1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * Kernel Probes (KProbes)
6 * Copyright (C) IBM Corporation, 2002, 2004
8 * 2002-Oct Created by Vamsi Krishna S <vamsi_krishna@in.ibm.com> Kernel
9 * Probes initial implementation (includes suggestions from
11 * 2004-Aug Updated by Prasanna S Panchamukhi <prasanna@in.ibm.com> with
12 * hlists and exceptions notifier as suggested by Andi Kleen.
13 * 2004-July Suparna Bhattacharya <suparna@in.ibm.com> added jumper probes
14 * interface to access function arguments.
15 * 2004-Sep Prasanna S Panchamukhi <prasanna@in.ibm.com> Changed Kprobes
16 * exceptions notifier to be first on the priority list.
17 * 2005-May Hien Nguyen <hien@us.ibm.com>, Jim Keniston
18 * <jkenisto@us.ibm.com> and Prasanna S Panchamukhi
19 * <prasanna@in.ibm.com> added function-return probes.
21 #include <linux/kprobes.h>
22 #include <linux/hash.h>
23 #include <linux/init.h>
24 #include <linux/slab.h>
25 #include <linux/stddef.h>
26 #include <linux/export.h>
27 #include <linux/moduleloader.h>
28 #include <linux/kallsyms.h>
29 #include <linux/freezer.h>
30 #include <linux/seq_file.h>
31 #include <linux/debugfs.h>
32 #include <linux/sysctl.h>
33 #include <linux/kdebug.h>
34 #include <linux/memory.h>
35 #include <linux/ftrace.h>
36 #include <linux/cpu.h>
37 #include <linux/jump_label.h>
39 #include <asm/sections.h>
40 #include <asm/cacheflush.h>
41 #include <asm/errno.h>
42 #include <linux/uaccess.h>
44 #define KPROBE_HASH_BITS 6
45 #define KPROBE_TABLE_SIZE (1 << KPROBE_HASH_BITS)
48 static int kprobes_initialized
;
49 /* kprobe_table can be accessed by
50 * - Normal hlist traversal and RCU add/del under kprobe_mutex is held.
52 * - RCU hlist traversal under disabling preempt (breakpoint handlers)
54 static struct hlist_head kprobe_table
[KPROBE_TABLE_SIZE
];
55 static struct hlist_head kretprobe_inst_table
[KPROBE_TABLE_SIZE
];
57 /* NOTE: change this value only with kprobe_mutex held */
58 static bool kprobes_all_disarmed
;
60 /* This protects kprobe_table and optimizing_list */
61 static DEFINE_MUTEX(kprobe_mutex
);
62 static DEFINE_PER_CPU(struct kprobe
*, kprobe_instance
) = NULL
;
64 raw_spinlock_t lock ____cacheline_aligned_in_smp
;
65 } kretprobe_table_locks
[KPROBE_TABLE_SIZE
];
67 kprobe_opcode_t
* __weak
kprobe_lookup_name(const char *name
,
68 unsigned int __unused
)
70 return ((kprobe_opcode_t
*)(kallsyms_lookup_name(name
)));
73 static raw_spinlock_t
*kretprobe_table_lock_ptr(unsigned long hash
)
75 return &(kretprobe_table_locks
[hash
].lock
);
78 /* Blacklist -- list of struct kprobe_blacklist_entry */
79 static LIST_HEAD(kprobe_blacklist
);
81 #ifdef __ARCH_WANT_KPROBES_INSN_SLOT
83 * kprobe->ainsn.insn points to the copy of the instruction to be
84 * single-stepped. x86_64, POWER4 and above have no-exec support and
85 * stepping on the instruction on a vmalloced/kmalloced/data page
86 * is a recipe for disaster
88 struct kprobe_insn_page
{
89 struct list_head list
;
90 kprobe_opcode_t
*insns
; /* Page of instruction slots */
91 struct kprobe_insn_cache
*cache
;
97 #define KPROBE_INSN_PAGE_SIZE(slots) \
98 (offsetof(struct kprobe_insn_page, slot_used) + \
99 (sizeof(char) * (slots)))
101 static int slots_per_page(struct kprobe_insn_cache
*c
)
103 return PAGE_SIZE
/(c
->insn_size
* sizeof(kprobe_opcode_t
));
106 enum kprobe_slot_state
{
112 void __weak
*alloc_insn_page(void)
114 return module_alloc(PAGE_SIZE
);
117 void __weak
free_insn_page(void *page
)
119 module_memfree(page
);
122 struct kprobe_insn_cache kprobe_insn_slots
= {
123 .mutex
= __MUTEX_INITIALIZER(kprobe_insn_slots
.mutex
),
124 .alloc
= alloc_insn_page
,
125 .free
= free_insn_page
,
126 .pages
= LIST_HEAD_INIT(kprobe_insn_slots
.pages
),
127 .insn_size
= MAX_INSN_SIZE
,
130 static int collect_garbage_slots(struct kprobe_insn_cache
*c
);
133 * __get_insn_slot() - Find a slot on an executable page for an instruction.
134 * We allocate an executable page if there's no room on existing ones.
136 kprobe_opcode_t
*__get_insn_slot(struct kprobe_insn_cache
*c
)
138 struct kprobe_insn_page
*kip
;
139 kprobe_opcode_t
*slot
= NULL
;
141 /* Since the slot array is not protected by rcu, we need a mutex */
142 mutex_lock(&c
->mutex
);
145 list_for_each_entry_rcu(kip
, &c
->pages
, list
) {
146 if (kip
->nused
< slots_per_page(c
)) {
148 for (i
= 0; i
< slots_per_page(c
); i
++) {
149 if (kip
->slot_used
[i
] == SLOT_CLEAN
) {
150 kip
->slot_used
[i
] = SLOT_USED
;
152 slot
= kip
->insns
+ (i
* c
->insn_size
);
157 /* kip->nused is broken. Fix it. */
158 kip
->nused
= slots_per_page(c
);
164 /* If there are any garbage slots, collect it and try again. */
165 if (c
->nr_garbage
&& collect_garbage_slots(c
) == 0)
168 /* All out of space. Need to allocate a new page. */
169 kip
= kmalloc(KPROBE_INSN_PAGE_SIZE(slots_per_page(c
)), GFP_KERNEL
);
174 * Use module_alloc so this page is within +/- 2GB of where the
175 * kernel image and loaded module images reside. This is required
176 * so x86_64 can correctly handle the %rip-relative fixups.
178 kip
->insns
= c
->alloc();
183 INIT_LIST_HEAD(&kip
->list
);
184 memset(kip
->slot_used
, SLOT_CLEAN
, slots_per_page(c
));
185 kip
->slot_used
[0] = SLOT_USED
;
189 list_add_rcu(&kip
->list
, &c
->pages
);
192 mutex_unlock(&c
->mutex
);
196 /* Return 1 if all garbages are collected, otherwise 0. */
197 static int collect_one_slot(struct kprobe_insn_page
*kip
, int idx
)
199 kip
->slot_used
[idx
] = SLOT_CLEAN
;
201 if (kip
->nused
== 0) {
203 * Page is no longer in use. Free it unless
204 * it's the last one. We keep the last one
205 * so as not to have to set it up again the
206 * next time somebody inserts a probe.
208 if (!list_is_singular(&kip
->list
)) {
209 list_del_rcu(&kip
->list
);
211 kip
->cache
->free(kip
->insns
);
219 static int collect_garbage_slots(struct kprobe_insn_cache
*c
)
221 struct kprobe_insn_page
*kip
, *next
;
223 /* Ensure no-one is interrupted on the garbages */
226 list_for_each_entry_safe(kip
, next
, &c
->pages
, list
) {
228 if (kip
->ngarbage
== 0)
230 kip
->ngarbage
= 0; /* we will collect all garbages */
231 for (i
= 0; i
< slots_per_page(c
); i
++) {
232 if (kip
->slot_used
[i
] == SLOT_DIRTY
&& collect_one_slot(kip
, i
))
240 void __free_insn_slot(struct kprobe_insn_cache
*c
,
241 kprobe_opcode_t
*slot
, int dirty
)
243 struct kprobe_insn_page
*kip
;
246 mutex_lock(&c
->mutex
);
248 list_for_each_entry_rcu(kip
, &c
->pages
, list
) {
249 idx
= ((long)slot
- (long)kip
->insns
) /
250 (c
->insn_size
* sizeof(kprobe_opcode_t
));
251 if (idx
>= 0 && idx
< slots_per_page(c
))
254 /* Could not find this slot. */
259 /* Mark and sweep: this may sleep */
261 /* Check double free */
262 WARN_ON(kip
->slot_used
[idx
] != SLOT_USED
);
264 kip
->slot_used
[idx
] = SLOT_DIRTY
;
266 if (++c
->nr_garbage
> slots_per_page(c
))
267 collect_garbage_slots(c
);
269 collect_one_slot(kip
, idx
);
272 mutex_unlock(&c
->mutex
);
276 * Check given address is on the page of kprobe instruction slots.
277 * This will be used for checking whether the address on a stack
278 * is on a text area or not.
280 bool __is_insn_slot_addr(struct kprobe_insn_cache
*c
, unsigned long addr
)
282 struct kprobe_insn_page
*kip
;
286 list_for_each_entry_rcu(kip
, &c
->pages
, list
) {
287 if (addr
>= (unsigned long)kip
->insns
&&
288 addr
< (unsigned long)kip
->insns
+ PAGE_SIZE
) {
298 #ifdef CONFIG_OPTPROBES
299 /* For optimized_kprobe buffer */
300 struct kprobe_insn_cache kprobe_optinsn_slots
= {
301 .mutex
= __MUTEX_INITIALIZER(kprobe_optinsn_slots
.mutex
),
302 .alloc
= alloc_insn_page
,
303 .free
= free_insn_page
,
304 .pages
= LIST_HEAD_INIT(kprobe_optinsn_slots
.pages
),
305 /* .insn_size is initialized later */
311 /* We have preemption disabled.. so it is safe to use __ versions */
312 static inline void set_kprobe_instance(struct kprobe
*kp
)
314 __this_cpu_write(kprobe_instance
, kp
);
317 static inline void reset_kprobe_instance(void)
319 __this_cpu_write(kprobe_instance
, NULL
);
323 * This routine is called either:
324 * - under the kprobe_mutex - during kprobe_[un]register()
326 * - with preemption disabled - from arch/xxx/kernel/kprobes.c
328 struct kprobe
*get_kprobe(void *addr
)
330 struct hlist_head
*head
;
333 head
= &kprobe_table
[hash_ptr(addr
, KPROBE_HASH_BITS
)];
334 hlist_for_each_entry_rcu(p
, head
, hlist
,
335 lockdep_is_held(&kprobe_mutex
)) {
342 NOKPROBE_SYMBOL(get_kprobe
);
344 static int aggr_pre_handler(struct kprobe
*p
, struct pt_regs
*regs
);
346 /* Return true if the kprobe is an aggregator */
347 static inline int kprobe_aggrprobe(struct kprobe
*p
)
349 return p
->pre_handler
== aggr_pre_handler
;
352 /* Return true(!0) if the kprobe is unused */
353 static inline int kprobe_unused(struct kprobe
*p
)
355 return kprobe_aggrprobe(p
) && kprobe_disabled(p
) &&
356 list_empty(&p
->list
);
360 * Keep all fields in the kprobe consistent
362 static inline void copy_kprobe(struct kprobe
*ap
, struct kprobe
*p
)
364 memcpy(&p
->opcode
, &ap
->opcode
, sizeof(kprobe_opcode_t
));
365 memcpy(&p
->ainsn
, &ap
->ainsn
, sizeof(struct arch_specific_insn
));
368 #ifdef CONFIG_OPTPROBES
369 /* NOTE: change this value only with kprobe_mutex held */
370 static bool kprobes_allow_optimization
;
373 * Call all pre_handler on the list, but ignores its return value.
374 * This must be called from arch-dep optimized caller.
376 void opt_pre_handler(struct kprobe
*p
, struct pt_regs
*regs
)
380 list_for_each_entry_rcu(kp
, &p
->list
, list
) {
381 if (kp
->pre_handler
&& likely(!kprobe_disabled(kp
))) {
382 set_kprobe_instance(kp
);
383 kp
->pre_handler(kp
, regs
);
385 reset_kprobe_instance();
388 NOKPROBE_SYMBOL(opt_pre_handler
);
390 /* Free optimized instructions and optimized_kprobe */
391 static void free_aggr_kprobe(struct kprobe
*p
)
393 struct optimized_kprobe
*op
;
395 op
= container_of(p
, struct optimized_kprobe
, kp
);
396 arch_remove_optimized_kprobe(op
);
397 arch_remove_kprobe(p
);
401 /* Return true(!0) if the kprobe is ready for optimization. */
402 static inline int kprobe_optready(struct kprobe
*p
)
404 struct optimized_kprobe
*op
;
406 if (kprobe_aggrprobe(p
)) {
407 op
= container_of(p
, struct optimized_kprobe
, kp
);
408 return arch_prepared_optinsn(&op
->optinsn
);
414 /* Return true(!0) if the kprobe is disarmed. Note: p must be on hash list */
415 static inline int kprobe_disarmed(struct kprobe
*p
)
417 struct optimized_kprobe
*op
;
419 /* If kprobe is not aggr/opt probe, just return kprobe is disabled */
420 if (!kprobe_aggrprobe(p
))
421 return kprobe_disabled(p
);
423 op
= container_of(p
, struct optimized_kprobe
, kp
);
425 return kprobe_disabled(p
) && list_empty(&op
->list
);
428 /* Return true(!0) if the probe is queued on (un)optimizing lists */
429 static int kprobe_queued(struct kprobe
*p
)
431 struct optimized_kprobe
*op
;
433 if (kprobe_aggrprobe(p
)) {
434 op
= container_of(p
, struct optimized_kprobe
, kp
);
435 if (!list_empty(&op
->list
))
442 * Return an optimized kprobe whose optimizing code replaces
443 * instructions including addr (exclude breakpoint).
445 static struct kprobe
*get_optimized_kprobe(unsigned long addr
)
448 struct kprobe
*p
= NULL
;
449 struct optimized_kprobe
*op
;
451 /* Don't check i == 0, since that is a breakpoint case. */
452 for (i
= 1; !p
&& i
< MAX_OPTIMIZED_LENGTH
; i
++)
453 p
= get_kprobe((void *)(addr
- i
));
455 if (p
&& kprobe_optready(p
)) {
456 op
= container_of(p
, struct optimized_kprobe
, kp
);
457 if (arch_within_optimized_kprobe(op
, addr
))
464 /* Optimization staging list, protected by kprobe_mutex */
465 static LIST_HEAD(optimizing_list
);
466 static LIST_HEAD(unoptimizing_list
);
467 static LIST_HEAD(freeing_list
);
469 static void kprobe_optimizer(struct work_struct
*work
);
470 static DECLARE_DELAYED_WORK(optimizing_work
, kprobe_optimizer
);
471 #define OPTIMIZE_DELAY 5
474 * Optimize (replace a breakpoint with a jump) kprobes listed on
477 static void do_optimize_kprobes(void)
479 lockdep_assert_held(&text_mutex
);
481 * The optimization/unoptimization refers online_cpus via
482 * stop_machine() and cpu-hotplug modifies online_cpus.
483 * And same time, text_mutex will be held in cpu-hotplug and here.
484 * This combination can cause a deadlock (cpu-hotplug try to lock
485 * text_mutex but stop_machine can not be done because online_cpus
487 * To avoid this deadlock, caller must have locked cpu hotplug
488 * for preventing cpu-hotplug outside of text_mutex locking.
490 lockdep_assert_cpus_held();
492 /* Optimization never be done when disarmed */
493 if (kprobes_all_disarmed
|| !kprobes_allow_optimization
||
494 list_empty(&optimizing_list
))
497 arch_optimize_kprobes(&optimizing_list
);
501 * Unoptimize (replace a jump with a breakpoint and remove the breakpoint
502 * if need) kprobes listed on unoptimizing_list.
504 static void do_unoptimize_kprobes(void)
506 struct optimized_kprobe
*op
, *tmp
;
508 lockdep_assert_held(&text_mutex
);
509 /* See comment in do_optimize_kprobes() */
510 lockdep_assert_cpus_held();
512 /* Unoptimization must be done anytime */
513 if (list_empty(&unoptimizing_list
))
516 arch_unoptimize_kprobes(&unoptimizing_list
, &freeing_list
);
517 /* Loop free_list for disarming */
518 list_for_each_entry_safe(op
, tmp
, &freeing_list
, list
) {
519 /* Switching from detour code to origin */
520 op
->kp
.flags
&= ~KPROBE_FLAG_OPTIMIZED
;
521 /* Disarm probes if marked disabled */
522 if (kprobe_disabled(&op
->kp
))
523 arch_disarm_kprobe(&op
->kp
);
524 if (kprobe_unused(&op
->kp
)) {
526 * Remove unused probes from hash list. After waiting
527 * for synchronization, these probes are reclaimed.
528 * (reclaiming is done by do_free_cleaned_kprobes.)
530 hlist_del_rcu(&op
->kp
.hlist
);
532 list_del_init(&op
->list
);
536 /* Reclaim all kprobes on the free_list */
537 static void do_free_cleaned_kprobes(void)
539 struct optimized_kprobe
*op
, *tmp
;
541 list_for_each_entry_safe(op
, tmp
, &freeing_list
, list
) {
542 list_del_init(&op
->list
);
543 if (WARN_ON_ONCE(!kprobe_unused(&op
->kp
))) {
545 * This must not happen, but if there is a kprobe
546 * still in use, keep it on kprobes hash list.
550 free_aggr_kprobe(&op
->kp
);
554 /* Start optimizer after OPTIMIZE_DELAY passed */
555 static void kick_kprobe_optimizer(void)
557 schedule_delayed_work(&optimizing_work
, OPTIMIZE_DELAY
);
560 /* Kprobe jump optimizer */
561 static void kprobe_optimizer(struct work_struct
*work
)
563 mutex_lock(&kprobe_mutex
);
565 mutex_lock(&text_mutex
);
566 /* Lock modules while optimizing kprobes */
567 mutex_lock(&module_mutex
);
570 * Step 1: Unoptimize kprobes and collect cleaned (unused and disarmed)
571 * kprobes before waiting for quiesence period.
573 do_unoptimize_kprobes();
576 * Step 2: Wait for quiesence period to ensure all potentially
577 * preempted tasks to have normally scheduled. Because optprobe
578 * may modify multiple instructions, there is a chance that Nth
579 * instruction is preempted. In that case, such tasks can return
580 * to 2nd-Nth byte of jump instruction. This wait is for avoiding it.
581 * Note that on non-preemptive kernel, this is transparently converted
582 * to synchronoze_sched() to wait for all interrupts to have completed.
584 synchronize_rcu_tasks();
586 /* Step 3: Optimize kprobes after quiesence period */
587 do_optimize_kprobes();
589 /* Step 4: Free cleaned kprobes after quiesence period */
590 do_free_cleaned_kprobes();
592 mutex_unlock(&module_mutex
);
593 mutex_unlock(&text_mutex
);
595 mutex_unlock(&kprobe_mutex
);
597 /* Step 5: Kick optimizer again if needed */
598 if (!list_empty(&optimizing_list
) || !list_empty(&unoptimizing_list
))
599 kick_kprobe_optimizer();
602 /* Wait for completing optimization and unoptimization */
603 void wait_for_kprobe_optimizer(void)
605 mutex_lock(&kprobe_mutex
);
607 while (!list_empty(&optimizing_list
) || !list_empty(&unoptimizing_list
)) {
608 mutex_unlock(&kprobe_mutex
);
610 /* this will also make optimizing_work execute immmediately */
611 flush_delayed_work(&optimizing_work
);
612 /* @optimizing_work might not have been queued yet, relax */
615 mutex_lock(&kprobe_mutex
);
618 mutex_unlock(&kprobe_mutex
);
621 static bool optprobe_queued_unopt(struct optimized_kprobe
*op
)
623 struct optimized_kprobe
*_op
;
625 list_for_each_entry(_op
, &unoptimizing_list
, list
) {
633 /* Optimize kprobe if p is ready to be optimized */
634 static void optimize_kprobe(struct kprobe
*p
)
636 struct optimized_kprobe
*op
;
638 /* Check if the kprobe is disabled or not ready for optimization. */
639 if (!kprobe_optready(p
) || !kprobes_allow_optimization
||
640 (kprobe_disabled(p
) || kprobes_all_disarmed
))
643 /* kprobes with post_handler can not be optimized */
647 op
= container_of(p
, struct optimized_kprobe
, kp
);
649 /* Check there is no other kprobes at the optimized instructions */
650 if (arch_check_optimized_kprobe(op
) < 0)
653 /* Check if it is already optimized. */
654 if (op
->kp
.flags
& KPROBE_FLAG_OPTIMIZED
) {
655 if (optprobe_queued_unopt(op
)) {
656 /* This is under unoptimizing. Just dequeue the probe */
657 list_del_init(&op
->list
);
661 op
->kp
.flags
|= KPROBE_FLAG_OPTIMIZED
;
663 /* On unoptimizing/optimizing_list, op must have OPTIMIZED flag */
664 if (WARN_ON_ONCE(!list_empty(&op
->list
)))
667 list_add(&op
->list
, &optimizing_list
);
668 kick_kprobe_optimizer();
671 /* Short cut to direct unoptimizing */
672 static void force_unoptimize_kprobe(struct optimized_kprobe
*op
)
674 lockdep_assert_cpus_held();
675 arch_unoptimize_kprobe(op
);
676 op
->kp
.flags
&= ~KPROBE_FLAG_OPTIMIZED
;
677 if (kprobe_disabled(&op
->kp
))
678 arch_disarm_kprobe(&op
->kp
);
681 /* Unoptimize a kprobe if p is optimized */
682 static void unoptimize_kprobe(struct kprobe
*p
, bool force
)
684 struct optimized_kprobe
*op
;
686 if (!kprobe_aggrprobe(p
) || kprobe_disarmed(p
))
687 return; /* This is not an optprobe nor optimized */
689 op
= container_of(p
, struct optimized_kprobe
, kp
);
690 if (!kprobe_optimized(p
))
693 if (!list_empty(&op
->list
)) {
694 if (optprobe_queued_unopt(op
)) {
695 /* Queued in unoptimizing queue */
698 * Forcibly unoptimize the kprobe here, and queue it
699 * in the freeing list for release afterwards.
701 force_unoptimize_kprobe(op
);
702 list_move(&op
->list
, &freeing_list
);
705 /* Dequeue from the optimizing queue */
706 list_del_init(&op
->list
);
707 op
->kp
.flags
&= ~KPROBE_FLAG_OPTIMIZED
;
712 /* Optimized kprobe case */
714 /* Forcibly update the code: this is a special case */
715 force_unoptimize_kprobe(op
);
717 list_add(&op
->list
, &unoptimizing_list
);
718 kick_kprobe_optimizer();
722 /* Cancel unoptimizing for reusing */
723 static int reuse_unused_kprobe(struct kprobe
*ap
)
725 struct optimized_kprobe
*op
;
728 * Unused kprobe MUST be on the way of delayed unoptimizing (means
729 * there is still a relative jump) and disabled.
731 op
= container_of(ap
, struct optimized_kprobe
, kp
);
732 WARN_ON_ONCE(list_empty(&op
->list
));
733 /* Enable the probe again */
734 ap
->flags
&= ~KPROBE_FLAG_DISABLED
;
735 /* Optimize it again (remove from op->list) */
736 if (!kprobe_optready(ap
))
743 /* Remove optimized instructions */
744 static void kill_optimized_kprobe(struct kprobe
*p
)
746 struct optimized_kprobe
*op
;
748 op
= container_of(p
, struct optimized_kprobe
, kp
);
749 if (!list_empty(&op
->list
))
750 /* Dequeue from the (un)optimization queue */
751 list_del_init(&op
->list
);
752 op
->kp
.flags
&= ~KPROBE_FLAG_OPTIMIZED
;
754 if (kprobe_unused(p
)) {
755 /* Enqueue if it is unused */
756 list_add(&op
->list
, &freeing_list
);
758 * Remove unused probes from the hash list. After waiting
759 * for synchronization, this probe is reclaimed.
760 * (reclaiming is done by do_free_cleaned_kprobes().)
762 hlist_del_rcu(&op
->kp
.hlist
);
765 /* Don't touch the code, because it is already freed. */
766 arch_remove_optimized_kprobe(op
);
770 void __prepare_optimized_kprobe(struct optimized_kprobe
*op
, struct kprobe
*p
)
772 if (!kprobe_ftrace(p
))
773 arch_prepare_optimized_kprobe(op
, p
);
776 /* Try to prepare optimized instructions */
777 static void prepare_optimized_kprobe(struct kprobe
*p
)
779 struct optimized_kprobe
*op
;
781 op
= container_of(p
, struct optimized_kprobe
, kp
);
782 __prepare_optimized_kprobe(op
, p
);
785 /* Allocate new optimized_kprobe and try to prepare optimized instructions */
786 static struct kprobe
*alloc_aggr_kprobe(struct kprobe
*p
)
788 struct optimized_kprobe
*op
;
790 op
= kzalloc(sizeof(struct optimized_kprobe
), GFP_KERNEL
);
794 INIT_LIST_HEAD(&op
->list
);
795 op
->kp
.addr
= p
->addr
;
796 __prepare_optimized_kprobe(op
, p
);
801 static void init_aggr_kprobe(struct kprobe
*ap
, struct kprobe
*p
);
804 * Prepare an optimized_kprobe and optimize it
805 * NOTE: p must be a normal registered kprobe
807 static void try_to_optimize_kprobe(struct kprobe
*p
)
810 struct optimized_kprobe
*op
;
812 /* Impossible to optimize ftrace-based kprobe */
813 if (kprobe_ftrace(p
))
816 /* For preparing optimization, jump_label_text_reserved() is called */
819 mutex_lock(&text_mutex
);
821 ap
= alloc_aggr_kprobe(p
);
825 op
= container_of(ap
, struct optimized_kprobe
, kp
);
826 if (!arch_prepared_optinsn(&op
->optinsn
)) {
827 /* If failed to setup optimizing, fallback to kprobe */
828 arch_remove_optimized_kprobe(op
);
833 init_aggr_kprobe(ap
, p
);
834 optimize_kprobe(ap
); /* This just kicks optimizer thread */
837 mutex_unlock(&text_mutex
);
843 static void optimize_all_kprobes(void)
845 struct hlist_head
*head
;
849 mutex_lock(&kprobe_mutex
);
850 /* If optimization is already allowed, just return */
851 if (kprobes_allow_optimization
)
855 kprobes_allow_optimization
= true;
856 for (i
= 0; i
< KPROBE_TABLE_SIZE
; i
++) {
857 head
= &kprobe_table
[i
];
858 hlist_for_each_entry(p
, head
, hlist
)
859 if (!kprobe_disabled(p
))
863 printk(KERN_INFO
"Kprobes globally optimized\n");
865 mutex_unlock(&kprobe_mutex
);
868 static void unoptimize_all_kprobes(void)
870 struct hlist_head
*head
;
874 mutex_lock(&kprobe_mutex
);
875 /* If optimization is already prohibited, just return */
876 if (!kprobes_allow_optimization
) {
877 mutex_unlock(&kprobe_mutex
);
882 kprobes_allow_optimization
= false;
883 for (i
= 0; i
< KPROBE_TABLE_SIZE
; i
++) {
884 head
= &kprobe_table
[i
];
885 hlist_for_each_entry(p
, head
, hlist
) {
886 if (!kprobe_disabled(p
))
887 unoptimize_kprobe(p
, false);
891 mutex_unlock(&kprobe_mutex
);
893 /* Wait for unoptimizing completion */
894 wait_for_kprobe_optimizer();
895 printk(KERN_INFO
"Kprobes globally unoptimized\n");
898 static DEFINE_MUTEX(kprobe_sysctl_mutex
);
899 int sysctl_kprobes_optimization
;
900 int proc_kprobes_optimization_handler(struct ctl_table
*table
, int write
,
901 void *buffer
, size_t *length
,
906 mutex_lock(&kprobe_sysctl_mutex
);
907 sysctl_kprobes_optimization
= kprobes_allow_optimization
? 1 : 0;
908 ret
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
910 if (sysctl_kprobes_optimization
)
911 optimize_all_kprobes();
913 unoptimize_all_kprobes();
914 mutex_unlock(&kprobe_sysctl_mutex
);
918 #endif /* CONFIG_SYSCTL */
920 /* Put a breakpoint for a probe. Must be called with text_mutex locked */
921 static void __arm_kprobe(struct kprobe
*p
)
925 /* Check collision with other optimized kprobes */
926 _p
= get_optimized_kprobe((unsigned long)p
->addr
);
928 /* Fallback to unoptimized kprobe */
929 unoptimize_kprobe(_p
, true);
932 optimize_kprobe(p
); /* Try to optimize (add kprobe to a list) */
935 /* Remove the breakpoint of a probe. Must be called with text_mutex locked */
936 static void __disarm_kprobe(struct kprobe
*p
, bool reopt
)
940 /* Try to unoptimize */
941 unoptimize_kprobe(p
, kprobes_all_disarmed
);
943 if (!kprobe_queued(p
)) {
944 arch_disarm_kprobe(p
);
945 /* If another kprobe was blocked, optimize it. */
946 _p
= get_optimized_kprobe((unsigned long)p
->addr
);
947 if (unlikely(_p
) && reopt
)
950 /* TODO: reoptimize others after unoptimized this probe */
953 #else /* !CONFIG_OPTPROBES */
955 #define optimize_kprobe(p) do {} while (0)
956 #define unoptimize_kprobe(p, f) do {} while (0)
957 #define kill_optimized_kprobe(p) do {} while (0)
958 #define prepare_optimized_kprobe(p) do {} while (0)
959 #define try_to_optimize_kprobe(p) do {} while (0)
960 #define __arm_kprobe(p) arch_arm_kprobe(p)
961 #define __disarm_kprobe(p, o) arch_disarm_kprobe(p)
962 #define kprobe_disarmed(p) kprobe_disabled(p)
963 #define wait_for_kprobe_optimizer() do {} while (0)
965 static int reuse_unused_kprobe(struct kprobe
*ap
)
968 * If the optimized kprobe is NOT supported, the aggr kprobe is
969 * released at the same time that the last aggregated kprobe is
971 * Thus there should be no chance to reuse unused kprobe.
973 printk(KERN_ERR
"Error: There should be no unused kprobe here.\n");
977 static void free_aggr_kprobe(struct kprobe
*p
)
979 arch_remove_kprobe(p
);
983 static struct kprobe
*alloc_aggr_kprobe(struct kprobe
*p
)
985 return kzalloc(sizeof(struct kprobe
), GFP_KERNEL
);
987 #endif /* CONFIG_OPTPROBES */
989 #ifdef CONFIG_KPROBES_ON_FTRACE
990 static struct ftrace_ops kprobe_ftrace_ops __read_mostly
= {
991 .func
= kprobe_ftrace_handler
,
992 .flags
= FTRACE_OPS_FL_SAVE_REGS
,
995 static struct ftrace_ops kprobe_ipmodify_ops __read_mostly
= {
996 .func
= kprobe_ftrace_handler
,
997 .flags
= FTRACE_OPS_FL_SAVE_REGS
| FTRACE_OPS_FL_IPMODIFY
,
1000 static int kprobe_ipmodify_enabled
;
1001 static int kprobe_ftrace_enabled
;
1003 /* Must ensure p->addr is really on ftrace */
1004 static int prepare_kprobe(struct kprobe
*p
)
1006 if (!kprobe_ftrace(p
))
1007 return arch_prepare_kprobe(p
);
1009 return arch_prepare_kprobe_ftrace(p
);
1012 /* Caller must lock kprobe_mutex */
1013 static int __arm_kprobe_ftrace(struct kprobe
*p
, struct ftrace_ops
*ops
,
1018 ret
= ftrace_set_filter_ip(ops
, (unsigned long)p
->addr
, 0, 0);
1020 pr_debug("Failed to arm kprobe-ftrace at %pS (%d)\n",
1026 ret
= register_ftrace_function(ops
);
1028 pr_debug("Failed to init kprobe-ftrace (%d)\n", ret
);
1038 * At this point, sinec ops is not registered, we should be sefe from
1039 * registering empty filter.
1041 ftrace_set_filter_ip(ops
, (unsigned long)p
->addr
, 1, 0);
1045 static int arm_kprobe_ftrace(struct kprobe
*p
)
1047 bool ipmodify
= (p
->post_handler
!= NULL
);
1049 return __arm_kprobe_ftrace(p
,
1050 ipmodify
? &kprobe_ipmodify_ops
: &kprobe_ftrace_ops
,
1051 ipmodify
? &kprobe_ipmodify_enabled
: &kprobe_ftrace_enabled
);
1054 /* Caller must lock kprobe_mutex */
1055 static int __disarm_kprobe_ftrace(struct kprobe
*p
, struct ftrace_ops
*ops
,
1061 ret
= unregister_ftrace_function(ops
);
1062 if (WARN(ret
< 0, "Failed to unregister kprobe-ftrace (%d)\n", ret
))
1068 ret
= ftrace_set_filter_ip(ops
, (unsigned long)p
->addr
, 1, 0);
1069 WARN_ONCE(ret
< 0, "Failed to disarm kprobe-ftrace at %pS (%d)\n",
1074 static int disarm_kprobe_ftrace(struct kprobe
*p
)
1076 bool ipmodify
= (p
->post_handler
!= NULL
);
1078 return __disarm_kprobe_ftrace(p
,
1079 ipmodify
? &kprobe_ipmodify_ops
: &kprobe_ftrace_ops
,
1080 ipmodify
? &kprobe_ipmodify_enabled
: &kprobe_ftrace_enabled
);
1082 #else /* !CONFIG_KPROBES_ON_FTRACE */
1083 #define prepare_kprobe(p) arch_prepare_kprobe(p)
1084 #define arm_kprobe_ftrace(p) (-ENODEV)
1085 #define disarm_kprobe_ftrace(p) (-ENODEV)
1088 /* Arm a kprobe with text_mutex */
1089 static int arm_kprobe(struct kprobe
*kp
)
1091 if (unlikely(kprobe_ftrace(kp
)))
1092 return arm_kprobe_ftrace(kp
);
1095 mutex_lock(&text_mutex
);
1097 mutex_unlock(&text_mutex
);
1103 /* Disarm a kprobe with text_mutex */
1104 static int disarm_kprobe(struct kprobe
*kp
, bool reopt
)
1106 if (unlikely(kprobe_ftrace(kp
)))
1107 return disarm_kprobe_ftrace(kp
);
1110 mutex_lock(&text_mutex
);
1111 __disarm_kprobe(kp
, reopt
);
1112 mutex_unlock(&text_mutex
);
1119 * Aggregate handlers for multiple kprobes support - these handlers
1120 * take care of invoking the individual kprobe handlers on p->list
1122 static int aggr_pre_handler(struct kprobe
*p
, struct pt_regs
*regs
)
1126 list_for_each_entry_rcu(kp
, &p
->list
, list
) {
1127 if (kp
->pre_handler
&& likely(!kprobe_disabled(kp
))) {
1128 set_kprobe_instance(kp
);
1129 if (kp
->pre_handler(kp
, regs
))
1132 reset_kprobe_instance();
1136 NOKPROBE_SYMBOL(aggr_pre_handler
);
1138 static void aggr_post_handler(struct kprobe
*p
, struct pt_regs
*regs
,
1139 unsigned long flags
)
1143 list_for_each_entry_rcu(kp
, &p
->list
, list
) {
1144 if (kp
->post_handler
&& likely(!kprobe_disabled(kp
))) {
1145 set_kprobe_instance(kp
);
1146 kp
->post_handler(kp
, regs
, flags
);
1147 reset_kprobe_instance();
1151 NOKPROBE_SYMBOL(aggr_post_handler
);
1153 static int aggr_fault_handler(struct kprobe
*p
, struct pt_regs
*regs
,
1156 struct kprobe
*cur
= __this_cpu_read(kprobe_instance
);
1159 * if we faulted "during" the execution of a user specified
1160 * probe handler, invoke just that probe's fault handler
1162 if (cur
&& cur
->fault_handler
) {
1163 if (cur
->fault_handler(cur
, regs
, trapnr
))
1168 NOKPROBE_SYMBOL(aggr_fault_handler
);
1170 /* Walks the list and increments nmissed count for multiprobe case */
1171 void kprobes_inc_nmissed_count(struct kprobe
*p
)
1174 if (!kprobe_aggrprobe(p
)) {
1177 list_for_each_entry_rcu(kp
, &p
->list
, list
)
1182 NOKPROBE_SYMBOL(kprobes_inc_nmissed_count
);
1184 void recycle_rp_inst(struct kretprobe_instance
*ri
,
1185 struct hlist_head
*head
)
1187 struct kretprobe
*rp
= ri
->rp
;
1189 /* remove rp inst off the rprobe_inst_table */
1190 hlist_del(&ri
->hlist
);
1191 INIT_HLIST_NODE(&ri
->hlist
);
1193 raw_spin_lock(&rp
->lock
);
1194 hlist_add_head(&ri
->hlist
, &rp
->free_instances
);
1195 raw_spin_unlock(&rp
->lock
);
1198 hlist_add_head(&ri
->hlist
, head
);
1200 NOKPROBE_SYMBOL(recycle_rp_inst
);
1202 void kretprobe_hash_lock(struct task_struct
*tsk
,
1203 struct hlist_head
**head
, unsigned long *flags
)
1204 __acquires(hlist_lock
)
1206 unsigned long hash
= hash_ptr(tsk
, KPROBE_HASH_BITS
);
1207 raw_spinlock_t
*hlist_lock
;
1209 *head
= &kretprobe_inst_table
[hash
];
1210 hlist_lock
= kretprobe_table_lock_ptr(hash
);
1211 raw_spin_lock_irqsave(hlist_lock
, *flags
);
1213 NOKPROBE_SYMBOL(kretprobe_hash_lock
);
1215 static void kretprobe_table_lock(unsigned long hash
,
1216 unsigned long *flags
)
1217 __acquires(hlist_lock
)
1219 raw_spinlock_t
*hlist_lock
= kretprobe_table_lock_ptr(hash
);
1220 raw_spin_lock_irqsave(hlist_lock
, *flags
);
1222 NOKPROBE_SYMBOL(kretprobe_table_lock
);
1224 void kretprobe_hash_unlock(struct task_struct
*tsk
,
1225 unsigned long *flags
)
1226 __releases(hlist_lock
)
1228 unsigned long hash
= hash_ptr(tsk
, KPROBE_HASH_BITS
);
1229 raw_spinlock_t
*hlist_lock
;
1231 hlist_lock
= kretprobe_table_lock_ptr(hash
);
1232 raw_spin_unlock_irqrestore(hlist_lock
, *flags
);
1234 NOKPROBE_SYMBOL(kretprobe_hash_unlock
);
1236 static void kretprobe_table_unlock(unsigned long hash
,
1237 unsigned long *flags
)
1238 __releases(hlist_lock
)
1240 raw_spinlock_t
*hlist_lock
= kretprobe_table_lock_ptr(hash
);
1241 raw_spin_unlock_irqrestore(hlist_lock
, *flags
);
1243 NOKPROBE_SYMBOL(kretprobe_table_unlock
);
1246 * This function is called from finish_task_switch when task tk becomes dead,
1247 * so that we can recycle any function-return probe instances associated
1248 * with this task. These left over instances represent probed functions
1249 * that have been called but will never return.
1251 void kprobe_flush_task(struct task_struct
*tk
)
1253 struct kretprobe_instance
*ri
;
1254 struct hlist_head
*head
, empty_rp
;
1255 struct hlist_node
*tmp
;
1256 unsigned long hash
, flags
= 0;
1258 if (unlikely(!kprobes_initialized
))
1259 /* Early boot. kretprobe_table_locks not yet initialized. */
1262 INIT_HLIST_HEAD(&empty_rp
);
1263 hash
= hash_ptr(tk
, KPROBE_HASH_BITS
);
1264 head
= &kretprobe_inst_table
[hash
];
1265 kretprobe_table_lock(hash
, &flags
);
1266 hlist_for_each_entry_safe(ri
, tmp
, head
, hlist
) {
1268 recycle_rp_inst(ri
, &empty_rp
);
1270 kretprobe_table_unlock(hash
, &flags
);
1271 hlist_for_each_entry_safe(ri
, tmp
, &empty_rp
, hlist
) {
1272 hlist_del(&ri
->hlist
);
1276 NOKPROBE_SYMBOL(kprobe_flush_task
);
1278 static inline void free_rp_inst(struct kretprobe
*rp
)
1280 struct kretprobe_instance
*ri
;
1281 struct hlist_node
*next
;
1283 hlist_for_each_entry_safe(ri
, next
, &rp
->free_instances
, hlist
) {
1284 hlist_del(&ri
->hlist
);
1289 static void cleanup_rp_inst(struct kretprobe
*rp
)
1291 unsigned long flags
, hash
;
1292 struct kretprobe_instance
*ri
;
1293 struct hlist_node
*next
;
1294 struct hlist_head
*head
;
1297 for (hash
= 0; hash
< KPROBE_TABLE_SIZE
; hash
++) {
1298 kretprobe_table_lock(hash
, &flags
);
1299 head
= &kretprobe_inst_table
[hash
];
1300 hlist_for_each_entry_safe(ri
, next
, head
, hlist
) {
1304 kretprobe_table_unlock(hash
, &flags
);
1308 NOKPROBE_SYMBOL(cleanup_rp_inst
);
1310 /* Add the new probe to ap->list */
1311 static int add_new_kprobe(struct kprobe
*ap
, struct kprobe
*p
)
1313 if (p
->post_handler
)
1314 unoptimize_kprobe(ap
, true); /* Fall back to normal kprobe */
1316 list_add_rcu(&p
->list
, &ap
->list
);
1317 if (p
->post_handler
&& !ap
->post_handler
)
1318 ap
->post_handler
= aggr_post_handler
;
1324 * Fill in the required fields of the "manager kprobe". Replace the
1325 * earlier kprobe in the hlist with the manager kprobe
1327 static void init_aggr_kprobe(struct kprobe
*ap
, struct kprobe
*p
)
1329 /* Copy p's insn slot to ap */
1331 flush_insn_slot(ap
);
1333 ap
->flags
= p
->flags
& ~KPROBE_FLAG_OPTIMIZED
;
1334 ap
->pre_handler
= aggr_pre_handler
;
1335 ap
->fault_handler
= aggr_fault_handler
;
1336 /* We don't care the kprobe which has gone. */
1337 if (p
->post_handler
&& !kprobe_gone(p
))
1338 ap
->post_handler
= aggr_post_handler
;
1340 INIT_LIST_HEAD(&ap
->list
);
1341 INIT_HLIST_NODE(&ap
->hlist
);
1343 list_add_rcu(&p
->list
, &ap
->list
);
1344 hlist_replace_rcu(&p
->hlist
, &ap
->hlist
);
1348 * This is the second or subsequent kprobe at the address - handle
1351 static int register_aggr_kprobe(struct kprobe
*orig_p
, struct kprobe
*p
)
1354 struct kprobe
*ap
= orig_p
;
1358 /* For preparing optimization, jump_label_text_reserved() is called */
1360 mutex_lock(&text_mutex
);
1362 if (!kprobe_aggrprobe(orig_p
)) {
1363 /* If orig_p is not an aggr_kprobe, create new aggr_kprobe. */
1364 ap
= alloc_aggr_kprobe(orig_p
);
1369 init_aggr_kprobe(ap
, orig_p
);
1370 } else if (kprobe_unused(ap
)) {
1371 /* This probe is going to die. Rescue it */
1372 ret
= reuse_unused_kprobe(ap
);
1377 if (kprobe_gone(ap
)) {
1379 * Attempting to insert new probe at the same location that
1380 * had a probe in the module vaddr area which already
1381 * freed. So, the instruction slot has already been
1382 * released. We need a new slot for the new probe.
1384 ret
= arch_prepare_kprobe(ap
);
1387 * Even if fail to allocate new slot, don't need to
1388 * free aggr_probe. It will be used next time, or
1389 * freed by unregister_kprobe.
1393 /* Prepare optimized instructions if possible. */
1394 prepare_optimized_kprobe(ap
);
1397 * Clear gone flag to prevent allocating new slot again, and
1398 * set disabled flag because it is not armed yet.
1400 ap
->flags
= (ap
->flags
& ~KPROBE_FLAG_GONE
)
1401 | KPROBE_FLAG_DISABLED
;
1404 /* Copy ap's insn slot to p */
1406 ret
= add_new_kprobe(ap
, p
);
1409 mutex_unlock(&text_mutex
);
1410 jump_label_unlock();
1413 if (ret
== 0 && kprobe_disabled(ap
) && !kprobe_disabled(p
)) {
1414 ap
->flags
&= ~KPROBE_FLAG_DISABLED
;
1415 if (!kprobes_all_disarmed
) {
1416 /* Arm the breakpoint again. */
1417 ret
= arm_kprobe(ap
);
1419 ap
->flags
|= KPROBE_FLAG_DISABLED
;
1420 list_del_rcu(&p
->list
);
1428 bool __weak
arch_within_kprobe_blacklist(unsigned long addr
)
1430 /* The __kprobes marked functions and entry code must not be probed */
1431 return addr
>= (unsigned long)__kprobes_text_start
&&
1432 addr
< (unsigned long)__kprobes_text_end
;
1435 static bool __within_kprobe_blacklist(unsigned long addr
)
1437 struct kprobe_blacklist_entry
*ent
;
1439 if (arch_within_kprobe_blacklist(addr
))
1442 * If there exists a kprobe_blacklist, verify and
1443 * fail any probe registration in the prohibited area
1445 list_for_each_entry(ent
, &kprobe_blacklist
, list
) {
1446 if (addr
>= ent
->start_addr
&& addr
< ent
->end_addr
)
1452 bool within_kprobe_blacklist(unsigned long addr
)
1454 char symname
[KSYM_NAME_LEN
], *p
;
1456 if (__within_kprobe_blacklist(addr
))
1459 /* Check if the address is on a suffixed-symbol */
1460 if (!lookup_symbol_name(addr
, symname
)) {
1461 p
= strchr(symname
, '.');
1465 addr
= (unsigned long)kprobe_lookup_name(symname
, 0);
1467 return __within_kprobe_blacklist(addr
);
1473 * If we have a symbol_name argument, look it up and add the offset field
1474 * to it. This way, we can specify a relative address to a symbol.
1475 * This returns encoded errors if it fails to look up symbol or invalid
1476 * combination of parameters.
1478 static kprobe_opcode_t
*_kprobe_addr(kprobe_opcode_t
*addr
,
1479 const char *symbol_name
, unsigned int offset
)
1481 if ((symbol_name
&& addr
) || (!symbol_name
&& !addr
))
1485 addr
= kprobe_lookup_name(symbol_name
, offset
);
1487 return ERR_PTR(-ENOENT
);
1490 addr
= (kprobe_opcode_t
*)(((char *)addr
) + offset
);
1495 return ERR_PTR(-EINVAL
);
1498 static kprobe_opcode_t
*kprobe_addr(struct kprobe
*p
)
1500 return _kprobe_addr(p
->addr
, p
->symbol_name
, p
->offset
);
1503 /* Check passed kprobe is valid and return kprobe in kprobe_table. */
1504 static struct kprobe
*__get_valid_kprobe(struct kprobe
*p
)
1506 struct kprobe
*ap
, *list_p
;
1508 lockdep_assert_held(&kprobe_mutex
);
1510 ap
= get_kprobe(p
->addr
);
1515 list_for_each_entry(list_p
, &ap
->list
, list
)
1517 /* kprobe p is a valid probe */
1525 /* Return error if the kprobe is being re-registered */
1526 static inline int check_kprobe_rereg(struct kprobe
*p
)
1530 mutex_lock(&kprobe_mutex
);
1531 if (__get_valid_kprobe(p
))
1533 mutex_unlock(&kprobe_mutex
);
1538 int __weak
arch_check_ftrace_location(struct kprobe
*p
)
1540 unsigned long ftrace_addr
;
1542 ftrace_addr
= ftrace_location((unsigned long)p
->addr
);
1544 #ifdef CONFIG_KPROBES_ON_FTRACE
1545 /* Given address is not on the instruction boundary */
1546 if ((unsigned long)p
->addr
!= ftrace_addr
)
1548 p
->flags
|= KPROBE_FLAG_FTRACE
;
1549 #else /* !CONFIG_KPROBES_ON_FTRACE */
1556 static int check_kprobe_address_safe(struct kprobe
*p
,
1557 struct module
**probed_mod
)
1561 ret
= arch_check_ftrace_location(p
);
1567 /* Ensure it is not in reserved area nor out of text */
1568 if (!kernel_text_address((unsigned long) p
->addr
) ||
1569 within_kprobe_blacklist((unsigned long) p
->addr
) ||
1570 jump_label_text_reserved(p
->addr
, p
->addr
) ||
1571 find_bug((unsigned long)p
->addr
)) {
1576 /* Check if are we probing a module */
1577 *probed_mod
= __module_text_address((unsigned long) p
->addr
);
1580 * We must hold a refcount of the probed module while updating
1581 * its code to prohibit unexpected unloading.
1583 if (unlikely(!try_module_get(*probed_mod
))) {
1589 * If the module freed .init.text, we couldn't insert
1592 if (within_module_init((unsigned long)p
->addr
, *probed_mod
) &&
1593 (*probed_mod
)->state
!= MODULE_STATE_COMING
) {
1594 module_put(*probed_mod
);
1601 jump_label_unlock();
1606 int register_kprobe(struct kprobe
*p
)
1609 struct kprobe
*old_p
;
1610 struct module
*probed_mod
;
1611 kprobe_opcode_t
*addr
;
1613 /* Adjust probe address from symbol */
1614 addr
= kprobe_addr(p
);
1616 return PTR_ERR(addr
);
1619 ret
= check_kprobe_rereg(p
);
1623 /* User can pass only KPROBE_FLAG_DISABLED to register_kprobe */
1624 p
->flags
&= KPROBE_FLAG_DISABLED
;
1626 INIT_LIST_HEAD(&p
->list
);
1628 ret
= check_kprobe_address_safe(p
, &probed_mod
);
1632 mutex_lock(&kprobe_mutex
);
1634 old_p
= get_kprobe(p
->addr
);
1636 /* Since this may unoptimize old_p, locking text_mutex. */
1637 ret
= register_aggr_kprobe(old_p
, p
);
1642 /* Prevent text modification */
1643 mutex_lock(&text_mutex
);
1644 ret
= prepare_kprobe(p
);
1645 mutex_unlock(&text_mutex
);
1650 INIT_HLIST_NODE(&p
->hlist
);
1651 hlist_add_head_rcu(&p
->hlist
,
1652 &kprobe_table
[hash_ptr(p
->addr
, KPROBE_HASH_BITS
)]);
1654 if (!kprobes_all_disarmed
&& !kprobe_disabled(p
)) {
1655 ret
= arm_kprobe(p
);
1657 hlist_del_rcu(&p
->hlist
);
1663 /* Try to optimize kprobe */
1664 try_to_optimize_kprobe(p
);
1666 mutex_unlock(&kprobe_mutex
);
1669 module_put(probed_mod
);
1673 EXPORT_SYMBOL_GPL(register_kprobe
);
1675 /* Check if all probes on the aggrprobe are disabled */
1676 static int aggr_kprobe_disabled(struct kprobe
*ap
)
1680 lockdep_assert_held(&kprobe_mutex
);
1682 list_for_each_entry(kp
, &ap
->list
, list
)
1683 if (!kprobe_disabled(kp
))
1685 * There is an active probe on the list.
1686 * We can't disable this ap.
1693 /* Disable one kprobe: Make sure called under kprobe_mutex is locked */
1694 static struct kprobe
*__disable_kprobe(struct kprobe
*p
)
1696 struct kprobe
*orig_p
;
1699 /* Get an original kprobe for return */
1700 orig_p
= __get_valid_kprobe(p
);
1701 if (unlikely(orig_p
== NULL
))
1702 return ERR_PTR(-EINVAL
);
1704 if (!kprobe_disabled(p
)) {
1705 /* Disable probe if it is a child probe */
1707 p
->flags
|= KPROBE_FLAG_DISABLED
;
1709 /* Try to disarm and disable this/parent probe */
1710 if (p
== orig_p
|| aggr_kprobe_disabled(orig_p
)) {
1712 * If kprobes_all_disarmed is set, orig_p
1713 * should have already been disarmed, so
1714 * skip unneed disarming process.
1716 if (!kprobes_all_disarmed
) {
1717 ret
= disarm_kprobe(orig_p
, true);
1719 p
->flags
&= ~KPROBE_FLAG_DISABLED
;
1720 return ERR_PTR(ret
);
1723 orig_p
->flags
|= KPROBE_FLAG_DISABLED
;
1731 * Unregister a kprobe without a scheduler synchronization.
1733 static int __unregister_kprobe_top(struct kprobe
*p
)
1735 struct kprobe
*ap
, *list_p
;
1737 /* Disable kprobe. This will disarm it if needed. */
1738 ap
= __disable_kprobe(p
);
1744 * This probe is an independent(and non-optimized) kprobe
1745 * (not an aggrprobe). Remove from the hash list.
1749 /* Following process expects this probe is an aggrprobe */
1750 WARN_ON(!kprobe_aggrprobe(ap
));
1752 if (list_is_singular(&ap
->list
) && kprobe_disarmed(ap
))
1754 * !disarmed could be happen if the probe is under delayed
1759 /* If disabling probe has special handlers, update aggrprobe */
1760 if (p
->post_handler
&& !kprobe_gone(p
)) {
1761 list_for_each_entry(list_p
, &ap
->list
, list
) {
1762 if ((list_p
!= p
) && (list_p
->post_handler
))
1765 ap
->post_handler
= NULL
;
1769 * Remove from the aggrprobe: this path will do nothing in
1770 * __unregister_kprobe_bottom().
1772 list_del_rcu(&p
->list
);
1773 if (!kprobe_disabled(ap
) && !kprobes_all_disarmed
)
1775 * Try to optimize this probe again, because post
1776 * handler may have been changed.
1778 optimize_kprobe(ap
);
1783 hlist_del_rcu(&ap
->hlist
);
1787 static void __unregister_kprobe_bottom(struct kprobe
*p
)
1791 if (list_empty(&p
->list
))
1792 /* This is an independent kprobe */
1793 arch_remove_kprobe(p
);
1794 else if (list_is_singular(&p
->list
)) {
1795 /* This is the last child of an aggrprobe */
1796 ap
= list_entry(p
->list
.next
, struct kprobe
, list
);
1798 free_aggr_kprobe(ap
);
1800 /* Otherwise, do nothing. */
1803 int register_kprobes(struct kprobe
**kps
, int num
)
1809 for (i
= 0; i
< num
; i
++) {
1810 ret
= register_kprobe(kps
[i
]);
1813 unregister_kprobes(kps
, i
);
1819 EXPORT_SYMBOL_GPL(register_kprobes
);
1821 void unregister_kprobe(struct kprobe
*p
)
1823 unregister_kprobes(&p
, 1);
1825 EXPORT_SYMBOL_GPL(unregister_kprobe
);
1827 void unregister_kprobes(struct kprobe
**kps
, int num
)
1833 mutex_lock(&kprobe_mutex
);
1834 for (i
= 0; i
< num
; i
++)
1835 if (__unregister_kprobe_top(kps
[i
]) < 0)
1836 kps
[i
]->addr
= NULL
;
1837 mutex_unlock(&kprobe_mutex
);
1840 for (i
= 0; i
< num
; i
++)
1842 __unregister_kprobe_bottom(kps
[i
]);
1844 EXPORT_SYMBOL_GPL(unregister_kprobes
);
1846 int __weak
kprobe_exceptions_notify(struct notifier_block
*self
,
1847 unsigned long val
, void *data
)
1851 NOKPROBE_SYMBOL(kprobe_exceptions_notify
);
1853 static struct notifier_block kprobe_exceptions_nb
= {
1854 .notifier_call
= kprobe_exceptions_notify
,
1855 .priority
= 0x7fffffff /* we need to be notified first */
1858 unsigned long __weak
arch_deref_entry_point(void *entry
)
1860 return (unsigned long)entry
;
1863 #ifdef CONFIG_KRETPROBES
1865 * This kprobe pre_handler is registered with every kretprobe. When probe
1866 * hits it will set up the return probe.
1868 static int pre_handler_kretprobe(struct kprobe
*p
, struct pt_regs
*regs
)
1870 struct kretprobe
*rp
= container_of(p
, struct kretprobe
, kp
);
1871 unsigned long hash
, flags
= 0;
1872 struct kretprobe_instance
*ri
;
1875 * To avoid deadlocks, prohibit return probing in NMI contexts,
1876 * just skip the probe and increase the (inexact) 'nmissed'
1877 * statistical counter, so that the user is informed that
1878 * something happened:
1880 if (unlikely(in_nmi())) {
1885 /* TODO: consider to only swap the RA after the last pre_handler fired */
1886 hash
= hash_ptr(current
, KPROBE_HASH_BITS
);
1887 raw_spin_lock_irqsave(&rp
->lock
, flags
);
1888 if (!hlist_empty(&rp
->free_instances
)) {
1889 ri
= hlist_entry(rp
->free_instances
.first
,
1890 struct kretprobe_instance
, hlist
);
1891 hlist_del(&ri
->hlist
);
1892 raw_spin_unlock_irqrestore(&rp
->lock
, flags
);
1897 if (rp
->entry_handler
&& rp
->entry_handler(ri
, regs
)) {
1898 raw_spin_lock_irqsave(&rp
->lock
, flags
);
1899 hlist_add_head(&ri
->hlist
, &rp
->free_instances
);
1900 raw_spin_unlock_irqrestore(&rp
->lock
, flags
);
1904 arch_prepare_kretprobe(ri
, regs
);
1906 /* XXX(hch): why is there no hlist_move_head? */
1907 INIT_HLIST_NODE(&ri
->hlist
);
1908 kretprobe_table_lock(hash
, &flags
);
1909 hlist_add_head(&ri
->hlist
, &kretprobe_inst_table
[hash
]);
1910 kretprobe_table_unlock(hash
, &flags
);
1913 raw_spin_unlock_irqrestore(&rp
->lock
, flags
);
1917 NOKPROBE_SYMBOL(pre_handler_kretprobe
);
1919 bool __weak
arch_kprobe_on_func_entry(unsigned long offset
)
1924 bool kprobe_on_func_entry(kprobe_opcode_t
*addr
, const char *sym
, unsigned long offset
)
1926 kprobe_opcode_t
*kp_addr
= _kprobe_addr(addr
, sym
, offset
);
1928 if (IS_ERR(kp_addr
))
1931 if (!kallsyms_lookup_size_offset((unsigned long)kp_addr
, NULL
, &offset
) ||
1932 !arch_kprobe_on_func_entry(offset
))
1938 int register_kretprobe(struct kretprobe
*rp
)
1941 struct kretprobe_instance
*inst
;
1945 if (!kprobe_on_func_entry(rp
->kp
.addr
, rp
->kp
.symbol_name
, rp
->kp
.offset
))
1948 if (kretprobe_blacklist_size
) {
1949 addr
= kprobe_addr(&rp
->kp
);
1951 return PTR_ERR(addr
);
1953 for (i
= 0; kretprobe_blacklist
[i
].name
!= NULL
; i
++) {
1954 if (kretprobe_blacklist
[i
].addr
== addr
)
1959 rp
->kp
.pre_handler
= pre_handler_kretprobe
;
1960 rp
->kp
.post_handler
= NULL
;
1961 rp
->kp
.fault_handler
= NULL
;
1963 /* Pre-allocate memory for max kretprobe instances */
1964 if (rp
->maxactive
<= 0) {
1965 #ifdef CONFIG_PREEMPTION
1966 rp
->maxactive
= max_t(unsigned int, 10, 2*num_possible_cpus());
1968 rp
->maxactive
= num_possible_cpus();
1971 raw_spin_lock_init(&rp
->lock
);
1972 INIT_HLIST_HEAD(&rp
->free_instances
);
1973 for (i
= 0; i
< rp
->maxactive
; i
++) {
1974 inst
= kmalloc(sizeof(struct kretprobe_instance
) +
1975 rp
->data_size
, GFP_KERNEL
);
1980 INIT_HLIST_NODE(&inst
->hlist
);
1981 hlist_add_head(&inst
->hlist
, &rp
->free_instances
);
1985 /* Establish function entry probe point */
1986 ret
= register_kprobe(&rp
->kp
);
1991 EXPORT_SYMBOL_GPL(register_kretprobe
);
1993 int register_kretprobes(struct kretprobe
**rps
, int num
)
1999 for (i
= 0; i
< num
; i
++) {
2000 ret
= register_kretprobe(rps
[i
]);
2003 unregister_kretprobes(rps
, i
);
2009 EXPORT_SYMBOL_GPL(register_kretprobes
);
2011 void unregister_kretprobe(struct kretprobe
*rp
)
2013 unregister_kretprobes(&rp
, 1);
2015 EXPORT_SYMBOL_GPL(unregister_kretprobe
);
2017 void unregister_kretprobes(struct kretprobe
**rps
, int num
)
2023 mutex_lock(&kprobe_mutex
);
2024 for (i
= 0; i
< num
; i
++)
2025 if (__unregister_kprobe_top(&rps
[i
]->kp
) < 0)
2026 rps
[i
]->kp
.addr
= NULL
;
2027 mutex_unlock(&kprobe_mutex
);
2030 for (i
= 0; i
< num
; i
++) {
2031 if (rps
[i
]->kp
.addr
) {
2032 __unregister_kprobe_bottom(&rps
[i
]->kp
);
2033 cleanup_rp_inst(rps
[i
]);
2037 EXPORT_SYMBOL_GPL(unregister_kretprobes
);
2039 #else /* CONFIG_KRETPROBES */
2040 int register_kretprobe(struct kretprobe
*rp
)
2044 EXPORT_SYMBOL_GPL(register_kretprobe
);
2046 int register_kretprobes(struct kretprobe
**rps
, int num
)
2050 EXPORT_SYMBOL_GPL(register_kretprobes
);
2052 void unregister_kretprobe(struct kretprobe
*rp
)
2055 EXPORT_SYMBOL_GPL(unregister_kretprobe
);
2057 void unregister_kretprobes(struct kretprobe
**rps
, int num
)
2060 EXPORT_SYMBOL_GPL(unregister_kretprobes
);
2062 static int pre_handler_kretprobe(struct kprobe
*p
, struct pt_regs
*regs
)
2066 NOKPROBE_SYMBOL(pre_handler_kretprobe
);
2068 #endif /* CONFIG_KRETPROBES */
2070 /* Set the kprobe gone and remove its instruction buffer. */
2071 static void kill_kprobe(struct kprobe
*p
)
2075 lockdep_assert_held(&kprobe_mutex
);
2077 p
->flags
|= KPROBE_FLAG_GONE
;
2078 if (kprobe_aggrprobe(p
)) {
2080 * If this is an aggr_kprobe, we have to list all the
2081 * chained probes and mark them GONE.
2083 list_for_each_entry(kp
, &p
->list
, list
)
2084 kp
->flags
|= KPROBE_FLAG_GONE
;
2085 p
->post_handler
= NULL
;
2086 kill_optimized_kprobe(p
);
2089 * Here, we can remove insn_slot safely, because no thread calls
2090 * the original probed function (which will be freed soon) any more.
2092 arch_remove_kprobe(p
);
2095 /* Disable one kprobe */
2096 int disable_kprobe(struct kprobe
*kp
)
2101 mutex_lock(&kprobe_mutex
);
2103 /* Disable this kprobe */
2104 p
= __disable_kprobe(kp
);
2108 mutex_unlock(&kprobe_mutex
);
2111 EXPORT_SYMBOL_GPL(disable_kprobe
);
2113 /* Enable one kprobe */
2114 int enable_kprobe(struct kprobe
*kp
)
2119 mutex_lock(&kprobe_mutex
);
2121 /* Check whether specified probe is valid. */
2122 p
= __get_valid_kprobe(kp
);
2123 if (unlikely(p
== NULL
)) {
2128 if (kprobe_gone(kp
)) {
2129 /* This kprobe has gone, we couldn't enable it. */
2135 kp
->flags
&= ~KPROBE_FLAG_DISABLED
;
2137 if (!kprobes_all_disarmed
&& kprobe_disabled(p
)) {
2138 p
->flags
&= ~KPROBE_FLAG_DISABLED
;
2139 ret
= arm_kprobe(p
);
2141 p
->flags
|= KPROBE_FLAG_DISABLED
;
2144 mutex_unlock(&kprobe_mutex
);
2147 EXPORT_SYMBOL_GPL(enable_kprobe
);
2149 /* Caller must NOT call this in usual path. This is only for critical case */
2150 void dump_kprobe(struct kprobe
*kp
)
2152 pr_err("Dumping kprobe:\n");
2153 pr_err("Name: %s\nOffset: %x\nAddress: %pS\n",
2154 kp
->symbol_name
, kp
->offset
, kp
->addr
);
2156 NOKPROBE_SYMBOL(dump_kprobe
);
2158 int kprobe_add_ksym_blacklist(unsigned long entry
)
2160 struct kprobe_blacklist_entry
*ent
;
2161 unsigned long offset
= 0, size
= 0;
2163 if (!kernel_text_address(entry
) ||
2164 !kallsyms_lookup_size_offset(entry
, &size
, &offset
))
2167 ent
= kmalloc(sizeof(*ent
), GFP_KERNEL
);
2170 ent
->start_addr
= entry
;
2171 ent
->end_addr
= entry
+ size
;
2172 INIT_LIST_HEAD(&ent
->list
);
2173 list_add_tail(&ent
->list
, &kprobe_blacklist
);
2178 /* Add all symbols in given area into kprobe blacklist */
2179 int kprobe_add_area_blacklist(unsigned long start
, unsigned long end
)
2181 unsigned long entry
;
2184 for (entry
= start
; entry
< end
; entry
+= ret
) {
2185 ret
= kprobe_add_ksym_blacklist(entry
);
2188 if (ret
== 0) /* In case of alias symbol */
2194 /* Remove all symbols in given area from kprobe blacklist */
2195 static void kprobe_remove_area_blacklist(unsigned long start
, unsigned long end
)
2197 struct kprobe_blacklist_entry
*ent
, *n
;
2199 list_for_each_entry_safe(ent
, n
, &kprobe_blacklist
, list
) {
2200 if (ent
->start_addr
< start
|| ent
->start_addr
>= end
)
2202 list_del(&ent
->list
);
2207 static void kprobe_remove_ksym_blacklist(unsigned long entry
)
2209 kprobe_remove_area_blacklist(entry
, entry
+ 1);
2212 int __init __weak
arch_populate_kprobe_blacklist(void)
2218 * Lookup and populate the kprobe_blacklist.
2220 * Unlike the kretprobe blacklist, we'll need to determine
2221 * the range of addresses that belong to the said functions,
2222 * since a kprobe need not necessarily be at the beginning
2225 static int __init
populate_kprobe_blacklist(unsigned long *start
,
2228 unsigned long entry
;
2229 unsigned long *iter
;
2232 for (iter
= start
; iter
< end
; iter
++) {
2233 entry
= arch_deref_entry_point((void *)*iter
);
2234 ret
= kprobe_add_ksym_blacklist(entry
);
2241 /* Symbols in __kprobes_text are blacklisted */
2242 ret
= kprobe_add_area_blacklist((unsigned long)__kprobes_text_start
,
2243 (unsigned long)__kprobes_text_end
);
2247 /* Symbols in noinstr section are blacklisted */
2248 ret
= kprobe_add_area_blacklist((unsigned long)__noinstr_text_start
,
2249 (unsigned long)__noinstr_text_end
);
2251 return ret
? : arch_populate_kprobe_blacklist();
2254 static void add_module_kprobe_blacklist(struct module
*mod
)
2256 unsigned long start
, end
;
2259 if (mod
->kprobe_blacklist
) {
2260 for (i
= 0; i
< mod
->num_kprobe_blacklist
; i
++)
2261 kprobe_add_ksym_blacklist(mod
->kprobe_blacklist
[i
]);
2264 start
= (unsigned long)mod
->kprobes_text_start
;
2266 end
= start
+ mod
->kprobes_text_size
;
2267 kprobe_add_area_blacklist(start
, end
);
2270 start
= (unsigned long)mod
->noinstr_text_start
;
2272 end
= start
+ mod
->noinstr_text_size
;
2273 kprobe_add_area_blacklist(start
, end
);
2277 static void remove_module_kprobe_blacklist(struct module
*mod
)
2279 unsigned long start
, end
;
2282 if (mod
->kprobe_blacklist
) {
2283 for (i
= 0; i
< mod
->num_kprobe_blacklist
; i
++)
2284 kprobe_remove_ksym_blacklist(mod
->kprobe_blacklist
[i
]);
2287 start
= (unsigned long)mod
->kprobes_text_start
;
2289 end
= start
+ mod
->kprobes_text_size
;
2290 kprobe_remove_area_blacklist(start
, end
);
2293 start
= (unsigned long)mod
->noinstr_text_start
;
2295 end
= start
+ mod
->noinstr_text_size
;
2296 kprobe_remove_area_blacklist(start
, end
);
2300 /* Module notifier call back, checking kprobes on the module */
2301 static int kprobes_module_callback(struct notifier_block
*nb
,
2302 unsigned long val
, void *data
)
2304 struct module
*mod
= data
;
2305 struct hlist_head
*head
;
2308 int checkcore
= (val
== MODULE_STATE_GOING
);
2310 if (val
== MODULE_STATE_COMING
) {
2311 mutex_lock(&kprobe_mutex
);
2312 add_module_kprobe_blacklist(mod
);
2313 mutex_unlock(&kprobe_mutex
);
2315 if (val
!= MODULE_STATE_GOING
&& val
!= MODULE_STATE_LIVE
)
2319 * When MODULE_STATE_GOING was notified, both of module .text and
2320 * .init.text sections would be freed. When MODULE_STATE_LIVE was
2321 * notified, only .init.text section would be freed. We need to
2322 * disable kprobes which have been inserted in the sections.
2324 mutex_lock(&kprobe_mutex
);
2325 for (i
= 0; i
< KPROBE_TABLE_SIZE
; i
++) {
2326 head
= &kprobe_table
[i
];
2327 hlist_for_each_entry(p
, head
, hlist
)
2328 if (within_module_init((unsigned long)p
->addr
, mod
) ||
2330 within_module_core((unsigned long)p
->addr
, mod
))) {
2332 * The vaddr this probe is installed will soon
2333 * be vfreed buy not synced to disk. Hence,
2334 * disarming the breakpoint isn't needed.
2336 * Note, this will also move any optimized probes
2337 * that are pending to be removed from their
2338 * corresponding lists to the freeing_list and
2339 * will not be touched by the delayed
2340 * kprobe_optimizer work handler.
2345 if (val
== MODULE_STATE_GOING
)
2346 remove_module_kprobe_blacklist(mod
);
2347 mutex_unlock(&kprobe_mutex
);
2351 static struct notifier_block kprobe_module_nb
= {
2352 .notifier_call
= kprobes_module_callback
,
2356 /* Markers of _kprobe_blacklist section */
2357 extern unsigned long __start_kprobe_blacklist
[];
2358 extern unsigned long __stop_kprobe_blacklist
[];
2360 static int __init
init_kprobes(void)
2364 /* FIXME allocate the probe table, currently defined statically */
2365 /* initialize all list heads */
2366 for (i
= 0; i
< KPROBE_TABLE_SIZE
; i
++) {
2367 INIT_HLIST_HEAD(&kprobe_table
[i
]);
2368 INIT_HLIST_HEAD(&kretprobe_inst_table
[i
]);
2369 raw_spin_lock_init(&(kretprobe_table_locks
[i
].lock
));
2372 err
= populate_kprobe_blacklist(__start_kprobe_blacklist
,
2373 __stop_kprobe_blacklist
);
2375 pr_err("kprobes: failed to populate blacklist: %d\n", err
);
2376 pr_err("Please take care of using kprobes.\n");
2379 if (kretprobe_blacklist_size
) {
2380 /* lookup the function address from its name */
2381 for (i
= 0; kretprobe_blacklist
[i
].name
!= NULL
; i
++) {
2382 kretprobe_blacklist
[i
].addr
=
2383 kprobe_lookup_name(kretprobe_blacklist
[i
].name
, 0);
2384 if (!kretprobe_blacklist
[i
].addr
)
2385 printk("kretprobe: lookup failed: %s\n",
2386 kretprobe_blacklist
[i
].name
);
2390 #if defined(CONFIG_OPTPROBES)
2391 #if defined(__ARCH_WANT_KPROBES_INSN_SLOT)
2392 /* Init kprobe_optinsn_slots */
2393 kprobe_optinsn_slots
.insn_size
= MAX_OPTINSN_SIZE
;
2395 /* By default, kprobes can be optimized */
2396 kprobes_allow_optimization
= true;
2399 /* By default, kprobes are armed */
2400 kprobes_all_disarmed
= false;
2402 err
= arch_init_kprobes();
2404 err
= register_die_notifier(&kprobe_exceptions_nb
);
2406 err
= register_module_notifier(&kprobe_module_nb
);
2408 kprobes_initialized
= (err
== 0);
2414 subsys_initcall(init_kprobes
);
2416 #ifdef CONFIG_DEBUG_FS
2417 static void report_probe(struct seq_file
*pi
, struct kprobe
*p
,
2418 const char *sym
, int offset
, char *modname
, struct kprobe
*pp
)
2421 void *addr
= p
->addr
;
2423 if (p
->pre_handler
== pre_handler_kretprobe
)
2428 if (!kallsyms_show_value())
2432 seq_printf(pi
, "%px %s %s+0x%x %s ",
2433 addr
, kprobe_type
, sym
, offset
,
2434 (modname
? modname
: " "));
2435 else /* try to use %pS */
2436 seq_printf(pi
, "%px %s %pS ",
2437 addr
, kprobe_type
, p
->addr
);
2441 seq_printf(pi
, "%s%s%s%s\n",
2442 (kprobe_gone(p
) ? "[GONE]" : ""),
2443 ((kprobe_disabled(p
) && !kprobe_gone(p
)) ? "[DISABLED]" : ""),
2444 (kprobe_optimized(pp
) ? "[OPTIMIZED]" : ""),
2445 (kprobe_ftrace(pp
) ? "[FTRACE]" : ""));
2448 static void *kprobe_seq_start(struct seq_file
*f
, loff_t
*pos
)
2450 return (*pos
< KPROBE_TABLE_SIZE
) ? pos
: NULL
;
2453 static void *kprobe_seq_next(struct seq_file
*f
, void *v
, loff_t
*pos
)
2456 if (*pos
>= KPROBE_TABLE_SIZE
)
2461 static void kprobe_seq_stop(struct seq_file
*f
, void *v
)
2466 static int show_kprobe_addr(struct seq_file
*pi
, void *v
)
2468 struct hlist_head
*head
;
2469 struct kprobe
*p
, *kp
;
2470 const char *sym
= NULL
;
2471 unsigned int i
= *(loff_t
*) v
;
2472 unsigned long offset
= 0;
2473 char *modname
, namebuf
[KSYM_NAME_LEN
];
2475 head
= &kprobe_table
[i
];
2477 hlist_for_each_entry_rcu(p
, head
, hlist
) {
2478 sym
= kallsyms_lookup((unsigned long)p
->addr
, NULL
,
2479 &offset
, &modname
, namebuf
);
2480 if (kprobe_aggrprobe(p
)) {
2481 list_for_each_entry_rcu(kp
, &p
->list
, list
)
2482 report_probe(pi
, kp
, sym
, offset
, modname
, p
);
2484 report_probe(pi
, p
, sym
, offset
, modname
, NULL
);
2490 static const struct seq_operations kprobes_sops
= {
2491 .start
= kprobe_seq_start
,
2492 .next
= kprobe_seq_next
,
2493 .stop
= kprobe_seq_stop
,
2494 .show
= show_kprobe_addr
2497 DEFINE_SEQ_ATTRIBUTE(kprobes
);
2499 /* kprobes/blacklist -- shows which functions can not be probed */
2500 static void *kprobe_blacklist_seq_start(struct seq_file
*m
, loff_t
*pos
)
2502 mutex_lock(&kprobe_mutex
);
2503 return seq_list_start(&kprobe_blacklist
, *pos
);
2506 static void *kprobe_blacklist_seq_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
2508 return seq_list_next(v
, &kprobe_blacklist
, pos
);
2511 static int kprobe_blacklist_seq_show(struct seq_file
*m
, void *v
)
2513 struct kprobe_blacklist_entry
*ent
=
2514 list_entry(v
, struct kprobe_blacklist_entry
, list
);
2517 * If /proc/kallsyms is not showing kernel address, we won't
2518 * show them here either.
2520 if (!kallsyms_show_value())
2521 seq_printf(m
, "0x%px-0x%px\t%ps\n", NULL
, NULL
,
2522 (void *)ent
->start_addr
);
2524 seq_printf(m
, "0x%px-0x%px\t%ps\n", (void *)ent
->start_addr
,
2525 (void *)ent
->end_addr
, (void *)ent
->start_addr
);
2529 static void kprobe_blacklist_seq_stop(struct seq_file
*f
, void *v
)
2531 mutex_unlock(&kprobe_mutex
);
2534 static const struct seq_operations kprobe_blacklist_sops
= {
2535 .start
= kprobe_blacklist_seq_start
,
2536 .next
= kprobe_blacklist_seq_next
,
2537 .stop
= kprobe_blacklist_seq_stop
,
2538 .show
= kprobe_blacklist_seq_show
,
2540 DEFINE_SEQ_ATTRIBUTE(kprobe_blacklist
);
2542 static int arm_all_kprobes(void)
2544 struct hlist_head
*head
;
2546 unsigned int i
, total
= 0, errors
= 0;
2549 mutex_lock(&kprobe_mutex
);
2551 /* If kprobes are armed, just return */
2552 if (!kprobes_all_disarmed
)
2553 goto already_enabled
;
2556 * optimize_kprobe() called by arm_kprobe() checks
2557 * kprobes_all_disarmed, so set kprobes_all_disarmed before
2560 kprobes_all_disarmed
= false;
2561 /* Arming kprobes doesn't optimize kprobe itself */
2562 for (i
= 0; i
< KPROBE_TABLE_SIZE
; i
++) {
2563 head
= &kprobe_table
[i
];
2564 /* Arm all kprobes on a best-effort basis */
2565 hlist_for_each_entry(p
, head
, hlist
) {
2566 if (!kprobe_disabled(p
)) {
2567 err
= arm_kprobe(p
);
2578 pr_warn("Kprobes globally enabled, but failed to arm %d out of %d probes\n",
2581 pr_info("Kprobes globally enabled\n");
2584 mutex_unlock(&kprobe_mutex
);
2588 static int disarm_all_kprobes(void)
2590 struct hlist_head
*head
;
2592 unsigned int i
, total
= 0, errors
= 0;
2595 mutex_lock(&kprobe_mutex
);
2597 /* If kprobes are already disarmed, just return */
2598 if (kprobes_all_disarmed
) {
2599 mutex_unlock(&kprobe_mutex
);
2603 kprobes_all_disarmed
= true;
2605 for (i
= 0; i
< KPROBE_TABLE_SIZE
; i
++) {
2606 head
= &kprobe_table
[i
];
2607 /* Disarm all kprobes on a best-effort basis */
2608 hlist_for_each_entry(p
, head
, hlist
) {
2609 if (!arch_trampoline_kprobe(p
) && !kprobe_disabled(p
)) {
2610 err
= disarm_kprobe(p
, false);
2621 pr_warn("Kprobes globally disabled, but failed to disarm %d out of %d probes\n",
2624 pr_info("Kprobes globally disabled\n");
2626 mutex_unlock(&kprobe_mutex
);
2628 /* Wait for disarming all kprobes by optimizer */
2629 wait_for_kprobe_optimizer();
2635 * XXX: The debugfs bool file interface doesn't allow for callbacks
2636 * when the bool state is switched. We can reuse that facility when
2639 static ssize_t
read_enabled_file_bool(struct file
*file
,
2640 char __user
*user_buf
, size_t count
, loff_t
*ppos
)
2644 if (!kprobes_all_disarmed
)
2650 return simple_read_from_buffer(user_buf
, count
, ppos
, buf
, 2);
2653 static ssize_t
write_enabled_file_bool(struct file
*file
,
2654 const char __user
*user_buf
, size_t count
, loff_t
*ppos
)
2660 buf_size
= min(count
, (sizeof(buf
)-1));
2661 if (copy_from_user(buf
, user_buf
, buf_size
))
2664 buf
[buf_size
] = '\0';
2669 ret
= arm_all_kprobes();
2674 ret
= disarm_all_kprobes();
2686 static const struct file_operations fops_kp
= {
2687 .read
= read_enabled_file_bool
,
2688 .write
= write_enabled_file_bool
,
2689 .llseek
= default_llseek
,
2692 static int __init
debugfs_kprobe_init(void)
2695 unsigned int value
= 1;
2697 dir
= debugfs_create_dir("kprobes", NULL
);
2699 debugfs_create_file("list", 0400, dir
, NULL
, &kprobes_fops
);
2701 debugfs_create_file("enabled", 0600, dir
, &value
, &fops_kp
);
2703 debugfs_create_file("blacklist", 0400, dir
, NULL
,
2704 &kprobe_blacklist_fops
);
2709 late_initcall(debugfs_kprobe_init
);
2710 #endif /* CONFIG_DEBUG_FS */