Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * Copyright (C) 2008-2014 Mathieu Desnoyers
4 */
5#include <linux/module.h>
6#include <linux/mutex.h>
7#include <linux/types.h>
8#include <linux/jhash.h>
9#include <linux/list.h>
10#include <linux/rcupdate.h>
11#include <linux/tracepoint.h>
12#include <linux/err.h>
13#include <linux/slab.h>
14#include <linux/sched/signal.h>
15#include <linux/sched/task.h>
16#include <linux/static_key.h>
17
18extern tracepoint_ptr_t __start___tracepoints_ptrs[];
19extern tracepoint_ptr_t __stop___tracepoints_ptrs[];
20
21DEFINE_SRCU(tracepoint_srcu);
22EXPORT_SYMBOL_GPL(tracepoint_srcu);
23
24/* Set to 1 to enable tracepoint debug output */
25static const int tracepoint_debug;
26
27#ifdef CONFIG_MODULES
28/*
29 * Tracepoint module list mutex protects the local module list.
30 */
31static DEFINE_MUTEX(tracepoint_module_list_mutex);
32
33/* Local list of struct tp_module */
34static LIST_HEAD(tracepoint_module_list);
35#endif /* CONFIG_MODULES */
36
37/*
38 * tracepoints_mutex protects the builtin and module tracepoints.
39 * tracepoints_mutex nests inside tracepoint_module_list_mutex.
40 */
41static DEFINE_MUTEX(tracepoints_mutex);
42
43static struct rcu_head *early_probes;
44static bool ok_to_free_tracepoints;
45
46/*
47 * Note about RCU :
48 * It is used to delay the free of multiple probes array until a quiescent
49 * state is reached.
50 */
51struct tp_probes {
52 struct rcu_head rcu;
53 struct tracepoint_func probes[0];
54};
55
56static inline void *allocate_probes(int count)
57{
58 struct tp_probes *p = kmalloc(struct_size(p, probes, count),
59 GFP_KERNEL);
60 return p == NULL ? NULL : p->probes;
61}
62
63static void srcu_free_old_probes(struct rcu_head *head)
64{
65 kfree(container_of(head, struct tp_probes, rcu));
66}
67
68static void rcu_free_old_probes(struct rcu_head *head)
69{
70 call_srcu(&tracepoint_srcu, head, srcu_free_old_probes);
71}
72
73static __init int release_early_probes(void)
74{
75 struct rcu_head *tmp;
76
77 ok_to_free_tracepoints = true;
78
79 while (early_probes) {
80 tmp = early_probes;
81 early_probes = tmp->next;
82 call_rcu(tmp, rcu_free_old_probes);
83 }
84
85 return 0;
86}
87
88/* SRCU is initialized at core_initcall */
89postcore_initcall(release_early_probes);
90
91static inline void release_probes(struct tracepoint_func *old)
92{
93 if (old) {
94 struct tp_probes *tp_probes = container_of(old,
95 struct tp_probes, probes[0]);
96
97 /*
98 * We can't free probes if SRCU is not initialized yet.
99 * Postpone the freeing till after SRCU is initialized.
100 */
101 if (unlikely(!ok_to_free_tracepoints)) {
102 tp_probes->rcu.next = early_probes;
103 early_probes = &tp_probes->rcu;
104 return;
105 }
106
107 /*
108 * Tracepoint probes are protected by both sched RCU and SRCU,
109 * by calling the SRCU callback in the sched RCU callback we
110 * cover both cases. So let us chain the SRCU and sched RCU
111 * callbacks to wait for both grace periods.
112 */
113 call_rcu(&tp_probes->rcu, rcu_free_old_probes);
114 }
115}
116
117static void debug_print_probes(struct tracepoint_func *funcs)
118{
119 int i;
120
121 if (!tracepoint_debug || !funcs)
122 return;
123
124 for (i = 0; funcs[i].func; i++)
125 printk(KERN_DEBUG "Probe %d : %p\n", i, funcs[i].func);
126}
127
128static struct tracepoint_func *
129func_add(struct tracepoint_func **funcs, struct tracepoint_func *tp_func,
130 int prio)
131{
132 struct tracepoint_func *old, *new;
133 int nr_probes = 0;
134 int pos = -1;
135
136 if (WARN_ON(!tp_func->func))
137 return ERR_PTR(-EINVAL);
138
139 debug_print_probes(*funcs);
140 old = *funcs;
141 if (old) {
142 /* (N -> N+1), (N != 0, 1) probes */
143 for (nr_probes = 0; old[nr_probes].func; nr_probes++) {
144 /* Insert before probes of lower priority */
145 if (pos < 0 && old[nr_probes].prio < prio)
146 pos = nr_probes;
147 if (old[nr_probes].func == tp_func->func &&
148 old[nr_probes].data == tp_func->data)
149 return ERR_PTR(-EEXIST);
150 }
151 }
152 /* + 2 : one for new probe, one for NULL func */
153 new = allocate_probes(nr_probes + 2);
154 if (new == NULL)
155 return ERR_PTR(-ENOMEM);
156 if (old) {
157 if (pos < 0) {
158 pos = nr_probes;
159 memcpy(new, old, nr_probes * sizeof(struct tracepoint_func));
160 } else {
161 /* Copy higher priority probes ahead of the new probe */
162 memcpy(new, old, pos * sizeof(struct tracepoint_func));
163 /* Copy the rest after it. */
164 memcpy(new + pos + 1, old + pos,
165 (nr_probes - pos) * sizeof(struct tracepoint_func));
166 }
167 } else
168 pos = 0;
169 new[pos] = *tp_func;
170 new[nr_probes + 1].func = NULL;
171 *funcs = new;
172 debug_print_probes(*funcs);
173 return old;
174}
175
176static void *func_remove(struct tracepoint_func **funcs,
177 struct tracepoint_func *tp_func)
178{
179 int nr_probes = 0, nr_del = 0, i;
180 struct tracepoint_func *old, *new;
181
182 old = *funcs;
183
184 if (!old)
185 return ERR_PTR(-ENOENT);
186
187 debug_print_probes(*funcs);
188 /* (N -> M), (N > 1, M >= 0) probes */
189 if (tp_func->func) {
190 for (nr_probes = 0; old[nr_probes].func; nr_probes++) {
191 if (old[nr_probes].func == tp_func->func &&
192 old[nr_probes].data == tp_func->data)
193 nr_del++;
194 }
195 }
196
197 /*
198 * If probe is NULL, then nr_probes = nr_del = 0, and then the
199 * entire entry will be removed.
200 */
201 if (nr_probes - nr_del == 0) {
202 /* N -> 0, (N > 1) */
203 *funcs = NULL;
204 debug_print_probes(*funcs);
205 return old;
206 } else {
207 int j = 0;
208 /* N -> M, (N > 1, M > 0) */
209 /* + 1 for NULL */
210 new = allocate_probes(nr_probes - nr_del + 1);
211 if (new == NULL)
212 return ERR_PTR(-ENOMEM);
213 for (i = 0; old[i].func; i++)
214 if (old[i].func != tp_func->func
215 || old[i].data != tp_func->data)
216 new[j++] = old[i];
217 new[nr_probes - nr_del].func = NULL;
218 *funcs = new;
219 }
220 debug_print_probes(*funcs);
221 return old;
222}
223
224/*
225 * Add the probe function to a tracepoint.
226 */
227static int tracepoint_add_func(struct tracepoint *tp,
228 struct tracepoint_func *func, int prio)
229{
230 struct tracepoint_func *old, *tp_funcs;
231 int ret;
232
233 if (tp->regfunc && !static_key_enabled(&tp->key)) {
234 ret = tp->regfunc();
235 if (ret < 0)
236 return ret;
237 }
238
239 tp_funcs = rcu_dereference_protected(tp->funcs,
240 lockdep_is_held(&tracepoints_mutex));
241 old = func_add(&tp_funcs, func, prio);
242 if (IS_ERR(old)) {
243 WARN_ON_ONCE(PTR_ERR(old) != -ENOMEM);
244 return PTR_ERR(old);
245 }
246
247 /*
248 * rcu_assign_pointer has as smp_store_release() which makes sure
249 * that the new probe callbacks array is consistent before setting
250 * a pointer to it. This array is referenced by __DO_TRACE from
251 * include/linux/tracepoint.h using rcu_dereference_sched().
252 */
253 rcu_assign_pointer(tp->funcs, tp_funcs);
254 if (!static_key_enabled(&tp->key))
255 static_key_slow_inc(&tp->key);
256 release_probes(old);
257 return 0;
258}
259
260/*
261 * Remove a probe function from a tracepoint.
262 * Note: only waiting an RCU period after setting elem->call to the empty
263 * function insures that the original callback is not used anymore. This insured
264 * by preempt_disable around the call site.
265 */
266static int tracepoint_remove_func(struct tracepoint *tp,
267 struct tracepoint_func *func)
268{
269 struct tracepoint_func *old, *tp_funcs;
270
271 tp_funcs = rcu_dereference_protected(tp->funcs,
272 lockdep_is_held(&tracepoints_mutex));
273 old = func_remove(&tp_funcs, func);
274 if (IS_ERR(old)) {
275 WARN_ON_ONCE(PTR_ERR(old) != -ENOMEM);
276 return PTR_ERR(old);
277 }
278
279 if (!tp_funcs) {
280 /* Removed last function */
281 if (tp->unregfunc && static_key_enabled(&tp->key))
282 tp->unregfunc();
283
284 if (static_key_enabled(&tp->key))
285 static_key_slow_dec(&tp->key);
286 }
287 rcu_assign_pointer(tp->funcs, tp_funcs);
288 release_probes(old);
289 return 0;
290}
291
292/**
293 * tracepoint_probe_register_prio - Connect a probe to a tracepoint with priority
294 * @tp: tracepoint
295 * @probe: probe handler
296 * @data: tracepoint data
297 * @prio: priority of this function over other registered functions
298 *
299 * Returns 0 if ok, error value on error.
300 * Note: if @tp is within a module, the caller is responsible for
301 * unregistering the probe before the module is gone. This can be
302 * performed either with a tracepoint module going notifier, or from
303 * within module exit functions.
304 */
305int tracepoint_probe_register_prio(struct tracepoint *tp, void *probe,
306 void *data, int prio)
307{
308 struct tracepoint_func tp_func;
309 int ret;
310
311 mutex_lock(&tracepoints_mutex);
312 tp_func.func = probe;
313 tp_func.data = data;
314 tp_func.prio = prio;
315 ret = tracepoint_add_func(tp, &tp_func, prio);
316 mutex_unlock(&tracepoints_mutex);
317 return ret;
318}
319EXPORT_SYMBOL_GPL(tracepoint_probe_register_prio);
320
321/**
322 * tracepoint_probe_register - Connect a probe to a tracepoint
323 * @tp: tracepoint
324 * @probe: probe handler
325 * @data: tracepoint data
326 *
327 * Returns 0 if ok, error value on error.
328 * Note: if @tp is within a module, the caller is responsible for
329 * unregistering the probe before the module is gone. This can be
330 * performed either with a tracepoint module going notifier, or from
331 * within module exit functions.
332 */
333int tracepoint_probe_register(struct tracepoint *tp, void *probe, void *data)
334{
335 return tracepoint_probe_register_prio(tp, probe, data, TRACEPOINT_DEFAULT_PRIO);
336}
337EXPORT_SYMBOL_GPL(tracepoint_probe_register);
338
339/**
340 * tracepoint_probe_unregister - Disconnect a probe from a tracepoint
341 * @tp: tracepoint
342 * @probe: probe function pointer
343 * @data: tracepoint data
344 *
345 * Returns 0 if ok, error value on error.
346 */
347int tracepoint_probe_unregister(struct tracepoint *tp, void *probe, void *data)
348{
349 struct tracepoint_func tp_func;
350 int ret;
351
352 mutex_lock(&tracepoints_mutex);
353 tp_func.func = probe;
354 tp_func.data = data;
355 ret = tracepoint_remove_func(tp, &tp_func);
356 mutex_unlock(&tracepoints_mutex);
357 return ret;
358}
359EXPORT_SYMBOL_GPL(tracepoint_probe_unregister);
360
361static void for_each_tracepoint_range(
362 tracepoint_ptr_t *begin, tracepoint_ptr_t *end,
363 void (*fct)(struct tracepoint *tp, void *priv),
364 void *priv)
365{
366 tracepoint_ptr_t *iter;
367
368 if (!begin)
369 return;
370 for (iter = begin; iter < end; iter++)
371 fct(tracepoint_ptr_deref(iter), priv);
372}
373
374#ifdef CONFIG_MODULES
375bool trace_module_has_bad_taint(struct module *mod)
376{
377 return mod->taints & ~((1 << TAINT_OOT_MODULE) | (1 << TAINT_CRAP) |
378 (1 << TAINT_UNSIGNED_MODULE));
379}
380
381static BLOCKING_NOTIFIER_HEAD(tracepoint_notify_list);
382
383/**
384 * register_tracepoint_notifier - register tracepoint coming/going notifier
385 * @nb: notifier block
386 *
387 * Notifiers registered with this function are called on module
388 * coming/going with the tracepoint_module_list_mutex held.
389 * The notifier block callback should expect a "struct tp_module" data
390 * pointer.
391 */
392int register_tracepoint_module_notifier(struct notifier_block *nb)
393{
394 struct tp_module *tp_mod;
395 int ret;
396
397 mutex_lock(&tracepoint_module_list_mutex);
398 ret = blocking_notifier_chain_register(&tracepoint_notify_list, nb);
399 if (ret)
400 goto end;
401 list_for_each_entry(tp_mod, &tracepoint_module_list, list)
402 (void) nb->notifier_call(nb, MODULE_STATE_COMING, tp_mod);
403end:
404 mutex_unlock(&tracepoint_module_list_mutex);
405 return ret;
406}
407EXPORT_SYMBOL_GPL(register_tracepoint_module_notifier);
408
409/**
410 * unregister_tracepoint_notifier - unregister tracepoint coming/going notifier
411 * @nb: notifier block
412 *
413 * The notifier block callback should expect a "struct tp_module" data
414 * pointer.
415 */
416int unregister_tracepoint_module_notifier(struct notifier_block *nb)
417{
418 struct tp_module *tp_mod;
419 int ret;
420
421 mutex_lock(&tracepoint_module_list_mutex);
422 ret = blocking_notifier_chain_unregister(&tracepoint_notify_list, nb);
423 if (ret)
424 goto end;
425 list_for_each_entry(tp_mod, &tracepoint_module_list, list)
426 (void) nb->notifier_call(nb, MODULE_STATE_GOING, tp_mod);
427end:
428 mutex_unlock(&tracepoint_module_list_mutex);
429 return ret;
430
431}
432EXPORT_SYMBOL_GPL(unregister_tracepoint_module_notifier);
433
434/*
435 * Ensure the tracer unregistered the module's probes before the module
436 * teardown is performed. Prevents leaks of probe and data pointers.
437 */
438static void tp_module_going_check_quiescent(struct tracepoint *tp, void *priv)
439{
440 WARN_ON_ONCE(tp->funcs);
441}
442
443static int tracepoint_module_coming(struct module *mod)
444{
445 struct tp_module *tp_mod;
446 int ret = 0;
447
448 if (!mod->num_tracepoints)
449 return 0;
450
451 /*
452 * We skip modules that taint the kernel, especially those with different
453 * module headers (for forced load), to make sure we don't cause a crash.
454 * Staging, out-of-tree, and unsigned GPL modules are fine.
455 */
456 if (trace_module_has_bad_taint(mod))
457 return 0;
458 mutex_lock(&tracepoint_module_list_mutex);
459 tp_mod = kmalloc(sizeof(struct tp_module), GFP_KERNEL);
460 if (!tp_mod) {
461 ret = -ENOMEM;
462 goto end;
463 }
464 tp_mod->mod = mod;
465 list_add_tail(&tp_mod->list, &tracepoint_module_list);
466 blocking_notifier_call_chain(&tracepoint_notify_list,
467 MODULE_STATE_COMING, tp_mod);
468end:
469 mutex_unlock(&tracepoint_module_list_mutex);
470 return ret;
471}
472
473static void tracepoint_module_going(struct module *mod)
474{
475 struct tp_module *tp_mod;
476
477 if (!mod->num_tracepoints)
478 return;
479
480 mutex_lock(&tracepoint_module_list_mutex);
481 list_for_each_entry(tp_mod, &tracepoint_module_list, list) {
482 if (tp_mod->mod == mod) {
483 blocking_notifier_call_chain(&tracepoint_notify_list,
484 MODULE_STATE_GOING, tp_mod);
485 list_del(&tp_mod->list);
486 kfree(tp_mod);
487 /*
488 * Called the going notifier before checking for
489 * quiescence.
490 */
491 for_each_tracepoint_range(mod->tracepoints_ptrs,
492 mod->tracepoints_ptrs + mod->num_tracepoints,
493 tp_module_going_check_quiescent, NULL);
494 break;
495 }
496 }
497 /*
498 * In the case of modules that were tainted at "coming", we'll simply
499 * walk through the list without finding it. We cannot use the "tainted"
500 * flag on "going", in case a module taints the kernel only after being
501 * loaded.
502 */
503 mutex_unlock(&tracepoint_module_list_mutex);
504}
505
506static int tracepoint_module_notify(struct notifier_block *self,
507 unsigned long val, void *data)
508{
509 struct module *mod = data;
510 int ret = 0;
511
512 switch (val) {
513 case MODULE_STATE_COMING:
514 ret = tracepoint_module_coming(mod);
515 break;
516 case MODULE_STATE_LIVE:
517 break;
518 case MODULE_STATE_GOING:
519 tracepoint_module_going(mod);
520 break;
521 case MODULE_STATE_UNFORMED:
522 break;
523 }
524 return ret;
525}
526
527static struct notifier_block tracepoint_module_nb = {
528 .notifier_call = tracepoint_module_notify,
529 .priority = 0,
530};
531
532static __init int init_tracepoints(void)
533{
534 int ret;
535
536 ret = register_module_notifier(&tracepoint_module_nb);
537 if (ret)
538 pr_warn("Failed to register tracepoint module enter notifier\n");
539
540 return ret;
541}
542__initcall(init_tracepoints);
543#endif /* CONFIG_MODULES */
544
545/**
546 * for_each_kernel_tracepoint - iteration on all kernel tracepoints
547 * @fct: callback
548 * @priv: private data
549 */
550void for_each_kernel_tracepoint(void (*fct)(struct tracepoint *tp, void *priv),
551 void *priv)
552{
553 for_each_tracepoint_range(__start___tracepoints_ptrs,
554 __stop___tracepoints_ptrs, fct, priv);
555}
556EXPORT_SYMBOL_GPL(for_each_kernel_tracepoint);
557
558#ifdef CONFIG_HAVE_SYSCALL_TRACEPOINTS
559
560/* NB: reg/unreg are called while guarded with the tracepoints_mutex */
561static int sys_tracepoint_refcount;
562
563int syscall_regfunc(void)
564{
565 struct task_struct *p, *t;
566
567 if (!sys_tracepoint_refcount) {
568 read_lock(&tasklist_lock);
569 for_each_process_thread(p, t) {
570 set_tsk_thread_flag(t, TIF_SYSCALL_TRACEPOINT);
571 }
572 read_unlock(&tasklist_lock);
573 }
574 sys_tracepoint_refcount++;
575
576 return 0;
577}
578
579void syscall_unregfunc(void)
580{
581 struct task_struct *p, *t;
582
583 sys_tracepoint_refcount--;
584 if (!sys_tracepoint_refcount) {
585 read_lock(&tasklist_lock);
586 for_each_process_thread(p, t) {
587 clear_tsk_thread_flag(t, TIF_SYSCALL_TRACEPOINT);
588 }
589 read_unlock(&tasklist_lock);
590 }
591}
592#endif
1/*
2 * Copyright (C) 2008 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/jump_label.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/*
37 * tracepoints_mutex nests inside module_mutex. Tracepoints mutex protects the
38 * builtin and module tracepoints and the hash table.
39 */
40static DEFINE_MUTEX(tracepoints_mutex);
41
42/*
43 * Tracepoint hash table, containing the active tracepoints.
44 * Protected by tracepoints_mutex.
45 */
46#define TRACEPOINT_HASH_BITS 6
47#define TRACEPOINT_TABLE_SIZE (1 << TRACEPOINT_HASH_BITS)
48static struct hlist_head tracepoint_table[TRACEPOINT_TABLE_SIZE];
49
50/*
51 * Note about RCU :
52 * It is used to delay the free of multiple probes array until a quiescent
53 * state is reached.
54 * Tracepoint entries modifications are protected by the tracepoints_mutex.
55 */
56struct tracepoint_entry {
57 struct hlist_node hlist;
58 struct tracepoint_func *funcs;
59 int refcount; /* Number of times armed. 0 if disarmed. */
60 char name[0];
61};
62
63struct tp_probes {
64 union {
65 struct rcu_head rcu;
66 struct list_head list;
67 } u;
68 struct tracepoint_func probes[0];
69};
70
71static inline void *allocate_probes(int count)
72{
73 struct tp_probes *p = kmalloc(count * sizeof(struct tracepoint_func)
74 + sizeof(struct tp_probes), GFP_KERNEL);
75 return p == NULL ? NULL : p->probes;
76}
77
78static void rcu_free_old_probes(struct rcu_head *head)
79{
80 kfree(container_of(head, struct tp_probes, u.rcu));
81}
82
83static inline void release_probes(struct tracepoint_func *old)
84{
85 if (old) {
86 struct tp_probes *tp_probes = container_of(old,
87 struct tp_probes, probes[0]);
88 call_rcu_sched(&tp_probes->u.rcu, rcu_free_old_probes);
89 }
90}
91
92static void debug_print_probes(struct tracepoint_entry *entry)
93{
94 int i;
95
96 if (!tracepoint_debug || !entry->funcs)
97 return;
98
99 for (i = 0; entry->funcs[i].func; i++)
100 printk(KERN_DEBUG "Probe %d : %p\n", i, entry->funcs[i].func);
101}
102
103static struct tracepoint_func *
104tracepoint_entry_add_probe(struct tracepoint_entry *entry,
105 void *probe, void *data)
106{
107 int nr_probes = 0;
108 struct tracepoint_func *old, *new;
109
110 WARN_ON(!probe);
111
112 debug_print_probes(entry);
113 old = entry->funcs;
114 if (old) {
115 /* (N -> N+1), (N != 0, 1) probes */
116 for (nr_probes = 0; old[nr_probes].func; nr_probes++)
117 if (old[nr_probes].func == probe &&
118 old[nr_probes].data == data)
119 return ERR_PTR(-EEXIST);
120 }
121 /* + 2 : one for new probe, one for NULL func */
122 new = allocate_probes(nr_probes + 2);
123 if (new == NULL)
124 return ERR_PTR(-ENOMEM);
125 if (old)
126 memcpy(new, old, nr_probes * sizeof(struct tracepoint_func));
127 new[nr_probes].func = probe;
128 new[nr_probes].data = data;
129 new[nr_probes + 1].func = NULL;
130 entry->refcount = nr_probes + 1;
131 entry->funcs = new;
132 debug_print_probes(entry);
133 return old;
134}
135
136static void *
137tracepoint_entry_remove_probe(struct tracepoint_entry *entry,
138 void *probe, void *data)
139{
140 int nr_probes = 0, nr_del = 0, i;
141 struct tracepoint_func *old, *new;
142
143 old = entry->funcs;
144
145 if (!old)
146 return ERR_PTR(-ENOENT);
147
148 debug_print_probes(entry);
149 /* (N -> M), (N > 1, M >= 0) probes */
150 for (nr_probes = 0; old[nr_probes].func; nr_probes++) {
151 if (!probe ||
152 (old[nr_probes].func == probe &&
153 old[nr_probes].data == data))
154 nr_del++;
155 }
156
157 if (nr_probes - nr_del == 0) {
158 /* N -> 0, (N > 1) */
159 entry->funcs = NULL;
160 entry->refcount = 0;
161 debug_print_probes(entry);
162 return old;
163 } else {
164 int j = 0;
165 /* N -> M, (N > 1, M > 0) */
166 /* + 1 for NULL */
167 new = allocate_probes(nr_probes - nr_del + 1);
168 if (new == NULL)
169 return ERR_PTR(-ENOMEM);
170 for (i = 0; old[i].func; i++)
171 if (probe &&
172 (old[i].func != probe || old[i].data != data))
173 new[j++] = old[i];
174 new[nr_probes - nr_del].func = NULL;
175 entry->refcount = nr_probes - nr_del;
176 entry->funcs = new;
177 }
178 debug_print_probes(entry);
179 return old;
180}
181
182/*
183 * Get tracepoint if the tracepoint is present in the tracepoint hash table.
184 * Must be called with tracepoints_mutex held.
185 * Returns NULL if not present.
186 */
187static struct tracepoint_entry *get_tracepoint(const char *name)
188{
189 struct hlist_head *head;
190 struct hlist_node *node;
191 struct tracepoint_entry *e;
192 u32 hash = jhash(name, strlen(name), 0);
193
194 head = &tracepoint_table[hash & (TRACEPOINT_TABLE_SIZE - 1)];
195 hlist_for_each_entry(e, node, head, hlist) {
196 if (!strcmp(name, e->name))
197 return e;
198 }
199 return NULL;
200}
201
202/*
203 * Add the tracepoint to the tracepoint hash table. Must be called with
204 * tracepoints_mutex held.
205 */
206static struct tracepoint_entry *add_tracepoint(const char *name)
207{
208 struct hlist_head *head;
209 struct hlist_node *node;
210 struct tracepoint_entry *e;
211 size_t name_len = strlen(name) + 1;
212 u32 hash = jhash(name, name_len-1, 0);
213
214 head = &tracepoint_table[hash & (TRACEPOINT_TABLE_SIZE - 1)];
215 hlist_for_each_entry(e, node, head, hlist) {
216 if (!strcmp(name, e->name)) {
217 printk(KERN_NOTICE
218 "tracepoint %s busy\n", name);
219 return ERR_PTR(-EEXIST); /* Already there */
220 }
221 }
222 /*
223 * Using kmalloc here to allocate a variable length element. Could
224 * cause some memory fragmentation if overused.
225 */
226 e = kmalloc(sizeof(struct tracepoint_entry) + name_len, GFP_KERNEL);
227 if (!e)
228 return ERR_PTR(-ENOMEM);
229 memcpy(&e->name[0], name, name_len);
230 e->funcs = NULL;
231 e->refcount = 0;
232 hlist_add_head(&e->hlist, head);
233 return e;
234}
235
236/*
237 * Remove the tracepoint from the tracepoint hash table. Must be called with
238 * mutex_lock held.
239 */
240static inline void remove_tracepoint(struct tracepoint_entry *e)
241{
242 hlist_del(&e->hlist);
243 kfree(e);
244}
245
246/*
247 * Sets the probe callback corresponding to one tracepoint.
248 */
249static void set_tracepoint(struct tracepoint_entry **entry,
250 struct tracepoint *elem, int active)
251{
252 WARN_ON(strcmp((*entry)->name, elem->name) != 0);
253
254 if (elem->regfunc && !jump_label_enabled(&elem->key) && active)
255 elem->regfunc();
256 else if (elem->unregfunc && jump_label_enabled(&elem->key) && !active)
257 elem->unregfunc();
258
259 /*
260 * rcu_assign_pointer has a smp_wmb() which makes sure that the new
261 * probe callbacks array is consistent before setting a pointer to it.
262 * This array is referenced by __DO_TRACE from
263 * include/linux/tracepoints.h. A matching smp_read_barrier_depends()
264 * is used.
265 */
266 rcu_assign_pointer(elem->funcs, (*entry)->funcs);
267 if (active && !jump_label_enabled(&elem->key))
268 jump_label_inc(&elem->key);
269 else if (!active && jump_label_enabled(&elem->key))
270 jump_label_dec(&elem->key);
271}
272
273/*
274 * Disable a tracepoint and its probe callback.
275 * Note: only waiting an RCU period after setting elem->call to the empty
276 * function insures that the original callback is not used anymore. This insured
277 * by preempt_disable around the call site.
278 */
279static void disable_tracepoint(struct tracepoint *elem)
280{
281 if (elem->unregfunc && jump_label_enabled(&elem->key))
282 elem->unregfunc();
283
284 if (jump_label_enabled(&elem->key))
285 jump_label_dec(&elem->key);
286 rcu_assign_pointer(elem->funcs, NULL);
287}
288
289/**
290 * tracepoint_update_probe_range - Update a probe range
291 * @begin: beginning of the range
292 * @end: end of the range
293 *
294 * Updates the probe callback corresponding to a range of tracepoints.
295 */
296void tracepoint_update_probe_range(struct tracepoint * const *begin,
297 struct tracepoint * const *end)
298{
299 struct tracepoint * const *iter;
300 struct tracepoint_entry *mark_entry;
301
302 if (!begin)
303 return;
304
305 mutex_lock(&tracepoints_mutex);
306 for (iter = begin; iter < end; iter++) {
307 mark_entry = get_tracepoint((*iter)->name);
308 if (mark_entry) {
309 set_tracepoint(&mark_entry, *iter,
310 !!mark_entry->refcount);
311 } else {
312 disable_tracepoint(*iter);
313 }
314 }
315 mutex_unlock(&tracepoints_mutex);
316}
317
318/*
319 * Update probes, removing the faulty probes.
320 */
321static void tracepoint_update_probes(void)
322{
323 /* Core kernel tracepoints */
324 tracepoint_update_probe_range(__start___tracepoints_ptrs,
325 __stop___tracepoints_ptrs);
326 /* tracepoints in modules. */
327 module_update_tracepoints();
328}
329
330static struct tracepoint_func *
331tracepoint_add_probe(const char *name, void *probe, void *data)
332{
333 struct tracepoint_entry *entry;
334 struct tracepoint_func *old;
335
336 entry = get_tracepoint(name);
337 if (!entry) {
338 entry = add_tracepoint(name);
339 if (IS_ERR(entry))
340 return (struct tracepoint_func *)entry;
341 }
342 old = tracepoint_entry_add_probe(entry, probe, data);
343 if (IS_ERR(old) && !entry->refcount)
344 remove_tracepoint(entry);
345 return old;
346}
347
348/**
349 * tracepoint_probe_register - Connect a probe to a tracepoint
350 * @name: tracepoint name
351 * @probe: probe handler
352 *
353 * Returns 0 if ok, error value on error.
354 * The probe address must at least be aligned on the architecture pointer size.
355 */
356int tracepoint_probe_register(const char *name, void *probe, void *data)
357{
358 struct tracepoint_func *old;
359
360 mutex_lock(&tracepoints_mutex);
361 old = tracepoint_add_probe(name, probe, data);
362 mutex_unlock(&tracepoints_mutex);
363 if (IS_ERR(old))
364 return PTR_ERR(old);
365
366 tracepoint_update_probes(); /* may update entry */
367 release_probes(old);
368 return 0;
369}
370EXPORT_SYMBOL_GPL(tracepoint_probe_register);
371
372static struct tracepoint_func *
373tracepoint_remove_probe(const char *name, void *probe, void *data)
374{
375 struct tracepoint_entry *entry;
376 struct tracepoint_func *old;
377
378 entry = get_tracepoint(name);
379 if (!entry)
380 return ERR_PTR(-ENOENT);
381 old = tracepoint_entry_remove_probe(entry, probe, data);
382 if (IS_ERR(old))
383 return old;
384 if (!entry->refcount)
385 remove_tracepoint(entry);
386 return old;
387}
388
389/**
390 * tracepoint_probe_unregister - Disconnect a probe from a tracepoint
391 * @name: tracepoint name
392 * @probe: probe function pointer
393 *
394 * We do not need to call a synchronize_sched to make sure the probes have
395 * finished running before doing a module unload, because the module unload
396 * itself uses stop_machine(), which insures that every preempt disabled section
397 * have finished.
398 */
399int tracepoint_probe_unregister(const char *name, void *probe, void *data)
400{
401 struct tracepoint_func *old;
402
403 mutex_lock(&tracepoints_mutex);
404 old = tracepoint_remove_probe(name, probe, data);
405 mutex_unlock(&tracepoints_mutex);
406 if (IS_ERR(old))
407 return PTR_ERR(old);
408
409 tracepoint_update_probes(); /* may update entry */
410 release_probes(old);
411 return 0;
412}
413EXPORT_SYMBOL_GPL(tracepoint_probe_unregister);
414
415static LIST_HEAD(old_probes);
416static int need_update;
417
418static void tracepoint_add_old_probes(void *old)
419{
420 need_update = 1;
421 if (old) {
422 struct tp_probes *tp_probes = container_of(old,
423 struct tp_probes, probes[0]);
424 list_add(&tp_probes->u.list, &old_probes);
425 }
426}
427
428/**
429 * tracepoint_probe_register_noupdate - register a probe but not connect
430 * @name: tracepoint name
431 * @probe: probe handler
432 *
433 * caller must call tracepoint_probe_update_all()
434 */
435int tracepoint_probe_register_noupdate(const char *name, void *probe,
436 void *data)
437{
438 struct tracepoint_func *old;
439
440 mutex_lock(&tracepoints_mutex);
441 old = tracepoint_add_probe(name, probe, data);
442 if (IS_ERR(old)) {
443 mutex_unlock(&tracepoints_mutex);
444 return PTR_ERR(old);
445 }
446 tracepoint_add_old_probes(old);
447 mutex_unlock(&tracepoints_mutex);
448 return 0;
449}
450EXPORT_SYMBOL_GPL(tracepoint_probe_register_noupdate);
451
452/**
453 * tracepoint_probe_unregister_noupdate - remove a probe but not disconnect
454 * @name: tracepoint name
455 * @probe: probe function pointer
456 *
457 * caller must call tracepoint_probe_update_all()
458 */
459int tracepoint_probe_unregister_noupdate(const char *name, void *probe,
460 void *data)
461{
462 struct tracepoint_func *old;
463
464 mutex_lock(&tracepoints_mutex);
465 old = tracepoint_remove_probe(name, probe, data);
466 if (IS_ERR(old)) {
467 mutex_unlock(&tracepoints_mutex);
468 return PTR_ERR(old);
469 }
470 tracepoint_add_old_probes(old);
471 mutex_unlock(&tracepoints_mutex);
472 return 0;
473}
474EXPORT_SYMBOL_GPL(tracepoint_probe_unregister_noupdate);
475
476/**
477 * tracepoint_probe_update_all - update tracepoints
478 */
479void tracepoint_probe_update_all(void)
480{
481 LIST_HEAD(release_probes);
482 struct tp_probes *pos, *next;
483
484 mutex_lock(&tracepoints_mutex);
485 if (!need_update) {
486 mutex_unlock(&tracepoints_mutex);
487 return;
488 }
489 if (!list_empty(&old_probes))
490 list_replace_init(&old_probes, &release_probes);
491 need_update = 0;
492 mutex_unlock(&tracepoints_mutex);
493
494 tracepoint_update_probes();
495 list_for_each_entry_safe(pos, next, &release_probes, u.list) {
496 list_del(&pos->u.list);
497 call_rcu_sched(&pos->u.rcu, rcu_free_old_probes);
498 }
499}
500EXPORT_SYMBOL_GPL(tracepoint_probe_update_all);
501
502/**
503 * tracepoint_get_iter_range - Get a next tracepoint iterator given a range.
504 * @tracepoint: current tracepoints (in), next tracepoint (out)
505 * @begin: beginning of the range
506 * @end: end of the range
507 *
508 * Returns whether a next tracepoint has been found (1) or not (0).
509 * Will return the first tracepoint in the range if the input tracepoint is
510 * NULL.
511 */
512int tracepoint_get_iter_range(struct tracepoint * const **tracepoint,
513 struct tracepoint * const *begin, struct tracepoint * const *end)
514{
515 if (!*tracepoint && begin != end) {
516 *tracepoint = begin;
517 return 1;
518 }
519 if (*tracepoint >= begin && *tracepoint < end)
520 return 1;
521 return 0;
522}
523EXPORT_SYMBOL_GPL(tracepoint_get_iter_range);
524
525static void tracepoint_get_iter(struct tracepoint_iter *iter)
526{
527 int found = 0;
528
529 /* Core kernel tracepoints */
530 if (!iter->module) {
531 found = tracepoint_get_iter_range(&iter->tracepoint,
532 __start___tracepoints_ptrs,
533 __stop___tracepoints_ptrs);
534 if (found)
535 goto end;
536 }
537 /* tracepoints in modules. */
538 found = module_get_iter_tracepoints(iter);
539end:
540 if (!found)
541 tracepoint_iter_reset(iter);
542}
543
544void tracepoint_iter_start(struct tracepoint_iter *iter)
545{
546 tracepoint_get_iter(iter);
547}
548EXPORT_SYMBOL_GPL(tracepoint_iter_start);
549
550void tracepoint_iter_next(struct tracepoint_iter *iter)
551{
552 iter->tracepoint++;
553 /*
554 * iter->tracepoint may be invalid because we blindly incremented it.
555 * Make sure it is valid by marshalling on the tracepoints, getting the
556 * tracepoints from following modules if necessary.
557 */
558 tracepoint_get_iter(iter);
559}
560EXPORT_SYMBOL_GPL(tracepoint_iter_next);
561
562void tracepoint_iter_stop(struct tracepoint_iter *iter)
563{
564}
565EXPORT_SYMBOL_GPL(tracepoint_iter_stop);
566
567void tracepoint_iter_reset(struct tracepoint_iter *iter)
568{
569 iter->module = NULL;
570 iter->tracepoint = NULL;
571}
572EXPORT_SYMBOL_GPL(tracepoint_iter_reset);
573
574#ifdef CONFIG_MODULES
575
576int tracepoint_module_notify(struct notifier_block *self,
577 unsigned long val, void *data)
578{
579 struct module *mod = data;
580
581 switch (val) {
582 case MODULE_STATE_COMING:
583 case MODULE_STATE_GOING:
584 tracepoint_update_probe_range(mod->tracepoints_ptrs,
585 mod->tracepoints_ptrs + mod->num_tracepoints);
586 break;
587 }
588 return 0;
589}
590
591struct notifier_block tracepoint_module_nb = {
592 .notifier_call = tracepoint_module_notify,
593 .priority = 0,
594};
595
596static int init_tracepoints(void)
597{
598 return register_module_notifier(&tracepoint_module_nb);
599}
600__initcall(init_tracepoints);
601
602#endif /* CONFIG_MODULES */
603
604#ifdef CONFIG_HAVE_SYSCALL_TRACEPOINTS
605
606/* NB: reg/unreg are called while guarded with the tracepoints_mutex */
607static int sys_tracepoint_refcount;
608
609void syscall_regfunc(void)
610{
611 unsigned long flags;
612 struct task_struct *g, *t;
613
614 if (!sys_tracepoint_refcount) {
615 read_lock_irqsave(&tasklist_lock, flags);
616 do_each_thread(g, t) {
617 /* Skip kernel threads. */
618 if (t->mm)
619 set_tsk_thread_flag(t, TIF_SYSCALL_TRACEPOINT);
620 } while_each_thread(g, t);
621 read_unlock_irqrestore(&tasklist_lock, flags);
622 }
623 sys_tracepoint_refcount++;
624}
625
626void syscall_unregfunc(void)
627{
628 unsigned long flags;
629 struct task_struct *g, *t;
630
631 sys_tracepoint_refcount--;
632 if (!sys_tracepoint_refcount) {
633 read_lock_irqsave(&tasklist_lock, flags);
634 do_each_thread(g, t) {
635 clear_tsk_thread_flag(t, TIF_SYSCALL_TRACEPOINT);
636 } while_each_thread(g, t);
637 read_unlock_irqrestore(&tasklist_lock, flags);
638 }
639}
640#endif