kprobe_event_define_fields;
1502         }
1503
1504         call->flags = TRACE_EVENT_FL_KPROBE;
1505         call->class->reg = kprobe_register;
1506 }
1507
1508 static int register_kprobe_event(struct trace_kprobe *tk)
1509 {
1510         init_trace_event_call(tk);
1511
1512         return trace_probe_register_event_call(&tk->tp);
1513 }
1514
1515 static int unregister_kprobe_event(struct trace_kprobe *tk)
1516 {
1517         return trace_probe_unregister_event_call(&tk->tp);
1518 }
1519
1520 #ifdef CONFIG_PERF_EVENTS
1521 /* create a trace_kprobe, but don't add it to global lists */
1522 struct trace_event_call *
1523 create_local_trace_kprobe(char *func, void *addr, unsigned long offs,
1524                           bool is_return)
1525 {
1526         struct trace_kprobe *tk;
1527         int ret;
1528         char *event;
1529
1530         /*
1531          * local trace_kprobes are not added to dyn_event, so they are never
1532          * searched in find_trace_kprobe(). Therefore, there is no concern of
1533          * duplicated name here.
1534          */
1535         event = func ? func : "DUMMY_EVENT";
1536
1537         tk = alloc_trace_kprobe(KPROBE_EVENT_SYSTEM, event, (void *)addr, func,
1538                                 offs, 0 /* maxactive */, 0 /* nargs */,
1539                                 is_return);
1540
1541         if (IS_ERR(tk)) {
1542                 pr_info("Failed to allocate trace_probe.(%d)\n",
1543                         (int)PTR_ERR(tk));
1544                 return ERR_CAST(tk);
1545         }
1546
1547         init_trace_event_call(tk);
1548
1549         if (traceprobe_set_print_fmt(&tk->tp, trace_kprobe_is_return(tk)) < 0) {
1550                 ret = -ENOMEM;
1551                 goto error;
1552         }
1553
1554         ret = __register_trace_kprobe(tk);
1555         if (ret < 0)
1556                 goto error;
1557
1558         return trace_probe_event_call(&tk->tp);
1559 error:
1560         free_trace_kprobe(tk);
1561         return ERR_PTR(ret);
1562 }
1563
1564 void destroy_local_trace_kprobe(struct trace_event_call *event_call)
1565 {
1566         struct trace_kprobe *tk;
1567
1568         tk = trace_kprobe_primary_from_call(event_call);
1569         if (unlikely(!tk))
1570                 return;
1571
1572         if (trace_probe_is_enabled(&tk->tp)) {
1573                 WARN_ON(1);
1574                 return;
1575         }
1576
1577         __unregister_trace_kprobe(tk);
1578
1579         free_trace_kprobe(tk);
1580 }
1581 #endif /* CONFIG_PERF_EVENTS */
1582
1583 static __init void enable_boot_kprobe_events(void)
1584 {
1585         struct trace_array *tr = top_trace_array();
1586         struct trace_event_file *file;
1587         struct trace_kprobe *tk;
1588         struct dyn_event *pos;
1589
1590         mutex_lock(&event_mutex);
1591         for_each_trace_kprobe(tk, pos) {
1592                 list_for_each_entry(file, &tr->events, list)
1593                         if (file->event_call == trace_probe_event_call(&tk->tp))
1594                                 trace_event_enable_disable(file, 1, 0);
1595         }
1596         mutex_unlock(&event_mutex);
1597 }
1598
1599 static __init void setup_boot_kprobe_events(void)
1600 {
1601         char *p, *cmd = kprobe_boot_events_buf;
1602         int ret;
1603
1604         strreplace(kprobe_boot_events_buf, ',', ' ');
1605
1606         while (cmd && *cmd != '\0') {
1607                 p = strchr(cmd, ';');
1608                 if (p)
1609                         *p++ = '\0';
1610
1611                 ret = trace_run_command(cmd, create_or_delete_trace_kprobe);
1612                 if (ret)
1613                         pr_warn("Failed to add event(%d): %s\n", ret, cmd);
1614                 else
1615                         kprobe_boot_events_enabled = true;
1616
1617                 cmd = p;
1618         }
1619
1620         enable_boot_kprobe_events();
1621 }
1622
1623 /* Make a tracefs interface for controlling probe points */
1624 static __init int init_kprobe_trace(void)
1625 {
1626         struct dentry *d_tracer;
1627         struct dentry *entry;
1628         int ret;
1629
1630         ret = dyn_event_register(&trace_kprobe_ops);
1631         if (ret)
1632                 return ret;
1633
1634         if (register_module_notifier(&trace_kprobe_module_nb))
1635                 return -EINVAL;
1636
1637         d_tracer = tracing_init_dentry();
1638         if (IS_ERR(d_tracer))
1639                 return 0;
1640
1641         entry = tracefs_create_file("kprobe_events", 0644, d_tracer,
1642                                     NULL, &kprobe_events_ops);
1643
1644         /* Event list interface */
1645         if (!entry)
1646                 pr_warn("Could not create tracefs 'kprobe_events' entry\n");
1647
1648         /* Profile interface */
1649         entry = tracefs_create_file("kprobe_profile", 0444, d_tracer,
1650                                     NULL, &kprobe_profile_ops);
1651
1652         if (!entry)
1653                 pr_warn("Could not create tracefs 'kprobe_profile' entry\n");
1654
1655         setup_boot_kprobe_events();
1656
1657         return 0;
1658 }
1659 fs_initcall(init_kprobe_trace);
1660
1661
1662 #ifdef CONFIG_FTRACE_STARTUP_TEST
1663 static __init struct trace_event_file *
1664 find_trace_probe_file(struct trace_kprobe *tk, struct trace_array *tr)
1665 {
1666         struct trace_event_file *file;
1667
1668         list_for_each_entry(file, &tr->events, list)
1669                 if (file->event_call == trace_probe_event_call(&tk->tp))
1670                         return file;
1671
1672         return NULL;
1673 }
1674
1675 /*
1676  * Nobody but us can call enable_trace_kprobe/disable_trace_kprobe at this
1677  * stage, we can do this lockless.
1678  */
1679 static __init int kprobe_trace_self_tests_init(void)
1680 {
1681         int ret, warn = 0;
1682         int (*target)(int, int, int, int, int, int);
1683         struct trace_kprobe *tk;
1684         struct trace_event_file *file;
1685
1686         if (tracing_is_disabled())
1687                 return -ENODEV;
1688
1689         if (kprobe_boot_events_enabled) {
1690                 pr_info("Skipping kprobe tests due to kprobe_event on cmdline\n");
1691                 return 0;
1692         }
1693
1694         target = kprobe_trace_selftest_target;
1695
1696         pr_info("Testing kprobe tracing: ");
1697
1698         ret = trace_run_command("p:testprobe kprobe_trace_selftest_target $stack $stack0 +0($stack)",
1699                                 create_or_delete_trace_kprobe);
1700         if (WARN_ON_ONCE(ret)) {
1701                 pr_warn("error on probing function entry.\n");
1702                 warn++;
1703         } else {
1704                 /* Enable trace point */
1705                 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1706                 if (WARN_ON_ONCE(tk == NULL)) {
1707                         pr_warn("error on getting new probe.\n");
1708                         warn++;
1709                 } else {
1710                         file = find_trace_probe_file(tk, top_trace_array());
1711                         if (WARN_ON_ONCE(file == NULL)) {
1712                                 pr_warn("error on getting probe file.\n");
1713                                 warn++;
1714                         } else
1715                                 enable_trace_kprobe(
1716                                         trace_probe_event_call(&tk->tp), file);
1717                 }
1718         }
1719
1720         ret = trace_run_command("r:testprobe2 kprobe_trace_selftest_target $retval",
1721                                 create_or_delete_trace_kprobe);
1722         if (WARN_ON_ONCE(ret)) {
1723                 pr_warn("error on probing function return.\n");
1724                 warn++;
1725         } else {
1726                 /* Enable trace point */
1727                 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1728                 if (WARN_ON_ONCE(tk == NULL)) {
1729                         pr_warn("error on getting 2nd new probe.\n");
1730                         warn++;
1731                 } else {
1732                         file = find_trace_probe_file(tk, top_trace_array());
1733                         if (WARN_ON_ONCE(file == NULL)) {
1734                                 pr_warn("error on getting probe file.\n");
1735                                 warn++;
1736                         } else
1737                                 enable_trace_kprobe(
1738                                         trace_probe_event_call(&tk->tp), file);
1739                 }
1740         }
1741
1742         if (warn)
1743                 goto end;
1744
1745         ret = target(1, 2, 3, 4, 5, 6);
1746
1747         /*
1748          * Not expecting an error here, the check is only to prevent the
1749          * optimizer from removing the call to target() as otherwise there
1750          * are no side-effects and the call is never performed.
1751          */
1752         if (ret != 21)
1753                 warn++;
1754
1755         /* Disable trace points before removing it */
1756         tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1757         if (WARN_ON_ONCE(tk == NULL)) {
1758                 pr_warn("error on getting test probe.\n");
1759                 warn++;
1760         } else {
1761                 if (trace_kprobe_nhit(tk) != 1) {
1762                         pr_warn("incorrect number of testprobe hits\n");
1763                         warn++;
1764                 }
1765
1766                 file = find_trace_probe_file(tk, top_trace_array());
1767                 if (WARN_ON_ONCE(file == NULL)) {
1768                         pr_warn("error on getting probe file.\n");
1769                         warn++;
1770                 } else
1771                         disable_trace_kprobe(
1772                                 trace_probe_event_call(&tk->tp), file);
1773         }
1774
1775         tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1776         if (WARN_ON_ONCE(tk == NULL)) {
1777                 pr_warn("error on getting 2nd test probe.\n");
1778                 warn++;
1779         } else {
1780                 if (trace_kprobe_nhit(tk) != 1) {
1781                         pr_warn("incorrect number of testprobe2 hits\n");
1782                         warn++;
1783                 }
1784
1785                 file = find_trace_probe_file(tk, top_trace_array());
1786                 if (WARN_ON_ONCE(file == NULL)) {
1787                         pr_warn("error on getting probe file.\n");
1788                         warn++;
1789                 } else
1790                         disable_trace_kprobe(
1791                                 trace_probe_event_call(&tk->tp), file);
1792         }
1793
1794         ret = trace_run_command("-:testprobe", create_or_delete_trace_kprobe);
1795         if (WARN_ON_ONCE(ret)) {
1796                 pr_warn("error on deleting a probe.\n");
1797                 warn++;
1798         }
1799
1800         ret = trace_run_command("-:testprobe2", create_or_delete_trace_kprobe);
1801         if (WARN_ON_ONCE(ret)) {
1802                 pr_warn("error on deleting a probe.\n");
1803                 warn++;
1804         }
1805
1806 end:
1807         ret = dyn_events_release_all(&trace_kprobe_ops);
1808         if (WARN_ON_ONCE(ret)) {
1809                 pr_warn("error on cleaning up probes.\n");
1810                 warn++;
1811         }
1812         /*
1813          * Wait for the optimizer work to finish. Otherwise it might fiddle
1814          * with probes in already freed __init text.
1815          */
1816         wait_for_kprobe_optimizer();
1817         if (warn)
1818                 pr_cont("NG: Some tests are failed. Please check them.\n");
1819         else
1820                 pr_cont("OK\n");
1821         return 0;
1822 }
1823
1824 late_initcall(kprobe_trace_self_tests_init);
1825
1826 #endif