Loading...
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
30extern struct tracepoint * const __start___tracepoints_ptrs[];
31extern struct tracepoint * const __stop___tracepoints_ptrs[];
32
33/* Set to 1 to enable tracepoint debug output */
34static const int tracepoint_debug;
35
36#ifdef CONFIG_MODULES
37/*
38 * Tracepoint module list mutex protects the local module list.
39 */
40static DEFINE_MUTEX(tracepoint_module_list_mutex);
41
42/* Local list of struct tp_module */
43static 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 */
50static 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 */
57struct tp_probes {
58 struct rcu_head rcu;
59 struct tracepoint_func probes[0];
60};
61
62static 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
69static void rcu_free_old_probes(struct rcu_head *head)
70{
71 kfree(container_of(head, struct tp_probes, rcu));
72}
73
74static 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
83static 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
94static 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
125static 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 */
176static 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 */
212static 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 */
249int 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}
261EXPORT_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 */
270int 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}
282EXPORT_SYMBOL_GPL(tracepoint_probe_unregister);
283
284#ifdef CONFIG_MODULES
285bool 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
291static 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 */
302int 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);
313end:
314 mutex_unlock(&tracepoint_module_list_mutex);
315 return ret;
316}
317EXPORT_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 */
326int 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);
337end:
338 mutex_unlock(&tracepoint_module_list_mutex);
339 return ret;
340
341}
342EXPORT_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 */
348static 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
359static 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);
384end:
385 mutex_unlock(&tracepoint_module_list_mutex);
386 return ret;
387}
388
389static 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
421static 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
442static struct notifier_block tracepoint_module_nb = {
443 .notifier_call = tracepoint_module_notify,
444 .priority = 0,
445};
446
447static __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
460static 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 */
478void 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}
484EXPORT_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 */
489static int sys_tracepoint_refcount;
490
491void 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
508void 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
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
31extern struct tracepoint * const __start___tracepoints_ptrs[];
32extern struct tracepoint * const __stop___tracepoints_ptrs[];
33
34/* Set to 1 to enable tracepoint debug output */
35static const int tracepoint_debug;
36
37#ifdef CONFIG_MODULES
38/*
39 * Tracepoint module list mutex protects the local module list.
40 */
41static DEFINE_MUTEX(tracepoint_module_list_mutex);
42
43/* Local list of struct tp_module */
44static 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 */
51static 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 */
58struct tp_probes {
59 struct rcu_head rcu;
60 struct tracepoint_func probes[0];
61};
62
63static 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
70static void rcu_free_old_probes(struct rcu_head *head)
71{
72 kfree(container_of(head, struct tp_probes, rcu));
73}
74
75static 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
84static 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
95static struct tracepoint_func *
96func_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
143static 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 */
194static 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 */
233static 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 - Connect a probe to a tracepoint
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 */
272int 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}
286EXPORT_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 * @prio: priority of this function over other registered functions
294 *
295 * Returns 0 if ok, error value on error.
296 * Note: if @tp is within a module, the caller is responsible for
297 * unregistering the probe before the module is gone. This can be
298 * performed either with a tracepoint module going notifier, or from
299 * within module exit functions.
300 */
301int tracepoint_probe_register(struct tracepoint *tp, void *probe, void *data)
302{
303 return tracepoint_probe_register_prio(tp, probe, data, TRACEPOINT_DEFAULT_PRIO);
304}
305EXPORT_SYMBOL_GPL(tracepoint_probe_register);
306
307/**
308 * tracepoint_probe_unregister - Disconnect a probe from a tracepoint
309 * @tp: tracepoint
310 * @probe: probe function pointer
311 * @data: tracepoint data
312 *
313 * Returns 0 if ok, error value on error.
314 */
315int tracepoint_probe_unregister(struct tracepoint *tp, void *probe, void *data)
316{
317 struct tracepoint_func tp_func;
318 int ret;
319
320 mutex_lock(&tracepoints_mutex);
321 tp_func.func = probe;
322 tp_func.data = data;
323 ret = tracepoint_remove_func(tp, &tp_func);
324 mutex_unlock(&tracepoints_mutex);
325 return ret;
326}
327EXPORT_SYMBOL_GPL(tracepoint_probe_unregister);
328
329#ifdef CONFIG_MODULES
330bool trace_module_has_bad_taint(struct module *mod)
331{
332 return mod->taints & ~((1 << TAINT_OOT_MODULE) | (1 << TAINT_CRAP) |
333 (1 << TAINT_UNSIGNED_MODULE));
334}
335
336static BLOCKING_NOTIFIER_HEAD(tracepoint_notify_list);
337
338/**
339 * register_tracepoint_notifier - register tracepoint coming/going notifier
340 * @nb: notifier block
341 *
342 * Notifiers registered with this function are called on module
343 * coming/going with the tracepoint_module_list_mutex held.
344 * The notifier block callback should expect a "struct tp_module" data
345 * pointer.
346 */
347int register_tracepoint_module_notifier(struct notifier_block *nb)
348{
349 struct tp_module *tp_mod;
350 int ret;
351
352 mutex_lock(&tracepoint_module_list_mutex);
353 ret = blocking_notifier_chain_register(&tracepoint_notify_list, nb);
354 if (ret)
355 goto end;
356 list_for_each_entry(tp_mod, &tracepoint_module_list, list)
357 (void) nb->notifier_call(nb, MODULE_STATE_COMING, tp_mod);
358end:
359 mutex_unlock(&tracepoint_module_list_mutex);
360 return ret;
361}
362EXPORT_SYMBOL_GPL(register_tracepoint_module_notifier);
363
364/**
365 * unregister_tracepoint_notifier - unregister tracepoint coming/going notifier
366 * @nb: notifier block
367 *
368 * The notifier block callback should expect a "struct tp_module" data
369 * pointer.
370 */
371int unregister_tracepoint_module_notifier(struct notifier_block *nb)
372{
373 struct tp_module *tp_mod;
374 int ret;
375
376 mutex_lock(&tracepoint_module_list_mutex);
377 ret = blocking_notifier_chain_unregister(&tracepoint_notify_list, nb);
378 if (ret)
379 goto end;
380 list_for_each_entry(tp_mod, &tracepoint_module_list, list)
381 (void) nb->notifier_call(nb, MODULE_STATE_GOING, tp_mod);
382end:
383 mutex_unlock(&tracepoint_module_list_mutex);
384 return ret;
385
386}
387EXPORT_SYMBOL_GPL(unregister_tracepoint_module_notifier);
388
389/*
390 * Ensure the tracer unregistered the module's probes before the module
391 * teardown is performed. Prevents leaks of probe and data pointers.
392 */
393static void tp_module_going_check_quiescent(struct tracepoint * const *begin,
394 struct tracepoint * const *end)
395{
396 struct tracepoint * const *iter;
397
398 if (!begin)
399 return;
400 for (iter = begin; iter < end; iter++)
401 WARN_ON_ONCE((*iter)->funcs);
402}
403
404static int tracepoint_module_coming(struct module *mod)
405{
406 struct tp_module *tp_mod;
407 int ret = 0;
408
409 if (!mod->num_tracepoints)
410 return 0;
411
412 /*
413 * We skip modules that taint the kernel, especially those with different
414 * module headers (for forced load), to make sure we don't cause a crash.
415 * Staging, out-of-tree, and unsigned GPL modules are fine.
416 */
417 if (trace_module_has_bad_taint(mod))
418 return 0;
419 mutex_lock(&tracepoint_module_list_mutex);
420 tp_mod = kmalloc(sizeof(struct tp_module), GFP_KERNEL);
421 if (!tp_mod) {
422 ret = -ENOMEM;
423 goto end;
424 }
425 tp_mod->mod = mod;
426 list_add_tail(&tp_mod->list, &tracepoint_module_list);
427 blocking_notifier_call_chain(&tracepoint_notify_list,
428 MODULE_STATE_COMING, tp_mod);
429end:
430 mutex_unlock(&tracepoint_module_list_mutex);
431 return ret;
432}
433
434static void tracepoint_module_going(struct module *mod)
435{
436 struct tp_module *tp_mod;
437
438 if (!mod->num_tracepoints)
439 return;
440
441 mutex_lock(&tracepoint_module_list_mutex);
442 list_for_each_entry(tp_mod, &tracepoint_module_list, list) {
443 if (tp_mod->mod == mod) {
444 blocking_notifier_call_chain(&tracepoint_notify_list,
445 MODULE_STATE_GOING, tp_mod);
446 list_del(&tp_mod->list);
447 kfree(tp_mod);
448 /*
449 * Called the going notifier before checking for
450 * quiescence.
451 */
452 tp_module_going_check_quiescent(mod->tracepoints_ptrs,
453 mod->tracepoints_ptrs + mod->num_tracepoints);
454 break;
455 }
456 }
457 /*
458 * In the case of modules that were tainted at "coming", we'll simply
459 * walk through the list without finding it. We cannot use the "tainted"
460 * flag on "going", in case a module taints the kernel only after being
461 * loaded.
462 */
463 mutex_unlock(&tracepoint_module_list_mutex);
464}
465
466static int tracepoint_module_notify(struct notifier_block *self,
467 unsigned long val, void *data)
468{
469 struct module *mod = data;
470 int ret = 0;
471
472 switch (val) {
473 case MODULE_STATE_COMING:
474 ret = tracepoint_module_coming(mod);
475 break;
476 case MODULE_STATE_LIVE:
477 break;
478 case MODULE_STATE_GOING:
479 tracepoint_module_going(mod);
480 break;
481 case MODULE_STATE_UNFORMED:
482 break;
483 }
484 return ret;
485}
486
487static struct notifier_block tracepoint_module_nb = {
488 .notifier_call = tracepoint_module_notify,
489 .priority = 0,
490};
491
492static __init int init_tracepoints(void)
493{
494 int ret;
495
496 ret = register_module_notifier(&tracepoint_module_nb);
497 if (ret)
498 pr_warn("Failed to register tracepoint module enter notifier\n");
499
500 return ret;
501}
502__initcall(init_tracepoints);
503#endif /* CONFIG_MODULES */
504
505static void for_each_tracepoint_range(struct tracepoint * const *begin,
506 struct tracepoint * const *end,
507 void (*fct)(struct tracepoint *tp, void *priv),
508 void *priv)
509{
510 struct tracepoint * const *iter;
511
512 if (!begin)
513 return;
514 for (iter = begin; iter < end; iter++)
515 fct(*iter, priv);
516}
517
518/**
519 * for_each_kernel_tracepoint - iteration on all kernel tracepoints
520 * @fct: callback
521 * @priv: private data
522 */
523void for_each_kernel_tracepoint(void (*fct)(struct tracepoint *tp, void *priv),
524 void *priv)
525{
526 for_each_tracepoint_range(__start___tracepoints_ptrs,
527 __stop___tracepoints_ptrs, fct, priv);
528}
529EXPORT_SYMBOL_GPL(for_each_kernel_tracepoint);
530
531#ifdef CONFIG_HAVE_SYSCALL_TRACEPOINTS
532
533/* NB: reg/unreg are called while guarded with the tracepoints_mutex */
534static int sys_tracepoint_refcount;
535
536int syscall_regfunc(void)
537{
538 struct task_struct *p, *t;
539
540 if (!sys_tracepoint_refcount) {
541 read_lock(&tasklist_lock);
542 for_each_process_thread(p, t) {
543 set_tsk_thread_flag(t, TIF_SYSCALL_TRACEPOINT);
544 }
545 read_unlock(&tasklist_lock);
546 }
547 sys_tracepoint_refcount++;
548
549 return 0;
550}
551
552void syscall_unregfunc(void)
553{
554 struct task_struct *p, *t;
555
556 sys_tracepoint_refcount--;
557 if (!sys_tracepoint_refcount) {
558 read_lock(&tasklist_lock);
559 for_each_process_thread(p, t) {
560 clear_tsk_thread_flag(t, TIF_SYSCALL_TRACEPOINT);
561 }
562 read_unlock(&tasklist_lock);
563 }
564}
565#endif