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