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-generic/sections.h>
53 #include <asm/cacheflush.h>
54 #include <asm/errno.h>
55 #include <asm/uaccess.h>
57 #define KPROBE_HASH_BITS 6
58 #define KPROBE_TABLE_SIZE (1 << KPROBE_HASH_BITS)
62 * Some oddball architectures like 64bit powerpc have function descriptors
63 * so this must be overridable.
65 #ifndef kprobe_lookup_name
66 #define kprobe_lookup_name(name, addr) \
67 addr = ((kprobe_opcode_t *)(kallsyms_lookup_name(name)))
70 static int kprobes_initialized
;
71 static struct hlist_head kprobe_table
[KPROBE_TABLE_SIZE
];
72 static struct hlist_head kretprobe_inst_table
[KPROBE_TABLE_SIZE
];
74 /* NOTE: change this value only with kprobe_mutex held */
75 static bool kprobes_all_disarmed
;
77 /* This protects kprobe_table and optimizing_list */
78 static DEFINE_MUTEX(kprobe_mutex
);
79 static DEFINE_PER_CPU(struct kprobe
*, kprobe_instance
) = NULL
;
81 raw_spinlock_t lock ____cacheline_aligned_in_smp
;
82 } kretprobe_table_locks
[KPROBE_TABLE_SIZE
];
84 static raw_spinlock_t
*kretprobe_table_lock_ptr(unsigned long hash
)
86 return &(kretprobe_table_locks
[hash
].lock
);
90 * Normally, functions that we'd want to prohibit kprobes in, are marked
91 * __kprobes. But, there are cases where such functions already belong to
92 * a different section (__sched for preempt_schedule)
94 * For such cases, we now have a blacklist
96 static struct kprobe_blackpoint kprobe_blacklist
[] = {
97 {"preempt_schedule",},
98 {"native_get_debugreg",},
99 {NULL
} /* Terminator */
102 #ifdef __ARCH_WANT_KPROBES_INSN_SLOT
104 * kprobe->ainsn.insn points to the copy of the instruction to be
105 * single-stepped. x86_64, POWER4 and above have no-exec support and
106 * stepping on the instruction on a vmalloced/kmalloced/data page
107 * is a recipe for disaster
109 struct kprobe_insn_page
{
110 struct list_head list
;
111 kprobe_opcode_t
*insns
; /* Page of instruction slots */
112 struct kprobe_insn_cache
*cache
;
118 #define KPROBE_INSN_PAGE_SIZE(slots) \
119 (offsetof(struct kprobe_insn_page, slot_used) + \
120 (sizeof(char) * (slots)))
122 static int slots_per_page(struct kprobe_insn_cache
*c
)
124 return PAGE_SIZE
/(c
->insn_size
* sizeof(kprobe_opcode_t
));
127 enum kprobe_slot_state
{
133 static void *alloc_insn_page(void)
135 return module_alloc(PAGE_SIZE
);
138 static void free_insn_page(void *page
)
140 module_free(NULL
, page
);
143 struct kprobe_insn_cache kprobe_insn_slots
= {
144 .mutex
= __MUTEX_INITIALIZER(kprobe_insn_slots
.mutex
),
145 .alloc
= alloc_insn_page
,
146 .free
= free_insn_page
,
147 .pages
= LIST_HEAD_INIT(kprobe_insn_slots
.pages
),
148 .insn_size
= MAX_INSN_SIZE
,
151 static int __kprobes
collect_garbage_slots(struct kprobe_insn_cache
*c
);
154 * __get_insn_slot() - Find a slot on an executable page for an instruction.
155 * We allocate an executable page if there's no room on existing ones.
157 kprobe_opcode_t __kprobes
*__get_insn_slot(struct kprobe_insn_cache
*c
)
159 struct kprobe_insn_page
*kip
;
160 kprobe_opcode_t
*slot
= NULL
;
162 mutex_lock(&c
->mutex
);
164 list_for_each_entry(kip
, &c
->pages
, list
) {
165 if (kip
->nused
< slots_per_page(c
)) {
167 for (i
= 0; i
< slots_per_page(c
); i
++) {
168 if (kip
->slot_used
[i
] == SLOT_CLEAN
) {
169 kip
->slot_used
[i
] = SLOT_USED
;
171 slot
= kip
->insns
+ (i
* c
->insn_size
);
175 /* kip->nused is broken. Fix it. */
176 kip
->nused
= slots_per_page(c
);
181 /* If there are any garbage slots, collect it and try again. */
182 if (c
->nr_garbage
&& collect_garbage_slots(c
) == 0)
185 /* All out of space. Need to allocate a new page. */
186 kip
= kmalloc(KPROBE_INSN_PAGE_SIZE(slots_per_page(c
)), GFP_KERNEL
);
191 * Use module_alloc so this page is within +/- 2GB of where the
192 * kernel image and loaded module images reside. This is required
193 * so x86_64 can correctly handle the %rip-relative fixups.
195 kip
->insns
= c
->alloc();
200 INIT_LIST_HEAD(&kip
->list
);
201 memset(kip
->slot_used
, SLOT_CLEAN
, slots_per_page(c
));
202 kip
->slot_used
[0] = SLOT_USED
;
206 list_add(&kip
->list
, &c
->pages
);
209 mutex_unlock(&c
->mutex
);
213 /* Return 1 if all garbages are collected, otherwise 0. */
214 static int __kprobes
collect_one_slot(struct kprobe_insn_page
*kip
, int idx
)
216 kip
->slot_used
[idx
] = SLOT_CLEAN
;
218 if (kip
->nused
== 0) {
220 * Page is no longer in use. Free it unless
221 * it's the last one. We keep the last one
222 * so as not to have to set it up again the
223 * next time somebody inserts a probe.
225 if (!list_is_singular(&kip
->list
)) {
226 list_del(&kip
->list
);
227 kip
->cache
->free(kip
->insns
);
235 static int __kprobes
collect_garbage_slots(struct kprobe_insn_cache
*c
)
237 struct kprobe_insn_page
*kip
, *next
;
239 /* Ensure no-one is interrupted on the garbages */
242 list_for_each_entry_safe(kip
, next
, &c
->pages
, list
) {
244 if (kip
->ngarbage
== 0)
246 kip
->ngarbage
= 0; /* we will collect all garbages */
247 for (i
= 0; i
< slots_per_page(c
); i
++) {
248 if (kip
->slot_used
[i
] == SLOT_DIRTY
&&
249 collect_one_slot(kip
, i
))
257 void __kprobes
__free_insn_slot(struct kprobe_insn_cache
*c
,
258 kprobe_opcode_t
*slot
, int dirty
)
260 struct kprobe_insn_page
*kip
;
262 mutex_lock(&c
->mutex
);
263 list_for_each_entry(kip
, &c
->pages
, list
) {
264 long idx
= ((long)slot
- (long)kip
->insns
) /
265 (c
->insn_size
* sizeof(kprobe_opcode_t
));
266 if (idx
>= 0 && idx
< slots_per_page(c
)) {
267 WARN_ON(kip
->slot_used
[idx
] != SLOT_USED
);
269 kip
->slot_used
[idx
] = SLOT_DIRTY
;
271 if (++c
->nr_garbage
> slots_per_page(c
))
272 collect_garbage_slots(c
);
274 collect_one_slot(kip
, idx
);
278 /* Could not free this slot. */
281 mutex_unlock(&c
->mutex
);
284 #ifdef CONFIG_OPTPROBES
285 /* For optimized_kprobe buffer */
286 struct kprobe_insn_cache kprobe_optinsn_slots
= {
287 .mutex
= __MUTEX_INITIALIZER(kprobe_optinsn_slots
.mutex
),
288 .alloc
= alloc_insn_page
,
289 .free
= free_insn_page
,
290 .pages
= LIST_HEAD_INIT(kprobe_optinsn_slots
.pages
),
291 /* .insn_size is initialized later */
297 /* We have preemption disabled.. so it is safe to use __ versions */
298 static inline void set_kprobe_instance(struct kprobe
*kp
)
300 __this_cpu_write(kprobe_instance
, kp
);
303 static inline void reset_kprobe_instance(void)
305 __this_cpu_write(kprobe_instance
, NULL
);
309 * This routine is called either:
310 * - under the kprobe_mutex - during kprobe_[un]register()
312 * - with preemption disabled - from arch/xxx/kernel/kprobes.c
314 struct kprobe __kprobes
*get_kprobe(void *addr
)
316 struct hlist_head
*head
;
319 head
= &kprobe_table
[hash_ptr(addr
, KPROBE_HASH_BITS
)];
320 hlist_for_each_entry_rcu(p
, head
, hlist
) {
328 static int __kprobes
aggr_pre_handler(struct kprobe
*p
, struct pt_regs
*regs
);
330 /* Return true if the kprobe is an aggregator */
331 static inline int kprobe_aggrprobe(struct kprobe
*p
)
333 return p
->pre_handler
== aggr_pre_handler
;
336 /* Return true(!0) if the kprobe is unused */
337 static inline int kprobe_unused(struct kprobe
*p
)
339 return kprobe_aggrprobe(p
) && kprobe_disabled(p
) &&
340 list_empty(&p
->list
);
344 * Keep all fields in the kprobe consistent
346 static inline void copy_kprobe(struct kprobe
*ap
, struct kprobe
*p
)
348 memcpy(&p
->opcode
, &ap
->opcode
, sizeof(kprobe_opcode_t
));
349 memcpy(&p
->ainsn
, &ap
->ainsn
, sizeof(struct arch_specific_insn
));
352 #ifdef CONFIG_OPTPROBES
353 /* NOTE: change this value only with kprobe_mutex held */
354 static bool kprobes_allow_optimization
;
357 * Call all pre_handler on the list, but ignores its return value.
358 * This must be called from arch-dep optimized caller.
360 void __kprobes
opt_pre_handler(struct kprobe
*p
, struct pt_regs
*regs
)
364 list_for_each_entry_rcu(kp
, &p
->list
, list
) {
365 if (kp
->pre_handler
&& likely(!kprobe_disabled(kp
))) {
366 set_kprobe_instance(kp
);
367 kp
->pre_handler(kp
, regs
);
369 reset_kprobe_instance();
373 /* Free optimized instructions and optimized_kprobe */
374 static __kprobes
void free_aggr_kprobe(struct kprobe
*p
)
376 struct optimized_kprobe
*op
;
378 op
= container_of(p
, struct optimized_kprobe
, kp
);
379 arch_remove_optimized_kprobe(op
);
380 arch_remove_kprobe(p
);
384 /* Return true(!0) if the kprobe is ready for optimization. */
385 static inline int kprobe_optready(struct kprobe
*p
)
387 struct optimized_kprobe
*op
;
389 if (kprobe_aggrprobe(p
)) {
390 op
= container_of(p
, struct optimized_kprobe
, kp
);
391 return arch_prepared_optinsn(&op
->optinsn
);
397 /* Return true(!0) if the kprobe is disarmed. Note: p must be on hash list */
398 static inline int kprobe_disarmed(struct kprobe
*p
)
400 struct optimized_kprobe
*op
;
402 /* If kprobe is not aggr/opt probe, just return kprobe is disabled */
403 if (!kprobe_aggrprobe(p
))
404 return kprobe_disabled(p
);
406 op
= container_of(p
, struct optimized_kprobe
, kp
);
408 return kprobe_disabled(p
) && list_empty(&op
->list
);
411 /* Return true(!0) if the probe is queued on (un)optimizing lists */
412 static int __kprobes
kprobe_queued(struct kprobe
*p
)
414 struct optimized_kprobe
*op
;
416 if (kprobe_aggrprobe(p
)) {
417 op
= container_of(p
, struct optimized_kprobe
, kp
);
418 if (!list_empty(&op
->list
))
425 * Return an optimized kprobe whose optimizing code replaces
426 * instructions including addr (exclude breakpoint).
428 static struct kprobe
*__kprobes
get_optimized_kprobe(unsigned long addr
)
431 struct kprobe
*p
= NULL
;
432 struct optimized_kprobe
*op
;
434 /* Don't check i == 0, since that is a breakpoint case. */
435 for (i
= 1; !p
&& i
< MAX_OPTIMIZED_LENGTH
; i
++)
436 p
= get_kprobe((void *)(addr
- i
));
438 if (p
&& kprobe_optready(p
)) {
439 op
= container_of(p
, struct optimized_kprobe
, kp
);
440 if (arch_within_optimized_kprobe(op
, addr
))
447 /* Optimization staging list, protected by kprobe_mutex */
448 static LIST_HEAD(optimizing_list
);
449 static LIST_HEAD(unoptimizing_list
);
450 static LIST_HEAD(freeing_list
);
452 static void kprobe_optimizer(struct work_struct
*work
);
453 static DECLARE_DELAYED_WORK(optimizing_work
, kprobe_optimizer
);
454 #define OPTIMIZE_DELAY 5
457 * Optimize (replace a breakpoint with a jump) kprobes listed on
460 static __kprobes
void do_optimize_kprobes(void)
462 /* Optimization never be done when disarmed */
463 if (kprobes_all_disarmed
|| !kprobes_allow_optimization
||
464 list_empty(&optimizing_list
))
468 * The optimization/unoptimization refers online_cpus via
469 * stop_machine() and cpu-hotplug modifies online_cpus.
470 * And same time, text_mutex will be held in cpu-hotplug and here.
471 * This combination can cause a deadlock (cpu-hotplug try to lock
472 * text_mutex but stop_machine can not be done because online_cpus
474 * To avoid this deadlock, we need to call get_online_cpus()
475 * for preventing cpu-hotplug outside of text_mutex locking.
478 mutex_lock(&text_mutex
);
479 arch_optimize_kprobes(&optimizing_list
);
480 mutex_unlock(&text_mutex
);
485 * Unoptimize (replace a jump with a breakpoint and remove the breakpoint
486 * if need) kprobes listed on unoptimizing_list.
488 static __kprobes
void do_unoptimize_kprobes(void)
490 struct optimized_kprobe
*op
, *tmp
;
492 /* Unoptimization must be done anytime */
493 if (list_empty(&unoptimizing_list
))
496 /* Ditto to do_optimize_kprobes */
498 mutex_lock(&text_mutex
);
499 arch_unoptimize_kprobes(&unoptimizing_list
, &freeing_list
);
500 /* Loop free_list for disarming */
501 list_for_each_entry_safe(op
, tmp
, &freeing_list
, list
) {
502 /* Disarm probes if marked disabled */
503 if (kprobe_disabled(&op
->kp
))
504 arch_disarm_kprobe(&op
->kp
);
505 if (kprobe_unused(&op
->kp
)) {
507 * Remove unused probes from hash list. After waiting
508 * for synchronization, these probes are reclaimed.
509 * (reclaiming is done by do_free_cleaned_kprobes.)
511 hlist_del_rcu(&op
->kp
.hlist
);
513 list_del_init(&op
->list
);
515 mutex_unlock(&text_mutex
);
519 /* Reclaim all kprobes on the free_list */
520 static __kprobes
void do_free_cleaned_kprobes(void)
522 struct optimized_kprobe
*op
, *tmp
;
524 list_for_each_entry_safe(op
, tmp
, &freeing_list
, list
) {
525 BUG_ON(!kprobe_unused(&op
->kp
));
526 list_del_init(&op
->list
);
527 free_aggr_kprobe(&op
->kp
);
531 /* Start optimizer after OPTIMIZE_DELAY passed */
532 static __kprobes
void kick_kprobe_optimizer(void)
534 schedule_delayed_work(&optimizing_work
, OPTIMIZE_DELAY
);
537 /* Kprobe jump optimizer */
538 static __kprobes
void kprobe_optimizer(struct work_struct
*work
)
540 mutex_lock(&kprobe_mutex
);
541 /* Lock modules while optimizing kprobes */
542 mutex_lock(&module_mutex
);
545 * Step 1: Unoptimize kprobes and collect cleaned (unused and disarmed)
546 * kprobes before waiting for quiesence period.
548 do_unoptimize_kprobes();
551 * Step 2: Wait for quiesence period to ensure all running interrupts
552 * are done. Because optprobe may modify multiple instructions
553 * there is a chance that Nth instruction is interrupted. In that
554 * case, running interrupt can return to 2nd-Nth byte of jump
555 * instruction. This wait is for avoiding it.
559 /* Step 3: Optimize kprobes after quiesence period */
560 do_optimize_kprobes();
562 /* Step 4: Free cleaned kprobes after quiesence period */
563 do_free_cleaned_kprobes();
565 mutex_unlock(&module_mutex
);
566 mutex_unlock(&kprobe_mutex
);
568 /* Step 5: Kick optimizer again if needed */
569 if (!list_empty(&optimizing_list
) || !list_empty(&unoptimizing_list
))
570 kick_kprobe_optimizer();
573 /* Wait for completing optimization and unoptimization */
574 static __kprobes
void wait_for_kprobe_optimizer(void)
576 mutex_lock(&kprobe_mutex
);
578 while (!list_empty(&optimizing_list
) || !list_empty(&unoptimizing_list
)) {
579 mutex_unlock(&kprobe_mutex
);
581 /* this will also make optimizing_work execute immmediately */
582 flush_delayed_work(&optimizing_work
);
583 /* @optimizing_work might not have been queued yet, relax */
586 mutex_lock(&kprobe_mutex
);
589 mutex_unlock(&kprobe_mutex
);
592 /* Optimize kprobe if p is ready to be optimized */
593 static __kprobes
void optimize_kprobe(struct kprobe
*p
)
595 struct optimized_kprobe
*op
;
597 /* Check if the kprobe is disabled or not ready for optimization. */
598 if (!kprobe_optready(p
) || !kprobes_allow_optimization
||
599 (kprobe_disabled(p
) || kprobes_all_disarmed
))
602 /* Both of break_handler and post_handler are not supported. */
603 if (p
->break_handler
|| p
->post_handler
)
606 op
= container_of(p
, struct optimized_kprobe
, kp
);
608 /* Check there is no other kprobes at the optimized instructions */
609 if (arch_check_optimized_kprobe(op
) < 0)
612 /* Check if it is already optimized. */
613 if (op
->kp
.flags
& KPROBE_FLAG_OPTIMIZED
)
615 op
->kp
.flags
|= KPROBE_FLAG_OPTIMIZED
;
617 if (!list_empty(&op
->list
))
618 /* This is under unoptimizing. Just dequeue the probe */
619 list_del_init(&op
->list
);
621 list_add(&op
->list
, &optimizing_list
);
622 kick_kprobe_optimizer();
626 /* Short cut to direct unoptimizing */
627 static __kprobes
void force_unoptimize_kprobe(struct optimized_kprobe
*op
)
630 arch_unoptimize_kprobe(op
);
632 if (kprobe_disabled(&op
->kp
))
633 arch_disarm_kprobe(&op
->kp
);
636 /* Unoptimize a kprobe if p is optimized */
637 static __kprobes
void unoptimize_kprobe(struct kprobe
*p
, bool force
)
639 struct optimized_kprobe
*op
;
641 if (!kprobe_aggrprobe(p
) || kprobe_disarmed(p
))
642 return; /* This is not an optprobe nor optimized */
644 op
= container_of(p
, struct optimized_kprobe
, kp
);
645 if (!kprobe_optimized(p
)) {
646 /* Unoptimized or unoptimizing case */
647 if (force
&& !list_empty(&op
->list
)) {
649 * Only if this is unoptimizing kprobe and forced,
650 * forcibly unoptimize it. (No need to unoptimize
651 * unoptimized kprobe again :)
653 list_del_init(&op
->list
);
654 force_unoptimize_kprobe(op
);
659 op
->kp
.flags
&= ~KPROBE_FLAG_OPTIMIZED
;
660 if (!list_empty(&op
->list
)) {
661 /* Dequeue from the optimization queue */
662 list_del_init(&op
->list
);
665 /* Optimized kprobe case */
667 /* Forcibly update the code: this is a special case */
668 force_unoptimize_kprobe(op
);
670 list_add(&op
->list
, &unoptimizing_list
);
671 kick_kprobe_optimizer();
675 /* Cancel unoptimizing for reusing */
676 static void reuse_unused_kprobe(struct kprobe
*ap
)
678 struct optimized_kprobe
*op
;
680 BUG_ON(!kprobe_unused(ap
));
682 * Unused kprobe MUST be on the way of delayed unoptimizing (means
683 * there is still a relative jump) and disabled.
685 op
= container_of(ap
, struct optimized_kprobe
, kp
);
686 if (unlikely(list_empty(&op
->list
)))
687 printk(KERN_WARNING
"Warning: found a stray unused "
688 "aggrprobe@%p\n", ap
->addr
);
689 /* Enable the probe again */
690 ap
->flags
&= ~KPROBE_FLAG_DISABLED
;
691 /* Optimize it again (remove from op->list) */
692 BUG_ON(!kprobe_optready(ap
));
696 /* Remove optimized instructions */
697 static void __kprobes
kill_optimized_kprobe(struct kprobe
*p
)
699 struct optimized_kprobe
*op
;
701 op
= container_of(p
, struct optimized_kprobe
, kp
);
702 if (!list_empty(&op
->list
))
703 /* Dequeue from the (un)optimization queue */
704 list_del_init(&op
->list
);
705 op
->kp
.flags
&= ~KPROBE_FLAG_OPTIMIZED
;
707 if (kprobe_unused(p
)) {
708 /* Enqueue if it is unused */
709 list_add(&op
->list
, &freeing_list
);
711 * Remove unused probes from the hash list. After waiting
712 * for synchronization, this probe is reclaimed.
713 * (reclaiming is done by do_free_cleaned_kprobes().)
715 hlist_del_rcu(&op
->kp
.hlist
);
718 /* Don't touch the code, because it is already freed. */
719 arch_remove_optimized_kprobe(op
);
722 /* Try to prepare optimized instructions */
723 static __kprobes
void prepare_optimized_kprobe(struct kprobe
*p
)
725 struct optimized_kprobe
*op
;
727 op
= container_of(p
, struct optimized_kprobe
, kp
);
728 arch_prepare_optimized_kprobe(op
);
731 /* Allocate new optimized_kprobe and try to prepare optimized instructions */
732 static __kprobes
struct kprobe
*alloc_aggr_kprobe(struct kprobe
*p
)
734 struct optimized_kprobe
*op
;
736 op
= kzalloc(sizeof(struct optimized_kprobe
), GFP_KERNEL
);
740 INIT_LIST_HEAD(&op
->list
);
741 op
->kp
.addr
= p
->addr
;
742 arch_prepare_optimized_kprobe(op
);
747 static void __kprobes
init_aggr_kprobe(struct kprobe
*ap
, struct kprobe
*p
);
750 * Prepare an optimized_kprobe and optimize it
751 * NOTE: p must be a normal registered kprobe
753 static __kprobes
void try_to_optimize_kprobe(struct kprobe
*p
)
756 struct optimized_kprobe
*op
;
758 /* Impossible to optimize ftrace-based kprobe */
759 if (kprobe_ftrace(p
))
762 /* For preparing optimization, jump_label_text_reserved() is called */
764 mutex_lock(&text_mutex
);
766 ap
= alloc_aggr_kprobe(p
);
770 op
= container_of(ap
, struct optimized_kprobe
, kp
);
771 if (!arch_prepared_optinsn(&op
->optinsn
)) {
772 /* If failed to setup optimizing, fallback to kprobe */
773 arch_remove_optimized_kprobe(op
);
778 init_aggr_kprobe(ap
, p
);
779 optimize_kprobe(ap
); /* This just kicks optimizer thread */
782 mutex_unlock(&text_mutex
);
787 static void __kprobes
optimize_all_kprobes(void)
789 struct hlist_head
*head
;
793 mutex_lock(&kprobe_mutex
);
794 /* If optimization is already allowed, just return */
795 if (kprobes_allow_optimization
)
798 kprobes_allow_optimization
= true;
799 for (i
= 0; i
< KPROBE_TABLE_SIZE
; i
++) {
800 head
= &kprobe_table
[i
];
801 hlist_for_each_entry_rcu(p
, head
, hlist
)
802 if (!kprobe_disabled(p
))
805 printk(KERN_INFO
"Kprobes globally optimized\n");
807 mutex_unlock(&kprobe_mutex
);
810 static void __kprobes
unoptimize_all_kprobes(void)
812 struct hlist_head
*head
;
816 mutex_lock(&kprobe_mutex
);
817 /* If optimization is already prohibited, just return */
818 if (!kprobes_allow_optimization
) {
819 mutex_unlock(&kprobe_mutex
);
823 kprobes_allow_optimization
= false;
824 for (i
= 0; i
< KPROBE_TABLE_SIZE
; i
++) {
825 head
= &kprobe_table
[i
];
826 hlist_for_each_entry_rcu(p
, head
, hlist
) {
827 if (!kprobe_disabled(p
))
828 unoptimize_kprobe(p
, false);
831 mutex_unlock(&kprobe_mutex
);
833 /* Wait for unoptimizing completion */
834 wait_for_kprobe_optimizer();
835 printk(KERN_INFO
"Kprobes globally unoptimized\n");
838 static DEFINE_MUTEX(kprobe_sysctl_mutex
);
839 int sysctl_kprobes_optimization
;
840 int proc_kprobes_optimization_handler(struct ctl_table
*table
, int write
,
841 void __user
*buffer
, size_t *length
,
846 mutex_lock(&kprobe_sysctl_mutex
);
847 sysctl_kprobes_optimization
= kprobes_allow_optimization
? 1 : 0;
848 ret
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
850 if (sysctl_kprobes_optimization
)
851 optimize_all_kprobes();
853 unoptimize_all_kprobes();
854 mutex_unlock(&kprobe_sysctl_mutex
);
858 #endif /* CONFIG_SYSCTL */
860 /* Put a breakpoint for a probe. Must be called with text_mutex locked */
861 static void __kprobes
__arm_kprobe(struct kprobe
*p
)
865 /* Check collision with other optimized kprobes */
866 _p
= get_optimized_kprobe((unsigned long)p
->addr
);
868 /* Fallback to unoptimized kprobe */
869 unoptimize_kprobe(_p
, true);
872 optimize_kprobe(p
); /* Try to optimize (add kprobe to a list) */
875 /* Remove the breakpoint of a probe. Must be called with text_mutex locked */
876 static void __kprobes
__disarm_kprobe(struct kprobe
*p
, bool reopt
)
880 unoptimize_kprobe(p
, false); /* Try to unoptimize */
882 if (!kprobe_queued(p
)) {
883 arch_disarm_kprobe(p
);
884 /* If another kprobe was blocked, optimize it. */
885 _p
= get_optimized_kprobe((unsigned long)p
->addr
);
886 if (unlikely(_p
) && reopt
)
889 /* TODO: reoptimize others after unoptimized this probe */
892 #else /* !CONFIG_OPTPROBES */
894 #define optimize_kprobe(p) do {} while (0)
895 #define unoptimize_kprobe(p, f) do {} while (0)
896 #define kill_optimized_kprobe(p) do {} while (0)
897 #define prepare_optimized_kprobe(p) do {} while (0)
898 #define try_to_optimize_kprobe(p) do {} while (0)
899 #define __arm_kprobe(p) arch_arm_kprobe(p)
900 #define __disarm_kprobe(p, o) arch_disarm_kprobe(p)
901 #define kprobe_disarmed(p) kprobe_disabled(p)
902 #define wait_for_kprobe_optimizer() do {} while (0)
904 /* There should be no unused kprobes can be reused without optimization */
905 static void reuse_unused_kprobe(struct kprobe
*ap
)
907 printk(KERN_ERR
"Error: There should be no unused kprobe here.\n");
908 BUG_ON(kprobe_unused(ap
));
911 static __kprobes
void free_aggr_kprobe(struct kprobe
*p
)
913 arch_remove_kprobe(p
);
917 static __kprobes
struct kprobe
*alloc_aggr_kprobe(struct kprobe
*p
)
919 return kzalloc(sizeof(struct kprobe
), GFP_KERNEL
);
921 #endif /* CONFIG_OPTPROBES */
923 #ifdef CONFIG_KPROBES_ON_FTRACE
924 static struct ftrace_ops kprobe_ftrace_ops __read_mostly
= {
925 .func
= kprobe_ftrace_handler
,
926 .flags
= FTRACE_OPS_FL_SAVE_REGS
,
928 static int kprobe_ftrace_enabled
;
930 /* Must ensure p->addr is really on ftrace */
931 static int __kprobes
prepare_kprobe(struct kprobe
*p
)
933 if (!kprobe_ftrace(p
))
934 return arch_prepare_kprobe(p
);
936 return arch_prepare_kprobe_ftrace(p
);
939 /* Caller must lock kprobe_mutex */
940 static void __kprobes
arm_kprobe_ftrace(struct kprobe
*p
)
944 ret
= ftrace_set_filter_ip(&kprobe_ftrace_ops
,
945 (unsigned long)p
->addr
, 0, 0);
946 WARN(ret
< 0, "Failed to arm kprobe-ftrace at %p (%d)\n", p
->addr
, ret
);
947 kprobe_ftrace_enabled
++;
948 if (kprobe_ftrace_enabled
== 1) {
949 ret
= register_ftrace_function(&kprobe_ftrace_ops
);
950 WARN(ret
< 0, "Failed to init kprobe-ftrace (%d)\n", ret
);
954 /* Caller must lock kprobe_mutex */
955 static void __kprobes
disarm_kprobe_ftrace(struct kprobe
*p
)
959 kprobe_ftrace_enabled
--;
960 if (kprobe_ftrace_enabled
== 0) {
961 ret
= unregister_ftrace_function(&kprobe_ftrace_ops
);
962 WARN(ret
< 0, "Failed to init kprobe-ftrace (%d)\n", ret
);
964 ret
= ftrace_set_filter_ip(&kprobe_ftrace_ops
,
965 (unsigned long)p
->addr
, 1, 0);
966 WARN(ret
< 0, "Failed to disarm kprobe-ftrace at %p (%d)\n", p
->addr
, ret
);
968 #else /* !CONFIG_KPROBES_ON_FTRACE */
969 #define prepare_kprobe(p) arch_prepare_kprobe(p)
970 #define arm_kprobe_ftrace(p) do {} while (0)
971 #define disarm_kprobe_ftrace(p) do {} while (0)
974 /* Arm a kprobe with text_mutex */
975 static void __kprobes
arm_kprobe(struct kprobe
*kp
)
977 if (unlikely(kprobe_ftrace(kp
))) {
978 arm_kprobe_ftrace(kp
);
982 * Here, since __arm_kprobe() doesn't use stop_machine(),
983 * this doesn't cause deadlock on text_mutex. So, we don't
984 * need get_online_cpus().
986 mutex_lock(&text_mutex
);
988 mutex_unlock(&text_mutex
);
991 /* Disarm a kprobe with text_mutex */
992 static void __kprobes
disarm_kprobe(struct kprobe
*kp
, bool reopt
)
994 if (unlikely(kprobe_ftrace(kp
))) {
995 disarm_kprobe_ftrace(kp
);
999 mutex_lock(&text_mutex
);
1000 __disarm_kprobe(kp
, reopt
);
1001 mutex_unlock(&text_mutex
);
1005 * Aggregate handlers for multiple kprobes support - these handlers
1006 * take care of invoking the individual kprobe handlers on p->list
1008 static int __kprobes
aggr_pre_handler(struct kprobe
*p
, struct pt_regs
*regs
)
1012 list_for_each_entry_rcu(kp
, &p
->list
, list
) {
1013 if (kp
->pre_handler
&& likely(!kprobe_disabled(kp
))) {
1014 set_kprobe_instance(kp
);
1015 if (kp
->pre_handler(kp
, regs
))
1018 reset_kprobe_instance();
1023 static void __kprobes
aggr_post_handler(struct kprobe
*p
, struct pt_regs
*regs
,
1024 unsigned long flags
)
1028 list_for_each_entry_rcu(kp
, &p
->list
, list
) {
1029 if (kp
->post_handler
&& likely(!kprobe_disabled(kp
))) {
1030 set_kprobe_instance(kp
);
1031 kp
->post_handler(kp
, regs
, flags
);
1032 reset_kprobe_instance();
1037 static int __kprobes
aggr_fault_handler(struct kprobe
*p
, struct pt_regs
*regs
,
1040 struct kprobe
*cur
= __this_cpu_read(kprobe_instance
);
1043 * if we faulted "during" the execution of a user specified
1044 * probe handler, invoke just that probe's fault handler
1046 if (cur
&& cur
->fault_handler
) {
1047 if (cur
->fault_handler(cur
, regs
, trapnr
))
1053 static int __kprobes
aggr_break_handler(struct kprobe
*p
, struct pt_regs
*regs
)
1055 struct kprobe
*cur
= __this_cpu_read(kprobe_instance
);
1058 if (cur
&& cur
->break_handler
) {
1059 if (cur
->break_handler(cur
, regs
))
1062 reset_kprobe_instance();
1066 /* Walks the list and increments nmissed count for multiprobe case */
1067 void __kprobes
kprobes_inc_nmissed_count(struct kprobe
*p
)
1070 if (!kprobe_aggrprobe(p
)) {
1073 list_for_each_entry_rcu(kp
, &p
->list
, list
)
1079 void __kprobes
recycle_rp_inst(struct kretprobe_instance
*ri
,
1080 struct hlist_head
*head
)
1082 struct kretprobe
*rp
= ri
->rp
;
1084 /* remove rp inst off the rprobe_inst_table */
1085 hlist_del(&ri
->hlist
);
1086 INIT_HLIST_NODE(&ri
->hlist
);
1088 raw_spin_lock(&rp
->lock
);
1089 hlist_add_head(&ri
->hlist
, &rp
->free_instances
);
1090 raw_spin_unlock(&rp
->lock
);
1093 hlist_add_head(&ri
->hlist
, head
);
1096 void __kprobes
kretprobe_hash_lock(struct task_struct
*tsk
,
1097 struct hlist_head
**head
, unsigned long *flags
)
1098 __acquires(hlist_lock
)
1100 unsigned long hash
= hash_ptr(tsk
, KPROBE_HASH_BITS
);
1101 raw_spinlock_t
*hlist_lock
;
1103 *head
= &kretprobe_inst_table
[hash
];
1104 hlist_lock
= kretprobe_table_lock_ptr(hash
);
1105 raw_spin_lock_irqsave(hlist_lock
, *flags
);
1108 static void __kprobes
kretprobe_table_lock(unsigned long hash
,
1109 unsigned long *flags
)
1110 __acquires(hlist_lock
)
1112 raw_spinlock_t
*hlist_lock
= kretprobe_table_lock_ptr(hash
);
1113 raw_spin_lock_irqsave(hlist_lock
, *flags
);
1116 void __kprobes
kretprobe_hash_unlock(struct task_struct
*tsk
,
1117 unsigned long *flags
)
1118 __releases(hlist_lock
)
1120 unsigned long hash
= hash_ptr(tsk
, KPROBE_HASH_BITS
);
1121 raw_spinlock_t
*hlist_lock
;
1123 hlist_lock
= kretprobe_table_lock_ptr(hash
);
1124 raw_spin_unlock_irqrestore(hlist_lock
, *flags
);
1127 static void __kprobes
kretprobe_table_unlock(unsigned long hash
,
1128 unsigned long *flags
)
1129 __releases(hlist_lock
)
1131 raw_spinlock_t
*hlist_lock
= kretprobe_table_lock_ptr(hash
);
1132 raw_spin_unlock_irqrestore(hlist_lock
, *flags
);
1136 * This function is called from finish_task_switch when task tk becomes dead,
1137 * so that we can recycle any function-return probe instances associated
1138 * with this task. These left over instances represent probed functions
1139 * that have been called but will never return.
1141 void __kprobes
kprobe_flush_task(struct task_struct
*tk
)
1143 struct kretprobe_instance
*ri
;
1144 struct hlist_head
*head
, empty_rp
;
1145 struct hlist_node
*tmp
;
1146 unsigned long hash
, flags
= 0;
1148 if (unlikely(!kprobes_initialized
))
1149 /* Early boot. kretprobe_table_locks not yet initialized. */
1152 INIT_HLIST_HEAD(&empty_rp
);
1153 hash
= hash_ptr(tk
, KPROBE_HASH_BITS
);
1154 head
= &kretprobe_inst_table
[hash
];
1155 kretprobe_table_lock(hash
, &flags
);
1156 hlist_for_each_entry_safe(ri
, tmp
, head
, hlist
) {
1158 recycle_rp_inst(ri
, &empty_rp
);
1160 kretprobe_table_unlock(hash
, &flags
);
1161 hlist_for_each_entry_safe(ri
, tmp
, &empty_rp
, hlist
) {
1162 hlist_del(&ri
->hlist
);
1167 static inline void free_rp_inst(struct kretprobe
*rp
)
1169 struct kretprobe_instance
*ri
;
1170 struct hlist_node
*next
;
1172 hlist_for_each_entry_safe(ri
, next
, &rp
->free_instances
, hlist
) {
1173 hlist_del(&ri
->hlist
);
1178 static void __kprobes
cleanup_rp_inst(struct kretprobe
*rp
)
1180 unsigned long flags
, hash
;
1181 struct kretprobe_instance
*ri
;
1182 struct hlist_node
*next
;
1183 struct hlist_head
*head
;
1186 for (hash
= 0; hash
< KPROBE_TABLE_SIZE
; hash
++) {
1187 kretprobe_table_lock(hash
, &flags
);
1188 head
= &kretprobe_inst_table
[hash
];
1189 hlist_for_each_entry_safe(ri
, next
, head
, hlist
) {
1193 kretprobe_table_unlock(hash
, &flags
);
1199 * Add the new probe to ap->list. Fail if this is the
1200 * second jprobe at the address - two jprobes can't coexist
1202 static int __kprobes
add_new_kprobe(struct kprobe
*ap
, struct kprobe
*p
)
1204 BUG_ON(kprobe_gone(ap
) || kprobe_gone(p
));
1206 if (p
->break_handler
|| p
->post_handler
)
1207 unoptimize_kprobe(ap
, true); /* Fall back to normal kprobe */
1209 if (p
->break_handler
) {
1210 if (ap
->break_handler
)
1212 list_add_tail_rcu(&p
->list
, &ap
->list
);
1213 ap
->break_handler
= aggr_break_handler
;
1215 list_add_rcu(&p
->list
, &ap
->list
);
1216 if (p
->post_handler
&& !ap
->post_handler
)
1217 ap
->post_handler
= aggr_post_handler
;
1223 * Fill in the required fields of the "manager kprobe". Replace the
1224 * earlier kprobe in the hlist with the manager kprobe
1226 static void __kprobes
init_aggr_kprobe(struct kprobe
*ap
, struct kprobe
*p
)
1228 /* Copy p's insn slot to ap */
1230 flush_insn_slot(ap
);
1232 ap
->flags
= p
->flags
& ~KPROBE_FLAG_OPTIMIZED
;
1233 ap
->pre_handler
= aggr_pre_handler
;
1234 ap
->fault_handler
= aggr_fault_handler
;
1235 /* We don't care the kprobe which has gone. */
1236 if (p
->post_handler
&& !kprobe_gone(p
))
1237 ap
->post_handler
= aggr_post_handler
;
1238 if (p
->break_handler
&& !kprobe_gone(p
))
1239 ap
->break_handler
= aggr_break_handler
;
1241 INIT_LIST_HEAD(&ap
->list
);
1242 INIT_HLIST_NODE(&ap
->hlist
);
1244 list_add_rcu(&p
->list
, &ap
->list
);
1245 hlist_replace_rcu(&p
->hlist
, &ap
->hlist
);
1249 * This is the second or subsequent kprobe at the address - handle
1252 static int __kprobes
register_aggr_kprobe(struct kprobe
*orig_p
,
1256 struct kprobe
*ap
= orig_p
;
1258 /* For preparing optimization, jump_label_text_reserved() is called */
1261 * Get online CPUs to avoid text_mutex deadlock.with stop machine,
1262 * which is invoked by unoptimize_kprobe() in add_new_kprobe()
1265 mutex_lock(&text_mutex
);
1267 if (!kprobe_aggrprobe(orig_p
)) {
1268 /* If orig_p is not an aggr_kprobe, create new aggr_kprobe. */
1269 ap
= alloc_aggr_kprobe(orig_p
);
1274 init_aggr_kprobe(ap
, orig_p
);
1275 } else if (kprobe_unused(ap
))
1276 /* This probe is going to die. Rescue it */
1277 reuse_unused_kprobe(ap
);
1279 if (kprobe_gone(ap
)) {
1281 * Attempting to insert new probe at the same location that
1282 * had a probe in the module vaddr area which already
1283 * freed. So, the instruction slot has already been
1284 * released. We need a new slot for the new probe.
1286 ret
= arch_prepare_kprobe(ap
);
1289 * Even if fail to allocate new slot, don't need to
1290 * free aggr_probe. It will be used next time, or
1291 * freed by unregister_kprobe.
1295 /* Prepare optimized instructions if possible. */
1296 prepare_optimized_kprobe(ap
);
1299 * Clear gone flag to prevent allocating new slot again, and
1300 * set disabled flag because it is not armed yet.
1302 ap
->flags
= (ap
->flags
& ~KPROBE_FLAG_GONE
)
1303 | KPROBE_FLAG_DISABLED
;
1306 /* Copy ap's insn slot to p */
1308 ret
= add_new_kprobe(ap
, p
);
1311 mutex_unlock(&text_mutex
);
1313 jump_label_unlock();
1315 if (ret
== 0 && kprobe_disabled(ap
) && !kprobe_disabled(p
)) {
1316 ap
->flags
&= ~KPROBE_FLAG_DISABLED
;
1317 if (!kprobes_all_disarmed
)
1318 /* Arm the breakpoint again. */
1324 bool __weak
arch_within_kprobe_blacklist(unsigned long addr
)
1326 /* The __kprobes marked functions and entry code must not be probed */
1327 return addr
>= (unsigned long)__kprobes_text_start
&&
1328 addr
< (unsigned long)__kprobes_text_end
;
1331 static int __kprobes
in_kprobes_functions(unsigned long addr
)
1333 struct kprobe_blackpoint
*kb
;
1335 if (arch_within_kprobe_blacklist(addr
))
1338 * If there exists a kprobe_blacklist, verify and
1339 * fail any probe registration in the prohibited area
1341 for (kb
= kprobe_blacklist
; kb
->name
!= NULL
; kb
++) {
1342 if (kb
->start_addr
) {
1343 if (addr
>= kb
->start_addr
&&
1344 addr
< (kb
->start_addr
+ kb
->range
))
1352 * If we have a symbol_name argument, look it up and add the offset field
1353 * to it. This way, we can specify a relative address to a symbol.
1354 * This returns encoded errors if it fails to look up symbol or invalid
1355 * combination of parameters.
1357 static kprobe_opcode_t __kprobes
*kprobe_addr(struct kprobe
*p
)
1359 kprobe_opcode_t
*addr
= p
->addr
;
1361 if ((p
->symbol_name
&& p
->addr
) ||
1362 (!p
->symbol_name
&& !p
->addr
))
1365 if (p
->symbol_name
) {
1366 kprobe_lookup_name(p
->symbol_name
, addr
);
1368 return ERR_PTR(-ENOENT
);
1371 addr
= (kprobe_opcode_t
*)(((char *)addr
) + p
->offset
);
1376 return ERR_PTR(-EINVAL
);
1379 /* Check passed kprobe is valid and return kprobe in kprobe_table. */
1380 static struct kprobe
* __kprobes
__get_valid_kprobe(struct kprobe
*p
)
1382 struct kprobe
*ap
, *list_p
;
1384 ap
= get_kprobe(p
->addr
);
1389 list_for_each_entry_rcu(list_p
, &ap
->list
, list
)
1391 /* kprobe p is a valid probe */
1399 /* Return error if the kprobe is being re-registered */
1400 static inline int check_kprobe_rereg(struct kprobe
*p
)
1404 mutex_lock(&kprobe_mutex
);
1405 if (__get_valid_kprobe(p
))
1407 mutex_unlock(&kprobe_mutex
);
1412 static __kprobes
int check_kprobe_address_safe(struct kprobe
*p
,
1413 struct module
**probed_mod
)
1416 unsigned long ftrace_addr
;
1419 * If the address is located on a ftrace nop, set the
1420 * breakpoint to the following instruction.
1422 ftrace_addr
= ftrace_location((unsigned long)p
->addr
);
1424 #ifdef CONFIG_KPROBES_ON_FTRACE
1425 /* Given address is not on the instruction boundary */
1426 if ((unsigned long)p
->addr
!= ftrace_addr
)
1428 p
->flags
|= KPROBE_FLAG_FTRACE
;
1429 #else /* !CONFIG_KPROBES_ON_FTRACE */
1437 /* Ensure it is not in reserved area nor out of text */
1438 if (!kernel_text_address((unsigned long) p
->addr
) ||
1439 in_kprobes_functions((unsigned long) p
->addr
) ||
1440 jump_label_text_reserved(p
->addr
, p
->addr
)) {
1445 /* Check if are we probing a module */
1446 *probed_mod
= __module_text_address((unsigned long) p
->addr
);
1449 * We must hold a refcount of the probed module while updating
1450 * its code to prohibit unexpected unloading.
1452 if (unlikely(!try_module_get(*probed_mod
))) {
1458 * If the module freed .init.text, we couldn't insert
1461 if (within_module_init((unsigned long)p
->addr
, *probed_mod
) &&
1462 (*probed_mod
)->state
!= MODULE_STATE_COMING
) {
1463 module_put(*probed_mod
);
1470 jump_label_unlock();
1475 int __kprobes
register_kprobe(struct kprobe
*p
)
1478 struct kprobe
*old_p
;
1479 struct module
*probed_mod
;
1480 kprobe_opcode_t
*addr
;
1482 /* Adjust probe address from symbol */
1483 addr
= kprobe_addr(p
);
1485 return PTR_ERR(addr
);
1488 ret
= check_kprobe_rereg(p
);
1492 /* User can pass only KPROBE_FLAG_DISABLED to register_kprobe */
1493 p
->flags
&= KPROBE_FLAG_DISABLED
;
1495 INIT_LIST_HEAD(&p
->list
);
1497 ret
= check_kprobe_address_safe(p
, &probed_mod
);
1501 mutex_lock(&kprobe_mutex
);
1503 old_p
= get_kprobe(p
->addr
);
1505 /* Since this may unoptimize old_p, locking text_mutex. */
1506 ret
= register_aggr_kprobe(old_p
, p
);
1510 mutex_lock(&text_mutex
); /* Avoiding text modification */
1511 ret
= prepare_kprobe(p
);
1512 mutex_unlock(&text_mutex
);
1516 INIT_HLIST_NODE(&p
->hlist
);
1517 hlist_add_head_rcu(&p
->hlist
,
1518 &kprobe_table
[hash_ptr(p
->addr
, KPROBE_HASH_BITS
)]);
1520 if (!kprobes_all_disarmed
&& !kprobe_disabled(p
))
1523 /* Try to optimize kprobe */
1524 try_to_optimize_kprobe(p
);
1527 mutex_unlock(&kprobe_mutex
);
1530 module_put(probed_mod
);
1534 EXPORT_SYMBOL_GPL(register_kprobe
);
1536 /* Check if all probes on the aggrprobe are disabled */
1537 static int __kprobes
aggr_kprobe_disabled(struct kprobe
*ap
)
1541 list_for_each_entry_rcu(kp
, &ap
->list
, list
)
1542 if (!kprobe_disabled(kp
))
1544 * There is an active probe on the list.
1545 * We can't disable this ap.
1552 /* Disable one kprobe: Make sure called under kprobe_mutex is locked */
1553 static struct kprobe
*__kprobes
__disable_kprobe(struct kprobe
*p
)
1555 struct kprobe
*orig_p
;
1557 /* Get an original kprobe for return */
1558 orig_p
= __get_valid_kprobe(p
);
1559 if (unlikely(orig_p
== NULL
))
1562 if (!kprobe_disabled(p
)) {
1563 /* Disable probe if it is a child probe */
1565 p
->flags
|= KPROBE_FLAG_DISABLED
;
1567 /* Try to disarm and disable this/parent probe */
1568 if (p
== orig_p
|| aggr_kprobe_disabled(orig_p
)) {
1569 disarm_kprobe(orig_p
, true);
1570 orig_p
->flags
|= KPROBE_FLAG_DISABLED
;
1578 * Unregister a kprobe without a scheduler synchronization.
1580 static int __kprobes
__unregister_kprobe_top(struct kprobe
*p
)
1582 struct kprobe
*ap
, *list_p
;
1584 /* Disable kprobe. This will disarm it if needed. */
1585 ap
= __disable_kprobe(p
);
1591 * This probe is an independent(and non-optimized) kprobe
1592 * (not an aggrprobe). Remove from the hash list.
1596 /* Following process expects this probe is an aggrprobe */
1597 WARN_ON(!kprobe_aggrprobe(ap
));
1599 if (list_is_singular(&ap
->list
) && kprobe_disarmed(ap
))
1601 * !disarmed could be happen if the probe is under delayed
1606 /* If disabling probe has special handlers, update aggrprobe */
1607 if (p
->break_handler
&& !kprobe_gone(p
))
1608 ap
->break_handler
= NULL
;
1609 if (p
->post_handler
&& !kprobe_gone(p
)) {
1610 list_for_each_entry_rcu(list_p
, &ap
->list
, list
) {
1611 if ((list_p
!= p
) && (list_p
->post_handler
))
1614 ap
->post_handler
= NULL
;
1618 * Remove from the aggrprobe: this path will do nothing in
1619 * __unregister_kprobe_bottom().
1621 list_del_rcu(&p
->list
);
1622 if (!kprobe_disabled(ap
) && !kprobes_all_disarmed
)
1624 * Try to optimize this probe again, because post
1625 * handler may have been changed.
1627 optimize_kprobe(ap
);
1632 BUG_ON(!kprobe_disarmed(ap
));
1633 hlist_del_rcu(&ap
->hlist
);
1637 static void __kprobes
__unregister_kprobe_bottom(struct kprobe
*p
)
1641 if (list_empty(&p
->list
))
1642 /* This is an independent kprobe */
1643 arch_remove_kprobe(p
);
1644 else if (list_is_singular(&p
->list
)) {
1645 /* This is the last child of an aggrprobe */
1646 ap
= list_entry(p
->list
.next
, struct kprobe
, list
);
1648 free_aggr_kprobe(ap
);
1650 /* Otherwise, do nothing. */
1653 int __kprobes
register_kprobes(struct kprobe
**kps
, int num
)
1659 for (i
= 0; i
< num
; i
++) {
1660 ret
= register_kprobe(kps
[i
]);
1663 unregister_kprobes(kps
, i
);
1669 EXPORT_SYMBOL_GPL(register_kprobes
);
1671 void __kprobes
unregister_kprobe(struct kprobe
*p
)
1673 unregister_kprobes(&p
, 1);
1675 EXPORT_SYMBOL_GPL(unregister_kprobe
);
1677 void __kprobes
unregister_kprobes(struct kprobe
**kps
, int num
)
1683 mutex_lock(&kprobe_mutex
);
1684 for (i
= 0; i
< num
; i
++)
1685 if (__unregister_kprobe_top(kps
[i
]) < 0)
1686 kps
[i
]->addr
= NULL
;
1687 mutex_unlock(&kprobe_mutex
);
1689 synchronize_sched();
1690 for (i
= 0; i
< num
; i
++)
1692 __unregister_kprobe_bottom(kps
[i
]);
1694 EXPORT_SYMBOL_GPL(unregister_kprobes
);
1696 static struct notifier_block kprobe_exceptions_nb
= {
1697 .notifier_call
= kprobe_exceptions_notify
,
1698 .priority
= 0x7fffffff /* we need to be notified first */
1701 unsigned long __weak
arch_deref_entry_point(void *entry
)
1703 return (unsigned long)entry
;
1706 int __kprobes
register_jprobes(struct jprobe
**jps
, int num
)
1713 for (i
= 0; i
< num
; i
++) {
1714 unsigned long addr
, offset
;
1716 addr
= arch_deref_entry_point(jp
->entry
);
1718 /* Verify probepoint is a function entry point */
1719 if (kallsyms_lookup_size_offset(addr
, NULL
, &offset
) &&
1721 jp
->kp
.pre_handler
= setjmp_pre_handler
;
1722 jp
->kp
.break_handler
= longjmp_break_handler
;
1723 ret
= register_kprobe(&jp
->kp
);
1729 unregister_jprobes(jps
, i
);
1735 EXPORT_SYMBOL_GPL(register_jprobes
);
1737 int __kprobes
register_jprobe(struct jprobe
*jp
)
1739 return register_jprobes(&jp
, 1);
1741 EXPORT_SYMBOL_GPL(register_jprobe
);
1743 void __kprobes
unregister_jprobe(struct jprobe
*jp
)
1745 unregister_jprobes(&jp
, 1);
1747 EXPORT_SYMBOL_GPL(unregister_jprobe
);
1749 void __kprobes
unregister_jprobes(struct jprobe
**jps
, int num
)
1755 mutex_lock(&kprobe_mutex
);
1756 for (i
= 0; i
< num
; i
++)
1757 if (__unregister_kprobe_top(&jps
[i
]->kp
) < 0)
1758 jps
[i
]->kp
.addr
= NULL
;
1759 mutex_unlock(&kprobe_mutex
);
1761 synchronize_sched();
1762 for (i
= 0; i
< num
; i
++) {
1763 if (jps
[i
]->kp
.addr
)
1764 __unregister_kprobe_bottom(&jps
[i
]->kp
);
1767 EXPORT_SYMBOL_GPL(unregister_jprobes
);
1769 #ifdef CONFIG_KRETPROBES
1771 * This kprobe pre_handler is registered with every kretprobe. When probe
1772 * hits it will set up the return probe.
1774 static int __kprobes
pre_handler_kretprobe(struct kprobe
*p
,
1775 struct pt_regs
*regs
)
1777 struct kretprobe
*rp
= container_of(p
, struct kretprobe
, kp
);
1778 unsigned long hash
, flags
= 0;
1779 struct kretprobe_instance
*ri
;
1781 /*TODO: consider to only swap the RA after the last pre_handler fired */
1782 hash
= hash_ptr(current
, KPROBE_HASH_BITS
);
1783 raw_spin_lock_irqsave(&rp
->lock
, flags
);
1784 if (!hlist_empty(&rp
->free_instances
)) {
1785 ri
= hlist_entry(rp
->free_instances
.first
,
1786 struct kretprobe_instance
, hlist
);
1787 hlist_del(&ri
->hlist
);
1788 raw_spin_unlock_irqrestore(&rp
->lock
, flags
);
1793 if (rp
->entry_handler
&& rp
->entry_handler(ri
, regs
)) {
1794 raw_spin_lock_irqsave(&rp
->lock
, flags
);
1795 hlist_add_head(&ri
->hlist
, &rp
->free_instances
);
1796 raw_spin_unlock_irqrestore(&rp
->lock
, flags
);
1800 arch_prepare_kretprobe(ri
, regs
);
1802 /* XXX(hch): why is there no hlist_move_head? */
1803 INIT_HLIST_NODE(&ri
->hlist
);
1804 kretprobe_table_lock(hash
, &flags
);
1805 hlist_add_head(&ri
->hlist
, &kretprobe_inst_table
[hash
]);
1806 kretprobe_table_unlock(hash
, &flags
);
1809 raw_spin_unlock_irqrestore(&rp
->lock
, flags
);
1814 int __kprobes
register_kretprobe(struct kretprobe
*rp
)
1817 struct kretprobe_instance
*inst
;
1821 if (kretprobe_blacklist_size
) {
1822 addr
= kprobe_addr(&rp
->kp
);
1824 return PTR_ERR(addr
);
1826 for (i
= 0; kretprobe_blacklist
[i
].name
!= NULL
; i
++) {
1827 if (kretprobe_blacklist
[i
].addr
== addr
)
1832 rp
->kp
.pre_handler
= pre_handler_kretprobe
;
1833 rp
->kp
.post_handler
= NULL
;
1834 rp
->kp
.fault_handler
= NULL
;
1835 rp
->kp
.break_handler
= NULL
;
1837 /* Pre-allocate memory for max kretprobe instances */
1838 if (rp
->maxactive
<= 0) {
1839 #ifdef CONFIG_PREEMPT
1840 rp
->maxactive
= max_t(unsigned int, 10, 2*num_possible_cpus());
1842 rp
->maxactive
= num_possible_cpus();
1845 raw_spin_lock_init(&rp
->lock
);
1846 INIT_HLIST_HEAD(&rp
->free_instances
);
1847 for (i
= 0; i
< rp
->maxactive
; i
++) {
1848 inst
= kmalloc(sizeof(struct kretprobe_instance
) +
1849 rp
->data_size
, GFP_KERNEL
);
1854 INIT_HLIST_NODE(&inst
->hlist
);
1855 hlist_add_head(&inst
->hlist
, &rp
->free_instances
);
1859 /* Establish function entry probe point */
1860 ret
= register_kprobe(&rp
->kp
);
1865 EXPORT_SYMBOL_GPL(register_kretprobe
);
1867 int __kprobes
register_kretprobes(struct kretprobe
**rps
, int num
)
1873 for (i
= 0; i
< num
; i
++) {
1874 ret
= register_kretprobe(rps
[i
]);
1877 unregister_kretprobes(rps
, i
);
1883 EXPORT_SYMBOL_GPL(register_kretprobes
);
1885 void __kprobes
unregister_kretprobe(struct kretprobe
*rp
)
1887 unregister_kretprobes(&rp
, 1);
1889 EXPORT_SYMBOL_GPL(unregister_kretprobe
);
1891 void __kprobes
unregister_kretprobes(struct kretprobe
**rps
, int num
)
1897 mutex_lock(&kprobe_mutex
);
1898 for (i
= 0; i
< num
; i
++)
1899 if (__unregister_kprobe_top(&rps
[i
]->kp
) < 0)
1900 rps
[i
]->kp
.addr
= NULL
;
1901 mutex_unlock(&kprobe_mutex
);
1903 synchronize_sched();
1904 for (i
= 0; i
< num
; i
++) {
1905 if (rps
[i
]->kp
.addr
) {
1906 __unregister_kprobe_bottom(&rps
[i
]->kp
);
1907 cleanup_rp_inst(rps
[i
]);
1911 EXPORT_SYMBOL_GPL(unregister_kretprobes
);
1913 #else /* CONFIG_KRETPROBES */
1914 int __kprobes
register_kretprobe(struct kretprobe
*rp
)
1918 EXPORT_SYMBOL_GPL(register_kretprobe
);
1920 int __kprobes
register_kretprobes(struct kretprobe
**rps
, int num
)
1924 EXPORT_SYMBOL_GPL(register_kretprobes
);
1926 void __kprobes
unregister_kretprobe(struct kretprobe
*rp
)
1929 EXPORT_SYMBOL_GPL(unregister_kretprobe
);
1931 void __kprobes
unregister_kretprobes(struct kretprobe
**rps
, int num
)
1934 EXPORT_SYMBOL_GPL(unregister_kretprobes
);
1936 static int __kprobes
pre_handler_kretprobe(struct kprobe
*p
,
1937 struct pt_regs
*regs
)
1942 #endif /* CONFIG_KRETPROBES */
1944 /* Set the kprobe gone and remove its instruction buffer. */
1945 static void __kprobes
kill_kprobe(struct kprobe
*p
)
1949 p
->flags
|= KPROBE_FLAG_GONE
;
1950 if (kprobe_aggrprobe(p
)) {
1952 * If this is an aggr_kprobe, we have to list all the
1953 * chained probes and mark them GONE.
1955 list_for_each_entry_rcu(kp
, &p
->list
, list
)
1956 kp
->flags
|= KPROBE_FLAG_GONE
;
1957 p
->post_handler
= NULL
;
1958 p
->break_handler
= NULL
;
1959 kill_optimized_kprobe(p
);
1962 * Here, we can remove insn_slot safely, because no thread calls
1963 * the original probed function (which will be freed soon) any more.
1965 arch_remove_kprobe(p
);
1968 /* Disable one kprobe */
1969 int __kprobes
disable_kprobe(struct kprobe
*kp
)
1973 mutex_lock(&kprobe_mutex
);
1975 /* Disable this kprobe */
1976 if (__disable_kprobe(kp
) == NULL
)
1979 mutex_unlock(&kprobe_mutex
);
1982 EXPORT_SYMBOL_GPL(disable_kprobe
);
1984 /* Enable one kprobe */
1985 int __kprobes
enable_kprobe(struct kprobe
*kp
)
1990 mutex_lock(&kprobe_mutex
);
1992 /* Check whether specified probe is valid. */
1993 p
= __get_valid_kprobe(kp
);
1994 if (unlikely(p
== NULL
)) {
1999 if (kprobe_gone(kp
)) {
2000 /* This kprobe has gone, we couldn't enable it. */
2006 kp
->flags
&= ~KPROBE_FLAG_DISABLED
;
2008 if (!kprobes_all_disarmed
&& kprobe_disabled(p
)) {
2009 p
->flags
&= ~KPROBE_FLAG_DISABLED
;
2013 mutex_unlock(&kprobe_mutex
);
2016 EXPORT_SYMBOL_GPL(enable_kprobe
);
2018 void __kprobes
dump_kprobe(struct kprobe
*kp
)
2020 printk(KERN_WARNING
"Dumping kprobe:\n");
2021 printk(KERN_WARNING
"Name: %s\nAddress: %p\nOffset: %x\n",
2022 kp
->symbol_name
, kp
->addr
, kp
->offset
);
2025 /* Module notifier call back, checking kprobes on the module */
2026 static int __kprobes
kprobes_module_callback(struct notifier_block
*nb
,
2027 unsigned long val
, void *data
)
2029 struct module
*mod
= data
;
2030 struct hlist_head
*head
;
2033 int checkcore
= (val
== MODULE_STATE_GOING
);
2035 if (val
!= MODULE_STATE_GOING
&& val
!= MODULE_STATE_LIVE
)
2039 * When MODULE_STATE_GOING was notified, both of module .text and
2040 * .init.text sections would be freed. When MODULE_STATE_LIVE was
2041 * notified, only .init.text section would be freed. We need to
2042 * disable kprobes which have been inserted in the sections.
2044 mutex_lock(&kprobe_mutex
);
2045 for (i
= 0; i
< KPROBE_TABLE_SIZE
; i
++) {
2046 head
= &kprobe_table
[i
];
2047 hlist_for_each_entry_rcu(p
, head
, hlist
)
2048 if (within_module_init((unsigned long)p
->addr
, mod
) ||
2050 within_module_core((unsigned long)p
->addr
, mod
))) {
2052 * The vaddr this probe is installed will soon
2053 * be vfreed buy not synced to disk. Hence,
2054 * disarming the breakpoint isn't needed.
2059 mutex_unlock(&kprobe_mutex
);
2063 static struct notifier_block kprobe_module_nb
= {
2064 .notifier_call
= kprobes_module_callback
,
2068 static int __init
init_kprobes(void)
2071 unsigned long offset
= 0, size
= 0;
2072 char *modname
, namebuf
[KSYM_NAME_LEN
];
2073 const char *symbol_name
;
2075 struct kprobe_blackpoint
*kb
;
2077 /* FIXME allocate the probe table, currently defined statically */
2078 /* initialize all list heads */
2079 for (i
= 0; i
< KPROBE_TABLE_SIZE
; i
++) {
2080 INIT_HLIST_HEAD(&kprobe_table
[i
]);
2081 INIT_HLIST_HEAD(&kretprobe_inst_table
[i
]);
2082 raw_spin_lock_init(&(kretprobe_table_locks
[i
].lock
));
2086 * Lookup and populate the kprobe_blacklist.
2088 * Unlike the kretprobe blacklist, we'll need to determine
2089 * the range of addresses that belong to the said functions,
2090 * since a kprobe need not necessarily be at the beginning
2093 for (kb
= kprobe_blacklist
; kb
->name
!= NULL
; kb
++) {
2094 kprobe_lookup_name(kb
->name
, addr
);
2098 kb
->start_addr
= (unsigned long)addr
;
2099 symbol_name
= kallsyms_lookup(kb
->start_addr
,
2100 &size
, &offset
, &modname
, namebuf
);
2107 if (kretprobe_blacklist_size
) {
2108 /* lookup the function address from its name */
2109 for (i
= 0; kretprobe_blacklist
[i
].name
!= NULL
; i
++) {
2110 kprobe_lookup_name(kretprobe_blacklist
[i
].name
,
2111 kretprobe_blacklist
[i
].addr
);
2112 if (!kretprobe_blacklist
[i
].addr
)
2113 printk("kretprobe: lookup failed: %s\n",
2114 kretprobe_blacklist
[i
].name
);
2118 #if defined(CONFIG_OPTPROBES)
2119 #if defined(__ARCH_WANT_KPROBES_INSN_SLOT)
2120 /* Init kprobe_optinsn_slots */
2121 kprobe_optinsn_slots
.insn_size
= MAX_OPTINSN_SIZE
;
2123 /* By default, kprobes can be optimized */
2124 kprobes_allow_optimization
= true;
2127 /* By default, kprobes are armed */
2128 kprobes_all_disarmed
= false;
2130 err
= arch_init_kprobes();
2132 err
= register_die_notifier(&kprobe_exceptions_nb
);
2134 err
= register_module_notifier(&kprobe_module_nb
);
2136 kprobes_initialized
= (err
== 0);
2143 #ifdef CONFIG_DEBUG_FS
2144 static void __kprobes
report_probe(struct seq_file
*pi
, struct kprobe
*p
,
2145 const char *sym
, int offset
, char *modname
, struct kprobe
*pp
)
2149 if (p
->pre_handler
== pre_handler_kretprobe
)
2151 else if (p
->pre_handler
== setjmp_pre_handler
)
2157 seq_printf(pi
, "%p %s %s+0x%x %s ",
2158 p
->addr
, kprobe_type
, sym
, offset
,
2159 (modname
? modname
: " "));
2161 seq_printf(pi
, "%p %s %p ",
2162 p
->addr
, kprobe_type
, p
->addr
);
2166 seq_printf(pi
, "%s%s%s%s\n",
2167 (kprobe_gone(p
) ? "[GONE]" : ""),
2168 ((kprobe_disabled(p
) && !kprobe_gone(p
)) ? "[DISABLED]" : ""),
2169 (kprobe_optimized(pp
) ? "[OPTIMIZED]" : ""),
2170 (kprobe_ftrace(pp
) ? "[FTRACE]" : ""));
2173 static void __kprobes
*kprobe_seq_start(struct seq_file
*f
, loff_t
*pos
)
2175 return (*pos
< KPROBE_TABLE_SIZE
) ? pos
: NULL
;
2178 static void __kprobes
*kprobe_seq_next(struct seq_file
*f
, void *v
, loff_t
*pos
)
2181 if (*pos
>= KPROBE_TABLE_SIZE
)
2186 static void __kprobes
kprobe_seq_stop(struct seq_file
*f
, void *v
)
2191 static int __kprobes
show_kprobe_addr(struct seq_file
*pi
, void *v
)
2193 struct hlist_head
*head
;
2194 struct kprobe
*p
, *kp
;
2195 const char *sym
= NULL
;
2196 unsigned int i
= *(loff_t
*) v
;
2197 unsigned long offset
= 0;
2198 char *modname
, namebuf
[KSYM_NAME_LEN
];
2200 head
= &kprobe_table
[i
];
2202 hlist_for_each_entry_rcu(p
, head
, hlist
) {
2203 sym
= kallsyms_lookup((unsigned long)p
->addr
, NULL
,
2204 &offset
, &modname
, namebuf
);
2205 if (kprobe_aggrprobe(p
)) {
2206 list_for_each_entry_rcu(kp
, &p
->list
, list
)
2207 report_probe(pi
, kp
, sym
, offset
, modname
, p
);
2209 report_probe(pi
, p
, sym
, offset
, modname
, NULL
);
2215 static const struct seq_operations kprobes_seq_ops
= {
2216 .start
= kprobe_seq_start
,
2217 .next
= kprobe_seq_next
,
2218 .stop
= kprobe_seq_stop
,
2219 .show
= show_kprobe_addr
2222 static int __kprobes
kprobes_open(struct inode
*inode
, struct file
*filp
)
2224 return seq_open(filp
, &kprobes_seq_ops
);
2227 static const struct file_operations debugfs_kprobes_operations
= {
2228 .open
= kprobes_open
,
2230 .llseek
= seq_lseek
,
2231 .release
= seq_release
,
2234 static void __kprobes
arm_all_kprobes(void)
2236 struct hlist_head
*head
;
2240 mutex_lock(&kprobe_mutex
);
2242 /* If kprobes are armed, just return */
2243 if (!kprobes_all_disarmed
)
2244 goto already_enabled
;
2246 /* Arming kprobes doesn't optimize kprobe itself */
2247 for (i
= 0; i
< KPROBE_TABLE_SIZE
; i
++) {
2248 head
= &kprobe_table
[i
];
2249 hlist_for_each_entry_rcu(p
, head
, hlist
)
2250 if (!kprobe_disabled(p
))
2254 kprobes_all_disarmed
= false;
2255 printk(KERN_INFO
"Kprobes globally enabled\n");
2258 mutex_unlock(&kprobe_mutex
);
2262 static void __kprobes
disarm_all_kprobes(void)
2264 struct hlist_head
*head
;
2268 mutex_lock(&kprobe_mutex
);
2270 /* If kprobes are already disarmed, just return */
2271 if (kprobes_all_disarmed
) {
2272 mutex_unlock(&kprobe_mutex
);
2276 kprobes_all_disarmed
= true;
2277 printk(KERN_INFO
"Kprobes globally disabled\n");
2279 for (i
= 0; i
< KPROBE_TABLE_SIZE
; i
++) {
2280 head
= &kprobe_table
[i
];
2281 hlist_for_each_entry_rcu(p
, head
, hlist
) {
2282 if (!arch_trampoline_kprobe(p
) && !kprobe_disabled(p
))
2283 disarm_kprobe(p
, false);
2286 mutex_unlock(&kprobe_mutex
);
2288 /* Wait for disarming all kprobes by optimizer */
2289 wait_for_kprobe_optimizer();
2293 * XXX: The debugfs bool file interface doesn't allow for callbacks
2294 * when the bool state is switched. We can reuse that facility when
2297 static ssize_t
read_enabled_file_bool(struct file
*file
,
2298 char __user
*user_buf
, size_t count
, loff_t
*ppos
)
2302 if (!kprobes_all_disarmed
)
2308 return simple_read_from_buffer(user_buf
, count
, ppos
, buf
, 2);
2311 static ssize_t
write_enabled_file_bool(struct file
*file
,
2312 const char __user
*user_buf
, size_t count
, loff_t
*ppos
)
2317 buf_size
= min(count
, (sizeof(buf
)-1));
2318 if (copy_from_user(buf
, user_buf
, buf_size
))
2321 buf
[buf_size
] = '\0';
2331 disarm_all_kprobes();
2340 static const struct file_operations fops_kp
= {
2341 .read
= read_enabled_file_bool
,
2342 .write
= write_enabled_file_bool
,
2343 .llseek
= default_llseek
,
2346 static int __kprobes
debugfs_kprobe_init(void)
2348 struct dentry
*dir
, *file
;
2349 unsigned int value
= 1;
2351 dir
= debugfs_create_dir("kprobes", NULL
);
2355 file
= debugfs_create_file("list", 0444, dir
, NULL
,
2356 &debugfs_kprobes_operations
);
2358 debugfs_remove(dir
);
2362 file
= debugfs_create_file("enabled", 0600, dir
,
2365 debugfs_remove(dir
);
2372 late_initcall(debugfs_kprobe_init
);
2373 #endif /* CONFIG_DEBUG_FS */
2375 module_init(init_kprobes
);
2377 /* defined in arch/.../kernel/kprobes.c */
2378 EXPORT_SYMBOL_GPL(jprobe_return
);