kernel/tracepoint.c

   1 /*
   2  * Copyright (C) 2008-2014 Mathieu Desnoyers
   3  *
   4  * This program is free software; you can redistribute it and/or modify
   5  * it under the terms of the GNU General Public License as published by
   6  * the Free Software Foundation; either version 2 of the License, or
   7  * (at your option) any later version.
   8  *
   9  * This program is distributed in the hope that it will be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write to the Free Software
  16  * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
  17  */
  18 #include <linux/module.h>
  19 #include <linux/mutex.h>
  20 #include <linux/types.h>
  21 #include <linux/jhash.h>
  22 #include <linux/list.h>
  23 #include <linux/rcupdate.h>
  24 #include <linux/tracepoint.h>
  25 #include <linux/err.h>
  26 #include <linux/slab.h>
  27 #include <linux/sched/signal.h>
  28 #include <linux/sched/task.h>
  29 #include <linux/static_key.h>
  30
  31 extern struct tracepoint * const __start___tracepoints_ptrs[];
  32 extern struct tracepoint * const __stop___tracepoints_ptrs[];
  33
  34 DEFINE_SRCU(tracepoint_srcu);
  35 EXPORT_SYMBOL_GPL(tracepoint_srcu);
  36
  37 /* Set to 1 to enable tracepoint debug output */
  38 static const int tracepoint_debug;
  39
  40 #ifdef CONFIG_MODULES
  41 /*
  42  * Tracepoint module list mutex protects the local module list.
  43  */
  44 static DEFINE_MUTEX(tracepoint_module_list_mutex);
  45
  46 /* Local list of struct tp_module */
  47 static LIST_HEAD(tracepoint_module_list);
  48 #endif /* CONFIG_MODULES */
  49
  50 /*
  51  * tracepoints_mutex protects the builtin and module tracepoints.
  52  * tracepoints_mutex nests inside tracepoint_module_list_mutex.
  53  */
  54 static DEFINE_MUTEX(tracepoints_mutex);
  55
  56 static struct rcu_head *early_probes;
  57 static bool ok_to_free_tracepoints;
  58
  59 /*
  60  * Note about RCU :
  61  * It is used to delay the free of multiple probes array until a quiescent
  62  * state is reached.
  63  */
  64 struct tp_probes {
  65         struct rcu_head rcu;
  66         struct tracepoint_func probes[0];
  67 };
  68
  69 static inline void *allocate_probes(int count)
  70 {
  71         struct tp_probes *p  = kmalloc(count * sizeof(struct tracepoint_func)
  72                         + sizeof(struct tp_probes), GFP_KERNEL);
  73         return p == NULL ? NULL : p->probes;
  74 }
  75
  76 static void srcu_free_old_probes(struct rcu_head *head)
  77 {
  78         kfree(container_of(head, struct tp_probes, rcu));
  79 }
  80
  81 static void rcu_free_old_probes(struct rcu_head *head)
  82 {
  83         call_srcu(&tracepoint_srcu, head, srcu_free_old_probes);
  84 }
  85
  86 static __init int release_early_probes(void)
  87 {
  88         struct rcu_head *tmp;
  89
  90         ok_to_free_tracepoints = true;
  91
  92         while (early_probes) {
  93                 tmp = early_probes;
  94                 early_probes = tmp->next;
  95                 call_rcu_sched(tmp, rcu_free_old_probes);
  96         }
  97
  98         return 0;
  99 }
 100
 101 /* SRCU is initialized at core_initcall */
 102 postcore_initcall(release_early_probes);
 103
 104 static inline void release_probes(struct tracepoint_func *old)
 105 {
 106         if (old) {
 107                 struct tp_probes *tp_probes = container_of(old,
 108                         struct tp_probes, probes[0]);
 109
 110                 /*
 111                  * We can't free probes if SRCU is not initialized yet.
 112                  * Postpone the freeing till after SRCU is initialized.
 113                  */
 114                 if (unlikely(!ok_to_free_tracepoints)) {
 115                         tp_probes->rcu.next = early_probes;
 116                         early_probes = &tp_probes->rcu;
 117                         return;
 118                 }
 119
 120                 /*
 121                  * Tracepoint probes are protected by both sched RCU and SRCU,
 122                  * by calling the SRCU callback in the sched RCU callback we
 123                  * cover both cases. So let us chain the SRCU and sched RCU
 124                  * callbacks to wait for both grace periods.
 125                  */
 126                 call_rcu_sched(&tp_probes->rcu, rcu_free_old_probes);
 127         }
 128 }
 129
 130 static void debug_print_probes(struct tracepoint_func *funcs)
 131 {
 132         int i;
 133
 134         if (!tracepoint_debug || !funcs)
 135                 return;
 136
 137         for (i = 0; funcs[i].func; i++)
 138                 printk(KERN_DEBUG "Probe %d : %p\n", i, funcs[i].func);
 139 }
 140
 141 static struct tracepoint_func *
 142 func_add(struct tracepoint_func **funcs, struct tracepoint_func *tp_func,
 143          int prio)
 144 {
 145         struct tracepoint_func *old, *new;
 146         int nr_probes = 0;
 147         int pos = -1;
 148
 149         if (WARN_ON(!tp_func->func))
 150                 return ERR_PTR(-EINVAL);
 151
 152         debug_print_probes(*funcs);
 153         old = *funcs;
 154         if (old) {
 155                 /* (N -> N+1), (N != 0, 1) probes */
 156                 for (nr_probes = 0; old[nr_probes].func; nr_probes++) {
 157                         /* Insert before probes of lower priority */
 158                         if (pos < 0 && old[nr_probes].prio < prio)
 159                                 pos = nr_probes;
 160                         if (old[nr_probes].func == tp_func->func &&
 161                             old[nr_probes].data == tp_func->data)
 162                                 return ERR_PTR(-EEXIST);
 163                 }
 164         }
 165         /* + 2 : one for new probe, one for NULL func */
 166         new = allocate_probes(nr_probes + 2);
 167         if (new == NULL)
 168                 return ERR_PTR(-ENOMEM);
 169         if (old) {
 170                 if (pos < 0) {
 171                         pos = nr_probes;
 172                         memcpy(new, old, nr_probes * sizeof(struct tracepoint_func));
 173                 } else {
 174                         /* Copy higher priority probes ahead of the new probe */
 175                         memcpy(new, old, pos * sizeof(struct tracepoint_func));
 176                         /* Copy the rest after it. */
 177                         memcpy(new + pos + 1, old + pos,
 178                                (nr_probes - pos) * sizeof(struct tracepoint_func));
 179                 }
 180         } else
 181                 pos = 0;
 182         new[pos] = *tp_func;
 183         new[nr_probes + 1].func = NULL;
 184         *funcs = new;
 185         debug_print_probes(*funcs);
 186         return old;
 187 }
 188
 189 static void *func_remove(struct tracepoint_func **funcs,
 190                 struct tracepoint_func *tp_func)
 191 {
 192         int nr_probes = 0, nr_del = 0, i;
 193         struct tracepoint_func *old, *new;
 194
 195         old = *funcs;
 196
 197         if (!old)
 198                 return ERR_PTR(-ENOENT);
 199
 200         debug_print_probes(*funcs);
 201         /* (N -> M), (N > 1, M >= 0) probes */
 202         if (tp_func->func) {
 203                 for (nr_probes = 0; old[nr_probes].func; nr_probes++) {
 204                         if (old[nr_probes].func == tp_func->func &&
 205                              old[nr_probes].data == tp_func->data)
 206                                 nr_del++;
 207                 }
 208         }
 209
 210         /*
 211          * If probe is NULL, then nr_probes = nr_del = 0, and then the
 212          * entire entry will be removed.
 213          */
 214         if (nr_probes - nr_del == 0) {
 215                 /* N -> 0, (N > 1) */
 216                 *funcs = NULL;
 217                 debug_print_probes(*funcs);
 218                 return old;
 219         } else {
 220                 int j = 0;
 221                 /* N -> M, (N > 1, M > 0) */
 222                 /* + 1 for NULL */
 223                 new = allocate_probes(nr_probes - nr_del + 1);
 224                 if (new == NULL)
 225                         return ERR_PTR(-ENOMEM);
 226                 for (i = 0; old[i].func; i++)
 227                         if (old[i].func != tp_func->func
 228                                         || old[i].data != tp_func->data)
 229                                 new[j++] = old[i];
 230                 new[nr_probes - nr_del].func = NULL;
 231                 *funcs = new;
 232         }
 233         debug_print_probes(*funcs);
 234         return old;
 235 }
 236
 237 /*
 238  * Add the probe function to a tracepoint.
 239  */
 240 static int tracepoint_add_func(struct tracepoint *tp,
 241                                struct tracepoint_func *func, int prio)
 242 {
 243         struct tracepoint_func *old, *tp_funcs;
 244         int ret;
 245
 246         if (tp->regfunc && !static_key_enabled(&tp->key)) {
 247                 ret = tp->regfunc();
 248                 if (ret < 0)
 249                         return ret;
 250         }
 251
 252         tp_funcs = rcu_dereference_protected(tp->funcs,
 253                         lockdep_is_held(&tracepoints_mutex));
 254         old = func_add(&tp_funcs, func, prio);
 255         if (IS_ERR(old)) {
 256                 WARN_ON_ONCE(PTR_ERR(old) != -ENOMEM);
 257                 return PTR_ERR(old);
 258         }
 259
 260         /*
 261          * rcu_assign_pointer has as smp_store_release() which makes sure
 262          * that the new probe callbacks array is consistent before setting
 263          * a pointer to it.  This array is referenced by __DO_TRACE from
 264          * include/linux/tracepoint.h using rcu_dereference_sched().
 265          */
 266         rcu_assign_pointer(tp->funcs, tp_funcs);
 267         if (!static_key_enabled(&tp->key))
 268                 static_key_slow_inc(&tp->key);
 269         release_probes(old);
 270         return 0;
 271 }
 272
 273 /*
 274  * Remove a probe function from a tracepoint.
 275  * Note: only waiting an RCU period after setting elem->call to the empty
 276  * function insures that the original callback is not used anymore. This insured
 277  * by preempt_disable around the call site.
 278  */
 279 static int tracepoint_remove_func(struct tracepoint *tp,
 280                 struct tracepoint_func *func)
 281 {
 282         struct tracepoint_func *old, *tp_funcs;
 283
 284         tp_funcs = rcu_dereference_protected(tp->funcs,
 285                         lockdep_is_held(&tracepoints_mutex));
 286         old = func_remove(&tp_funcs, func);
 287         if (IS_ERR(old)) {
 288                 WARN_ON_ONCE(PTR_ERR(old) != -ENOMEM);
 289                 return PTR_ERR(old);
 290         }
 291
 292         if (!tp_funcs) {
 293                 /* Removed last function */
 294                 if (tp->unregfunc && static_key_enabled(&tp->key))
 295                         tp->unregfunc();
 296
 297                 if (static_key_enabled(&tp->key))
 298                         static_key_slow_dec(&tp->key);
 299         }
 300         rcu_assign_pointer(tp->funcs, tp_funcs);
 301         release_probes(old);
 302         return 0;
 303 }
 304
 305 /**
 306  * tracepoint_probe_register_prio -  Connect a probe to a tracepoint with priority
 307  * @tp: tracepoint
 308  * @probe: probe handler
 309  * @data: tracepoint data
 310  * @prio: priority of this function over other registered functions
 311  *
 312  * Returns 0 if ok, error value on error.
 313  * Note: if @tp is within a module, the caller is responsible for
 314  * unregistering the probe before the module is gone. This can be
 315  * performed either with a tracepoint module going notifier, or from
 316  * within module exit functions.
 317  */
 318 int tracepoint_probe_register_prio(struct tracepoint *tp, void *probe,
 319                                    void *data, int prio)
 320 {
 321         struct tracepoint_func tp_func;
 322         int ret;
 323
 324         mutex_lock(&tracepoints_mutex);
 325         tp_func.func = probe;
 326         tp_func.data = data;
 327         tp_func.prio = prio;
 328         ret = tracepoint_add_func(tp, &tp_func, prio);
 329         mutex_unlock(&tracepoints_mutex);
 330         return ret;
 331 }
 332 EXPORT_SYMBOL_GPL(tracepoint_probe_register_prio);
 333
 334 /**
 335  * tracepoint_probe_register -  Connect a probe to a tracepoint
 336  * @tp: tracepoint
 337  * @probe: probe handler
 338  * @data: tracepoint data
 339  *
 340  * Returns 0 if ok, error value on error.
 341  * Note: if @tp is within a module, the caller is responsible for
 342  * unregistering the probe before the module is gone. This can be
 343  * performed either with a tracepoint module going notifier, or from
 344  * within module exit functions.
 345  */
 346 int tracepoint_probe_register(struct tracepoint *tp, void *probe, void *data)
 347 {
 348         return tracepoint_probe_register_prio(tp, probe, data, TRACEPOINT_DEFAULT_PRIO);
 349 }
 350 EXPORT_SYMBOL_GPL(tracepoint_probe_register);
 351
 352 /**
 353  * tracepoint_probe_unregister -  Disconnect a probe from a tracepoint
 354  * @tp: tracepoint
 355  * @probe: probe function pointer
 356  * @data: tracepoint data
 357  *
 358  * Returns 0 if ok, error value on error.
 359  */
 360 int tracepoint_probe_unregister(struct tracepoint *tp, void *probe, void *data)
 361 {
 362         struct tracepoint_func tp_func;
 363         int ret;
 364
 365         mutex_lock(&tracepoints_mutex);
 366         tp_func.func = probe;
 367         tp_func.data = data;
 368         ret = tracepoint_remove_func(tp, &tp_func);
 369         mutex_unlock(&tracepoints_mutex);
 370         return ret;
 371 }
 372 EXPORT_SYMBOL_GPL(tracepoint_probe_unregister);
 373
 374 static void for_each_tracepoint_range(struct tracepoint * const *begin,
 375                 struct tracepoint * const *end,
 376                 void (*fct)(struct tracepoint *tp, void *priv),
 377                 void *priv)
 378 {
 379         if (!begin)
 380                 return;
 381
 382         if (IS_ENABLED(CONFIG_HAVE_ARCH_PREL32_RELOCATIONS)) {
 383                 const int *iter;
 384
 385                 for (iter = (const int *)begin; iter < (const int *)end; iter++)
 386                         fct(offset_to_ptr(iter), priv);
 387         } else {
 388                 struct tracepoint * const *iter;
 389
 390                 for (iter = begin; iter < end; iter++)
 391                         fct(*iter, priv);
 392         }
 393 }
 394
 395 #ifdef CONFIG_MODULES
 396 bool trace_module_has_bad_taint(struct module *mod)
 397 {
 398         return mod->taints & ~((1 << TAINT_OOT_MODULE) | (1 << TAINT_CRAP) |
 399                                (1 << TAINT_UNSIGNED_MODULE));
 400 }
 401
 402 static BLOCKING_NOTIFIER_HEAD(tracepoint_notify_list);
 403
 404 /**
 405  * register_tracepoint_notifier - register tracepoint coming/going notifier
 406  * @nb: notifier block
 407  *
 408  * Notifiers registered with this function are called on module
 409  * coming/going with the tracepoint_module_list_mutex held.
 410  * The notifier block callback should expect a "struct tp_module" data
 411  * pointer.
 412  */
 413 int register_tracepoint_module_notifier(struct notifier_block *nb)
 414 {
 415         struct tp_module *tp_mod;
 416         int ret;
 417
 418         mutex_lock(&tracepoint_module_list_mutex);
 419         ret = blocking_notifier_chain_register(&tracepoint_notify_list, nb);
 420         if (ret)
 421                 goto end;
 422         list_for_each_entry(tp_mod, &tracepoint_module_list, list)
 423                 (void) nb->notifier_call(nb, MODULE_STATE_COMING, tp_mod);
 424 end:
 425         mutex_unlock(&tracepoint_module_list_mutex);
 426         return ret;
 427 }
 428 EXPORT_SYMBOL_GPL(register_tracepoint_module_notifier);
 429
 430 /**
 431  * unregister_tracepoint_notifier - unregister tracepoint coming/going notifier
 432  * @nb: notifier block
 433  *
 434  * The notifier block callback should expect a "struct tp_module" data
 435  * pointer.
 436  */
 437 int unregister_tracepoint_module_notifier(struct notifier_block *nb)
 438 {
 439         struct tp_module *tp_mod;
 440         int ret;
 441
 442         mutex_lock(&tracepoint_module_list_mutex);
 443         ret = blocking_notifier_chain_unregister(&tracepoint_notify_list, nb);
 444         if (ret)
 445                 goto end;
 446         list_for_each_entry(tp_mod, &tracepoint_module_list, list)
 447                 (void) nb->notifier_call(nb, MODULE_STATE_GOING, tp_mod);
 448 end:
 449         mutex_unlock(&tracepoint_module_list_mutex);
 450         return ret;
 451
 452 }
 453 EXPORT_SYMBOL_GPL(unregister_tracepoint_module_notifier);
 454
 455 /*
 456  * Ensure the tracer unregistered the module's probes before the module
 457  * teardown is performed. Prevents leaks of probe and data pointers.
 458  */
 459 static void tp_module_going_check_quiescent(struct tracepoint *tp, void *priv)
 460 {
 461         WARN_ON_ONCE(tp->funcs);
 462 }
 463
 464 static int tracepoint_module_coming(struct module *mod)
 465 {
 466         struct tp_module *tp_mod;
 467         int ret = 0;
 468
 469         if (!mod->num_tracepoints)
 470                 return 0;
 471
 472         /*
 473          * We skip modules that taint the kernel, especially those with different
 474          * module headers (for forced load), to make sure we don't cause a crash.
 475          * Staging, out-of-tree, and unsigned GPL modules are fine.
 476          */
 477         if (trace_module_has_bad_taint(mod))
 478                 return 0;
 479         mutex_lock(&tracepoint_module_list_mutex);
 480         tp_mod = kmalloc(sizeof(struct tp_module), GFP_KERNEL);
 481         if (!tp_mod) {
 482                 ret = -ENOMEM;
 483                 goto end;
 484         }
 485         tp_mod->mod = mod;
 486         list_add_tail(&tp_mod->list, &tracepoint_module_list);
 487         blocking_notifier_call_chain(&tracepoint_notify_list,
 488                         MODULE_STATE_COMING, tp_mod);
 489 end:
 490         mutex_unlock(&tracepoint_module_list_mutex);
 491         return ret;
 492 }
 493
 494 static void tracepoint_module_going(struct module *mod)
 495 {
 496         struct tp_module *tp_mod;
 497
 498         if (!mod->num_tracepoints)
 499                 return;
 500
 501         mutex_lock(&tracepoint_module_list_mutex);
 502         list_for_each_entry(tp_mod, &tracepoint_module_list, list) {
 503                 if (tp_mod->mod == mod) {
 504                         blocking_notifier_call_chain(&tracepoint_notify_list,
 505                                         MODULE_STATE_GOING, tp_mod);
 506                         list_del(&tp_mod->list);
 507                         kfree(tp_mod);
 508                         /*
 509                          * Called the going notifier before checking for
 510                          * quiescence.
 511                          */
 512                         for_each_tracepoint_range(mod->tracepoints_ptrs,
 513                                 mod->tracepoints_ptrs + mod->num_tracepoints,
 514                                 tp_module_going_check_quiescent, NULL);
 515                         break;
 516                 }
 517         }
 518         /*
 519          * In the case of modules that were tainted at "coming", we'll simply
 520          * walk through the list without finding it. We cannot use the "tainted"
 521          * flag on "going", in case a module taints the kernel only after being
 522          * loaded.
 523          */
 524         mutex_unlock(&tracepoint_module_list_mutex);
 525 }
 526
 527 static int tracepoint_module_notify(struct notifier_block *self,
 528                 unsigned long val, void *data)
 529 {
 530         struct module *mod = data;
 531         int ret = 0;
 532
 533         switch (val) {
 534         case MODULE_STATE_COMING:
 535                 ret = tracepoint_module_coming(mod);
 536                 break;
 537         case MODULE_STATE_LIVE:
 538                 break;
 539         case MODULE_STATE_GOING:
 540                 tracepoint_module_going(mod);
 541                 break;
 542         case MODULE_STATE_UNFORMED:
 543                 break;
 544         }
 545         return ret;
 546 }
 547
 548 static struct notifier_block tracepoint_module_nb = {
 549         .notifier_call = tracepoint_module_notify,
 550         .priority = 0,
 551 };
 552
 553 static __init int init_tracepoints(void)
 554 {
 555         int ret;
 556
 557         ret = register_module_notifier(&tracepoint_module_nb);
 558         if (ret)
 559                 pr_warn("Failed to register tracepoint module enter notifier\n");
 560
 561         return ret;
 562 }
 563 __initcall(init_tracepoints);
 564 #endif /* CONFIG_MODULES */
 565
 566 /**
 567  * for_each_kernel_tracepoint - iteration on all kernel tracepoints
 568  * @fct: callback
 569  * @priv: private data
 570  */
 571 void for_each_kernel_tracepoint(void (*fct)(struct tracepoint *tp, void *priv),
 572                 void *priv)
 573 {
 574         for_each_tracepoint_range(__start___tracepoints_ptrs,
 575                 __stop___tracepoints_ptrs, fct, priv);
 576 }
 577 EXPORT_SYMBOL_GPL(for_each_kernel_tracepoint);
 578
 579 #ifdef CONFIG_HAVE_SYSCALL_TRACEPOINTS
 580
 581 /* NB: reg/unreg are called while guarded with the tracepoints_mutex */
 582 static int sys_tracepoint_refcount;
 583
 584 int syscall_regfunc(void)
 585 {
 586         struct task_struct *p, *t;
 587
 588         if (!sys_tracepoint_refcount) {
 589                 read_lock(&tasklist_lock);
 590                 for_each_process_thread(p, t) {
 591                         set_tsk_thread_flag(t, TIF_SYSCALL_TRACEPOINT);
 592                 }
 593                 read_unlock(&tasklist_lock);
 594         }
 595         sys_tracepoint_refcount++;
 596
 597         return 0;
 598 }
 599
 600 void syscall_unregfunc(void)
 601 {
 602         struct task_struct *p, *t;
 603
 604         sys_tracepoint_refcount--;
 605         if (!sys_tracepoint_refcount) {
 606                 read_lock(&tasklist_lock);
 607                 for_each_process_thread(p, t) {
 608                         clear_tsk_thread_flag(t, TIF_SYSCALL_TRACEPOINT);
 609                 }
 610                 read_unlock(&tasklist_lock);
 611         }
 612 }
 613 #endif