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/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/*
37 * Tracepoints mutex protects the builtin and module tracepoints and the hash
38 * table, as well as the local module list.
39 */
40static DEFINE_MUTEX(tracepoints_mutex);
41
42#ifdef CONFIG_MODULES
43/* Local list of struct module */
44static LIST_HEAD(tracepoint_module_list);
45#endif /* CONFIG_MODULES */
46
47/*
48 * Tracepoint hash table, containing the active tracepoints.
49 * Protected by tracepoints_mutex.
50 */
51#define TRACEPOINT_HASH_BITS 6
52#define TRACEPOINT_TABLE_SIZE (1 << TRACEPOINT_HASH_BITS)
53static struct hlist_head tracepoint_table[TRACEPOINT_TABLE_SIZE];
54
55/*
56 * Note about RCU :
57 * It is used to delay the free of multiple probes array until a quiescent
58 * state is reached.
59 * Tracepoint entries modifications are protected by the tracepoints_mutex.
60 */
61struct tracepoint_entry {
62 struct hlist_node hlist;
63 struct tracepoint_func *funcs;
64 int refcount; /* Number of times armed. 0 if disarmed. */
65 char name[0];
66};
67
68struct tp_probes {
69 union {
70 struct rcu_head rcu;
71 struct list_head list;
72 } u;
73 struct tracepoint_func probes[0];
74};
75
76static inline void *allocate_probes(int count)
77{
78 struct tp_probes *p = kmalloc(count * sizeof(struct tracepoint_func)
79 + sizeof(struct tp_probes), GFP_KERNEL);
80 return p == NULL ? NULL : p->probes;
81}
82
83static void rcu_free_old_probes(struct rcu_head *head)
84{
85 kfree(container_of(head, struct tp_probes, u.rcu));
86}
87
88static inline void release_probes(struct tracepoint_func *old)
89{
90 if (old) {
91 struct tp_probes *tp_probes = container_of(old,
92 struct tp_probes, probes[0]);
93 call_rcu_sched(&tp_probes->u.rcu, rcu_free_old_probes);
94 }
95}
96
97static void debug_print_probes(struct tracepoint_entry *entry)
98{
99 int i;
100
101 if (!tracepoint_debug || !entry->funcs)
102 return;
103
104 for (i = 0; entry->funcs[i].func; i++)
105 printk(KERN_DEBUG "Probe %d : %p\n", i, entry->funcs[i].func);
106}
107
108static struct tracepoint_func *
109tracepoint_entry_add_probe(struct tracepoint_entry *entry,
110 void *probe, void *data)
111{
112 int nr_probes = 0;
113 struct tracepoint_func *old, *new;
114
115 WARN_ON(!probe);
116
117 debug_print_probes(entry);
118 old = entry->funcs;
119 if (old) {
120 /* (N -> N+1), (N != 0, 1) probes */
121 for (nr_probes = 0; old[nr_probes].func; nr_probes++)
122 if (old[nr_probes].func == probe &&
123 old[nr_probes].data == data)
124 return ERR_PTR(-EEXIST);
125 }
126 /* + 2 : one for new probe, one for NULL func */
127 new = allocate_probes(nr_probes + 2);
128 if (new == NULL)
129 return ERR_PTR(-ENOMEM);
130 if (old)
131 memcpy(new, old, nr_probes * sizeof(struct tracepoint_func));
132 new[nr_probes].func = probe;
133 new[nr_probes].data = data;
134 new[nr_probes + 1].func = NULL;
135 entry->refcount = nr_probes + 1;
136 entry->funcs = new;
137 debug_print_probes(entry);
138 return old;
139}
140
141static void *
142tracepoint_entry_remove_probe(struct tracepoint_entry *entry,
143 void *probe, void *data)
144{
145 int nr_probes = 0, nr_del = 0, i;
146 struct tracepoint_func *old, *new;
147
148 old = entry->funcs;
149
150 if (!old)
151 return ERR_PTR(-ENOENT);
152
153 debug_print_probes(entry);
154 /* (N -> M), (N > 1, M >= 0) probes */
155 for (nr_probes = 0; old[nr_probes].func; nr_probes++) {
156 if (!probe ||
157 (old[nr_probes].func == probe &&
158 old[nr_probes].data == data))
159 nr_del++;
160 }
161
162 if (nr_probes - nr_del == 0) {
163 /* N -> 0, (N > 1) */
164 entry->funcs = NULL;
165 entry->refcount = 0;
166 debug_print_probes(entry);
167 return old;
168 } else {
169 int j = 0;
170 /* N -> M, (N > 1, M > 0) */
171 /* + 1 for NULL */
172 new = allocate_probes(nr_probes - nr_del + 1);
173 if (new == NULL)
174 return ERR_PTR(-ENOMEM);
175 for (i = 0; old[i].func; i++)
176 if (probe &&
177 (old[i].func != probe || old[i].data != data))
178 new[j++] = old[i];
179 new[nr_probes - nr_del].func = NULL;
180 entry->refcount = nr_probes - nr_del;
181 entry->funcs = new;
182 }
183 debug_print_probes(entry);
184 return old;
185}
186
187/*
188 * Get tracepoint if the tracepoint is present in the tracepoint hash table.
189 * Must be called with tracepoints_mutex held.
190 * Returns NULL if not present.
191 */
192static struct tracepoint_entry *get_tracepoint(const char *name)
193{
194 struct hlist_head *head;
195 struct hlist_node *node;
196 struct tracepoint_entry *e;
197 u32 hash = jhash(name, strlen(name), 0);
198
199 head = &tracepoint_table[hash & (TRACEPOINT_TABLE_SIZE - 1)];
200 hlist_for_each_entry(e, node, head, hlist) {
201 if (!strcmp(name, e->name))
202 return e;
203 }
204 return NULL;
205}
206
207/*
208 * Add the tracepoint to the tracepoint hash table. Must be called with
209 * tracepoints_mutex held.
210 */
211static struct tracepoint_entry *add_tracepoint(const char *name)
212{
213 struct hlist_head *head;
214 struct hlist_node *node;
215 struct tracepoint_entry *e;
216 size_t name_len = strlen(name) + 1;
217 u32 hash = jhash(name, name_len-1, 0);
218
219 head = &tracepoint_table[hash & (TRACEPOINT_TABLE_SIZE - 1)];
220 hlist_for_each_entry(e, node, head, hlist) {
221 if (!strcmp(name, e->name)) {
222 printk(KERN_NOTICE
223 "tracepoint %s busy\n", name);
224 return ERR_PTR(-EEXIST); /* Already there */
225 }
226 }
227 /*
228 * Using kmalloc here to allocate a variable length element. Could
229 * cause some memory fragmentation if overused.
230 */
231 e = kmalloc(sizeof(struct tracepoint_entry) + name_len, GFP_KERNEL);
232 if (!e)
233 return ERR_PTR(-ENOMEM);
234 memcpy(&e->name[0], name, name_len);
235 e->funcs = NULL;
236 e->refcount = 0;
237 hlist_add_head(&e->hlist, head);
238 return e;
239}
240
241/*
242 * Remove the tracepoint from the tracepoint hash table. Must be called with
243 * mutex_lock held.
244 */
245static inline void remove_tracepoint(struct tracepoint_entry *e)
246{
247 hlist_del(&e->hlist);
248 kfree(e);
249}
250
251/*
252 * Sets the probe callback corresponding to one tracepoint.
253 */
254static void set_tracepoint(struct tracepoint_entry **entry,
255 struct tracepoint *elem, int active)
256{
257 WARN_ON(strcmp((*entry)->name, elem->name) != 0);
258
259 if (elem->regfunc && !static_key_enabled(&elem->key) && active)
260 elem->regfunc();
261 else if (elem->unregfunc && static_key_enabled(&elem->key) && !active)
262 elem->unregfunc();
263
264 /*
265 * rcu_assign_pointer has a smp_wmb() which makes sure that the new
266 * probe callbacks array is consistent before setting a pointer to it.
267 * This array is referenced by __DO_TRACE from
268 * include/linux/tracepoints.h. A matching smp_read_barrier_depends()
269 * is used.
270 */
271 rcu_assign_pointer(elem->funcs, (*entry)->funcs);
272 if (active && !static_key_enabled(&elem->key))
273 static_key_slow_inc(&elem->key);
274 else if (!active && static_key_enabled(&elem->key))
275 static_key_slow_dec(&elem->key);
276}
277
278/*
279 * Disable a tracepoint and its probe callback.
280 * Note: only waiting an RCU period after setting elem->call to the empty
281 * function insures that the original callback is not used anymore. This insured
282 * by preempt_disable around the call site.
283 */
284static void disable_tracepoint(struct tracepoint *elem)
285{
286 if (elem->unregfunc && static_key_enabled(&elem->key))
287 elem->unregfunc();
288
289 if (static_key_enabled(&elem->key))
290 static_key_slow_dec(&elem->key);
291 rcu_assign_pointer(elem->funcs, NULL);
292}
293
294/**
295 * tracepoint_update_probe_range - Update a probe range
296 * @begin: beginning of the range
297 * @end: end of the range
298 *
299 * Updates the probe callback corresponding to a range of tracepoints.
300 * Called with tracepoints_mutex held.
301 */
302static void tracepoint_update_probe_range(struct tracepoint * const *begin,
303 struct tracepoint * const *end)
304{
305 struct tracepoint * const *iter;
306 struct tracepoint_entry *mark_entry;
307
308 if (!begin)
309 return;
310
311 for (iter = begin; iter < end; iter++) {
312 mark_entry = get_tracepoint((*iter)->name);
313 if (mark_entry) {
314 set_tracepoint(&mark_entry, *iter,
315 !!mark_entry->refcount);
316 } else {
317 disable_tracepoint(*iter);
318 }
319 }
320}
321
322#ifdef CONFIG_MODULES
323void module_update_tracepoints(void)
324{
325 struct tp_module *tp_mod;
326
327 list_for_each_entry(tp_mod, &tracepoint_module_list, list)
328 tracepoint_update_probe_range(tp_mod->tracepoints_ptrs,
329 tp_mod->tracepoints_ptrs + tp_mod->num_tracepoints);
330}
331#else /* CONFIG_MODULES */
332void module_update_tracepoints(void)
333{
334}
335#endif /* CONFIG_MODULES */
336
337
338/*
339 * Update probes, removing the faulty probes.
340 * Called with tracepoints_mutex held.
341 */
342static void tracepoint_update_probes(void)
343{
344 /* Core kernel tracepoints */
345 tracepoint_update_probe_range(__start___tracepoints_ptrs,
346 __stop___tracepoints_ptrs);
347 /* tracepoints in modules. */
348 module_update_tracepoints();
349}
350
351static struct tracepoint_func *
352tracepoint_add_probe(const char *name, void *probe, void *data)
353{
354 struct tracepoint_entry *entry;
355 struct tracepoint_func *old;
356
357 entry = get_tracepoint(name);
358 if (!entry) {
359 entry = add_tracepoint(name);
360 if (IS_ERR(entry))
361 return (struct tracepoint_func *)entry;
362 }
363 old = tracepoint_entry_add_probe(entry, probe, data);
364 if (IS_ERR(old) && !entry->refcount)
365 remove_tracepoint(entry);
366 return old;
367}
368
369/**
370 * tracepoint_probe_register - Connect a probe to a tracepoint
371 * @name: tracepoint name
372 * @probe: probe handler
373 *
374 * Returns 0 if ok, error value on error.
375 * The probe address must at least be aligned on the architecture pointer size.
376 */
377int tracepoint_probe_register(const char *name, void *probe, void *data)
378{
379 struct tracepoint_func *old;
380
381 mutex_lock(&tracepoints_mutex);
382 old = tracepoint_add_probe(name, probe, data);
383 if (IS_ERR(old)) {
384 mutex_unlock(&tracepoints_mutex);
385 return PTR_ERR(old);
386 }
387 tracepoint_update_probes(); /* may update entry */
388 mutex_unlock(&tracepoints_mutex);
389 release_probes(old);
390 return 0;
391}
392EXPORT_SYMBOL_GPL(tracepoint_probe_register);
393
394static struct tracepoint_func *
395tracepoint_remove_probe(const char *name, void *probe, void *data)
396{
397 struct tracepoint_entry *entry;
398 struct tracepoint_func *old;
399
400 entry = get_tracepoint(name);
401 if (!entry)
402 return ERR_PTR(-ENOENT);
403 old = tracepoint_entry_remove_probe(entry, probe, data);
404 if (IS_ERR(old))
405 return old;
406 if (!entry->refcount)
407 remove_tracepoint(entry);
408 return old;
409}
410
411/**
412 * tracepoint_probe_unregister - Disconnect a probe from a tracepoint
413 * @name: tracepoint name
414 * @probe: probe function pointer
415 *
416 * We do not need to call a synchronize_sched to make sure the probes have
417 * finished running before doing a module unload, because the module unload
418 * itself uses stop_machine(), which insures that every preempt disabled section
419 * have finished.
420 */
421int tracepoint_probe_unregister(const char *name, void *probe, void *data)
422{
423 struct tracepoint_func *old;
424
425 mutex_lock(&tracepoints_mutex);
426 old = tracepoint_remove_probe(name, probe, data);
427 if (IS_ERR(old)) {
428 mutex_unlock(&tracepoints_mutex);
429 return PTR_ERR(old);
430 }
431 tracepoint_update_probes(); /* may update entry */
432 mutex_unlock(&tracepoints_mutex);
433 release_probes(old);
434 return 0;
435}
436EXPORT_SYMBOL_GPL(tracepoint_probe_unregister);
437
438static LIST_HEAD(old_probes);
439static int need_update;
440
441static void tracepoint_add_old_probes(void *old)
442{
443 need_update = 1;
444 if (old) {
445 struct tp_probes *tp_probes = container_of(old,
446 struct tp_probes, probes[0]);
447 list_add(&tp_probes->u.list, &old_probes);
448 }
449}
450
451/**
452 * tracepoint_probe_register_noupdate - register a probe but not connect
453 * @name: tracepoint name
454 * @probe: probe handler
455 *
456 * caller must call tracepoint_probe_update_all()
457 */
458int tracepoint_probe_register_noupdate(const char *name, void *probe,
459 void *data)
460{
461 struct tracepoint_func *old;
462
463 mutex_lock(&tracepoints_mutex);
464 old = tracepoint_add_probe(name, probe, data);
465 if (IS_ERR(old)) {
466 mutex_unlock(&tracepoints_mutex);
467 return PTR_ERR(old);
468 }
469 tracepoint_add_old_probes(old);
470 mutex_unlock(&tracepoints_mutex);
471 return 0;
472}
473EXPORT_SYMBOL_GPL(tracepoint_probe_register_noupdate);
474
475/**
476 * tracepoint_probe_unregister_noupdate - remove a probe but not disconnect
477 * @name: tracepoint name
478 * @probe: probe function pointer
479 *
480 * caller must call tracepoint_probe_update_all()
481 */
482int tracepoint_probe_unregister_noupdate(const char *name, void *probe,
483 void *data)
484{
485 struct tracepoint_func *old;
486
487 mutex_lock(&tracepoints_mutex);
488 old = tracepoint_remove_probe(name, probe, data);
489 if (IS_ERR(old)) {
490 mutex_unlock(&tracepoints_mutex);
491 return PTR_ERR(old);
492 }
493 tracepoint_add_old_probes(old);
494 mutex_unlock(&tracepoints_mutex);
495 return 0;
496}
497EXPORT_SYMBOL_GPL(tracepoint_probe_unregister_noupdate);
498
499/**
500 * tracepoint_probe_update_all - update tracepoints
501 */
502void tracepoint_probe_update_all(void)
503{
504 LIST_HEAD(release_probes);
505 struct tp_probes *pos, *next;
506
507 mutex_lock(&tracepoints_mutex);
508 if (!need_update) {
509 mutex_unlock(&tracepoints_mutex);
510 return;
511 }
512 if (!list_empty(&old_probes))
513 list_replace_init(&old_probes, &release_probes);
514 need_update = 0;
515 tracepoint_update_probes();
516 mutex_unlock(&tracepoints_mutex);
517 list_for_each_entry_safe(pos, next, &release_probes, u.list) {
518 list_del(&pos->u.list);
519 call_rcu_sched(&pos->u.rcu, rcu_free_old_probes);
520 }
521}
522EXPORT_SYMBOL_GPL(tracepoint_probe_update_all);
523
524/**
525 * tracepoint_get_iter_range - Get a next tracepoint iterator given a range.
526 * @tracepoint: current tracepoints (in), next tracepoint (out)
527 * @begin: beginning of the range
528 * @end: end of the range
529 *
530 * Returns whether a next tracepoint has been found (1) or not (0).
531 * Will return the first tracepoint in the range if the input tracepoint is
532 * NULL.
533 */
534static int tracepoint_get_iter_range(struct tracepoint * const **tracepoint,
535 struct tracepoint * const *begin, struct tracepoint * const *end)
536{
537 if (!*tracepoint && begin != end) {
538 *tracepoint = begin;
539 return 1;
540 }
541 if (*tracepoint >= begin && *tracepoint < end)
542 return 1;
543 return 0;
544}
545
546#ifdef CONFIG_MODULES
547static void tracepoint_get_iter(struct tracepoint_iter *iter)
548{
549 int found = 0;
550 struct tp_module *iter_mod;
551
552 /* Core kernel tracepoints */
553 if (!iter->module) {
554 found = tracepoint_get_iter_range(&iter->tracepoint,
555 __start___tracepoints_ptrs,
556 __stop___tracepoints_ptrs);
557 if (found)
558 goto end;
559 }
560 /* Tracepoints in modules */
561 mutex_lock(&tracepoints_mutex);
562 list_for_each_entry(iter_mod, &tracepoint_module_list, list) {
563 /*
564 * Sorted module list
565 */
566 if (iter_mod < iter->module)
567 continue;
568 else if (iter_mod > iter->module)
569 iter->tracepoint = NULL;
570 found = tracepoint_get_iter_range(&iter->tracepoint,
571 iter_mod->tracepoints_ptrs,
572 iter_mod->tracepoints_ptrs
573 + iter_mod->num_tracepoints);
574 if (found) {
575 iter->module = iter_mod;
576 break;
577 }
578 }
579 mutex_unlock(&tracepoints_mutex);
580end:
581 if (!found)
582 tracepoint_iter_reset(iter);
583}
584#else /* CONFIG_MODULES */
585static void tracepoint_get_iter(struct tracepoint_iter *iter)
586{
587 int found = 0;
588
589 /* Core kernel tracepoints */
590 found = tracepoint_get_iter_range(&iter->tracepoint,
591 __start___tracepoints_ptrs,
592 __stop___tracepoints_ptrs);
593 if (!found)
594 tracepoint_iter_reset(iter);
595}
596#endif /* CONFIG_MODULES */
597
598void tracepoint_iter_start(struct tracepoint_iter *iter)
599{
600 tracepoint_get_iter(iter);
601}
602EXPORT_SYMBOL_GPL(tracepoint_iter_start);
603
604void tracepoint_iter_next(struct tracepoint_iter *iter)
605{
606 iter->tracepoint++;
607 /*
608 * iter->tracepoint may be invalid because we blindly incremented it.
609 * Make sure it is valid by marshalling on the tracepoints, getting the
610 * tracepoints from following modules if necessary.
611 */
612 tracepoint_get_iter(iter);
613}
614EXPORT_SYMBOL_GPL(tracepoint_iter_next);
615
616void tracepoint_iter_stop(struct tracepoint_iter *iter)
617{
618}
619EXPORT_SYMBOL_GPL(tracepoint_iter_stop);
620
621void tracepoint_iter_reset(struct tracepoint_iter *iter)
622{
623#ifdef CONFIG_MODULES
624 iter->module = NULL;
625#endif /* CONFIG_MODULES */
626 iter->tracepoint = NULL;
627}
628EXPORT_SYMBOL_GPL(tracepoint_iter_reset);
629
630#ifdef CONFIG_MODULES
631static int tracepoint_module_coming(struct module *mod)
632{
633 struct tp_module *tp_mod, *iter;
634 int ret = 0;
635
636 /*
637 * We skip modules that taint the kernel, especially those with different
638 * module headers (for forced load), to make sure we don't cause a crash.
639 * Staging and out-of-tree GPL modules are fine.
640 */
641 if (mod->taints & ~((1 << TAINT_OOT_MODULE) | (1 << TAINT_CRAP)))
642 return 0;
643 mutex_lock(&tracepoints_mutex);
644 tp_mod = kmalloc(sizeof(struct tp_module), GFP_KERNEL);
645 if (!tp_mod) {
646 ret = -ENOMEM;
647 goto end;
648 }
649 tp_mod->num_tracepoints = mod->num_tracepoints;
650 tp_mod->tracepoints_ptrs = mod->tracepoints_ptrs;
651
652 /*
653 * tracepoint_module_list is kept sorted by struct module pointer
654 * address for iteration on tracepoints from a seq_file that can release
655 * the mutex between calls.
656 */
657 list_for_each_entry_reverse(iter, &tracepoint_module_list, list) {
658 BUG_ON(iter == tp_mod); /* Should never be in the list twice */
659 if (iter < tp_mod) {
660 /* We belong to the location right after iter. */
661 list_add(&tp_mod->list, &iter->list);
662 goto module_added;
663 }
664 }
665 /* We belong to the beginning of the list */
666 list_add(&tp_mod->list, &tracepoint_module_list);
667module_added:
668 tracepoint_update_probe_range(mod->tracepoints_ptrs,
669 mod->tracepoints_ptrs + mod->num_tracepoints);
670end:
671 mutex_unlock(&tracepoints_mutex);
672 return ret;
673}
674
675static int tracepoint_module_going(struct module *mod)
676{
677 struct tp_module *pos;
678
679 mutex_lock(&tracepoints_mutex);
680 tracepoint_update_probe_range(mod->tracepoints_ptrs,
681 mod->tracepoints_ptrs + mod->num_tracepoints);
682 list_for_each_entry(pos, &tracepoint_module_list, list) {
683 if (pos->tracepoints_ptrs == mod->tracepoints_ptrs) {
684 list_del(&pos->list);
685 kfree(pos);
686 break;
687 }
688 }
689 /*
690 * In the case of modules that were tainted at "coming", we'll simply
691 * walk through the list without finding it. We cannot use the "tainted"
692 * flag on "going", in case a module taints the kernel only after being
693 * loaded.
694 */
695 mutex_unlock(&tracepoints_mutex);
696 return 0;
697}
698
699int tracepoint_module_notify(struct notifier_block *self,
700 unsigned long val, void *data)
701{
702 struct module *mod = data;
703 int ret = 0;
704
705 switch (val) {
706 case MODULE_STATE_COMING:
707 ret = tracepoint_module_coming(mod);
708 break;
709 case MODULE_STATE_LIVE:
710 break;
711 case MODULE_STATE_GOING:
712 ret = tracepoint_module_going(mod);
713 break;
714 }
715 return ret;
716}
717
718struct notifier_block tracepoint_module_nb = {
719 .notifier_call = tracepoint_module_notify,
720 .priority = 0,
721};
722
723static int init_tracepoints(void)
724{
725 return register_module_notifier(&tracepoint_module_nb);
726}
727__initcall(init_tracepoints);
728#endif /* CONFIG_MODULES */
729
730#ifdef CONFIG_HAVE_SYSCALL_TRACEPOINTS
731
732/* NB: reg/unreg are called while guarded with the tracepoints_mutex */
733static int sys_tracepoint_refcount;
734
735void syscall_regfunc(void)
736{
737 unsigned long flags;
738 struct task_struct *g, *t;
739
740 if (!sys_tracepoint_refcount) {
741 read_lock_irqsave(&tasklist_lock, flags);
742 do_each_thread(g, t) {
743 /* Skip kernel threads. */
744 if (t->mm)
745 set_tsk_thread_flag(t, TIF_SYSCALL_TRACEPOINT);
746 } while_each_thread(g, t);
747 read_unlock_irqrestore(&tasklist_lock, flags);
748 }
749 sys_tracepoint_refcount++;
750}
751
752void syscall_unregfunc(void)
753{
754 unsigned long flags;
755 struct task_struct *g, *t;
756
757 sys_tracepoint_refcount--;
758 if (!sys_tracepoint_refcount) {
759 read_lock_irqsave(&tasklist_lock, flags);
760 do_each_thread(g, t) {
761 clear_tsk_thread_flag(t, TIF_SYSCALL_TRACEPOINT);
762 } while_each_thread(g, t);
763 read_unlock_irqrestore(&tasklist_lock, flags);
764 }
765}
766#endif