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>
38 #include <linux/perf_event.h>
39 #include <linux/static_call.h>
41 #include <asm/sections.h>
42 #include <asm/cacheflush.h>
43 #include <asm/errno.h>
44 #include <linux/uaccess.h>
46 #define KPROBE_HASH_BITS 6
47 #define KPROBE_TABLE_SIZE (1 << KPROBE_HASH_BITS)
50 static int kprobes_initialized
;
51 /* kprobe_table can be accessed by
52 * - Normal hlist traversal and RCU add/del under kprobe_mutex is held.
54 * - RCU hlist traversal under disabling preempt (breakpoint handlers)
56 static struct hlist_head kprobe_table
[KPROBE_TABLE_SIZE
];
57 static struct hlist_head kretprobe_inst_table
[KPROBE_TABLE_SIZE
];
59 /* NOTE: change this value only with kprobe_mutex held */
60 static bool kprobes_all_disarmed
;
62 /* This protects kprobe_table and optimizing_list */
63 static DEFINE_MUTEX(kprobe_mutex
);
64 static DEFINE_PER_CPU(struct kprobe
*, kprobe_instance
) = NULL
;
66 raw_spinlock_t lock ____cacheline_aligned_in_smp
;
67 } kretprobe_table_locks
[KPROBE_TABLE_SIZE
];
69 kprobe_opcode_t
* __weak
kprobe_lookup_name(const char *name
,
70 unsigned int __unused
)
72 return ((kprobe_opcode_t
*)(kallsyms_lookup_name(name
)));
75 static raw_spinlock_t
*kretprobe_table_lock_ptr(unsigned long hash
)
77 return &(kretprobe_table_locks
[hash
].lock
);
80 /* Blacklist -- list of struct kprobe_blacklist_entry */
81 static LIST_HEAD(kprobe_blacklist
);
83 #ifdef __ARCH_WANT_KPROBES_INSN_SLOT
85 * kprobe->ainsn.insn points to the copy of the instruction to be
86 * single-stepped. x86_64, POWER4 and above have no-exec support and
87 * stepping on the instruction on a vmalloced/kmalloced/data page
88 * is a recipe for disaster
90 struct kprobe_insn_page
{
91 struct list_head list
;
92 kprobe_opcode_t
*insns
; /* Page of instruction slots */
93 struct kprobe_insn_cache
*cache
;
99 #define KPROBE_INSN_PAGE_SIZE(slots) \
100 (offsetof(struct kprobe_insn_page, slot_used) + \
101 (sizeof(char) * (slots)))
103 static int slots_per_page(struct kprobe_insn_cache
*c
)
105 return PAGE_SIZE
/(c
->insn_size
* sizeof(kprobe_opcode_t
));
108 enum kprobe_slot_state
{
114 void __weak
*alloc_insn_page(void)
116 return module_alloc(PAGE_SIZE
);
119 void __weak
free_insn_page(void *page
)
121 module_memfree(page
);
124 struct kprobe_insn_cache kprobe_insn_slots
= {
125 .mutex
= __MUTEX_INITIALIZER(kprobe_insn_slots
.mutex
),
126 .alloc
= alloc_insn_page
,
127 .free
= free_insn_page
,
128 .sym
= KPROBE_INSN_PAGE_SYM
,
129 .pages
= LIST_HEAD_INIT(kprobe_insn_slots
.pages
),
130 .insn_size
= MAX_INSN_SIZE
,
133 static int collect_garbage_slots(struct kprobe_insn_cache
*c
);
136 * __get_insn_slot() - Find a slot on an executable page for an instruction.
137 * We allocate an executable page if there's no room on existing ones.
139 kprobe_opcode_t
*__get_insn_slot(struct kprobe_insn_cache
*c
)
141 struct kprobe_insn_page
*kip
;
142 kprobe_opcode_t
*slot
= NULL
;
144 /* Since the slot array is not protected by rcu, we need a mutex */
145 mutex_lock(&c
->mutex
);
148 list_for_each_entry_rcu(kip
, &c
->pages
, list
) {
149 if (kip
->nused
< slots_per_page(c
)) {
151 for (i
= 0; i
< slots_per_page(c
); i
++) {
152 if (kip
->slot_used
[i
] == SLOT_CLEAN
) {
153 kip
->slot_used
[i
] = SLOT_USED
;
155 slot
= kip
->insns
+ (i
* c
->insn_size
);
160 /* kip->nused is broken. Fix it. */
161 kip
->nused
= slots_per_page(c
);
167 /* If there are any garbage slots, collect it and try again. */
168 if (c
->nr_garbage
&& collect_garbage_slots(c
) == 0)
171 /* All out of space. Need to allocate a new page. */
172 kip
= kmalloc(KPROBE_INSN_PAGE_SIZE(slots_per_page(c
)), GFP_KERNEL
);
177 * Use module_alloc so this page is within +/- 2GB of where the
178 * kernel image and loaded module images reside. This is required
179 * so x86_64 can correctly handle the %rip-relative fixups.
181 kip
->insns
= c
->alloc();
186 INIT_LIST_HEAD(&kip
->list
);
187 memset(kip
->slot_used
, SLOT_CLEAN
, slots_per_page(c
));
188 kip
->slot_used
[0] = SLOT_USED
;
192 list_add_rcu(&kip
->list
, &c
->pages
);
195 /* Record the perf ksymbol register event after adding the page */
196 perf_event_ksymbol(PERF_RECORD_KSYMBOL_TYPE_OOL
, (unsigned long)kip
->insns
,
197 PAGE_SIZE
, false, c
->sym
);
199 mutex_unlock(&c
->mutex
);
203 /* Return 1 if all garbages are collected, otherwise 0. */
204 static int collect_one_slot(struct kprobe_insn_page
*kip
, int idx
)
206 kip
->slot_used
[idx
] = SLOT_CLEAN
;
208 if (kip
->nused
== 0) {
210 * Page is no longer in use. Free it unless
211 * it's the last one. We keep the last one
212 * so as not to have to set it up again the
213 * next time somebody inserts a probe.
215 if (!list_is_singular(&kip
->list
)) {
217 * Record perf ksymbol unregister event before removing
220 perf_event_ksymbol(PERF_RECORD_KSYMBOL_TYPE_OOL
,
221 (unsigned long)kip
->insns
, PAGE_SIZE
, true,
223 list_del_rcu(&kip
->list
);
225 kip
->cache
->free(kip
->insns
);
233 static int collect_garbage_slots(struct kprobe_insn_cache
*c
)
235 struct kprobe_insn_page
*kip
, *next
;
237 /* Ensure no-one is interrupted on the garbages */
240 list_for_each_entry_safe(kip
, next
, &c
->pages
, list
) {
242 if (kip
->ngarbage
== 0)
244 kip
->ngarbage
= 0; /* we will collect all garbages */
245 for (i
= 0; i
< slots_per_page(c
); i
++) {
246 if (kip
->slot_used
[i
] == SLOT_DIRTY
&& collect_one_slot(kip
, i
))
254 void __free_insn_slot(struct kprobe_insn_cache
*c
,
255 kprobe_opcode_t
*slot
, int dirty
)
257 struct kprobe_insn_page
*kip
;
260 mutex_lock(&c
->mutex
);
262 list_for_each_entry_rcu(kip
, &c
->pages
, list
) {
263 idx
= ((long)slot
- (long)kip
->insns
) /
264 (c
->insn_size
* sizeof(kprobe_opcode_t
));
265 if (idx
>= 0 && idx
< slots_per_page(c
))
268 /* Could not find this slot. */
273 /* Mark and sweep: this may sleep */
275 /* Check double free */
276 WARN_ON(kip
->slot_used
[idx
] != SLOT_USED
);
278 kip
->slot_used
[idx
] = SLOT_DIRTY
;
280 if (++c
->nr_garbage
> slots_per_page(c
))
281 collect_garbage_slots(c
);
283 collect_one_slot(kip
, idx
);
286 mutex_unlock(&c
->mutex
);
290 * Check given address is on the page of kprobe instruction slots.
291 * This will be used for checking whether the address on a stack
292 * is on a text area or not.
294 bool __is_insn_slot_addr(struct kprobe_insn_cache
*c
, unsigned long addr
)
296 struct kprobe_insn_page
*kip
;
300 list_for_each_entry_rcu(kip
, &c
->pages
, list
) {
301 if (addr
>= (unsigned long)kip
->insns
&&
302 addr
< (unsigned long)kip
->insns
+ PAGE_SIZE
) {
312 int kprobe_cache_get_kallsym(struct kprobe_insn_cache
*c
, unsigned int *symnum
,
313 unsigned long *value
, char *type
, char *sym
)
315 struct kprobe_insn_page
*kip
;
319 list_for_each_entry_rcu(kip
, &c
->pages
, list
) {
322 strlcpy(sym
, c
->sym
, KSYM_NAME_LEN
);
324 *value
= (unsigned long)kip
->insns
;
333 #ifdef CONFIG_OPTPROBES
334 /* For optimized_kprobe buffer */
335 struct kprobe_insn_cache kprobe_optinsn_slots
= {
336 .mutex
= __MUTEX_INITIALIZER(kprobe_optinsn_slots
.mutex
),
337 .alloc
= alloc_insn_page
,
338 .free
= free_insn_page
,
339 .sym
= KPROBE_OPTINSN_PAGE_SYM
,
340 .pages
= LIST_HEAD_INIT(kprobe_optinsn_slots
.pages
),
341 /* .insn_size is initialized later */
347 /* We have preemption disabled.. so it is safe to use __ versions */
348 static inline void set_kprobe_instance(struct kprobe
*kp
)
350 __this_cpu_write(kprobe_instance
, kp
);
353 static inline void reset_kprobe_instance(void)
355 __this_cpu_write(kprobe_instance
, NULL
);
359 * This routine is called either:
360 * - under the kprobe_mutex - during kprobe_[un]register()
362 * - with preemption disabled - from arch/xxx/kernel/kprobes.c
364 struct kprobe
*get_kprobe(void *addr
)
366 struct hlist_head
*head
;
369 head
= &kprobe_table
[hash_ptr(addr
, KPROBE_HASH_BITS
)];
370 hlist_for_each_entry_rcu(p
, head
, hlist
,
371 lockdep_is_held(&kprobe_mutex
)) {
378 NOKPROBE_SYMBOL(get_kprobe
);
380 static int aggr_pre_handler(struct kprobe
*p
, struct pt_regs
*regs
);
382 /* Return true if the kprobe is an aggregator */
383 static inline int kprobe_aggrprobe(struct kprobe
*p
)
385 return p
->pre_handler
== aggr_pre_handler
;
388 /* Return true(!0) if the kprobe is unused */
389 static inline int kprobe_unused(struct kprobe
*p
)
391 return kprobe_aggrprobe(p
) && kprobe_disabled(p
) &&
392 list_empty(&p
->list
);
396 * Keep all fields in the kprobe consistent
398 static inline void copy_kprobe(struct kprobe
*ap
, struct kprobe
*p
)
400 memcpy(&p
->opcode
, &ap
->opcode
, sizeof(kprobe_opcode_t
));
401 memcpy(&p
->ainsn
, &ap
->ainsn
, sizeof(struct arch_specific_insn
));
404 #ifdef CONFIG_OPTPROBES
405 /* NOTE: change this value only with kprobe_mutex held */
406 static bool kprobes_allow_optimization
;
409 * Call all pre_handler on the list, but ignores its return value.
410 * This must be called from arch-dep optimized caller.
412 void opt_pre_handler(struct kprobe
*p
, struct pt_regs
*regs
)
416 list_for_each_entry_rcu(kp
, &p
->list
, list
) {
417 if (kp
->pre_handler
&& likely(!kprobe_disabled(kp
))) {
418 set_kprobe_instance(kp
);
419 kp
->pre_handler(kp
, regs
);
421 reset_kprobe_instance();
424 NOKPROBE_SYMBOL(opt_pre_handler
);
426 /* Free optimized instructions and optimized_kprobe */
427 static void free_aggr_kprobe(struct kprobe
*p
)
429 struct optimized_kprobe
*op
;
431 op
= container_of(p
, struct optimized_kprobe
, kp
);
432 arch_remove_optimized_kprobe(op
);
433 arch_remove_kprobe(p
);
437 /* Return true(!0) if the kprobe is ready for optimization. */
438 static inline int kprobe_optready(struct kprobe
*p
)
440 struct optimized_kprobe
*op
;
442 if (kprobe_aggrprobe(p
)) {
443 op
= container_of(p
, struct optimized_kprobe
, kp
);
444 return arch_prepared_optinsn(&op
->optinsn
);
450 /* Return true(!0) if the kprobe is disarmed. Note: p must be on hash list */
451 static inline int kprobe_disarmed(struct kprobe
*p
)
453 struct optimized_kprobe
*op
;
455 /* If kprobe is not aggr/opt probe, just return kprobe is disabled */
456 if (!kprobe_aggrprobe(p
))
457 return kprobe_disabled(p
);
459 op
= container_of(p
, struct optimized_kprobe
, kp
);
461 return kprobe_disabled(p
) && list_empty(&op
->list
);
464 /* Return true(!0) if the probe is queued on (un)optimizing lists */
465 static int kprobe_queued(struct kprobe
*p
)
467 struct optimized_kprobe
*op
;
469 if (kprobe_aggrprobe(p
)) {
470 op
= container_of(p
, struct optimized_kprobe
, kp
);
471 if (!list_empty(&op
->list
))
478 * Return an optimized kprobe whose optimizing code replaces
479 * instructions including addr (exclude breakpoint).
481 static struct kprobe
*get_optimized_kprobe(unsigned long addr
)
484 struct kprobe
*p
= NULL
;
485 struct optimized_kprobe
*op
;
487 /* Don't check i == 0, since that is a breakpoint case. */
488 for (i
= 1; !p
&& i
< MAX_OPTIMIZED_LENGTH
; i
++)
489 p
= get_kprobe((void *)(addr
- i
));
491 if (p
&& kprobe_optready(p
)) {
492 op
= container_of(p
, struct optimized_kprobe
, kp
);
493 if (arch_within_optimized_kprobe(op
, addr
))
500 /* Optimization staging list, protected by kprobe_mutex */
501 static LIST_HEAD(optimizing_list
);
502 static LIST_HEAD(unoptimizing_list
);
503 static LIST_HEAD(freeing_list
);
505 static void kprobe_optimizer(struct work_struct
*work
);
506 static DECLARE_DELAYED_WORK(optimizing_work
, kprobe_optimizer
);
507 #define OPTIMIZE_DELAY 5
510 * Optimize (replace a breakpoint with a jump) kprobes listed on
513 static void do_optimize_kprobes(void)
515 lockdep_assert_held(&text_mutex
);
517 * The optimization/unoptimization refers online_cpus via
518 * stop_machine() and cpu-hotplug modifies online_cpus.
519 * And same time, text_mutex will be held in cpu-hotplug and here.
520 * This combination can cause a deadlock (cpu-hotplug try to lock
521 * text_mutex but stop_machine can not be done because online_cpus
523 * To avoid this deadlock, caller must have locked cpu hotplug
524 * for preventing cpu-hotplug outside of text_mutex locking.
526 lockdep_assert_cpus_held();
528 /* Optimization never be done when disarmed */
529 if (kprobes_all_disarmed
|| !kprobes_allow_optimization
||
530 list_empty(&optimizing_list
))
533 arch_optimize_kprobes(&optimizing_list
);
537 * Unoptimize (replace a jump with a breakpoint and remove the breakpoint
538 * if need) kprobes listed on unoptimizing_list.
540 static void do_unoptimize_kprobes(void)
542 struct optimized_kprobe
*op
, *tmp
;
544 lockdep_assert_held(&text_mutex
);
545 /* See comment in do_optimize_kprobes() */
546 lockdep_assert_cpus_held();
548 /* Unoptimization must be done anytime */
549 if (list_empty(&unoptimizing_list
))
552 arch_unoptimize_kprobes(&unoptimizing_list
, &freeing_list
);
553 /* Loop free_list for disarming */
554 list_for_each_entry_safe(op
, tmp
, &freeing_list
, list
) {
555 /* Switching from detour code to origin */
556 op
->kp
.flags
&= ~KPROBE_FLAG_OPTIMIZED
;
557 /* Disarm probes if marked disabled */
558 if (kprobe_disabled(&op
->kp
))
559 arch_disarm_kprobe(&op
->kp
);
560 if (kprobe_unused(&op
->kp
)) {
562 * Remove unused probes from hash list. After waiting
563 * for synchronization, these probes are reclaimed.
564 * (reclaiming is done by do_free_cleaned_kprobes.)
566 hlist_del_rcu(&op
->kp
.hlist
);
568 list_del_init(&op
->list
);
572 /* Reclaim all kprobes on the free_list */
573 static void do_free_cleaned_kprobes(void)
575 struct optimized_kprobe
*op
, *tmp
;
577 list_for_each_entry_safe(op
, tmp
, &freeing_list
, list
) {
578 list_del_init(&op
->list
);
579 if (WARN_ON_ONCE(!kprobe_unused(&op
->kp
))) {
581 * This must not happen, but if there is a kprobe
582 * still in use, keep it on kprobes hash list.
586 free_aggr_kprobe(&op
->kp
);
590 /* Start optimizer after OPTIMIZE_DELAY passed */
591 static void kick_kprobe_optimizer(void)
593 schedule_delayed_work(&optimizing_work
, OPTIMIZE_DELAY
);
596 /* Kprobe jump optimizer */
597 static void kprobe_optimizer(struct work_struct
*work
)
599 mutex_lock(&kprobe_mutex
);
601 mutex_lock(&text_mutex
);
604 * Step 1: Unoptimize kprobes and collect cleaned (unused and disarmed)
605 * kprobes before waiting for quiesence period.
607 do_unoptimize_kprobes();
610 * Step 2: Wait for quiesence period to ensure all potentially
611 * preempted tasks to have normally scheduled. Because optprobe
612 * may modify multiple instructions, there is a chance that Nth
613 * instruction is preempted. In that case, such tasks can return
614 * to 2nd-Nth byte of jump instruction. This wait is for avoiding it.
615 * Note that on non-preemptive kernel, this is transparently converted
616 * to synchronoze_sched() to wait for all interrupts to have completed.
618 synchronize_rcu_tasks();
620 /* Step 3: Optimize kprobes after quiesence period */
621 do_optimize_kprobes();
623 /* Step 4: Free cleaned kprobes after quiesence period */
624 do_free_cleaned_kprobes();
626 mutex_unlock(&text_mutex
);
629 /* Step 5: Kick optimizer again if needed */
630 if (!list_empty(&optimizing_list
) || !list_empty(&unoptimizing_list
))
631 kick_kprobe_optimizer();
633 mutex_unlock(&kprobe_mutex
);
636 /* Wait for completing optimization and unoptimization */
637 void wait_for_kprobe_optimizer(void)
639 mutex_lock(&kprobe_mutex
);
641 while (!list_empty(&optimizing_list
) || !list_empty(&unoptimizing_list
)) {
642 mutex_unlock(&kprobe_mutex
);
644 /* this will also make optimizing_work execute immmediately */
645 flush_delayed_work(&optimizing_work
);
646 /* @optimizing_work might not have been queued yet, relax */
649 mutex_lock(&kprobe_mutex
);
652 mutex_unlock(&kprobe_mutex
);
655 static bool optprobe_queued_unopt(struct optimized_kprobe
*op
)
657 struct optimized_kprobe
*_op
;
659 list_for_each_entry(_op
, &unoptimizing_list
, list
) {
667 /* Optimize kprobe if p is ready to be optimized */
668 static void optimize_kprobe(struct kprobe
*p
)
670 struct optimized_kprobe
*op
;
672 /* Check if the kprobe is disabled or not ready for optimization. */
673 if (!kprobe_optready(p
) || !kprobes_allow_optimization
||
674 (kprobe_disabled(p
) || kprobes_all_disarmed
))
677 /* kprobes with post_handler can not be optimized */
681 op
= container_of(p
, struct optimized_kprobe
, kp
);
683 /* Check there is no other kprobes at the optimized instructions */
684 if (arch_check_optimized_kprobe(op
) < 0)
687 /* Check if it is already optimized. */
688 if (op
->kp
.flags
& KPROBE_FLAG_OPTIMIZED
) {
689 if (optprobe_queued_unopt(op
)) {
690 /* This is under unoptimizing. Just dequeue the probe */
691 list_del_init(&op
->list
);
695 op
->kp
.flags
|= KPROBE_FLAG_OPTIMIZED
;
697 /* On unoptimizing/optimizing_list, op must have OPTIMIZED flag */
698 if (WARN_ON_ONCE(!list_empty(&op
->list
)))
701 list_add(&op
->list
, &optimizing_list
);
702 kick_kprobe_optimizer();
705 /* Short cut to direct unoptimizing */
706 static void force_unoptimize_kprobe(struct optimized_kprobe
*op
)
708 lockdep_assert_cpus_held();
709 arch_unoptimize_kprobe(op
);
710 op
->kp
.flags
&= ~KPROBE_FLAG_OPTIMIZED
;
713 /* Unoptimize a kprobe if p is optimized */
714 static void unoptimize_kprobe(struct kprobe
*p
, bool force
)
716 struct optimized_kprobe
*op
;
718 if (!kprobe_aggrprobe(p
) || kprobe_disarmed(p
))
719 return; /* This is not an optprobe nor optimized */
721 op
= container_of(p
, struct optimized_kprobe
, kp
);
722 if (!kprobe_optimized(p
))
725 if (!list_empty(&op
->list
)) {
726 if (optprobe_queued_unopt(op
)) {
727 /* Queued in unoptimizing queue */
730 * Forcibly unoptimize the kprobe here, and queue it
731 * in the freeing list for release afterwards.
733 force_unoptimize_kprobe(op
);
734 list_move(&op
->list
, &freeing_list
);
737 /* Dequeue from the optimizing queue */
738 list_del_init(&op
->list
);
739 op
->kp
.flags
&= ~KPROBE_FLAG_OPTIMIZED
;
744 /* Optimized kprobe case */
746 /* Forcibly update the code: this is a special case */
747 force_unoptimize_kprobe(op
);
749 list_add(&op
->list
, &unoptimizing_list
);
750 kick_kprobe_optimizer();
754 /* Cancel unoptimizing for reusing */
755 static int reuse_unused_kprobe(struct kprobe
*ap
)
757 struct optimized_kprobe
*op
;
760 * Unused kprobe MUST be on the way of delayed unoptimizing (means
761 * there is still a relative jump) and disabled.
763 op
= container_of(ap
, struct optimized_kprobe
, kp
);
764 WARN_ON_ONCE(list_empty(&op
->list
));
765 /* Enable the probe again */
766 ap
->flags
&= ~KPROBE_FLAG_DISABLED
;
767 /* Optimize it again (remove from op->list) */
768 if (!kprobe_optready(ap
))
775 /* Remove optimized instructions */
776 static void kill_optimized_kprobe(struct kprobe
*p
)
778 struct optimized_kprobe
*op
;
780 op
= container_of(p
, struct optimized_kprobe
, kp
);
781 if (!list_empty(&op
->list
))
782 /* Dequeue from the (un)optimization queue */
783 list_del_init(&op
->list
);
784 op
->kp
.flags
&= ~KPROBE_FLAG_OPTIMIZED
;
786 if (kprobe_unused(p
)) {
787 /* Enqueue if it is unused */
788 list_add(&op
->list
, &freeing_list
);
790 * Remove unused probes from the hash list. After waiting
791 * for synchronization, this probe is reclaimed.
792 * (reclaiming is done by do_free_cleaned_kprobes().)
794 hlist_del_rcu(&op
->kp
.hlist
);
797 /* Don't touch the code, because it is already freed. */
798 arch_remove_optimized_kprobe(op
);
802 void __prepare_optimized_kprobe(struct optimized_kprobe
*op
, struct kprobe
*p
)
804 if (!kprobe_ftrace(p
))
805 arch_prepare_optimized_kprobe(op
, p
);
808 /* Try to prepare optimized instructions */
809 static void prepare_optimized_kprobe(struct kprobe
*p
)
811 struct optimized_kprobe
*op
;
813 op
= container_of(p
, struct optimized_kprobe
, kp
);
814 __prepare_optimized_kprobe(op
, p
);
817 /* Allocate new optimized_kprobe and try to prepare optimized instructions */
818 static struct kprobe
*alloc_aggr_kprobe(struct kprobe
*p
)
820 struct optimized_kprobe
*op
;
822 op
= kzalloc(sizeof(struct optimized_kprobe
), GFP_KERNEL
);
826 INIT_LIST_HEAD(&op
->list
);
827 op
->kp
.addr
= p
->addr
;
828 __prepare_optimized_kprobe(op
, p
);
833 static void init_aggr_kprobe(struct kprobe
*ap
, struct kprobe
*p
);
836 * Prepare an optimized_kprobe and optimize it
837 * NOTE: p must be a normal registered kprobe
839 static void try_to_optimize_kprobe(struct kprobe
*p
)
842 struct optimized_kprobe
*op
;
844 /* Impossible to optimize ftrace-based kprobe */
845 if (kprobe_ftrace(p
))
848 /* For preparing optimization, jump_label_text_reserved() is called */
851 mutex_lock(&text_mutex
);
853 ap
= alloc_aggr_kprobe(p
);
857 op
= container_of(ap
, struct optimized_kprobe
, kp
);
858 if (!arch_prepared_optinsn(&op
->optinsn
)) {
859 /* If failed to setup optimizing, fallback to kprobe */
860 arch_remove_optimized_kprobe(op
);
865 init_aggr_kprobe(ap
, p
);
866 optimize_kprobe(ap
); /* This just kicks optimizer thread */
869 mutex_unlock(&text_mutex
);
875 static void optimize_all_kprobes(void)
877 struct hlist_head
*head
;
881 mutex_lock(&kprobe_mutex
);
882 /* If optimization is already allowed, just return */
883 if (kprobes_allow_optimization
)
887 kprobes_allow_optimization
= true;
888 for (i
= 0; i
< KPROBE_TABLE_SIZE
; i
++) {
889 head
= &kprobe_table
[i
];
890 hlist_for_each_entry(p
, head
, hlist
)
891 if (!kprobe_disabled(p
))
895 printk(KERN_INFO
"Kprobes globally optimized\n");
897 mutex_unlock(&kprobe_mutex
);
900 static void unoptimize_all_kprobes(void)
902 struct hlist_head
*head
;
906 mutex_lock(&kprobe_mutex
);
907 /* If optimization is already prohibited, just return */
908 if (!kprobes_allow_optimization
) {
909 mutex_unlock(&kprobe_mutex
);
914 kprobes_allow_optimization
= false;
915 for (i
= 0; i
< KPROBE_TABLE_SIZE
; i
++) {
916 head
= &kprobe_table
[i
];
917 hlist_for_each_entry(p
, head
, hlist
) {
918 if (!kprobe_disabled(p
))
919 unoptimize_kprobe(p
, false);
923 mutex_unlock(&kprobe_mutex
);
925 /* Wait for unoptimizing completion */
926 wait_for_kprobe_optimizer();
927 printk(KERN_INFO
"Kprobes globally unoptimized\n");
930 static DEFINE_MUTEX(kprobe_sysctl_mutex
);
931 int sysctl_kprobes_optimization
;
932 int proc_kprobes_optimization_handler(struct ctl_table
*table
, int write
,
933 void *buffer
, size_t *length
,
938 mutex_lock(&kprobe_sysctl_mutex
);
939 sysctl_kprobes_optimization
= kprobes_allow_optimization
? 1 : 0;
940 ret
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
942 if (sysctl_kprobes_optimization
)
943 optimize_all_kprobes();
945 unoptimize_all_kprobes();
946 mutex_unlock(&kprobe_sysctl_mutex
);
950 #endif /* CONFIG_SYSCTL */
952 /* Put a breakpoint for a probe. Must be called with text_mutex locked */
953 static void __arm_kprobe(struct kprobe
*p
)
957 /* Check collision with other optimized kprobes */
958 _p
= get_optimized_kprobe((unsigned long)p
->addr
);
960 /* Fallback to unoptimized kprobe */
961 unoptimize_kprobe(_p
, true);
964 optimize_kprobe(p
); /* Try to optimize (add kprobe to a list) */
967 /* Remove the breakpoint of a probe. Must be called with text_mutex locked */
968 static void __disarm_kprobe(struct kprobe
*p
, bool reopt
)
972 /* Try to unoptimize */
973 unoptimize_kprobe(p
, kprobes_all_disarmed
);
975 if (!kprobe_queued(p
)) {
976 arch_disarm_kprobe(p
);
977 /* If another kprobe was blocked, optimize it. */
978 _p
= get_optimized_kprobe((unsigned long)p
->addr
);
979 if (unlikely(_p
) && reopt
)
982 /* TODO: reoptimize others after unoptimized this probe */
985 #else /* !CONFIG_OPTPROBES */
987 #define optimize_kprobe(p) do {} while (0)
988 #define unoptimize_kprobe(p, f) do {} while (0)
989 #define kill_optimized_kprobe(p) do {} while (0)
990 #define prepare_optimized_kprobe(p) do {} while (0)
991 #define try_to_optimize_kprobe(p) do {} while (0)
992 #define __arm_kprobe(p) arch_arm_kprobe(p)
993 #define __disarm_kprobe(p, o) arch_disarm_kprobe(p)
994 #define kprobe_disarmed(p) kprobe_disabled(p)
995 #define wait_for_kprobe_optimizer() do {} while (0)
997 static int reuse_unused_kprobe(struct kprobe
*ap
)
1000 * If the optimized kprobe is NOT supported, the aggr kprobe is
1001 * released at the same time that the last aggregated kprobe is
1003 * Thus there should be no chance to reuse unused kprobe.
1005 printk(KERN_ERR
"Error: There should be no unused kprobe here.\n");
1009 static void free_aggr_kprobe(struct kprobe
*p
)
1011 arch_remove_kprobe(p
);
1015 static struct kprobe
*alloc_aggr_kprobe(struct kprobe
*p
)
1017 return kzalloc(sizeof(struct kprobe
), GFP_KERNEL
);
1019 #endif /* CONFIG_OPTPROBES */
1021 #ifdef CONFIG_KPROBES_ON_FTRACE
1022 static struct ftrace_ops kprobe_ftrace_ops __read_mostly
= {
1023 .func
= kprobe_ftrace_handler
,
1024 .flags
= FTRACE_OPS_FL_SAVE_REGS
,
1027 static struct ftrace_ops kprobe_ipmodify_ops __read_mostly
= {
1028 .func
= kprobe_ftrace_handler
,
1029 .flags
= FTRACE_OPS_FL_SAVE_REGS
| FTRACE_OPS_FL_IPMODIFY
,
1032 static int kprobe_ipmodify_enabled
;
1033 static int kprobe_ftrace_enabled
;
1035 /* Must ensure p->addr is really on ftrace */
1036 static int prepare_kprobe(struct kprobe
*p
)
1038 if (!kprobe_ftrace(p
))
1039 return arch_prepare_kprobe(p
);
1041 return arch_prepare_kprobe_ftrace(p
);
1044 /* Caller must lock kprobe_mutex */
1045 static int __arm_kprobe_ftrace(struct kprobe
*p
, struct ftrace_ops
*ops
,
1050 ret
= ftrace_set_filter_ip(ops
, (unsigned long)p
->addr
, 0, 0);
1052 pr_debug("Failed to arm kprobe-ftrace at %pS (%d)\n",
1058 ret
= register_ftrace_function(ops
);
1060 pr_debug("Failed to init kprobe-ftrace (%d)\n", ret
);
1070 * At this point, sinec ops is not registered, we should be sefe from
1071 * registering empty filter.
1073 ftrace_set_filter_ip(ops
, (unsigned long)p
->addr
, 1, 0);
1077 static int arm_kprobe_ftrace(struct kprobe
*p
)
1079 bool ipmodify
= (p
->post_handler
!= NULL
);
1081 return __arm_kprobe_ftrace(p
,
1082 ipmodify
? &kprobe_ipmodify_ops
: &kprobe_ftrace_ops
,
1083 ipmodify
? &kprobe_ipmodify_enabled
: &kprobe_ftrace_enabled
);
1086 /* Caller must lock kprobe_mutex */
1087 static int __disarm_kprobe_ftrace(struct kprobe
*p
, struct ftrace_ops
*ops
,
1093 ret
= unregister_ftrace_function(ops
);
1094 if (WARN(ret
< 0, "Failed to unregister kprobe-ftrace (%d)\n", ret
))
1100 ret
= ftrace_set_filter_ip(ops
, (unsigned long)p
->addr
, 1, 0);
1101 WARN_ONCE(ret
< 0, "Failed to disarm kprobe-ftrace at %pS (%d)\n",
1106 static int disarm_kprobe_ftrace(struct kprobe
*p
)
1108 bool ipmodify
= (p
->post_handler
!= NULL
);
1110 return __disarm_kprobe_ftrace(p
,
1111 ipmodify
? &kprobe_ipmodify_ops
: &kprobe_ftrace_ops
,
1112 ipmodify
? &kprobe_ipmodify_enabled
: &kprobe_ftrace_enabled
);
1114 #else /* !CONFIG_KPROBES_ON_FTRACE */
1115 static inline int prepare_kprobe(struct kprobe
*p
)
1117 return arch_prepare_kprobe(p
);
1120 static inline int arm_kprobe_ftrace(struct kprobe
*p
)
1125 static inline int disarm_kprobe_ftrace(struct kprobe
*p
)
1131 /* Arm a kprobe with text_mutex */
1132 static int arm_kprobe(struct kprobe
*kp
)
1134 if (unlikely(kprobe_ftrace(kp
)))
1135 return arm_kprobe_ftrace(kp
);
1138 mutex_lock(&text_mutex
);
1140 mutex_unlock(&text_mutex
);
1146 /* Disarm a kprobe with text_mutex */
1147 static int disarm_kprobe(struct kprobe
*kp
, bool reopt
)
1149 if (unlikely(kprobe_ftrace(kp
)))
1150 return disarm_kprobe_ftrace(kp
);
1153 mutex_lock(&text_mutex
);
1154 __disarm_kprobe(kp
, reopt
);
1155 mutex_unlock(&text_mutex
);
1162 * Aggregate handlers for multiple kprobes support - these handlers
1163 * take care of invoking the individual kprobe handlers on p->list
1165 static int aggr_pre_handler(struct kprobe
*p
, struct pt_regs
*regs
)
1169 list_for_each_entry_rcu(kp
, &p
->list
, list
) {
1170 if (kp
->pre_handler
&& likely(!kprobe_disabled(kp
))) {
1171 set_kprobe_instance(kp
);
1172 if (kp
->pre_handler(kp
, regs
))
1175 reset_kprobe_instance();
1179 NOKPROBE_SYMBOL(aggr_pre_handler
);
1181 static void aggr_post_handler(struct kprobe
*p
, struct pt_regs
*regs
,
1182 unsigned long flags
)
1186 list_for_each_entry_rcu(kp
, &p
->list
, list
) {
1187 if (kp
->post_handler
&& likely(!kprobe_disabled(kp
))) {
1188 set_kprobe_instance(kp
);
1189 kp
->post_handler(kp
, regs
, flags
);
1190 reset_kprobe_instance();
1194 NOKPROBE_SYMBOL(aggr_post_handler
);
1196 static int aggr_fault_handler(struct kprobe
*p
, struct pt_regs
*regs
,
1199 struct kprobe
*cur
= __this_cpu_read(kprobe_instance
);
1202 * if we faulted "during" the execution of a user specified
1203 * probe handler, invoke just that probe's fault handler
1205 if (cur
&& cur
->fault_handler
) {
1206 if (cur
->fault_handler(cur
, regs
, trapnr
))
1211 NOKPROBE_SYMBOL(aggr_fault_handler
);
1213 /* Walks the list and increments nmissed count for multiprobe case */
1214 void kprobes_inc_nmissed_count(struct kprobe
*p
)
1217 if (!kprobe_aggrprobe(p
)) {
1220 list_for_each_entry_rcu(kp
, &p
->list
, list
)
1225 NOKPROBE_SYMBOL(kprobes_inc_nmissed_count
);
1227 static void recycle_rp_inst(struct kretprobe_instance
*ri
)
1229 struct kretprobe
*rp
= ri
->rp
;
1231 /* remove rp inst off the rprobe_inst_table */
1232 hlist_del(&ri
->hlist
);
1233 INIT_HLIST_NODE(&ri
->hlist
);
1235 raw_spin_lock(&rp
->lock
);
1236 hlist_add_head(&ri
->hlist
, &rp
->free_instances
);
1237 raw_spin_unlock(&rp
->lock
);
1241 NOKPROBE_SYMBOL(recycle_rp_inst
);
1243 static void kretprobe_hash_lock(struct task_struct
*tsk
,
1244 struct hlist_head
**head
, unsigned long *flags
)
1245 __acquires(hlist_lock
)
1247 unsigned long hash
= hash_ptr(tsk
, KPROBE_HASH_BITS
);
1248 raw_spinlock_t
*hlist_lock
;
1250 *head
= &kretprobe_inst_table
[hash
];
1251 hlist_lock
= kretprobe_table_lock_ptr(hash
);
1252 raw_spin_lock_irqsave(hlist_lock
, *flags
);
1254 NOKPROBE_SYMBOL(kretprobe_hash_lock
);
1256 static void kretprobe_table_lock(unsigned long hash
,
1257 unsigned long *flags
)
1258 __acquires(hlist_lock
)
1260 raw_spinlock_t
*hlist_lock
= kretprobe_table_lock_ptr(hash
);
1261 raw_spin_lock_irqsave(hlist_lock
, *flags
);
1263 NOKPROBE_SYMBOL(kretprobe_table_lock
);
1265 static void kretprobe_hash_unlock(struct task_struct
*tsk
,
1266 unsigned long *flags
)
1267 __releases(hlist_lock
)
1269 unsigned long hash
= hash_ptr(tsk
, KPROBE_HASH_BITS
);
1270 raw_spinlock_t
*hlist_lock
;
1272 hlist_lock
= kretprobe_table_lock_ptr(hash
);
1273 raw_spin_unlock_irqrestore(hlist_lock
, *flags
);
1275 NOKPROBE_SYMBOL(kretprobe_hash_unlock
);
1277 static void kretprobe_table_unlock(unsigned long hash
,
1278 unsigned long *flags
)
1279 __releases(hlist_lock
)
1281 raw_spinlock_t
*hlist_lock
= kretprobe_table_lock_ptr(hash
);
1282 raw_spin_unlock_irqrestore(hlist_lock
, *flags
);
1284 NOKPROBE_SYMBOL(kretprobe_table_unlock
);
1286 static struct kprobe kprobe_busy
= {
1287 .addr
= (void *) get_kprobe
,
1290 void kprobe_busy_begin(void)
1292 struct kprobe_ctlblk
*kcb
;
1295 __this_cpu_write(current_kprobe
, &kprobe_busy
);
1296 kcb
= get_kprobe_ctlblk();
1297 kcb
->kprobe_status
= KPROBE_HIT_ACTIVE
;
1300 void kprobe_busy_end(void)
1302 __this_cpu_write(current_kprobe
, NULL
);
1307 * This function is called from finish_task_switch when task tk becomes dead,
1308 * so that we can recycle any function-return probe instances associated
1309 * with this task. These left over instances represent probed functions
1310 * that have been called but will never return.
1312 void kprobe_flush_task(struct task_struct
*tk
)
1314 struct kretprobe_instance
*ri
;
1315 struct hlist_head
*head
;
1316 struct hlist_node
*tmp
;
1317 unsigned long hash
, flags
= 0;
1319 if (unlikely(!kprobes_initialized
))
1320 /* Early boot. kretprobe_table_locks not yet initialized. */
1323 kprobe_busy_begin();
1325 hash
= hash_ptr(tk
, KPROBE_HASH_BITS
);
1326 head
= &kretprobe_inst_table
[hash
];
1327 kretprobe_table_lock(hash
, &flags
);
1328 hlist_for_each_entry_safe(ri
, tmp
, head
, hlist
) {
1330 recycle_rp_inst(ri
);
1332 kretprobe_table_unlock(hash
, &flags
);
1336 NOKPROBE_SYMBOL(kprobe_flush_task
);
1338 static inline void free_rp_inst(struct kretprobe
*rp
)
1340 struct kretprobe_instance
*ri
;
1341 struct hlist_node
*next
;
1343 hlist_for_each_entry_safe(ri
, next
, &rp
->free_instances
, hlist
) {
1344 hlist_del(&ri
->hlist
);
1349 static void cleanup_rp_inst(struct kretprobe
*rp
)
1351 unsigned long flags
, hash
;
1352 struct kretprobe_instance
*ri
;
1353 struct hlist_node
*next
;
1354 struct hlist_head
*head
;
1356 /* To avoid recursive kretprobe by NMI, set kprobe busy here */
1357 kprobe_busy_begin();
1358 for (hash
= 0; hash
< KPROBE_TABLE_SIZE
; hash
++) {
1359 kretprobe_table_lock(hash
, &flags
);
1360 head
= &kretprobe_inst_table
[hash
];
1361 hlist_for_each_entry_safe(ri
, next
, head
, hlist
) {
1365 kretprobe_table_unlock(hash
, &flags
);
1371 NOKPROBE_SYMBOL(cleanup_rp_inst
);
1373 /* Add the new probe to ap->list */
1374 static int add_new_kprobe(struct kprobe
*ap
, struct kprobe
*p
)
1376 if (p
->post_handler
)
1377 unoptimize_kprobe(ap
, true); /* Fall back to normal kprobe */
1379 list_add_rcu(&p
->list
, &ap
->list
);
1380 if (p
->post_handler
&& !ap
->post_handler
)
1381 ap
->post_handler
= aggr_post_handler
;
1387 * Fill in the required fields of the "manager kprobe". Replace the
1388 * earlier kprobe in the hlist with the manager kprobe
1390 static void init_aggr_kprobe(struct kprobe
*ap
, struct kprobe
*p
)
1392 /* Copy p's insn slot to ap */
1394 flush_insn_slot(ap
);
1396 ap
->flags
= p
->flags
& ~KPROBE_FLAG_OPTIMIZED
;
1397 ap
->pre_handler
= aggr_pre_handler
;
1398 ap
->fault_handler
= aggr_fault_handler
;
1399 /* We don't care the kprobe which has gone. */
1400 if (p
->post_handler
&& !kprobe_gone(p
))
1401 ap
->post_handler
= aggr_post_handler
;
1403 INIT_LIST_HEAD(&ap
->list
);
1404 INIT_HLIST_NODE(&ap
->hlist
);
1406 list_add_rcu(&p
->list
, &ap
->list
);
1407 hlist_replace_rcu(&p
->hlist
, &ap
->hlist
);
1411 * This is the second or subsequent kprobe at the address - handle
1414 static int register_aggr_kprobe(struct kprobe
*orig_p
, struct kprobe
*p
)
1417 struct kprobe
*ap
= orig_p
;
1421 /* For preparing optimization, jump_label_text_reserved() is called */
1423 mutex_lock(&text_mutex
);
1425 if (!kprobe_aggrprobe(orig_p
)) {
1426 /* If orig_p is not an aggr_kprobe, create new aggr_kprobe. */
1427 ap
= alloc_aggr_kprobe(orig_p
);
1432 init_aggr_kprobe(ap
, orig_p
);
1433 } else if (kprobe_unused(ap
)) {
1434 /* This probe is going to die. Rescue it */
1435 ret
= reuse_unused_kprobe(ap
);
1440 if (kprobe_gone(ap
)) {
1442 * Attempting to insert new probe at the same location that
1443 * had a probe in the module vaddr area which already
1444 * freed. So, the instruction slot has already been
1445 * released. We need a new slot for the new probe.
1447 ret
= arch_prepare_kprobe(ap
);
1450 * Even if fail to allocate new slot, don't need to
1451 * free aggr_probe. It will be used next time, or
1452 * freed by unregister_kprobe.
1456 /* Prepare optimized instructions if possible. */
1457 prepare_optimized_kprobe(ap
);
1460 * Clear gone flag to prevent allocating new slot again, and
1461 * set disabled flag because it is not armed yet.
1463 ap
->flags
= (ap
->flags
& ~KPROBE_FLAG_GONE
)
1464 | KPROBE_FLAG_DISABLED
;
1467 /* Copy ap's insn slot to p */
1469 ret
= add_new_kprobe(ap
, p
);
1472 mutex_unlock(&text_mutex
);
1473 jump_label_unlock();
1476 if (ret
== 0 && kprobe_disabled(ap
) && !kprobe_disabled(p
)) {
1477 ap
->flags
&= ~KPROBE_FLAG_DISABLED
;
1478 if (!kprobes_all_disarmed
) {
1479 /* Arm the breakpoint again. */
1480 ret
= arm_kprobe(ap
);
1482 ap
->flags
|= KPROBE_FLAG_DISABLED
;
1483 list_del_rcu(&p
->list
);
1491 bool __weak
arch_within_kprobe_blacklist(unsigned long addr
)
1493 /* The __kprobes marked functions and entry code must not be probed */
1494 return addr
>= (unsigned long)__kprobes_text_start
&&
1495 addr
< (unsigned long)__kprobes_text_end
;
1498 static bool __within_kprobe_blacklist(unsigned long addr
)
1500 struct kprobe_blacklist_entry
*ent
;
1502 if (arch_within_kprobe_blacklist(addr
))
1505 * If there exists a kprobe_blacklist, verify and
1506 * fail any probe registration in the prohibited area
1508 list_for_each_entry(ent
, &kprobe_blacklist
, list
) {
1509 if (addr
>= ent
->start_addr
&& addr
< ent
->end_addr
)
1515 bool within_kprobe_blacklist(unsigned long addr
)
1517 char symname
[KSYM_NAME_LEN
], *p
;
1519 if (__within_kprobe_blacklist(addr
))
1522 /* Check if the address is on a suffixed-symbol */
1523 if (!lookup_symbol_name(addr
, symname
)) {
1524 p
= strchr(symname
, '.');
1528 addr
= (unsigned long)kprobe_lookup_name(symname
, 0);
1530 return __within_kprobe_blacklist(addr
);
1536 * If we have a symbol_name argument, look it up and add the offset field
1537 * to it. This way, we can specify a relative address to a symbol.
1538 * This returns encoded errors if it fails to look up symbol or invalid
1539 * combination of parameters.
1541 static kprobe_opcode_t
*_kprobe_addr(kprobe_opcode_t
*addr
,
1542 const char *symbol_name
, unsigned int offset
)
1544 if ((symbol_name
&& addr
) || (!symbol_name
&& !addr
))
1548 addr
= kprobe_lookup_name(symbol_name
, offset
);
1550 return ERR_PTR(-ENOENT
);
1553 addr
= (kprobe_opcode_t
*)(((char *)addr
) + offset
);
1558 return ERR_PTR(-EINVAL
);
1561 static kprobe_opcode_t
*kprobe_addr(struct kprobe
*p
)
1563 return _kprobe_addr(p
->addr
, p
->symbol_name
, p
->offset
);
1566 /* Check passed kprobe is valid and return kprobe in kprobe_table. */
1567 static struct kprobe
*__get_valid_kprobe(struct kprobe
*p
)
1569 struct kprobe
*ap
, *list_p
;
1571 lockdep_assert_held(&kprobe_mutex
);
1573 ap
= get_kprobe(p
->addr
);
1578 list_for_each_entry(list_p
, &ap
->list
, list
)
1580 /* kprobe p is a valid probe */
1588 /* Return error if the kprobe is being re-registered */
1589 static inline int check_kprobe_rereg(struct kprobe
*p
)
1593 mutex_lock(&kprobe_mutex
);
1594 if (__get_valid_kprobe(p
))
1596 mutex_unlock(&kprobe_mutex
);
1601 int __weak
arch_check_ftrace_location(struct kprobe
*p
)
1603 unsigned long ftrace_addr
;
1605 ftrace_addr
= ftrace_location((unsigned long)p
->addr
);
1607 #ifdef CONFIG_KPROBES_ON_FTRACE
1608 /* Given address is not on the instruction boundary */
1609 if ((unsigned long)p
->addr
!= ftrace_addr
)
1611 p
->flags
|= KPROBE_FLAG_FTRACE
;
1612 #else /* !CONFIG_KPROBES_ON_FTRACE */
1619 static int check_kprobe_address_safe(struct kprobe
*p
,
1620 struct module
**probed_mod
)
1624 ret
= arch_check_ftrace_location(p
);
1630 /* Ensure it is not in reserved area nor out of text */
1631 if (!kernel_text_address((unsigned long) p
->addr
) ||
1632 within_kprobe_blacklist((unsigned long) p
->addr
) ||
1633 jump_label_text_reserved(p
->addr
, p
->addr
) ||
1634 static_call_text_reserved(p
->addr
, p
->addr
) ||
1635 find_bug((unsigned long)p
->addr
)) {
1640 /* Check if are we probing a module */
1641 *probed_mod
= __module_text_address((unsigned long) p
->addr
);
1644 * We must hold a refcount of the probed module while updating
1645 * its code to prohibit unexpected unloading.
1647 if (unlikely(!try_module_get(*probed_mod
))) {
1653 * If the module freed .init.text, we couldn't insert
1656 if (within_module_init((unsigned long)p
->addr
, *probed_mod
) &&
1657 (*probed_mod
)->state
!= MODULE_STATE_COMING
) {
1658 module_put(*probed_mod
);
1665 jump_label_unlock();
1670 int register_kprobe(struct kprobe
*p
)
1673 struct kprobe
*old_p
;
1674 struct module
*probed_mod
;
1675 kprobe_opcode_t
*addr
;
1677 /* Adjust probe address from symbol */
1678 addr
= kprobe_addr(p
);
1680 return PTR_ERR(addr
);
1683 ret
= check_kprobe_rereg(p
);
1687 /* User can pass only KPROBE_FLAG_DISABLED to register_kprobe */
1688 p
->flags
&= KPROBE_FLAG_DISABLED
;
1690 INIT_LIST_HEAD(&p
->list
);
1692 ret
= check_kprobe_address_safe(p
, &probed_mod
);
1696 mutex_lock(&kprobe_mutex
);
1698 old_p
= get_kprobe(p
->addr
);
1700 /* Since this may unoptimize old_p, locking text_mutex. */
1701 ret
= register_aggr_kprobe(old_p
, p
);
1706 /* Prevent text modification */
1707 mutex_lock(&text_mutex
);
1708 ret
= prepare_kprobe(p
);
1709 mutex_unlock(&text_mutex
);
1714 INIT_HLIST_NODE(&p
->hlist
);
1715 hlist_add_head_rcu(&p
->hlist
,
1716 &kprobe_table
[hash_ptr(p
->addr
, KPROBE_HASH_BITS
)]);
1718 if (!kprobes_all_disarmed
&& !kprobe_disabled(p
)) {
1719 ret
= arm_kprobe(p
);
1721 hlist_del_rcu(&p
->hlist
);
1727 /* Try to optimize kprobe */
1728 try_to_optimize_kprobe(p
);
1730 mutex_unlock(&kprobe_mutex
);
1733 module_put(probed_mod
);
1737 EXPORT_SYMBOL_GPL(register_kprobe
);
1739 /* Check if all probes on the aggrprobe are disabled */
1740 static int aggr_kprobe_disabled(struct kprobe
*ap
)
1744 lockdep_assert_held(&kprobe_mutex
);
1746 list_for_each_entry(kp
, &ap
->list
, list
)
1747 if (!kprobe_disabled(kp
))
1749 * There is an active probe on the list.
1750 * We can't disable this ap.
1757 /* Disable one kprobe: Make sure called under kprobe_mutex is locked */
1758 static struct kprobe
*__disable_kprobe(struct kprobe
*p
)
1760 struct kprobe
*orig_p
;
1763 /* Get an original kprobe for return */
1764 orig_p
= __get_valid_kprobe(p
);
1765 if (unlikely(orig_p
== NULL
))
1766 return ERR_PTR(-EINVAL
);
1768 if (!kprobe_disabled(p
)) {
1769 /* Disable probe if it is a child probe */
1771 p
->flags
|= KPROBE_FLAG_DISABLED
;
1773 /* Try to disarm and disable this/parent probe */
1774 if (p
== orig_p
|| aggr_kprobe_disabled(orig_p
)) {
1776 * If kprobes_all_disarmed is set, orig_p
1777 * should have already been disarmed, so
1778 * skip unneed disarming process.
1780 if (!kprobes_all_disarmed
) {
1781 ret
= disarm_kprobe(orig_p
, true);
1783 p
->flags
&= ~KPROBE_FLAG_DISABLED
;
1784 return ERR_PTR(ret
);
1787 orig_p
->flags
|= KPROBE_FLAG_DISABLED
;
1795 * Unregister a kprobe without a scheduler synchronization.
1797 static int __unregister_kprobe_top(struct kprobe
*p
)
1799 struct kprobe
*ap
, *list_p
;
1801 /* Disable kprobe. This will disarm it if needed. */
1802 ap
= __disable_kprobe(p
);
1808 * This probe is an independent(and non-optimized) kprobe
1809 * (not an aggrprobe). Remove from the hash list.
1813 /* Following process expects this probe is an aggrprobe */
1814 WARN_ON(!kprobe_aggrprobe(ap
));
1816 if (list_is_singular(&ap
->list
) && kprobe_disarmed(ap
))
1818 * !disarmed could be happen if the probe is under delayed
1823 /* If disabling probe has special handlers, update aggrprobe */
1824 if (p
->post_handler
&& !kprobe_gone(p
)) {
1825 list_for_each_entry(list_p
, &ap
->list
, list
) {
1826 if ((list_p
!= p
) && (list_p
->post_handler
))
1829 ap
->post_handler
= NULL
;
1833 * Remove from the aggrprobe: this path will do nothing in
1834 * __unregister_kprobe_bottom().
1836 list_del_rcu(&p
->list
);
1837 if (!kprobe_disabled(ap
) && !kprobes_all_disarmed
)
1839 * Try to optimize this probe again, because post
1840 * handler may have been changed.
1842 optimize_kprobe(ap
);
1847 hlist_del_rcu(&ap
->hlist
);
1851 static void __unregister_kprobe_bottom(struct kprobe
*p
)
1855 if (list_empty(&p
->list
))
1856 /* This is an independent kprobe */
1857 arch_remove_kprobe(p
);
1858 else if (list_is_singular(&p
->list
)) {
1859 /* This is the last child of an aggrprobe */
1860 ap
= list_entry(p
->list
.next
, struct kprobe
, list
);
1862 free_aggr_kprobe(ap
);
1864 /* Otherwise, do nothing. */
1867 int register_kprobes(struct kprobe
**kps
, int num
)
1873 for (i
= 0; i
< num
; i
++) {
1874 ret
= register_kprobe(kps
[i
]);
1877 unregister_kprobes(kps
, i
);
1883 EXPORT_SYMBOL_GPL(register_kprobes
);
1885 void unregister_kprobe(struct kprobe
*p
)
1887 unregister_kprobes(&p
, 1);
1889 EXPORT_SYMBOL_GPL(unregister_kprobe
);
1891 void unregister_kprobes(struct kprobe
**kps
, int num
)
1897 mutex_lock(&kprobe_mutex
);
1898 for (i
= 0; i
< num
; i
++)
1899 if (__unregister_kprobe_top(kps
[i
]) < 0)
1900 kps
[i
]->addr
= NULL
;
1901 mutex_unlock(&kprobe_mutex
);
1904 for (i
= 0; i
< num
; i
++)
1906 __unregister_kprobe_bottom(kps
[i
]);
1908 EXPORT_SYMBOL_GPL(unregister_kprobes
);
1910 int __weak
kprobe_exceptions_notify(struct notifier_block
*self
,
1911 unsigned long val
, void *data
)
1915 NOKPROBE_SYMBOL(kprobe_exceptions_notify
);
1917 static struct notifier_block kprobe_exceptions_nb
= {
1918 .notifier_call
= kprobe_exceptions_notify
,
1919 .priority
= 0x7fffffff /* we need to be notified first */
1922 unsigned long __weak
arch_deref_entry_point(void *entry
)
1924 return (unsigned long)entry
;
1927 #ifdef CONFIG_KRETPROBES
1929 unsigned long __kretprobe_trampoline_handler(struct pt_regs
*regs
,
1930 void *trampoline_address
,
1931 void *frame_pointer
)
1933 struct kretprobe_instance
*ri
= NULL
, *last
= NULL
;
1934 struct hlist_head
*head
;
1935 struct hlist_node
*tmp
;
1936 unsigned long flags
;
1937 kprobe_opcode_t
*correct_ret_addr
= NULL
;
1938 bool skipped
= false;
1940 kretprobe_hash_lock(current
, &head
, &flags
);
1943 * It is possible to have multiple instances associated with a given
1944 * task either because multiple functions in the call path have
1945 * return probes installed on them, and/or more than one
1946 * return probe was registered for a target function.
1948 * We can handle this because:
1949 * - instances are always pushed into the head of the list
1950 * - when multiple return probes are registered for the same
1951 * function, the (chronologically) first instance's ret_addr
1952 * will be the real return address, and all the rest will
1953 * point to kretprobe_trampoline.
1955 hlist_for_each_entry(ri
, head
, hlist
) {
1956 if (ri
->task
!= current
)
1957 /* another task is sharing our hash bucket */
1960 * Return probes must be pushed on this hash list correct
1961 * order (same as return order) so that it can be popped
1962 * correctly. However, if we find it is pushed it incorrect
1963 * order, this means we find a function which should not be
1964 * probed, because the wrong order entry is pushed on the
1965 * path of processing other kretprobe itself.
1967 if (ri
->fp
!= frame_pointer
) {
1969 pr_warn("kretprobe is stacked incorrectly. Trying to fixup.\n");
1974 correct_ret_addr
= ri
->ret_addr
;
1976 pr_warn("%ps must be blacklisted because of incorrect kretprobe order\n",
1979 if (correct_ret_addr
!= trampoline_address
)
1981 * This is the real return address. Any other
1982 * instances associated with this task are for
1983 * other calls deeper on the call stack
1988 BUG_ON(!correct_ret_addr
|| (correct_ret_addr
== trampoline_address
));
1991 hlist_for_each_entry_safe(ri
, tmp
, head
, hlist
) {
1992 if (ri
->task
!= current
)
1993 /* another task is sharing our hash bucket */
1995 if (ri
->fp
!= frame_pointer
)
1998 if (ri
->rp
&& ri
->rp
->handler
) {
1999 struct kprobe
*prev
= kprobe_running();
2001 __this_cpu_write(current_kprobe
, &ri
->rp
->kp
);
2002 ri
->ret_addr
= correct_ret_addr
;
2003 ri
->rp
->handler(ri
, regs
);
2004 __this_cpu_write(current_kprobe
, prev
);
2007 recycle_rp_inst(ri
);
2013 kretprobe_hash_unlock(current
, &flags
);
2015 return (unsigned long)correct_ret_addr
;
2017 NOKPROBE_SYMBOL(__kretprobe_trampoline_handler
)
2020 * This kprobe pre_handler is registered with every kretprobe. When probe
2021 * hits it will set up the return probe.
2023 static int pre_handler_kretprobe(struct kprobe
*p
, struct pt_regs
*regs
)
2025 struct kretprobe
*rp
= container_of(p
, struct kretprobe
, kp
);
2026 unsigned long hash
, flags
= 0;
2027 struct kretprobe_instance
*ri
;
2029 /* TODO: consider to only swap the RA after the last pre_handler fired */
2030 hash
= hash_ptr(current
, KPROBE_HASH_BITS
);
2031 raw_spin_lock_irqsave(&rp
->lock
, flags
);
2032 if (!hlist_empty(&rp
->free_instances
)) {
2033 ri
= hlist_entry(rp
->free_instances
.first
,
2034 struct kretprobe_instance
, hlist
);
2035 hlist_del(&ri
->hlist
);
2036 raw_spin_unlock_irqrestore(&rp
->lock
, flags
);
2041 if (rp
->entry_handler
&& rp
->entry_handler(ri
, regs
)) {
2042 raw_spin_lock_irqsave(&rp
->lock
, flags
);
2043 hlist_add_head(&ri
->hlist
, &rp
->free_instances
);
2044 raw_spin_unlock_irqrestore(&rp
->lock
, flags
);
2048 arch_prepare_kretprobe(ri
, regs
);
2050 /* XXX(hch): why is there no hlist_move_head? */
2051 INIT_HLIST_NODE(&ri
->hlist
);
2052 kretprobe_table_lock(hash
, &flags
);
2053 hlist_add_head(&ri
->hlist
, &kretprobe_inst_table
[hash
]);
2054 kretprobe_table_unlock(hash
, &flags
);
2057 raw_spin_unlock_irqrestore(&rp
->lock
, flags
);
2061 NOKPROBE_SYMBOL(pre_handler_kretprobe
);
2063 bool __weak
arch_kprobe_on_func_entry(unsigned long offset
)
2068 bool kprobe_on_func_entry(kprobe_opcode_t
*addr
, const char *sym
, unsigned long offset
)
2070 kprobe_opcode_t
*kp_addr
= _kprobe_addr(addr
, sym
, offset
);
2072 if (IS_ERR(kp_addr
))
2075 if (!kallsyms_lookup_size_offset((unsigned long)kp_addr
, NULL
, &offset
) ||
2076 !arch_kprobe_on_func_entry(offset
))
2082 int register_kretprobe(struct kretprobe
*rp
)
2085 struct kretprobe_instance
*inst
;
2089 if (!kprobe_on_func_entry(rp
->kp
.addr
, rp
->kp
.symbol_name
, rp
->kp
.offset
))
2092 if (kretprobe_blacklist_size
) {
2093 addr
= kprobe_addr(&rp
->kp
);
2095 return PTR_ERR(addr
);
2097 for (i
= 0; kretprobe_blacklist
[i
].name
!= NULL
; i
++) {
2098 if (kretprobe_blacklist
[i
].addr
== addr
)
2103 rp
->kp
.pre_handler
= pre_handler_kretprobe
;
2104 rp
->kp
.post_handler
= NULL
;
2105 rp
->kp
.fault_handler
= NULL
;
2107 /* Pre-allocate memory for max kretprobe instances */
2108 if (rp
->maxactive
<= 0) {
2109 #ifdef CONFIG_PREEMPTION
2110 rp
->maxactive
= max_t(unsigned int, 10, 2*num_possible_cpus());
2112 rp
->maxactive
= num_possible_cpus();
2115 raw_spin_lock_init(&rp
->lock
);
2116 INIT_HLIST_HEAD(&rp
->free_instances
);
2117 for (i
= 0; i
< rp
->maxactive
; i
++) {
2118 inst
= kmalloc(sizeof(struct kretprobe_instance
) +
2119 rp
->data_size
, GFP_KERNEL
);
2124 INIT_HLIST_NODE(&inst
->hlist
);
2125 hlist_add_head(&inst
->hlist
, &rp
->free_instances
);
2129 /* Establish function entry probe point */
2130 ret
= register_kprobe(&rp
->kp
);
2135 EXPORT_SYMBOL_GPL(register_kretprobe
);
2137 int register_kretprobes(struct kretprobe
**rps
, int num
)
2143 for (i
= 0; i
< num
; i
++) {
2144 ret
= register_kretprobe(rps
[i
]);
2147 unregister_kretprobes(rps
, i
);
2153 EXPORT_SYMBOL_GPL(register_kretprobes
);
2155 void unregister_kretprobe(struct kretprobe
*rp
)
2157 unregister_kretprobes(&rp
, 1);
2159 EXPORT_SYMBOL_GPL(unregister_kretprobe
);
2161 void unregister_kretprobes(struct kretprobe
**rps
, int num
)
2167 mutex_lock(&kprobe_mutex
);
2168 for (i
= 0; i
< num
; i
++)
2169 if (__unregister_kprobe_top(&rps
[i
]->kp
) < 0)
2170 rps
[i
]->kp
.addr
= NULL
;
2171 mutex_unlock(&kprobe_mutex
);
2174 for (i
= 0; i
< num
; i
++) {
2175 if (rps
[i
]->kp
.addr
) {
2176 __unregister_kprobe_bottom(&rps
[i
]->kp
);
2177 cleanup_rp_inst(rps
[i
]);
2181 EXPORT_SYMBOL_GPL(unregister_kretprobes
);
2183 #else /* CONFIG_KRETPROBES */
2184 int register_kretprobe(struct kretprobe
*rp
)
2188 EXPORT_SYMBOL_GPL(register_kretprobe
);
2190 int register_kretprobes(struct kretprobe
**rps
, int num
)
2194 EXPORT_SYMBOL_GPL(register_kretprobes
);
2196 void unregister_kretprobe(struct kretprobe
*rp
)
2199 EXPORT_SYMBOL_GPL(unregister_kretprobe
);
2201 void unregister_kretprobes(struct kretprobe
**rps
, int num
)
2204 EXPORT_SYMBOL_GPL(unregister_kretprobes
);
2206 static int pre_handler_kretprobe(struct kprobe
*p
, struct pt_regs
*regs
)
2210 NOKPROBE_SYMBOL(pre_handler_kretprobe
);
2212 #endif /* CONFIG_KRETPROBES */
2214 /* Set the kprobe gone and remove its instruction buffer. */
2215 static void kill_kprobe(struct kprobe
*p
)
2219 lockdep_assert_held(&kprobe_mutex
);
2221 if (WARN_ON_ONCE(kprobe_gone(p
)))
2224 p
->flags
|= KPROBE_FLAG_GONE
;
2225 if (kprobe_aggrprobe(p
)) {
2227 * If this is an aggr_kprobe, we have to list all the
2228 * chained probes and mark them GONE.
2230 list_for_each_entry(kp
, &p
->list
, list
)
2231 kp
->flags
|= KPROBE_FLAG_GONE
;
2232 p
->post_handler
= NULL
;
2233 kill_optimized_kprobe(p
);
2236 * Here, we can remove insn_slot safely, because no thread calls
2237 * the original probed function (which will be freed soon) any more.
2239 arch_remove_kprobe(p
);
2242 * The module is going away. We should disarm the kprobe which
2243 * is using ftrace, because ftrace framework is still available at
2244 * MODULE_STATE_GOING notification.
2246 if (kprobe_ftrace(p
) && !kprobe_disabled(p
) && !kprobes_all_disarmed
)
2247 disarm_kprobe_ftrace(p
);
2250 /* Disable one kprobe */
2251 int disable_kprobe(struct kprobe
*kp
)
2256 mutex_lock(&kprobe_mutex
);
2258 /* Disable this kprobe */
2259 p
= __disable_kprobe(kp
);
2263 mutex_unlock(&kprobe_mutex
);
2266 EXPORT_SYMBOL_GPL(disable_kprobe
);
2268 /* Enable one kprobe */
2269 int enable_kprobe(struct kprobe
*kp
)
2274 mutex_lock(&kprobe_mutex
);
2276 /* Check whether specified probe is valid. */
2277 p
= __get_valid_kprobe(kp
);
2278 if (unlikely(p
== NULL
)) {
2283 if (kprobe_gone(kp
)) {
2284 /* This kprobe has gone, we couldn't enable it. */
2290 kp
->flags
&= ~KPROBE_FLAG_DISABLED
;
2292 if (!kprobes_all_disarmed
&& kprobe_disabled(p
)) {
2293 p
->flags
&= ~KPROBE_FLAG_DISABLED
;
2294 ret
= arm_kprobe(p
);
2296 p
->flags
|= KPROBE_FLAG_DISABLED
;
2299 mutex_unlock(&kprobe_mutex
);
2302 EXPORT_SYMBOL_GPL(enable_kprobe
);
2304 /* Caller must NOT call this in usual path. This is only for critical case */
2305 void dump_kprobe(struct kprobe
*kp
)
2307 pr_err("Dumping kprobe:\n");
2308 pr_err("Name: %s\nOffset: %x\nAddress: %pS\n",
2309 kp
->symbol_name
, kp
->offset
, kp
->addr
);
2311 NOKPROBE_SYMBOL(dump_kprobe
);
2313 int kprobe_add_ksym_blacklist(unsigned long entry
)
2315 struct kprobe_blacklist_entry
*ent
;
2316 unsigned long offset
= 0, size
= 0;
2318 if (!kernel_text_address(entry
) ||
2319 !kallsyms_lookup_size_offset(entry
, &size
, &offset
))
2322 ent
= kmalloc(sizeof(*ent
), GFP_KERNEL
);
2325 ent
->start_addr
= entry
;
2326 ent
->end_addr
= entry
+ size
;
2327 INIT_LIST_HEAD(&ent
->list
);
2328 list_add_tail(&ent
->list
, &kprobe_blacklist
);
2333 /* Add all symbols in given area into kprobe blacklist */
2334 int kprobe_add_area_blacklist(unsigned long start
, unsigned long end
)
2336 unsigned long entry
;
2339 for (entry
= start
; entry
< end
; entry
+= ret
) {
2340 ret
= kprobe_add_ksym_blacklist(entry
);
2343 if (ret
== 0) /* In case of alias symbol */
2349 /* Remove all symbols in given area from kprobe blacklist */
2350 static void kprobe_remove_area_blacklist(unsigned long start
, unsigned long end
)
2352 struct kprobe_blacklist_entry
*ent
, *n
;
2354 list_for_each_entry_safe(ent
, n
, &kprobe_blacklist
, list
) {
2355 if (ent
->start_addr
< start
|| ent
->start_addr
>= end
)
2357 list_del(&ent
->list
);
2362 static void kprobe_remove_ksym_blacklist(unsigned long entry
)
2364 kprobe_remove_area_blacklist(entry
, entry
+ 1);
2367 int __weak
arch_kprobe_get_kallsym(unsigned int *symnum
, unsigned long *value
,
2368 char *type
, char *sym
)
2373 int kprobe_get_kallsym(unsigned int symnum
, unsigned long *value
, char *type
,
2376 #ifdef __ARCH_WANT_KPROBES_INSN_SLOT
2377 if (!kprobe_cache_get_kallsym(&kprobe_insn_slots
, &symnum
, value
, type
, sym
))
2379 #ifdef CONFIG_OPTPROBES
2380 if (!kprobe_cache_get_kallsym(&kprobe_optinsn_slots
, &symnum
, value
, type
, sym
))
2384 if (!arch_kprobe_get_kallsym(&symnum
, value
, type
, sym
))
2389 int __init __weak
arch_populate_kprobe_blacklist(void)
2395 * Lookup and populate the kprobe_blacklist.
2397 * Unlike the kretprobe blacklist, we'll need to determine
2398 * the range of addresses that belong to the said functions,
2399 * since a kprobe need not necessarily be at the beginning
2402 static int __init
populate_kprobe_blacklist(unsigned long *start
,
2405 unsigned long entry
;
2406 unsigned long *iter
;
2409 for (iter
= start
; iter
< end
; iter
++) {
2410 entry
= arch_deref_entry_point((void *)*iter
);
2411 ret
= kprobe_add_ksym_blacklist(entry
);
2418 /* Symbols in __kprobes_text are blacklisted */
2419 ret
= kprobe_add_area_blacklist((unsigned long)__kprobes_text_start
,
2420 (unsigned long)__kprobes_text_end
);
2424 /* Symbols in noinstr section are blacklisted */
2425 ret
= kprobe_add_area_blacklist((unsigned long)__noinstr_text_start
,
2426 (unsigned long)__noinstr_text_end
);
2428 return ret
? : arch_populate_kprobe_blacklist();
2431 static void add_module_kprobe_blacklist(struct module
*mod
)
2433 unsigned long start
, end
;
2436 if (mod
->kprobe_blacklist
) {
2437 for (i
= 0; i
< mod
->num_kprobe_blacklist
; i
++)
2438 kprobe_add_ksym_blacklist(mod
->kprobe_blacklist
[i
]);
2441 start
= (unsigned long)mod
->kprobes_text_start
;
2443 end
= start
+ mod
->kprobes_text_size
;
2444 kprobe_add_area_blacklist(start
, end
);
2447 start
= (unsigned long)mod
->noinstr_text_start
;
2449 end
= start
+ mod
->noinstr_text_size
;
2450 kprobe_add_area_blacklist(start
, end
);
2454 static void remove_module_kprobe_blacklist(struct module
*mod
)
2456 unsigned long start
, end
;
2459 if (mod
->kprobe_blacklist
) {
2460 for (i
= 0; i
< mod
->num_kprobe_blacklist
; i
++)
2461 kprobe_remove_ksym_blacklist(mod
->kprobe_blacklist
[i
]);
2464 start
= (unsigned long)mod
->kprobes_text_start
;
2466 end
= start
+ mod
->kprobes_text_size
;
2467 kprobe_remove_area_blacklist(start
, end
);
2470 start
= (unsigned long)mod
->noinstr_text_start
;
2472 end
= start
+ mod
->noinstr_text_size
;
2473 kprobe_remove_area_blacklist(start
, end
);
2477 /* Module notifier call back, checking kprobes on the module */
2478 static int kprobes_module_callback(struct notifier_block
*nb
,
2479 unsigned long val
, void *data
)
2481 struct module
*mod
= data
;
2482 struct hlist_head
*head
;
2485 int checkcore
= (val
== MODULE_STATE_GOING
);
2487 if (val
== MODULE_STATE_COMING
) {
2488 mutex_lock(&kprobe_mutex
);
2489 add_module_kprobe_blacklist(mod
);
2490 mutex_unlock(&kprobe_mutex
);
2492 if (val
!= MODULE_STATE_GOING
&& val
!= MODULE_STATE_LIVE
)
2496 * When MODULE_STATE_GOING was notified, both of module .text and
2497 * .init.text sections would be freed. When MODULE_STATE_LIVE was
2498 * notified, only .init.text section would be freed. We need to
2499 * disable kprobes which have been inserted in the sections.
2501 mutex_lock(&kprobe_mutex
);
2502 for (i
= 0; i
< KPROBE_TABLE_SIZE
; i
++) {
2503 head
= &kprobe_table
[i
];
2504 hlist_for_each_entry(p
, head
, hlist
) {
2508 if (within_module_init((unsigned long)p
->addr
, mod
) ||
2510 within_module_core((unsigned long)p
->addr
, mod
))) {
2512 * The vaddr this probe is installed will soon
2513 * be vfreed buy not synced to disk. Hence,
2514 * disarming the breakpoint isn't needed.
2516 * Note, this will also move any optimized probes
2517 * that are pending to be removed from their
2518 * corresponding lists to the freeing_list and
2519 * will not be touched by the delayed
2520 * kprobe_optimizer work handler.
2526 if (val
== MODULE_STATE_GOING
)
2527 remove_module_kprobe_blacklist(mod
);
2528 mutex_unlock(&kprobe_mutex
);
2532 static struct notifier_block kprobe_module_nb
= {
2533 .notifier_call
= kprobes_module_callback
,
2537 /* Markers of _kprobe_blacklist section */
2538 extern unsigned long __start_kprobe_blacklist
[];
2539 extern unsigned long __stop_kprobe_blacklist
[];
2541 void kprobe_free_init_mem(void)
2543 void *start
= (void *)(&__init_begin
);
2544 void *end
= (void *)(&__init_end
);
2545 struct hlist_head
*head
;
2549 mutex_lock(&kprobe_mutex
);
2551 /* Kill all kprobes on initmem */
2552 for (i
= 0; i
< KPROBE_TABLE_SIZE
; i
++) {
2553 head
= &kprobe_table
[i
];
2554 hlist_for_each_entry(p
, head
, hlist
) {
2555 if (start
<= (void *)p
->addr
&& (void *)p
->addr
< end
)
2560 mutex_unlock(&kprobe_mutex
);
2563 static int __init
init_kprobes(void)
2567 /* FIXME allocate the probe table, currently defined statically */
2568 /* initialize all list heads */
2569 for (i
= 0; i
< KPROBE_TABLE_SIZE
; i
++) {
2570 INIT_HLIST_HEAD(&kprobe_table
[i
]);
2571 INIT_HLIST_HEAD(&kretprobe_inst_table
[i
]);
2572 raw_spin_lock_init(&(kretprobe_table_locks
[i
].lock
));
2575 err
= populate_kprobe_blacklist(__start_kprobe_blacklist
,
2576 __stop_kprobe_blacklist
);
2578 pr_err("kprobes: failed to populate blacklist: %d\n", err
);
2579 pr_err("Please take care of using kprobes.\n");
2582 if (kretprobe_blacklist_size
) {
2583 /* lookup the function address from its name */
2584 for (i
= 0; kretprobe_blacklist
[i
].name
!= NULL
; i
++) {
2585 kretprobe_blacklist
[i
].addr
=
2586 kprobe_lookup_name(kretprobe_blacklist
[i
].name
, 0);
2587 if (!kretprobe_blacklist
[i
].addr
)
2588 printk("kretprobe: lookup failed: %s\n",
2589 kretprobe_blacklist
[i
].name
);
2593 #if defined(CONFIG_OPTPROBES)
2594 #if defined(__ARCH_WANT_KPROBES_INSN_SLOT)
2595 /* Init kprobe_optinsn_slots */
2596 kprobe_optinsn_slots
.insn_size
= MAX_OPTINSN_SIZE
;
2598 /* By default, kprobes can be optimized */
2599 kprobes_allow_optimization
= true;
2602 /* By default, kprobes are armed */
2603 kprobes_all_disarmed
= false;
2605 err
= arch_init_kprobes();
2607 err
= register_die_notifier(&kprobe_exceptions_nb
);
2609 err
= register_module_notifier(&kprobe_module_nb
);
2611 kprobes_initialized
= (err
== 0);
2617 subsys_initcall(init_kprobes
);
2619 #ifdef CONFIG_DEBUG_FS
2620 static void report_probe(struct seq_file
*pi
, struct kprobe
*p
,
2621 const char *sym
, int offset
, char *modname
, struct kprobe
*pp
)
2624 void *addr
= p
->addr
;
2626 if (p
->pre_handler
== pre_handler_kretprobe
)
2631 if (!kallsyms_show_value(pi
->file
->f_cred
))
2635 seq_printf(pi
, "%px %s %s+0x%x %s ",
2636 addr
, kprobe_type
, sym
, offset
,
2637 (modname
? modname
: " "));
2638 else /* try to use %pS */
2639 seq_printf(pi
, "%px %s %pS ",
2640 addr
, kprobe_type
, p
->addr
);
2644 seq_printf(pi
, "%s%s%s%s\n",
2645 (kprobe_gone(p
) ? "[GONE]" : ""),
2646 ((kprobe_disabled(p
) && !kprobe_gone(p
)) ? "[DISABLED]" : ""),
2647 (kprobe_optimized(pp
) ? "[OPTIMIZED]" : ""),
2648 (kprobe_ftrace(pp
) ? "[FTRACE]" : ""));
2651 static void *kprobe_seq_start(struct seq_file
*f
, loff_t
*pos
)
2653 return (*pos
< KPROBE_TABLE_SIZE
) ? pos
: NULL
;
2656 static void *kprobe_seq_next(struct seq_file
*f
, void *v
, loff_t
*pos
)
2659 if (*pos
>= KPROBE_TABLE_SIZE
)
2664 static void kprobe_seq_stop(struct seq_file
*f
, void *v
)
2669 static int show_kprobe_addr(struct seq_file
*pi
, void *v
)
2671 struct hlist_head
*head
;
2672 struct kprobe
*p
, *kp
;
2673 const char *sym
= NULL
;
2674 unsigned int i
= *(loff_t
*) v
;
2675 unsigned long offset
= 0;
2676 char *modname
, namebuf
[KSYM_NAME_LEN
];
2678 head
= &kprobe_table
[i
];
2680 hlist_for_each_entry_rcu(p
, head
, hlist
) {
2681 sym
= kallsyms_lookup((unsigned long)p
->addr
, NULL
,
2682 &offset
, &modname
, namebuf
);
2683 if (kprobe_aggrprobe(p
)) {
2684 list_for_each_entry_rcu(kp
, &p
->list
, list
)
2685 report_probe(pi
, kp
, sym
, offset
, modname
, p
);
2687 report_probe(pi
, p
, sym
, offset
, modname
, NULL
);
2693 static const struct seq_operations kprobes_sops
= {
2694 .start
= kprobe_seq_start
,
2695 .next
= kprobe_seq_next
,
2696 .stop
= kprobe_seq_stop
,
2697 .show
= show_kprobe_addr
2700 DEFINE_SEQ_ATTRIBUTE(kprobes
);
2702 /* kprobes/blacklist -- shows which functions can not be probed */
2703 static void *kprobe_blacklist_seq_start(struct seq_file
*m
, loff_t
*pos
)
2705 mutex_lock(&kprobe_mutex
);
2706 return seq_list_start(&kprobe_blacklist
, *pos
);
2709 static void *kprobe_blacklist_seq_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
2711 return seq_list_next(v
, &kprobe_blacklist
, pos
);
2714 static int kprobe_blacklist_seq_show(struct seq_file
*m
, void *v
)
2716 struct kprobe_blacklist_entry
*ent
=
2717 list_entry(v
, struct kprobe_blacklist_entry
, list
);
2720 * If /proc/kallsyms is not showing kernel address, we won't
2721 * show them here either.
2723 if (!kallsyms_show_value(m
->file
->f_cred
))
2724 seq_printf(m
, "0x%px-0x%px\t%ps\n", NULL
, NULL
,
2725 (void *)ent
->start_addr
);
2727 seq_printf(m
, "0x%px-0x%px\t%ps\n", (void *)ent
->start_addr
,
2728 (void *)ent
->end_addr
, (void *)ent
->start_addr
);
2732 static void kprobe_blacklist_seq_stop(struct seq_file
*f
, void *v
)
2734 mutex_unlock(&kprobe_mutex
);
2737 static const struct seq_operations kprobe_blacklist_sops
= {
2738 .start
= kprobe_blacklist_seq_start
,
2739 .next
= kprobe_blacklist_seq_next
,
2740 .stop
= kprobe_blacklist_seq_stop
,
2741 .show
= kprobe_blacklist_seq_show
,
2743 DEFINE_SEQ_ATTRIBUTE(kprobe_blacklist
);
2745 static int arm_all_kprobes(void)
2747 struct hlist_head
*head
;
2749 unsigned int i
, total
= 0, errors
= 0;
2752 mutex_lock(&kprobe_mutex
);
2754 /* If kprobes are armed, just return */
2755 if (!kprobes_all_disarmed
)
2756 goto already_enabled
;
2759 * optimize_kprobe() called by arm_kprobe() checks
2760 * kprobes_all_disarmed, so set kprobes_all_disarmed before
2763 kprobes_all_disarmed
= false;
2764 /* Arming kprobes doesn't optimize kprobe itself */
2765 for (i
= 0; i
< KPROBE_TABLE_SIZE
; i
++) {
2766 head
= &kprobe_table
[i
];
2767 /* Arm all kprobes on a best-effort basis */
2768 hlist_for_each_entry(p
, head
, hlist
) {
2769 if (!kprobe_disabled(p
)) {
2770 err
= arm_kprobe(p
);
2781 pr_warn("Kprobes globally enabled, but failed to arm %d out of %d probes\n",
2784 pr_info("Kprobes globally enabled\n");
2787 mutex_unlock(&kprobe_mutex
);
2791 static int disarm_all_kprobes(void)
2793 struct hlist_head
*head
;
2795 unsigned int i
, total
= 0, errors
= 0;
2798 mutex_lock(&kprobe_mutex
);
2800 /* If kprobes are already disarmed, just return */
2801 if (kprobes_all_disarmed
) {
2802 mutex_unlock(&kprobe_mutex
);
2806 kprobes_all_disarmed
= true;
2808 for (i
= 0; i
< KPROBE_TABLE_SIZE
; i
++) {
2809 head
= &kprobe_table
[i
];
2810 /* Disarm all kprobes on a best-effort basis */
2811 hlist_for_each_entry(p
, head
, hlist
) {
2812 if (!arch_trampoline_kprobe(p
) && !kprobe_disabled(p
)) {
2813 err
= disarm_kprobe(p
, false);
2824 pr_warn("Kprobes globally disabled, but failed to disarm %d out of %d probes\n",
2827 pr_info("Kprobes globally disabled\n");
2829 mutex_unlock(&kprobe_mutex
);
2831 /* Wait for disarming all kprobes by optimizer */
2832 wait_for_kprobe_optimizer();
2838 * XXX: The debugfs bool file interface doesn't allow for callbacks
2839 * when the bool state is switched. We can reuse that facility when
2842 static ssize_t
read_enabled_file_bool(struct file
*file
,
2843 char __user
*user_buf
, size_t count
, loff_t
*ppos
)
2847 if (!kprobes_all_disarmed
)
2853 return simple_read_from_buffer(user_buf
, count
, ppos
, buf
, 2);
2856 static ssize_t
write_enabled_file_bool(struct file
*file
,
2857 const char __user
*user_buf
, size_t count
, loff_t
*ppos
)
2863 buf_size
= min(count
, (sizeof(buf
)-1));
2864 if (copy_from_user(buf
, user_buf
, buf_size
))
2867 buf
[buf_size
] = '\0';
2872 ret
= arm_all_kprobes();
2877 ret
= disarm_all_kprobes();
2889 static const struct file_operations fops_kp
= {
2890 .read
= read_enabled_file_bool
,
2891 .write
= write_enabled_file_bool
,
2892 .llseek
= default_llseek
,
2895 static int __init
debugfs_kprobe_init(void)
2898 unsigned int value
= 1;
2900 dir
= debugfs_create_dir("kprobes", NULL
);
2902 debugfs_create_file("list", 0400, dir
, NULL
, &kprobes_fops
);
2904 debugfs_create_file("enabled", 0600, dir
, &value
, &fops_kp
);
2906 debugfs_create_file("blacklist", 0400, dir
, NULL
,
2907 &kprobe_blacklist_fops
);
2912 late_initcall(debugfs_kprobe_init
);
2913 #endif /* CONFIG_DEBUG_FS */