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/static_call.h>
39 #include <linux/perf_event.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
];
58 /* NOTE: change this value only with kprobe_mutex held */
59 static bool kprobes_all_disarmed
;
61 /* This protects kprobe_table and optimizing_list */
62 static DEFINE_MUTEX(kprobe_mutex
);
63 static DEFINE_PER_CPU(struct kprobe
*, kprobe_instance
) = NULL
;
65 kprobe_opcode_t
* __weak
kprobe_lookup_name(const char *name
,
66 unsigned int __unused
)
68 return ((kprobe_opcode_t
*)(kallsyms_lookup_name(name
)));
71 /* Blacklist -- list of struct kprobe_blacklist_entry */
72 static LIST_HEAD(kprobe_blacklist
);
74 #ifdef __ARCH_WANT_KPROBES_INSN_SLOT
76 * kprobe->ainsn.insn points to the copy of the instruction to be
77 * single-stepped. x86_64, POWER4 and above have no-exec support and
78 * stepping on the instruction on a vmalloced/kmalloced/data page
79 * is a recipe for disaster
81 struct kprobe_insn_page
{
82 struct list_head list
;
83 kprobe_opcode_t
*insns
; /* Page of instruction slots */
84 struct kprobe_insn_cache
*cache
;
90 #define KPROBE_INSN_PAGE_SIZE(slots) \
91 (offsetof(struct kprobe_insn_page, slot_used) + \
92 (sizeof(char) * (slots)))
94 static int slots_per_page(struct kprobe_insn_cache
*c
)
96 return PAGE_SIZE
/(c
->insn_size
* sizeof(kprobe_opcode_t
));
99 enum kprobe_slot_state
{
105 void __weak
*alloc_insn_page(void)
107 return module_alloc(PAGE_SIZE
);
110 static void free_insn_page(void *page
)
112 module_memfree(page
);
115 struct kprobe_insn_cache kprobe_insn_slots
= {
116 .mutex
= __MUTEX_INITIALIZER(kprobe_insn_slots
.mutex
),
117 .alloc
= alloc_insn_page
,
118 .free
= free_insn_page
,
119 .sym
= KPROBE_INSN_PAGE_SYM
,
120 .pages
= LIST_HEAD_INIT(kprobe_insn_slots
.pages
),
121 .insn_size
= MAX_INSN_SIZE
,
124 static int collect_garbage_slots(struct kprobe_insn_cache
*c
);
127 * __get_insn_slot() - Find a slot on an executable page for an instruction.
128 * We allocate an executable page if there's no room on existing ones.
130 kprobe_opcode_t
*__get_insn_slot(struct kprobe_insn_cache
*c
)
132 struct kprobe_insn_page
*kip
;
133 kprobe_opcode_t
*slot
= NULL
;
135 /* Since the slot array is not protected by rcu, we need a mutex */
136 mutex_lock(&c
->mutex
);
139 list_for_each_entry_rcu(kip
, &c
->pages
, list
) {
140 if (kip
->nused
< slots_per_page(c
)) {
142 for (i
= 0; i
< slots_per_page(c
); i
++) {
143 if (kip
->slot_used
[i
] == SLOT_CLEAN
) {
144 kip
->slot_used
[i
] = SLOT_USED
;
146 slot
= kip
->insns
+ (i
* c
->insn_size
);
151 /* kip->nused is broken. Fix it. */
152 kip
->nused
= slots_per_page(c
);
158 /* If there are any garbage slots, collect it and try again. */
159 if (c
->nr_garbage
&& collect_garbage_slots(c
) == 0)
162 /* All out of space. Need to allocate a new page. */
163 kip
= kmalloc(KPROBE_INSN_PAGE_SIZE(slots_per_page(c
)), GFP_KERNEL
);
168 * Use module_alloc so this page is within +/- 2GB of where the
169 * kernel image and loaded module images reside. This is required
170 * so x86_64 can correctly handle the %rip-relative fixups.
172 kip
->insns
= c
->alloc();
177 INIT_LIST_HEAD(&kip
->list
);
178 memset(kip
->slot_used
, SLOT_CLEAN
, slots_per_page(c
));
179 kip
->slot_used
[0] = SLOT_USED
;
183 list_add_rcu(&kip
->list
, &c
->pages
);
186 /* Record the perf ksymbol register event after adding the page */
187 perf_event_ksymbol(PERF_RECORD_KSYMBOL_TYPE_OOL
, (unsigned long)kip
->insns
,
188 PAGE_SIZE
, false, c
->sym
);
190 mutex_unlock(&c
->mutex
);
194 /* Return 1 if all garbages are collected, otherwise 0. */
195 static int collect_one_slot(struct kprobe_insn_page
*kip
, int idx
)
197 kip
->slot_used
[idx
] = SLOT_CLEAN
;
199 if (kip
->nused
== 0) {
201 * Page is no longer in use. Free it unless
202 * it's the last one. We keep the last one
203 * so as not to have to set it up again the
204 * next time somebody inserts a probe.
206 if (!list_is_singular(&kip
->list
)) {
208 * Record perf ksymbol unregister event before removing
211 perf_event_ksymbol(PERF_RECORD_KSYMBOL_TYPE_OOL
,
212 (unsigned long)kip
->insns
, PAGE_SIZE
, true,
214 list_del_rcu(&kip
->list
);
216 kip
->cache
->free(kip
->insns
);
224 static int collect_garbage_slots(struct kprobe_insn_cache
*c
)
226 struct kprobe_insn_page
*kip
, *next
;
228 /* Ensure no-one is interrupted on the garbages */
231 list_for_each_entry_safe(kip
, next
, &c
->pages
, list
) {
233 if (kip
->ngarbage
== 0)
235 kip
->ngarbage
= 0; /* we will collect all garbages */
236 for (i
= 0; i
< slots_per_page(c
); i
++) {
237 if (kip
->slot_used
[i
] == SLOT_DIRTY
&& collect_one_slot(kip
, i
))
245 void __free_insn_slot(struct kprobe_insn_cache
*c
,
246 kprobe_opcode_t
*slot
, int dirty
)
248 struct kprobe_insn_page
*kip
;
251 mutex_lock(&c
->mutex
);
253 list_for_each_entry_rcu(kip
, &c
->pages
, list
) {
254 idx
= ((long)slot
- (long)kip
->insns
) /
255 (c
->insn_size
* sizeof(kprobe_opcode_t
));
256 if (idx
>= 0 && idx
< slots_per_page(c
))
259 /* Could not find this slot. */
264 /* Mark and sweep: this may sleep */
266 /* Check double free */
267 WARN_ON(kip
->slot_used
[idx
] != SLOT_USED
);
269 kip
->slot_used
[idx
] = SLOT_DIRTY
;
271 if (++c
->nr_garbage
> slots_per_page(c
))
272 collect_garbage_slots(c
);
274 collect_one_slot(kip
, idx
);
277 mutex_unlock(&c
->mutex
);
281 * Check given address is on the page of kprobe instruction slots.
282 * This will be used for checking whether the address on a stack
283 * is on a text area or not.
285 bool __is_insn_slot_addr(struct kprobe_insn_cache
*c
, unsigned long addr
)
287 struct kprobe_insn_page
*kip
;
291 list_for_each_entry_rcu(kip
, &c
->pages
, list
) {
292 if (addr
>= (unsigned long)kip
->insns
&&
293 addr
< (unsigned long)kip
->insns
+ PAGE_SIZE
) {
303 int kprobe_cache_get_kallsym(struct kprobe_insn_cache
*c
, unsigned int *symnum
,
304 unsigned long *value
, char *type
, char *sym
)
306 struct kprobe_insn_page
*kip
;
310 list_for_each_entry_rcu(kip
, &c
->pages
, list
) {
313 strlcpy(sym
, c
->sym
, KSYM_NAME_LEN
);
315 *value
= (unsigned long)kip
->insns
;
324 #ifdef CONFIG_OPTPROBES
325 void __weak
*alloc_optinsn_page(void)
327 return alloc_insn_page();
330 void __weak
free_optinsn_page(void *page
)
332 free_insn_page(page
);
335 /* For optimized_kprobe buffer */
336 struct kprobe_insn_cache kprobe_optinsn_slots
= {
337 .mutex
= __MUTEX_INITIALIZER(kprobe_optinsn_slots
.mutex
),
338 .alloc
= alloc_optinsn_page
,
339 .free
= free_optinsn_page
,
340 .sym
= KPROBE_OPTINSN_PAGE_SYM
,
341 .pages
= LIST_HEAD_INIT(kprobe_optinsn_slots
.pages
),
342 /* .insn_size is initialized later */
348 /* We have preemption disabled.. so it is safe to use __ versions */
349 static inline void set_kprobe_instance(struct kprobe
*kp
)
351 __this_cpu_write(kprobe_instance
, kp
);
354 static inline void reset_kprobe_instance(void)
356 __this_cpu_write(kprobe_instance
, NULL
);
360 * This routine is called either:
361 * - under the kprobe_mutex - during kprobe_[un]register()
363 * - with preemption disabled - from arch/xxx/kernel/kprobes.c
365 struct kprobe
*get_kprobe(void *addr
)
367 struct hlist_head
*head
;
370 head
= &kprobe_table
[hash_ptr(addr
, KPROBE_HASH_BITS
)];
371 hlist_for_each_entry_rcu(p
, head
, hlist
,
372 lockdep_is_held(&kprobe_mutex
)) {
379 NOKPROBE_SYMBOL(get_kprobe
);
381 static int aggr_pre_handler(struct kprobe
*p
, struct pt_regs
*regs
);
383 /* Return true if the kprobe is an aggregator */
384 static inline int kprobe_aggrprobe(struct kprobe
*p
)
386 return p
->pre_handler
== aggr_pre_handler
;
389 /* Return true(!0) if the kprobe is unused */
390 static inline int kprobe_unused(struct kprobe
*p
)
392 return kprobe_aggrprobe(p
) && kprobe_disabled(p
) &&
393 list_empty(&p
->list
);
397 * Keep all fields in the kprobe consistent
399 static inline void copy_kprobe(struct kprobe
*ap
, struct kprobe
*p
)
401 memcpy(&p
->opcode
, &ap
->opcode
, sizeof(kprobe_opcode_t
));
402 memcpy(&p
->ainsn
, &ap
->ainsn
, sizeof(struct arch_specific_insn
));
405 #ifdef CONFIG_OPTPROBES
406 /* NOTE: change this value only with kprobe_mutex held */
407 static bool kprobes_allow_optimization
;
410 * Call all pre_handler on the list, but ignores its return value.
411 * This must be called from arch-dep optimized caller.
413 void opt_pre_handler(struct kprobe
*p
, struct pt_regs
*regs
)
417 list_for_each_entry_rcu(kp
, &p
->list
, list
) {
418 if (kp
->pre_handler
&& likely(!kprobe_disabled(kp
))) {
419 set_kprobe_instance(kp
);
420 kp
->pre_handler(kp
, regs
);
422 reset_kprobe_instance();
425 NOKPROBE_SYMBOL(opt_pre_handler
);
427 /* Free optimized instructions and optimized_kprobe */
428 static void free_aggr_kprobe(struct kprobe
*p
)
430 struct optimized_kprobe
*op
;
432 op
= container_of(p
, struct optimized_kprobe
, kp
);
433 arch_remove_optimized_kprobe(op
);
434 arch_remove_kprobe(p
);
438 /* Return true(!0) if the kprobe is ready for optimization. */
439 static inline int kprobe_optready(struct kprobe
*p
)
441 struct optimized_kprobe
*op
;
443 if (kprobe_aggrprobe(p
)) {
444 op
= container_of(p
, struct optimized_kprobe
, kp
);
445 return arch_prepared_optinsn(&op
->optinsn
);
451 /* Return true(!0) if the kprobe is disarmed. Note: p must be on hash list */
452 static inline int kprobe_disarmed(struct kprobe
*p
)
454 struct optimized_kprobe
*op
;
456 /* If kprobe is not aggr/opt probe, just return kprobe is disabled */
457 if (!kprobe_aggrprobe(p
))
458 return kprobe_disabled(p
);
460 op
= container_of(p
, struct optimized_kprobe
, kp
);
462 return kprobe_disabled(p
) && list_empty(&op
->list
);
465 /* Return true(!0) if the probe is queued on (un)optimizing lists */
466 static int kprobe_queued(struct kprobe
*p
)
468 struct optimized_kprobe
*op
;
470 if (kprobe_aggrprobe(p
)) {
471 op
= container_of(p
, struct optimized_kprobe
, kp
);
472 if (!list_empty(&op
->list
))
479 * Return an optimized kprobe whose optimizing code replaces
480 * instructions including addr (exclude breakpoint).
482 static struct kprobe
*get_optimized_kprobe(unsigned long addr
)
485 struct kprobe
*p
= NULL
;
486 struct optimized_kprobe
*op
;
488 /* Don't check i == 0, since that is a breakpoint case. */
489 for (i
= 1; !p
&& i
< MAX_OPTIMIZED_LENGTH
; i
++)
490 p
= get_kprobe((void *)(addr
- i
));
492 if (p
&& kprobe_optready(p
)) {
493 op
= container_of(p
, struct optimized_kprobe
, kp
);
494 if (arch_within_optimized_kprobe(op
, addr
))
501 /* Optimization staging list, protected by kprobe_mutex */
502 static LIST_HEAD(optimizing_list
);
503 static LIST_HEAD(unoptimizing_list
);
504 static LIST_HEAD(freeing_list
);
506 static void kprobe_optimizer(struct work_struct
*work
);
507 static DECLARE_DELAYED_WORK(optimizing_work
, kprobe_optimizer
);
508 #define OPTIMIZE_DELAY 5
511 * Optimize (replace a breakpoint with a jump) kprobes listed on
514 static void do_optimize_kprobes(void)
516 lockdep_assert_held(&text_mutex
);
518 * The optimization/unoptimization refers online_cpus via
519 * stop_machine() and cpu-hotplug modifies online_cpus.
520 * And same time, text_mutex will be held in cpu-hotplug and here.
521 * This combination can cause a deadlock (cpu-hotplug try to lock
522 * text_mutex but stop_machine can not be done because online_cpus
524 * To avoid this deadlock, caller must have locked cpu hotplug
525 * for preventing cpu-hotplug outside of text_mutex locking.
527 lockdep_assert_cpus_held();
529 /* Optimization never be done when disarmed */
530 if (kprobes_all_disarmed
|| !kprobes_allow_optimization
||
531 list_empty(&optimizing_list
))
534 arch_optimize_kprobes(&optimizing_list
);
538 * Unoptimize (replace a jump with a breakpoint and remove the breakpoint
539 * if need) kprobes listed on unoptimizing_list.
541 static void do_unoptimize_kprobes(void)
543 struct optimized_kprobe
*op
, *tmp
;
545 lockdep_assert_held(&text_mutex
);
546 /* See comment in do_optimize_kprobes() */
547 lockdep_assert_cpus_held();
549 /* Unoptimization must be done anytime */
550 if (list_empty(&unoptimizing_list
))
553 arch_unoptimize_kprobes(&unoptimizing_list
, &freeing_list
);
554 /* Loop free_list for disarming */
555 list_for_each_entry_safe(op
, tmp
, &freeing_list
, list
) {
556 /* Switching from detour code to origin */
557 op
->kp
.flags
&= ~KPROBE_FLAG_OPTIMIZED
;
558 /* Disarm probes if marked disabled */
559 if (kprobe_disabled(&op
->kp
))
560 arch_disarm_kprobe(&op
->kp
);
561 if (kprobe_unused(&op
->kp
)) {
563 * Remove unused probes from hash list. After waiting
564 * for synchronization, these probes are reclaimed.
565 * (reclaiming is done by do_free_cleaned_kprobes.)
567 hlist_del_rcu(&op
->kp
.hlist
);
569 list_del_init(&op
->list
);
573 /* Reclaim all kprobes on the free_list */
574 static void do_free_cleaned_kprobes(void)
576 struct optimized_kprobe
*op
, *tmp
;
578 list_for_each_entry_safe(op
, tmp
, &freeing_list
, list
) {
579 list_del_init(&op
->list
);
580 if (WARN_ON_ONCE(!kprobe_unused(&op
->kp
))) {
582 * This must not happen, but if there is a kprobe
583 * still in use, keep it on kprobes hash list.
587 free_aggr_kprobe(&op
->kp
);
591 /* Start optimizer after OPTIMIZE_DELAY passed */
592 static void kick_kprobe_optimizer(void)
594 schedule_delayed_work(&optimizing_work
, OPTIMIZE_DELAY
);
597 /* Kprobe jump optimizer */
598 static void kprobe_optimizer(struct work_struct
*work
)
600 mutex_lock(&kprobe_mutex
);
602 mutex_lock(&text_mutex
);
605 * Step 1: Unoptimize kprobes and collect cleaned (unused and disarmed)
606 * kprobes before waiting for quiesence period.
608 do_unoptimize_kprobes();
611 * Step 2: Wait for quiesence period to ensure all potentially
612 * preempted tasks to have normally scheduled. Because optprobe
613 * may modify multiple instructions, there is a chance that Nth
614 * instruction is preempted. In that case, such tasks can return
615 * to 2nd-Nth byte of jump instruction. This wait is for avoiding it.
616 * Note that on non-preemptive kernel, this is transparently converted
617 * to synchronoze_sched() to wait for all interrupts to have completed.
619 synchronize_rcu_tasks();
621 /* Step 3: Optimize kprobes after quiesence period */
622 do_optimize_kprobes();
624 /* Step 4: Free cleaned kprobes after quiesence period */
625 do_free_cleaned_kprobes();
627 mutex_unlock(&text_mutex
);
630 /* Step 5: Kick optimizer again if needed */
631 if (!list_empty(&optimizing_list
) || !list_empty(&unoptimizing_list
))
632 kick_kprobe_optimizer();
634 mutex_unlock(&kprobe_mutex
);
637 /* Wait for completing optimization and unoptimization */
638 void wait_for_kprobe_optimizer(void)
640 mutex_lock(&kprobe_mutex
);
642 while (!list_empty(&optimizing_list
) || !list_empty(&unoptimizing_list
)) {
643 mutex_unlock(&kprobe_mutex
);
645 /* this will also make optimizing_work execute immmediately */
646 flush_delayed_work(&optimizing_work
);
647 /* @optimizing_work might not have been queued yet, relax */
650 mutex_lock(&kprobe_mutex
);
653 mutex_unlock(&kprobe_mutex
);
656 static bool optprobe_queued_unopt(struct optimized_kprobe
*op
)
658 struct optimized_kprobe
*_op
;
660 list_for_each_entry(_op
, &unoptimizing_list
, list
) {
668 /* Optimize kprobe if p is ready to be optimized */
669 static void optimize_kprobe(struct kprobe
*p
)
671 struct optimized_kprobe
*op
;
673 /* Check if the kprobe is disabled or not ready for optimization. */
674 if (!kprobe_optready(p
) || !kprobes_allow_optimization
||
675 (kprobe_disabled(p
) || kprobes_all_disarmed
))
678 /* kprobes with post_handler can not be optimized */
682 op
= container_of(p
, struct optimized_kprobe
, kp
);
684 /* Check there is no other kprobes at the optimized instructions */
685 if (arch_check_optimized_kprobe(op
) < 0)
688 /* Check if it is already optimized. */
689 if (op
->kp
.flags
& KPROBE_FLAG_OPTIMIZED
) {
690 if (optprobe_queued_unopt(op
)) {
691 /* This is under unoptimizing. Just dequeue the probe */
692 list_del_init(&op
->list
);
696 op
->kp
.flags
|= KPROBE_FLAG_OPTIMIZED
;
698 /* On unoptimizing/optimizing_list, op must have OPTIMIZED flag */
699 if (WARN_ON_ONCE(!list_empty(&op
->list
)))
702 list_add(&op
->list
, &optimizing_list
);
703 kick_kprobe_optimizer();
706 /* Short cut to direct unoptimizing */
707 static void force_unoptimize_kprobe(struct optimized_kprobe
*op
)
709 lockdep_assert_cpus_held();
710 arch_unoptimize_kprobe(op
);
711 op
->kp
.flags
&= ~KPROBE_FLAG_OPTIMIZED
;
714 /* Unoptimize a kprobe if p is optimized */
715 static void unoptimize_kprobe(struct kprobe
*p
, bool force
)
717 struct optimized_kprobe
*op
;
719 if (!kprobe_aggrprobe(p
) || kprobe_disarmed(p
))
720 return; /* This is not an optprobe nor optimized */
722 op
= container_of(p
, struct optimized_kprobe
, kp
);
723 if (!kprobe_optimized(p
))
726 if (!list_empty(&op
->list
)) {
727 if (optprobe_queued_unopt(op
)) {
728 /* Queued in unoptimizing queue */
731 * Forcibly unoptimize the kprobe here, and queue it
732 * in the freeing list for release afterwards.
734 force_unoptimize_kprobe(op
);
735 list_move(&op
->list
, &freeing_list
);
738 /* Dequeue from the optimizing queue */
739 list_del_init(&op
->list
);
740 op
->kp
.flags
&= ~KPROBE_FLAG_OPTIMIZED
;
745 /* Optimized kprobe case */
747 /* Forcibly update the code: this is a special case */
748 force_unoptimize_kprobe(op
);
750 list_add(&op
->list
, &unoptimizing_list
);
751 kick_kprobe_optimizer();
755 /* Cancel unoptimizing for reusing */
756 static int reuse_unused_kprobe(struct kprobe
*ap
)
758 struct optimized_kprobe
*op
;
761 * Unused kprobe MUST be on the way of delayed unoptimizing (means
762 * there is still a relative jump) and disabled.
764 op
= container_of(ap
, struct optimized_kprobe
, kp
);
765 WARN_ON_ONCE(list_empty(&op
->list
));
766 /* Enable the probe again */
767 ap
->flags
&= ~KPROBE_FLAG_DISABLED
;
768 /* Optimize it again (remove from op->list) */
769 if (!kprobe_optready(ap
))
776 /* Remove optimized instructions */
777 static void kill_optimized_kprobe(struct kprobe
*p
)
779 struct optimized_kprobe
*op
;
781 op
= container_of(p
, struct optimized_kprobe
, kp
);
782 if (!list_empty(&op
->list
))
783 /* Dequeue from the (un)optimization queue */
784 list_del_init(&op
->list
);
785 op
->kp
.flags
&= ~KPROBE_FLAG_OPTIMIZED
;
787 if (kprobe_unused(p
)) {
788 /* Enqueue if it is unused */
789 list_add(&op
->list
, &freeing_list
);
791 * Remove unused probes from the hash list. After waiting
792 * for synchronization, this probe is reclaimed.
793 * (reclaiming is done by do_free_cleaned_kprobes().)
795 hlist_del_rcu(&op
->kp
.hlist
);
798 /* Don't touch the code, because it is already freed. */
799 arch_remove_optimized_kprobe(op
);
803 void __prepare_optimized_kprobe(struct optimized_kprobe
*op
, struct kprobe
*p
)
805 if (!kprobe_ftrace(p
))
806 arch_prepare_optimized_kprobe(op
, p
);
809 /* Try to prepare optimized instructions */
810 static void prepare_optimized_kprobe(struct kprobe
*p
)
812 struct optimized_kprobe
*op
;
814 op
= container_of(p
, struct optimized_kprobe
, kp
);
815 __prepare_optimized_kprobe(op
, p
);
818 /* Allocate new optimized_kprobe and try to prepare optimized instructions */
819 static struct kprobe
*alloc_aggr_kprobe(struct kprobe
*p
)
821 struct optimized_kprobe
*op
;
823 op
= kzalloc(sizeof(struct optimized_kprobe
), GFP_KERNEL
);
827 INIT_LIST_HEAD(&op
->list
);
828 op
->kp
.addr
= p
->addr
;
829 __prepare_optimized_kprobe(op
, p
);
834 static void init_aggr_kprobe(struct kprobe
*ap
, struct kprobe
*p
);
837 * Prepare an optimized_kprobe and optimize it
838 * NOTE: p must be a normal registered kprobe
840 static void try_to_optimize_kprobe(struct kprobe
*p
)
843 struct optimized_kprobe
*op
;
845 /* Impossible to optimize ftrace-based kprobe */
846 if (kprobe_ftrace(p
))
849 /* For preparing optimization, jump_label_text_reserved() is called */
852 mutex_lock(&text_mutex
);
854 ap
= alloc_aggr_kprobe(p
);
858 op
= container_of(ap
, struct optimized_kprobe
, kp
);
859 if (!arch_prepared_optinsn(&op
->optinsn
)) {
860 /* If failed to setup optimizing, fallback to kprobe */
861 arch_remove_optimized_kprobe(op
);
866 init_aggr_kprobe(ap
, p
);
867 optimize_kprobe(ap
); /* This just kicks optimizer thread */
870 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
);
901 static void unoptimize_all_kprobes(void)
903 struct hlist_head
*head
;
907 mutex_lock(&kprobe_mutex
);
908 /* If optimization is already prohibited, just return */
909 if (!kprobes_allow_optimization
) {
910 mutex_unlock(&kprobe_mutex
);
915 kprobes_allow_optimization
= false;
916 for (i
= 0; i
< KPROBE_TABLE_SIZE
; i
++) {
917 head
= &kprobe_table
[i
];
918 hlist_for_each_entry(p
, head
, hlist
) {
919 if (!kprobe_disabled(p
))
920 unoptimize_kprobe(p
, false);
924 mutex_unlock(&kprobe_mutex
);
926 /* Wait for unoptimizing completion */
927 wait_for_kprobe_optimizer();
928 printk(KERN_INFO
"Kprobes globally unoptimized\n");
931 static DEFINE_MUTEX(kprobe_sysctl_mutex
);
932 int sysctl_kprobes_optimization
;
933 int proc_kprobes_optimization_handler(struct ctl_table
*table
, int write
,
934 void *buffer
, size_t *length
,
939 mutex_lock(&kprobe_sysctl_mutex
);
940 sysctl_kprobes_optimization
= kprobes_allow_optimization
? 1 : 0;
941 ret
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
943 if (sysctl_kprobes_optimization
)
944 optimize_all_kprobes();
946 unoptimize_all_kprobes();
947 mutex_unlock(&kprobe_sysctl_mutex
);
951 #endif /* CONFIG_SYSCTL */
953 /* Put a breakpoint for a probe. Must be called with text_mutex locked */
954 static void __arm_kprobe(struct kprobe
*p
)
958 /* Check collision with other optimized kprobes */
959 _p
= get_optimized_kprobe((unsigned long)p
->addr
);
961 /* Fallback to unoptimized kprobe */
962 unoptimize_kprobe(_p
, true);
965 optimize_kprobe(p
); /* Try to optimize (add kprobe to a list) */
968 /* Remove the breakpoint of a probe. Must be called with text_mutex locked */
969 static void __disarm_kprobe(struct kprobe
*p
, bool reopt
)
973 /* Try to unoptimize */
974 unoptimize_kprobe(p
, kprobes_all_disarmed
);
976 if (!kprobe_queued(p
)) {
977 arch_disarm_kprobe(p
);
978 /* If another kprobe was blocked, optimize it. */
979 _p
= get_optimized_kprobe((unsigned long)p
->addr
);
980 if (unlikely(_p
) && reopt
)
983 /* TODO: reoptimize others after unoptimized this probe */
986 #else /* !CONFIG_OPTPROBES */
988 #define optimize_kprobe(p) do {} while (0)
989 #define unoptimize_kprobe(p, f) do {} while (0)
990 #define kill_optimized_kprobe(p) do {} while (0)
991 #define prepare_optimized_kprobe(p) do {} while (0)
992 #define try_to_optimize_kprobe(p) do {} while (0)
993 #define __arm_kprobe(p) arch_arm_kprobe(p)
994 #define __disarm_kprobe(p, o) arch_disarm_kprobe(p)
995 #define kprobe_disarmed(p) kprobe_disabled(p)
996 #define wait_for_kprobe_optimizer() do {} while (0)
998 static int reuse_unused_kprobe(struct kprobe
*ap
)
1001 * If the optimized kprobe is NOT supported, the aggr kprobe is
1002 * released at the same time that the last aggregated kprobe is
1004 * Thus there should be no chance to reuse unused kprobe.
1006 printk(KERN_ERR
"Error: There should be no unused kprobe here.\n");
1010 static void free_aggr_kprobe(struct kprobe
*p
)
1012 arch_remove_kprobe(p
);
1016 static struct kprobe
*alloc_aggr_kprobe(struct kprobe
*p
)
1018 return kzalloc(sizeof(struct kprobe
), GFP_KERNEL
);
1020 #endif /* CONFIG_OPTPROBES */
1022 #ifdef CONFIG_KPROBES_ON_FTRACE
1023 static struct ftrace_ops kprobe_ftrace_ops __read_mostly
= {
1024 .func
= kprobe_ftrace_handler
,
1025 .flags
= FTRACE_OPS_FL_SAVE_REGS
,
1028 static struct ftrace_ops kprobe_ipmodify_ops __read_mostly
= {
1029 .func
= kprobe_ftrace_handler
,
1030 .flags
= FTRACE_OPS_FL_SAVE_REGS
| FTRACE_OPS_FL_IPMODIFY
,
1033 static int kprobe_ipmodify_enabled
;
1034 static int kprobe_ftrace_enabled
;
1036 /* Must ensure p->addr is really on ftrace */
1037 static int prepare_kprobe(struct kprobe
*p
)
1039 if (!kprobe_ftrace(p
))
1040 return arch_prepare_kprobe(p
);
1042 return arch_prepare_kprobe_ftrace(p
);
1045 /* Caller must lock kprobe_mutex */
1046 static int __arm_kprobe_ftrace(struct kprobe
*p
, struct ftrace_ops
*ops
,
1051 ret
= ftrace_set_filter_ip(ops
, (unsigned long)p
->addr
, 0, 0);
1053 pr_debug("Failed to arm kprobe-ftrace at %pS (%d)\n",
1059 ret
= register_ftrace_function(ops
);
1061 pr_debug("Failed to init kprobe-ftrace (%d)\n", ret
);
1071 * At this point, sinec ops is not registered, we should be sefe from
1072 * registering empty filter.
1074 ftrace_set_filter_ip(ops
, (unsigned long)p
->addr
, 1, 0);
1078 static int arm_kprobe_ftrace(struct kprobe
*p
)
1080 bool ipmodify
= (p
->post_handler
!= NULL
);
1082 return __arm_kprobe_ftrace(p
,
1083 ipmodify
? &kprobe_ipmodify_ops
: &kprobe_ftrace_ops
,
1084 ipmodify
? &kprobe_ipmodify_enabled
: &kprobe_ftrace_enabled
);
1087 /* Caller must lock kprobe_mutex */
1088 static int __disarm_kprobe_ftrace(struct kprobe
*p
, struct ftrace_ops
*ops
,
1094 ret
= unregister_ftrace_function(ops
);
1095 if (WARN(ret
< 0, "Failed to unregister kprobe-ftrace (%d)\n", ret
))
1101 ret
= ftrace_set_filter_ip(ops
, (unsigned long)p
->addr
, 1, 0);
1102 WARN_ONCE(ret
< 0, "Failed to disarm kprobe-ftrace at %pS (%d)\n",
1107 static int disarm_kprobe_ftrace(struct kprobe
*p
)
1109 bool ipmodify
= (p
->post_handler
!= NULL
);
1111 return __disarm_kprobe_ftrace(p
,
1112 ipmodify
? &kprobe_ipmodify_ops
: &kprobe_ftrace_ops
,
1113 ipmodify
? &kprobe_ipmodify_enabled
: &kprobe_ftrace_enabled
);
1115 #else /* !CONFIG_KPROBES_ON_FTRACE */
1116 static inline int prepare_kprobe(struct kprobe
*p
)
1118 return arch_prepare_kprobe(p
);
1121 static inline int arm_kprobe_ftrace(struct kprobe
*p
)
1126 static inline int disarm_kprobe_ftrace(struct kprobe
*p
)
1132 /* Arm a kprobe with text_mutex */
1133 static int arm_kprobe(struct kprobe
*kp
)
1135 if (unlikely(kprobe_ftrace(kp
)))
1136 return arm_kprobe_ftrace(kp
);
1139 mutex_lock(&text_mutex
);
1141 mutex_unlock(&text_mutex
);
1147 /* Disarm a kprobe with text_mutex */
1148 static int disarm_kprobe(struct kprobe
*kp
, bool reopt
)
1150 if (unlikely(kprobe_ftrace(kp
)))
1151 return disarm_kprobe_ftrace(kp
);
1154 mutex_lock(&text_mutex
);
1155 __disarm_kprobe(kp
, reopt
);
1156 mutex_unlock(&text_mutex
);
1163 * Aggregate handlers for multiple kprobes support - these handlers
1164 * take care of invoking the individual kprobe handlers on p->list
1166 static int aggr_pre_handler(struct kprobe
*p
, struct pt_regs
*regs
)
1170 list_for_each_entry_rcu(kp
, &p
->list
, list
) {
1171 if (kp
->pre_handler
&& likely(!kprobe_disabled(kp
))) {
1172 set_kprobe_instance(kp
);
1173 if (kp
->pre_handler(kp
, regs
))
1176 reset_kprobe_instance();
1180 NOKPROBE_SYMBOL(aggr_pre_handler
);
1182 static void aggr_post_handler(struct kprobe
*p
, struct pt_regs
*regs
,
1183 unsigned long flags
)
1187 list_for_each_entry_rcu(kp
, &p
->list
, list
) {
1188 if (kp
->post_handler
&& likely(!kprobe_disabled(kp
))) {
1189 set_kprobe_instance(kp
);
1190 kp
->post_handler(kp
, regs
, flags
);
1191 reset_kprobe_instance();
1195 NOKPROBE_SYMBOL(aggr_post_handler
);
1197 /* Walks the list and increments nmissed count for multiprobe case */
1198 void kprobes_inc_nmissed_count(struct kprobe
*p
)
1201 if (!kprobe_aggrprobe(p
)) {
1204 list_for_each_entry_rcu(kp
, &p
->list
, list
)
1209 NOKPROBE_SYMBOL(kprobes_inc_nmissed_count
);
1211 static void free_rp_inst_rcu(struct rcu_head
*head
)
1213 struct kretprobe_instance
*ri
= container_of(head
, struct kretprobe_instance
, rcu
);
1215 if (refcount_dec_and_test(&ri
->rph
->ref
))
1219 NOKPROBE_SYMBOL(free_rp_inst_rcu
);
1221 static void recycle_rp_inst(struct kretprobe_instance
*ri
)
1223 struct kretprobe
*rp
= get_kretprobe(ri
);
1226 freelist_add(&ri
->freelist
, &rp
->freelist
);
1228 call_rcu(&ri
->rcu
, free_rp_inst_rcu
);
1230 NOKPROBE_SYMBOL(recycle_rp_inst
);
1232 static struct kprobe kprobe_busy
= {
1233 .addr
= (void *) get_kprobe
,
1236 void kprobe_busy_begin(void)
1238 struct kprobe_ctlblk
*kcb
;
1241 __this_cpu_write(current_kprobe
, &kprobe_busy
);
1242 kcb
= get_kprobe_ctlblk();
1243 kcb
->kprobe_status
= KPROBE_HIT_ACTIVE
;
1246 void kprobe_busy_end(void)
1248 __this_cpu_write(current_kprobe
, NULL
);
1253 * This function is called from finish_task_switch when task tk becomes dead,
1254 * so that we can recycle any function-return probe instances associated
1255 * with this task. These left over instances represent probed functions
1256 * that have been called but will never return.
1258 void kprobe_flush_task(struct task_struct
*tk
)
1260 struct kretprobe_instance
*ri
;
1261 struct llist_node
*node
;
1263 /* Early boot, not yet initialized. */
1264 if (unlikely(!kprobes_initialized
))
1267 kprobe_busy_begin();
1269 node
= __llist_del_all(&tk
->kretprobe_instances
);
1271 ri
= container_of(node
, struct kretprobe_instance
, llist
);
1274 recycle_rp_inst(ri
);
1279 NOKPROBE_SYMBOL(kprobe_flush_task
);
1281 static inline void free_rp_inst(struct kretprobe
*rp
)
1283 struct kretprobe_instance
*ri
;
1284 struct freelist_node
*node
;
1287 node
= rp
->freelist
.head
;
1289 ri
= container_of(node
, struct kretprobe_instance
, freelist
);
1296 if (refcount_sub_and_test(count
, &rp
->rph
->ref
)) {
1302 /* Add the new probe to ap->list */
1303 static int add_new_kprobe(struct kprobe
*ap
, struct kprobe
*p
)
1305 if (p
->post_handler
)
1306 unoptimize_kprobe(ap
, true); /* Fall back to normal kprobe */
1308 list_add_rcu(&p
->list
, &ap
->list
);
1309 if (p
->post_handler
&& !ap
->post_handler
)
1310 ap
->post_handler
= aggr_post_handler
;
1316 * Fill in the required fields of the "manager kprobe". Replace the
1317 * earlier kprobe in the hlist with the manager kprobe
1319 static void init_aggr_kprobe(struct kprobe
*ap
, struct kprobe
*p
)
1321 /* Copy p's insn slot to ap */
1323 flush_insn_slot(ap
);
1325 ap
->flags
= p
->flags
& ~KPROBE_FLAG_OPTIMIZED
;
1326 ap
->pre_handler
= aggr_pre_handler
;
1327 /* We don't care the kprobe which has gone. */
1328 if (p
->post_handler
&& !kprobe_gone(p
))
1329 ap
->post_handler
= aggr_post_handler
;
1331 INIT_LIST_HEAD(&ap
->list
);
1332 INIT_HLIST_NODE(&ap
->hlist
);
1334 list_add_rcu(&p
->list
, &ap
->list
);
1335 hlist_replace_rcu(&p
->hlist
, &ap
->hlist
);
1339 * This is the second or subsequent kprobe at the address - handle
1342 static int register_aggr_kprobe(struct kprobe
*orig_p
, struct kprobe
*p
)
1345 struct kprobe
*ap
= orig_p
;
1349 /* For preparing optimization, jump_label_text_reserved() is called */
1351 mutex_lock(&text_mutex
);
1353 if (!kprobe_aggrprobe(orig_p
)) {
1354 /* If orig_p is not an aggr_kprobe, create new aggr_kprobe. */
1355 ap
= alloc_aggr_kprobe(orig_p
);
1360 init_aggr_kprobe(ap
, orig_p
);
1361 } else if (kprobe_unused(ap
)) {
1362 /* This probe is going to die. Rescue it */
1363 ret
= reuse_unused_kprobe(ap
);
1368 if (kprobe_gone(ap
)) {
1370 * Attempting to insert new probe at the same location that
1371 * had a probe in the module vaddr area which already
1372 * freed. So, the instruction slot has already been
1373 * released. We need a new slot for the new probe.
1375 ret
= arch_prepare_kprobe(ap
);
1378 * Even if fail to allocate new slot, don't need to
1379 * free aggr_probe. It will be used next time, or
1380 * freed by unregister_kprobe.
1384 /* Prepare optimized instructions if possible. */
1385 prepare_optimized_kprobe(ap
);
1388 * Clear gone flag to prevent allocating new slot again, and
1389 * set disabled flag because it is not armed yet.
1391 ap
->flags
= (ap
->flags
& ~KPROBE_FLAG_GONE
)
1392 | KPROBE_FLAG_DISABLED
;
1395 /* Copy ap's insn slot to p */
1397 ret
= add_new_kprobe(ap
, p
);
1400 mutex_unlock(&text_mutex
);
1401 jump_label_unlock();
1404 if (ret
== 0 && kprobe_disabled(ap
) && !kprobe_disabled(p
)) {
1405 ap
->flags
&= ~KPROBE_FLAG_DISABLED
;
1406 if (!kprobes_all_disarmed
) {
1407 /* Arm the breakpoint again. */
1408 ret
= arm_kprobe(ap
);
1410 ap
->flags
|= KPROBE_FLAG_DISABLED
;
1411 list_del_rcu(&p
->list
);
1419 bool __weak
arch_within_kprobe_blacklist(unsigned long addr
)
1421 /* The __kprobes marked functions and entry code must not be probed */
1422 return addr
>= (unsigned long)__kprobes_text_start
&&
1423 addr
< (unsigned long)__kprobes_text_end
;
1426 static bool __within_kprobe_blacklist(unsigned long addr
)
1428 struct kprobe_blacklist_entry
*ent
;
1430 if (arch_within_kprobe_blacklist(addr
))
1433 * If there exists a kprobe_blacklist, verify and
1434 * fail any probe registration in the prohibited area
1436 list_for_each_entry(ent
, &kprobe_blacklist
, list
) {
1437 if (addr
>= ent
->start_addr
&& addr
< ent
->end_addr
)
1443 bool within_kprobe_blacklist(unsigned long addr
)
1445 char symname
[KSYM_NAME_LEN
], *p
;
1447 if (__within_kprobe_blacklist(addr
))
1450 /* Check if the address is on a suffixed-symbol */
1451 if (!lookup_symbol_name(addr
, symname
)) {
1452 p
= strchr(symname
, '.');
1456 addr
= (unsigned long)kprobe_lookup_name(symname
, 0);
1458 return __within_kprobe_blacklist(addr
);
1464 * If we have a symbol_name argument, look it up and add the offset field
1465 * to it. This way, we can specify a relative address to a symbol.
1466 * This returns encoded errors if it fails to look up symbol or invalid
1467 * combination of parameters.
1469 static kprobe_opcode_t
*_kprobe_addr(kprobe_opcode_t
*addr
,
1470 const char *symbol_name
, unsigned int offset
)
1472 if ((symbol_name
&& addr
) || (!symbol_name
&& !addr
))
1476 addr
= kprobe_lookup_name(symbol_name
, offset
);
1478 return ERR_PTR(-ENOENT
);
1481 addr
= (kprobe_opcode_t
*)(((char *)addr
) + offset
);
1486 return ERR_PTR(-EINVAL
);
1489 static kprobe_opcode_t
*kprobe_addr(struct kprobe
*p
)
1491 return _kprobe_addr(p
->addr
, p
->symbol_name
, p
->offset
);
1494 /* Check passed kprobe is valid and return kprobe in kprobe_table. */
1495 static struct kprobe
*__get_valid_kprobe(struct kprobe
*p
)
1497 struct kprobe
*ap
, *list_p
;
1499 lockdep_assert_held(&kprobe_mutex
);
1501 ap
= get_kprobe(p
->addr
);
1506 list_for_each_entry(list_p
, &ap
->list
, list
)
1508 /* kprobe p is a valid probe */
1517 * Warn and return error if the kprobe is being re-registered since
1518 * there must be a software bug.
1520 static inline int warn_kprobe_rereg(struct kprobe
*p
)
1524 mutex_lock(&kprobe_mutex
);
1525 if (WARN_ON_ONCE(__get_valid_kprobe(p
)))
1527 mutex_unlock(&kprobe_mutex
);
1532 int __weak
arch_check_ftrace_location(struct kprobe
*p
)
1534 unsigned long ftrace_addr
;
1536 ftrace_addr
= ftrace_location((unsigned long)p
->addr
);
1538 #ifdef CONFIG_KPROBES_ON_FTRACE
1539 /* Given address is not on the instruction boundary */
1540 if ((unsigned long)p
->addr
!= ftrace_addr
)
1542 p
->flags
|= KPROBE_FLAG_FTRACE
;
1543 #else /* !CONFIG_KPROBES_ON_FTRACE */
1550 static int check_kprobe_address_safe(struct kprobe
*p
,
1551 struct module
**probed_mod
)
1555 ret
= arch_check_ftrace_location(p
);
1561 /* Ensure it is not in reserved area nor out of text */
1562 if (!kernel_text_address((unsigned long) p
->addr
) ||
1563 within_kprobe_blacklist((unsigned long) p
->addr
) ||
1564 jump_label_text_reserved(p
->addr
, p
->addr
) ||
1565 static_call_text_reserved(p
->addr
, p
->addr
) ||
1566 find_bug((unsigned long)p
->addr
)) {
1571 /* Check if are we probing a module */
1572 *probed_mod
= __module_text_address((unsigned long) p
->addr
);
1575 * We must hold a refcount of the probed module while updating
1576 * its code to prohibit unexpected unloading.
1578 if (unlikely(!try_module_get(*probed_mod
))) {
1584 * If the module freed .init.text, we couldn't insert
1587 if (within_module_init((unsigned long)p
->addr
, *probed_mod
) &&
1588 (*probed_mod
)->state
!= MODULE_STATE_COMING
) {
1589 module_put(*probed_mod
);
1596 jump_label_unlock();
1601 int register_kprobe(struct kprobe
*p
)
1604 struct kprobe
*old_p
;
1605 struct module
*probed_mod
;
1606 kprobe_opcode_t
*addr
;
1608 /* Adjust probe address from symbol */
1609 addr
= kprobe_addr(p
);
1611 return PTR_ERR(addr
);
1614 ret
= warn_kprobe_rereg(p
);
1618 /* User can pass only KPROBE_FLAG_DISABLED to register_kprobe */
1619 p
->flags
&= KPROBE_FLAG_DISABLED
;
1621 INIT_LIST_HEAD(&p
->list
);
1623 ret
= check_kprobe_address_safe(p
, &probed_mod
);
1627 mutex_lock(&kprobe_mutex
);
1629 old_p
= get_kprobe(p
->addr
);
1631 /* Since this may unoptimize old_p, locking text_mutex. */
1632 ret
= register_aggr_kprobe(old_p
, p
);
1637 /* Prevent text modification */
1638 mutex_lock(&text_mutex
);
1639 ret
= prepare_kprobe(p
);
1640 mutex_unlock(&text_mutex
);
1645 INIT_HLIST_NODE(&p
->hlist
);
1646 hlist_add_head_rcu(&p
->hlist
,
1647 &kprobe_table
[hash_ptr(p
->addr
, KPROBE_HASH_BITS
)]);
1649 if (!kprobes_all_disarmed
&& !kprobe_disabled(p
)) {
1650 ret
= arm_kprobe(p
);
1652 hlist_del_rcu(&p
->hlist
);
1658 /* Try to optimize kprobe */
1659 try_to_optimize_kprobe(p
);
1661 mutex_unlock(&kprobe_mutex
);
1664 module_put(probed_mod
);
1668 EXPORT_SYMBOL_GPL(register_kprobe
);
1670 /* Check if all probes on the aggrprobe are disabled */
1671 static int aggr_kprobe_disabled(struct kprobe
*ap
)
1675 lockdep_assert_held(&kprobe_mutex
);
1677 list_for_each_entry(kp
, &ap
->list
, list
)
1678 if (!kprobe_disabled(kp
))
1680 * There is an active probe on the list.
1681 * We can't disable this ap.
1688 /* Disable one kprobe: Make sure called under kprobe_mutex is locked */
1689 static struct kprobe
*__disable_kprobe(struct kprobe
*p
)
1691 struct kprobe
*orig_p
;
1694 /* Get an original kprobe for return */
1695 orig_p
= __get_valid_kprobe(p
);
1696 if (unlikely(orig_p
== NULL
))
1697 return ERR_PTR(-EINVAL
);
1699 if (!kprobe_disabled(p
)) {
1700 /* Disable probe if it is a child probe */
1702 p
->flags
|= KPROBE_FLAG_DISABLED
;
1704 /* Try to disarm and disable this/parent probe */
1705 if (p
== orig_p
|| aggr_kprobe_disabled(orig_p
)) {
1707 * If kprobes_all_disarmed is set, orig_p
1708 * should have already been disarmed, so
1709 * skip unneed disarming process.
1711 if (!kprobes_all_disarmed
) {
1712 ret
= disarm_kprobe(orig_p
, true);
1714 p
->flags
&= ~KPROBE_FLAG_DISABLED
;
1715 return ERR_PTR(ret
);
1718 orig_p
->flags
|= KPROBE_FLAG_DISABLED
;
1726 * Unregister a kprobe without a scheduler synchronization.
1728 static int __unregister_kprobe_top(struct kprobe
*p
)
1730 struct kprobe
*ap
, *list_p
;
1732 /* Disable kprobe. This will disarm it if needed. */
1733 ap
= __disable_kprobe(p
);
1739 * This probe is an independent(and non-optimized) kprobe
1740 * (not an aggrprobe). Remove from the hash list.
1744 /* Following process expects this probe is an aggrprobe */
1745 WARN_ON(!kprobe_aggrprobe(ap
));
1747 if (list_is_singular(&ap
->list
) && kprobe_disarmed(ap
))
1749 * !disarmed could be happen if the probe is under delayed
1754 /* If disabling probe has special handlers, update aggrprobe */
1755 if (p
->post_handler
&& !kprobe_gone(p
)) {
1756 list_for_each_entry(list_p
, &ap
->list
, list
) {
1757 if ((list_p
!= p
) && (list_p
->post_handler
))
1760 ap
->post_handler
= NULL
;
1764 * Remove from the aggrprobe: this path will do nothing in
1765 * __unregister_kprobe_bottom().
1767 list_del_rcu(&p
->list
);
1768 if (!kprobe_disabled(ap
) && !kprobes_all_disarmed
)
1770 * Try to optimize this probe again, because post
1771 * handler may have been changed.
1773 optimize_kprobe(ap
);
1778 hlist_del_rcu(&ap
->hlist
);
1782 static void __unregister_kprobe_bottom(struct kprobe
*p
)
1786 if (list_empty(&p
->list
))
1787 /* This is an independent kprobe */
1788 arch_remove_kprobe(p
);
1789 else if (list_is_singular(&p
->list
)) {
1790 /* This is the last child of an aggrprobe */
1791 ap
= list_entry(p
->list
.next
, struct kprobe
, list
);
1793 free_aggr_kprobe(ap
);
1795 /* Otherwise, do nothing. */
1798 int register_kprobes(struct kprobe
**kps
, int num
)
1804 for (i
= 0; i
< num
; i
++) {
1805 ret
= register_kprobe(kps
[i
]);
1808 unregister_kprobes(kps
, i
);
1814 EXPORT_SYMBOL_GPL(register_kprobes
);
1816 void unregister_kprobe(struct kprobe
*p
)
1818 unregister_kprobes(&p
, 1);
1820 EXPORT_SYMBOL_GPL(unregister_kprobe
);
1822 void unregister_kprobes(struct kprobe
**kps
, int num
)
1828 mutex_lock(&kprobe_mutex
);
1829 for (i
= 0; i
< num
; i
++)
1830 if (__unregister_kprobe_top(kps
[i
]) < 0)
1831 kps
[i
]->addr
= NULL
;
1832 mutex_unlock(&kprobe_mutex
);
1835 for (i
= 0; i
< num
; i
++)
1837 __unregister_kprobe_bottom(kps
[i
]);
1839 EXPORT_SYMBOL_GPL(unregister_kprobes
);
1841 int __weak
kprobe_exceptions_notify(struct notifier_block
*self
,
1842 unsigned long val
, void *data
)
1846 NOKPROBE_SYMBOL(kprobe_exceptions_notify
);
1848 static struct notifier_block kprobe_exceptions_nb
= {
1849 .notifier_call
= kprobe_exceptions_notify
,
1850 .priority
= 0x7fffffff /* we need to be notified first */
1853 unsigned long __weak
arch_deref_entry_point(void *entry
)
1855 return (unsigned long)entry
;
1858 #ifdef CONFIG_KRETPROBES
1860 unsigned long __kretprobe_trampoline_handler(struct pt_regs
*regs
,
1861 void *trampoline_address
,
1862 void *frame_pointer
)
1864 kprobe_opcode_t
*correct_ret_addr
= NULL
;
1865 struct kretprobe_instance
*ri
= NULL
;
1866 struct llist_node
*first
, *node
;
1867 struct kretprobe
*rp
;
1869 /* Find all nodes for this frame. */
1870 first
= node
= current
->kretprobe_instances
.first
;
1872 ri
= container_of(node
, struct kretprobe_instance
, llist
);
1874 BUG_ON(ri
->fp
!= frame_pointer
);
1876 if (ri
->ret_addr
!= trampoline_address
) {
1877 correct_ret_addr
= ri
->ret_addr
;
1879 * This is the real return address. Any other
1880 * instances associated with this task are for
1881 * other calls deeper on the call stack
1888 pr_err("Oops! Kretprobe fails to find correct return address.\n");
1892 /* Unlink all nodes for this frame. */
1893 current
->kretprobe_instances
.first
= node
->next
;
1898 ri
= container_of(first
, struct kretprobe_instance
, llist
);
1899 first
= first
->next
;
1901 rp
= get_kretprobe(ri
);
1902 if (rp
&& rp
->handler
) {
1903 struct kprobe
*prev
= kprobe_running();
1905 __this_cpu_write(current_kprobe
, &rp
->kp
);
1906 ri
->ret_addr
= correct_ret_addr
;
1907 rp
->handler(ri
, regs
);
1908 __this_cpu_write(current_kprobe
, prev
);
1911 recycle_rp_inst(ri
);
1914 return (unsigned long)correct_ret_addr
;
1916 NOKPROBE_SYMBOL(__kretprobe_trampoline_handler
)
1919 * This kprobe pre_handler is registered with every kretprobe. When probe
1920 * hits it will set up the return probe.
1922 static int pre_handler_kretprobe(struct kprobe
*p
, struct pt_regs
*regs
)
1924 struct kretprobe
*rp
= container_of(p
, struct kretprobe
, kp
);
1925 struct kretprobe_instance
*ri
;
1926 struct freelist_node
*fn
;
1928 fn
= freelist_try_get(&rp
->freelist
);
1934 ri
= container_of(fn
, struct kretprobe_instance
, freelist
);
1936 if (rp
->entry_handler
&& rp
->entry_handler(ri
, regs
)) {
1937 freelist_add(&ri
->freelist
, &rp
->freelist
);
1941 arch_prepare_kretprobe(ri
, regs
);
1943 __llist_add(&ri
->llist
, ¤t
->kretprobe_instances
);
1947 NOKPROBE_SYMBOL(pre_handler_kretprobe
);
1949 bool __weak
arch_kprobe_on_func_entry(unsigned long offset
)
1955 * kprobe_on_func_entry() -- check whether given address is function entry
1956 * @addr: Target address
1957 * @sym: Target symbol name
1958 * @offset: The offset from the symbol or the address
1960 * This checks whether the given @addr+@offset or @sym+@offset is on the
1961 * function entry address or not.
1962 * This returns 0 if it is the function entry, or -EINVAL if it is not.
1963 * And also it returns -ENOENT if it fails the symbol or address lookup.
1964 * Caller must pass @addr or @sym (either one must be NULL), or this
1967 int kprobe_on_func_entry(kprobe_opcode_t
*addr
, const char *sym
, unsigned long offset
)
1969 kprobe_opcode_t
*kp_addr
= _kprobe_addr(addr
, sym
, offset
);
1971 if (IS_ERR(kp_addr
))
1972 return PTR_ERR(kp_addr
);
1974 if (!kallsyms_lookup_size_offset((unsigned long)kp_addr
, NULL
, &offset
))
1977 if (!arch_kprobe_on_func_entry(offset
))
1983 int register_kretprobe(struct kretprobe
*rp
)
1986 struct kretprobe_instance
*inst
;
1990 ret
= kprobe_on_func_entry(rp
->kp
.addr
, rp
->kp
.symbol_name
, rp
->kp
.offset
);
1994 /* If only rp->kp.addr is specified, check reregistering kprobes */
1995 if (rp
->kp
.addr
&& warn_kprobe_rereg(&rp
->kp
))
1998 if (kretprobe_blacklist_size
) {
1999 addr
= kprobe_addr(&rp
->kp
);
2001 return PTR_ERR(addr
);
2003 for (i
= 0; kretprobe_blacklist
[i
].name
!= NULL
; i
++) {
2004 if (kretprobe_blacklist
[i
].addr
== addr
)
2009 if (rp
->data_size
> KRETPROBE_MAX_DATA_SIZE
)
2012 rp
->kp
.pre_handler
= pre_handler_kretprobe
;
2013 rp
->kp
.post_handler
= NULL
;
2015 /* Pre-allocate memory for max kretprobe instances */
2016 if (rp
->maxactive
<= 0) {
2017 #ifdef CONFIG_PREEMPTION
2018 rp
->maxactive
= max_t(unsigned int, 10, 2*num_possible_cpus());
2020 rp
->maxactive
= num_possible_cpus();
2023 rp
->freelist
.head
= NULL
;
2024 rp
->rph
= kzalloc(sizeof(struct kretprobe_holder
), GFP_KERNEL
);
2029 for (i
= 0; i
< rp
->maxactive
; i
++) {
2030 inst
= kzalloc(sizeof(struct kretprobe_instance
) +
2031 rp
->data_size
, GFP_KERNEL
);
2033 refcount_set(&rp
->rph
->ref
, i
);
2037 inst
->rph
= rp
->rph
;
2038 freelist_add(&inst
->freelist
, &rp
->freelist
);
2040 refcount_set(&rp
->rph
->ref
, i
);
2043 /* Establish function entry probe point */
2044 ret
= register_kprobe(&rp
->kp
);
2049 EXPORT_SYMBOL_GPL(register_kretprobe
);
2051 int register_kretprobes(struct kretprobe
**rps
, int num
)
2057 for (i
= 0; i
< num
; i
++) {
2058 ret
= register_kretprobe(rps
[i
]);
2061 unregister_kretprobes(rps
, i
);
2067 EXPORT_SYMBOL_GPL(register_kretprobes
);
2069 void unregister_kretprobe(struct kretprobe
*rp
)
2071 unregister_kretprobes(&rp
, 1);
2073 EXPORT_SYMBOL_GPL(unregister_kretprobe
);
2075 void unregister_kretprobes(struct kretprobe
**rps
, int num
)
2081 mutex_lock(&kprobe_mutex
);
2082 for (i
= 0; i
< num
; i
++) {
2083 if (__unregister_kprobe_top(&rps
[i
]->kp
) < 0)
2084 rps
[i
]->kp
.addr
= NULL
;
2085 rps
[i
]->rph
->rp
= NULL
;
2087 mutex_unlock(&kprobe_mutex
);
2090 for (i
= 0; i
< num
; i
++) {
2091 if (rps
[i
]->kp
.addr
) {
2092 __unregister_kprobe_bottom(&rps
[i
]->kp
);
2093 free_rp_inst(rps
[i
]);
2097 EXPORT_SYMBOL_GPL(unregister_kretprobes
);
2099 #else /* CONFIG_KRETPROBES */
2100 int register_kretprobe(struct kretprobe
*rp
)
2104 EXPORT_SYMBOL_GPL(register_kretprobe
);
2106 int register_kretprobes(struct kretprobe
**rps
, int num
)
2110 EXPORT_SYMBOL_GPL(register_kretprobes
);
2112 void unregister_kretprobe(struct kretprobe
*rp
)
2115 EXPORT_SYMBOL_GPL(unregister_kretprobe
);
2117 void unregister_kretprobes(struct kretprobe
**rps
, int num
)
2120 EXPORT_SYMBOL_GPL(unregister_kretprobes
);
2122 static int pre_handler_kretprobe(struct kprobe
*p
, struct pt_regs
*regs
)
2126 NOKPROBE_SYMBOL(pre_handler_kretprobe
);
2128 #endif /* CONFIG_KRETPROBES */
2130 /* Set the kprobe gone and remove its instruction buffer. */
2131 static void kill_kprobe(struct kprobe
*p
)
2135 lockdep_assert_held(&kprobe_mutex
);
2137 p
->flags
|= KPROBE_FLAG_GONE
;
2138 if (kprobe_aggrprobe(p
)) {
2140 * If this is an aggr_kprobe, we have to list all the
2141 * chained probes and mark them GONE.
2143 list_for_each_entry(kp
, &p
->list
, list
)
2144 kp
->flags
|= KPROBE_FLAG_GONE
;
2145 p
->post_handler
= NULL
;
2146 kill_optimized_kprobe(p
);
2149 * Here, we can remove insn_slot safely, because no thread calls
2150 * the original probed function (which will be freed soon) any more.
2152 arch_remove_kprobe(p
);
2155 * The module is going away. We should disarm the kprobe which
2156 * is using ftrace, because ftrace framework is still available at
2157 * MODULE_STATE_GOING notification.
2159 if (kprobe_ftrace(p
) && !kprobe_disabled(p
) && !kprobes_all_disarmed
)
2160 disarm_kprobe_ftrace(p
);
2163 /* Disable one kprobe */
2164 int disable_kprobe(struct kprobe
*kp
)
2169 mutex_lock(&kprobe_mutex
);
2171 /* Disable this kprobe */
2172 p
= __disable_kprobe(kp
);
2176 mutex_unlock(&kprobe_mutex
);
2179 EXPORT_SYMBOL_GPL(disable_kprobe
);
2181 /* Enable one kprobe */
2182 int enable_kprobe(struct kprobe
*kp
)
2187 mutex_lock(&kprobe_mutex
);
2189 /* Check whether specified probe is valid. */
2190 p
= __get_valid_kprobe(kp
);
2191 if (unlikely(p
== NULL
)) {
2196 if (kprobe_gone(kp
)) {
2197 /* This kprobe has gone, we couldn't enable it. */
2203 kp
->flags
&= ~KPROBE_FLAG_DISABLED
;
2205 if (!kprobes_all_disarmed
&& kprobe_disabled(p
)) {
2206 p
->flags
&= ~KPROBE_FLAG_DISABLED
;
2207 ret
= arm_kprobe(p
);
2209 p
->flags
|= KPROBE_FLAG_DISABLED
;
2212 mutex_unlock(&kprobe_mutex
);
2215 EXPORT_SYMBOL_GPL(enable_kprobe
);
2217 /* Caller must NOT call this in usual path. This is only for critical case */
2218 void dump_kprobe(struct kprobe
*kp
)
2220 pr_err("Dumping kprobe:\n");
2221 pr_err("Name: %s\nOffset: %x\nAddress: %pS\n",
2222 kp
->symbol_name
, kp
->offset
, kp
->addr
);
2224 NOKPROBE_SYMBOL(dump_kprobe
);
2226 int kprobe_add_ksym_blacklist(unsigned long entry
)
2228 struct kprobe_blacklist_entry
*ent
;
2229 unsigned long offset
= 0, size
= 0;
2231 if (!kernel_text_address(entry
) ||
2232 !kallsyms_lookup_size_offset(entry
, &size
, &offset
))
2235 ent
= kmalloc(sizeof(*ent
), GFP_KERNEL
);
2238 ent
->start_addr
= entry
;
2239 ent
->end_addr
= entry
+ size
;
2240 INIT_LIST_HEAD(&ent
->list
);
2241 list_add_tail(&ent
->list
, &kprobe_blacklist
);
2246 /* Add all symbols in given area into kprobe blacklist */
2247 int kprobe_add_area_blacklist(unsigned long start
, unsigned long end
)
2249 unsigned long entry
;
2252 for (entry
= start
; entry
< end
; entry
+= ret
) {
2253 ret
= kprobe_add_ksym_blacklist(entry
);
2256 if (ret
== 0) /* In case of alias symbol */
2262 /* Remove all symbols in given area from kprobe blacklist */
2263 static void kprobe_remove_area_blacklist(unsigned long start
, unsigned long end
)
2265 struct kprobe_blacklist_entry
*ent
, *n
;
2267 list_for_each_entry_safe(ent
, n
, &kprobe_blacklist
, list
) {
2268 if (ent
->start_addr
< start
|| ent
->start_addr
>= end
)
2270 list_del(&ent
->list
);
2275 static void kprobe_remove_ksym_blacklist(unsigned long entry
)
2277 kprobe_remove_area_blacklist(entry
, entry
+ 1);
2280 int __weak
arch_kprobe_get_kallsym(unsigned int *symnum
, unsigned long *value
,
2281 char *type
, char *sym
)
2286 int kprobe_get_kallsym(unsigned int symnum
, unsigned long *value
, char *type
,
2289 #ifdef __ARCH_WANT_KPROBES_INSN_SLOT
2290 if (!kprobe_cache_get_kallsym(&kprobe_insn_slots
, &symnum
, value
, type
, sym
))
2292 #ifdef CONFIG_OPTPROBES
2293 if (!kprobe_cache_get_kallsym(&kprobe_optinsn_slots
, &symnum
, value
, type
, sym
))
2297 if (!arch_kprobe_get_kallsym(&symnum
, value
, type
, sym
))
2302 int __init __weak
arch_populate_kprobe_blacklist(void)
2308 * Lookup and populate the kprobe_blacklist.
2310 * Unlike the kretprobe blacklist, we'll need to determine
2311 * the range of addresses that belong to the said functions,
2312 * since a kprobe need not necessarily be at the beginning
2315 static int __init
populate_kprobe_blacklist(unsigned long *start
,
2318 unsigned long entry
;
2319 unsigned long *iter
;
2322 for (iter
= start
; iter
< end
; iter
++) {
2323 entry
= arch_deref_entry_point((void *)*iter
);
2324 ret
= kprobe_add_ksym_blacklist(entry
);
2331 /* Symbols in __kprobes_text are blacklisted */
2332 ret
= kprobe_add_area_blacklist((unsigned long)__kprobes_text_start
,
2333 (unsigned long)__kprobes_text_end
);
2337 /* Symbols in noinstr section are blacklisted */
2338 ret
= kprobe_add_area_blacklist((unsigned long)__noinstr_text_start
,
2339 (unsigned long)__noinstr_text_end
);
2341 return ret
? : arch_populate_kprobe_blacklist();
2344 static void add_module_kprobe_blacklist(struct module
*mod
)
2346 unsigned long start
, end
;
2349 if (mod
->kprobe_blacklist
) {
2350 for (i
= 0; i
< mod
->num_kprobe_blacklist
; i
++)
2351 kprobe_add_ksym_blacklist(mod
->kprobe_blacklist
[i
]);
2354 start
= (unsigned long)mod
->kprobes_text_start
;
2356 end
= start
+ mod
->kprobes_text_size
;
2357 kprobe_add_area_blacklist(start
, end
);
2360 start
= (unsigned long)mod
->noinstr_text_start
;
2362 end
= start
+ mod
->noinstr_text_size
;
2363 kprobe_add_area_blacklist(start
, end
);
2367 static void remove_module_kprobe_blacklist(struct module
*mod
)
2369 unsigned long start
, end
;
2372 if (mod
->kprobe_blacklist
) {
2373 for (i
= 0; i
< mod
->num_kprobe_blacklist
; i
++)
2374 kprobe_remove_ksym_blacklist(mod
->kprobe_blacklist
[i
]);
2377 start
= (unsigned long)mod
->kprobes_text_start
;
2379 end
= start
+ mod
->kprobes_text_size
;
2380 kprobe_remove_area_blacklist(start
, end
);
2383 start
= (unsigned long)mod
->noinstr_text_start
;
2385 end
= start
+ mod
->noinstr_text_size
;
2386 kprobe_remove_area_blacklist(start
, end
);
2390 /* Module notifier call back, checking kprobes on the module */
2391 static int kprobes_module_callback(struct notifier_block
*nb
,
2392 unsigned long val
, void *data
)
2394 struct module
*mod
= data
;
2395 struct hlist_head
*head
;
2398 int checkcore
= (val
== MODULE_STATE_GOING
);
2400 if (val
== MODULE_STATE_COMING
) {
2401 mutex_lock(&kprobe_mutex
);
2402 add_module_kprobe_blacklist(mod
);
2403 mutex_unlock(&kprobe_mutex
);
2405 if (val
!= MODULE_STATE_GOING
&& val
!= MODULE_STATE_LIVE
)
2409 * When MODULE_STATE_GOING was notified, both of module .text and
2410 * .init.text sections would be freed. When MODULE_STATE_LIVE was
2411 * notified, only .init.text section would be freed. We need to
2412 * disable kprobes which have been inserted in the sections.
2414 mutex_lock(&kprobe_mutex
);
2415 for (i
= 0; i
< KPROBE_TABLE_SIZE
; i
++) {
2416 head
= &kprobe_table
[i
];
2417 hlist_for_each_entry(p
, head
, hlist
)
2418 if (within_module_init((unsigned long)p
->addr
, mod
) ||
2420 within_module_core((unsigned long)p
->addr
, mod
))) {
2422 * The vaddr this probe is installed will soon
2423 * be vfreed buy not synced to disk. Hence,
2424 * disarming the breakpoint isn't needed.
2426 * Note, this will also move any optimized probes
2427 * that are pending to be removed from their
2428 * corresponding lists to the freeing_list and
2429 * will not be touched by the delayed
2430 * kprobe_optimizer work handler.
2435 if (val
== MODULE_STATE_GOING
)
2436 remove_module_kprobe_blacklist(mod
);
2437 mutex_unlock(&kprobe_mutex
);
2441 static struct notifier_block kprobe_module_nb
= {
2442 .notifier_call
= kprobes_module_callback
,
2446 /* Markers of _kprobe_blacklist section */
2447 extern unsigned long __start_kprobe_blacklist
[];
2448 extern unsigned long __stop_kprobe_blacklist
[];
2450 void kprobe_free_init_mem(void)
2452 void *start
= (void *)(&__init_begin
);
2453 void *end
= (void *)(&__init_end
);
2454 struct hlist_head
*head
;
2458 mutex_lock(&kprobe_mutex
);
2460 /* Kill all kprobes on initmem */
2461 for (i
= 0; i
< KPROBE_TABLE_SIZE
; i
++) {
2462 head
= &kprobe_table
[i
];
2463 hlist_for_each_entry(p
, head
, hlist
) {
2464 if (start
<= (void *)p
->addr
&& (void *)p
->addr
< end
)
2469 mutex_unlock(&kprobe_mutex
);
2472 static int __init
init_kprobes(void)
2476 /* FIXME allocate the probe table, currently defined statically */
2477 /* initialize all list heads */
2478 for (i
= 0; i
< KPROBE_TABLE_SIZE
; i
++)
2479 INIT_HLIST_HEAD(&kprobe_table
[i
]);
2481 err
= populate_kprobe_blacklist(__start_kprobe_blacklist
,
2482 __stop_kprobe_blacklist
);
2484 pr_err("kprobes: failed to populate blacklist: %d\n", err
);
2485 pr_err("Please take care of using kprobes.\n");
2488 if (kretprobe_blacklist_size
) {
2489 /* lookup the function address from its name */
2490 for (i
= 0; kretprobe_blacklist
[i
].name
!= NULL
; i
++) {
2491 kretprobe_blacklist
[i
].addr
=
2492 kprobe_lookup_name(kretprobe_blacklist
[i
].name
, 0);
2493 if (!kretprobe_blacklist
[i
].addr
)
2494 printk("kretprobe: lookup failed: %s\n",
2495 kretprobe_blacklist
[i
].name
);
2499 /* By default, kprobes are armed */
2500 kprobes_all_disarmed
= false;
2502 #if defined(CONFIG_OPTPROBES) && defined(__ARCH_WANT_KPROBES_INSN_SLOT)
2503 /* Init kprobe_optinsn_slots for allocation */
2504 kprobe_optinsn_slots
.insn_size
= MAX_OPTINSN_SIZE
;
2507 err
= arch_init_kprobes();
2509 err
= register_die_notifier(&kprobe_exceptions_nb
);
2511 err
= register_module_notifier(&kprobe_module_nb
);
2513 kprobes_initialized
= (err
== 0);
2519 early_initcall(init_kprobes
);
2521 #if defined(CONFIG_OPTPROBES)
2522 static int __init
init_optprobes(void)
2525 * Enable kprobe optimization - this kicks the optimizer which
2526 * depends on synchronize_rcu_tasks() and ksoftirqd, that is
2527 * not spawned in early initcall. So delay the optimization.
2529 optimize_all_kprobes();
2533 subsys_initcall(init_optprobes
);
2536 #ifdef CONFIG_DEBUG_FS
2537 static void report_probe(struct seq_file
*pi
, struct kprobe
*p
,
2538 const char *sym
, int offset
, char *modname
, struct kprobe
*pp
)
2541 void *addr
= p
->addr
;
2543 if (p
->pre_handler
== pre_handler_kretprobe
)
2548 if (!kallsyms_show_value(pi
->file
->f_cred
))
2552 seq_printf(pi
, "%px %s %s+0x%x %s ",
2553 addr
, kprobe_type
, sym
, offset
,
2554 (modname
? modname
: " "));
2555 else /* try to use %pS */
2556 seq_printf(pi
, "%px %s %pS ",
2557 addr
, kprobe_type
, p
->addr
);
2561 seq_printf(pi
, "%s%s%s%s\n",
2562 (kprobe_gone(p
) ? "[GONE]" : ""),
2563 ((kprobe_disabled(p
) && !kprobe_gone(p
)) ? "[DISABLED]" : ""),
2564 (kprobe_optimized(pp
) ? "[OPTIMIZED]" : ""),
2565 (kprobe_ftrace(pp
) ? "[FTRACE]" : ""));
2568 static void *kprobe_seq_start(struct seq_file
*f
, loff_t
*pos
)
2570 return (*pos
< KPROBE_TABLE_SIZE
) ? pos
: NULL
;
2573 static void *kprobe_seq_next(struct seq_file
*f
, void *v
, loff_t
*pos
)
2576 if (*pos
>= KPROBE_TABLE_SIZE
)
2581 static void kprobe_seq_stop(struct seq_file
*f
, void *v
)
2586 static int show_kprobe_addr(struct seq_file
*pi
, void *v
)
2588 struct hlist_head
*head
;
2589 struct kprobe
*p
, *kp
;
2590 const char *sym
= NULL
;
2591 unsigned int i
= *(loff_t
*) v
;
2592 unsigned long offset
= 0;
2593 char *modname
, namebuf
[KSYM_NAME_LEN
];
2595 head
= &kprobe_table
[i
];
2597 hlist_for_each_entry_rcu(p
, head
, hlist
) {
2598 sym
= kallsyms_lookup((unsigned long)p
->addr
, NULL
,
2599 &offset
, &modname
, namebuf
);
2600 if (kprobe_aggrprobe(p
)) {
2601 list_for_each_entry_rcu(kp
, &p
->list
, list
)
2602 report_probe(pi
, kp
, sym
, offset
, modname
, p
);
2604 report_probe(pi
, p
, sym
, offset
, modname
, NULL
);
2610 static const struct seq_operations kprobes_sops
= {
2611 .start
= kprobe_seq_start
,
2612 .next
= kprobe_seq_next
,
2613 .stop
= kprobe_seq_stop
,
2614 .show
= show_kprobe_addr
2617 DEFINE_SEQ_ATTRIBUTE(kprobes
);
2619 /* kprobes/blacklist -- shows which functions can not be probed */
2620 static void *kprobe_blacklist_seq_start(struct seq_file
*m
, loff_t
*pos
)
2622 mutex_lock(&kprobe_mutex
);
2623 return seq_list_start(&kprobe_blacklist
, *pos
);
2626 static void *kprobe_blacklist_seq_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
2628 return seq_list_next(v
, &kprobe_blacklist
, pos
);
2631 static int kprobe_blacklist_seq_show(struct seq_file
*m
, void *v
)
2633 struct kprobe_blacklist_entry
*ent
=
2634 list_entry(v
, struct kprobe_blacklist_entry
, list
);
2637 * If /proc/kallsyms is not showing kernel address, we won't
2638 * show them here either.
2640 if (!kallsyms_show_value(m
->file
->f_cred
))
2641 seq_printf(m
, "0x%px-0x%px\t%ps\n", NULL
, NULL
,
2642 (void *)ent
->start_addr
);
2644 seq_printf(m
, "0x%px-0x%px\t%ps\n", (void *)ent
->start_addr
,
2645 (void *)ent
->end_addr
, (void *)ent
->start_addr
);
2649 static void kprobe_blacklist_seq_stop(struct seq_file
*f
, void *v
)
2651 mutex_unlock(&kprobe_mutex
);
2654 static const struct seq_operations kprobe_blacklist_sops
= {
2655 .start
= kprobe_blacklist_seq_start
,
2656 .next
= kprobe_blacklist_seq_next
,
2657 .stop
= kprobe_blacklist_seq_stop
,
2658 .show
= kprobe_blacklist_seq_show
,
2660 DEFINE_SEQ_ATTRIBUTE(kprobe_blacklist
);
2662 static int arm_all_kprobes(void)
2664 struct hlist_head
*head
;
2666 unsigned int i
, total
= 0, errors
= 0;
2669 mutex_lock(&kprobe_mutex
);
2671 /* If kprobes are armed, just return */
2672 if (!kprobes_all_disarmed
)
2673 goto already_enabled
;
2676 * optimize_kprobe() called by arm_kprobe() checks
2677 * kprobes_all_disarmed, so set kprobes_all_disarmed before
2680 kprobes_all_disarmed
= false;
2681 /* Arming kprobes doesn't optimize kprobe itself */
2682 for (i
= 0; i
< KPROBE_TABLE_SIZE
; i
++) {
2683 head
= &kprobe_table
[i
];
2684 /* Arm all kprobes on a best-effort basis */
2685 hlist_for_each_entry(p
, head
, hlist
) {
2686 if (!kprobe_disabled(p
)) {
2687 err
= arm_kprobe(p
);
2698 pr_warn("Kprobes globally enabled, but failed to arm %d out of %d probes\n",
2701 pr_info("Kprobes globally enabled\n");
2704 mutex_unlock(&kprobe_mutex
);
2708 static int disarm_all_kprobes(void)
2710 struct hlist_head
*head
;
2712 unsigned int i
, total
= 0, errors
= 0;
2715 mutex_lock(&kprobe_mutex
);
2717 /* If kprobes are already disarmed, just return */
2718 if (kprobes_all_disarmed
) {
2719 mutex_unlock(&kprobe_mutex
);
2723 kprobes_all_disarmed
= true;
2725 for (i
= 0; i
< KPROBE_TABLE_SIZE
; i
++) {
2726 head
= &kprobe_table
[i
];
2727 /* Disarm all kprobes on a best-effort basis */
2728 hlist_for_each_entry(p
, head
, hlist
) {
2729 if (!arch_trampoline_kprobe(p
) && !kprobe_disabled(p
)) {
2730 err
= disarm_kprobe(p
, false);
2741 pr_warn("Kprobes globally disabled, but failed to disarm %d out of %d probes\n",
2744 pr_info("Kprobes globally disabled\n");
2746 mutex_unlock(&kprobe_mutex
);
2748 /* Wait for disarming all kprobes by optimizer */
2749 wait_for_kprobe_optimizer();
2755 * XXX: The debugfs bool file interface doesn't allow for callbacks
2756 * when the bool state is switched. We can reuse that facility when
2759 static ssize_t
read_enabled_file_bool(struct file
*file
,
2760 char __user
*user_buf
, size_t count
, loff_t
*ppos
)
2764 if (!kprobes_all_disarmed
)
2770 return simple_read_from_buffer(user_buf
, count
, ppos
, buf
, 2);
2773 static ssize_t
write_enabled_file_bool(struct file
*file
,
2774 const char __user
*user_buf
, size_t count
, loff_t
*ppos
)
2780 buf_size
= min(count
, (sizeof(buf
)-1));
2781 if (copy_from_user(buf
, user_buf
, buf_size
))
2784 buf
[buf_size
] = '\0';
2789 ret
= arm_all_kprobes();
2794 ret
= disarm_all_kprobes();
2806 static const struct file_operations fops_kp
= {
2807 .read
= read_enabled_file_bool
,
2808 .write
= write_enabled_file_bool
,
2809 .llseek
= default_llseek
,
2812 static int __init
debugfs_kprobe_init(void)
2816 dir
= debugfs_create_dir("kprobes", NULL
);
2818 debugfs_create_file("list", 0400, dir
, NULL
, &kprobes_fops
);
2820 debugfs_create_file("enabled", 0600, dir
, NULL
, &fops_kp
);
2822 debugfs_create_file("blacklist", 0400, dir
, NULL
,
2823 &kprobe_blacklist_fops
);
2828 late_initcall(debugfs_kprobe_init
);
2829 #endif /* CONFIG_DEBUG_FS */