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