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 a smp_wmb() which makes sure that the new
 216          * probe callbacks array is consistent before setting a pointer to it.
 217          * This array is referenced by __DO_TRACE from
 218          * include/linux/tracepoints.h. A matching smp_read_barrier_depends()
 219          * is used.
 220          */
 221         rcu_assign_pointer(tp->funcs, tp_funcs);
 222         if (!static_key_enabled(&tp->key))
 223                 static_key_slow_inc(&tp->key);
 224         release_probes(old);
 225         return 0;
 226 }
 227
 228 /*
 229  * Remove a probe function from a tracepoint.
 230  * Note: only waiting an RCU period after setting elem->call to the empty
 231  * function insures that the original callback is not used anymore. This insured
 232  * by preempt_disable around the call site.
 233  */
 234 static int tracepoint_remove_func(struct tracepoint *tp,
 235                 struct tracepoint_func *func)
 236 {
 237         struct tracepoint_func *old, *tp_funcs;
 238
 239         tp_funcs = rcu_dereference_protected(tp->funcs,
 240                         lockdep_is_held(&tracepoints_mutex));
 241         old = func_remove(&tp_funcs, func);
 242         if (IS_ERR(old)) {
 243                 WARN_ON_ONCE(PTR_ERR(old) != -ENOMEM);
 244                 return PTR_ERR(old);
 245         }
 246
 247         if (!tp_funcs) {
 248                 /* Removed last function */
 249                 if (tp->unregfunc && static_key_enabled(&tp->key))
 250                         tp->unregfunc();
 251
 252                 if (static_key_enabled(&tp->key))
 253                         static_key_slow_dec(&tp->key);
 254         }
 255         rcu_assign_pointer(tp->funcs, tp_funcs);
 256         release_probes(old);
 257         return 0;
 258 }
 259
 260 /**
 261  * tracepoint_probe_register -  Connect a probe to a tracepoint
 262  * @tp: tracepoint
 263  * @probe: probe handler
 264  * @data: tracepoint data
 265  * @prio: priority of this function over other registered functions
 266  *
 267  * Returns 0 if ok, error value on error.
 268  * Note: if @tp is within a module, the caller is responsible for
 269  * unregistering the probe before the module is gone. This can be
 270  * performed either with a tracepoint module going notifier, or from
 271  * within module exit functions.
 272  */
 273 int tracepoint_probe_register_prio(struct tracepoint *tp, void *probe,
 274                                    void *data, int prio)
 275 {
 276         struct tracepoint_func tp_func;
 277         int ret;
 278
 279         mutex_lock(&tracepoints_mutex);
 280         tp_func.func = probe;
 281         tp_func.data = data;
 282         tp_func.prio = prio;
 283         ret = tracepoint_add_func(tp, &tp_func, prio);
 284         mutex_unlock(&tracepoints_mutex);
 285         return ret;
 286 }
 287 EXPORT_SYMBOL_GPL(tracepoint_probe_register_prio);
 288
 289 /**
 290  * tracepoint_probe_register -  Connect a probe to a tracepoint
 291  * @tp: tracepoint
 292  * @probe: probe handler
 293  * @data: tracepoint data
 294  * @prio: priority of this function over other registered functions
 295  *
 296  * Returns 0 if ok, error value on error.
 297  * Note: if @tp is within a module, the caller is responsible for
 298  * unregistering the probe before the module is gone. This can be
 299  * performed either with a tracepoint module going notifier, or from
 300  * within module exit functions.
 301  */
 302 int tracepoint_probe_register(struct tracepoint *tp, void *probe, void *data)
 303 {
 304         return tracepoint_probe_register_prio(tp, probe, data, TRACEPOINT_DEFAULT_PRIO);
 305 }
 306 EXPORT_SYMBOL_GPL(tracepoint_probe_register);
 307
 308 /**
 309  * tracepoint_probe_unregister -  Disconnect a probe from a tracepoint
 310  * @tp: tracepoint
 311  * @probe: probe function pointer
 312  * @data: tracepoint data
 313  *
 314  * Returns 0 if ok, error value on error.
 315  */
 316 int tracepoint_probe_unregister(struct tracepoint *tp, void *probe, void *data)
 317 {
 318         struct tracepoint_func tp_func;
 319         int ret;
 320
 321         mutex_lock(&tracepoints_mutex);
 322         tp_func.func = probe;
 323         tp_func.data = data;
 324         ret = tracepoint_remove_func(tp, &tp_func);
 325         mutex_unlock(&tracepoints_mutex);
 326         return ret;
 327 }
 328 EXPORT_SYMBOL_GPL(tracepoint_probe_unregister);
 329
 330 #ifdef CONFIG_MODULES
 331 bool trace_module_has_bad_taint(struct module *mod)
 332 {
 333         return mod->taints & ~((1 << TAINT_OOT_MODULE) | (1 << TAINT_CRAP) |
 334                                (1 << TAINT_UNSIGNED_MODULE));
 335 }
 336
 337 static BLOCKING_NOTIFIER_HEAD(tracepoint_notify_list);
 338
 339 /**
 340  * register_tracepoint_notifier - register tracepoint coming/going notifier
 341  * @nb: notifier block
 342  *
 343  * Notifiers registered with this function are called on module
 344  * coming/going with the tracepoint_module_list_mutex held.
 345  * The notifier block callback should expect a "struct tp_module" data
 346  * pointer.
 347  */
 348 int register_tracepoint_module_notifier(struct notifier_block *nb)
 349 {
 350         struct tp_module *tp_mod;
 351         int ret;
 352
 353         mutex_lock(&tracepoint_module_list_mutex);
 354         ret = blocking_notifier_chain_register(&tracepoint_notify_list, nb);
 355         if (ret)
 356                 goto end;
 357         list_for_each_entry(tp_mod, &tracepoint_module_list, list)
 358                 (void) nb->notifier_call(nb, MODULE_STATE_COMING, tp_mod);
 359 end:
 360         mutex_unlock(&tracepoint_module_list_mutex);
 361         return ret;
 362 }
 363 EXPORT_SYMBOL_GPL(register_tracepoint_module_notifier);
 364
 365 /**
 366  * unregister_tracepoint_notifier - unregister tracepoint coming/going notifier
 367  * @nb: notifier block
 368  *
 369  * The notifier block callback should expect a "struct tp_module" data
 370  * pointer.
 371  */
 372 int unregister_tracepoint_module_notifier(struct notifier_block *nb)
 373 {
 374         struct tp_module *tp_mod;
 375         int ret;
 376
 377         mutex_lock(&tracepoint_module_list_mutex);
 378         ret = blocking_notifier_chain_unregister(&tracepoint_notify_list, nb);
 379         if (ret)
 380                 goto end;
 381         list_for_each_entry(tp_mod, &tracepoint_module_list, list)
 382                 (void) nb->notifier_call(nb, MODULE_STATE_GOING, tp_mod);
 383 end:
 384         mutex_unlock(&tracepoint_module_list_mutex);
 385         return ret;
 386
 387 }
 388 EXPORT_SYMBOL_GPL(unregister_tracepoint_module_notifier);
 389
 390 /*
 391  * Ensure the tracer unregistered the module's probes before the module
 392  * teardown is performed. Prevents leaks of probe and data pointers.
 393  */
 394 static void tp_module_going_check_quiescent(struct tracepoint * const *begin,
 395                 struct tracepoint * const *end)
 396 {
 397         struct tracepoint * const *iter;
 398
 399         if (!begin)
 400                 return;
 401         for (iter = begin; iter < end; iter++)
 402                 WARN_ON_ONCE((*iter)->funcs);
 403 }
 404
 405 static int tracepoint_module_coming(struct module *mod)
 406 {
 407         struct tp_module *tp_mod;
 408         int ret = 0;
 409
 410         if (!mod->num_tracepoints)
 411                 return 0;
 412
 413         /*
 414          * We skip modules that taint the kernel, especially those with different
 415          * module headers (for forced load), to make sure we don't cause a crash.
 416          * Staging, out-of-tree, and unsigned GPL modules are fine.
 417          */
 418         if (trace_module_has_bad_taint(mod))
 419                 return 0;
 420         mutex_lock(&tracepoint_module_list_mutex);
 421         tp_mod = kmalloc(sizeof(struct tp_module), GFP_KERNEL);
 422         if (!tp_mod) {
 423                 ret = -ENOMEM;
 424                 goto end;
 425         }
 426         tp_mod->mod = mod;
 427         list_add_tail(&tp_mod->list, &tracepoint_module_list);
 428         blocking_notifier_call_chain(&tracepoint_notify_list,
 429                         MODULE_STATE_COMING, tp_mod);
 430 end:
 431         mutex_unlock(&tracepoint_module_list_mutex);
 432         return ret;
 433 }
 434
 435 static void tracepoint_module_going(struct module *mod)
 436 {
 437         struct tp_module *tp_mod;
 438
 439         if (!mod->num_tracepoints)
 440                 return;
 441
 442         mutex_lock(&tracepoint_module_list_mutex);
 443         list_for_each_entry(tp_mod, &tracepoint_module_list, list) {
 444                 if (tp_mod->mod == mod) {
 445                         blocking_notifier_call_chain(&tracepoint_notify_list,
 446                                         MODULE_STATE_GOING, tp_mod);
 447                         list_del(&tp_mod->list);
 448                         kfree(tp_mod);
 449                         /*
 450                          * Called the going notifier before checking for
 451                          * quiescence.
 452                          */
 453                         tp_module_going_check_quiescent(mod->tracepoints_ptrs,
 454                                 mod->tracepoints_ptrs + mod->num_tracepoints);
 455                         break;
 456                 }
 457         }
 458         /*
 459          * In the case of modules that were tainted at "coming", we'll simply
 460          * walk through the list without finding it. We cannot use the "tainted"
 461          * flag on "going", in case a module taints the kernel only after being
 462          * loaded.
 463          */
 464         mutex_unlock(&tracepoint_module_list_mutex);
 465 }
 466
 467 static int tracepoint_module_notify(struct notifier_block *self,
 468                 unsigned long val, void *data)
 469 {
 470         struct module *mod = data;
 471         int ret = 0;
 472
 473         switch (val) {
 474         case MODULE_STATE_COMING:
 475                 ret = tracepoint_module_coming(mod);
 476                 break;
 477         case MODULE_STATE_LIVE:
 478                 break;
 479         case MODULE_STATE_GOING:
 480                 tracepoint_module_going(mod);
 481                 break;
 482         case MODULE_STATE_UNFORMED:
 483                 break;
 484         }
 485         return ret;
 486 }
 487
 488 static struct notifier_block tracepoint_module_nb = {
 489         .notifier_call = tracepoint_module_notify,
 490         .priority = 0,
 491 };
 492
 493 static __init int init_tracepoints(void)
 494 {
 495         int ret;
 496
 497         ret = register_module_notifier(&tracepoint_module_nb);
 498         if (ret)
 499                 pr_warn("Failed to register tracepoint module enter notifier\n");
 500
 501         return ret;
 502 }
 503 __initcall(init_tracepoints);
 504 #endif /* CONFIG_MODULES */
 505
 506 static void for_each_tracepoint_range(struct tracepoint * const *begin,
 507                 struct tracepoint * const *end,
 508                 void (*fct)(struct tracepoint *tp, void *priv),
 509                 void *priv)
 510 {
 511         struct tracepoint * const *iter;
 512
 513         if (!begin)
 514                 return;
 515         for (iter = begin; iter < end; iter++)
 516                 fct(*iter, priv);
 517 }
 518
 519 /**
 520  * for_each_kernel_tracepoint - iteration on all kernel tracepoints
 521  * @fct: callback
 522  * @priv: private data
 523  */
 524 void for_each_kernel_tracepoint(void (*fct)(struct tracepoint *tp, void *priv),
 525                 void *priv)
 526 {
 527         for_each_tracepoint_range(__start___tracepoints_ptrs,
 528                 __stop___tracepoints_ptrs, fct, priv);
 529 }
 530 EXPORT_SYMBOL_GPL(for_each_kernel_tracepoint);
 531
 532 #ifdef CONFIG_HAVE_SYSCALL_TRACEPOINTS
 533
 534 /* NB: reg/unreg are called while guarded with the tracepoints_mutex */
 535 static int sys_tracepoint_refcount;
 536
 537 int syscall_regfunc(void)
 538 {
 539         struct task_struct *p, *t;
 540
 541         if (!sys_tracepoint_refcount) {
 542                 read_lock(&tasklist_lock);
 543                 for_each_process_thread(p, t) {
 544                         set_tsk_thread_flag(t, TIF_SYSCALL_TRACEPOINT);
 545                 }
 546                 read_unlock(&tasklist_lock);
 547         }
 548         sys_tracepoint_refcount++;
 549
 550         return 0;
 551 }
 552
 553 void syscall_unregfunc(void)
 554 {
 555         struct task_struct *p, *t;
 556
 557         sys_tracepoint_refcount--;
 558         if (!sys_tracepoint_refcount) {
 559                 read_lock(&tasklist_lock);
 560                 for_each_process_thread(p, t) {
 561                         clear_tsk_thread_flag(t, TIF_SYSCALL_TRACEPOINT);
 562                 }
 563                 read_unlock(&tasklist_lock);
 564         }
 565 }
 566 #endif