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         list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
 375                 tk = container_of(pos, struct trace_kprobe, tp);
 376                 if (trace_kprobe_has_gone(tk))
 377                         continue;
 378                 ret = __enable_trace_kprobe(tk);
 379                 if (ret)
 380                         break;
 381                 enabled = true;
 382         }
 383
 384         if (ret) {
 385                 /* Failed to enable one of them. Roll back all */
 386                 if (enabled)
 387                         __disable_trace_kprobe(tp);
 388                 if (file)
 389                         trace_probe_remove_file(tp, file);
 390                 else
 391                         trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
 392         }
 393
 394         return ret;
 395 }
 396
 397 /*
 398  * Disable trace_probe
 399  * if the file is NULL, disable "perf" handler, or disable "trace" handler.
 400  */
 401 static int disable_trace_kprobe(struct trace_event_call *call,
 402                                 struct trace_event_file *file)
 403 {
 404         struct trace_probe *tp;
 405
 406         tp = trace_probe_primary_from_call(call);
 407         if (WARN_ON_ONCE(!tp))
 408                 return -ENODEV;
 409
 410         if (file) {
 411                 if (!trace_probe_get_file_link(tp, file))
 412                         return -ENOENT;
 413                 if (!trace_probe_has_single_file(tp))
 414                         goto out;
 415                 trace_probe_clear_flag(tp, TP_FLAG_TRACE);
 416         } else
 417                 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
 418
 419         if (!trace_probe_is_enabled(tp))
 420                 __disable_trace_kprobe(tp);
 421
 422  out:
 423         if (file)
 424                 /*
 425                  * Synchronization is done in below function. For perf event,
 426                  * file == NULL and perf_trace_event_unreg() calls
 427                  * tracepoint_synchronize_unregister() to ensure synchronize
 428                  * event. We don't need to care about it.
 429                  */
 430                 trace_probe_remove_file(tp, file);
 431
 432         return 0;
 433 }
 434
 435 #if defined(CONFIG_KPROBES_ON_FTRACE) && \
 436         !defined(CONFIG_KPROBE_EVENTS_ON_NOTRACE)
 437 static bool within_notrace_func(struct trace_kprobe *tk)
 438 {
 439         unsigned long offset, size, addr;
 440
 441         addr = trace_kprobe_address(tk);
 442         if (!addr || !kallsyms_lookup_size_offset(addr, &size, &offset))
 443                 return false;
 444
 445         /* Get the entry address of the target function */
 446         addr -= offset;
 447
 448         /*
 449          * Since ftrace_location_range() does inclusive range check, we need
 450          * to subtract 1 byte from the end address.
 451          */
 452         return !ftrace_location_range(addr, addr + size - 1);
 453 }
 454 #else
 455 #define within_notrace_func(tk) (false)
 456 #endif
 457
 458 /* Internal register function - just handle k*probes and flags */
 459 static int __register_trace_kprobe(struct trace_kprobe *tk)
 460 {
 461         int i, ret;
 462
 463         if (trace_kprobe_is_registered(tk))
 464                 return -EINVAL;
 465
 466         if (within_notrace_func(tk)) {
 467                 pr_warn("Could not probe notrace function %s\n",
 468                         trace_kprobe_symbol(tk));
 469                 return -EINVAL;
 470         }
 471
 472         for (i = 0; i < tk->tp.nr_args; i++) {
 473                 ret = traceprobe_update_arg(&tk->tp.args[i]);
 474                 if (ret)
 475                         return ret;
 476         }
 477
 478         /* Set/clear disabled flag according to tp->flag */
 479         if (trace_probe_is_enabled(&tk->tp))
 480                 tk->rp.kp.flags &= ~KPROBE_FLAG_DISABLED;
 481         else
 482                 tk->rp.kp.flags |= KPROBE_FLAG_DISABLED;
 483
 484         if (trace_kprobe_is_return(tk))
 485                 ret = register_kretprobe(&tk->rp);
 486         else
 487                 ret = register_kprobe(&tk->rp.kp);
 488
 489         return ret;
 490 }
 491
 492 /* Internal unregister function - just handle k*probes and flags */
 493 static void __unregister_trace_kprobe(struct trace_kprobe *tk)
 494 {
 495         if (trace_kprobe_is_registered(tk)) {
 496                 if (trace_kprobe_is_return(tk))
 497                         unregister_kretprobe(&tk->rp);
 498                 else
 499                         unregister_kprobe(&tk->rp.kp);
 500                 /* Cleanup kprobe for reuse and mark it unregistered */
 501                 INIT_HLIST_NODE(&tk->rp.kp.hlist);
 502                 INIT_LIST_HEAD(&tk->rp.kp.list);
 503                 if (tk->rp.kp.symbol_name)
 504                         tk->rp.kp.addr = NULL;
 505         }
 506 }
 507
 508 /* Unregister a trace_probe and probe_event */
 509 static int unregister_trace_kprobe(struct trace_kprobe *tk)
 510 {
 511         /* If other probes are on the event, just unregister kprobe */
 512         if (trace_probe_has_sibling(&tk->tp))
 513                 goto unreg;
 514
 515         /* Enabled event can not be unregistered */
 516         if (trace_probe_is_enabled(&tk->tp))
 517                 return -EBUSY;
 518
 519         /* Will fail if probe is being used by ftrace or perf */
 520         if (unregister_kprobe_event(tk))
 521                 return -EBUSY;
 522
 523 unreg:
 524         __unregister_trace_kprobe(tk);
 525         dyn_event_remove(&tk->devent);
 526         trace_probe_unlink(&tk->tp);
 527
 528         return 0;
 529 }
 530
 531 static bool trace_kprobe_has_same_kprobe(struct trace_kprobe *orig,
 532                                          struct trace_kprobe *comp)
 533 {
 534         struct trace_probe_event *tpe = orig->tp.event;
 535         struct trace_probe *pos;
 536         int i;
 537
 538         list_for_each_entry(pos, &tpe->probes, list) {
 539                 orig = container_of(pos, struct trace_kprobe, tp);
 540                 if (strcmp(trace_kprobe_symbol(orig),
 541                            trace_kprobe_symbol(comp)) ||
 542                     trace_kprobe_offset(orig) != trace_kprobe_offset(comp))
 543                         continue;
 544
 545                 /*
 546                  * trace_probe_compare_arg_type() ensured that nr_args and
 547                  * each argument name and type are same. Let's compare comm.
 548                  */
 549                 for (i = 0; i < orig->tp.nr_args; i++) {
 550                         if (strcmp(orig->tp.args[i].comm,
 551                                    comp->tp.args[i].comm))
 552                                 continue;
 553                 }
 554
 555                 return true;
 556         }
 557
 558         return false;
 559 }
 560
 561 static int append_trace_kprobe(struct trace_kprobe *tk, struct trace_kprobe *to)
 562 {
 563         int ret;
 564
 565         ret = trace_probe_compare_arg_type(&tk->tp, &to->tp);
 566         if (ret) {
 567                 /* Note that argument starts index = 2 */
 568                 trace_probe_log_set_index(ret + 1);
 569                 trace_probe_log_err(0, DIFF_ARG_TYPE);
 570                 return -EEXIST;
 571         }
 572         if (trace_kprobe_has_same_kprobe(to, tk)) {
 573                 trace_probe_log_set_index(0);
 574                 trace_probe_log_err(0, SAME_PROBE);
 575                 return -EEXIST;
 576         }
 577
 578         /* Append to existing event */
 579         ret = trace_probe_append(&tk->tp, &to->tp);
 580         if (ret)
 581                 return ret;
 582
 583         /* Register k*probe */
 584         ret = __register_trace_kprobe(tk);
 585         if (ret == -ENOENT && !trace_kprobe_module_exist(tk)) {
 586                 pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
 587                 ret = 0;
 588         }
 589
 590         if (ret)
 591                 trace_probe_unlink(&tk->tp);
 592         else
 593                 dyn_event_add(&tk->devent);
 594
 595         return ret;
 596 }
 597
 598 /* Register a trace_probe and probe_event */
 599 static int register_trace_kprobe(struct trace_kprobe *tk)
 600 {
 601         struct trace_kprobe *old_tk;
 602         int ret;
 603
 604         mutex_lock(&event_mutex);
 605
 606         old_tk = find_trace_kprobe(trace_probe_name(&tk->tp),
 607                                    trace_probe_group_name(&tk->tp));
 608         if (old_tk) {
 609                 if (trace_kprobe_is_return(tk) != trace_kprobe_is_return(old_tk)) {
 610                         trace_probe_log_set_index(0);
 611                         trace_probe_log_err(0, DIFF_PROBE_TYPE);
 612                         ret = -EEXIST;
 613                 } else {
 614                         ret = append_trace_kprobe(tk, old_tk);
 615                 }
 616                 goto end;
 617         }
 618
 619         /* Register new event */
 620         ret = register_kprobe_event(tk);
 621         if (ret) {
 622                 pr_warn("Failed to register probe event(%d)\n", ret);
 623                 goto end;
 624         }
 625
 626         /* Register k*probe */
 627         ret = __register_trace_kprobe(tk);
 628         if (ret == -ENOENT && !trace_kprobe_module_exist(tk)) {
 629                 pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
 630                 ret = 0;
 631         }
 632
 633         if (ret < 0)
 634                 unregister_kprobe_event(tk);
 635         else
 636                 dyn_event_add(&tk->devent);
 637
 638 end:
 639         mutex_unlock(&event_mutex);
 640         return ret;
 641 }
 642
 643 /* Module notifier call back, checking event on the module */
 644 static int trace_kprobe_module_callback(struct notifier_block *nb,
 645                                        unsigned long val, void *data)
 646 {
 647         struct module *mod = data;
 648         struct dyn_event *pos;
 649         struct trace_kprobe *tk;
 650         int ret;
 651
 652         if (val != MODULE_STATE_COMING)
 653                 return NOTIFY_DONE;
 654
 655         /* Update probes on coming module */
 656         mutex_lock(&event_mutex);
 657         for_each_trace_kprobe(tk, pos) {
 658                 if (trace_kprobe_within_module(tk, mod)) {
 659                         /* Don't need to check busy - this should have gone. */
 660                         __unregister_trace_kprobe(tk);
 661                         ret = __register_trace_kprobe(tk);
 662                         if (ret)
 663                                 pr_warn("Failed to re-register probe %s on %s: %d\n",
 664                                         trace_probe_name(&tk->tp),
 665                                         mod->name, ret);
 666                 }
 667         }
 668         mutex_unlock(&event_mutex);
 669
 670         return NOTIFY_DONE;
 671 }
 672
 673 static struct notifier_block trace_kprobe_module_nb = {
 674         .notifier_call = trace_kprobe_module_callback,
 675         .priority = 1   /* Invoked after kprobe module callback */
 676 };
 677
 678 /* Convert certain expected symbols into '_' when generating event names */
 679 static inline void sanitize_event_name(char *name)
 680 {
 681         while (*name++ != '\0')
 682                 if (*name == ':' || *name == '.')
 683                         *name = '_';
 684 }
 685
 686 static int trace_kprobe_create(int argc, const char *argv[])
 687 {
 688         /*
 689          * Argument syntax:
 690          *  - Add kprobe:
 691          *      p[:[GRP/]EVENT] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
 692          *  - Add kretprobe:
 693          *      r[MAXACTIVE][:[GRP/]EVENT] [MOD:]KSYM[+0] [FETCHARGS]
 694          * Fetch args:
 695          *  $retval     : fetch return value
 696          *  $stack      : fetch stack address
 697          *  $stackN     : fetch Nth of stack (N:0-)
 698          *  $comm       : fetch current task comm
 699          *  @ADDR       : fetch memory at ADDR (ADDR should be in kernel)
 700          *  @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
 701          *  %REG        : fetch register REG
 702          * Dereferencing memory fetch:
 703          *  +|-offs(ARG) : fetch memory at ARG +|- offs address.
 704          * Alias name of args:
 705          *  NAME=FETCHARG : set NAME as alias of FETCHARG.
 706          * Type of args:
 707          *  FETCHARG:TYPE : use TYPE instead of unsigned long.
 708          */
 709         struct trace_kprobe *tk = NULL;
 710         int i, len, ret = 0;
 711         bool is_return = false;
 712         char *symbol = NULL, *tmp = NULL;
 713         const char *event = NULL, *group = KPROBE_EVENT_SYSTEM;
 714         int maxactive = 0;
 715         long offset = 0;
 716         void *addr = NULL;
 717         char buf[MAX_EVENT_NAME_LEN];
 718         unsigned int flags = TPARG_FL_KERNEL;
 719
 720         switch (argv[0][0]) {
 721         case 'r':
 722                 is_return = true;
 723                 flags |= TPARG_FL_RETURN;
 724                 break;
 725         case 'p':
 726                 break;
 727         default:
 728                 return -ECANCELED;
 729         }
 730         if (argc < 2)
 731                 return -ECANCELED;
 732
 733         trace_probe_log_init("trace_kprobe", argc, argv);
 734
 735         event = strchr(&argv[0][1], ':');
 736         if (event)
 737                 event++;
 738
 739         if (isdigit(argv[0][1])) {
 740                 if (!is_return) {
 741                         trace_probe_log_err(1, MAXACT_NO_KPROBE);
 742                         goto parse_error;
 743                 }
 744                 if (event)
 745                         len = event - &argv[0][1] - 1;
 746                 else
 747                         len = strlen(&argv[0][1]);
 748                 if (len > MAX_EVENT_NAME_LEN - 1) {
 749                         trace_probe_log_err(1, BAD_MAXACT);
 750                         goto parse_error;
 751                 }
 752                 memcpy(buf, &argv[0][1], len);
 753                 buf[len] = '\0';
 754                 ret = kstrtouint(buf, 0, &maxactive);
 755                 if (ret || !maxactive) {
 756                         trace_probe_log_err(1, BAD_MAXACT);
 757                         goto parse_error;
 758                 }
 759                 /* kretprobes instances are iterated over via a list. The
 760                  * maximum should stay reasonable.
 761                  */
 762                 if (maxactive > KRETPROBE_MAXACTIVE_MAX) {
 763                         trace_probe_log_err(1, MAXACT_TOO_BIG);
 764                         goto parse_error;
 765                 }
 766         }
 767
 768         /* try to parse an address. if that fails, try to read the
 769          * input as a symbol. */
 770         if (kstrtoul(argv[1], 0, (unsigned long *)&addr)) {
 771                 trace_probe_log_set_index(1);
 772                 /* Check whether uprobe event specified */
 773                 if (strchr(argv[1], '/') && strchr(argv[1], ':')) {
 774                         ret = -ECANCELED;
 775                         goto error;
 776                 }
 777                 /* a symbol specified */
 778                 symbol = kstrdup(argv[1], GFP_KERNEL);
 779                 if (!symbol)
 780                         return -ENOMEM;
 781                 /* TODO: support .init module functions */
 782                 ret = traceprobe_split_symbol_offset(symbol, &offset);
 783                 if (ret || offset < 0 || offset > UINT_MAX) {
 784                         trace_probe_log_err(0, BAD_PROBE_ADDR);
 785                         goto parse_error;
 786                 }
 787                 if (kprobe_on_func_entry(NULL, symbol, offset))
 788                         flags |= TPARG_FL_FENTRY;
 789                 if (offset && is_return && !(flags & TPARG_FL_FENTRY)) {
 790                         trace_probe_log_err(0, BAD_RETPROBE);
 791                         goto parse_error;
 792                 }
 793         }
 794
 795         trace_probe_log_set_index(0);
 796         if (event) {
 797                 ret = traceprobe_parse_event_name(&event, &group, buf,
 798                                                   event - argv[0]);
 799                 if (ret)
 800                         goto parse_error;
 801         } else {
 802                 /* Make a new event name */
 803                 if (symbol)
 804                         snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
 805                                  is_return ? 'r' : 'p', symbol, offset);
 806                 else
 807                         snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
 808                                  is_return ? 'r' : 'p', addr);
 809                 sanitize_event_name(buf);
 810                 event = buf;
 811         }
 812
 813         /* setup a probe */
 814         tk = alloc_trace_kprobe(group, event, addr, symbol, offset, maxactive,
 815                                argc - 2, is_return);
 816         if (IS_ERR(tk)) {
 817                 ret = PTR_ERR(tk);
 818                 /* This must return -ENOMEM, else there is a bug */
 819                 WARN_ON_ONCE(ret != -ENOMEM);
 820                 goto out;       /* We know tk is not allocated */
 821         }
 822         argc -= 2; argv += 2;
 823
 824         /* parse arguments */
 825         for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
 826                 tmp = kstrdup(argv[i], GFP_KERNEL);
 827                 if (!tmp) {
 828                         ret = -ENOMEM;
 829                         goto error;
 830                 }
 831
 832                 trace_probe_log_set_index(i + 2);
 833                 ret = traceprobe_parse_probe_arg(&tk->tp, i, tmp, flags);
 834                 kfree(tmp);
 835                 if (ret)
 836                         goto error;     /* This can be -ENOMEM */
 837         }
 838
 839         ret = traceprobe_set_print_fmt(&tk->tp, is_return);
 840         if (ret < 0)
 841                 goto error;
 842
 843         ret = register_trace_kprobe(tk);
 844         if (ret) {
 845                 trace_probe_log_set_index(1);
 846                 if (ret == -EILSEQ)
 847                         trace_probe_log_err(0, BAD_INSN_BNDRY);
 848                 else if (ret == -ENOENT)
 849                         trace_probe_log_err(0, BAD_PROBE_ADDR);
 850                 else if (ret != -ENOMEM && ret != -EEXIST)
 851                         trace_probe_log_err(0, FAIL_REG_PROBE);
 852                 goto error;
 853         }
 854
 855 out:
 856         trace_probe_log_clear();
 857         kfree(symbol);
 858         return ret;
 859
 860 parse_error:
 861         ret = -EINVAL;
 862 error:
 863         free_trace_kprobe(tk);
 864         goto out;
 865 }
 866
 867 static int create_or_delete_trace_kprobe(int argc, char **argv)
 868 {
 869         int ret;
 870
 871         if (argv[0][0] == '-')
 872                 return dyn_event_release(argc, argv, &trace_kprobe_ops);
 873
 874         ret = trace_kprobe_create(argc, (const char **)argv);
 875         return ret == -ECANCELED ? -EINVAL : ret;
 876 }
 877
 878 static int trace_kprobe_release(struct dyn_event *ev)
 879 {
 880         struct trace_kprobe *tk = to_trace_kprobe(ev);
 881         int ret = unregister_trace_kprobe(tk);
 882
 883         if (!ret)
 884                 free_trace_kprobe(tk);
 885         return ret;
 886 }
 887
 888 static int trace_kprobe_show(struct seq_file *m, struct dyn_event *ev)
 889 {
 890         struct trace_kprobe *tk = to_trace_kprobe(ev);
 891         int i;
 892
 893         seq_putc(m, trace_kprobe_is_return(tk) ? 'r' : 'p');
 894         seq_printf(m, ":%s/%s", trace_probe_group_name(&tk->tp),
 895                                 trace_probe_name(&tk->tp));
 896
 897         if (!tk->symbol)
 898                 seq_printf(m, " 0x%p", tk->rp.kp.addr);
 899         else if (tk->rp.kp.offset)
 900                 seq_printf(m, " %s+%u", trace_kprobe_symbol(tk),
 901                            tk->rp.kp.offset);
 902         else
 903                 seq_printf(m, " %s", trace_kprobe_symbol(tk));
 904
 905         for (i = 0; i < tk->tp.nr_args; i++)
 906                 seq_printf(m, " %s=%s", tk->tp.args[i].name, tk->tp.args[i].comm);
 907         seq_putc(m, '\n');
 908
 909         return 0;
 910 }
 911
 912 static int probes_seq_show(struct seq_file *m, void *v)
 913 {
 914         struct dyn_event *ev = v;
 915
 916         if (!is_trace_kprobe(ev))
 917                 return 0;
 918
 919         return trace_kprobe_show(m, ev);
 920 }
 921
 922 static const struct seq_operations probes_seq_op = {
 923         .start  = dyn_event_seq_start,
 924         .next   = dyn_event_seq_next,
 925         .stop   = dyn_event_seq_stop,
 926         .show   = probes_seq_show
 927 };
 928
 929 static int probes_open(struct inode *inode, struct file *file)
 930 {
 931         int ret;
 932
 933         if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
 934                 ret = dyn_events_release_all(&trace_kprobe_ops);
 935                 if (ret < 0)
 936                         return ret;
 937         }
 938
 939         return seq_open(file, &probes_seq_op);
 940 }
 941
 942 static ssize_t probes_write(struct file *file, const char __user *buffer,
 943                             size_t count, loff_t *ppos)
 944 {
 945         return trace_parse_run_command(file, buffer, count, ppos,
 946                                        create_or_delete_trace_kprobe);
 947 }
 948
 949 static const struct file_operations kprobe_events_ops = {
 950         .owner          = THIS_MODULE,
 951         .open           = probes_open,
 952         .read           = seq_read,
 953         .llseek         = seq_lseek,
 954         .release        = seq_release,
 955         .write          = probes_write,
 956 };
 957
 958 /* Probes profiling interfaces */
 959 static int probes_profile_seq_show(struct seq_file *m, void *v)
 960 {
 961         struct dyn_event *ev = v;
 962         struct trace_kprobe *tk;
 963
 964         if (!is_trace_kprobe(ev))
 965                 return 0;
 966
 967         tk = to_trace_kprobe(ev);
 968         seq_printf(m, "  %-44s %15lu %15lu\n",
 969                    trace_probe_name(&tk->tp),
 970                    trace_kprobe_nhit(tk),
 971                    tk->rp.kp.nmissed);
 972
 973         return 0;
 974 }
 975
 976 static const struct seq_operations profile_seq_op = {
 977         .start  = dyn_event_seq_start,
 978         .next   = dyn_event_seq_next,
 979         .stop   = dyn_event_seq_stop,
 980         .show   = probes_profile_seq_show
 981 };
 982
 983 static int profile_open(struct inode *inode, struct file *file)
 984 {
 985         return seq_open(file, &profile_seq_op);
 986 }
 987
 988 static const struct file_operations kprobe_profile_ops = {
 989         .owner          = THIS_MODULE,
 990         .open           = profile_open,
 991         .read           = seq_read,
 992         .llseek         = seq_lseek,
 993         .release        = seq_release,
 994 };
 995
 996 /* Kprobe specific fetch functions */
 997
 998 /* Return the length of string -- including null terminal byte */
 999 static nokprobe_inline int
1000 fetch_store_strlen(unsigned long addr)
1001 {
1002         int ret, len = 0;
1003         u8 c;
1004
1005         do {
1006                 ret = probe_kernel_read(&c, (u8 *)addr + len, 1);
1007                 len++;
1008         } while (c && ret == 0 && len < MAX_STRING_SIZE);
1009
1010         return (ret < 0) ? ret : len;
1011 }
1012
1013 /* Return the length of string -- including null terminal byte */
1014 static nokprobe_inline int
1015 fetch_store_strlen_user(unsigned long addr)
1016 {
1017         const void __user *uaddr =  (__force const void __user *)addr;
1018
1019         return strnlen_unsafe_user(uaddr, MAX_STRING_SIZE);
1020 }
1021
1022 /*
1023  * Fetch a null-terminated string. Caller MUST set *(u32 *)buf with max
1024  * length and relative data location.
1025  */
1026 static nokprobe_inline int
1027 fetch_store_string(unsigned long addr, void *dest, void *base)
1028 {
1029         int maxlen = get_loc_len(*(u32 *)dest);
1030         void *__dest;
1031         long ret;
1032
1033         if (unlikely(!maxlen))
1034                 return -ENOMEM;
1035
1036         __dest = get_loc_data(dest, base);
1037
1038         /*
1039          * Try to get string again, since the string can be changed while
1040          * probing.
1041          */
1042         ret = strncpy_from_unsafe(__dest, (void *)addr, maxlen);
1043         if (ret >= 0)
1044                 *(u32 *)dest = make_data_loc(ret, __dest - base);
1045
1046         return ret;
1047 }
1048
1049 /*
1050  * Fetch a null-terminated string from user. Caller MUST set *(u32 *)buf
1051  * with max length and relative data location.
1052  */
1053 static nokprobe_inline int
1054 fetch_store_string_user(unsigned long addr, void *dest, void *base)
1055 {
1056         const void __user *uaddr =  (__force const void __user *)addr;
1057         int maxlen = get_loc_len(*(u32 *)dest);
1058         void *__dest;
1059         long ret;
1060
1061         if (unlikely(!maxlen))
1062                 return -ENOMEM;
1063
1064         __dest = get_loc_data(dest, base);
1065
1066         ret = strncpy_from_unsafe_user(__dest, uaddr, maxlen);
1067         if (ret >= 0)
1068                 *(u32 *)dest = make_data_loc(ret, __dest - base);
1069
1070         return ret;
1071 }
1072
1073 static nokprobe_inline int
1074 probe_mem_read(void *dest, void *src, size_t size)
1075 {
1076         return probe_kernel_read(dest, src, size);
1077 }
1078
1079 static nokprobe_inline int
1080 probe_mem_read_user(void *dest, void *src, size_t size)
1081 {
1082         const void __user *uaddr =  (__force const void __user *)src;
1083
1084         return probe_user_read(dest, uaddr, size);
1085 }
1086
1087 /* Note that we don't verify it, since the code does not come from user space */
1088 static int
1089 process_fetch_insn(struct fetch_insn *code, struct pt_regs *regs, void *dest,
1090                    void *base)
1091 {
1092         unsigned long val;
1093
1094 retry:
1095         /* 1st stage: get value from context */
1096         switch (code->op) {
1097         case FETCH_OP_REG:
1098                 val = regs_get_register(regs, code->param);
1099                 break;
1100         case FETCH_OP_STACK:
1101                 val = regs_get_kernel_stack_nth(regs, code->param);
1102                 break;
1103         case FETCH_OP_STACKP:
1104                 val = kernel_stack_pointer(regs);
1105                 break;
1106         case FETCH_OP_RETVAL:
1107                 val = regs_return_value(regs);
1108                 break;
1109         case FETCH_OP_IMM:
1110                 val = code->immediate;
1111                 break;
1112         case FETCH_OP_COMM:
1113                 val = (unsigned long)current->comm;
1114                 break;
1115         case FETCH_OP_DATA:
1116                 val = (unsigned long)code->data;
1117                 break;
1118 #ifdef CONFIG_HAVE_FUNCTION_ARG_ACCESS_API
1119         case FETCH_OP_ARG:
1120                 val = regs_get_kernel_argument(regs, code->param);
1121                 break;
1122 #endif
1123         case FETCH_NOP_SYMBOL:  /* Ignore a place holder */
1124                 code++;
1125                 goto retry;
1126         default:
1127                 return -EILSEQ;
1128         }
1129         code++;
1130
1131         return process_fetch_insn_bottom(code, val, dest, base);
1132 }
1133 NOKPROBE_SYMBOL(process_fetch_insn)
1134
1135 /* Kprobe handler */
1136 static nokprobe_inline void
1137 __kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs,
1138                     struct trace_event_file *trace_file)
1139 {
1140         struct kprobe_trace_entry_head *entry;
1141         struct ring_buffer_event *event;
1142         struct ring_buffer *buffer;
1143         int size, dsize, pc;
1144         unsigned long irq_flags;
1145         struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1146
1147         WARN_ON(call != trace_file->event_call);
1148
1149         if (trace_trigger_soft_disabled(trace_file))
1150                 return;
1151
1152         local_save_flags(irq_flags);
1153         pc = preempt_count();
1154
1155         dsize = __get_data_size(&tk->tp, regs);
1156         size = sizeof(*entry) + tk->tp.size + dsize;
1157
1158         event = trace_event_buffer_lock_reserve(&buffer, trace_file,
1159                                                 call->event.type,
1160                                                 size, irq_flags, pc);
1161         if (!event)
1162                 return;
1163
1164         entry = ring_buffer_event_data(event);
1165         entry->ip = (unsigned long)tk->rp.kp.addr;
1166         store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1167
1168         event_trigger_unlock_commit_regs(trace_file, buffer, event,
1169                                          entry, irq_flags, pc, regs);
1170 }
1171
1172 static void
1173 kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs)
1174 {
1175         struct event_file_link *link;
1176
1177         trace_probe_for_each_link_rcu(link, &tk->tp)
1178                 __kprobe_trace_func(tk, regs, link->file);
1179 }
1180 NOKPROBE_SYMBOL(kprobe_trace_func);
1181
1182 /* Kretprobe handler */
1183 static nokprobe_inline void
1184 __kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1185                        struct pt_regs *regs,
1186                        struct trace_event_file *trace_file)
1187 {
1188         struct kretprobe_trace_entry_head *entry;
1189         struct ring_buffer_event *event;
1190         struct ring_buffer *buffer;
1191         int size, pc, dsize;
1192         unsigned long irq_flags;
1193         struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1194
1195         WARN_ON(call != trace_file->event_call);
1196
1197         if (trace_trigger_soft_disabled(trace_file))
1198                 return;
1199
1200         local_save_flags(irq_flags);
1201         pc = preempt_count();
1202
1203         dsize = __get_data_size(&tk->tp, regs);
1204         size = sizeof(*entry) + tk->tp.size + dsize;
1205
1206         event = trace_event_buffer_lock_reserve(&buffer, trace_file,
1207                                                 call->event.type,
1208                                                 size, irq_flags, pc);
1209         if (!event)
1210                 return;
1211
1212         entry = ring_buffer_event_data(event);
1213         entry->func = (unsigned long)tk->rp.kp.addr;
1214         entry->ret_ip = (unsigned long)ri->ret_addr;
1215         store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1216
1217         event_trigger_unlock_commit_regs(trace_file, buffer, event,
1218                                          entry, irq_flags, pc, regs);
1219 }
1220
1221 static void
1222 kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1223                      struct pt_regs *regs)
1224 {
1225         struct event_file_link *link;
1226
1227         trace_probe_for_each_link_rcu(link, &tk->tp)
1228                 __kretprobe_trace_func(tk, ri, regs, link->file);
1229 }
1230 NOKPROBE_SYMBOL(kretprobe_trace_func);
1231
1232 /* Event entry printers */
1233 static enum print_line_t
1234 print_kprobe_event(struct trace_iterator *iter, int flags,
1235                    struct trace_event *event)
1236 {
1237         struct kprobe_trace_entry_head *field;
1238         struct trace_seq *s = &iter->seq;
1239         struct trace_probe *tp;
1240
1241         field = (struct kprobe_trace_entry_head *)iter->ent;
1242         tp = trace_probe_primary_from_call(
1243                 container_of(event, struct trace_event_call, event));
1244         if (WARN_ON_ONCE(!tp))
1245                 goto out;
1246
1247         trace_seq_printf(s, "%s: (", trace_probe_name(tp));
1248
1249         if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
1250                 goto out;
1251
1252         trace_seq_putc(s, ')');
1253
1254         if (print_probe_args(s, tp->args, tp->nr_args,
1255                              (u8 *)&field[1], field) < 0)
1256                 goto out;
1257
1258         trace_seq_putc(s, '\n');
1259  out:
1260         return trace_handle_return(s);
1261 }
1262
1263 static enum print_line_t
1264 print_kretprobe_event(struct trace_iterator *iter, int flags,
1265                       struct trace_event *event)
1266 {
1267         struct kretprobe_trace_entry_head *field;
1268         struct trace_seq *s = &iter->seq;
1269         struct trace_probe *tp;
1270
1271         field = (struct kretprobe_trace_entry_head *)iter->ent;
1272         tp = trace_probe_primary_from_call(
1273                 container_of(event, struct trace_event_call, event));
1274         if (WARN_ON_ONCE(!tp))
1275                 goto out;
1276
1277         trace_seq_printf(s, "%s: (", trace_probe_name(tp));
1278
1279         if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1280                 goto out;
1281
1282         trace_seq_puts(s, " <- ");
1283
1284         if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1285                 goto out;
1286
1287         trace_seq_putc(s, ')');
1288
1289         if (print_probe_args(s, tp->args, tp->nr_args,
1290                              (u8 *)&field[1], field) < 0)
1291                 goto out;
1292
1293         trace_seq_putc(s, '\n');
1294
1295  out:
1296         return trace_handle_return(s);
1297 }
1298
1299
1300 static int kprobe_event_define_fields(struct trace_event_call *event_call)
1301 {
1302         int ret;
1303         struct kprobe_trace_entry_head field;
1304         struct trace_probe *tp;
1305
1306         tp = trace_probe_primary_from_call(event_call);
1307         if (WARN_ON_ONCE(!tp))
1308                 return -ENOENT;
1309
1310         DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1311
1312         return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
1313 }
1314
1315 static int kretprobe_event_define_fields(struct trace_event_call *event_call)
1316 {
1317         int ret;
1318         struct kretprobe_trace_entry_head field;
1319         struct trace_probe *tp;
1320
1321         tp = trace_probe_primary_from_call(event_call);
1322         if (WARN_ON_ONCE(!tp))
1323                 return -ENOENT;
1324
1325         DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1326         DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1327
1328         return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
1329 }
1330
1331 #ifdef CONFIG_PERF_EVENTS
1332
1333 /* Kprobe profile handler */
1334 static int
1335 kprobe_perf_func(struct trace_kprobe *tk, struct pt_regs *regs)
1336 {
1337         struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1338         struct kprobe_trace_entry_head *entry;
1339         struct hlist_head *head;
1340         int size, __size, dsize;
1341         int rctx;
1342
1343         if (bpf_prog_array_valid(call)) {
1344                 unsigned long orig_ip = instruction_pointer(regs);
1345                 int ret;
1346
1347                 ret = trace_call_bpf(call, regs);
1348
1349                 /*
1350                  * We need to check and see if we modified the pc of the
1351                  * pt_regs, and if so return 1 so that we don't do the
1352                  * single stepping.
1353                  */
1354                 if (orig_ip != instruction_pointer(regs))
1355                         return 1;
1356                 if (!ret)
1357                         return 0;
1358         }
1359
1360         head = this_cpu_ptr(call->perf_events);
1361         if (hlist_empty(head))
1362                 return 0;
1363
1364         dsize = __get_data_size(&tk->tp, regs);
1365         __size = sizeof(*entry) + tk->tp.size + dsize;
1366         size = ALIGN(__size + sizeof(u32), sizeof(u64));
1367         size -= sizeof(u32);
1368
1369         entry = perf_trace_buf_alloc(size, NULL, &rctx);
1370         if (!entry)
1371                 return 0;
1372
1373         entry->ip = (unsigned long)tk->rp.kp.addr;
1374         memset(&entry[1], 0, dsize);
1375         store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1376         perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1377                               head, NULL);
1378         return 0;
1379 }
1380 NOKPROBE_SYMBOL(kprobe_perf_func);
1381
1382 /* Kretprobe profile handler */
1383 static void
1384 kretprobe_perf_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1385                     struct pt_regs *regs)
1386 {
1387         struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1388         struct kretprobe_trace_entry_head *entry;
1389         struct hlist_head *head;
1390         int size, __size, dsize;
1391         int rctx;
1392
1393         if (bpf_prog_array_valid(call) && !trace_call_bpf(call, regs))
1394                 return;
1395
1396         head = this_cpu_ptr(call->perf_events);
1397         if (hlist_empty(head))
1398                 return;
1399
1400         dsize = __get_data_size(&tk->tp, regs);
1401         __size = sizeof(*entry) + tk->tp.size + dsize;
1402         size = ALIGN(__size + sizeof(u32), sizeof(u64));
1403         size -= sizeof(u32);
1404
1405         entry = perf_trace_buf_alloc(size, NULL, &rctx);
1406         if (!entry)
1407                 return;
1408
1409         entry->func = (unsigned long)tk->rp.kp.addr;
1410         entry->ret_ip = (unsigned long)ri->ret_addr;
1411         store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1412         perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1413                               head, NULL);
1414 }
1415 NOKPROBE_SYMBOL(kretprobe_perf_func);
1416
1417 int bpf_get_kprobe_info(const struct perf_event *event, u32 *fd_type,
1418                         const char **symbol, u64 *probe_offset,
1419                         u64 *probe_addr, bool perf_type_tracepoint)
1420 {
1421         const char *pevent = trace_event_name(event->tp_event);
1422         const char *group = event->tp_event->class->system;
1423         struct trace_kprobe *tk;
1424
1425         if (perf_type_tracepoint)
1426                 tk = find_trace_kprobe(pevent, group);
1427         else
1428                 tk = event->tp_event->data;
1429         if (!tk)
1430                 return -EINVAL;
1431
1432         *fd_type = trace_kprobe_is_return(tk) ? BPF_FD_TYPE_KRETPROBE
1433                                               : BPF_FD_TYPE_KPROBE;
1434         if (tk->symbol) {
1435                 *symbol = tk->symbol;
1436                 *probe_offset = tk->rp.kp.offset;
1437                 *probe_addr = 0;
1438         } else {
1439                 *symbol = NULL;
1440                 *probe_offset = 0;
1441                 *probe_addr = (unsigned long)tk->rp.kp.addr;
1442         }
1443         return 0;
1444 }
1445 #endif  /* CONFIG_PERF_EVENTS */
1446
1447 /*
1448  * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex.
1449  *
1450  * kprobe_trace_self_tests_init() does enable_trace_probe/disable_trace_probe
1451  * lockless, but we can't race with this __init function.
1452  */
1453 static int kprobe_register(struct trace_event_call *event,
1454                            enum trace_reg type, void *data)
1455 {
1456         struct trace_event_file *file = data;
1457
1458         switch (type) {
1459         case TRACE_REG_REGISTER:
1460                 return enable_trace_kprobe(event, file);
1461         case TRACE_REG_UNREGISTER:
1462                 return disable_trace_kprobe(event, file);
1463
1464 #ifdef CONFIG_PERF_EVENTS
1465         case TRACE_REG_PERF_REGISTER:
1466                 return enable_trace_kprobe(event, NULL);
1467         case TRACE_REG_PERF_UNREGISTER:
1468                 return disable_trace_kprobe(event, NULL);
1469         case TRACE_REG_PERF_OPEN:
1470         case TRACE_REG_PERF_CLOSE:
1471         case TRACE_REG_PERF_ADD:
1472         case TRACE_REG_PERF_DEL:
1473                 return 0;
1474 #endif
1475         }
1476         return 0;
1477 }
1478
1479 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1480 {
1481         struct trace_kprobe *tk = container_of(kp, struct trace_kprobe, rp.kp);
1482         int ret = 0;
1483
1484         raw_cpu_inc(*tk->nhit);
1485
1486         if (trace_probe_test_flag(&tk->tp, TP_FLAG_TRACE))
1487                 kprobe_trace_func(tk, regs);
1488 #ifdef CONFIG_PERF_EVENTS
1489         if (trace_probe_test_flag(&tk->tp, TP_FLAG_PROFILE))
1490                 ret = kprobe_perf_func(tk, regs);
1491 #endif
1492         return ret;
1493 }
1494 NOKPROBE_SYMBOL(kprobe_dispatcher);
1495
1496 static int
1497 kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1498 {
1499         struct trace_kprobe *tk = container_of(ri->rp, struct trace_kprobe, rp);
1500
1501         raw_cpu_inc(*tk->nhit);
1502
1503         if (trace_probe_test_flag(&tk->tp, TP_FLAG_TRACE))
1504                 kretprobe_trace_func(tk, ri, regs);
1505 #ifdef CONFIG_PERF_EVENTS
1506         if (trace_probe_test_flag(&tk->tp, TP_FLAG_PROFILE))
1507                 kretprobe_perf_func(tk, ri, regs);
1508 #endif
1509         return 0;       /* We don't tweek kernel, so just return 0 */
1510 }
1511 NOKPROBE_SYMBOL(kretprobe_dispatcher);
1512
1513 static struct trace_event_functions kretprobe_funcs = {
1514         .trace          = print_kretprobe_event
1515 };
1516
1517 static struct trace_event_functions kprobe_funcs = {
1518         .trace          = print_kprobe_event
1519 };
1520
1521 static inline void init_trace_event_call(struct trace_kprobe *tk)
1522 {
1523         struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1524
1525         if (trace_kprobe_is_return(tk)) {
1526                 call->event.funcs = &kretprobe_funcs;
1527                 call->class->define_fields = kretprobe_event_define_fields;
1528         } else {
1529                 call->event.funcs = &kprobe_funcs;
1530                 call->class->define_fields = kprobe_event_define_fields;
1531         }
1532
1533         call->flags = TRACE_EVENT_FL_KPROBE;
1534         call->class->reg = kprobe_register;
1535 }
1536
1537 static int register_kprobe_event(struct trace_kprobe *tk)
1538 {
1539         init_trace_event_call(tk);
1540
1541         return trace_probe_register_event_call(&tk->tp);
1542 }
1543
1544 static int unregister_kprobe_event(struct trace_kprobe *tk)
1545 {
1546         return trace_probe_unregister_event_call(&tk->tp);
1547 }
1548
1549 #ifdef CONFIG_PERF_EVENTS
1550 /* create a trace_kprobe, but don't add it to global lists */
1551 struct trace_event_call *
1552 create_local_trace_kprobe(char *func, void *addr, unsigned long offs,
1553                           bool is_return)
1554 {
1555         struct trace_kprobe *tk;
1556         int ret;
1557         char *event;
1558
1559         /*
1560          * local trace_kprobes are not added to dyn_event, so they are never
1561          * searched in find_trace_kprobe(). Therefore, there is no concern of
1562          * duplicated name here.
1563          */
1564         event = func ? func : "DUMMY_EVENT";
1565
1566         tk = alloc_trace_kprobe(KPROBE_EVENT_SYSTEM, event, (void *)addr, func,
1567                                 offs, 0 /* maxactive */, 0 /* nargs */,
1568                                 is_return);
1569
1570         if (IS_ERR(tk)) {
1571                 pr_info("Failed to allocate trace_probe.(%d)\n",
1572                         (int)PTR_ERR(tk));
1573                 return ERR_CAST(tk);
1574         }
1575
1576         init_trace_event_call(tk);
1577
1578         if (traceprobe_set_print_fmt(&tk->tp, trace_kprobe_is_return(tk)) < 0) {
1579                 ret = -ENOMEM;
1580                 goto error;
1581         }
1582
1583         ret = __register_trace_kprobe(tk);
1584         if (ret < 0)
1585                 goto error;
1586
1587         return trace_probe_event_call(&tk->tp);
1588 error:
1589         free_trace_kprobe(tk);
1590         return ERR_PTR(ret);
1591 }
1592
1593 void destroy_local_trace_kprobe(struct trace_event_call *event_call)
1594 {
1595         struct trace_kprobe *tk;
1596
1597         tk = trace_kprobe_primary_from_call(event_call);
1598         if (unlikely(!tk))
1599                 return;
1600
1601         if (trace_probe_is_enabled(&tk->tp)) {
1602                 WARN_ON(1);
1603                 return;
1604         }
1605
1606         __unregister_trace_kprobe(tk);
1607
1608         free_trace_kprobe(tk);
1609 }
1610 #endif /* CONFIG_PERF_EVENTS */
1611
1612 static __init void enable_boot_kprobe_events(void)
1613 {
1614         struct trace_array *tr = top_trace_array();
1615         struct trace_event_file *file;
1616         struct trace_kprobe *tk;
1617         struct dyn_event *pos;
1618
1619         mutex_lock(&event_mutex);
1620         for_each_trace_kprobe(tk, pos) {
1621                 list_for_each_entry(file, &tr->events, list)
1622                         if (file->event_call == trace_probe_event_call(&tk->tp))
1623                                 trace_event_enable_disable(file, 1, 0);
1624         }
1625         mutex_unlock(&event_mutex);
1626 }
1627
1628 static __init void setup_boot_kprobe_events(void)
1629 {
1630         char *p, *cmd = kprobe_boot_events_buf;
1631         int ret;
1632
1633         strreplace(kprobe_boot_events_buf, ',', ' ');
1634
1635         while (cmd && *cmd != '\0') {
1636                 p = strchr(cmd, ';');
1637                 if (p)
1638                         *p++ = '\0';
1639
1640                 ret = trace_run_command(cmd, create_or_delete_trace_kprobe);
1641                 if (ret)
1642                         pr_warn("Failed to add event(%d): %s\n", ret, cmd);
1643                 else
1644                         kprobe_boot_events_enabled = true;
1645
1646                 cmd = p;
1647         }
1648
1649         enable_boot_kprobe_events();
1650 }
1651
1652 /* Make a tracefs interface for controlling probe points */
1653 static __init int init_kprobe_trace(void)
1654 {
1655         struct dentry *d_tracer;
1656         struct dentry *entry;
1657         int ret;
1658
1659         ret = dyn_event_register(&trace_kprobe_ops);
1660         if (ret)
1661                 return ret;
1662
1663         if (register_module_notifier(&trace_kprobe_module_nb))
1664                 return -EINVAL;
1665
1666         d_tracer = tracing_init_dentry();
1667         if (IS_ERR(d_tracer))
1668                 return 0;
1669
1670         entry = tracefs_create_file("kprobe_events", 0644, d_tracer,
1671                                     NULL, &kprobe_events_ops);
1672
1673         /* Event list interface */
1674         if (!entry)
1675                 pr_warn("Could not create tracefs 'kprobe_events' entry\n");
1676
1677         /* Profile interface */
1678         entry = tracefs_create_file("kprobe_profile", 0444, d_tracer,
1679                                     NULL, &kprobe_profile_ops);
1680
1681         if (!entry)
1682                 pr_warn("Could not create tracefs 'kprobe_profile' entry\n");
1683
1684         setup_boot_kprobe_events();
1685
1686         return 0;
1687 }
1688 fs_initcall(init_kprobe_trace);
1689
1690
1691 #ifdef CONFIG_FTRACE_STARTUP_TEST
1692 static __init struct trace_event_file *
1693 find_trace_probe_file(struct trace_kprobe *tk, struct trace_array *tr)
1694 {
1695         struct trace_event_file *file;
1696
1697         list_for_each_entry(file, &tr->events, list)
1698                 if (file->event_call == trace_probe_event_call(&tk->tp))
1699                         return file;
1700
1701         return NULL;
1702 }
1703
1704 /*
1705  * Nobody but us can call enable_trace_kprobe/disable_trace_kprobe at this
1706  * stage, we can do this lockless.
1707  */
1708 static __init int kprobe_trace_self_tests_init(void)
1709 {
1710         int ret, warn = 0;
1711         int (*target)(int, int, int, int, int, int);
1712         struct trace_kprobe *tk;
1713         struct trace_event_file *file;
1714
1715         if (tracing_is_disabled())
1716                 return -ENODEV;
1717
1718         if (kprobe_boot_events_enabled) {
1719                 pr_info("Skipping kprobe tests due to kprobe_event on cmdline\n");
1720                 return 0;
1721         }
1722
1723         target = kprobe_trace_selftest_target;
1724
1725         pr_info("Testing kprobe tracing: ");
1726
1727         ret = trace_run_command("p:testprobe kprobe_trace_selftest_target $stack $stack0 +0($stack)",
1728                                 create_or_delete_trace_kprobe);
1729         if (WARN_ON_ONCE(ret)) {
1730                 pr_warn("error on probing function entry.\n");
1731                 warn++;
1732         } else {
1733                 /* Enable trace point */
1734                 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1735                 if (WARN_ON_ONCE(tk == NULL)) {
1736                         pr_warn("error on getting new probe.\n");
1737                         warn++;
1738                 } else {
1739                         file = find_trace_probe_file(tk, top_trace_array());
1740                         if (WARN_ON_ONCE(file == NULL)) {
1741                                 pr_warn("error on getting probe file.\n");
1742                                 warn++;
1743                         } else
1744                                 enable_trace_kprobe(
1745                                         trace_probe_event_call(&tk->tp), file);
1746                 }
1747         }
1748
1749         ret = trace_run_command("r:testprobe2 kprobe_trace_selftest_target $retval",
1750                                 create_or_delete_trace_kprobe);
1751         if (WARN_ON_ONCE(ret)) {
1752                 pr_warn("error on probing function return.\n");
1753                 warn++;
1754         } else {
1755                 /* Enable trace point */
1756                 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1757                 if (WARN_ON_ONCE(tk == NULL)) {
1758                         pr_warn("error on getting 2nd new probe.\n");
1759                         warn++;
1760                 } else {
1761                         file = find_trace_probe_file(tk, top_trace_array());
1762                         if (WARN_ON_ONCE(file == NULL)) {
1763                                 pr_warn("error on getting probe file.\n");
1764                                 warn++;
1765                         } else
1766                                 enable_trace_kprobe(
1767                                         trace_probe_event_call(&tk->tp), file);
1768                 }
1769         }
1770
1771         if (warn)
1772                 goto end;
1773
1774         ret = target(1, 2, 3, 4, 5, 6);
1775
1776         /*
1777          * Not expecting an error here, the check is only to prevent the
1778          * optimizer from removing the call to target() as otherwise there
1779          * are no side-effects and the call is never performed.
1780          */
1781         if (ret != 21)
1782                 warn++;
1783
1784         /* Disable trace points before removing it */
1785         tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1786         if (WARN_ON_ONCE(tk == NULL)) {
1787                 pr_warn("error on getting test probe.\n");
1788                 warn++;
1789         } else {
1790                 if (trace_kprobe_nhit(tk) != 1) {
1791                         pr_warn("incorrect number of testprobe hits\n");
1792                         warn++;
1793                 }
1794
1795                 file = find_trace_probe_file(tk, top_trace_array());
1796                 if (WARN_ON_ONCE(file == NULL)) {
1797                         pr_warn("error on getting probe file.\n");
1798                         warn++;
1799                 } else
1800                         disable_trace_kprobe(
1801                                 trace_probe_event_call(&tk->tp), file);
1802         }
1803
1804         tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1805         if (WARN_ON_ONCE(tk == NULL)) {
1806                 pr_warn("error on getting 2nd test probe.\n");
1807                 warn++;
1808         } else {
1809                 if (trace_kprobe_nhit(tk) != 1) {
1810                         pr_warn("incorrect number of testprobe2 hits\n");
1811                         warn++;
1812                 }
1813
1814                 file = find_trace_probe_file(tk, top_trace_array());
1815                 if (WARN_ON_ONCE(file == NULL)) {
1816                         pr_warn("error on getting probe file.\n");
1817                         warn++;
1818                 } else
1819                         disable_trace_kprobe(
1820                                 trace_probe_event_call(&tk->tp), file);
1821         }
1822
1823         ret = trace_run_command("-:testprobe", create_or_delete_trace_kprobe);
1824         if (WARN_ON_ONCE(ret)) {
1825                 pr_warn("error on deleting a probe.\n");
1826                 warn++;
1827         }
1828
1829         ret = trace_run_command("-:testprobe2", create_or_delete_trace_kprobe);
1830         if (WARN_ON_ONCE(ret)) {
1831                 pr_warn("error on deleting a probe.\n");
1832                 warn++;
1833         }
1834
1835 end:
1836         ret = dyn_events_release_all(&trace_kprobe_ops);
1837         if (WARN_ON_ONCE(ret)) {
1838                 pr_warn("error on cleaning up probes.\n");
1839                 warn++;
1840         }
1841         /*
1842          * Wait for the optimizer work to finish. Otherwise it might fiddle
1843          * with probes in already freed __init text.
1844          */
1845         wait_for_kprobe_optimizer();
1846         if (warn)
1847                 pr_cont("NG: Some tests are failed. Please check them.\n");
1848         else
1849                 pr_cont("OK\n");
1850         return 0;
1851 }
1852
1853 late_initcall(kprobe_trace_self_tests_init);
1854
1855 #endif