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