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 void __weak
*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)
487 * The optimization/unoptimization refers online_cpus via
488 * stop_machine() and cpu-hotplug modifies online_cpus.
489 * And same time, text_mutex will be held in cpu-hotplug and here.
490 * This combination can cause a deadlock (cpu-hotplug try to lock
491 * text_mutex but stop_machine can not be done because online_cpus
493 * To avoid this deadlock, caller must have locked cpu hotplug
494 * for preventing cpu-hotplug outside of text_mutex locking.
496 lockdep_assert_cpus_held();
498 /* Optimization never be done when disarmed */
499 if (kprobes_all_disarmed
|| !kprobes_allow_optimization
||
500 list_empty(&optimizing_list
))
503 mutex_lock(&text_mutex
);
504 arch_optimize_kprobes(&optimizing_list
);
505 mutex_unlock(&text_mutex
);
509 * Unoptimize (replace a jump with a breakpoint and remove the breakpoint
510 * if need) kprobes listed on unoptimizing_list.
512 static void do_unoptimize_kprobes(void)
514 struct optimized_kprobe
*op
, *tmp
;
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 mutex_lock(&text_mutex
);
524 arch_unoptimize_kprobes(&unoptimizing_list
, &freeing_list
);
525 /* Loop free_list for disarming */
526 list_for_each_entry_safe(op
, tmp
, &freeing_list
, list
) {
527 /* Disarm probes if marked disabled */
528 if (kprobe_disabled(&op
->kp
))
529 arch_disarm_kprobe(&op
->kp
);
530 if (kprobe_unused(&op
->kp
)) {
532 * Remove unused probes from hash list. After waiting
533 * for synchronization, these probes are reclaimed.
534 * (reclaiming is done by do_free_cleaned_kprobes.)
536 hlist_del_rcu(&op
->kp
.hlist
);
538 list_del_init(&op
->list
);
540 mutex_unlock(&text_mutex
);
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 /* Lock modules while optimizing kprobes */
573 mutex_lock(&module_mutex
);
576 * Step 1: Unoptimize kprobes and collect cleaned (unused and disarmed)
577 * kprobes before waiting for quiesence period.
579 do_unoptimize_kprobes();
582 * Step 2: Wait for quiesence period to ensure all potentially
583 * preempted tasks to have normally scheduled. Because optprobe
584 * may modify multiple instructions, there is a chance that Nth
585 * instruction is preempted. In that case, such tasks can return
586 * to 2nd-Nth byte of jump instruction. This wait is for avoiding it.
587 * Note that on non-preemptive kernel, this is transparently converted
588 * to synchronoze_sched() to wait for all interrupts to have completed.
590 synchronize_rcu_tasks();
592 /* Step 3: Optimize kprobes after quiesence period */
593 do_optimize_kprobes();
595 /* Step 4: Free cleaned kprobes after quiesence period */
596 do_free_cleaned_kprobes();
598 mutex_unlock(&module_mutex
);
600 mutex_unlock(&kprobe_mutex
);
602 /* Step 5: Kick optimizer again if needed */
603 if (!list_empty(&optimizing_list
) || !list_empty(&unoptimizing_list
))
604 kick_kprobe_optimizer();
607 /* Wait for completing optimization and unoptimization */
608 void wait_for_kprobe_optimizer(void)
610 mutex_lock(&kprobe_mutex
);
612 while (!list_empty(&optimizing_list
) || !list_empty(&unoptimizing_list
)) {
613 mutex_unlock(&kprobe_mutex
);
615 /* this will also make optimizing_work execute immmediately */
616 flush_delayed_work(&optimizing_work
);
617 /* @optimizing_work might not have been queued yet, relax */
620 mutex_lock(&kprobe_mutex
);
623 mutex_unlock(&kprobe_mutex
);
626 /* Optimize kprobe if p is ready to be optimized */
627 static void optimize_kprobe(struct kprobe
*p
)
629 struct optimized_kprobe
*op
;
631 /* Check if the kprobe is disabled or not ready for optimization. */
632 if (!kprobe_optready(p
) || !kprobes_allow_optimization
||
633 (kprobe_disabled(p
) || kprobes_all_disarmed
))
636 /* kprobes with post_handler can not be optimized */
640 op
= container_of(p
, struct optimized_kprobe
, kp
);
642 /* Check there is no other kprobes at the optimized instructions */
643 if (arch_check_optimized_kprobe(op
) < 0)
646 /* Check if it is already optimized. */
647 if (op
->kp
.flags
& KPROBE_FLAG_OPTIMIZED
)
649 op
->kp
.flags
|= KPROBE_FLAG_OPTIMIZED
;
651 if (!list_empty(&op
->list
))
652 /* This is under unoptimizing. Just dequeue the probe */
653 list_del_init(&op
->list
);
655 list_add(&op
->list
, &optimizing_list
);
656 kick_kprobe_optimizer();
660 /* Short cut to direct unoptimizing */
661 static void force_unoptimize_kprobe(struct optimized_kprobe
*op
)
663 lockdep_assert_cpus_held();
664 arch_unoptimize_kprobe(op
);
665 if (kprobe_disabled(&op
->kp
))
666 arch_disarm_kprobe(&op
->kp
);
669 /* Unoptimize a kprobe if p is optimized */
670 static void unoptimize_kprobe(struct kprobe
*p
, bool force
)
672 struct optimized_kprobe
*op
;
674 if (!kprobe_aggrprobe(p
) || kprobe_disarmed(p
))
675 return; /* This is not an optprobe nor optimized */
677 op
= container_of(p
, struct optimized_kprobe
, kp
);
678 if (!kprobe_optimized(p
)) {
679 /* Unoptimized or unoptimizing case */
680 if (force
&& !list_empty(&op
->list
)) {
682 * Only if this is unoptimizing kprobe and forced,
683 * forcibly unoptimize it. (No need to unoptimize
684 * unoptimized kprobe again :)
686 list_del_init(&op
->list
);
687 force_unoptimize_kprobe(op
);
692 op
->kp
.flags
&= ~KPROBE_FLAG_OPTIMIZED
;
693 if (!list_empty(&op
->list
)) {
694 /* Dequeue from the optimization queue */
695 list_del_init(&op
->list
);
698 /* Optimized kprobe case */
700 /* Forcibly update the code: this is a special case */
701 force_unoptimize_kprobe(op
);
703 list_add(&op
->list
, &unoptimizing_list
);
704 kick_kprobe_optimizer();
708 /* Cancel unoptimizing for reusing */
709 static int reuse_unused_kprobe(struct kprobe
*ap
)
711 struct optimized_kprobe
*op
;
715 * Unused kprobe MUST be on the way of delayed unoptimizing (means
716 * there is still a relative jump) and disabled.
718 op
= container_of(ap
, struct optimized_kprobe
, kp
);
719 WARN_ON_ONCE(list_empty(&op
->list
));
720 /* Enable the probe again */
721 ap
->flags
&= ~KPROBE_FLAG_DISABLED
;
722 /* Optimize it again (remove from op->list) */
723 ret
= kprobe_optready(ap
);
731 /* Remove optimized instructions */
732 static void kill_optimized_kprobe(struct kprobe
*p
)
734 struct optimized_kprobe
*op
;
736 op
= container_of(p
, struct optimized_kprobe
, kp
);
737 if (!list_empty(&op
->list
))
738 /* Dequeue from the (un)optimization queue */
739 list_del_init(&op
->list
);
740 op
->kp
.flags
&= ~KPROBE_FLAG_OPTIMIZED
;
742 if (kprobe_unused(p
)) {
743 /* Enqueue if it is unused */
744 list_add(&op
->list
, &freeing_list
);
746 * Remove unused probes from the hash list. After waiting
747 * for synchronization, this probe is reclaimed.
748 * (reclaiming is done by do_free_cleaned_kprobes().)
750 hlist_del_rcu(&op
->kp
.hlist
);
753 /* Don't touch the code, because it is already freed. */
754 arch_remove_optimized_kprobe(op
);
758 void __prepare_optimized_kprobe(struct optimized_kprobe
*op
, struct kprobe
*p
)
760 if (!kprobe_ftrace(p
))
761 arch_prepare_optimized_kprobe(op
, p
);
764 /* Try to prepare optimized instructions */
765 static void prepare_optimized_kprobe(struct kprobe
*p
)
767 struct optimized_kprobe
*op
;
769 op
= container_of(p
, struct optimized_kprobe
, kp
);
770 __prepare_optimized_kprobe(op
, p
);
773 /* Allocate new optimized_kprobe and try to prepare optimized instructions */
774 static struct kprobe
*alloc_aggr_kprobe(struct kprobe
*p
)
776 struct optimized_kprobe
*op
;
778 op
= kzalloc(sizeof(struct optimized_kprobe
), GFP_KERNEL
);
782 INIT_LIST_HEAD(&op
->list
);
783 op
->kp
.addr
= p
->addr
;
784 __prepare_optimized_kprobe(op
, p
);
789 static void init_aggr_kprobe(struct kprobe
*ap
, struct kprobe
*p
);
792 * Prepare an optimized_kprobe and optimize it
793 * NOTE: p must be a normal registered kprobe
795 static void try_to_optimize_kprobe(struct kprobe
*p
)
798 struct optimized_kprobe
*op
;
800 /* Impossible to optimize ftrace-based kprobe */
801 if (kprobe_ftrace(p
))
804 /* For preparing optimization, jump_label_text_reserved() is called */
807 mutex_lock(&text_mutex
);
809 ap
= alloc_aggr_kprobe(p
);
813 op
= container_of(ap
, struct optimized_kprobe
, kp
);
814 if (!arch_prepared_optinsn(&op
->optinsn
)) {
815 /* If failed to setup optimizing, fallback to kprobe */
816 arch_remove_optimized_kprobe(op
);
821 init_aggr_kprobe(ap
, p
);
822 optimize_kprobe(ap
); /* This just kicks optimizer thread */
825 mutex_unlock(&text_mutex
);
831 static void optimize_all_kprobes(void)
833 struct hlist_head
*head
;
837 mutex_lock(&kprobe_mutex
);
838 /* If optimization is already allowed, just return */
839 if (kprobes_allow_optimization
)
843 kprobes_allow_optimization
= true;
844 for (i
= 0; i
< KPROBE_TABLE_SIZE
; i
++) {
845 head
= &kprobe_table
[i
];
846 hlist_for_each_entry_rcu(p
, head
, hlist
)
847 if (!kprobe_disabled(p
))
851 printk(KERN_INFO
"Kprobes globally optimized\n");
853 mutex_unlock(&kprobe_mutex
);
856 static void unoptimize_all_kprobes(void)
858 struct hlist_head
*head
;
862 mutex_lock(&kprobe_mutex
);
863 /* If optimization is already prohibited, just return */
864 if (!kprobes_allow_optimization
) {
865 mutex_unlock(&kprobe_mutex
);
870 kprobes_allow_optimization
= false;
871 for (i
= 0; i
< KPROBE_TABLE_SIZE
; i
++) {
872 head
= &kprobe_table
[i
];
873 hlist_for_each_entry_rcu(p
, head
, hlist
) {
874 if (!kprobe_disabled(p
))
875 unoptimize_kprobe(p
, false);
879 mutex_unlock(&kprobe_mutex
);
881 /* Wait for unoptimizing completion */
882 wait_for_kprobe_optimizer();
883 printk(KERN_INFO
"Kprobes globally unoptimized\n");
886 static DEFINE_MUTEX(kprobe_sysctl_mutex
);
887 int sysctl_kprobes_optimization
;
888 int proc_kprobes_optimization_handler(struct ctl_table
*table
, int write
,
889 void __user
*buffer
, size_t *length
,
894 mutex_lock(&kprobe_sysctl_mutex
);
895 sysctl_kprobes_optimization
= kprobes_allow_optimization
? 1 : 0;
896 ret
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
898 if (sysctl_kprobes_optimization
)
899 optimize_all_kprobes();
901 unoptimize_all_kprobes();
902 mutex_unlock(&kprobe_sysctl_mutex
);
906 #endif /* CONFIG_SYSCTL */
908 /* Put a breakpoint for a probe. Must be called with text_mutex locked */
909 static void __arm_kprobe(struct kprobe
*p
)
913 /* Check collision with other optimized kprobes */
914 _p
= get_optimized_kprobe((unsigned long)p
->addr
);
916 /* Fallback to unoptimized kprobe */
917 unoptimize_kprobe(_p
, true);
920 optimize_kprobe(p
); /* Try to optimize (add kprobe to a list) */
923 /* Remove the breakpoint of a probe. Must be called with text_mutex locked */
924 static void __disarm_kprobe(struct kprobe
*p
, bool reopt
)
928 /* Try to unoptimize */
929 unoptimize_kprobe(p
, kprobes_all_disarmed
);
931 if (!kprobe_queued(p
)) {
932 arch_disarm_kprobe(p
);
933 /* If another kprobe was blocked, optimize it. */
934 _p
= get_optimized_kprobe((unsigned long)p
->addr
);
935 if (unlikely(_p
) && reopt
)
938 /* TODO: reoptimize others after unoptimized this probe */
941 #else /* !CONFIG_OPTPROBES */
943 #define optimize_kprobe(p) do {} while (0)
944 #define unoptimize_kprobe(p, f) do {} while (0)
945 #define kill_optimized_kprobe(p) do {} while (0)
946 #define prepare_optimized_kprobe(p) do {} while (0)
947 #define try_to_optimize_kprobe(p) do {} while (0)
948 #define __arm_kprobe(p) arch_arm_kprobe(p)
949 #define __disarm_kprobe(p, o) arch_disarm_kprobe(p)
950 #define kprobe_disarmed(p) kprobe_disabled(p)
951 #define wait_for_kprobe_optimizer() do {} while (0)
953 static int reuse_unused_kprobe(struct kprobe
*ap
)
956 * If the optimized kprobe is NOT supported, the aggr kprobe is
957 * released at the same time that the last aggregated kprobe is
959 * Thus there should be no chance to reuse unused kprobe.
961 printk(KERN_ERR
"Error: There should be no unused kprobe here.\n");
965 static void free_aggr_kprobe(struct kprobe
*p
)
967 arch_remove_kprobe(p
);
971 static struct kprobe
*alloc_aggr_kprobe(struct kprobe
*p
)
973 return kzalloc(sizeof(struct kprobe
), GFP_KERNEL
);
975 #endif /* CONFIG_OPTPROBES */
977 #ifdef CONFIG_KPROBES_ON_FTRACE
978 static struct ftrace_ops kprobe_ftrace_ops __read_mostly
= {
979 .func
= kprobe_ftrace_handler
,
980 .flags
= FTRACE_OPS_FL_SAVE_REGS
| FTRACE_OPS_FL_IPMODIFY
,
982 static int kprobe_ftrace_enabled
;
984 /* Must ensure p->addr is really on ftrace */
985 static int prepare_kprobe(struct kprobe
*p
)
987 if (!kprobe_ftrace(p
))
988 return arch_prepare_kprobe(p
);
990 return arch_prepare_kprobe_ftrace(p
);
993 /* Caller must lock kprobe_mutex */
994 static int arm_kprobe_ftrace(struct kprobe
*p
)
998 ret
= ftrace_set_filter_ip(&kprobe_ftrace_ops
,
999 (unsigned long)p
->addr
, 0, 0);
1001 pr_debug("Failed to arm kprobe-ftrace at %pS (%d)\n",
1006 if (kprobe_ftrace_enabled
== 0) {
1007 ret
= register_ftrace_function(&kprobe_ftrace_ops
);
1009 pr_debug("Failed to init kprobe-ftrace (%d)\n", ret
);
1014 kprobe_ftrace_enabled
++;
1019 * Note: Since kprobe_ftrace_ops has IPMODIFY set, and ftrace requires a
1020 * non-empty filter_hash for IPMODIFY ops, we're safe from an accidental
1021 * empty filter_hash which would undesirably trace all functions.
1023 ftrace_set_filter_ip(&kprobe_ftrace_ops
, (unsigned long)p
->addr
, 1, 0);
1027 /* Caller must lock kprobe_mutex */
1028 static int disarm_kprobe_ftrace(struct kprobe
*p
)
1032 if (kprobe_ftrace_enabled
== 1) {
1033 ret
= unregister_ftrace_function(&kprobe_ftrace_ops
);
1034 if (WARN(ret
< 0, "Failed to unregister kprobe-ftrace (%d)\n", ret
))
1038 kprobe_ftrace_enabled
--;
1040 ret
= ftrace_set_filter_ip(&kprobe_ftrace_ops
,
1041 (unsigned long)p
->addr
, 1, 0);
1042 WARN_ONCE(ret
< 0, "Failed to disarm kprobe-ftrace at %pS (%d)\n",
1046 #else /* !CONFIG_KPROBES_ON_FTRACE */
1047 #define prepare_kprobe(p) arch_prepare_kprobe(p)
1048 #define arm_kprobe_ftrace(p) (-ENODEV)
1049 #define disarm_kprobe_ftrace(p) (-ENODEV)
1052 /* Arm a kprobe with text_mutex */
1053 static int arm_kprobe(struct kprobe
*kp
)
1055 if (unlikely(kprobe_ftrace(kp
)))
1056 return arm_kprobe_ftrace(kp
);
1059 mutex_lock(&text_mutex
);
1061 mutex_unlock(&text_mutex
);
1067 /* Disarm a kprobe with text_mutex */
1068 static int disarm_kprobe(struct kprobe
*kp
, bool reopt
)
1070 if (unlikely(kprobe_ftrace(kp
)))
1071 return disarm_kprobe_ftrace(kp
);
1074 mutex_lock(&text_mutex
);
1075 __disarm_kprobe(kp
, reopt
);
1076 mutex_unlock(&text_mutex
);
1083 * Aggregate handlers for multiple kprobes support - these handlers
1084 * take care of invoking the individual kprobe handlers on p->list
1086 static int aggr_pre_handler(struct kprobe
*p
, struct pt_regs
*regs
)
1090 list_for_each_entry_rcu(kp
, &p
->list
, list
) {
1091 if (kp
->pre_handler
&& likely(!kprobe_disabled(kp
))) {
1092 set_kprobe_instance(kp
);
1093 if (kp
->pre_handler(kp
, regs
))
1096 reset_kprobe_instance();
1100 NOKPROBE_SYMBOL(aggr_pre_handler
);
1102 static void aggr_post_handler(struct kprobe
*p
, struct pt_regs
*regs
,
1103 unsigned long flags
)
1107 list_for_each_entry_rcu(kp
, &p
->list
, list
) {
1108 if (kp
->post_handler
&& likely(!kprobe_disabled(kp
))) {
1109 set_kprobe_instance(kp
);
1110 kp
->post_handler(kp
, regs
, flags
);
1111 reset_kprobe_instance();
1115 NOKPROBE_SYMBOL(aggr_post_handler
);
1117 static int aggr_fault_handler(struct kprobe
*p
, struct pt_regs
*regs
,
1120 struct kprobe
*cur
= __this_cpu_read(kprobe_instance
);
1123 * if we faulted "during" the execution of a user specified
1124 * probe handler, invoke just that probe's fault handler
1126 if (cur
&& cur
->fault_handler
) {
1127 if (cur
->fault_handler(cur
, regs
, trapnr
))
1132 NOKPROBE_SYMBOL(aggr_fault_handler
);
1134 /* Walks the list and increments nmissed count for multiprobe case */
1135 void kprobes_inc_nmissed_count(struct kprobe
*p
)
1138 if (!kprobe_aggrprobe(p
)) {
1141 list_for_each_entry_rcu(kp
, &p
->list
, list
)
1146 NOKPROBE_SYMBOL(kprobes_inc_nmissed_count
);
1148 void recycle_rp_inst(struct kretprobe_instance
*ri
,
1149 struct hlist_head
*head
)
1151 struct kretprobe
*rp
= ri
->rp
;
1153 /* remove rp inst off the rprobe_inst_table */
1154 hlist_del(&ri
->hlist
);
1155 INIT_HLIST_NODE(&ri
->hlist
);
1157 raw_spin_lock(&rp
->lock
);
1158 hlist_add_head(&ri
->hlist
, &rp
->free_instances
);
1159 raw_spin_unlock(&rp
->lock
);
1162 hlist_add_head(&ri
->hlist
, head
);
1164 NOKPROBE_SYMBOL(recycle_rp_inst
);
1166 void kretprobe_hash_lock(struct task_struct
*tsk
,
1167 struct hlist_head
**head
, unsigned long *flags
)
1168 __acquires(hlist_lock
)
1170 unsigned long hash
= hash_ptr(tsk
, KPROBE_HASH_BITS
);
1171 raw_spinlock_t
*hlist_lock
;
1173 *head
= &kretprobe_inst_table
[hash
];
1174 hlist_lock
= kretprobe_table_lock_ptr(hash
);
1175 raw_spin_lock_irqsave(hlist_lock
, *flags
);
1177 NOKPROBE_SYMBOL(kretprobe_hash_lock
);
1179 static void kretprobe_table_lock(unsigned long hash
,
1180 unsigned long *flags
)
1181 __acquires(hlist_lock
)
1183 raw_spinlock_t
*hlist_lock
= kretprobe_table_lock_ptr(hash
);
1184 raw_spin_lock_irqsave(hlist_lock
, *flags
);
1186 NOKPROBE_SYMBOL(kretprobe_table_lock
);
1188 void kretprobe_hash_unlock(struct task_struct
*tsk
,
1189 unsigned long *flags
)
1190 __releases(hlist_lock
)
1192 unsigned long hash
= hash_ptr(tsk
, KPROBE_HASH_BITS
);
1193 raw_spinlock_t
*hlist_lock
;
1195 hlist_lock
= kretprobe_table_lock_ptr(hash
);
1196 raw_spin_unlock_irqrestore(hlist_lock
, *flags
);
1198 NOKPROBE_SYMBOL(kretprobe_hash_unlock
);
1200 static void kretprobe_table_unlock(unsigned long hash
,
1201 unsigned long *flags
)
1202 __releases(hlist_lock
)
1204 raw_spinlock_t
*hlist_lock
= kretprobe_table_lock_ptr(hash
);
1205 raw_spin_unlock_irqrestore(hlist_lock
, *flags
);
1207 NOKPROBE_SYMBOL(kretprobe_table_unlock
);
1210 * This function is called from finish_task_switch when task tk becomes dead,
1211 * so that we can recycle any function-return probe instances associated
1212 * with this task. These left over instances represent probed functions
1213 * that have been called but will never return.
1215 void kprobe_flush_task(struct task_struct
*tk
)
1217 struct kretprobe_instance
*ri
;
1218 struct hlist_head
*head
, empty_rp
;
1219 struct hlist_node
*tmp
;
1220 unsigned long hash
, flags
= 0;
1222 if (unlikely(!kprobes_initialized
))
1223 /* Early boot. kretprobe_table_locks not yet initialized. */
1226 INIT_HLIST_HEAD(&empty_rp
);
1227 hash
= hash_ptr(tk
, KPROBE_HASH_BITS
);
1228 head
= &kretprobe_inst_table
[hash
];
1229 kretprobe_table_lock(hash
, &flags
);
1230 hlist_for_each_entry_safe(ri
, tmp
, head
, hlist
) {
1232 recycle_rp_inst(ri
, &empty_rp
);
1234 kretprobe_table_unlock(hash
, &flags
);
1235 hlist_for_each_entry_safe(ri
, tmp
, &empty_rp
, hlist
) {
1236 hlist_del(&ri
->hlist
);
1240 NOKPROBE_SYMBOL(kprobe_flush_task
);
1242 static inline void free_rp_inst(struct kretprobe
*rp
)
1244 struct kretprobe_instance
*ri
;
1245 struct hlist_node
*next
;
1247 hlist_for_each_entry_safe(ri
, next
, &rp
->free_instances
, hlist
) {
1248 hlist_del(&ri
->hlist
);
1253 static void cleanup_rp_inst(struct kretprobe
*rp
)
1255 unsigned long flags
, hash
;
1256 struct kretprobe_instance
*ri
;
1257 struct hlist_node
*next
;
1258 struct hlist_head
*head
;
1261 for (hash
= 0; hash
< KPROBE_TABLE_SIZE
; hash
++) {
1262 kretprobe_table_lock(hash
, &flags
);
1263 head
= &kretprobe_inst_table
[hash
];
1264 hlist_for_each_entry_safe(ri
, next
, head
, hlist
) {
1268 kretprobe_table_unlock(hash
, &flags
);
1272 NOKPROBE_SYMBOL(cleanup_rp_inst
);
1274 /* Add the new probe to ap->list */
1275 static int add_new_kprobe(struct kprobe
*ap
, struct kprobe
*p
)
1277 if (p
->post_handler
)
1278 unoptimize_kprobe(ap
, true); /* Fall back to normal kprobe */
1280 list_add_rcu(&p
->list
, &ap
->list
);
1281 if (p
->post_handler
&& !ap
->post_handler
)
1282 ap
->post_handler
= aggr_post_handler
;
1288 * Fill in the required fields of the "manager kprobe". Replace the
1289 * earlier kprobe in the hlist with the manager kprobe
1291 static void init_aggr_kprobe(struct kprobe
*ap
, struct kprobe
*p
)
1293 /* Copy p's insn slot to ap */
1295 flush_insn_slot(ap
);
1297 ap
->flags
= p
->flags
& ~KPROBE_FLAG_OPTIMIZED
;
1298 ap
->pre_handler
= aggr_pre_handler
;
1299 ap
->fault_handler
= aggr_fault_handler
;
1300 /* We don't care the kprobe which has gone. */
1301 if (p
->post_handler
&& !kprobe_gone(p
))
1302 ap
->post_handler
= aggr_post_handler
;
1304 INIT_LIST_HEAD(&ap
->list
);
1305 INIT_HLIST_NODE(&ap
->hlist
);
1307 list_add_rcu(&p
->list
, &ap
->list
);
1308 hlist_replace_rcu(&p
->hlist
, &ap
->hlist
);
1312 * This is the second or subsequent kprobe at the address - handle
1315 static int register_aggr_kprobe(struct kprobe
*orig_p
, struct kprobe
*p
)
1318 struct kprobe
*ap
= orig_p
;
1322 /* For preparing optimization, jump_label_text_reserved() is called */
1324 mutex_lock(&text_mutex
);
1326 if (!kprobe_aggrprobe(orig_p
)) {
1327 /* If orig_p is not an aggr_kprobe, create new aggr_kprobe. */
1328 ap
= alloc_aggr_kprobe(orig_p
);
1333 init_aggr_kprobe(ap
, orig_p
);
1334 } else if (kprobe_unused(ap
)) {
1335 /* This probe is going to die. Rescue it */
1336 ret
= reuse_unused_kprobe(ap
);
1341 if (kprobe_gone(ap
)) {
1343 * Attempting to insert new probe at the same location that
1344 * had a probe in the module vaddr area which already
1345 * freed. So, the instruction slot has already been
1346 * released. We need a new slot for the new probe.
1348 ret
= arch_prepare_kprobe(ap
);
1351 * Even if fail to allocate new slot, don't need to
1352 * free aggr_probe. It will be used next time, or
1353 * freed by unregister_kprobe.
1357 /* Prepare optimized instructions if possible. */
1358 prepare_optimized_kprobe(ap
);
1361 * Clear gone flag to prevent allocating new slot again, and
1362 * set disabled flag because it is not armed yet.
1364 ap
->flags
= (ap
->flags
& ~KPROBE_FLAG_GONE
)
1365 | KPROBE_FLAG_DISABLED
;
1368 /* Copy ap's insn slot to p */
1370 ret
= add_new_kprobe(ap
, p
);
1373 mutex_unlock(&text_mutex
);
1374 jump_label_unlock();
1377 if (ret
== 0 && kprobe_disabled(ap
) && !kprobe_disabled(p
)) {
1378 ap
->flags
&= ~KPROBE_FLAG_DISABLED
;
1379 if (!kprobes_all_disarmed
) {
1380 /* Arm the breakpoint again. */
1381 ret
= arm_kprobe(ap
);
1383 ap
->flags
|= KPROBE_FLAG_DISABLED
;
1384 list_del_rcu(&p
->list
);
1392 bool __weak
arch_within_kprobe_blacklist(unsigned long addr
)
1394 /* The __kprobes marked functions and entry code must not be probed */
1395 return addr
>= (unsigned long)__kprobes_text_start
&&
1396 addr
< (unsigned long)__kprobes_text_end
;
1399 static bool __within_kprobe_blacklist(unsigned long addr
)
1401 struct kprobe_blacklist_entry
*ent
;
1403 if (arch_within_kprobe_blacklist(addr
))
1406 * If there exists a kprobe_blacklist, verify and
1407 * fail any probe registration in the prohibited area
1409 list_for_each_entry(ent
, &kprobe_blacklist
, list
) {
1410 if (addr
>= ent
->start_addr
&& addr
< ent
->end_addr
)
1416 bool within_kprobe_blacklist(unsigned long addr
)
1418 char symname
[KSYM_NAME_LEN
], *p
;
1420 if (__within_kprobe_blacklist(addr
))
1423 /* Check if the address is on a suffixed-symbol */
1424 if (!lookup_symbol_name(addr
, symname
)) {
1425 p
= strchr(symname
, '.');
1429 addr
= (unsigned long)kprobe_lookup_name(symname
, 0);
1431 return __within_kprobe_blacklist(addr
);
1437 * If we have a symbol_name argument, look it up and add the offset field
1438 * to it. This way, we can specify a relative address to a symbol.
1439 * This returns encoded errors if it fails to look up symbol or invalid
1440 * combination of parameters.
1442 static kprobe_opcode_t
*_kprobe_addr(kprobe_opcode_t
*addr
,
1443 const char *symbol_name
, unsigned int offset
)
1445 if ((symbol_name
&& addr
) || (!symbol_name
&& !addr
))
1449 addr
= kprobe_lookup_name(symbol_name
, offset
);
1451 return ERR_PTR(-ENOENT
);
1454 addr
= (kprobe_opcode_t
*)(((char *)addr
) + offset
);
1459 return ERR_PTR(-EINVAL
);
1462 static kprobe_opcode_t
*kprobe_addr(struct kprobe
*p
)
1464 return _kprobe_addr(p
->addr
, p
->symbol_name
, p
->offset
);
1467 /* Check passed kprobe is valid and return kprobe in kprobe_table. */
1468 static struct kprobe
*__get_valid_kprobe(struct kprobe
*p
)
1470 struct kprobe
*ap
, *list_p
;
1472 ap
= get_kprobe(p
->addr
);
1477 list_for_each_entry_rcu(list_p
, &ap
->list
, list
)
1479 /* kprobe p is a valid probe */
1487 /* Return error if the kprobe is being re-registered */
1488 static inline int check_kprobe_rereg(struct kprobe
*p
)
1492 mutex_lock(&kprobe_mutex
);
1493 if (__get_valid_kprobe(p
))
1495 mutex_unlock(&kprobe_mutex
);
1500 int __weak
arch_check_ftrace_location(struct kprobe
*p
)
1502 unsigned long ftrace_addr
;
1504 ftrace_addr
= ftrace_location((unsigned long)p
->addr
);
1506 #ifdef CONFIG_KPROBES_ON_FTRACE
1507 /* Given address is not on the instruction boundary */
1508 if ((unsigned long)p
->addr
!= ftrace_addr
)
1510 p
->flags
|= KPROBE_FLAG_FTRACE
;
1511 #else /* !CONFIG_KPROBES_ON_FTRACE */
1518 static int check_kprobe_address_safe(struct kprobe
*p
,
1519 struct module
**probed_mod
)
1523 ret
= arch_check_ftrace_location(p
);
1529 /* Ensure it is not in reserved area nor out of text */
1530 if (!kernel_text_address((unsigned long) p
->addr
) ||
1531 within_kprobe_blacklist((unsigned long) p
->addr
) ||
1532 jump_label_text_reserved(p
->addr
, 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
)) {
1616 ret
= arm_kprobe(p
);
1618 hlist_del_rcu(&p
->hlist
);
1624 /* Try to optimize kprobe */
1625 try_to_optimize_kprobe(p
);
1627 mutex_unlock(&kprobe_mutex
);
1630 module_put(probed_mod
);
1634 EXPORT_SYMBOL_GPL(register_kprobe
);
1636 /* Check if all probes on the aggrprobe are disabled */
1637 static int aggr_kprobe_disabled(struct kprobe
*ap
)
1641 list_for_each_entry_rcu(kp
, &ap
->list
, list
)
1642 if (!kprobe_disabled(kp
))
1644 * There is an active probe on the list.
1645 * We can't disable this ap.
1652 /* Disable one kprobe: Make sure called under kprobe_mutex is locked */
1653 static struct kprobe
*__disable_kprobe(struct kprobe
*p
)
1655 struct kprobe
*orig_p
;
1658 /* Get an original kprobe for return */
1659 orig_p
= __get_valid_kprobe(p
);
1660 if (unlikely(orig_p
== NULL
))
1661 return ERR_PTR(-EINVAL
);
1663 if (!kprobe_disabled(p
)) {
1664 /* Disable probe if it is a child probe */
1666 p
->flags
|= KPROBE_FLAG_DISABLED
;
1668 /* Try to disarm and disable this/parent probe */
1669 if (p
== orig_p
|| aggr_kprobe_disabled(orig_p
)) {
1671 * If kprobes_all_disarmed is set, orig_p
1672 * should have already been disarmed, so
1673 * skip unneed disarming process.
1675 if (!kprobes_all_disarmed
) {
1676 ret
= disarm_kprobe(orig_p
, true);
1678 p
->flags
&= ~KPROBE_FLAG_DISABLED
;
1679 return ERR_PTR(ret
);
1682 orig_p
->flags
|= KPROBE_FLAG_DISABLED
;
1690 * Unregister a kprobe without a scheduler synchronization.
1692 static int __unregister_kprobe_top(struct kprobe
*p
)
1694 struct kprobe
*ap
, *list_p
;
1696 /* Disable kprobe. This will disarm it if needed. */
1697 ap
= __disable_kprobe(p
);
1703 * This probe is an independent(and non-optimized) kprobe
1704 * (not an aggrprobe). Remove from the hash list.
1708 /* Following process expects this probe is an aggrprobe */
1709 WARN_ON(!kprobe_aggrprobe(ap
));
1711 if (list_is_singular(&ap
->list
) && kprobe_disarmed(ap
))
1713 * !disarmed could be happen if the probe is under delayed
1718 /* If disabling probe has special handlers, update aggrprobe */
1719 if (p
->post_handler
&& !kprobe_gone(p
)) {
1720 list_for_each_entry_rcu(list_p
, &ap
->list
, list
) {
1721 if ((list_p
!= p
) && (list_p
->post_handler
))
1724 ap
->post_handler
= NULL
;
1728 * Remove from the aggrprobe: this path will do nothing in
1729 * __unregister_kprobe_bottom().
1731 list_del_rcu(&p
->list
);
1732 if (!kprobe_disabled(ap
) && !kprobes_all_disarmed
)
1734 * Try to optimize this probe again, because post
1735 * handler may have been changed.
1737 optimize_kprobe(ap
);
1742 hlist_del_rcu(&ap
->hlist
);
1746 static void __unregister_kprobe_bottom(struct kprobe
*p
)
1750 if (list_empty(&p
->list
))
1751 /* This is an independent kprobe */
1752 arch_remove_kprobe(p
);
1753 else if (list_is_singular(&p
->list
)) {
1754 /* This is the last child of an aggrprobe */
1755 ap
= list_entry(p
->list
.next
, struct kprobe
, list
);
1757 free_aggr_kprobe(ap
);
1759 /* Otherwise, do nothing. */
1762 int register_kprobes(struct kprobe
**kps
, int num
)
1768 for (i
= 0; i
< num
; i
++) {
1769 ret
= register_kprobe(kps
[i
]);
1772 unregister_kprobes(kps
, i
);
1778 EXPORT_SYMBOL_GPL(register_kprobes
);
1780 void unregister_kprobe(struct kprobe
*p
)
1782 unregister_kprobes(&p
, 1);
1784 EXPORT_SYMBOL_GPL(unregister_kprobe
);
1786 void unregister_kprobes(struct kprobe
**kps
, int num
)
1792 mutex_lock(&kprobe_mutex
);
1793 for (i
= 0; i
< num
; i
++)
1794 if (__unregister_kprobe_top(kps
[i
]) < 0)
1795 kps
[i
]->addr
= NULL
;
1796 mutex_unlock(&kprobe_mutex
);
1799 for (i
= 0; i
< num
; i
++)
1801 __unregister_kprobe_bottom(kps
[i
]);
1803 EXPORT_SYMBOL_GPL(unregister_kprobes
);
1805 int __weak
kprobe_exceptions_notify(struct notifier_block
*self
,
1806 unsigned long val
, void *data
)
1810 NOKPROBE_SYMBOL(kprobe_exceptions_notify
);
1812 static struct notifier_block kprobe_exceptions_nb
= {
1813 .notifier_call
= kprobe_exceptions_notify
,
1814 .priority
= 0x7fffffff /* we need to be notified first */
1817 unsigned long __weak
arch_deref_entry_point(void *entry
)
1819 return (unsigned long)entry
;
1822 #ifdef CONFIG_KRETPROBES
1824 * This kprobe pre_handler is registered with every kretprobe. When probe
1825 * hits it will set up the return probe.
1827 static int pre_handler_kretprobe(struct kprobe
*p
, struct pt_regs
*regs
)
1829 struct kretprobe
*rp
= container_of(p
, struct kretprobe
, kp
);
1830 unsigned long hash
, flags
= 0;
1831 struct kretprobe_instance
*ri
;
1834 * To avoid deadlocks, prohibit return probing in NMI contexts,
1835 * just skip the probe and increase the (inexact) 'nmissed'
1836 * statistical counter, so that the user is informed that
1837 * something happened:
1839 if (unlikely(in_nmi())) {
1844 /* TODO: consider to only swap the RA after the last pre_handler fired */
1845 hash
= hash_ptr(current
, KPROBE_HASH_BITS
);
1846 raw_spin_lock_irqsave(&rp
->lock
, flags
);
1847 if (!hlist_empty(&rp
->free_instances
)) {
1848 ri
= hlist_entry(rp
->free_instances
.first
,
1849 struct kretprobe_instance
, hlist
);
1850 hlist_del(&ri
->hlist
);
1851 raw_spin_unlock_irqrestore(&rp
->lock
, flags
);
1856 if (rp
->entry_handler
&& rp
->entry_handler(ri
, regs
)) {
1857 raw_spin_lock_irqsave(&rp
->lock
, flags
);
1858 hlist_add_head(&ri
->hlist
, &rp
->free_instances
);
1859 raw_spin_unlock_irqrestore(&rp
->lock
, flags
);
1863 arch_prepare_kretprobe(ri
, regs
);
1865 /* XXX(hch): why is there no hlist_move_head? */
1866 INIT_HLIST_NODE(&ri
->hlist
);
1867 kretprobe_table_lock(hash
, &flags
);
1868 hlist_add_head(&ri
->hlist
, &kretprobe_inst_table
[hash
]);
1869 kretprobe_table_unlock(hash
, &flags
);
1872 raw_spin_unlock_irqrestore(&rp
->lock
, flags
);
1876 NOKPROBE_SYMBOL(pre_handler_kretprobe
);
1878 bool __weak
arch_kprobe_on_func_entry(unsigned long offset
)
1883 bool kprobe_on_func_entry(kprobe_opcode_t
*addr
, const char *sym
, unsigned long offset
)
1885 kprobe_opcode_t
*kp_addr
= _kprobe_addr(addr
, sym
, offset
);
1887 if (IS_ERR(kp_addr
))
1890 if (!kallsyms_lookup_size_offset((unsigned long)kp_addr
, NULL
, &offset
) ||
1891 !arch_kprobe_on_func_entry(offset
))
1897 int register_kretprobe(struct kretprobe
*rp
)
1900 struct kretprobe_instance
*inst
;
1904 if (!kprobe_on_func_entry(rp
->kp
.addr
, rp
->kp
.symbol_name
, rp
->kp
.offset
))
1907 if (kretprobe_blacklist_size
) {
1908 addr
= kprobe_addr(&rp
->kp
);
1910 return PTR_ERR(addr
);
1912 for (i
= 0; kretprobe_blacklist
[i
].name
!= NULL
; i
++) {
1913 if (kretprobe_blacklist
[i
].addr
== addr
)
1918 rp
->kp
.pre_handler
= pre_handler_kretprobe
;
1919 rp
->kp
.post_handler
= NULL
;
1920 rp
->kp
.fault_handler
= NULL
;
1922 /* Pre-allocate memory for max kretprobe instances */
1923 if (rp
->maxactive
<= 0) {
1924 #ifdef CONFIG_PREEMPT
1925 rp
->maxactive
= max_t(unsigned int, 10, 2*num_possible_cpus());
1927 rp
->maxactive
= num_possible_cpus();
1930 raw_spin_lock_init(&rp
->lock
);
1931 INIT_HLIST_HEAD(&rp
->free_instances
);
1932 for (i
= 0; i
< rp
->maxactive
; i
++) {
1933 inst
= kmalloc(sizeof(struct kretprobe_instance
) +
1934 rp
->data_size
, GFP_KERNEL
);
1939 INIT_HLIST_NODE(&inst
->hlist
);
1940 hlist_add_head(&inst
->hlist
, &rp
->free_instances
);
1944 /* Establish function entry probe point */
1945 ret
= register_kprobe(&rp
->kp
);
1950 EXPORT_SYMBOL_GPL(register_kretprobe
);
1952 int register_kretprobes(struct kretprobe
**rps
, int num
)
1958 for (i
= 0; i
< num
; i
++) {
1959 ret
= register_kretprobe(rps
[i
]);
1962 unregister_kretprobes(rps
, i
);
1968 EXPORT_SYMBOL_GPL(register_kretprobes
);
1970 void unregister_kretprobe(struct kretprobe
*rp
)
1972 unregister_kretprobes(&rp
, 1);
1974 EXPORT_SYMBOL_GPL(unregister_kretprobe
);
1976 void unregister_kretprobes(struct kretprobe
**rps
, int num
)
1982 mutex_lock(&kprobe_mutex
);
1983 for (i
= 0; i
< num
; i
++)
1984 if (__unregister_kprobe_top(&rps
[i
]->kp
) < 0)
1985 rps
[i
]->kp
.addr
= NULL
;
1986 mutex_unlock(&kprobe_mutex
);
1989 for (i
= 0; i
< num
; i
++) {
1990 if (rps
[i
]->kp
.addr
) {
1991 __unregister_kprobe_bottom(&rps
[i
]->kp
);
1992 cleanup_rp_inst(rps
[i
]);
1996 EXPORT_SYMBOL_GPL(unregister_kretprobes
);
1998 #else /* CONFIG_KRETPROBES */
1999 int register_kretprobe(struct kretprobe
*rp
)
2003 EXPORT_SYMBOL_GPL(register_kretprobe
);
2005 int register_kretprobes(struct kretprobe
**rps
, int num
)
2009 EXPORT_SYMBOL_GPL(register_kretprobes
);
2011 void unregister_kretprobe(struct kretprobe
*rp
)
2014 EXPORT_SYMBOL_GPL(unregister_kretprobe
);
2016 void unregister_kretprobes(struct kretprobe
**rps
, int num
)
2019 EXPORT_SYMBOL_GPL(unregister_kretprobes
);
2021 static int pre_handler_kretprobe(struct kprobe
*p
, struct pt_regs
*regs
)
2025 NOKPROBE_SYMBOL(pre_handler_kretprobe
);
2027 #endif /* CONFIG_KRETPROBES */
2029 /* Set the kprobe gone and remove its instruction buffer. */
2030 static void kill_kprobe(struct kprobe
*p
)
2034 p
->flags
|= KPROBE_FLAG_GONE
;
2035 if (kprobe_aggrprobe(p
)) {
2037 * If this is an aggr_kprobe, we have to list all the
2038 * chained probes and mark them GONE.
2040 list_for_each_entry_rcu(kp
, &p
->list
, list
)
2041 kp
->flags
|= KPROBE_FLAG_GONE
;
2042 p
->post_handler
= NULL
;
2043 kill_optimized_kprobe(p
);
2046 * Here, we can remove insn_slot safely, because no thread calls
2047 * the original probed function (which will be freed soon) any more.
2049 arch_remove_kprobe(p
);
2052 /* Disable one kprobe */
2053 int disable_kprobe(struct kprobe
*kp
)
2058 mutex_lock(&kprobe_mutex
);
2060 /* Disable this kprobe */
2061 p
= __disable_kprobe(kp
);
2065 mutex_unlock(&kprobe_mutex
);
2068 EXPORT_SYMBOL_GPL(disable_kprobe
);
2070 /* Enable one kprobe */
2071 int enable_kprobe(struct kprobe
*kp
)
2076 mutex_lock(&kprobe_mutex
);
2078 /* Check whether specified probe is valid. */
2079 p
= __get_valid_kprobe(kp
);
2080 if (unlikely(p
== NULL
)) {
2085 if (kprobe_gone(kp
)) {
2086 /* This kprobe has gone, we couldn't enable it. */
2092 kp
->flags
&= ~KPROBE_FLAG_DISABLED
;
2094 if (!kprobes_all_disarmed
&& kprobe_disabled(p
)) {
2095 p
->flags
&= ~KPROBE_FLAG_DISABLED
;
2096 ret
= arm_kprobe(p
);
2098 p
->flags
|= KPROBE_FLAG_DISABLED
;
2101 mutex_unlock(&kprobe_mutex
);
2104 EXPORT_SYMBOL_GPL(enable_kprobe
);
2106 /* Caller must NOT call this in usual path. This is only for critical case */
2107 void dump_kprobe(struct kprobe
*kp
)
2109 pr_err("Dumping kprobe:\n");
2110 pr_err("Name: %s\nOffset: %x\nAddress: %pS\n",
2111 kp
->symbol_name
, kp
->offset
, kp
->addr
);
2113 NOKPROBE_SYMBOL(dump_kprobe
);
2115 int kprobe_add_ksym_blacklist(unsigned long entry
)
2117 struct kprobe_blacklist_entry
*ent
;
2118 unsigned long offset
= 0, size
= 0;
2120 if (!kernel_text_address(entry
) ||
2121 !kallsyms_lookup_size_offset(entry
, &size
, &offset
))
2124 ent
= kmalloc(sizeof(*ent
), GFP_KERNEL
);
2127 ent
->start_addr
= entry
;
2128 ent
->end_addr
= entry
+ size
;
2129 INIT_LIST_HEAD(&ent
->list
);
2130 list_add_tail(&ent
->list
, &kprobe_blacklist
);
2135 /* Add all symbols in given area into kprobe blacklist */
2136 int kprobe_add_area_blacklist(unsigned long start
, unsigned long end
)
2138 unsigned long entry
;
2141 for (entry
= start
; entry
< end
; entry
+= ret
) {
2142 ret
= kprobe_add_ksym_blacklist(entry
);
2145 if (ret
== 0) /* In case of alias symbol */
2151 int __init __weak
arch_populate_kprobe_blacklist(void)
2157 * Lookup and populate the kprobe_blacklist.
2159 * Unlike the kretprobe blacklist, we'll need to determine
2160 * the range of addresses that belong to the said functions,
2161 * since a kprobe need not necessarily be at the beginning
2164 static int __init
populate_kprobe_blacklist(unsigned long *start
,
2167 unsigned long entry
;
2168 unsigned long *iter
;
2171 for (iter
= start
; iter
< end
; iter
++) {
2172 entry
= arch_deref_entry_point((void *)*iter
);
2173 ret
= kprobe_add_ksym_blacklist(entry
);
2180 /* Symbols in __kprobes_text are blacklisted */
2181 ret
= kprobe_add_area_blacklist((unsigned long)__kprobes_text_start
,
2182 (unsigned long)__kprobes_text_end
);
2184 return ret
? : arch_populate_kprobe_blacklist();
2187 /* Module notifier call back, checking kprobes on the module */
2188 static int kprobes_module_callback(struct notifier_block
*nb
,
2189 unsigned long val
, void *data
)
2191 struct module
*mod
= data
;
2192 struct hlist_head
*head
;
2195 int checkcore
= (val
== MODULE_STATE_GOING
);
2197 if (val
!= MODULE_STATE_GOING
&& val
!= MODULE_STATE_LIVE
)
2201 * When MODULE_STATE_GOING was notified, both of module .text and
2202 * .init.text sections would be freed. When MODULE_STATE_LIVE was
2203 * notified, only .init.text section would be freed. We need to
2204 * disable kprobes which have been inserted in the sections.
2206 mutex_lock(&kprobe_mutex
);
2207 for (i
= 0; i
< KPROBE_TABLE_SIZE
; i
++) {
2208 head
= &kprobe_table
[i
];
2209 hlist_for_each_entry_rcu(p
, head
, hlist
)
2210 if (within_module_init((unsigned long)p
->addr
, mod
) ||
2212 within_module_core((unsigned long)p
->addr
, mod
))) {
2214 * The vaddr this probe is installed will soon
2215 * be vfreed buy not synced to disk. Hence,
2216 * disarming the breakpoint isn't needed.
2218 * Note, this will also move any optimized probes
2219 * that are pending to be removed from their
2220 * corresponding lists to the freeing_list and
2221 * will not be touched by the delayed
2222 * kprobe_optimizer work handler.
2227 mutex_unlock(&kprobe_mutex
);
2231 static struct notifier_block kprobe_module_nb
= {
2232 .notifier_call
= kprobes_module_callback
,
2236 /* Markers of _kprobe_blacklist section */
2237 extern unsigned long __start_kprobe_blacklist
[];
2238 extern unsigned long __stop_kprobe_blacklist
[];
2240 static int __init
init_kprobes(void)
2244 /* FIXME allocate the probe table, currently defined statically */
2245 /* initialize all list heads */
2246 for (i
= 0; i
< KPROBE_TABLE_SIZE
; i
++) {
2247 INIT_HLIST_HEAD(&kprobe_table
[i
]);
2248 INIT_HLIST_HEAD(&kretprobe_inst_table
[i
]);
2249 raw_spin_lock_init(&(kretprobe_table_locks
[i
].lock
));
2252 err
= populate_kprobe_blacklist(__start_kprobe_blacklist
,
2253 __stop_kprobe_blacklist
);
2255 pr_err("kprobes: failed to populate blacklist: %d\n", err
);
2256 pr_err("Please take care of using kprobes.\n");
2259 if (kretprobe_blacklist_size
) {
2260 /* lookup the function address from its name */
2261 for (i
= 0; kretprobe_blacklist
[i
].name
!= NULL
; i
++) {
2262 kretprobe_blacklist
[i
].addr
=
2263 kprobe_lookup_name(kretprobe_blacklist
[i
].name
, 0);
2264 if (!kretprobe_blacklist
[i
].addr
)
2265 printk("kretprobe: lookup failed: %s\n",
2266 kretprobe_blacklist
[i
].name
);
2270 #if defined(CONFIG_OPTPROBES)
2271 #if defined(__ARCH_WANT_KPROBES_INSN_SLOT)
2272 /* Init kprobe_optinsn_slots */
2273 kprobe_optinsn_slots
.insn_size
= MAX_OPTINSN_SIZE
;
2275 /* By default, kprobes can be optimized */
2276 kprobes_allow_optimization
= true;
2279 /* By default, kprobes are armed */
2280 kprobes_all_disarmed
= false;
2282 err
= arch_init_kprobes();
2284 err
= register_die_notifier(&kprobe_exceptions_nb
);
2286 err
= register_module_notifier(&kprobe_module_nb
);
2288 kprobes_initialized
= (err
== 0);
2295 #ifdef CONFIG_DEBUG_FS
2296 static void report_probe(struct seq_file
*pi
, struct kprobe
*p
,
2297 const char *sym
, int offset
, char *modname
, struct kprobe
*pp
)
2300 void *addr
= p
->addr
;
2302 if (p
->pre_handler
== pre_handler_kretprobe
)
2307 if (!kallsyms_show_value())
2311 seq_printf(pi
, "%px %s %s+0x%x %s ",
2312 addr
, kprobe_type
, sym
, offset
,
2313 (modname
? modname
: " "));
2314 else /* try to use %pS */
2315 seq_printf(pi
, "%px %s %pS ",
2316 addr
, kprobe_type
, p
->addr
);
2320 seq_printf(pi
, "%s%s%s%s\n",
2321 (kprobe_gone(p
) ? "[GONE]" : ""),
2322 ((kprobe_disabled(p
) && !kprobe_gone(p
)) ? "[DISABLED]" : ""),
2323 (kprobe_optimized(pp
) ? "[OPTIMIZED]" : ""),
2324 (kprobe_ftrace(pp
) ? "[FTRACE]" : ""));
2327 static void *kprobe_seq_start(struct seq_file
*f
, loff_t
*pos
)
2329 return (*pos
< KPROBE_TABLE_SIZE
) ? pos
: NULL
;
2332 static void *kprobe_seq_next(struct seq_file
*f
, void *v
, loff_t
*pos
)
2335 if (*pos
>= KPROBE_TABLE_SIZE
)
2340 static void kprobe_seq_stop(struct seq_file
*f
, void *v
)
2345 static int show_kprobe_addr(struct seq_file
*pi
, void *v
)
2347 struct hlist_head
*head
;
2348 struct kprobe
*p
, *kp
;
2349 const char *sym
= NULL
;
2350 unsigned int i
= *(loff_t
*) v
;
2351 unsigned long offset
= 0;
2352 char *modname
, namebuf
[KSYM_NAME_LEN
];
2354 head
= &kprobe_table
[i
];
2356 hlist_for_each_entry_rcu(p
, head
, hlist
) {
2357 sym
= kallsyms_lookup((unsigned long)p
->addr
, NULL
,
2358 &offset
, &modname
, namebuf
);
2359 if (kprobe_aggrprobe(p
)) {
2360 list_for_each_entry_rcu(kp
, &p
->list
, list
)
2361 report_probe(pi
, kp
, sym
, offset
, modname
, p
);
2363 report_probe(pi
, p
, sym
, offset
, modname
, NULL
);
2369 static const struct seq_operations kprobes_seq_ops
= {
2370 .start
= kprobe_seq_start
,
2371 .next
= kprobe_seq_next
,
2372 .stop
= kprobe_seq_stop
,
2373 .show
= show_kprobe_addr
2376 static int kprobes_open(struct inode
*inode
, struct file
*filp
)
2378 return seq_open(filp
, &kprobes_seq_ops
);
2381 static const struct file_operations debugfs_kprobes_operations
= {
2382 .open
= kprobes_open
,
2384 .llseek
= seq_lseek
,
2385 .release
= seq_release
,
2388 /* kprobes/blacklist -- shows which functions can not be probed */
2389 static void *kprobe_blacklist_seq_start(struct seq_file
*m
, loff_t
*pos
)
2391 return seq_list_start(&kprobe_blacklist
, *pos
);
2394 static void *kprobe_blacklist_seq_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
2396 return seq_list_next(v
, &kprobe_blacklist
, pos
);
2399 static int kprobe_blacklist_seq_show(struct seq_file
*m
, void *v
)
2401 struct kprobe_blacklist_entry
*ent
=
2402 list_entry(v
, struct kprobe_blacklist_entry
, list
);
2405 * If /proc/kallsyms is not showing kernel address, we won't
2406 * show them here either.
2408 if (!kallsyms_show_value())
2409 seq_printf(m
, "0x%px-0x%px\t%ps\n", NULL
, NULL
,
2410 (void *)ent
->start_addr
);
2412 seq_printf(m
, "0x%px-0x%px\t%ps\n", (void *)ent
->start_addr
,
2413 (void *)ent
->end_addr
, (void *)ent
->start_addr
);
2417 static const struct seq_operations kprobe_blacklist_seq_ops
= {
2418 .start
= kprobe_blacklist_seq_start
,
2419 .next
= kprobe_blacklist_seq_next
,
2420 .stop
= kprobe_seq_stop
, /* Reuse void function */
2421 .show
= kprobe_blacklist_seq_show
,
2424 static int kprobe_blacklist_open(struct inode
*inode
, struct file
*filp
)
2426 return seq_open(filp
, &kprobe_blacklist_seq_ops
);
2429 static const struct file_operations debugfs_kprobe_blacklist_ops
= {
2430 .open
= kprobe_blacklist_open
,
2432 .llseek
= seq_lseek
,
2433 .release
= seq_release
,
2436 static int arm_all_kprobes(void)
2438 struct hlist_head
*head
;
2440 unsigned int i
, total
= 0, errors
= 0;
2443 mutex_lock(&kprobe_mutex
);
2445 /* If kprobes are armed, just return */
2446 if (!kprobes_all_disarmed
)
2447 goto already_enabled
;
2450 * optimize_kprobe() called by arm_kprobe() checks
2451 * kprobes_all_disarmed, so set kprobes_all_disarmed before
2454 kprobes_all_disarmed
= false;
2455 /* Arming kprobes doesn't optimize kprobe itself */
2456 for (i
= 0; i
< KPROBE_TABLE_SIZE
; i
++) {
2457 head
= &kprobe_table
[i
];
2458 /* Arm all kprobes on a best-effort basis */
2459 hlist_for_each_entry_rcu(p
, head
, hlist
) {
2460 if (!kprobe_disabled(p
)) {
2461 err
= arm_kprobe(p
);
2472 pr_warn("Kprobes globally enabled, but failed to arm %d out of %d probes\n",
2475 pr_info("Kprobes globally enabled\n");
2478 mutex_unlock(&kprobe_mutex
);
2482 static int disarm_all_kprobes(void)
2484 struct hlist_head
*head
;
2486 unsigned int i
, total
= 0, errors
= 0;
2489 mutex_lock(&kprobe_mutex
);
2491 /* If kprobes are already disarmed, just return */
2492 if (kprobes_all_disarmed
) {
2493 mutex_unlock(&kprobe_mutex
);
2497 kprobes_all_disarmed
= true;
2499 for (i
= 0; i
< KPROBE_TABLE_SIZE
; i
++) {
2500 head
= &kprobe_table
[i
];
2501 /* Disarm all kprobes on a best-effort basis */
2502 hlist_for_each_entry_rcu(p
, head
, hlist
) {
2503 if (!arch_trampoline_kprobe(p
) && !kprobe_disabled(p
)) {
2504 err
= disarm_kprobe(p
, false);
2515 pr_warn("Kprobes globally disabled, but failed to disarm %d out of %d probes\n",
2518 pr_info("Kprobes globally disabled\n");
2520 mutex_unlock(&kprobe_mutex
);
2522 /* Wait for disarming all kprobes by optimizer */
2523 wait_for_kprobe_optimizer();
2529 * XXX: The debugfs bool file interface doesn't allow for callbacks
2530 * when the bool state is switched. We can reuse that facility when
2533 static ssize_t
read_enabled_file_bool(struct file
*file
,
2534 char __user
*user_buf
, size_t count
, loff_t
*ppos
)
2538 if (!kprobes_all_disarmed
)
2544 return simple_read_from_buffer(user_buf
, count
, ppos
, buf
, 2);
2547 static ssize_t
write_enabled_file_bool(struct file
*file
,
2548 const char __user
*user_buf
, size_t count
, loff_t
*ppos
)
2554 buf_size
= min(count
, (sizeof(buf
)-1));
2555 if (copy_from_user(buf
, user_buf
, buf_size
))
2558 buf
[buf_size
] = '\0';
2563 ret
= arm_all_kprobes();
2568 ret
= disarm_all_kprobes();
2580 static const struct file_operations fops_kp
= {
2581 .read
= read_enabled_file_bool
,
2582 .write
= write_enabled_file_bool
,
2583 .llseek
= default_llseek
,
2586 static int __init
debugfs_kprobe_init(void)
2588 struct dentry
*dir
, *file
;
2589 unsigned int value
= 1;
2591 dir
= debugfs_create_dir("kprobes", NULL
);
2595 file
= debugfs_create_file("list", 0400, dir
, NULL
,
2596 &debugfs_kprobes_operations
);
2600 file
= debugfs_create_file("enabled", 0600, dir
,
2605 file
= debugfs_create_file("blacklist", 0400, dir
, NULL
,
2606 &debugfs_kprobe_blacklist_ops
);
2613 debugfs_remove(dir
);
2617 late_initcall(debugfs_kprobe_init
);
2618 #endif /* CONFIG_DEBUG_FS */
2620 module_init(init_kprobes
);