1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * Kernel Probes (KProbes)
5 * Copyright (C) IBM Corporation, 2002, 2004
7 * 2002-Oct Created by Vamsi Krishna S <vamsi_krishna@in.ibm.com> Kernel
8 * Probes initial implementation (includes suggestions from
10 * 2004-Aug Updated by Prasanna S Panchamukhi <prasanna@in.ibm.com> with
11 * hlists and exceptions notifier as suggested by Andi Kleen.
12 * 2004-July Suparna Bhattacharya <suparna@in.ibm.com> added jumper probes
13 * interface to access function arguments.
14 * 2004-Sep Prasanna S Panchamukhi <prasanna@in.ibm.com> Changed Kprobes
15 * exceptions notifier to be first on the priority list.
16 * 2005-May Hien Nguyen <hien@us.ibm.com>, Jim Keniston
17 * <jkenisto@us.ibm.com> and Prasanna S Panchamukhi
18 * <prasanna@in.ibm.com> added function-return probes.
21 #define pr_fmt(fmt) "kprobes: " fmt
23 #include <linux/kprobes.h>
24 #include <linux/hash.h>
25 #include <linux/init.h>
26 #include <linux/slab.h>
27 #include <linux/stddef.h>
28 #include <linux/export.h>
29 #include <linux/moduleloader.h>
30 #include <linux/kallsyms.h>
31 #include <linux/freezer.h>
32 #include <linux/seq_file.h>
33 #include <linux/debugfs.h>
34 #include <linux/sysctl.h>
35 #include <linux/kdebug.h>
36 #include <linux/memory.h>
37 #include <linux/ftrace.h>
38 #include <linux/cpu.h>
39 #include <linux/jump_label.h>
40 #include <linux/static_call.h>
41 #include <linux/perf_event.h>
43 #include <asm/sections.h>
44 #include <asm/cacheflush.h>
45 #include <asm/errno.h>
46 #include <linux/uaccess.h>
48 #define KPROBE_HASH_BITS 6
49 #define KPROBE_TABLE_SIZE (1 << KPROBE_HASH_BITS)
51 #if !defined(CONFIG_OPTPROBES) || !defined(CONFIG_SYSCTL)
52 #define kprobe_sysctls_init() do { } while (0)
55 static int kprobes_initialized
;
56 /* kprobe_table can be accessed by
57 * - Normal hlist traversal and RCU add/del under 'kprobe_mutex' is held.
59 * - RCU hlist traversal under disabling preempt (breakpoint handlers)
61 static struct hlist_head kprobe_table
[KPROBE_TABLE_SIZE
];
63 /* NOTE: change this value only with 'kprobe_mutex' held */
64 static bool kprobes_all_disarmed
;
66 /* This protects 'kprobe_table' and 'optimizing_list' */
67 static DEFINE_MUTEX(kprobe_mutex
);
68 static DEFINE_PER_CPU(struct kprobe
*, kprobe_instance
);
70 kprobe_opcode_t
* __weak
kprobe_lookup_name(const char *name
,
71 unsigned int __unused
)
73 return ((kprobe_opcode_t
*)(kallsyms_lookup_name(name
)));
77 * Blacklist -- list of 'struct kprobe_blacklist_entry' to store info where
78 * kprobes can not probe.
80 static LIST_HEAD(kprobe_blacklist
);
82 #ifdef __ARCH_WANT_KPROBES_INSN_SLOT
84 * 'kprobe::ainsn.insn' points to the copy of the instruction to be
85 * single-stepped. x86_64, POWER4 and above have no-exec support and
86 * stepping on the instruction on a vmalloced/kmalloced/data page
87 * is a recipe for disaster
89 struct kprobe_insn_page
{
90 struct list_head list
;
91 kprobe_opcode_t
*insns
; /* Page of instruction slots */
92 struct kprobe_insn_cache
*cache
;
98 #define KPROBE_INSN_PAGE_SIZE(slots) \
99 (offsetof(struct kprobe_insn_page, slot_used) + \
100 (sizeof(char) * (slots)))
102 static int slots_per_page(struct kprobe_insn_cache
*c
)
104 return PAGE_SIZE
/(c
->insn_size
* sizeof(kprobe_opcode_t
));
107 enum kprobe_slot_state
{
113 void __weak
*alloc_insn_page(void)
116 * Use module_alloc() so this page is within +/- 2GB of where the
117 * kernel image and loaded module images reside. This is required
118 * for most of the architectures.
119 * (e.g. x86-64 needs this to handle the %rip-relative fixups.)
121 return module_alloc(PAGE_SIZE
);
124 static void free_insn_page(void *page
)
126 module_memfree(page
);
129 struct kprobe_insn_cache kprobe_insn_slots
= {
130 .mutex
= __MUTEX_INITIALIZER(kprobe_insn_slots
.mutex
),
131 .alloc
= alloc_insn_page
,
132 .free
= free_insn_page
,
133 .sym
= KPROBE_INSN_PAGE_SYM
,
134 .pages
= LIST_HEAD_INIT(kprobe_insn_slots
.pages
),
135 .insn_size
= MAX_INSN_SIZE
,
138 static int collect_garbage_slots(struct kprobe_insn_cache
*c
);
141 * __get_insn_slot() - Find a slot on an executable page for an instruction.
142 * We allocate an executable page if there's no room on existing ones.
144 kprobe_opcode_t
*__get_insn_slot(struct kprobe_insn_cache
*c
)
146 struct kprobe_insn_page
*kip
;
147 kprobe_opcode_t
*slot
= NULL
;
149 /* Since the slot array is not protected by rcu, we need a mutex */
150 mutex_lock(&c
->mutex
);
153 list_for_each_entry_rcu(kip
, &c
->pages
, list
) {
154 if (kip
->nused
< slots_per_page(c
)) {
157 for (i
= 0; i
< slots_per_page(c
); i
++) {
158 if (kip
->slot_used
[i
] == SLOT_CLEAN
) {
159 kip
->slot_used
[i
] = SLOT_USED
;
161 slot
= kip
->insns
+ (i
* c
->insn_size
);
166 /* kip->nused is broken. Fix it. */
167 kip
->nused
= slots_per_page(c
);
173 /* If there are any garbage slots, collect it and try again. */
174 if (c
->nr_garbage
&& collect_garbage_slots(c
) == 0)
177 /* All out of space. Need to allocate a new page. */
178 kip
= kmalloc(KPROBE_INSN_PAGE_SIZE(slots_per_page(c
)), GFP_KERNEL
);
182 kip
->insns
= c
->alloc();
187 INIT_LIST_HEAD(&kip
->list
);
188 memset(kip
->slot_used
, SLOT_CLEAN
, slots_per_page(c
));
189 kip
->slot_used
[0] = SLOT_USED
;
193 list_add_rcu(&kip
->list
, &c
->pages
);
196 /* Record the perf ksymbol register event after adding the page */
197 perf_event_ksymbol(PERF_RECORD_KSYMBOL_TYPE_OOL
, (unsigned long)kip
->insns
,
198 PAGE_SIZE
, false, c
->sym
);
200 mutex_unlock(&c
->mutex
);
204 /* Return true if all garbages are collected, otherwise false. */
205 static bool collect_one_slot(struct kprobe_insn_page
*kip
, int idx
)
207 kip
->slot_used
[idx
] = SLOT_CLEAN
;
209 if (kip
->nused
== 0) {
211 * Page is no longer in use. Free it unless
212 * it's the last one. We keep the last one
213 * so as not to have to set it up again the
214 * next time somebody inserts a probe.
216 if (!list_is_singular(&kip
->list
)) {
218 * Record perf ksymbol unregister event before removing
221 perf_event_ksymbol(PERF_RECORD_KSYMBOL_TYPE_OOL
,
222 (unsigned long)kip
->insns
, PAGE_SIZE
, true,
224 list_del_rcu(&kip
->list
);
226 kip
->cache
->free(kip
->insns
);
234 static int collect_garbage_slots(struct kprobe_insn_cache
*c
)
236 struct kprobe_insn_page
*kip
, *next
;
238 /* Ensure no-one is interrupted on the garbages */
241 list_for_each_entry_safe(kip
, next
, &c
->pages
, list
) {
244 if (kip
->ngarbage
== 0)
246 kip
->ngarbage
= 0; /* we will collect all garbages */
247 for (i
= 0; i
< slots_per_page(c
); i
++) {
248 if (kip
->slot_used
[i
] == SLOT_DIRTY
&& collect_one_slot(kip
, i
))
256 void __free_insn_slot(struct kprobe_insn_cache
*c
,
257 kprobe_opcode_t
*slot
, int dirty
)
259 struct kprobe_insn_page
*kip
;
262 mutex_lock(&c
->mutex
);
264 list_for_each_entry_rcu(kip
, &c
->pages
, list
) {
265 idx
= ((long)slot
- (long)kip
->insns
) /
266 (c
->insn_size
* sizeof(kprobe_opcode_t
));
267 if (idx
>= 0 && idx
< slots_per_page(c
))
270 /* Could not find this slot. */
275 /* Mark and sweep: this may sleep */
277 /* Check double free */
278 WARN_ON(kip
->slot_used
[idx
] != SLOT_USED
);
280 kip
->slot_used
[idx
] = SLOT_DIRTY
;
282 if (++c
->nr_garbage
> slots_per_page(c
))
283 collect_garbage_slots(c
);
285 collect_one_slot(kip
, idx
);
288 mutex_unlock(&c
->mutex
);
292 * Check given address is on the page of kprobe instruction slots.
293 * This will be used for checking whether the address on a stack
294 * is on a text area or not.
296 bool __is_insn_slot_addr(struct kprobe_insn_cache
*c
, unsigned long addr
)
298 struct kprobe_insn_page
*kip
;
302 list_for_each_entry_rcu(kip
, &c
->pages
, list
) {
303 if (addr
>= (unsigned long)kip
->insns
&&
304 addr
< (unsigned long)kip
->insns
+ PAGE_SIZE
) {
314 int kprobe_cache_get_kallsym(struct kprobe_insn_cache
*c
, unsigned int *symnum
,
315 unsigned long *value
, char *type
, char *sym
)
317 struct kprobe_insn_page
*kip
;
321 list_for_each_entry_rcu(kip
, &c
->pages
, list
) {
324 strscpy(sym
, c
->sym
, KSYM_NAME_LEN
);
326 *value
= (unsigned long)kip
->insns
;
335 #ifdef CONFIG_OPTPROBES
336 void __weak
*alloc_optinsn_page(void)
338 return alloc_insn_page();
341 void __weak
free_optinsn_page(void *page
)
343 free_insn_page(page
);
346 /* For optimized_kprobe buffer */
347 struct kprobe_insn_cache kprobe_optinsn_slots
= {
348 .mutex
= __MUTEX_INITIALIZER(kprobe_optinsn_slots
.mutex
),
349 .alloc
= alloc_optinsn_page
,
350 .free
= free_optinsn_page
,
351 .sym
= KPROBE_OPTINSN_PAGE_SYM
,
352 .pages
= LIST_HEAD_INIT(kprobe_optinsn_slots
.pages
),
353 /* .insn_size is initialized later */
359 /* We have preemption disabled.. so it is safe to use __ versions */
360 static inline void set_kprobe_instance(struct kprobe
*kp
)
362 __this_cpu_write(kprobe_instance
, kp
);
365 static inline void reset_kprobe_instance(void)
367 __this_cpu_write(kprobe_instance
, NULL
);
371 * This routine is called either:
372 * - under the 'kprobe_mutex' - during kprobe_[un]register().
374 * - with preemption disabled - from architecture specific code.
376 struct kprobe
*get_kprobe(void *addr
)
378 struct hlist_head
*head
;
381 head
= &kprobe_table
[hash_ptr(addr
, KPROBE_HASH_BITS
)];
382 hlist_for_each_entry_rcu(p
, head
, hlist
,
383 lockdep_is_held(&kprobe_mutex
)) {
390 NOKPROBE_SYMBOL(get_kprobe
);
392 static int aggr_pre_handler(struct kprobe
*p
, struct pt_regs
*regs
);
394 /* Return true if 'p' is an aggregator */
395 static inline bool kprobe_aggrprobe(struct kprobe
*p
)
397 return p
->pre_handler
== aggr_pre_handler
;
400 /* Return true if 'p' is unused */
401 static inline bool kprobe_unused(struct kprobe
*p
)
403 return kprobe_aggrprobe(p
) && kprobe_disabled(p
) &&
404 list_empty(&p
->list
);
407 /* Keep all fields in the kprobe consistent. */
408 static inline void copy_kprobe(struct kprobe
*ap
, struct kprobe
*p
)
410 memcpy(&p
->opcode
, &ap
->opcode
, sizeof(kprobe_opcode_t
));
411 memcpy(&p
->ainsn
, &ap
->ainsn
, sizeof(struct arch_specific_insn
));
414 #ifdef CONFIG_OPTPROBES
415 /* NOTE: This is protected by 'kprobe_mutex'. */
416 static bool kprobes_allow_optimization
;
419 * Call all 'kprobe::pre_handler' on the list, but ignores its return value.
420 * This must be called from arch-dep optimized caller.
422 void opt_pre_handler(struct kprobe
*p
, struct pt_regs
*regs
)
426 list_for_each_entry_rcu(kp
, &p
->list
, list
) {
427 if (kp
->pre_handler
&& likely(!kprobe_disabled(kp
))) {
428 set_kprobe_instance(kp
);
429 kp
->pre_handler(kp
, regs
);
431 reset_kprobe_instance();
434 NOKPROBE_SYMBOL(opt_pre_handler
);
436 /* Free optimized instructions and optimized_kprobe */
437 static void free_aggr_kprobe(struct kprobe
*p
)
439 struct optimized_kprobe
*op
;
441 op
= container_of(p
, struct optimized_kprobe
, kp
);
442 arch_remove_optimized_kprobe(op
);
443 arch_remove_kprobe(p
);
447 /* Return true if the kprobe is ready for optimization. */
448 static inline int kprobe_optready(struct kprobe
*p
)
450 struct optimized_kprobe
*op
;
452 if (kprobe_aggrprobe(p
)) {
453 op
= container_of(p
, struct optimized_kprobe
, kp
);
454 return arch_prepared_optinsn(&op
->optinsn
);
460 /* Return true if the kprobe is disarmed. Note: p must be on hash list */
461 static inline bool kprobe_disarmed(struct kprobe
*p
)
463 struct optimized_kprobe
*op
;
465 /* If kprobe is not aggr/opt probe, just return kprobe is disabled */
466 if (!kprobe_aggrprobe(p
))
467 return kprobe_disabled(p
);
469 op
= container_of(p
, struct optimized_kprobe
, kp
);
471 return kprobe_disabled(p
) && list_empty(&op
->list
);
474 /* Return true if the probe is queued on (un)optimizing lists */
475 static bool kprobe_queued(struct kprobe
*p
)
477 struct optimized_kprobe
*op
;
479 if (kprobe_aggrprobe(p
)) {
480 op
= container_of(p
, struct optimized_kprobe
, kp
);
481 if (!list_empty(&op
->list
))
488 * Return an optimized kprobe whose optimizing code replaces
489 * instructions including 'addr' (exclude breakpoint).
491 static struct kprobe
*get_optimized_kprobe(kprobe_opcode_t
*addr
)
494 struct kprobe
*p
= NULL
;
495 struct optimized_kprobe
*op
;
497 /* Don't check i == 0, since that is a breakpoint case. */
498 for (i
= 1; !p
&& i
< MAX_OPTIMIZED_LENGTH
/ sizeof(kprobe_opcode_t
); i
++)
499 p
= get_kprobe(addr
- i
);
501 if (p
&& kprobe_optready(p
)) {
502 op
= container_of(p
, struct optimized_kprobe
, kp
);
503 if (arch_within_optimized_kprobe(op
, addr
))
510 /* Optimization staging list, protected by 'kprobe_mutex' */
511 static LIST_HEAD(optimizing_list
);
512 static LIST_HEAD(unoptimizing_list
);
513 static LIST_HEAD(freeing_list
);
515 static void kprobe_optimizer(struct work_struct
*work
);
516 static DECLARE_DELAYED_WORK(optimizing_work
, kprobe_optimizer
);
517 #define OPTIMIZE_DELAY 5
520 * Optimize (replace a breakpoint with a jump) kprobes listed on
523 static void do_optimize_kprobes(void)
525 lockdep_assert_held(&text_mutex
);
527 * The optimization/unoptimization refers 'online_cpus' via
528 * stop_machine() and cpu-hotplug modifies the 'online_cpus'.
529 * And same time, 'text_mutex' will be held in cpu-hotplug and here.
530 * This combination can cause a deadlock (cpu-hotplug tries to lock
531 * 'text_mutex' but stop_machine() can not be done because
532 * the 'online_cpus' has been changed)
533 * To avoid this deadlock, caller must have locked cpu-hotplug
534 * for preventing cpu-hotplug outside of 'text_mutex' locking.
536 lockdep_assert_cpus_held();
538 /* Optimization never be done when disarmed */
539 if (kprobes_all_disarmed
|| !kprobes_allow_optimization
||
540 list_empty(&optimizing_list
))
543 arch_optimize_kprobes(&optimizing_list
);
547 * Unoptimize (replace a jump with a breakpoint and remove the breakpoint
548 * if need) kprobes listed on 'unoptimizing_list'.
550 static void do_unoptimize_kprobes(void)
552 struct optimized_kprobe
*op
, *tmp
;
554 lockdep_assert_held(&text_mutex
);
555 /* See comment in do_optimize_kprobes() */
556 lockdep_assert_cpus_held();
558 /* Unoptimization must be done anytime */
559 if (list_empty(&unoptimizing_list
))
562 arch_unoptimize_kprobes(&unoptimizing_list
, &freeing_list
);
563 /* Loop on 'freeing_list' for disarming */
564 list_for_each_entry_safe(op
, tmp
, &freeing_list
, list
) {
565 /* Switching from detour code to origin */
566 op
->kp
.flags
&= ~KPROBE_FLAG_OPTIMIZED
;
567 /* Disarm probes if marked disabled */
568 if (kprobe_disabled(&op
->kp
))
569 arch_disarm_kprobe(&op
->kp
);
570 if (kprobe_unused(&op
->kp
)) {
572 * Remove unused probes from hash list. After waiting
573 * for synchronization, these probes are reclaimed.
574 * (reclaiming is done by do_free_cleaned_kprobes().)
576 hlist_del_rcu(&op
->kp
.hlist
);
578 list_del_init(&op
->list
);
582 /* Reclaim all kprobes on the 'freeing_list' */
583 static void do_free_cleaned_kprobes(void)
585 struct optimized_kprobe
*op
, *tmp
;
587 list_for_each_entry_safe(op
, tmp
, &freeing_list
, list
) {
588 list_del_init(&op
->list
);
589 if (WARN_ON_ONCE(!kprobe_unused(&op
->kp
))) {
591 * This must not happen, but if there is a kprobe
592 * still in use, keep it on kprobes hash list.
596 free_aggr_kprobe(&op
->kp
);
600 /* Start optimizer after OPTIMIZE_DELAY passed */
601 static void kick_kprobe_optimizer(void)
603 schedule_delayed_work(&optimizing_work
, OPTIMIZE_DELAY
);
606 /* Kprobe jump optimizer */
607 static void kprobe_optimizer(struct work_struct
*work
)
609 mutex_lock(&kprobe_mutex
);
611 mutex_lock(&text_mutex
);
614 * Step 1: Unoptimize kprobes and collect cleaned (unused and disarmed)
615 * kprobes before waiting for quiesence period.
617 do_unoptimize_kprobes();
620 * Step 2: Wait for quiesence period to ensure all potentially
621 * preempted tasks to have normally scheduled. Because optprobe
622 * may modify multiple instructions, there is a chance that Nth
623 * instruction is preempted. In that case, such tasks can return
624 * to 2nd-Nth byte of jump instruction. This wait is for avoiding it.
625 * Note that on non-preemptive kernel, this is transparently converted
626 * to synchronoze_sched() to wait for all interrupts to have completed.
628 synchronize_rcu_tasks();
630 /* Step 3: Optimize kprobes after quiesence period */
631 do_optimize_kprobes();
633 /* Step 4: Free cleaned kprobes after quiesence period */
634 do_free_cleaned_kprobes();
636 mutex_unlock(&text_mutex
);
639 /* Step 5: Kick optimizer again if needed */
640 if (!list_empty(&optimizing_list
) || !list_empty(&unoptimizing_list
))
641 kick_kprobe_optimizer();
643 mutex_unlock(&kprobe_mutex
);
646 /* Wait for completing optimization and unoptimization */
647 void wait_for_kprobe_optimizer(void)
649 mutex_lock(&kprobe_mutex
);
651 while (!list_empty(&optimizing_list
) || !list_empty(&unoptimizing_list
)) {
652 mutex_unlock(&kprobe_mutex
);
654 /* This will also make 'optimizing_work' execute immmediately */
655 flush_delayed_work(&optimizing_work
);
656 /* 'optimizing_work' might not have been queued yet, relax */
659 mutex_lock(&kprobe_mutex
);
662 mutex_unlock(&kprobe_mutex
);
665 static bool optprobe_queued_unopt(struct optimized_kprobe
*op
)
667 struct optimized_kprobe
*_op
;
669 list_for_each_entry(_op
, &unoptimizing_list
, list
) {
677 /* Optimize kprobe if p is ready to be optimized */
678 static void optimize_kprobe(struct kprobe
*p
)
680 struct optimized_kprobe
*op
;
682 /* Check if the kprobe is disabled or not ready for optimization. */
683 if (!kprobe_optready(p
) || !kprobes_allow_optimization
||
684 (kprobe_disabled(p
) || kprobes_all_disarmed
))
687 /* kprobes with 'post_handler' can not be optimized */
691 op
= container_of(p
, struct optimized_kprobe
, kp
);
693 /* Check there is no other kprobes at the optimized instructions */
694 if (arch_check_optimized_kprobe(op
) < 0)
697 /* Check if it is already optimized. */
698 if (op
->kp
.flags
& KPROBE_FLAG_OPTIMIZED
) {
699 if (optprobe_queued_unopt(op
)) {
700 /* This is under unoptimizing. Just dequeue the probe */
701 list_del_init(&op
->list
);
705 op
->kp
.flags
|= KPROBE_FLAG_OPTIMIZED
;
708 * On the 'unoptimizing_list' and 'optimizing_list',
709 * 'op' must have OPTIMIZED flag
711 if (WARN_ON_ONCE(!list_empty(&op
->list
)))
714 list_add(&op
->list
, &optimizing_list
);
715 kick_kprobe_optimizer();
718 /* Short cut to direct unoptimizing */
719 static void force_unoptimize_kprobe(struct optimized_kprobe
*op
)
721 lockdep_assert_cpus_held();
722 arch_unoptimize_kprobe(op
);
723 op
->kp
.flags
&= ~KPROBE_FLAG_OPTIMIZED
;
726 /* Unoptimize a kprobe if p is optimized */
727 static void unoptimize_kprobe(struct kprobe
*p
, bool force
)
729 struct optimized_kprobe
*op
;
731 if (!kprobe_aggrprobe(p
) || kprobe_disarmed(p
))
732 return; /* This is not an optprobe nor optimized */
734 op
= container_of(p
, struct optimized_kprobe
, kp
);
735 if (!kprobe_optimized(p
))
738 if (!list_empty(&op
->list
)) {
739 if (optprobe_queued_unopt(op
)) {
740 /* Queued in unoptimizing queue */
743 * Forcibly unoptimize the kprobe here, and queue it
744 * in the freeing list for release afterwards.
746 force_unoptimize_kprobe(op
);
747 list_move(&op
->list
, &freeing_list
);
750 /* Dequeue from the optimizing queue */
751 list_del_init(&op
->list
);
752 op
->kp
.flags
&= ~KPROBE_FLAG_OPTIMIZED
;
757 /* Optimized kprobe case */
759 /* Forcibly update the code: this is a special case */
760 force_unoptimize_kprobe(op
);
762 list_add(&op
->list
, &unoptimizing_list
);
763 kick_kprobe_optimizer();
767 /* Cancel unoptimizing for reusing */
768 static int reuse_unused_kprobe(struct kprobe
*ap
)
770 struct optimized_kprobe
*op
;
773 * Unused kprobe MUST be on the way of delayed unoptimizing (means
774 * there is still a relative jump) and disabled.
776 op
= container_of(ap
, struct optimized_kprobe
, kp
);
777 WARN_ON_ONCE(list_empty(&op
->list
));
778 /* Enable the probe again */
779 ap
->flags
&= ~KPROBE_FLAG_DISABLED
;
780 /* Optimize it again. (remove from 'op->list') */
781 if (!kprobe_optready(ap
))
788 /* Remove optimized instructions */
789 static void kill_optimized_kprobe(struct kprobe
*p
)
791 struct optimized_kprobe
*op
;
793 op
= container_of(p
, struct optimized_kprobe
, kp
);
794 if (!list_empty(&op
->list
))
795 /* Dequeue from the (un)optimization queue */
796 list_del_init(&op
->list
);
797 op
->kp
.flags
&= ~KPROBE_FLAG_OPTIMIZED
;
799 if (kprobe_unused(p
)) {
800 /* Enqueue if it is unused */
801 list_add(&op
->list
, &freeing_list
);
803 * Remove unused probes from the hash list. After waiting
804 * for synchronization, this probe is reclaimed.
805 * (reclaiming is done by do_free_cleaned_kprobes().)
807 hlist_del_rcu(&op
->kp
.hlist
);
810 /* Don't touch the code, because it is already freed. */
811 arch_remove_optimized_kprobe(op
);
815 void __prepare_optimized_kprobe(struct optimized_kprobe
*op
, struct kprobe
*p
)
817 if (!kprobe_ftrace(p
))
818 arch_prepare_optimized_kprobe(op
, p
);
821 /* Try to prepare optimized instructions */
822 static void prepare_optimized_kprobe(struct kprobe
*p
)
824 struct optimized_kprobe
*op
;
826 op
= container_of(p
, struct optimized_kprobe
, kp
);
827 __prepare_optimized_kprobe(op
, p
);
830 /* Allocate new optimized_kprobe and try to prepare optimized instructions. */
831 static struct kprobe
*alloc_aggr_kprobe(struct kprobe
*p
)
833 struct optimized_kprobe
*op
;
835 op
= kzalloc(sizeof(struct optimized_kprobe
), GFP_KERNEL
);
839 INIT_LIST_HEAD(&op
->list
);
840 op
->kp
.addr
= p
->addr
;
841 __prepare_optimized_kprobe(op
, p
);
846 static void init_aggr_kprobe(struct kprobe
*ap
, struct kprobe
*p
);
849 * Prepare an optimized_kprobe and optimize it.
850 * NOTE: 'p' must be a normal registered kprobe.
852 static void try_to_optimize_kprobe(struct kprobe
*p
)
855 struct optimized_kprobe
*op
;
857 /* Impossible to optimize ftrace-based kprobe. */
858 if (kprobe_ftrace(p
))
861 /* For preparing optimization, jump_label_text_reserved() is called. */
864 mutex_lock(&text_mutex
);
866 ap
= alloc_aggr_kprobe(p
);
870 op
= container_of(ap
, struct optimized_kprobe
, kp
);
871 if (!arch_prepared_optinsn(&op
->optinsn
)) {
872 /* If failed to setup optimizing, fallback to kprobe. */
873 arch_remove_optimized_kprobe(op
);
878 init_aggr_kprobe(ap
, p
);
879 optimize_kprobe(ap
); /* This just kicks optimizer thread. */
882 mutex_unlock(&text_mutex
);
887 static void optimize_all_kprobes(void)
889 struct hlist_head
*head
;
893 mutex_lock(&kprobe_mutex
);
894 /* If optimization is already allowed, just return. */
895 if (kprobes_allow_optimization
)
899 kprobes_allow_optimization
= true;
900 for (i
= 0; i
< KPROBE_TABLE_SIZE
; i
++) {
901 head
= &kprobe_table
[i
];
902 hlist_for_each_entry(p
, head
, hlist
)
903 if (!kprobe_disabled(p
))
907 pr_info("kprobe jump-optimization is enabled. All kprobes are optimized if possible.\n");
909 mutex_unlock(&kprobe_mutex
);
913 static void unoptimize_all_kprobes(void)
915 struct hlist_head
*head
;
919 mutex_lock(&kprobe_mutex
);
920 /* If optimization is already prohibited, just return. */
921 if (!kprobes_allow_optimization
) {
922 mutex_unlock(&kprobe_mutex
);
927 kprobes_allow_optimization
= false;
928 for (i
= 0; i
< KPROBE_TABLE_SIZE
; i
++) {
929 head
= &kprobe_table
[i
];
930 hlist_for_each_entry(p
, head
, hlist
) {
931 if (!kprobe_disabled(p
))
932 unoptimize_kprobe(p
, false);
936 mutex_unlock(&kprobe_mutex
);
938 /* Wait for unoptimizing completion. */
939 wait_for_kprobe_optimizer();
940 pr_info("kprobe jump-optimization is disabled. All kprobes are based on software breakpoint.\n");
943 static DEFINE_MUTEX(kprobe_sysctl_mutex
);
944 static int sysctl_kprobes_optimization
;
945 static int proc_kprobes_optimization_handler(struct ctl_table
*table
,
946 int write
, void *buffer
,
947 size_t *length
, loff_t
*ppos
)
951 mutex_lock(&kprobe_sysctl_mutex
);
952 sysctl_kprobes_optimization
= kprobes_allow_optimization
? 1 : 0;
953 ret
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
955 if (sysctl_kprobes_optimization
)
956 optimize_all_kprobes();
958 unoptimize_all_kprobes();
959 mutex_unlock(&kprobe_sysctl_mutex
);
964 static struct ctl_table kprobe_sysctls
[] = {
966 .procname
= "kprobes-optimization",
967 .data
= &sysctl_kprobes_optimization
,
968 .maxlen
= sizeof(int),
970 .proc_handler
= proc_kprobes_optimization_handler
,
971 .extra1
= SYSCTL_ZERO
,
972 .extra2
= SYSCTL_ONE
,
977 static void __init
kprobe_sysctls_init(void)
979 register_sysctl_init("debug", kprobe_sysctls
);
981 #endif /* CONFIG_SYSCTL */
983 /* Put a breakpoint for a probe. */
984 static void __arm_kprobe(struct kprobe
*p
)
988 lockdep_assert_held(&text_mutex
);
990 /* Find the overlapping optimized kprobes. */
991 _p
= get_optimized_kprobe(p
->addr
);
993 /* Fallback to unoptimized kprobe */
994 unoptimize_kprobe(_p
, true);
997 optimize_kprobe(p
); /* Try to optimize (add kprobe to a list) */
1000 /* Remove the breakpoint of a probe. */
1001 static void __disarm_kprobe(struct kprobe
*p
, bool reopt
)
1005 lockdep_assert_held(&text_mutex
);
1007 /* Try to unoptimize */
1008 unoptimize_kprobe(p
, kprobes_all_disarmed
);
1010 if (!kprobe_queued(p
)) {
1011 arch_disarm_kprobe(p
);
1012 /* If another kprobe was blocked, re-optimize it. */
1013 _p
= get_optimized_kprobe(p
->addr
);
1014 if (unlikely(_p
) && reopt
)
1015 optimize_kprobe(_p
);
1018 * TODO: Since unoptimization and real disarming will be done by
1019 * the worker thread, we can not check whether another probe are
1020 * unoptimized because of this probe here. It should be re-optimized
1021 * by the worker thread.
1025 #else /* !CONFIG_OPTPROBES */
1027 #define optimize_kprobe(p) do {} while (0)
1028 #define unoptimize_kprobe(p, f) do {} while (0)
1029 #define kill_optimized_kprobe(p) do {} while (0)
1030 #define prepare_optimized_kprobe(p) do {} while (0)
1031 #define try_to_optimize_kprobe(p) do {} while (0)
1032 #define __arm_kprobe(p) arch_arm_kprobe(p)
1033 #define __disarm_kprobe(p, o) arch_disarm_kprobe(p)
1034 #define kprobe_disarmed(p) kprobe_disabled(p)
1035 #define wait_for_kprobe_optimizer() do {} while (0)
1037 static int reuse_unused_kprobe(struct kprobe
*ap
)
1040 * If the optimized kprobe is NOT supported, the aggr kprobe is
1041 * released at the same time that the last aggregated kprobe is
1043 * Thus there should be no chance to reuse unused kprobe.
1049 static void free_aggr_kprobe(struct kprobe
*p
)
1051 arch_remove_kprobe(p
);
1055 static struct kprobe
*alloc_aggr_kprobe(struct kprobe
*p
)
1057 return kzalloc(sizeof(struct kprobe
), GFP_KERNEL
);
1059 #endif /* CONFIG_OPTPROBES */
1061 #ifdef CONFIG_KPROBES_ON_FTRACE
1062 static struct ftrace_ops kprobe_ftrace_ops __read_mostly
= {
1063 .func
= kprobe_ftrace_handler
,
1064 .flags
= FTRACE_OPS_FL_SAVE_REGS
,
1067 static struct ftrace_ops kprobe_ipmodify_ops __read_mostly
= {
1068 .func
= kprobe_ftrace_handler
,
1069 .flags
= FTRACE_OPS_FL_SAVE_REGS
| FTRACE_OPS_FL_IPMODIFY
,
1072 static int kprobe_ipmodify_enabled
;
1073 static int kprobe_ftrace_enabled
;
1075 static int __arm_kprobe_ftrace(struct kprobe
*p
, struct ftrace_ops
*ops
,
1080 lockdep_assert_held(&kprobe_mutex
);
1082 ret
= ftrace_set_filter_ip(ops
, (unsigned long)p
->addr
, 0, 0);
1083 if (WARN_ONCE(ret
< 0, "Failed to arm kprobe-ftrace at %pS (error %d)\n", p
->addr
, ret
))
1087 ret
= register_ftrace_function(ops
);
1088 if (WARN(ret
< 0, "Failed to register kprobe-ftrace (error %d)\n", ret
))
1097 * At this point, sinec ops is not registered, we should be sefe from
1098 * registering empty filter.
1100 ftrace_set_filter_ip(ops
, (unsigned long)p
->addr
, 1, 0);
1104 static int arm_kprobe_ftrace(struct kprobe
*p
)
1106 bool ipmodify
= (p
->post_handler
!= NULL
);
1108 return __arm_kprobe_ftrace(p
,
1109 ipmodify
? &kprobe_ipmodify_ops
: &kprobe_ftrace_ops
,
1110 ipmodify
? &kprobe_ipmodify_enabled
: &kprobe_ftrace_enabled
);
1113 static int __disarm_kprobe_ftrace(struct kprobe
*p
, struct ftrace_ops
*ops
,
1118 lockdep_assert_held(&kprobe_mutex
);
1121 ret
= unregister_ftrace_function(ops
);
1122 if (WARN(ret
< 0, "Failed to unregister kprobe-ftrace (error %d)\n", ret
))
1128 ret
= ftrace_set_filter_ip(ops
, (unsigned long)p
->addr
, 1, 0);
1129 WARN_ONCE(ret
< 0, "Failed to disarm kprobe-ftrace at %pS (error %d)\n",
1134 static int disarm_kprobe_ftrace(struct kprobe
*p
)
1136 bool ipmodify
= (p
->post_handler
!= NULL
);
1138 return __disarm_kprobe_ftrace(p
,
1139 ipmodify
? &kprobe_ipmodify_ops
: &kprobe_ftrace_ops
,
1140 ipmodify
? &kprobe_ipmodify_enabled
: &kprobe_ftrace_enabled
);
1142 #else /* !CONFIG_KPROBES_ON_FTRACE */
1143 static inline int arm_kprobe_ftrace(struct kprobe
*p
)
1148 static inline int disarm_kprobe_ftrace(struct kprobe
*p
)
1154 static int prepare_kprobe(struct kprobe
*p
)
1156 /* Must ensure p->addr is really on ftrace */
1157 if (kprobe_ftrace(p
))
1158 return arch_prepare_kprobe_ftrace(p
);
1160 return arch_prepare_kprobe(p
);
1163 static int arm_kprobe(struct kprobe
*kp
)
1165 if (unlikely(kprobe_ftrace(kp
)))
1166 return arm_kprobe_ftrace(kp
);
1169 mutex_lock(&text_mutex
);
1171 mutex_unlock(&text_mutex
);
1177 static int disarm_kprobe(struct kprobe
*kp
, bool reopt
)
1179 if (unlikely(kprobe_ftrace(kp
)))
1180 return disarm_kprobe_ftrace(kp
);
1183 mutex_lock(&text_mutex
);
1184 __disarm_kprobe(kp
, reopt
);
1185 mutex_unlock(&text_mutex
);
1192 * Aggregate handlers for multiple kprobes support - these handlers
1193 * take care of invoking the individual kprobe handlers on p->list
1195 static int aggr_pre_handler(struct kprobe
*p
, struct pt_regs
*regs
)
1199 list_for_each_entry_rcu(kp
, &p
->list
, list
) {
1200 if (kp
->pre_handler
&& likely(!kprobe_disabled(kp
))) {
1201 set_kprobe_instance(kp
);
1202 if (kp
->pre_handler(kp
, regs
))
1205 reset_kprobe_instance();
1209 NOKPROBE_SYMBOL(aggr_pre_handler
);
1211 static void aggr_post_handler(struct kprobe
*p
, struct pt_regs
*regs
,
1212 unsigned long flags
)
1216 list_for_each_entry_rcu(kp
, &p
->list
, list
) {
1217 if (kp
->post_handler
&& likely(!kprobe_disabled(kp
))) {
1218 set_kprobe_instance(kp
);
1219 kp
->post_handler(kp
, regs
, flags
);
1220 reset_kprobe_instance();
1224 NOKPROBE_SYMBOL(aggr_post_handler
);
1226 /* Walks the list and increments 'nmissed' if 'p' has child probes. */
1227 void kprobes_inc_nmissed_count(struct kprobe
*p
)
1231 if (!kprobe_aggrprobe(p
)) {
1234 list_for_each_entry_rcu(kp
, &p
->list
, list
)
1238 NOKPROBE_SYMBOL(kprobes_inc_nmissed_count
);
1240 static struct kprobe kprobe_busy
= {
1241 .addr
= (void *) get_kprobe
,
1244 void kprobe_busy_begin(void)
1246 struct kprobe_ctlblk
*kcb
;
1249 __this_cpu_write(current_kprobe
, &kprobe_busy
);
1250 kcb
= get_kprobe_ctlblk();
1251 kcb
->kprobe_status
= KPROBE_HIT_ACTIVE
;
1254 void kprobe_busy_end(void)
1256 __this_cpu_write(current_kprobe
, NULL
);
1260 /* Add the new probe to 'ap->list'. */
1261 static int add_new_kprobe(struct kprobe
*ap
, struct kprobe
*p
)
1263 if (p
->post_handler
)
1264 unoptimize_kprobe(ap
, true); /* Fall back to normal kprobe */
1266 list_add_rcu(&p
->list
, &ap
->list
);
1267 if (p
->post_handler
&& !ap
->post_handler
)
1268 ap
->post_handler
= aggr_post_handler
;
1274 * Fill in the required fields of the aggregator kprobe. Replace the
1275 * earlier kprobe in the hlist with the aggregator kprobe.
1277 static void init_aggr_kprobe(struct kprobe
*ap
, struct kprobe
*p
)
1279 /* Copy the insn slot of 'p' to 'ap'. */
1281 flush_insn_slot(ap
);
1283 ap
->flags
= p
->flags
& ~KPROBE_FLAG_OPTIMIZED
;
1284 ap
->pre_handler
= aggr_pre_handler
;
1285 /* We don't care the kprobe which has gone. */
1286 if (p
->post_handler
&& !kprobe_gone(p
))
1287 ap
->post_handler
= aggr_post_handler
;
1289 INIT_LIST_HEAD(&ap
->list
);
1290 INIT_HLIST_NODE(&ap
->hlist
);
1292 list_add_rcu(&p
->list
, &ap
->list
);
1293 hlist_replace_rcu(&p
->hlist
, &ap
->hlist
);
1297 * This registers the second or subsequent kprobe at the same address.
1299 static int register_aggr_kprobe(struct kprobe
*orig_p
, struct kprobe
*p
)
1302 struct kprobe
*ap
= orig_p
;
1306 /* For preparing optimization, jump_label_text_reserved() is called */
1308 mutex_lock(&text_mutex
);
1310 if (!kprobe_aggrprobe(orig_p
)) {
1311 /* If 'orig_p' is not an 'aggr_kprobe', create new one. */
1312 ap
= alloc_aggr_kprobe(orig_p
);
1317 init_aggr_kprobe(ap
, orig_p
);
1318 } else if (kprobe_unused(ap
)) {
1319 /* This probe is going to die. Rescue it */
1320 ret
= reuse_unused_kprobe(ap
);
1325 if (kprobe_gone(ap
)) {
1327 * Attempting to insert new probe at the same location that
1328 * had a probe in the module vaddr area which already
1329 * freed. So, the instruction slot has already been
1330 * released. We need a new slot for the new probe.
1332 ret
= arch_prepare_kprobe(ap
);
1335 * Even if fail to allocate new slot, don't need to
1336 * free the 'ap'. It will be used next time, or
1337 * freed by unregister_kprobe().
1341 /* Prepare optimized instructions if possible. */
1342 prepare_optimized_kprobe(ap
);
1345 * Clear gone flag to prevent allocating new slot again, and
1346 * set disabled flag because it is not armed yet.
1348 ap
->flags
= (ap
->flags
& ~KPROBE_FLAG_GONE
)
1349 | KPROBE_FLAG_DISABLED
;
1352 /* Copy the insn slot of 'p' to 'ap'. */
1354 ret
= add_new_kprobe(ap
, p
);
1357 mutex_unlock(&text_mutex
);
1358 jump_label_unlock();
1361 if (ret
== 0 && kprobe_disabled(ap
) && !kprobe_disabled(p
)) {
1362 ap
->flags
&= ~KPROBE_FLAG_DISABLED
;
1363 if (!kprobes_all_disarmed
) {
1364 /* Arm the breakpoint again. */
1365 ret
= arm_kprobe(ap
);
1367 ap
->flags
|= KPROBE_FLAG_DISABLED
;
1368 list_del_rcu(&p
->list
);
1376 bool __weak
arch_within_kprobe_blacklist(unsigned long addr
)
1378 /* The '__kprobes' functions and entry code must not be probed. */
1379 return addr
>= (unsigned long)__kprobes_text_start
&&
1380 addr
< (unsigned long)__kprobes_text_end
;
1383 static bool __within_kprobe_blacklist(unsigned long addr
)
1385 struct kprobe_blacklist_entry
*ent
;
1387 if (arch_within_kprobe_blacklist(addr
))
1390 * If 'kprobe_blacklist' is defined, check the address and
1391 * reject any probe registration in the prohibited area.
1393 list_for_each_entry(ent
, &kprobe_blacklist
, list
) {
1394 if (addr
>= ent
->start_addr
&& addr
< ent
->end_addr
)
1400 bool within_kprobe_blacklist(unsigned long addr
)
1402 char symname
[KSYM_NAME_LEN
], *p
;
1404 if (__within_kprobe_blacklist(addr
))
1407 /* Check if the address is on a suffixed-symbol */
1408 if (!lookup_symbol_name(addr
, symname
)) {
1409 p
= strchr(symname
, '.');
1413 addr
= (unsigned long)kprobe_lookup_name(symname
, 0);
1415 return __within_kprobe_blacklist(addr
);
1421 * arch_adjust_kprobe_addr - adjust the address
1422 * @addr: symbol base address
1423 * @offset: offset within the symbol
1424 * @on_func_entry: was this @addr+@offset on the function entry
1426 * Typically returns @addr + @offset, except for special cases where the
1427 * function might be prefixed by a CFI landing pad, in that case any offset
1428 * inside the landing pad is mapped to the first 'real' instruction of the
1431 * Specifically, for things like IBT/BTI, skip the resp. ENDBR/BTI.C
1432 * instruction at +0.
1434 kprobe_opcode_t
*__weak
arch_adjust_kprobe_addr(unsigned long addr
,
1435 unsigned long offset
,
1436 bool *on_func_entry
)
1438 *on_func_entry
= !offset
;
1439 return (kprobe_opcode_t
*)(addr
+ offset
);
1443 * If 'symbol_name' is specified, look it up and add the 'offset'
1444 * to it. This way, we can specify a relative address to a symbol.
1445 * This returns encoded errors if it fails to look up symbol or invalid
1446 * combination of parameters.
1448 static kprobe_opcode_t
*
1449 _kprobe_addr(kprobe_opcode_t
*addr
, const char *symbol_name
,
1450 unsigned long offset
, bool *on_func_entry
)
1452 if ((symbol_name
&& addr
) || (!symbol_name
&& !addr
))
1457 * Input: @sym + @offset
1458 * Output: @addr + @offset
1460 * NOTE: kprobe_lookup_name() does *NOT* fold the offset
1461 * argument into it's output!
1463 addr
= kprobe_lookup_name(symbol_name
, offset
);
1465 return ERR_PTR(-ENOENT
);
1469 * So here we have @addr + @offset, displace it into a new
1470 * @addr' + @offset' where @addr' is the symbol start address.
1472 addr
= (void *)addr
+ offset
;
1473 if (!kallsyms_lookup_size_offset((unsigned long)addr
, NULL
, &offset
))
1474 return ERR_PTR(-ENOENT
);
1475 addr
= (void *)addr
- offset
;
1478 * Then ask the architecture to re-combine them, taking care of
1479 * magical function entry details while telling us if this was indeed
1480 * at the start of the function.
1482 addr
= arch_adjust_kprobe_addr((unsigned long)addr
, offset
, on_func_entry
);
1487 return ERR_PTR(-EINVAL
);
1490 static kprobe_opcode_t
*kprobe_addr(struct kprobe
*p
)
1493 return _kprobe_addr(p
->addr
, p
->symbol_name
, p
->offset
, &on_func_entry
);
1497 * Check the 'p' is valid and return the aggregator kprobe
1498 * at the same address.
1500 static struct kprobe
*__get_valid_kprobe(struct kprobe
*p
)
1502 struct kprobe
*ap
, *list_p
;
1504 lockdep_assert_held(&kprobe_mutex
);
1506 ap
= get_kprobe(p
->addr
);
1511 list_for_each_entry(list_p
, &ap
->list
, list
)
1513 /* kprobe p is a valid probe */
1522 * Warn and return error if the kprobe is being re-registered since
1523 * there must be a software bug.
1525 static inline int warn_kprobe_rereg(struct kprobe
*p
)
1529 mutex_lock(&kprobe_mutex
);
1530 if (WARN_ON_ONCE(__get_valid_kprobe(p
)))
1532 mutex_unlock(&kprobe_mutex
);
1537 static int check_ftrace_location(struct kprobe
*p
)
1539 unsigned long addr
= (unsigned long)p
->addr
;
1541 if (ftrace_location(addr
) == addr
) {
1542 #ifdef CONFIG_KPROBES_ON_FTRACE
1543 p
->flags
|= KPROBE_FLAG_FTRACE
;
1544 #else /* !CONFIG_KPROBES_ON_FTRACE */
1551 static int check_kprobe_address_safe(struct kprobe
*p
,
1552 struct module
**probed_mod
)
1556 ret
= check_ftrace_location(p
);
1562 /* Ensure it is not in reserved area nor out of text */
1563 if (!(core_kernel_text((unsigned long) p
->addr
) ||
1564 is_module_text_address((unsigned long) p
->addr
)) ||
1565 in_gate_area_no_mm((unsigned long) p
->addr
) ||
1566 within_kprobe_blacklist((unsigned long) p
->addr
) ||
1567 jump_label_text_reserved(p
->addr
, p
->addr
) ||
1568 static_call_text_reserved(p
->addr
, p
->addr
) ||
1569 find_bug((unsigned long)p
->addr
)) {
1574 /* Check if 'p' is probing a module. */
1575 *probed_mod
= __module_text_address((unsigned long) p
->addr
);
1578 * We must hold a refcount of the probed module while updating
1579 * its code to prohibit unexpected unloading.
1581 if (unlikely(!try_module_get(*probed_mod
))) {
1587 * If the module freed '.init.text', we couldn't insert
1590 if (within_module_init((unsigned long)p
->addr
, *probed_mod
) &&
1591 (*probed_mod
)->state
!= MODULE_STATE_COMING
) {
1592 module_put(*probed_mod
);
1599 jump_label_unlock();
1604 int register_kprobe(struct kprobe
*p
)
1607 struct kprobe
*old_p
;
1608 struct module
*probed_mod
;
1609 kprobe_opcode_t
*addr
;
1612 /* Adjust probe address from symbol */
1613 addr
= _kprobe_addr(p
->addr
, p
->symbol_name
, p
->offset
, &on_func_entry
);
1615 return PTR_ERR(addr
);
1618 ret
= warn_kprobe_rereg(p
);
1622 /* User can pass only KPROBE_FLAG_DISABLED to register_kprobe */
1623 p
->flags
&= KPROBE_FLAG_DISABLED
;
1625 INIT_LIST_HEAD(&p
->list
);
1627 ret
= check_kprobe_address_safe(p
, &probed_mod
);
1631 mutex_lock(&kprobe_mutex
);
1634 p
->flags
|= KPROBE_FLAG_ON_FUNC_ENTRY
;
1636 old_p
= get_kprobe(p
->addr
);
1638 /* Since this may unoptimize 'old_p', locking 'text_mutex'. */
1639 ret
= register_aggr_kprobe(old_p
, p
);
1644 /* Prevent text modification */
1645 mutex_lock(&text_mutex
);
1646 ret
= prepare_kprobe(p
);
1647 mutex_unlock(&text_mutex
);
1652 INIT_HLIST_NODE(&p
->hlist
);
1653 hlist_add_head_rcu(&p
->hlist
,
1654 &kprobe_table
[hash_ptr(p
->addr
, KPROBE_HASH_BITS
)]);
1656 if (!kprobes_all_disarmed
&& !kprobe_disabled(p
)) {
1657 ret
= arm_kprobe(p
);
1659 hlist_del_rcu(&p
->hlist
);
1665 /* Try to optimize kprobe */
1666 try_to_optimize_kprobe(p
);
1668 mutex_unlock(&kprobe_mutex
);
1671 module_put(probed_mod
);
1675 EXPORT_SYMBOL_GPL(register_kprobe
);
1677 /* Check if all probes on the 'ap' are disabled. */
1678 static bool aggr_kprobe_disabled(struct kprobe
*ap
)
1682 lockdep_assert_held(&kprobe_mutex
);
1684 list_for_each_entry(kp
, &ap
->list
, list
)
1685 if (!kprobe_disabled(kp
))
1687 * Since there is an active probe on the list,
1688 * we can't disable this 'ap'.
1695 static struct kprobe
*__disable_kprobe(struct kprobe
*p
)
1697 struct kprobe
*orig_p
;
1700 lockdep_assert_held(&kprobe_mutex
);
1702 /* Get an original kprobe for return */
1703 orig_p
= __get_valid_kprobe(p
);
1704 if (unlikely(orig_p
== NULL
))
1705 return ERR_PTR(-EINVAL
);
1707 if (!kprobe_disabled(p
)) {
1708 /* Disable probe if it is a child probe */
1710 p
->flags
|= KPROBE_FLAG_DISABLED
;
1712 /* Try to disarm and disable this/parent probe */
1713 if (p
== orig_p
|| aggr_kprobe_disabled(orig_p
)) {
1715 * Don't be lazy here. Even if 'kprobes_all_disarmed'
1716 * is false, 'orig_p' might not have been armed yet.
1717 * Note arm_all_kprobes() __tries__ to arm all kprobes
1718 * on the best effort basis.
1720 if (!kprobes_all_disarmed
&& !kprobe_disabled(orig_p
)) {
1721 ret
= disarm_kprobe(orig_p
, true);
1723 p
->flags
&= ~KPROBE_FLAG_DISABLED
;
1724 return ERR_PTR(ret
);
1727 orig_p
->flags
|= KPROBE_FLAG_DISABLED
;
1735 * Unregister a kprobe without a scheduler synchronization.
1737 static int __unregister_kprobe_top(struct kprobe
*p
)
1739 struct kprobe
*ap
, *list_p
;
1741 /* Disable kprobe. This will disarm it if needed. */
1742 ap
= __disable_kprobe(p
);
1748 * This probe is an independent(and non-optimized) kprobe
1749 * (not an aggrprobe). Remove from the hash list.
1753 /* Following process expects this probe is an aggrprobe */
1754 WARN_ON(!kprobe_aggrprobe(ap
));
1756 if (list_is_singular(&ap
->list
) && kprobe_disarmed(ap
))
1758 * !disarmed could be happen if the probe is under delayed
1763 /* If disabling probe has special handlers, update aggrprobe */
1764 if (p
->post_handler
&& !kprobe_gone(p
)) {
1765 list_for_each_entry(list_p
, &ap
->list
, list
) {
1766 if ((list_p
!= p
) && (list_p
->post_handler
))
1770 * For the kprobe-on-ftrace case, we keep the
1771 * post_handler setting to identify this aggrprobe
1772 * armed with kprobe_ipmodify_ops.
1774 if (!kprobe_ftrace(ap
))
1775 ap
->post_handler
= NULL
;
1779 * Remove from the aggrprobe: this path will do nothing in
1780 * __unregister_kprobe_bottom().
1782 list_del_rcu(&p
->list
);
1783 if (!kprobe_disabled(ap
) && !kprobes_all_disarmed
)
1785 * Try to optimize this probe again, because post
1786 * handler may have been changed.
1788 optimize_kprobe(ap
);
1793 hlist_del_rcu(&ap
->hlist
);
1797 static void __unregister_kprobe_bottom(struct kprobe
*p
)
1801 if (list_empty(&p
->list
))
1802 /* This is an independent kprobe */
1803 arch_remove_kprobe(p
);
1804 else if (list_is_singular(&p
->list
)) {
1805 /* This is the last child of an aggrprobe */
1806 ap
= list_entry(p
->list
.next
, struct kprobe
, list
);
1808 free_aggr_kprobe(ap
);
1810 /* Otherwise, do nothing. */
1813 int register_kprobes(struct kprobe
**kps
, int num
)
1819 for (i
= 0; i
< num
; i
++) {
1820 ret
= register_kprobe(kps
[i
]);
1823 unregister_kprobes(kps
, i
);
1829 EXPORT_SYMBOL_GPL(register_kprobes
);
1831 void unregister_kprobe(struct kprobe
*p
)
1833 unregister_kprobes(&p
, 1);
1835 EXPORT_SYMBOL_GPL(unregister_kprobe
);
1837 void unregister_kprobes(struct kprobe
**kps
, int num
)
1843 mutex_lock(&kprobe_mutex
);
1844 for (i
= 0; i
< num
; i
++)
1845 if (__unregister_kprobe_top(kps
[i
]) < 0)
1846 kps
[i
]->addr
= NULL
;
1847 mutex_unlock(&kprobe_mutex
);
1850 for (i
= 0; i
< num
; i
++)
1852 __unregister_kprobe_bottom(kps
[i
]);
1854 EXPORT_SYMBOL_GPL(unregister_kprobes
);
1856 int __weak
kprobe_exceptions_notify(struct notifier_block
*self
,
1857 unsigned long val
, void *data
)
1861 NOKPROBE_SYMBOL(kprobe_exceptions_notify
);
1863 static struct notifier_block kprobe_exceptions_nb
= {
1864 .notifier_call
= kprobe_exceptions_notify
,
1865 .priority
= 0x7fffffff /* we need to be notified first */
1868 #ifdef CONFIG_KRETPROBES
1870 #if !defined(CONFIG_KRETPROBE_ON_RETHOOK)
1871 static void free_rp_inst_rcu(struct rcu_head
*head
)
1873 struct kretprobe_instance
*ri
= container_of(head
, struct kretprobe_instance
, rcu
);
1875 if (refcount_dec_and_test(&ri
->rph
->ref
))
1879 NOKPROBE_SYMBOL(free_rp_inst_rcu
);
1881 static void recycle_rp_inst(struct kretprobe_instance
*ri
)
1883 struct kretprobe
*rp
= get_kretprobe(ri
);
1886 freelist_add(&ri
->freelist
, &rp
->freelist
);
1888 call_rcu(&ri
->rcu
, free_rp_inst_rcu
);
1890 NOKPROBE_SYMBOL(recycle_rp_inst
);
1893 * This function is called from delayed_put_task_struct() when a task is
1894 * dead and cleaned up to recycle any kretprobe instances associated with
1895 * this task. These left over instances represent probed functions that
1896 * have been called but will never return.
1898 void kprobe_flush_task(struct task_struct
*tk
)
1900 struct kretprobe_instance
*ri
;
1901 struct llist_node
*node
;
1903 /* Early boot, not yet initialized. */
1904 if (unlikely(!kprobes_initialized
))
1907 kprobe_busy_begin();
1909 node
= __llist_del_all(&tk
->kretprobe_instances
);
1911 ri
= container_of(node
, struct kretprobe_instance
, llist
);
1914 recycle_rp_inst(ri
);
1919 NOKPROBE_SYMBOL(kprobe_flush_task
);
1921 static inline void free_rp_inst(struct kretprobe
*rp
)
1923 struct kretprobe_instance
*ri
;
1924 struct freelist_node
*node
;
1927 node
= rp
->freelist
.head
;
1929 ri
= container_of(node
, struct kretprobe_instance
, freelist
);
1936 if (refcount_sub_and_test(count
, &rp
->rph
->ref
)) {
1942 /* This assumes the 'tsk' is the current task or the is not running. */
1943 static kprobe_opcode_t
*__kretprobe_find_ret_addr(struct task_struct
*tsk
,
1944 struct llist_node
**cur
)
1946 struct kretprobe_instance
*ri
= NULL
;
1947 struct llist_node
*node
= *cur
;
1950 node
= tsk
->kretprobe_instances
.first
;
1955 ri
= container_of(node
, struct kretprobe_instance
, llist
);
1956 if (ri
->ret_addr
!= kretprobe_trampoline_addr()) {
1958 return ri
->ret_addr
;
1964 NOKPROBE_SYMBOL(__kretprobe_find_ret_addr
);
1967 * kretprobe_find_ret_addr -- Find correct return address modified by kretprobe
1969 * @fp: A frame pointer
1970 * @cur: a storage of the loop cursor llist_node pointer for next call
1972 * Find the correct return address modified by a kretprobe on @tsk in unsigned
1973 * long type. If it finds the return address, this returns that address value,
1974 * or this returns 0.
1975 * The @tsk must be 'current' or a task which is not running. @fp is a hint
1976 * to get the currect return address - which is compared with the
1977 * kretprobe_instance::fp field. The @cur is a loop cursor for searching the
1978 * kretprobe return addresses on the @tsk. The '*@cur' should be NULL at the
1979 * first call, but '@cur' itself must NOT NULL.
1981 unsigned long kretprobe_find_ret_addr(struct task_struct
*tsk
, void *fp
,
1982 struct llist_node
**cur
)
1984 struct kretprobe_instance
*ri
= NULL
;
1985 kprobe_opcode_t
*ret
;
1987 if (WARN_ON_ONCE(!cur
))
1991 ret
= __kretprobe_find_ret_addr(tsk
, cur
);
1994 ri
= container_of(*cur
, struct kretprobe_instance
, llist
);
1995 } while (ri
->fp
!= fp
);
1997 return (unsigned long)ret
;
1999 NOKPROBE_SYMBOL(kretprobe_find_ret_addr
);
2001 void __weak
arch_kretprobe_fixup_return(struct pt_regs
*regs
,
2002 kprobe_opcode_t
*correct_ret_addr
)
2005 * Do nothing by default. Please fill this to update the fake return
2006 * address on the stack with the correct one on each arch if possible.
2010 unsigned long __kretprobe_trampoline_handler(struct pt_regs
*regs
,
2011 void *frame_pointer
)
2013 kprobe_opcode_t
*correct_ret_addr
= NULL
;
2014 struct kretprobe_instance
*ri
= NULL
;
2015 struct llist_node
*first
, *node
= NULL
;
2016 struct kretprobe
*rp
;
2018 /* Find correct address and all nodes for this frame. */
2019 correct_ret_addr
= __kretprobe_find_ret_addr(current
, &node
);
2020 if (!correct_ret_addr
) {
2021 pr_err("kretprobe: Return address not found, not execute handler. Maybe there is a bug in the kernel.\n");
2026 * Set the return address as the instruction pointer, because if the
2027 * user handler calls stack_trace_save_regs() with this 'regs',
2028 * the stack trace will start from the instruction pointer.
2030 instruction_pointer_set(regs
, (unsigned long)correct_ret_addr
);
2032 /* Run the user handler of the nodes. */
2033 first
= current
->kretprobe_instances
.first
;
2035 ri
= container_of(first
, struct kretprobe_instance
, llist
);
2037 if (WARN_ON_ONCE(ri
->fp
!= frame_pointer
))
2040 rp
= get_kretprobe(ri
);
2041 if (rp
&& rp
->handler
) {
2042 struct kprobe
*prev
= kprobe_running();
2044 __this_cpu_write(current_kprobe
, &rp
->kp
);
2045 ri
->ret_addr
= correct_ret_addr
;
2046 rp
->handler(ri
, regs
);
2047 __this_cpu_write(current_kprobe
, prev
);
2052 first
= first
->next
;
2055 arch_kretprobe_fixup_return(regs
, correct_ret_addr
);
2057 /* Unlink all nodes for this frame. */
2058 first
= current
->kretprobe_instances
.first
;
2059 current
->kretprobe_instances
.first
= node
->next
;
2062 /* Recycle free instances. */
2064 ri
= container_of(first
, struct kretprobe_instance
, llist
);
2065 first
= first
->next
;
2067 recycle_rp_inst(ri
);
2070 return (unsigned long)correct_ret_addr
;
2072 NOKPROBE_SYMBOL(__kretprobe_trampoline_handler
)
2075 * This kprobe pre_handler is registered with every kretprobe. When probe
2076 * hits it will set up the return probe.
2078 static int pre_handler_kretprobe(struct kprobe
*p
, struct pt_regs
*regs
)
2080 struct kretprobe
*rp
= container_of(p
, struct kretprobe
, kp
);
2081 struct kretprobe_instance
*ri
;
2082 struct freelist_node
*fn
;
2084 fn
= freelist_try_get(&rp
->freelist
);
2090 ri
= container_of(fn
, struct kretprobe_instance
, freelist
);
2092 if (rp
->entry_handler
&& rp
->entry_handler(ri
, regs
)) {
2093 freelist_add(&ri
->freelist
, &rp
->freelist
);
2097 arch_prepare_kretprobe(ri
, regs
);
2099 __llist_add(&ri
->llist
, ¤t
->kretprobe_instances
);
2103 NOKPROBE_SYMBOL(pre_handler_kretprobe
);
2104 #else /* CONFIG_KRETPROBE_ON_RETHOOK */
2106 * This kprobe pre_handler is registered with every kretprobe. When probe
2107 * hits it will set up the return probe.
2109 static int pre_handler_kretprobe(struct kprobe
*p
, struct pt_regs
*regs
)
2111 struct kretprobe
*rp
= container_of(p
, struct kretprobe
, kp
);
2112 struct kretprobe_instance
*ri
;
2113 struct rethook_node
*rhn
;
2115 rhn
= rethook_try_get(rp
->rh
);
2121 ri
= container_of(rhn
, struct kretprobe_instance
, node
);
2123 if (rp
->entry_handler
&& rp
->entry_handler(ri
, regs
))
2124 rethook_recycle(rhn
);
2126 rethook_hook(rhn
, regs
, kprobe_ftrace(p
));
2130 NOKPROBE_SYMBOL(pre_handler_kretprobe
);
2132 static void kretprobe_rethook_handler(struct rethook_node
*rh
, void *data
,
2133 struct pt_regs
*regs
)
2135 struct kretprobe
*rp
= (struct kretprobe
*)data
;
2136 struct kretprobe_instance
*ri
;
2137 struct kprobe_ctlblk
*kcb
;
2139 /* The data must NOT be null. This means rethook data structure is broken. */
2140 if (WARN_ON_ONCE(!data
) || !rp
->handler
)
2143 __this_cpu_write(current_kprobe
, &rp
->kp
);
2144 kcb
= get_kprobe_ctlblk();
2145 kcb
->kprobe_status
= KPROBE_HIT_ACTIVE
;
2147 ri
= container_of(rh
, struct kretprobe_instance
, node
);
2148 rp
->handler(ri
, regs
);
2150 __this_cpu_write(current_kprobe
, NULL
);
2152 NOKPROBE_SYMBOL(kretprobe_rethook_handler
);
2154 #endif /* !CONFIG_KRETPROBE_ON_RETHOOK */
2157 * kprobe_on_func_entry() -- check whether given address is function entry
2158 * @addr: Target address
2159 * @sym: Target symbol name
2160 * @offset: The offset from the symbol or the address
2162 * This checks whether the given @addr+@offset or @sym+@offset is on the
2163 * function entry address or not.
2164 * This returns 0 if it is the function entry, or -EINVAL if it is not.
2165 * And also it returns -ENOENT if it fails the symbol or address lookup.
2166 * Caller must pass @addr or @sym (either one must be NULL), or this
2169 int kprobe_on_func_entry(kprobe_opcode_t
*addr
, const char *sym
, unsigned long offset
)
2172 kprobe_opcode_t
*kp_addr
= _kprobe_addr(addr
, sym
, offset
, &on_func_entry
);
2174 if (IS_ERR(kp_addr
))
2175 return PTR_ERR(kp_addr
);
2183 int register_kretprobe(struct kretprobe
*rp
)
2186 struct kretprobe_instance
*inst
;
2190 ret
= kprobe_on_func_entry(rp
->kp
.addr
, rp
->kp
.symbol_name
, rp
->kp
.offset
);
2194 /* If only 'rp->kp.addr' is specified, check reregistering kprobes */
2195 if (rp
->kp
.addr
&& warn_kprobe_rereg(&rp
->kp
))
2198 if (kretprobe_blacklist_size
) {
2199 addr
= kprobe_addr(&rp
->kp
);
2201 return PTR_ERR(addr
);
2203 for (i
= 0; kretprobe_blacklist
[i
].name
!= NULL
; i
++) {
2204 if (kretprobe_blacklist
[i
].addr
== addr
)
2209 if (rp
->data_size
> KRETPROBE_MAX_DATA_SIZE
)
2212 rp
->kp
.pre_handler
= pre_handler_kretprobe
;
2213 rp
->kp
.post_handler
= NULL
;
2215 /* Pre-allocate memory for max kretprobe instances */
2216 if (rp
->maxactive
<= 0)
2217 rp
->maxactive
= max_t(unsigned int, 10, 2*num_possible_cpus());
2219 #ifdef CONFIG_KRETPROBE_ON_RETHOOK
2220 rp
->rh
= rethook_alloc((void *)rp
, kretprobe_rethook_handler
);
2224 for (i
= 0; i
< rp
->maxactive
; i
++) {
2225 inst
= kzalloc(sizeof(struct kretprobe_instance
) +
2226 rp
->data_size
, GFP_KERNEL
);
2228 rethook_free(rp
->rh
);
2232 rethook_add_node(rp
->rh
, &inst
->node
);
2235 /* Establish function entry probe point */
2236 ret
= register_kprobe(&rp
->kp
);
2238 rethook_free(rp
->rh
);
2241 #else /* !CONFIG_KRETPROBE_ON_RETHOOK */
2242 rp
->freelist
.head
= NULL
;
2243 rp
->rph
= kzalloc(sizeof(struct kretprobe_holder
), GFP_KERNEL
);
2248 for (i
= 0; i
< rp
->maxactive
; i
++) {
2249 inst
= kzalloc(sizeof(struct kretprobe_instance
) +
2250 rp
->data_size
, GFP_KERNEL
);
2252 refcount_set(&rp
->rph
->ref
, i
);
2256 inst
->rph
= rp
->rph
;
2257 freelist_add(&inst
->freelist
, &rp
->freelist
);
2259 refcount_set(&rp
->rph
->ref
, i
);
2262 /* Establish function entry probe point */
2263 ret
= register_kprobe(&rp
->kp
);
2269 EXPORT_SYMBOL_GPL(register_kretprobe
);
2271 int register_kretprobes(struct kretprobe
**rps
, int num
)
2277 for (i
= 0; i
< num
; i
++) {
2278 ret
= register_kretprobe(rps
[i
]);
2281 unregister_kretprobes(rps
, i
);
2287 EXPORT_SYMBOL_GPL(register_kretprobes
);
2289 void unregister_kretprobe(struct kretprobe
*rp
)
2291 unregister_kretprobes(&rp
, 1);
2293 EXPORT_SYMBOL_GPL(unregister_kretprobe
);
2295 void unregister_kretprobes(struct kretprobe
**rps
, int num
)
2301 mutex_lock(&kprobe_mutex
);
2302 for (i
= 0; i
< num
; i
++) {
2303 if (__unregister_kprobe_top(&rps
[i
]->kp
) < 0)
2304 rps
[i
]->kp
.addr
= NULL
;
2305 #ifdef CONFIG_KRETPROBE_ON_RETHOOK
2306 rethook_free(rps
[i
]->rh
);
2308 rps
[i
]->rph
->rp
= NULL
;
2311 mutex_unlock(&kprobe_mutex
);
2314 for (i
= 0; i
< num
; i
++) {
2315 if (rps
[i
]->kp
.addr
) {
2316 __unregister_kprobe_bottom(&rps
[i
]->kp
);
2317 #ifndef CONFIG_KRETPROBE_ON_RETHOOK
2318 free_rp_inst(rps
[i
]);
2323 EXPORT_SYMBOL_GPL(unregister_kretprobes
);
2325 #else /* CONFIG_KRETPROBES */
2326 int register_kretprobe(struct kretprobe
*rp
)
2330 EXPORT_SYMBOL_GPL(register_kretprobe
);
2332 int register_kretprobes(struct kretprobe
**rps
, int num
)
2336 EXPORT_SYMBOL_GPL(register_kretprobes
);
2338 void unregister_kretprobe(struct kretprobe
*rp
)
2341 EXPORT_SYMBOL_GPL(unregister_kretprobe
);
2343 void unregister_kretprobes(struct kretprobe
**rps
, int num
)
2346 EXPORT_SYMBOL_GPL(unregister_kretprobes
);
2348 static int pre_handler_kretprobe(struct kprobe
*p
, struct pt_regs
*regs
)
2352 NOKPROBE_SYMBOL(pre_handler_kretprobe
);
2354 #endif /* CONFIG_KRETPROBES */
2356 /* Set the kprobe gone and remove its instruction buffer. */
2357 static void kill_kprobe(struct kprobe
*p
)
2361 lockdep_assert_held(&kprobe_mutex
);
2364 * The module is going away. We should disarm the kprobe which
2365 * is using ftrace, because ftrace framework is still available at
2366 * 'MODULE_STATE_GOING' notification.
2368 if (kprobe_ftrace(p
) && !kprobe_disabled(p
) && !kprobes_all_disarmed
)
2369 disarm_kprobe_ftrace(p
);
2371 p
->flags
|= KPROBE_FLAG_GONE
;
2372 if (kprobe_aggrprobe(p
)) {
2374 * If this is an aggr_kprobe, we have to list all the
2375 * chained probes and mark them GONE.
2377 list_for_each_entry(kp
, &p
->list
, list
)
2378 kp
->flags
|= KPROBE_FLAG_GONE
;
2379 p
->post_handler
= NULL
;
2380 kill_optimized_kprobe(p
);
2383 * Here, we can remove insn_slot safely, because no thread calls
2384 * the original probed function (which will be freed soon) any more.
2386 arch_remove_kprobe(p
);
2389 /* Disable one kprobe */
2390 int disable_kprobe(struct kprobe
*kp
)
2395 mutex_lock(&kprobe_mutex
);
2397 /* Disable this kprobe */
2398 p
= __disable_kprobe(kp
);
2402 mutex_unlock(&kprobe_mutex
);
2405 EXPORT_SYMBOL_GPL(disable_kprobe
);
2407 /* Enable one kprobe */
2408 int enable_kprobe(struct kprobe
*kp
)
2413 mutex_lock(&kprobe_mutex
);
2415 /* Check whether specified probe is valid. */
2416 p
= __get_valid_kprobe(kp
);
2417 if (unlikely(p
== NULL
)) {
2422 if (kprobe_gone(kp
)) {
2423 /* This kprobe has gone, we couldn't enable it. */
2429 kp
->flags
&= ~KPROBE_FLAG_DISABLED
;
2431 if (!kprobes_all_disarmed
&& kprobe_disabled(p
)) {
2432 p
->flags
&= ~KPROBE_FLAG_DISABLED
;
2433 ret
= arm_kprobe(p
);
2435 p
->flags
|= KPROBE_FLAG_DISABLED
;
2437 kp
->flags
|= KPROBE_FLAG_DISABLED
;
2441 mutex_unlock(&kprobe_mutex
);
2444 EXPORT_SYMBOL_GPL(enable_kprobe
);
2446 /* Caller must NOT call this in usual path. This is only for critical case */
2447 void dump_kprobe(struct kprobe
*kp
)
2449 pr_err("Dump kprobe:\n.symbol_name = %s, .offset = %x, .addr = %pS\n",
2450 kp
->symbol_name
, kp
->offset
, kp
->addr
);
2452 NOKPROBE_SYMBOL(dump_kprobe
);
2454 int kprobe_add_ksym_blacklist(unsigned long entry
)
2456 struct kprobe_blacklist_entry
*ent
;
2457 unsigned long offset
= 0, size
= 0;
2459 if (!kernel_text_address(entry
) ||
2460 !kallsyms_lookup_size_offset(entry
, &size
, &offset
))
2463 ent
= kmalloc(sizeof(*ent
), GFP_KERNEL
);
2466 ent
->start_addr
= entry
;
2467 ent
->end_addr
= entry
+ size
;
2468 INIT_LIST_HEAD(&ent
->list
);
2469 list_add_tail(&ent
->list
, &kprobe_blacklist
);
2474 /* Add all symbols in given area into kprobe blacklist */
2475 int kprobe_add_area_blacklist(unsigned long start
, unsigned long end
)
2477 unsigned long entry
;
2480 for (entry
= start
; entry
< end
; entry
+= ret
) {
2481 ret
= kprobe_add_ksym_blacklist(entry
);
2484 if (ret
== 0) /* In case of alias symbol */
2490 /* Remove all symbols in given area from kprobe blacklist */
2491 static void kprobe_remove_area_blacklist(unsigned long start
, unsigned long end
)
2493 struct kprobe_blacklist_entry
*ent
, *n
;
2495 list_for_each_entry_safe(ent
, n
, &kprobe_blacklist
, list
) {
2496 if (ent
->start_addr
< start
|| ent
->start_addr
>= end
)
2498 list_del(&ent
->list
);
2503 static void kprobe_remove_ksym_blacklist(unsigned long entry
)
2505 kprobe_remove_area_blacklist(entry
, entry
+ 1);
2508 int __weak
arch_kprobe_get_kallsym(unsigned int *symnum
, unsigned long *value
,
2509 char *type
, char *sym
)
2514 int kprobe_get_kallsym(unsigned int symnum
, unsigned long *value
, char *type
,
2517 #ifdef __ARCH_WANT_KPROBES_INSN_SLOT
2518 if (!kprobe_cache_get_kallsym(&kprobe_insn_slots
, &symnum
, value
, type
, sym
))
2520 #ifdef CONFIG_OPTPROBES
2521 if (!kprobe_cache_get_kallsym(&kprobe_optinsn_slots
, &symnum
, value
, type
, sym
))
2525 if (!arch_kprobe_get_kallsym(&symnum
, value
, type
, sym
))
2530 int __init __weak
arch_populate_kprobe_blacklist(void)
2536 * Lookup and populate the kprobe_blacklist.
2538 * Unlike the kretprobe blacklist, we'll need to determine
2539 * the range of addresses that belong to the said functions,
2540 * since a kprobe need not necessarily be at the beginning
2543 static int __init
populate_kprobe_blacklist(unsigned long *start
,
2546 unsigned long entry
;
2547 unsigned long *iter
;
2550 for (iter
= start
; iter
< end
; iter
++) {
2551 entry
= (unsigned long)dereference_symbol_descriptor((void *)*iter
);
2552 ret
= kprobe_add_ksym_blacklist(entry
);
2559 /* Symbols in '__kprobes_text' are blacklisted */
2560 ret
= kprobe_add_area_blacklist((unsigned long)__kprobes_text_start
,
2561 (unsigned long)__kprobes_text_end
);
2565 /* Symbols in 'noinstr' section are blacklisted */
2566 ret
= kprobe_add_area_blacklist((unsigned long)__noinstr_text_start
,
2567 (unsigned long)__noinstr_text_end
);
2569 return ret
? : arch_populate_kprobe_blacklist();
2572 static void add_module_kprobe_blacklist(struct module
*mod
)
2574 unsigned long start
, end
;
2577 if (mod
->kprobe_blacklist
) {
2578 for (i
= 0; i
< mod
->num_kprobe_blacklist
; i
++)
2579 kprobe_add_ksym_blacklist(mod
->kprobe_blacklist
[i
]);
2582 start
= (unsigned long)mod
->kprobes_text_start
;
2584 end
= start
+ mod
->kprobes_text_size
;
2585 kprobe_add_area_blacklist(start
, end
);
2588 start
= (unsigned long)mod
->noinstr_text_start
;
2590 end
= start
+ mod
->noinstr_text_size
;
2591 kprobe_add_area_blacklist(start
, end
);
2595 static void remove_module_kprobe_blacklist(struct module
*mod
)
2597 unsigned long start
, end
;
2600 if (mod
->kprobe_blacklist
) {
2601 for (i
= 0; i
< mod
->num_kprobe_blacklist
; i
++)
2602 kprobe_remove_ksym_blacklist(mod
->kprobe_blacklist
[i
]);
2605 start
= (unsigned long)mod
->kprobes_text_start
;
2607 end
= start
+ mod
->kprobes_text_size
;
2608 kprobe_remove_area_blacklist(start
, end
);
2611 start
= (unsigned long)mod
->noinstr_text_start
;
2613 end
= start
+ mod
->noinstr_text_size
;
2614 kprobe_remove_area_blacklist(start
, end
);
2618 /* Module notifier call back, checking kprobes on the module */
2619 static int kprobes_module_callback(struct notifier_block
*nb
,
2620 unsigned long val
, void *data
)
2622 struct module
*mod
= data
;
2623 struct hlist_head
*head
;
2626 int checkcore
= (val
== MODULE_STATE_GOING
);
2628 if (val
== MODULE_STATE_COMING
) {
2629 mutex_lock(&kprobe_mutex
);
2630 add_module_kprobe_blacklist(mod
);
2631 mutex_unlock(&kprobe_mutex
);
2633 if (val
!= MODULE_STATE_GOING
&& val
!= MODULE_STATE_LIVE
)
2637 * When 'MODULE_STATE_GOING' was notified, both of module '.text' and
2638 * '.init.text' sections would be freed. When 'MODULE_STATE_LIVE' was
2639 * notified, only '.init.text' section would be freed. We need to
2640 * disable kprobes which have been inserted in the sections.
2642 mutex_lock(&kprobe_mutex
);
2643 for (i
= 0; i
< KPROBE_TABLE_SIZE
; i
++) {
2644 head
= &kprobe_table
[i
];
2645 hlist_for_each_entry(p
, head
, hlist
)
2646 if (within_module_init((unsigned long)p
->addr
, mod
) ||
2648 within_module_core((unsigned long)p
->addr
, mod
))) {
2650 * The vaddr this probe is installed will soon
2651 * be vfreed buy not synced to disk. Hence,
2652 * disarming the breakpoint isn't needed.
2654 * Note, this will also move any optimized probes
2655 * that are pending to be removed from their
2656 * corresponding lists to the 'freeing_list' and
2657 * will not be touched by the delayed
2658 * kprobe_optimizer() work handler.
2663 if (val
== MODULE_STATE_GOING
)
2664 remove_module_kprobe_blacklist(mod
);
2665 mutex_unlock(&kprobe_mutex
);
2669 static struct notifier_block kprobe_module_nb
= {
2670 .notifier_call
= kprobes_module_callback
,
2674 void kprobe_free_init_mem(void)
2676 void *start
= (void *)(&__init_begin
);
2677 void *end
= (void *)(&__init_end
);
2678 struct hlist_head
*head
;
2682 mutex_lock(&kprobe_mutex
);
2684 /* Kill all kprobes on initmem because the target code has been freed. */
2685 for (i
= 0; i
< KPROBE_TABLE_SIZE
; i
++) {
2686 head
= &kprobe_table
[i
];
2687 hlist_for_each_entry(p
, head
, hlist
) {
2688 if (start
<= (void *)p
->addr
&& (void *)p
->addr
< end
)
2693 mutex_unlock(&kprobe_mutex
);
2696 static int __init
init_kprobes(void)
2700 /* FIXME allocate the probe table, currently defined statically */
2701 /* initialize all list heads */
2702 for (i
= 0; i
< KPROBE_TABLE_SIZE
; i
++)
2703 INIT_HLIST_HEAD(&kprobe_table
[i
]);
2705 err
= populate_kprobe_blacklist(__start_kprobe_blacklist
,
2706 __stop_kprobe_blacklist
);
2708 pr_err("Failed to populate blacklist (error %d), kprobes not restricted, be careful using them!\n", err
);
2710 if (kretprobe_blacklist_size
) {
2711 /* lookup the function address from its name */
2712 for (i
= 0; kretprobe_blacklist
[i
].name
!= NULL
; i
++) {
2713 kretprobe_blacklist
[i
].addr
=
2714 kprobe_lookup_name(kretprobe_blacklist
[i
].name
, 0);
2715 if (!kretprobe_blacklist
[i
].addr
)
2716 pr_err("Failed to lookup symbol '%s' for kretprobe blacklist. Maybe the target function is removed or renamed.\n",
2717 kretprobe_blacklist
[i
].name
);
2721 /* By default, kprobes are armed */
2722 kprobes_all_disarmed
= false;
2724 #if defined(CONFIG_OPTPROBES) && defined(__ARCH_WANT_KPROBES_INSN_SLOT)
2725 /* Init 'kprobe_optinsn_slots' for allocation */
2726 kprobe_optinsn_slots
.insn_size
= MAX_OPTINSN_SIZE
;
2729 err
= arch_init_kprobes();
2731 err
= register_die_notifier(&kprobe_exceptions_nb
);
2733 err
= register_module_notifier(&kprobe_module_nb
);
2735 kprobes_initialized
= (err
== 0);
2736 kprobe_sysctls_init();
2739 early_initcall(init_kprobes
);
2741 #if defined(CONFIG_OPTPROBES)
2742 static int __init
init_optprobes(void)
2745 * Enable kprobe optimization - this kicks the optimizer which
2746 * depends on synchronize_rcu_tasks() and ksoftirqd, that is
2747 * not spawned in early initcall. So delay the optimization.
2749 optimize_all_kprobes();
2753 subsys_initcall(init_optprobes
);
2756 #ifdef CONFIG_DEBUG_FS
2757 static void report_probe(struct seq_file
*pi
, struct kprobe
*p
,
2758 const char *sym
, int offset
, char *modname
, struct kprobe
*pp
)
2761 void *addr
= p
->addr
;
2763 if (p
->pre_handler
== pre_handler_kretprobe
)
2768 if (!kallsyms_show_value(pi
->file
->f_cred
))
2772 seq_printf(pi
, "%px %s %s+0x%x %s ",
2773 addr
, kprobe_type
, sym
, offset
,
2774 (modname
? modname
: " "));
2775 else /* try to use %pS */
2776 seq_printf(pi
, "%px %s %pS ",
2777 addr
, kprobe_type
, p
->addr
);
2781 seq_printf(pi
, "%s%s%s%s\n",
2782 (kprobe_gone(p
) ? "[GONE]" : ""),
2783 ((kprobe_disabled(p
) && !kprobe_gone(p
)) ? "[DISABLED]" : ""),
2784 (kprobe_optimized(pp
) ? "[OPTIMIZED]" : ""),
2785 (kprobe_ftrace(pp
) ? "[FTRACE]" : ""));
2788 static void *kprobe_seq_start(struct seq_file
*f
, loff_t
*pos
)
2790 return (*pos
< KPROBE_TABLE_SIZE
) ? pos
: NULL
;
2793 static void *kprobe_seq_next(struct seq_file
*f
, void *v
, loff_t
*pos
)
2796 if (*pos
>= KPROBE_TABLE_SIZE
)
2801 static void kprobe_seq_stop(struct seq_file
*f
, void *v
)
2806 static int show_kprobe_addr(struct seq_file
*pi
, void *v
)
2808 struct hlist_head
*head
;
2809 struct kprobe
*p
, *kp
;
2810 const char *sym
= NULL
;
2811 unsigned int i
= *(loff_t
*) v
;
2812 unsigned long offset
= 0;
2813 char *modname
, namebuf
[KSYM_NAME_LEN
];
2815 head
= &kprobe_table
[i
];
2817 hlist_for_each_entry_rcu(p
, head
, hlist
) {
2818 sym
= kallsyms_lookup((unsigned long)p
->addr
, NULL
,
2819 &offset
, &modname
, namebuf
);
2820 if (kprobe_aggrprobe(p
)) {
2821 list_for_each_entry_rcu(kp
, &p
->list
, list
)
2822 report_probe(pi
, kp
, sym
, offset
, modname
, p
);
2824 report_probe(pi
, p
, sym
, offset
, modname
, NULL
);
2830 static const struct seq_operations kprobes_sops
= {
2831 .start
= kprobe_seq_start
,
2832 .next
= kprobe_seq_next
,
2833 .stop
= kprobe_seq_stop
,
2834 .show
= show_kprobe_addr
2837 DEFINE_SEQ_ATTRIBUTE(kprobes
);
2839 /* kprobes/blacklist -- shows which functions can not be probed */
2840 static void *kprobe_blacklist_seq_start(struct seq_file
*m
, loff_t
*pos
)
2842 mutex_lock(&kprobe_mutex
);
2843 return seq_list_start(&kprobe_blacklist
, *pos
);
2846 static void *kprobe_blacklist_seq_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
2848 return seq_list_next(v
, &kprobe_blacklist
, pos
);
2851 static int kprobe_blacklist_seq_show(struct seq_file
*m
, void *v
)
2853 struct kprobe_blacklist_entry
*ent
=
2854 list_entry(v
, struct kprobe_blacklist_entry
, list
);
2857 * If '/proc/kallsyms' is not showing kernel address, we won't
2858 * show them here either.
2860 if (!kallsyms_show_value(m
->file
->f_cred
))
2861 seq_printf(m
, "0x%px-0x%px\t%ps\n", NULL
, NULL
,
2862 (void *)ent
->start_addr
);
2864 seq_printf(m
, "0x%px-0x%px\t%ps\n", (void *)ent
->start_addr
,
2865 (void *)ent
->end_addr
, (void *)ent
->start_addr
);
2869 static void kprobe_blacklist_seq_stop(struct seq_file
*f
, void *v
)
2871 mutex_unlock(&kprobe_mutex
);
2874 static const struct seq_operations kprobe_blacklist_sops
= {
2875 .start
= kprobe_blacklist_seq_start
,
2876 .next
= kprobe_blacklist_seq_next
,
2877 .stop
= kprobe_blacklist_seq_stop
,
2878 .show
= kprobe_blacklist_seq_show
,
2880 DEFINE_SEQ_ATTRIBUTE(kprobe_blacklist
);
2882 static int arm_all_kprobes(void)
2884 struct hlist_head
*head
;
2886 unsigned int i
, total
= 0, errors
= 0;
2889 mutex_lock(&kprobe_mutex
);
2891 /* If kprobes are armed, just return */
2892 if (!kprobes_all_disarmed
)
2893 goto already_enabled
;
2896 * optimize_kprobe() called by arm_kprobe() checks
2897 * kprobes_all_disarmed, so set kprobes_all_disarmed before
2900 kprobes_all_disarmed
= false;
2901 /* Arming kprobes doesn't optimize kprobe itself */
2902 for (i
= 0; i
< KPROBE_TABLE_SIZE
; i
++) {
2903 head
= &kprobe_table
[i
];
2904 /* Arm all kprobes on a best-effort basis */
2905 hlist_for_each_entry(p
, head
, hlist
) {
2906 if (!kprobe_disabled(p
)) {
2907 err
= arm_kprobe(p
);
2918 pr_warn("Kprobes globally enabled, but failed to enable %d out of %d probes. Please check which kprobes are kept disabled via debugfs.\n",
2921 pr_info("Kprobes globally enabled\n");
2924 mutex_unlock(&kprobe_mutex
);
2928 static int disarm_all_kprobes(void)
2930 struct hlist_head
*head
;
2932 unsigned int i
, total
= 0, errors
= 0;
2935 mutex_lock(&kprobe_mutex
);
2937 /* If kprobes are already disarmed, just return */
2938 if (kprobes_all_disarmed
) {
2939 mutex_unlock(&kprobe_mutex
);
2943 kprobes_all_disarmed
= true;
2945 for (i
= 0; i
< KPROBE_TABLE_SIZE
; i
++) {
2946 head
= &kprobe_table
[i
];
2947 /* Disarm all kprobes on a best-effort basis */
2948 hlist_for_each_entry(p
, head
, hlist
) {
2949 if (!arch_trampoline_kprobe(p
) && !kprobe_disabled(p
)) {
2950 err
= disarm_kprobe(p
, false);
2961 pr_warn("Kprobes globally disabled, but failed to disable %d out of %d probes. Please check which kprobes are kept enabled via debugfs.\n",
2964 pr_info("Kprobes globally disabled\n");
2966 mutex_unlock(&kprobe_mutex
);
2968 /* Wait for disarming all kprobes by optimizer */
2969 wait_for_kprobe_optimizer();
2975 * XXX: The debugfs bool file interface doesn't allow for callbacks
2976 * when the bool state is switched. We can reuse that facility when
2979 static ssize_t
read_enabled_file_bool(struct file
*file
,
2980 char __user
*user_buf
, size_t count
, loff_t
*ppos
)
2984 if (!kprobes_all_disarmed
)
2990 return simple_read_from_buffer(user_buf
, count
, ppos
, buf
, 2);
2993 static ssize_t
write_enabled_file_bool(struct file
*file
,
2994 const char __user
*user_buf
, size_t count
, loff_t
*ppos
)
2999 ret
= kstrtobool_from_user(user_buf
, count
, &enable
);
3003 ret
= enable
? arm_all_kprobes() : disarm_all_kprobes();
3010 static const struct file_operations fops_kp
= {
3011 .read
= read_enabled_file_bool
,
3012 .write
= write_enabled_file_bool
,
3013 .llseek
= default_llseek
,
3016 static int __init
debugfs_kprobe_init(void)
3020 dir
= debugfs_create_dir("kprobes", NULL
);
3022 debugfs_create_file("list", 0400, dir
, NULL
, &kprobes_fops
);
3024 debugfs_create_file("enabled", 0600, dir
, NULL
, &fops_kp
);
3026 debugfs_create_file("blacklist", 0400, dir
, NULL
,
3027 &kprobe_blacklist_fops
);
3032 late_initcall(debugfs_kprobe_init
);
3033 #endif /* CONFIG_DEBUG_FS */