2 * Kernel Probes (KProbes)
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write to the Free Software
17 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
19 * Copyright (C) IBM Corporation, 2002, 2004
21 * 2002-Oct Created by Vamsi Krishna S <vamsi_krishna@in.ibm.com> Kernel
22 * Probes initial implementation (includes suggestions from
24 * 2004-Aug Updated by Prasanna S Panchamukhi <prasanna@in.ibm.com> with
25 * hlists and exceptions notifier as suggested by Andi Kleen.
26 * 2004-July Suparna Bhattacharya <suparna@in.ibm.com> added jumper probes
27 * interface to access function arguments.
28 * 2004-Sep Prasanna S Panchamukhi <prasanna@in.ibm.com> Changed Kprobes
29 * exceptions notifier to be first on the priority list.
30 * 2005-May Hien Nguyen <hien@us.ibm.com>, Jim Keniston
31 * <jkenisto@us.ibm.com> and Prasanna S Panchamukhi
32 * <prasanna@in.ibm.com> added function-return probes.
34 #include <linux/kprobes.h>
35 #include <linux/hash.h>
36 #include <linux/init.h>
37 #include <linux/slab.h>
38 #include <linux/stddef.h>
39 #include <linux/export.h>
40 #include <linux/moduleloader.h>
41 #include <linux/kallsyms.h>
42 #include <linux/freezer.h>
43 #include <linux/seq_file.h>
44 #include <linux/debugfs.h>
45 #include <linux/sysctl.h>
46 #include <linux/kdebug.h>
47 #include <linux/memory.h>
48 #include <linux/ftrace.h>
49 #include <linux/cpu.h>
50 #include <linux/jump_label.h>
52 #include <asm/sections.h>
53 #include <asm/cacheflush.h>
54 #include <asm/errno.h>
55 #include <linux/uaccess.h>
57 #define KPROBE_HASH_BITS 6
58 #define KPROBE_TABLE_SIZE (1 << KPROBE_HASH_BITS)
61 static int kprobes_initialized
;
62 static struct hlist_head kprobe_table
[KPROBE_TABLE_SIZE
];
63 static struct hlist_head kretprobe_inst_table
[KPROBE_TABLE_SIZE
];
65 /* NOTE: change this value only with kprobe_mutex held */
66 static bool kprobes_all_disarmed
;
68 /* This protects kprobe_table and optimizing_list */
69 static DEFINE_MUTEX(kprobe_mutex
);
70 static DEFINE_PER_CPU(struct kprobe
*, kprobe_instance
) = NULL
;
72 raw_spinlock_t lock ____cacheline_aligned_in_smp
;
73 } kretprobe_table_locks
[KPROBE_TABLE_SIZE
];
75 kprobe_opcode_t
* __weak
kprobe_lookup_name(const char *name
,
76 unsigned int __unused
)
78 return ((kprobe_opcode_t
*)(kallsyms_lookup_name(name
)));
81 static raw_spinlock_t
*kretprobe_table_lock_ptr(unsigned long hash
)
83 return &(kretprobe_table_locks
[hash
].lock
);
86 /* Blacklist -- list of struct kprobe_blacklist_entry */
87 static LIST_HEAD(kprobe_blacklist
);
89 #ifdef __ARCH_WANT_KPROBES_INSN_SLOT
91 * kprobe->ainsn.insn points to the copy of the instruction to be
92 * single-stepped. x86_64, POWER4 and above have no-exec support and
93 * stepping on the instruction on a vmalloced/kmalloced/data page
94 * is a recipe for disaster
96 struct kprobe_insn_page
{
97 struct list_head list
;
98 kprobe_opcode_t
*insns
; /* Page of instruction slots */
99 struct kprobe_insn_cache
*cache
;
105 #define KPROBE_INSN_PAGE_SIZE(slots) \
106 (offsetof(struct kprobe_insn_page, slot_used) + \
107 (sizeof(char) * (slots)))
109 static int slots_per_page(struct kprobe_insn_cache
*c
)
111 return PAGE_SIZE
/(c
->insn_size
* sizeof(kprobe_opcode_t
));
114 enum kprobe_slot_state
{
120 static void *alloc_insn_page(void)
122 return module_alloc(PAGE_SIZE
);
125 void __weak
free_insn_page(void *page
)
127 module_memfree(page
);
130 struct kprobe_insn_cache kprobe_insn_slots
= {
131 .mutex
= __MUTEX_INITIALIZER(kprobe_insn_slots
.mutex
),
132 .alloc
= alloc_insn_page
,
133 .free
= free_insn_page
,
134 .pages
= LIST_HEAD_INIT(kprobe_insn_slots
.pages
),
135 .insn_size
= MAX_INSN_SIZE
,
138 static int collect_garbage_slots(struct kprobe_insn_cache
*c
);
141 * __get_insn_slot() - Find a slot on an executable page for an instruction.
142 * We allocate an executable page if there's no room on existing ones.
144 kprobe_opcode_t
*__get_insn_slot(struct kprobe_insn_cache
*c
)
146 struct kprobe_insn_page
*kip
;
147 kprobe_opcode_t
*slot
= NULL
;
149 /* Since the slot array is not protected by rcu, we need a mutex */
150 mutex_lock(&c
->mutex
);
153 list_for_each_entry_rcu(kip
, &c
->pages
, list
) {
154 if (kip
->nused
< slots_per_page(c
)) {
156 for (i
= 0; i
< slots_per_page(c
); i
++) {
157 if (kip
->slot_used
[i
] == SLOT_CLEAN
) {
158 kip
->slot_used
[i
] = SLOT_USED
;
160 slot
= kip
->insns
+ (i
* c
->insn_size
);
165 /* kip->nused is broken. Fix it. */
166 kip
->nused
= slots_per_page(c
);
172 /* If there are any garbage slots, collect it and try again. */
173 if (c
->nr_garbage
&& collect_garbage_slots(c
) == 0)
176 /* All out of space. Need to allocate a new page. */
177 kip
= kmalloc(KPROBE_INSN_PAGE_SIZE(slots_per_page(c
)), GFP_KERNEL
);
182 * Use module_alloc so this page is within +/- 2GB of where the
183 * kernel image and loaded module images reside. This is required
184 * so x86_64 can correctly handle the %rip-relative fixups.
186 kip
->insns
= c
->alloc();
191 INIT_LIST_HEAD(&kip
->list
);
192 memset(kip
->slot_used
, SLOT_CLEAN
, slots_per_page(c
));
193 kip
->slot_used
[0] = SLOT_USED
;
197 list_add_rcu(&kip
->list
, &c
->pages
);
200 mutex_unlock(&c
->mutex
);
204 /* Return 1 if all garbages are collected, otherwise 0. */
205 static int collect_one_slot(struct kprobe_insn_page
*kip
, int idx
)
207 kip
->slot_used
[idx
] = SLOT_CLEAN
;
209 if (kip
->nused
== 0) {
211 * Page is no longer in use. Free it unless
212 * it's the last one. We keep the last one
213 * so as not to have to set it up again the
214 * next time somebody inserts a probe.
216 if (!list_is_singular(&kip
->list
)) {
217 list_del_rcu(&kip
->list
);
219 kip
->cache
->free(kip
->insns
);
227 static int collect_garbage_slots(struct kprobe_insn_cache
*c
)
229 struct kprobe_insn_page
*kip
, *next
;
231 /* Ensure no-one is interrupted on the garbages */
234 list_for_each_entry_safe(kip
, next
, &c
->pages
, list
) {
236 if (kip
->ngarbage
== 0)
238 kip
->ngarbage
= 0; /* we will collect all garbages */
239 for (i
= 0; i
< slots_per_page(c
); i
++) {
240 if (kip
->slot_used
[i
] == SLOT_DIRTY
&& collect_one_slot(kip
, i
))
248 void __free_insn_slot(struct kprobe_insn_cache
*c
,
249 kprobe_opcode_t
*slot
, int dirty
)
251 struct kprobe_insn_page
*kip
;
254 mutex_lock(&c
->mutex
);
256 list_for_each_entry_rcu(kip
, &c
->pages
, list
) {
257 idx
= ((long)slot
- (long)kip
->insns
) /
258 (c
->insn_size
* sizeof(kprobe_opcode_t
));
259 if (idx
>= 0 && idx
< slots_per_page(c
))
262 /* Could not find this slot. */
267 /* Mark and sweep: this may sleep */
269 /* Check double free */
270 WARN_ON(kip
->slot_used
[idx
] != SLOT_USED
);
272 kip
->slot_used
[idx
] = SLOT_DIRTY
;
274 if (++c
->nr_garbage
> slots_per_page(c
))
275 collect_garbage_slots(c
);
277 collect_one_slot(kip
, idx
);
280 mutex_unlock(&c
->mutex
);
284 * Check given address is on the page of kprobe instruction slots.
285 * This will be used for checking whether the address on a stack
286 * is on a text area or not.
288 bool __is_insn_slot_addr(struct kprobe_insn_cache
*c
, unsigned long addr
)
290 struct kprobe_insn_page
*kip
;
294 list_for_each_entry_rcu(kip
, &c
->pages
, list
) {
295 if (addr
>= (unsigned long)kip
->insns
&&
296 addr
< (unsigned long)kip
->insns
+ PAGE_SIZE
) {
306 #ifdef CONFIG_OPTPROBES
307 /* For optimized_kprobe buffer */
308 struct kprobe_insn_cache kprobe_optinsn_slots
= {
309 .mutex
= __MUTEX_INITIALIZER(kprobe_optinsn_slots
.mutex
),
310 .alloc
= alloc_insn_page
,
311 .free
= free_insn_page
,
312 .pages
= LIST_HEAD_INIT(kprobe_optinsn_slots
.pages
),
313 /* .insn_size is initialized later */
319 /* We have preemption disabled.. so it is safe to use __ versions */
320 static inline void set_kprobe_instance(struct kprobe
*kp
)
322 __this_cpu_write(kprobe_instance
, kp
);
325 static inline void reset_kprobe_instance(void)
327 __this_cpu_write(kprobe_instance
, NULL
);
331 * This routine is called either:
332 * - under the kprobe_mutex - during kprobe_[un]register()
334 * - with preemption disabled - from arch/xxx/kernel/kprobes.c
336 struct kprobe
*get_kprobe(void *addr
)
338 struct hlist_head
*head
;
341 head
= &kprobe_table
[hash_ptr(addr
, KPROBE_HASH_BITS
)];
342 hlist_for_each_entry_rcu(p
, head
, hlist
) {
349 NOKPROBE_SYMBOL(get_kprobe
);
351 static int aggr_pre_handler(struct kprobe
*p
, struct pt_regs
*regs
);
353 /* Return true if the kprobe is an aggregator */
354 static inline int kprobe_aggrprobe(struct kprobe
*p
)
356 return p
->pre_handler
== aggr_pre_handler
;
359 /* Return true(!0) if the kprobe is unused */
360 static inline int kprobe_unused(struct kprobe
*p
)
362 return kprobe_aggrprobe(p
) && kprobe_disabled(p
) &&
363 list_empty(&p
->list
);
367 * Keep all fields in the kprobe consistent
369 static inline void copy_kprobe(struct kprobe
*ap
, struct kprobe
*p
)
371 memcpy(&p
->opcode
, &ap
->opcode
, sizeof(kprobe_opcode_t
));
372 memcpy(&p
->ainsn
, &ap
->ainsn
, sizeof(struct arch_specific_insn
));
375 #ifdef CONFIG_OPTPROBES
376 /* NOTE: change this value only with kprobe_mutex held */
377 static bool kprobes_allow_optimization
;
380 * Call all pre_handler on the list, but ignores its return value.
381 * This must be called from arch-dep optimized caller.
383 void opt_pre_handler(struct kprobe
*p
, struct pt_regs
*regs
)
387 list_for_each_entry_rcu(kp
, &p
->list
, list
) {
388 if (kp
->pre_handler
&& likely(!kprobe_disabled(kp
))) {
389 set_kprobe_instance(kp
);
390 kp
->pre_handler(kp
, regs
);
392 reset_kprobe_instance();
395 NOKPROBE_SYMBOL(opt_pre_handler
);
397 /* Free optimized instructions and optimized_kprobe */
398 static void free_aggr_kprobe(struct kprobe
*p
)
400 struct optimized_kprobe
*op
;
402 op
= container_of(p
, struct optimized_kprobe
, kp
);
403 arch_remove_optimized_kprobe(op
);
404 arch_remove_kprobe(p
);
408 /* Return true(!0) if the kprobe is ready for optimization. */
409 static inline int kprobe_optready(struct kprobe
*p
)
411 struct optimized_kprobe
*op
;
413 if (kprobe_aggrprobe(p
)) {
414 op
= container_of(p
, struct optimized_kprobe
, kp
);
415 return arch_prepared_optinsn(&op
->optinsn
);
421 /* Return true(!0) if the kprobe is disarmed. Note: p must be on hash list */
422 static inline int kprobe_disarmed(struct kprobe
*p
)
424 struct optimized_kprobe
*op
;
426 /* If kprobe is not aggr/opt probe, just return kprobe is disabled */
427 if (!kprobe_aggrprobe(p
))
428 return kprobe_disabled(p
);
430 op
= container_of(p
, struct optimized_kprobe
, kp
);
432 return kprobe_disabled(p
) && list_empty(&op
->list
);
435 /* Return true(!0) if the probe is queued on (un)optimizing lists */
436 static int kprobe_queued(struct kprobe
*p
)
438 struct optimized_kprobe
*op
;
440 if (kprobe_aggrprobe(p
)) {
441 op
= container_of(p
, struct optimized_kprobe
, kp
);
442 if (!list_empty(&op
->list
))
449 * Return an optimized kprobe whose optimizing code replaces
450 * instructions including addr (exclude breakpoint).
452 static struct kprobe
*get_optimized_kprobe(unsigned long addr
)
455 struct kprobe
*p
= NULL
;
456 struct optimized_kprobe
*op
;
458 /* Don't check i == 0, since that is a breakpoint case. */
459 for (i
= 1; !p
&& i
< MAX_OPTIMIZED_LENGTH
; i
++)
460 p
= get_kprobe((void *)(addr
- i
));
462 if (p
&& kprobe_optready(p
)) {
463 op
= container_of(p
, struct optimized_kprobe
, kp
);
464 if (arch_within_optimized_kprobe(op
, addr
))
471 /* Optimization staging list, protected by kprobe_mutex */
472 static LIST_HEAD(optimizing_list
);
473 static LIST_HEAD(unoptimizing_list
);
474 static LIST_HEAD(freeing_list
);
476 static void kprobe_optimizer(struct work_struct
*work
);
477 static DECLARE_DELAYED_WORK(optimizing_work
, kprobe_optimizer
);
478 #define OPTIMIZE_DELAY 5
481 * Optimize (replace a breakpoint with a jump) kprobes listed on
484 static void do_optimize_kprobes(void)
486 lockdep_assert_held(&text_mutex
);
488 * The optimization/unoptimization refers online_cpus via
489 * stop_machine() and cpu-hotplug modifies online_cpus.
490 * And same time, text_mutex will be held in cpu-hotplug and here.
491 * This combination can cause a deadlock (cpu-hotplug try to lock
492 * text_mutex but stop_machine can not be done because online_cpus
494 * To avoid this deadlock, caller must have locked cpu hotplug
495 * for preventing cpu-hotplug outside of text_mutex locking.
497 lockdep_assert_cpus_held();
499 /* Optimization never be done when disarmed */
500 if (kprobes_all_disarmed
|| !kprobes_allow_optimization
||
501 list_empty(&optimizing_list
))
504 arch_optimize_kprobes(&optimizing_list
);
508 * Unoptimize (replace a jump with a breakpoint and remove the breakpoint
509 * if need) kprobes listed on unoptimizing_list.
511 static void do_unoptimize_kprobes(void)
513 struct optimized_kprobe
*op
, *tmp
;
515 lockdep_assert_held(&text_mutex
);
516 /* See comment in do_optimize_kprobes() */
517 lockdep_assert_cpus_held();
519 /* Unoptimization must be done anytime */
520 if (list_empty(&unoptimizing_list
))
523 arch_unoptimize_kprobes(&unoptimizing_list
, &freeing_list
);
524 /* Loop free_list for disarming */
525 list_for_each_entry_safe(op
, tmp
, &freeing_list
, list
) {
526 /* Switching from detour code to origin */
527 op
->kp
.flags
&= ~KPROBE_FLAG_OPTIMIZED
;
528 /* Disarm probes if marked disabled */
529 if (kprobe_disabled(&op
->kp
))
530 arch_disarm_kprobe(&op
->kp
);
531 if (kprobe_unused(&op
->kp
)) {
533 * Remove unused probes from hash list. After waiting
534 * for synchronization, these probes are reclaimed.
535 * (reclaiming is done by do_free_cleaned_kprobes.)
537 hlist_del_rcu(&op
->kp
.hlist
);
539 list_del_init(&op
->list
);
543 /* Reclaim all kprobes on the free_list */
544 static void do_free_cleaned_kprobes(void)
546 struct optimized_kprobe
*op
, *tmp
;
548 list_for_each_entry_safe(op
, tmp
, &freeing_list
, list
) {
549 list_del_init(&op
->list
);
550 if (WARN_ON_ONCE(!kprobe_unused(&op
->kp
))) {
552 * This must not happen, but if there is a kprobe
553 * still in use, keep it on kprobes hash list.
557 free_aggr_kprobe(&op
->kp
);
561 /* Start optimizer after OPTIMIZE_DELAY passed */
562 static void kick_kprobe_optimizer(void)
564 schedule_delayed_work(&optimizing_work
, OPTIMIZE_DELAY
);
567 /* Kprobe jump optimizer */
568 static void kprobe_optimizer(struct work_struct
*work
)
570 mutex_lock(&kprobe_mutex
);
572 mutex_lock(&text_mutex
);
573 /* Lock modules while optimizing kprobes */
574 mutex_lock(&module_mutex
);
577 * Step 1: Unoptimize kprobes and collect cleaned (unused and disarmed)
578 * kprobes before waiting for quiesence period.
580 do_unoptimize_kprobes();
583 * Step 2: Wait for quiesence period to ensure all potentially
584 * preempted tasks to have normally scheduled. Because optprobe
585 * may modify multiple instructions, there is a chance that Nth
586 * instruction is preempted. In that case, such tasks can return
587 * to 2nd-Nth byte of jump instruction. This wait is for avoiding it.
588 * Note that on non-preemptive kernel, this is transparently converted
589 * to synchronoze_sched() to wait for all interrupts to have completed.
591 synchronize_rcu_tasks();
593 /* Step 3: Optimize kprobes after quiesence period */
594 do_optimize_kprobes();
596 /* Step 4: Free cleaned kprobes after quiesence period */
597 do_free_cleaned_kprobes();
599 mutex_unlock(&module_mutex
);
600 mutex_unlock(&text_mutex
);
602 mutex_unlock(&kprobe_mutex
);
604 /* Step 5: Kick optimizer again if needed */
605 if (!list_empty(&optimizing_list
) || !list_empty(&unoptimizing_list
))
606 kick_kprobe_optimizer();
609 /* Wait for completing optimization and unoptimization */
610 void wait_for_kprobe_optimizer(void)
612 mutex_lock(&kprobe_mutex
);
614 while (!list_empty(&optimizing_list
) || !list_empty(&unoptimizing_list
)) {
615 mutex_unlock(&kprobe_mutex
);
617 /* this will also make optimizing_work execute immmediately */
618 flush_delayed_work(&optimizing_work
);
619 /* @optimizing_work might not have been queued yet, relax */
622 mutex_lock(&kprobe_mutex
);
625 mutex_unlock(&kprobe_mutex
);
628 static bool optprobe_queued_unopt(struct optimized_kprobe
*op
)
630 struct optimized_kprobe
*_op
;
632 list_for_each_entry(_op
, &unoptimizing_list
, list
) {
640 /* Optimize kprobe if p is ready to be optimized */
641 static void optimize_kprobe(struct kprobe
*p
)
643 struct optimized_kprobe
*op
;
645 /* Check if the kprobe is disabled or not ready for optimization. */
646 if (!kprobe_optready(p
) || !kprobes_allow_optimization
||
647 (kprobe_disabled(p
) || kprobes_all_disarmed
))
650 /* Both of break_handler and post_handler are not supported. */
651 if (p
->break_handler
|| p
->post_handler
)
654 op
= container_of(p
, struct optimized_kprobe
, kp
);
656 /* Check there is no other kprobes at the optimized instructions */
657 if (arch_check_optimized_kprobe(op
) < 0)
660 /* Check if it is already optimized. */
661 if (op
->kp
.flags
& KPROBE_FLAG_OPTIMIZED
) {
662 if (optprobe_queued_unopt(op
)) {
663 /* This is under unoptimizing. Just dequeue the probe */
664 list_del_init(&op
->list
);
668 op
->kp
.flags
|= KPROBE_FLAG_OPTIMIZED
;
670 /* On unoptimizing/optimizing_list, op must have OPTIMIZED flag */
671 if (WARN_ON_ONCE(!list_empty(&op
->list
)))
674 list_add(&op
->list
, &optimizing_list
);
675 kick_kprobe_optimizer();
678 /* Short cut to direct unoptimizing */
679 static void force_unoptimize_kprobe(struct optimized_kprobe
*op
)
681 lockdep_assert_cpus_held();
682 arch_unoptimize_kprobe(op
);
683 op
->kp
.flags
&= ~KPROBE_FLAG_OPTIMIZED
;
684 if (kprobe_disabled(&op
->kp
))
685 arch_disarm_kprobe(&op
->kp
);
688 /* Unoptimize a kprobe if p is optimized */
689 static void unoptimize_kprobe(struct kprobe
*p
, bool force
)
691 struct optimized_kprobe
*op
;
693 if (!kprobe_aggrprobe(p
) || kprobe_disarmed(p
))
694 return; /* This is not an optprobe nor optimized */
696 op
= container_of(p
, struct optimized_kprobe
, kp
);
697 if (!kprobe_optimized(p
))
700 if (!list_empty(&op
->list
)) {
701 if (optprobe_queued_unopt(op
)) {
702 /* Queued in unoptimizing queue */
705 * Forcibly unoptimize the kprobe here, and queue it
706 * in the freeing list for release afterwards.
708 force_unoptimize_kprobe(op
);
709 list_move(&op
->list
, &freeing_list
);
712 /* Dequeue from the optimizing queue */
713 list_del_init(&op
->list
);
714 op
->kp
.flags
&= ~KPROBE_FLAG_OPTIMIZED
;
719 /* Optimized kprobe case */
721 /* Forcibly update the code: this is a special case */
722 force_unoptimize_kprobe(op
);
724 list_add(&op
->list
, &unoptimizing_list
);
725 kick_kprobe_optimizer();
729 /* Cancel unoptimizing for reusing */
730 static int reuse_unused_kprobe(struct kprobe
*ap
)
732 struct optimized_kprobe
*op
;
734 BUG_ON(!kprobe_unused(ap
));
736 * Unused kprobe MUST be on the way of delayed unoptimizing (means
737 * there is still a relative jump) and disabled.
739 op
= container_of(ap
, struct optimized_kprobe
, kp
);
740 if (unlikely(list_empty(&op
->list
)))
741 printk(KERN_WARNING
"Warning: found a stray unused "
742 "aggrprobe@%p\n", ap
->addr
);
743 /* Enable the probe again */
744 ap
->flags
&= ~KPROBE_FLAG_DISABLED
;
745 /* Optimize it again (remove from op->list) */
746 if (!kprobe_optready(ap
))
753 /* Remove optimized instructions */
754 static void kill_optimized_kprobe(struct kprobe
*p
)
756 struct optimized_kprobe
*op
;
758 op
= container_of(p
, struct optimized_kprobe
, kp
);
759 if (!list_empty(&op
->list
))
760 /* Dequeue from the (un)optimization queue */
761 list_del_init(&op
->list
);
762 op
->kp
.flags
&= ~KPROBE_FLAG_OPTIMIZED
;
764 if (kprobe_unused(p
)) {
765 /* Enqueue if it is unused */
766 list_add(&op
->list
, &freeing_list
);
768 * Remove unused probes from the hash list. After waiting
769 * for synchronization, this probe is reclaimed.
770 * (reclaiming is done by do_free_cleaned_kprobes().)
772 hlist_del_rcu(&op
->kp
.hlist
);
775 /* Don't touch the code, because it is already freed. */
776 arch_remove_optimized_kprobe(op
);
780 void __prepare_optimized_kprobe(struct optimized_kprobe
*op
, struct kprobe
*p
)
782 if (!kprobe_ftrace(p
))
783 arch_prepare_optimized_kprobe(op
, p
);
786 /* Try to prepare optimized instructions */
787 static void prepare_optimized_kprobe(struct kprobe
*p
)
789 struct optimized_kprobe
*op
;
791 op
= container_of(p
, struct optimized_kprobe
, kp
);
792 __prepare_optimized_kprobe(op
, p
);
795 /* Allocate new optimized_kprobe and try to prepare optimized instructions */
796 static struct kprobe
*alloc_aggr_kprobe(struct kprobe
*p
)
798 struct optimized_kprobe
*op
;
800 op
= kzalloc(sizeof(struct optimized_kprobe
), GFP_KERNEL
);
804 INIT_LIST_HEAD(&op
->list
);
805 op
->kp
.addr
= p
->addr
;
806 __prepare_optimized_kprobe(op
, p
);
811 static void init_aggr_kprobe(struct kprobe
*ap
, struct kprobe
*p
);
814 * Prepare an optimized_kprobe and optimize it
815 * NOTE: p must be a normal registered kprobe
817 static void try_to_optimize_kprobe(struct kprobe
*p
)
820 struct optimized_kprobe
*op
;
822 /* Impossible to optimize ftrace-based kprobe */
823 if (kprobe_ftrace(p
))
826 /* For preparing optimization, jump_label_text_reserved() is called */
829 mutex_lock(&text_mutex
);
831 ap
= alloc_aggr_kprobe(p
);
835 op
= container_of(ap
, struct optimized_kprobe
, kp
);
836 if (!arch_prepared_optinsn(&op
->optinsn
)) {
837 /* If failed to setup optimizing, fallback to kprobe */
838 arch_remove_optimized_kprobe(op
);
843 init_aggr_kprobe(ap
, p
);
844 optimize_kprobe(ap
); /* This just kicks optimizer thread */
847 mutex_unlock(&text_mutex
);
853 static void optimize_all_kprobes(void)
855 struct hlist_head
*head
;
859 mutex_lock(&kprobe_mutex
);
860 /* If optimization is already allowed, just return */
861 if (kprobes_allow_optimization
)
865 kprobes_allow_optimization
= true;
866 for (i
= 0; i
< KPROBE_TABLE_SIZE
; i
++) {
867 head
= &kprobe_table
[i
];
868 hlist_for_each_entry_rcu(p
, head
, hlist
)
869 if (!kprobe_disabled(p
))
873 printk(KERN_INFO
"Kprobes globally optimized\n");
875 mutex_unlock(&kprobe_mutex
);
878 static void unoptimize_all_kprobes(void)
880 struct hlist_head
*head
;
884 mutex_lock(&kprobe_mutex
);
885 /* If optimization is already prohibited, just return */
886 if (!kprobes_allow_optimization
) {
887 mutex_unlock(&kprobe_mutex
);
892 kprobes_allow_optimization
= false;
893 for (i
= 0; i
< KPROBE_TABLE_SIZE
; i
++) {
894 head
= &kprobe_table
[i
];
895 hlist_for_each_entry_rcu(p
, head
, hlist
) {
896 if (!kprobe_disabled(p
))
897 unoptimize_kprobe(p
, false);
901 mutex_unlock(&kprobe_mutex
);
903 /* Wait for unoptimizing completion */
904 wait_for_kprobe_optimizer();
905 printk(KERN_INFO
"Kprobes globally unoptimized\n");
908 static DEFINE_MUTEX(kprobe_sysctl_mutex
);
909 int sysctl_kprobes_optimization
;
910 int proc_kprobes_optimization_handler(struct ctl_table
*table
, int write
,
911 void __user
*buffer
, size_t *length
,
916 mutex_lock(&kprobe_sysctl_mutex
);
917 sysctl_kprobes_optimization
= kprobes_allow_optimization
? 1 : 0;
918 ret
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
920 if (sysctl_kprobes_optimization
)
921 optimize_all_kprobes();
923 unoptimize_all_kprobes();
924 mutex_unlock(&kprobe_sysctl_mutex
);
928 #endif /* CONFIG_SYSCTL */
930 /* Put a breakpoint for a probe. Must be called with text_mutex locked */
931 static void __arm_kprobe(struct kprobe
*p
)
935 /* Check collision with other optimized kprobes */
936 _p
= get_optimized_kprobe((unsigned long)p
->addr
);
938 /* Fallback to unoptimized kprobe */
939 unoptimize_kprobe(_p
, true);
942 optimize_kprobe(p
); /* Try to optimize (add kprobe to a list) */
945 /* Remove the breakpoint of a probe. Must be called with text_mutex locked */
946 static void __disarm_kprobe(struct kprobe
*p
, bool reopt
)
950 /* Try to unoptimize */
951 unoptimize_kprobe(p
, kprobes_all_disarmed
);
953 if (!kprobe_queued(p
)) {
954 arch_disarm_kprobe(p
);
955 /* If another kprobe was blocked, optimize it. */
956 _p
= get_optimized_kprobe((unsigned long)p
->addr
);
957 if (unlikely(_p
) && reopt
)
960 /* TODO: reoptimize others after unoptimized this probe */
963 #else /* !CONFIG_OPTPROBES */
965 #define optimize_kprobe(p) do {} while (0)
966 #define unoptimize_kprobe(p, f) do {} while (0)
967 #define kill_optimized_kprobe(p) do {} while (0)
968 #define prepare_optimized_kprobe(p) do {} while (0)
969 #define try_to_optimize_kprobe(p) do {} while (0)
970 #define __arm_kprobe(p) arch_arm_kprobe(p)
971 #define __disarm_kprobe(p, o) arch_disarm_kprobe(p)
972 #define kprobe_disarmed(p) kprobe_disabled(p)
973 #define wait_for_kprobe_optimizer() do {} while (0)
975 static int reuse_unused_kprobe(struct kprobe
*ap
)
978 * If the optimized kprobe is NOT supported, the aggr kprobe is
979 * released at the same time that the last aggregated kprobe is
981 * Thus there should be no chance to reuse unused kprobe.
983 printk(KERN_ERR
"Error: There should be no unused kprobe here.\n");
987 static void free_aggr_kprobe(struct kprobe
*p
)
989 arch_remove_kprobe(p
);
993 static struct kprobe
*alloc_aggr_kprobe(struct kprobe
*p
)
995 return kzalloc(sizeof(struct kprobe
), GFP_KERNEL
);
997 #endif /* CONFIG_OPTPROBES */
999 #ifdef CONFIG_KPROBES_ON_FTRACE
1000 static struct ftrace_ops kprobe_ftrace_ops __read_mostly
= {
1001 .func
= kprobe_ftrace_handler
,
1002 .flags
= FTRACE_OPS_FL_SAVE_REGS
| FTRACE_OPS_FL_IPMODIFY
,
1004 static int kprobe_ftrace_enabled
;
1006 /* Must ensure p->addr is really on ftrace */
1007 static int prepare_kprobe(struct kprobe
*p
)
1009 if (!kprobe_ftrace(p
))
1010 return arch_prepare_kprobe(p
);
1012 return arch_prepare_kprobe_ftrace(p
);
1015 /* Caller must lock kprobe_mutex */
1016 static void arm_kprobe_ftrace(struct kprobe
*p
)
1020 ret
= ftrace_set_filter_ip(&kprobe_ftrace_ops
,
1021 (unsigned long)p
->addr
, 0, 0);
1022 WARN(ret
< 0, "Failed to arm kprobe-ftrace at %p (%d)\n", p
->addr
, ret
);
1023 kprobe_ftrace_enabled
++;
1024 if (kprobe_ftrace_enabled
== 1) {
1025 ret
= register_ftrace_function(&kprobe_ftrace_ops
);
1026 WARN(ret
< 0, "Failed to init kprobe-ftrace (%d)\n", ret
);
1030 /* Caller must lock kprobe_mutex */
1031 static void disarm_kprobe_ftrace(struct kprobe
*p
)
1035 kprobe_ftrace_enabled
--;
1036 if (kprobe_ftrace_enabled
== 0) {
1037 ret
= unregister_ftrace_function(&kprobe_ftrace_ops
);
1038 WARN(ret
< 0, "Failed to init kprobe-ftrace (%d)\n", ret
);
1040 ret
= ftrace_set_filter_ip(&kprobe_ftrace_ops
,
1041 (unsigned long)p
->addr
, 1, 0);
1042 WARN(ret
< 0, "Failed to disarm kprobe-ftrace at %p (%d)\n", p
->addr
, ret
);
1044 #else /* !CONFIG_KPROBES_ON_FTRACE */
1045 #define prepare_kprobe(p) arch_prepare_kprobe(p)
1046 #define arm_kprobe_ftrace(p) do {} while (0)
1047 #define disarm_kprobe_ftrace(p) do {} while (0)
1050 /* Arm a kprobe with text_mutex */
1051 static void arm_kprobe(struct kprobe
*kp
)
1053 if (unlikely(kprobe_ftrace(kp
))) {
1054 arm_kprobe_ftrace(kp
);
1058 mutex_lock(&text_mutex
);
1060 mutex_unlock(&text_mutex
);
1064 /* Disarm a kprobe with text_mutex */
1065 static void disarm_kprobe(struct kprobe
*kp
, bool reopt
)
1067 if (unlikely(kprobe_ftrace(kp
))) {
1068 disarm_kprobe_ftrace(kp
);
1073 mutex_lock(&text_mutex
);
1074 __disarm_kprobe(kp
, reopt
);
1075 mutex_unlock(&text_mutex
);
1080 * Aggregate handlers for multiple kprobes support - these handlers
1081 * take care of invoking the individual kprobe handlers on p->list
1083 static int aggr_pre_handler(struct kprobe
*p
, struct pt_regs
*regs
)
1087 list_for_each_entry_rcu(kp
, &p
->list
, list
) {
1088 if (kp
->pre_handler
&& likely(!kprobe_disabled(kp
))) {
1089 set_kprobe_instance(kp
);
1090 if (kp
->pre_handler(kp
, regs
))
1093 reset_kprobe_instance();
1097 NOKPROBE_SYMBOL(aggr_pre_handler
);
1099 static void aggr_post_handler(struct kprobe
*p
, struct pt_regs
*regs
,
1100 unsigned long flags
)
1104 list_for_each_entry_rcu(kp
, &p
->list
, list
) {
1105 if (kp
->post_handler
&& likely(!kprobe_disabled(kp
))) {
1106 set_kprobe_instance(kp
);
1107 kp
->post_handler(kp
, regs
, flags
);
1108 reset_kprobe_instance();
1112 NOKPROBE_SYMBOL(aggr_post_handler
);
1114 static int aggr_fault_handler(struct kprobe
*p
, struct pt_regs
*regs
,
1117 struct kprobe
*cur
= __this_cpu_read(kprobe_instance
);
1120 * if we faulted "during" the execution of a user specified
1121 * probe handler, invoke just that probe's fault handler
1123 if (cur
&& cur
->fault_handler
) {
1124 if (cur
->fault_handler(cur
, regs
, trapnr
))
1129 NOKPROBE_SYMBOL(aggr_fault_handler
);
1131 static int aggr_break_handler(struct kprobe
*p
, struct pt_regs
*regs
)
1133 struct kprobe
*cur
= __this_cpu_read(kprobe_instance
);
1136 if (cur
&& cur
->break_handler
) {
1137 if (cur
->break_handler(cur
, regs
))
1140 reset_kprobe_instance();
1143 NOKPROBE_SYMBOL(aggr_break_handler
);
1145 /* Walks the list and increments nmissed count for multiprobe case */
1146 void kprobes_inc_nmissed_count(struct kprobe
*p
)
1149 if (!kprobe_aggrprobe(p
)) {
1152 list_for_each_entry_rcu(kp
, &p
->list
, list
)
1157 NOKPROBE_SYMBOL(kprobes_inc_nmissed_count
);
1159 void recycle_rp_inst(struct kretprobe_instance
*ri
,
1160 struct hlist_head
*head
)
1162 struct kretprobe
*rp
= ri
->rp
;
1164 /* remove rp inst off the rprobe_inst_table */
1165 hlist_del(&ri
->hlist
);
1166 INIT_HLIST_NODE(&ri
->hlist
);
1168 raw_spin_lock(&rp
->lock
);
1169 hlist_add_head(&ri
->hlist
, &rp
->free_instances
);
1170 raw_spin_unlock(&rp
->lock
);
1173 hlist_add_head(&ri
->hlist
, head
);
1175 NOKPROBE_SYMBOL(recycle_rp_inst
);
1177 void kretprobe_hash_lock(struct task_struct
*tsk
,
1178 struct hlist_head
**head
, unsigned long *flags
)
1179 __acquires(hlist_lock
)
1181 unsigned long hash
= hash_ptr(tsk
, KPROBE_HASH_BITS
);
1182 raw_spinlock_t
*hlist_lock
;
1184 *head
= &kretprobe_inst_table
[hash
];
1185 hlist_lock
= kretprobe_table_lock_ptr(hash
);
1186 raw_spin_lock_irqsave(hlist_lock
, *flags
);
1188 NOKPROBE_SYMBOL(kretprobe_hash_lock
);
1190 static void kretprobe_table_lock(unsigned long hash
,
1191 unsigned long *flags
)
1192 __acquires(hlist_lock
)
1194 raw_spinlock_t
*hlist_lock
= kretprobe_table_lock_ptr(hash
);
1195 raw_spin_lock_irqsave(hlist_lock
, *flags
);
1197 NOKPROBE_SYMBOL(kretprobe_table_lock
);
1199 void kretprobe_hash_unlock(struct task_struct
*tsk
,
1200 unsigned long *flags
)
1201 __releases(hlist_lock
)
1203 unsigned long hash
= hash_ptr(tsk
, KPROBE_HASH_BITS
);
1204 raw_spinlock_t
*hlist_lock
;
1206 hlist_lock
= kretprobe_table_lock_ptr(hash
);
1207 raw_spin_unlock_irqrestore(hlist_lock
, *flags
);
1209 NOKPROBE_SYMBOL(kretprobe_hash_unlock
);
1211 static void kretprobe_table_unlock(unsigned long hash
,
1212 unsigned long *flags
)
1213 __releases(hlist_lock
)
1215 raw_spinlock_t
*hlist_lock
= kretprobe_table_lock_ptr(hash
);
1216 raw_spin_unlock_irqrestore(hlist_lock
, *flags
);
1218 NOKPROBE_SYMBOL(kretprobe_table_unlock
);
1221 * This function is called from finish_task_switch when task tk becomes dead,
1222 * so that we can recycle any function-return probe instances associated
1223 * with this task. These left over instances represent probed functions
1224 * that have been called but will never return.
1226 void kprobe_flush_task(struct task_struct
*tk
)
1228 struct kretprobe_instance
*ri
;
1229 struct hlist_head
*head
, empty_rp
;
1230 struct hlist_node
*tmp
;
1231 unsigned long hash
, flags
= 0;
1233 if (unlikely(!kprobes_initialized
))
1234 /* Early boot. kretprobe_table_locks not yet initialized. */
1237 INIT_HLIST_HEAD(&empty_rp
);
1238 hash
= hash_ptr(tk
, KPROBE_HASH_BITS
);
1239 head
= &kretprobe_inst_table
[hash
];
1240 kretprobe_table_lock(hash
, &flags
);
1241 hlist_for_each_entry_safe(ri
, tmp
, head
, hlist
) {
1243 recycle_rp_inst(ri
, &empty_rp
);
1245 kretprobe_table_unlock(hash
, &flags
);
1246 hlist_for_each_entry_safe(ri
, tmp
, &empty_rp
, hlist
) {
1247 hlist_del(&ri
->hlist
);
1251 NOKPROBE_SYMBOL(kprobe_flush_task
);
1253 static inline void free_rp_inst(struct kretprobe
*rp
)
1255 struct kretprobe_instance
*ri
;
1256 struct hlist_node
*next
;
1258 hlist_for_each_entry_safe(ri
, next
, &rp
->free_instances
, hlist
) {
1259 hlist_del(&ri
->hlist
);
1264 static void cleanup_rp_inst(struct kretprobe
*rp
)
1266 unsigned long flags
, hash
;
1267 struct kretprobe_instance
*ri
;
1268 struct hlist_node
*next
;
1269 struct hlist_head
*head
;
1272 for (hash
= 0; hash
< KPROBE_TABLE_SIZE
; hash
++) {
1273 kretprobe_table_lock(hash
, &flags
);
1274 head
= &kretprobe_inst_table
[hash
];
1275 hlist_for_each_entry_safe(ri
, next
, head
, hlist
) {
1279 kretprobe_table_unlock(hash
, &flags
);
1283 NOKPROBE_SYMBOL(cleanup_rp_inst
);
1286 * Add the new probe to ap->list. Fail if this is the
1287 * second jprobe at the address - two jprobes can't coexist
1289 static int add_new_kprobe(struct kprobe
*ap
, struct kprobe
*p
)
1291 BUG_ON(kprobe_gone(ap
) || kprobe_gone(p
));
1293 if (p
->break_handler
|| p
->post_handler
)
1294 unoptimize_kprobe(ap
, true); /* Fall back to normal kprobe */
1296 if (p
->break_handler
) {
1297 if (ap
->break_handler
)
1299 list_add_tail_rcu(&p
->list
, &ap
->list
);
1300 ap
->break_handler
= aggr_break_handler
;
1302 list_add_rcu(&p
->list
, &ap
->list
);
1303 if (p
->post_handler
&& !ap
->post_handler
)
1304 ap
->post_handler
= aggr_post_handler
;
1310 * Fill in the required fields of the "manager kprobe". Replace the
1311 * earlier kprobe in the hlist with the manager kprobe
1313 static void init_aggr_kprobe(struct kprobe
*ap
, struct kprobe
*p
)
1315 /* Copy p's insn slot to ap */
1317 flush_insn_slot(ap
);
1319 ap
->flags
= p
->flags
& ~KPROBE_FLAG_OPTIMIZED
;
1320 ap
->pre_handler
= aggr_pre_handler
;
1321 ap
->fault_handler
= aggr_fault_handler
;
1322 /* We don't care the kprobe which has gone. */
1323 if (p
->post_handler
&& !kprobe_gone(p
))
1324 ap
->post_handler
= aggr_post_handler
;
1325 if (p
->break_handler
&& !kprobe_gone(p
))
1326 ap
->break_handler
= aggr_break_handler
;
1328 INIT_LIST_HEAD(&ap
->list
);
1329 INIT_HLIST_NODE(&ap
->hlist
);
1331 list_add_rcu(&p
->list
, &ap
->list
);
1332 hlist_replace_rcu(&p
->hlist
, &ap
->hlist
);
1336 * This is the second or subsequent kprobe at the address - handle
1339 static int register_aggr_kprobe(struct kprobe
*orig_p
, struct kprobe
*p
)
1342 struct kprobe
*ap
= orig_p
;
1346 /* For preparing optimization, jump_label_text_reserved() is called */
1348 mutex_lock(&text_mutex
);
1350 if (!kprobe_aggrprobe(orig_p
)) {
1351 /* If orig_p is not an aggr_kprobe, create new aggr_kprobe. */
1352 ap
= alloc_aggr_kprobe(orig_p
);
1357 init_aggr_kprobe(ap
, orig_p
);
1358 } else if (kprobe_unused(ap
)) {
1359 /* This probe is going to die. Rescue it */
1360 ret
= reuse_unused_kprobe(ap
);
1365 if (kprobe_gone(ap
)) {
1367 * Attempting to insert new probe at the same location that
1368 * had a probe in the module vaddr area which already
1369 * freed. So, the instruction slot has already been
1370 * released. We need a new slot for the new probe.
1372 ret
= arch_prepare_kprobe(ap
);
1375 * Even if fail to allocate new slot, don't need to
1376 * free aggr_probe. It will be used next time, or
1377 * freed by unregister_kprobe.
1381 /* Prepare optimized instructions if possible. */
1382 prepare_optimized_kprobe(ap
);
1385 * Clear gone flag to prevent allocating new slot again, and
1386 * set disabled flag because it is not armed yet.
1388 ap
->flags
= (ap
->flags
& ~KPROBE_FLAG_GONE
)
1389 | KPROBE_FLAG_DISABLED
;
1392 /* Copy ap's insn slot to p */
1394 ret
= add_new_kprobe(ap
, p
);
1397 mutex_unlock(&text_mutex
);
1398 jump_label_unlock();
1401 if (ret
== 0 && kprobe_disabled(ap
) && !kprobe_disabled(p
)) {
1402 ap
->flags
&= ~KPROBE_FLAG_DISABLED
;
1403 if (!kprobes_all_disarmed
)
1404 /* Arm the breakpoint again. */
1410 bool __weak
arch_within_kprobe_blacklist(unsigned long addr
)
1412 /* The __kprobes marked functions and entry code must not be probed */
1413 return addr
>= (unsigned long)__kprobes_text_start
&&
1414 addr
< (unsigned long)__kprobes_text_end
;
1417 bool within_kprobe_blacklist(unsigned long addr
)
1419 struct kprobe_blacklist_entry
*ent
;
1421 if (arch_within_kprobe_blacklist(addr
))
1424 * If there exists a kprobe_blacklist, verify and
1425 * fail any probe registration in the prohibited area
1427 list_for_each_entry(ent
, &kprobe_blacklist
, list
) {
1428 if (addr
>= ent
->start_addr
&& addr
< ent
->end_addr
)
1436 * If we have a symbol_name argument, look it up and add the offset field
1437 * to it. This way, we can specify a relative address to a symbol.
1438 * This returns encoded errors if it fails to look up symbol or invalid
1439 * combination of parameters.
1441 static kprobe_opcode_t
*_kprobe_addr(kprobe_opcode_t
*addr
,
1442 const char *symbol_name
, unsigned int offset
)
1444 if ((symbol_name
&& addr
) || (!symbol_name
&& !addr
))
1448 addr
= kprobe_lookup_name(symbol_name
, offset
);
1450 return ERR_PTR(-ENOENT
);
1453 addr
= (kprobe_opcode_t
*)(((char *)addr
) + offset
);
1458 return ERR_PTR(-EINVAL
);
1461 static kprobe_opcode_t
*kprobe_addr(struct kprobe
*p
)
1463 return _kprobe_addr(p
->addr
, p
->symbol_name
, p
->offset
);
1466 /* Check passed kprobe is valid and return kprobe in kprobe_table. */
1467 static struct kprobe
*__get_valid_kprobe(struct kprobe
*p
)
1469 struct kprobe
*ap
, *list_p
;
1471 ap
= get_kprobe(p
->addr
);
1476 list_for_each_entry_rcu(list_p
, &ap
->list
, list
)
1478 /* kprobe p is a valid probe */
1486 /* Return error if the kprobe is being re-registered */
1487 static inline int check_kprobe_rereg(struct kprobe
*p
)
1491 mutex_lock(&kprobe_mutex
);
1492 if (__get_valid_kprobe(p
))
1494 mutex_unlock(&kprobe_mutex
);
1499 int __weak
arch_check_ftrace_location(struct kprobe
*p
)
1501 unsigned long ftrace_addr
;
1503 ftrace_addr
= ftrace_location((unsigned long)p
->addr
);
1505 #ifdef CONFIG_KPROBES_ON_FTRACE
1506 /* Given address is not on the instruction boundary */
1507 if ((unsigned long)p
->addr
!= ftrace_addr
)
1509 p
->flags
|= KPROBE_FLAG_FTRACE
;
1510 #else /* !CONFIG_KPROBES_ON_FTRACE */
1517 static int check_kprobe_address_safe(struct kprobe
*p
,
1518 struct module
**probed_mod
)
1522 ret
= arch_check_ftrace_location(p
);
1528 /* Ensure it is not in reserved area nor out of text */
1529 if (!kernel_text_address((unsigned long) p
->addr
) ||
1530 within_kprobe_blacklist((unsigned long) p
->addr
) ||
1531 jump_label_text_reserved(p
->addr
, p
->addr
) ||
1532 find_bug((unsigned long)p
->addr
)) {
1537 /* Check if are we probing a module */
1538 *probed_mod
= __module_text_address((unsigned long) p
->addr
);
1541 * We must hold a refcount of the probed module while updating
1542 * its code to prohibit unexpected unloading.
1544 if (unlikely(!try_module_get(*probed_mod
))) {
1550 * If the module freed .init.text, we couldn't insert
1553 if (within_module_init((unsigned long)p
->addr
, *probed_mod
) &&
1554 (*probed_mod
)->state
!= MODULE_STATE_COMING
) {
1555 module_put(*probed_mod
);
1562 jump_label_unlock();
1567 int register_kprobe(struct kprobe
*p
)
1570 struct kprobe
*old_p
;
1571 struct module
*probed_mod
;
1572 kprobe_opcode_t
*addr
;
1574 /* Adjust probe address from symbol */
1575 addr
= kprobe_addr(p
);
1577 return PTR_ERR(addr
);
1580 ret
= check_kprobe_rereg(p
);
1584 /* User can pass only KPROBE_FLAG_DISABLED to register_kprobe */
1585 p
->flags
&= KPROBE_FLAG_DISABLED
;
1587 INIT_LIST_HEAD(&p
->list
);
1589 ret
= check_kprobe_address_safe(p
, &probed_mod
);
1593 mutex_lock(&kprobe_mutex
);
1595 old_p
= get_kprobe(p
->addr
);
1597 /* Since this may unoptimize old_p, locking text_mutex. */
1598 ret
= register_aggr_kprobe(old_p
, p
);
1603 /* Prevent text modification */
1604 mutex_lock(&text_mutex
);
1605 ret
= prepare_kprobe(p
);
1606 mutex_unlock(&text_mutex
);
1611 INIT_HLIST_NODE(&p
->hlist
);
1612 hlist_add_head_rcu(&p
->hlist
,
1613 &kprobe_table
[hash_ptr(p
->addr
, KPROBE_HASH_BITS
)]);
1615 if (!kprobes_all_disarmed
&& !kprobe_disabled(p
))
1618 /* Try to optimize kprobe */
1619 try_to_optimize_kprobe(p
);
1621 mutex_unlock(&kprobe_mutex
);
1624 module_put(probed_mod
);
1628 EXPORT_SYMBOL_GPL(register_kprobe
);
1630 /* Check if all probes on the aggrprobe are disabled */
1631 static int aggr_kprobe_disabled(struct kprobe
*ap
)
1635 list_for_each_entry_rcu(kp
, &ap
->list
, list
)
1636 if (!kprobe_disabled(kp
))
1638 * There is an active probe on the list.
1639 * We can't disable this ap.
1646 /* Disable one kprobe: Make sure called under kprobe_mutex is locked */
1647 static struct kprobe
*__disable_kprobe(struct kprobe
*p
)
1649 struct kprobe
*orig_p
;
1651 /* Get an original kprobe for return */
1652 orig_p
= __get_valid_kprobe(p
);
1653 if (unlikely(orig_p
== NULL
))
1656 if (!kprobe_disabled(p
)) {
1657 /* Disable probe if it is a child probe */
1659 p
->flags
|= KPROBE_FLAG_DISABLED
;
1661 /* Try to disarm and disable this/parent probe */
1662 if (p
== orig_p
|| aggr_kprobe_disabled(orig_p
)) {
1664 * If kprobes_all_disarmed is set, orig_p
1665 * should have already been disarmed, so
1666 * skip unneed disarming process.
1668 if (!kprobes_all_disarmed
)
1669 disarm_kprobe(orig_p
, true);
1670 orig_p
->flags
|= KPROBE_FLAG_DISABLED
;
1678 * Unregister a kprobe without a scheduler synchronization.
1680 static int __unregister_kprobe_top(struct kprobe
*p
)
1682 struct kprobe
*ap
, *list_p
;
1684 /* Disable kprobe. This will disarm it if needed. */
1685 ap
= __disable_kprobe(p
);
1691 * This probe is an independent(and non-optimized) kprobe
1692 * (not an aggrprobe). Remove from the hash list.
1696 /* Following process expects this probe is an aggrprobe */
1697 WARN_ON(!kprobe_aggrprobe(ap
));
1699 if (list_is_singular(&ap
->list
) && kprobe_disarmed(ap
))
1701 * !disarmed could be happen if the probe is under delayed
1706 /* If disabling probe has special handlers, update aggrprobe */
1707 if (p
->break_handler
&& !kprobe_gone(p
))
1708 ap
->break_handler
= NULL
;
1709 if (p
->post_handler
&& !kprobe_gone(p
)) {
1710 list_for_each_entry_rcu(list_p
, &ap
->list
, list
) {
1711 if ((list_p
!= p
) && (list_p
->post_handler
))
1714 ap
->post_handler
= NULL
;
1718 * Remove from the aggrprobe: this path will do nothing in
1719 * __unregister_kprobe_bottom().
1721 list_del_rcu(&p
->list
);
1722 if (!kprobe_disabled(ap
) && !kprobes_all_disarmed
)
1724 * Try to optimize this probe again, because post
1725 * handler may have been changed.
1727 optimize_kprobe(ap
);
1732 BUG_ON(!kprobe_disarmed(ap
));
1733 hlist_del_rcu(&ap
->hlist
);
1737 static void __unregister_kprobe_bottom(struct kprobe
*p
)
1741 if (list_empty(&p
->list
))
1742 /* This is an independent kprobe */
1743 arch_remove_kprobe(p
);
1744 else if (list_is_singular(&p
->list
)) {
1745 /* This is the last child of an aggrprobe */
1746 ap
= list_entry(p
->list
.next
, struct kprobe
, list
);
1748 free_aggr_kprobe(ap
);
1750 /* Otherwise, do nothing. */
1753 int register_kprobes(struct kprobe
**kps
, int num
)
1759 for (i
= 0; i
< num
; i
++) {
1760 ret
= register_kprobe(kps
[i
]);
1763 unregister_kprobes(kps
, i
);
1769 EXPORT_SYMBOL_GPL(register_kprobes
);
1771 void unregister_kprobe(struct kprobe
*p
)
1773 unregister_kprobes(&p
, 1);
1775 EXPORT_SYMBOL_GPL(unregister_kprobe
);
1777 void unregister_kprobes(struct kprobe
**kps
, int num
)
1783 mutex_lock(&kprobe_mutex
);
1784 for (i
= 0; i
< num
; i
++)
1785 if (__unregister_kprobe_top(kps
[i
]) < 0)
1786 kps
[i
]->addr
= NULL
;
1787 mutex_unlock(&kprobe_mutex
);
1789 synchronize_sched();
1790 for (i
= 0; i
< num
; i
++)
1792 __unregister_kprobe_bottom(kps
[i
]);
1794 EXPORT_SYMBOL_GPL(unregister_kprobes
);
1796 int __weak
kprobe_exceptions_notify(struct notifier_block
*self
,
1797 unsigned long val
, void *data
)
1801 NOKPROBE_SYMBOL(kprobe_exceptions_notify
);
1803 static struct notifier_block kprobe_exceptions_nb
= {
1804 .notifier_call
= kprobe_exceptions_notify
,
1805 .priority
= 0x7fffffff /* we need to be notified first */
1808 unsigned long __weak
arch_deref_entry_point(void *entry
)
1810 return (unsigned long)entry
;
1813 int register_jprobes(struct jprobe
**jps
, int num
)
1820 for (i
= 0; i
< num
; i
++) {
1821 ret
= register_jprobe(jps
[i
]);
1825 unregister_jprobes(jps
, i
);
1832 EXPORT_SYMBOL_GPL(register_jprobes
);
1834 int register_jprobe(struct jprobe
*jp
)
1836 unsigned long addr
, offset
;
1837 struct kprobe
*kp
= &jp
->kp
;
1840 * Verify probepoint as well as the jprobe handler are
1841 * valid function entry points.
1843 addr
= arch_deref_entry_point(jp
->entry
);
1845 if (kallsyms_lookup_size_offset(addr
, NULL
, &offset
) && offset
== 0 &&
1846 kprobe_on_func_entry(kp
->addr
, kp
->symbol_name
, kp
->offset
)) {
1847 kp
->pre_handler
= setjmp_pre_handler
;
1848 kp
->break_handler
= longjmp_break_handler
;
1849 return register_kprobe(kp
);
1854 EXPORT_SYMBOL_GPL(register_jprobe
);
1856 void unregister_jprobe(struct jprobe
*jp
)
1858 unregister_jprobes(&jp
, 1);
1860 EXPORT_SYMBOL_GPL(unregister_jprobe
);
1862 void unregister_jprobes(struct jprobe
**jps
, int num
)
1868 mutex_lock(&kprobe_mutex
);
1869 for (i
= 0; i
< num
; i
++)
1870 if (__unregister_kprobe_top(&jps
[i
]->kp
) < 0)
1871 jps
[i
]->kp
.addr
= NULL
;
1872 mutex_unlock(&kprobe_mutex
);
1874 synchronize_sched();
1875 for (i
= 0; i
< num
; i
++) {
1876 if (jps
[i
]->kp
.addr
)
1877 __unregister_kprobe_bottom(&jps
[i
]->kp
);
1880 EXPORT_SYMBOL_GPL(unregister_jprobes
);
1882 #ifdef CONFIG_KRETPROBES
1884 * This kprobe pre_handler is registered with every kretprobe. When probe
1885 * hits it will set up the return probe.
1887 static int pre_handler_kretprobe(struct kprobe
*p
, struct pt_regs
*regs
)
1889 struct kretprobe
*rp
= container_of(p
, struct kretprobe
, kp
);
1890 unsigned long hash
, flags
= 0;
1891 struct kretprobe_instance
*ri
;
1894 * To avoid deadlocks, prohibit return probing in NMI contexts,
1895 * just skip the probe and increase the (inexact) 'nmissed'
1896 * statistical counter, so that the user is informed that
1897 * something happened:
1899 if (unlikely(in_nmi())) {
1904 /* TODO: consider to only swap the RA after the last pre_handler fired */
1905 hash
= hash_ptr(current
, KPROBE_HASH_BITS
);
1906 raw_spin_lock_irqsave(&rp
->lock
, flags
);
1907 if (!hlist_empty(&rp
->free_instances
)) {
1908 ri
= hlist_entry(rp
->free_instances
.first
,
1909 struct kretprobe_instance
, hlist
);
1910 hlist_del(&ri
->hlist
);
1911 raw_spin_unlock_irqrestore(&rp
->lock
, flags
);
1916 if (rp
->entry_handler
&& rp
->entry_handler(ri
, regs
)) {
1917 raw_spin_lock_irqsave(&rp
->lock
, flags
);
1918 hlist_add_head(&ri
->hlist
, &rp
->free_instances
);
1919 raw_spin_unlock_irqrestore(&rp
->lock
, flags
);
1923 arch_prepare_kretprobe(ri
, regs
);
1925 /* XXX(hch): why is there no hlist_move_head? */
1926 INIT_HLIST_NODE(&ri
->hlist
);
1927 kretprobe_table_lock(hash
, &flags
);
1928 hlist_add_head(&ri
->hlist
, &kretprobe_inst_table
[hash
]);
1929 kretprobe_table_unlock(hash
, &flags
);
1932 raw_spin_unlock_irqrestore(&rp
->lock
, flags
);
1936 NOKPROBE_SYMBOL(pre_handler_kretprobe
);
1938 bool __weak
arch_kprobe_on_func_entry(unsigned long offset
)
1943 bool kprobe_on_func_entry(kprobe_opcode_t
*addr
, const char *sym
, unsigned long offset
)
1945 kprobe_opcode_t
*kp_addr
= _kprobe_addr(addr
, sym
, offset
);
1947 if (IS_ERR(kp_addr
))
1950 if (!kallsyms_lookup_size_offset((unsigned long)kp_addr
, NULL
, &offset
) ||
1951 !arch_kprobe_on_func_entry(offset
))
1957 int register_kretprobe(struct kretprobe
*rp
)
1960 struct kretprobe_instance
*inst
;
1964 if (!kprobe_on_func_entry(rp
->kp
.addr
, rp
->kp
.symbol_name
, rp
->kp
.offset
))
1967 if (kretprobe_blacklist_size
) {
1968 addr
= kprobe_addr(&rp
->kp
);
1970 return PTR_ERR(addr
);
1972 for (i
= 0; kretprobe_blacklist
[i
].name
!= NULL
; i
++) {
1973 if (kretprobe_blacklist
[i
].addr
== addr
)
1978 rp
->kp
.pre_handler
= pre_handler_kretprobe
;
1979 rp
->kp
.post_handler
= NULL
;
1980 rp
->kp
.fault_handler
= NULL
;
1981 rp
->kp
.break_handler
= NULL
;
1983 /* Pre-allocate memory for max kretprobe instances */
1984 if (rp
->maxactive
<= 0) {
1985 #ifdef CONFIG_PREEMPT
1986 rp
->maxactive
= max_t(unsigned int, 10, 2*num_possible_cpus());
1988 rp
->maxactive
= num_possible_cpus();
1991 raw_spin_lock_init(&rp
->lock
);
1992 INIT_HLIST_HEAD(&rp
->free_instances
);
1993 for (i
= 0; i
< rp
->maxactive
; i
++) {
1994 inst
= kmalloc(sizeof(struct kretprobe_instance
) +
1995 rp
->data_size
, GFP_KERNEL
);
2000 INIT_HLIST_NODE(&inst
->hlist
);
2001 hlist_add_head(&inst
->hlist
, &rp
->free_instances
);
2005 /* Establish function entry probe point */
2006 ret
= register_kprobe(&rp
->kp
);
2011 EXPORT_SYMBOL_GPL(register_kretprobe
);
2013 int register_kretprobes(struct kretprobe
**rps
, int num
)
2019 for (i
= 0; i
< num
; i
++) {
2020 ret
= register_kretprobe(rps
[i
]);
2023 unregister_kretprobes(rps
, i
);
2029 EXPORT_SYMBOL_GPL(register_kretprobes
);
2031 void unregister_kretprobe(struct kretprobe
*rp
)
2033 unregister_kretprobes(&rp
, 1);
2035 EXPORT_SYMBOL_GPL(unregister_kretprobe
);
2037 void unregister_kretprobes(struct kretprobe
**rps
, int num
)
2043 mutex_lock(&kprobe_mutex
);
2044 for (i
= 0; i
< num
; i
++)
2045 if (__unregister_kprobe_top(&rps
[i
]->kp
) < 0)
2046 rps
[i
]->kp
.addr
= NULL
;
2047 mutex_unlock(&kprobe_mutex
);
2049 synchronize_sched();
2050 for (i
= 0; i
< num
; i
++) {
2051 if (rps
[i
]->kp
.addr
) {
2052 __unregister_kprobe_bottom(&rps
[i
]->kp
);
2053 cleanup_rp_inst(rps
[i
]);
2057 EXPORT_SYMBOL_GPL(unregister_kretprobes
);
2059 #else /* CONFIG_KRETPROBES */
2060 int register_kretprobe(struct kretprobe
*rp
)
2064 EXPORT_SYMBOL_GPL(register_kretprobe
);
2066 int register_kretprobes(struct kretprobe
**rps
, int num
)
2070 EXPORT_SYMBOL_GPL(register_kretprobes
);
2072 void unregister_kretprobe(struct kretprobe
*rp
)
2075 EXPORT_SYMBOL_GPL(unregister_kretprobe
);
2077 void unregister_kretprobes(struct kretprobe
**rps
, int num
)
2080 EXPORT_SYMBOL_GPL(unregister_kretprobes
);
2082 static int pre_handler_kretprobe(struct kprobe
*p
, struct pt_regs
*regs
)
2086 NOKPROBE_SYMBOL(pre_handler_kretprobe
);
2088 #endif /* CONFIG_KRETPROBES */
2090 /* Set the kprobe gone and remove its instruction buffer. */
2091 static void kill_kprobe(struct kprobe
*p
)
2095 p
->flags
|= KPROBE_FLAG_GONE
;
2096 if (kprobe_aggrprobe(p
)) {
2098 * If this is an aggr_kprobe, we have to list all the
2099 * chained probes and mark them GONE.
2101 list_for_each_entry_rcu(kp
, &p
->list
, list
)
2102 kp
->flags
|= KPROBE_FLAG_GONE
;
2103 p
->post_handler
= NULL
;
2104 p
->break_handler
= NULL
;
2105 kill_optimized_kprobe(p
);
2108 * Here, we can remove insn_slot safely, because no thread calls
2109 * the original probed function (which will be freed soon) any more.
2111 arch_remove_kprobe(p
);
2114 /* Disable one kprobe */
2115 int disable_kprobe(struct kprobe
*kp
)
2119 mutex_lock(&kprobe_mutex
);
2121 /* Disable this kprobe */
2122 if (__disable_kprobe(kp
) == NULL
)
2125 mutex_unlock(&kprobe_mutex
);
2128 EXPORT_SYMBOL_GPL(disable_kprobe
);
2130 /* Enable one kprobe */
2131 int enable_kprobe(struct kprobe
*kp
)
2136 mutex_lock(&kprobe_mutex
);
2138 /* Check whether specified probe is valid. */
2139 p
= __get_valid_kprobe(kp
);
2140 if (unlikely(p
== NULL
)) {
2145 if (kprobe_gone(kp
)) {
2146 /* This kprobe has gone, we couldn't enable it. */
2152 kp
->flags
&= ~KPROBE_FLAG_DISABLED
;
2154 if (!kprobes_all_disarmed
&& kprobe_disabled(p
)) {
2155 p
->flags
&= ~KPROBE_FLAG_DISABLED
;
2159 mutex_unlock(&kprobe_mutex
);
2162 EXPORT_SYMBOL_GPL(enable_kprobe
);
2164 void dump_kprobe(struct kprobe
*kp
)
2166 printk(KERN_WARNING
"Dumping kprobe:\n");
2167 printk(KERN_WARNING
"Name: %s\nAddress: %p\nOffset: %x\n",
2168 kp
->symbol_name
, kp
->addr
, kp
->offset
);
2170 NOKPROBE_SYMBOL(dump_kprobe
);
2173 * Lookup and populate the kprobe_blacklist.
2175 * Unlike the kretprobe blacklist, we'll need to determine
2176 * the range of addresses that belong to the said functions,
2177 * since a kprobe need not necessarily be at the beginning
2180 static int __init
populate_kprobe_blacklist(unsigned long *start
,
2183 unsigned long *iter
;
2184 struct kprobe_blacklist_entry
*ent
;
2185 unsigned long entry
, offset
= 0, size
= 0;
2187 for (iter
= start
; iter
< end
; iter
++) {
2188 entry
= arch_deref_entry_point((void *)*iter
);
2190 if (!kernel_text_address(entry
) ||
2191 !kallsyms_lookup_size_offset(entry
, &size
, &offset
)) {
2192 pr_err("Failed to find blacklist at %p\n",
2197 ent
= kmalloc(sizeof(*ent
), GFP_KERNEL
);
2200 ent
->start_addr
= entry
;
2201 ent
->end_addr
= entry
+ size
;
2202 INIT_LIST_HEAD(&ent
->list
);
2203 list_add_tail(&ent
->list
, &kprobe_blacklist
);
2208 /* Module notifier call back, checking kprobes on the module */
2209 static int kprobes_module_callback(struct notifier_block
*nb
,
2210 unsigned long val
, void *data
)
2212 struct module
*mod
= data
;
2213 struct hlist_head
*head
;
2216 int checkcore
= (val
== MODULE_STATE_GOING
);
2218 if (val
!= MODULE_STATE_GOING
&& val
!= MODULE_STATE_LIVE
)
2222 * When MODULE_STATE_GOING was notified, both of module .text and
2223 * .init.text sections would be freed. When MODULE_STATE_LIVE was
2224 * notified, only .init.text section would be freed. We need to
2225 * disable kprobes which have been inserted in the sections.
2227 mutex_lock(&kprobe_mutex
);
2228 for (i
= 0; i
< KPROBE_TABLE_SIZE
; i
++) {
2229 head
= &kprobe_table
[i
];
2230 hlist_for_each_entry_rcu(p
, head
, hlist
)
2231 if (within_module_init((unsigned long)p
->addr
, mod
) ||
2233 within_module_core((unsigned long)p
->addr
, mod
))) {
2235 * The vaddr this probe is installed will soon
2236 * be vfreed buy not synced to disk. Hence,
2237 * disarming the breakpoint isn't needed.
2239 * Note, this will also move any optimized probes
2240 * that are pending to be removed from their
2241 * corresponding lists to the freeing_list and
2242 * will not be touched by the delayed
2243 * kprobe_optimizer work handler.
2248 mutex_unlock(&kprobe_mutex
);
2252 static struct notifier_block kprobe_module_nb
= {
2253 .notifier_call
= kprobes_module_callback
,
2257 /* Markers of _kprobe_blacklist section */
2258 extern unsigned long __start_kprobe_blacklist
[];
2259 extern unsigned long __stop_kprobe_blacklist
[];
2261 static int __init
init_kprobes(void)
2265 /* FIXME allocate the probe table, currently defined statically */
2266 /* initialize all list heads */
2267 for (i
= 0; i
< KPROBE_TABLE_SIZE
; i
++) {
2268 INIT_HLIST_HEAD(&kprobe_table
[i
]);
2269 INIT_HLIST_HEAD(&kretprobe_inst_table
[i
]);
2270 raw_spin_lock_init(&(kretprobe_table_locks
[i
].lock
));
2273 err
= populate_kprobe_blacklist(__start_kprobe_blacklist
,
2274 __stop_kprobe_blacklist
);
2276 pr_err("kprobes: failed to populate blacklist: %d\n", err
);
2277 pr_err("Please take care of using kprobes.\n");
2280 if (kretprobe_blacklist_size
) {
2281 /* lookup the function address from its name */
2282 for (i
= 0; kretprobe_blacklist
[i
].name
!= NULL
; i
++) {
2283 kretprobe_blacklist
[i
].addr
=
2284 kprobe_lookup_name(kretprobe_blacklist
[i
].name
, 0);
2285 if (!kretprobe_blacklist
[i
].addr
)
2286 printk("kretprobe: lookup failed: %s\n",
2287 kretprobe_blacklist
[i
].name
);
2291 #if defined(CONFIG_OPTPROBES)
2292 #if defined(__ARCH_WANT_KPROBES_INSN_SLOT)
2293 /* Init kprobe_optinsn_slots */
2294 kprobe_optinsn_slots
.insn_size
= MAX_OPTINSN_SIZE
;
2296 /* By default, kprobes can be optimized */
2297 kprobes_allow_optimization
= true;
2300 /* By default, kprobes are armed */
2301 kprobes_all_disarmed
= false;
2303 err
= arch_init_kprobes();
2305 err
= register_die_notifier(&kprobe_exceptions_nb
);
2307 err
= register_module_notifier(&kprobe_module_nb
);
2309 kprobes_initialized
= (err
== 0);
2316 #ifdef CONFIG_DEBUG_FS
2317 static void report_probe(struct seq_file
*pi
, struct kprobe
*p
,
2318 const char *sym
, int offset
, char *modname
, struct kprobe
*pp
)
2322 if (p
->pre_handler
== pre_handler_kretprobe
)
2324 else if (p
->pre_handler
== setjmp_pre_handler
)
2330 seq_printf(pi
, "%p %s %s+0x%x %s ",
2331 p
->addr
, kprobe_type
, sym
, offset
,
2332 (modname
? modname
: " "));
2334 seq_printf(pi
, "%p %s %p ",
2335 p
->addr
, kprobe_type
, p
->addr
);
2339 seq_printf(pi
, "%s%s%s%s\n",
2340 (kprobe_gone(p
) ? "[GONE]" : ""),
2341 ((kprobe_disabled(p
) && !kprobe_gone(p
)) ? "[DISABLED]" : ""),
2342 (kprobe_optimized(pp
) ? "[OPTIMIZED]" : ""),
2343 (kprobe_ftrace(pp
) ? "[FTRACE]" : ""));
2346 static void *kprobe_seq_start(struct seq_file
*f
, loff_t
*pos
)
2348 return (*pos
< KPROBE_TABLE_SIZE
) ? pos
: NULL
;
2351 static void *kprobe_seq_next(struct seq_file
*f
, void *v
, loff_t
*pos
)
2354 if (*pos
>= KPROBE_TABLE_SIZE
)
2359 static void kprobe_seq_stop(struct seq_file
*f
, void *v
)
2364 static int show_kprobe_addr(struct seq_file
*pi
, void *v
)
2366 struct hlist_head
*head
;
2367 struct kprobe
*p
, *kp
;
2368 const char *sym
= NULL
;
2369 unsigned int i
= *(loff_t
*) v
;
2370 unsigned long offset
= 0;
2371 char *modname
, namebuf
[KSYM_NAME_LEN
];
2373 head
= &kprobe_table
[i
];
2375 hlist_for_each_entry_rcu(p
, head
, hlist
) {
2376 sym
= kallsyms_lookup((unsigned long)p
->addr
, NULL
,
2377 &offset
, &modname
, namebuf
);
2378 if (kprobe_aggrprobe(p
)) {
2379 list_for_each_entry_rcu(kp
, &p
->list
, list
)
2380 report_probe(pi
, kp
, sym
, offset
, modname
, p
);
2382 report_probe(pi
, p
, sym
, offset
, modname
, NULL
);
2388 static const struct seq_operations kprobes_seq_ops
= {
2389 .start
= kprobe_seq_start
,
2390 .next
= kprobe_seq_next
,
2391 .stop
= kprobe_seq_stop
,
2392 .show
= show_kprobe_addr
2395 static int kprobes_open(struct inode
*inode
, struct file
*filp
)
2397 return seq_open(filp
, &kprobes_seq_ops
);
2400 static const struct file_operations debugfs_kprobes_operations
= {
2401 .open
= kprobes_open
,
2403 .llseek
= seq_lseek
,
2404 .release
= seq_release
,
2407 /* kprobes/blacklist -- shows which functions can not be probed */
2408 static void *kprobe_blacklist_seq_start(struct seq_file
*m
, loff_t
*pos
)
2410 return seq_list_start(&kprobe_blacklist
, *pos
);
2413 static void *kprobe_blacklist_seq_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
2415 return seq_list_next(v
, &kprobe_blacklist
, pos
);
2418 static int kprobe_blacklist_seq_show(struct seq_file
*m
, void *v
)
2420 struct kprobe_blacklist_entry
*ent
=
2421 list_entry(v
, struct kprobe_blacklist_entry
, list
);
2423 seq_printf(m
, "0x%p-0x%p\t%ps\n", (void *)ent
->start_addr
,
2424 (void *)ent
->end_addr
, (void *)ent
->start_addr
);
2428 static const struct seq_operations kprobe_blacklist_seq_ops
= {
2429 .start
= kprobe_blacklist_seq_start
,
2430 .next
= kprobe_blacklist_seq_next
,
2431 .stop
= kprobe_seq_stop
, /* Reuse void function */
2432 .show
= kprobe_blacklist_seq_show
,
2435 static int kprobe_blacklist_open(struct inode
*inode
, struct file
*filp
)
2437 return seq_open(filp
, &kprobe_blacklist_seq_ops
);
2440 static const struct file_operations debugfs_kprobe_blacklist_ops
= {
2441 .open
= kprobe_blacklist_open
,
2443 .llseek
= seq_lseek
,
2444 .release
= seq_release
,
2447 static void arm_all_kprobes(void)
2449 struct hlist_head
*head
;
2453 mutex_lock(&kprobe_mutex
);
2455 /* If kprobes are armed, just return */
2456 if (!kprobes_all_disarmed
)
2457 goto already_enabled
;
2460 * optimize_kprobe() called by arm_kprobe() checks
2461 * kprobes_all_disarmed, so set kprobes_all_disarmed before
2464 kprobes_all_disarmed
= false;
2465 /* Arming kprobes doesn't optimize kprobe itself */
2466 for (i
= 0; i
< KPROBE_TABLE_SIZE
; i
++) {
2467 head
= &kprobe_table
[i
];
2468 hlist_for_each_entry_rcu(p
, head
, hlist
)
2469 if (!kprobe_disabled(p
))
2473 printk(KERN_INFO
"Kprobes globally enabled\n");
2476 mutex_unlock(&kprobe_mutex
);
2480 static void disarm_all_kprobes(void)
2482 struct hlist_head
*head
;
2486 mutex_lock(&kprobe_mutex
);
2488 /* If kprobes are already disarmed, just return */
2489 if (kprobes_all_disarmed
) {
2490 mutex_unlock(&kprobe_mutex
);
2494 kprobes_all_disarmed
= true;
2495 printk(KERN_INFO
"Kprobes globally disabled\n");
2497 for (i
= 0; i
< KPROBE_TABLE_SIZE
; i
++) {
2498 head
= &kprobe_table
[i
];
2499 hlist_for_each_entry_rcu(p
, head
, hlist
) {
2500 if (!arch_trampoline_kprobe(p
) && !kprobe_disabled(p
))
2501 disarm_kprobe(p
, false);
2504 mutex_unlock(&kprobe_mutex
);
2506 /* Wait for disarming all kprobes by optimizer */
2507 wait_for_kprobe_optimizer();
2511 * XXX: The debugfs bool file interface doesn't allow for callbacks
2512 * when the bool state is switched. We can reuse that facility when
2515 static ssize_t
read_enabled_file_bool(struct file
*file
,
2516 char __user
*user_buf
, size_t count
, loff_t
*ppos
)
2520 if (!kprobes_all_disarmed
)
2526 return simple_read_from_buffer(user_buf
, count
, ppos
, buf
, 2);
2529 static ssize_t
write_enabled_file_bool(struct file
*file
,
2530 const char __user
*user_buf
, size_t count
, loff_t
*ppos
)
2535 buf_size
= min(count
, (sizeof(buf
)-1));
2536 if (copy_from_user(buf
, user_buf
, buf_size
))
2539 buf
[buf_size
] = '\0';
2549 disarm_all_kprobes();
2558 static const struct file_operations fops_kp
= {
2559 .read
= read_enabled_file_bool
,
2560 .write
= write_enabled_file_bool
,
2561 .llseek
= default_llseek
,
2564 static int __init
debugfs_kprobe_init(void)
2566 struct dentry
*dir
, *file
;
2567 unsigned int value
= 1;
2569 dir
= debugfs_create_dir("kprobes", NULL
);
2573 file
= debugfs_create_file("list", 0400, dir
, NULL
,
2574 &debugfs_kprobes_operations
);
2578 file
= debugfs_create_file("enabled", 0600, dir
,
2583 file
= debugfs_create_file("blacklist", 0400, dir
, NULL
,
2584 &debugfs_kprobe_blacklist_ops
);
2591 debugfs_remove(dir
);
2595 late_initcall(debugfs_kprobe_init
);
2596 #endif /* CONFIG_DEBUG_FS */
2598 module_init(init_kprobes
);
2600 /* defined in arch/.../kernel/kprobes.c */
2601 EXPORT_SYMBOL_GPL(jprobe_return
);