1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * Kernel Probes (KProbes)
6 * Copyright (C) IBM Corporation, 2002, 2004
8 * 2002-Oct Created by Vamsi Krishna S <vamsi_krishna@in.ibm.com> Kernel
9 * Probes initial implementation (includes suggestions from
11 * 2004-Aug Updated by Prasanna S Panchamukhi <prasanna@in.ibm.com> with
12 * hlists and exceptions notifier as suggested by Andi Kleen.
13 * 2004-July Suparna Bhattacharya <suparna@in.ibm.com> added jumper probes
14 * interface to access function arguments.
15 * 2004-Sep Prasanna S Panchamukhi <prasanna@in.ibm.com> Changed Kprobes
16 * exceptions notifier to be first on the priority list.
17 * 2005-May Hien Nguyen <hien@us.ibm.com>, Jim Keniston
18 * <jkenisto@us.ibm.com> and Prasanna S Panchamukhi
19 * <prasanna@in.ibm.com> added function-return probes.
21 #include <linux/kprobes.h>
22 #include <linux/hash.h>
23 #include <linux/init.h>
24 #include <linux/slab.h>
25 #include <linux/stddef.h>
26 #include <linux/export.h>
27 #include <linux/moduleloader.h>
28 #include <linux/kallsyms.h>
29 #include <linux/freezer.h>
30 #include <linux/seq_file.h>
31 #include <linux/debugfs.h>
32 #include <linux/sysctl.h>
33 #include <linux/kdebug.h>
34 #include <linux/memory.h>
35 #include <linux/ftrace.h>
36 #include <linux/cpu.h>
37 #include <linux/jump_label.h>
38 #include <linux/perf_event.h>
40 #include <asm/sections.h>
41 #include <asm/cacheflush.h>
42 #include <asm/errno.h>
43 #include <linux/uaccess.h>
45 #define KPROBE_HASH_BITS 6
46 #define KPROBE_TABLE_SIZE (1 << KPROBE_HASH_BITS)
49 static int kprobes_initialized;
50 /* kprobe_table can be accessed by
51 * - Normal hlist traversal and RCU add/del under kprobe_mutex is held.
53 * - RCU hlist traversal under disabling preempt (breakpoint handlers)
55 static struct hlist_head kprobe_table[KPROBE_TABLE_SIZE];
57 /* NOTE: change this value only with kprobe_mutex held */
58 static bool kprobes_all_disarmed;
60 /* This protects kprobe_table and optimizing_list */
61 static DEFINE_MUTEX(kprobe_mutex);
62 static DEFINE_PER_CPU(struct kprobe *, kprobe_instance) = NULL;
64 kprobe_opcode_t * __weak kprobe_lookup_name(const char *name,
65 unsigned int __unused)
67 return ((kprobe_opcode_t *)(kallsyms_lookup_name(name)));
70 /* Blacklist -- list of struct kprobe_blacklist_entry */
71 static LIST_HEAD(kprobe_blacklist);
73 #ifdef __ARCH_WANT_KPROBES_INSN_SLOT
75 * kprobe->ainsn.insn points to the copy of the instruction to be
76 * single-stepped. x86_64, POWER4 and above have no-exec support and
77 * stepping on the instruction on a vmalloced/kmalloced/data page
78 * is a recipe for disaster
80 struct kprobe_insn_page {
81 struct list_head list;
82 kprobe_opcode_t *insns; /* Page of instruction slots */
83 struct kprobe_insn_cache *cache;
89 #define KPROBE_INSN_PAGE_SIZE(slots) \
90 (offsetof(struct kprobe_insn_page, slot_used) + \
91 (sizeof(char) * (slots)))
93 static int slots_per_page(struct kprobe_insn_cache *c)
95 return PAGE_SIZE/(c->insn_size * sizeof(kprobe_opcode_t));
98 enum kprobe_slot_state {
104 void __weak *alloc_insn_page(void)
106 return module_alloc(PAGE_SIZE);
109 void __weak free_insn_page(void *page)
111 module_memfree(page);
114 struct kprobe_insn_cache kprobe_insn_slots = {
115 .mutex = __MUTEX_INITIALIZER(kprobe_insn_slots.mutex),
116 .alloc = alloc_insn_page,
117 .free = free_insn_page,
118 .sym = KPROBE_INSN_PAGE_SYM,
119 .pages = LIST_HEAD_INIT(kprobe_insn_slots.pages),
120 .insn_size = MAX_INSN_SIZE,
123 static int collect_garbage_slots(struct kprobe_insn_cache *c);
126 * __get_insn_slot() - Find a slot on an executable page for an instruction.
127 * We allocate an executable page if there's no room on existing ones.
129 kprobe_opcode_t *__get_insn_slot(struct kprobe_insn_cache *c)
131 struct kprobe_insn_page *kip;
132 kprobe_opcode_t *slot = NULL;
134 /* Since the slot array is not protected by rcu, we need a mutex */
135 mutex_lock(&c->mutex);
138 list_for_each_entry_rcu(kip, &c->pages, list) {
139 if (kip->nused < slots_per_page(c)) {
141 for (i = 0; i < slots_per_page(c); i++) {
142 if (kip->slot_used[i] == SLOT_CLEAN) {
143 kip->slot_used[i] = SLOT_USED;
145 slot = kip->insns + (i * c->insn_size);
150 /* kip->nused is broken. Fix it. */
151 kip->nused = slots_per_page(c);
157 /* If there are any garbage slots, collect it and try again. */
158 if (c->nr_garbage && collect_garbage_slots(c) == 0)
161 /* All out of space. Need to allocate a new page. */
162 kip = kmalloc(KPROBE_INSN_PAGE_SIZE(slots_per_page(c)), GFP_KERNEL);
167 * Use module_alloc so this page is within +/- 2GB of where the
168 * kernel image and loaded module images reside. This is required
169 * so x86_64 can correctly handle the %rip-relative fixups.
171 kip->insns = c->alloc();
176 INIT_LIST_HEAD(&kip->list);
177 memset(kip->slot_used, SLOT_CLEAN, slots_per_page(c));
178 kip->slot_used[0] = SLOT_USED;
182 list_add_rcu(&kip->list, &c->pages);
185 /* Record the perf ksymbol register event after adding the page */
186 perf_event_ksymbol(PERF_RECORD_KSYMBOL_TYPE_OOL, (unsigned long)kip->insns,
187 PAGE_SIZE, false, c->sym);
189 mutex_unlock(&c->mutex);
193 /* Return 1 if all garbages are collected, otherwise 0. */
194 static int collect_one_slot(struct kprobe_insn_page *kip, int idx)
196 kip->slot_used[idx] = SLOT_CLEAN;
198 if (kip->nused == 0) {
200 * Page is no longer in use. Free it unless
201 * it's the last one. We keep the last one
202 * so as not to have to set it up again the
203 * next time somebody inserts a probe.
205 if (!list_is_singular(&kip->list)) {
207 * Record perf ksymbol unregister event before removing
210 perf_event_ksymbol(PERF_RECORD_KSYMBOL_TYPE_OOL,
211 (unsigned long)kip->insns, PAGE_SIZE, true,
213 list_del_rcu(&kip->list);
215 kip->cache->free(kip->insns);
223 static int collect_garbage_slots(struct kprobe_insn_cache *c)
225 struct kprobe_insn_page *kip, *next;
227 /* Ensure no-one is interrupted on the garbages */
230 list_for_each_entry_safe(kip, next, &c->pages, list) {
232 if (kip->ngarbage == 0)
234 kip->ngarbage = 0; /* we will collect all garbages */
235 for (i = 0; i < slots_per_page(c); i++) {
236 if (kip->slot_used[i] == SLOT_DIRTY && collect_one_slot(kip, i))
244 void __free_insn_slot(struct kprobe_insn_cache *c,
245 kprobe_opcode_t *slot, int dirty)
247 struct kprobe_insn_page *kip;
250 mutex_lock(&c->mutex);
252 list_for_each_entry_rcu(kip, &c->pages, list) {
253 idx = ((long)slot - (long)kip->insns) /
254 (c->insn_size * sizeof(kprobe_opcode_t));
255 if (idx >= 0 && idx < slots_per_page(c))
258 /* Could not find this slot. */
263 /* Mark and sweep: this may sleep */
265 /* Check double free */
266 WARN_ON(kip->slot_used[idx] != SLOT_USED);
268 kip->slot_used[idx] = SLOT_DIRTY;
270 if (++c->nr_garbage > slots_per_page(c))
271 collect_garbage_slots(c);
273 collect_one_slot(kip, idx);
276 mutex_unlock(&c->mutex);
280 * Check given address is on the page of kprobe instruction slots.
281 * This will be used for checking whether the address on a stack
282 * is on a text area or not.
284 bool __is_insn_slot_addr(struct kprobe_insn_cache *c, unsigned long addr)
286 struct kprobe_insn_page *kip;
290 list_for_each_entry_rcu(kip, &c->pages, list) {
291 if (addr >= (unsigned long)kip->insns &&
292 addr < (unsigned long)kip->insns + PAGE_SIZE) {
302 int kprobe_cache_get_kallsym(struct kprobe_insn_cache *c, unsigned int *symnum,
303 unsigned long *value, char *type, char *sym)
305 struct kprobe_insn_page *kip;
309 list_for_each_entry_rcu(kip, &c->pages, list) {
312 strlcpy(sym, c->sym, KSYM_NAME_LEN);
314 *value = (unsigned long)kip->insns;
323 #ifdef CONFIG_OPTPROBES
324 void __weak *alloc_optinsn_page(void)
326 return alloc_insn_page();
329 void __weak free_optinsn_page(void *page)
331 free_insn_page(page);
334 /* For optimized_kprobe buffer */
335 struct kprobe_insn_cache kprobe_optinsn_slots = {
336 .mutex = __MUTEX_INITIALIZER(kprobe_optinsn_slots.mutex),
337 .alloc = alloc_optinsn_page,
338 .free = free_optinsn_page,
339 .sym = KPROBE_OPTINSN_PAGE_SYM,
340 .pages = LIST_HEAD_INIT(kprobe_optinsn_slots.pages),
341 /* .insn_size is initialized later */
347 /* We have preemption disabled.. so it is safe to use __ versions */
348 static inline void set_kprobe_instance(struct kprobe *kp)
350 __this_cpu_write(kprobe_instance, kp);
353 static inline void reset_kprobe_instance(void)
355 __this_cpu_write(kprobe_instance, NULL);
359 * This routine is called either:
360 * - under the kprobe_mutex - during kprobe_[un]register()
362 * - with preemption disabled - from arch/xxx/kernel/kprobes.c
364 struct kprobe *get_kprobe(void *addr)
366 struct hlist_head *head;
369 head = &kprobe_table[hash_ptr(addr, KPROBE_HASH_BITS)];
370 hlist_for_each_entry_rcu(p, head, hlist,
371 lockdep_is_held(&kprobe_mutex)) {
378 NOKPROBE_SYMBOL(get_kprobe);
380 static int aggr_pre_handler(struct kprobe *p, struct pt_regs *regs);
382 /* Return true if the kprobe is an aggregator */
383 static inline int kprobe_aggrprobe(struct kprobe *p)
385 return p->pre_handler == aggr_pre_handler;
388 /* Return true(!0) if the kprobe is unused */
389 static inline int kprobe_unused(struct kprobe *p)
391 return kprobe_aggrprobe(p) && kprobe_disabled(p) &&
392 list_empty(&p->list);
396 * Keep all fields in the kprobe consistent
398 static inline void copy_kprobe(struct kprobe *ap, struct kprobe *p)
400 memcpy(&p->opcode, &ap->opcode, sizeof(kprobe_opcode_t));
401 memcpy(&p->ainsn, &ap->ainsn, sizeof(struct arch_specific_insn));
404 #ifdef CONFIG_OPTPROBES
405 /* NOTE: change this value only with kprobe_mutex held */
406 static bool kprobes_allow_optimization;
409 * Call all pre_handler on the list, but ignores its return value.
410 * This must be called from arch-dep optimized caller.
412 void opt_pre_handler(struct kprobe *p, struct pt_regs *regs)
416 list_for_each_entry_rcu(kp, &p->list, list) {
417 if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
418 set_kprobe_instance(kp);
419 kp->pre_handler(kp, regs);
421 reset_kprobe_instance();
424 NOKPROBE_SYMBOL(opt_pre_handler);
426 /* Free optimized instructions and optimized_kprobe */
427 static void free_aggr_kprobe(struct kprobe *p)
429 struct optimized_kprobe *op;
431 op = container_of(p, struct optimized_kprobe, kp);
432 arch_remove_optimized_kprobe(op);
433 arch_remove_kprobe(p);
437 /* Return true(!0) if the kprobe is ready for optimization. */
438 static inline int kprobe_optready(struct kprobe *p)
440 struct optimized_kprobe *op;
442 if (kprobe_aggrprobe(p)) {
443 op = container_of(p, struct optimized_kprobe, kp);
444 return arch_prepared_optinsn(&op->optinsn);
450 /* Return true(!0) if the kprobe is disarmed. Note: p must be on hash list */
451 static inline int kprobe_disarmed(struct kprobe *p)
453 struct optimized_kprobe *op;
455 /* If kprobe is not aggr/opt probe, just return kprobe is disabled */
456 if (!kprobe_aggrprobe(p))
457 return kprobe_disabled(p);
459 op = container_of(p, struct optimized_kprobe, kp);
461 return kprobe_disabled(p) && list_empty(&op->list);
464 /* Return true(!0) if the probe is queued on (un)optimizing lists */
465 static int kprobe_queued(struct kprobe *p)
467 struct optimized_kprobe *op;
469 if (kprobe_aggrprobe(p)) {
470 op = container_of(p, struct optimized_kprobe, kp);
471 if (!list_empty(&op->list))
478 * Return an optimized kprobe whose optimizing code replaces
479 * instructions including addr (exclude breakpoint).
481 static struct kprobe *get_optimized_kprobe(unsigned long addr)
484 struct kprobe *p = NULL;
485 struct optimized_kprobe *op;
487 /* Don't check i == 0, since that is a breakpoint case. */
488 for (i = 1; !p && i < MAX_OPTIMIZED_LENGTH; i++)
489 p = get_kprobe((void *)(addr - i));
491 if (p && kprobe_optready(p)) {
492 op = container_of(p, struct optimized_kprobe, kp);
493 if (arch_within_optimized_kprobe(op, addr))
500 /* Optimization staging list, protected by kprobe_mutex */
501 static LIST_HEAD(optimizing_list);
502 static LIST_HEAD(unoptimizing_list);
503 static LIST_HEAD(freeing_list);
505 static void kprobe_optimizer(struct work_struct *work);
506 static DECLARE_DELAYED_WORK(optimizing_work, kprobe_optimizer);
507 #define OPTIMIZE_DELAY 5
510 * Optimize (replace a breakpoint with a jump) kprobes listed on
513 static void do_optimize_kprobes(void)
515 lockdep_assert_held(&text_mutex);
517 * The optimization/unoptimization refers online_cpus via
518 * stop_machine() and cpu-hotplug modifies online_cpus.
519 * And same time, text_mutex will be held in cpu-hotplug and here.
520 * This combination can cause a deadlock (cpu-hotplug try to lock
521 * text_mutex but stop_machine can not be done because online_cpus
523 * To avoid this deadlock, caller must have locked cpu hotplug
524 * for preventing cpu-hotplug outside of text_mutex locking.
526 lockdep_assert_cpus_held();
528 /* Optimization never be done when disarmed */
529 if (kprobes_all_disarmed || !kprobes_allow_optimization ||
530 list_empty(&optimizing_list))
533 arch_optimize_kprobes(&optimizing_list);
537 * Unoptimize (replace a jump with a breakpoint and remove the breakpoint
538 * if need) kprobes listed on unoptimizing_list.
540 static void do_unoptimize_kprobes(void)
542 struct optimized_kprobe *op, *tmp;
544 lockdep_assert_held(&text_mutex);
545 /* See comment in do_optimize_kprobes() */
546 lockdep_assert_cpus_held();
548 /* Unoptimization must be done anytime */
549 if (list_empty(&unoptimizing_list))
552 arch_unoptimize_kprobes(&unoptimizing_list, &freeing_list);
553 /* Loop free_list for disarming */
554 list_for_each_entry_safe(op, tmp, &freeing_list, list) {
555 /* Switching from detour code to origin */
556 op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
557 /* Disarm probes if marked disabled */
558 if (kprobe_disabled(&op->kp))
559 arch_disarm_kprobe(&op->kp);
560 if (kprobe_unused(&op->kp)) {
562 * Remove unused probes from hash list. After waiting
563 * for synchronization, these probes are reclaimed.
564 * (reclaiming is done by do_free_cleaned_kprobes.)
566 hlist_del_rcu(&op->kp.hlist);
568 list_del_init(&op->list);
572 /* Reclaim all kprobes on the free_list */
573 static void do_free_cleaned_kprobes(void)
575 struct optimized_kprobe *op, *tmp;
577 list_for_each_entry_safe(op, tmp, &freeing_list, list) {
578 list_del_init(&op->list);
579 if (WARN_ON_ONCE(!kprobe_unused(&op->kp))) {
581 * This must not happen, but if there is a kprobe
582 * still in use, keep it on kprobes hash list.
586 free_aggr_kprobe(&op->kp);
590 /* Start optimizer after OPTIMIZE_DELAY passed */
591 static void kick_kprobe_optimizer(void)
593 schedule_delayed_work(&optimizing_work, OPTIMIZE_DELAY);
596 /* Kprobe jump optimizer */
597 static void kprobe_optimizer(struct work_struct *work)
599 mutex_lock(&kprobe_mutex);
601 mutex_lock(&text_mutex);
604 * Step 1: Unoptimize kprobes and collect cleaned (unused and disarmed)
605 * kprobes before waiting for quiesence period.
607 do_unoptimize_kprobes();
610 * Step 2: Wait for quiesence period to ensure all potentially
611 * preempted tasks to have normally scheduled. Because optprobe
612 * may modify multiple instructions, there is a chance that Nth
613 * instruction is preempted. In that case, such tasks can return
614 * to 2nd-Nth byte of jump instruction. This wait is for avoiding it.
615 * Note that on non-preemptive kernel, this is transparently converted
616 * to synchronoze_sched() to wait for all interrupts to have completed.
618 synchronize_rcu_tasks();
620 /* Step 3: Optimize kprobes after quiesence period */
621 do_optimize_kprobes();
623 /* Step 4: Free cleaned kprobes after quiesence period */
624 do_free_cleaned_kprobes();
626 mutex_unlock(&text_mutex);
629 /* Step 5: Kick optimizer again if needed */
630 if (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list))
631 kick_kprobe_optimizer();
633 mutex_unlock(&kprobe_mutex);
636 /* Wait for completing optimization and unoptimization */
637 void wait_for_kprobe_optimizer(void)
639 mutex_lock(&kprobe_mutex);
641 while (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list)) {
642 mutex_unlock(&kprobe_mutex);
644 /* this will also make optimizing_work execute immmediately */
645 flush_delayed_work(&optimizing_work);
646 /* @optimizing_work might not have been queued yet, relax */
649 mutex_lock(&kprobe_mutex);
652 mutex_unlock(&kprobe_mutex);
655 static bool optprobe_queued_unopt(struct optimized_kprobe *op)
657 struct optimized_kprobe *_op;
659 list_for_each_entry(_op, &unoptimizing_list, list) {
667 /* Optimize kprobe if p is ready to be optimized */
668 static void optimize_kprobe(struct kprobe *p)
670 struct optimized_kprobe *op;
672 /* Check if the kprobe is disabled or not ready for optimization. */
673 if (!kprobe_optready(p) || !kprobes_allow_optimization ||
674 (kprobe_disabled(p) || kprobes_all_disarmed))
677 /* kprobes with post_handler can not be optimized */
681 op = container_of(p, struct optimized_kprobe, kp);
683 /* Check there is no other kprobes at the optimized instructions */
684 if (arch_check_optimized_kprobe(op) < 0)
687 /* Check if it is already optimized. */
688 if (op->kp.flags & KPROBE_FLAG_OPTIMIZED) {
689 if (optprobe_queued_unopt(op)) {
690 /* This is under unoptimizing. Just dequeue the probe */
691 list_del_init(&op->list);
695 op->kp.flags |= KPROBE_FLAG_OPTIMIZED;
697 /* On unoptimizing/optimizing_list, op must have OPTIMIZED flag */
698 if (WARN_ON_ONCE(!list_empty(&op->list)))
701 list_add(&op->list, &optimizing_list);
702 kick_kprobe_optimizer();
705 /* Short cut to direct unoptimizing */
706 static void force_unoptimize_kprobe(struct optimized_kprobe *op)
708 lockdep_assert_cpus_held();
709 arch_unoptimize_kprobe(op);
710 op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
713 /* Unoptimize a kprobe if p is optimized */
714 static void unoptimize_kprobe(struct kprobe *p, bool force)
716 struct optimized_kprobe *op;
718 if (!kprobe_aggrprobe(p) || kprobe_disarmed(p))
719 return; /* This is not an optprobe nor optimized */
721 op = container_of(p, struct optimized_kprobe, kp);
722 if (!kprobe_optimized(p))
725 if (!list_empty(&op->list)) {
726 if (optprobe_queued_unopt(op)) {
727 /* Queued in unoptimizing queue */
730 * Forcibly unoptimize the kprobe here, and queue it
731 * in the freeing list for release afterwards.
733 force_unoptimize_kprobe(op);
734 list_move(&op->list, &freeing_list);
737 /* Dequeue from the optimizing queue */
738 list_del_init(&op->list);
739 op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
744 /* Optimized kprobe case */
746 /* Forcibly update the code: this is a special case */
747 force_unoptimize_kprobe(op);
749 list_add(&op->list, &unoptimizing_list);
750 kick_kprobe_optimizer();
754 /* Cancel unoptimizing for reusing */
755 static int reuse_unused_kprobe(struct kprobe *ap)
757 struct optimized_kprobe *op;
760 * Unused kprobe MUST be on the way of delayed unoptimizing (means
761 * there is still a relative jump) and disabled.
763 op = container_of(ap, struct optimized_kprobe, kp);
764 WARN_ON_ONCE(list_empty(&op->list));
765 /* Enable the probe again */
766 ap->flags &= ~KPROBE_FLAG_DISABLED;
767 /* Optimize it again (remove from op->list) */
768 if (!kprobe_optready(ap))
775 /* Remove optimized instructions */
776 static void kill_optimized_kprobe(struct kprobe *p)
778 struct optimized_kprobe *op;
780 op = container_of(p, struct optimized_kprobe, kp);
781 if (!list_empty(&op->list))
782 /* Dequeue from the (un)optimization queue */
783 list_del_init(&op->list);
784 op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
786 if (kprobe_unused(p)) {
787 /* Enqueue if it is unused */
788 list_add(&op->list, &freeing_list);
790 * Remove unused probes from the hash list. After waiting
791 * for synchronization, this probe is reclaimed.
792 * (reclaiming is done by do_free_cleaned_kprobes().)
794 hlist_del_rcu(&op->kp.hlist);
797 /* Don't touch the code, because it is already freed. */
798 arch_remove_optimized_kprobe(op);
802 void __prepare_optimized_kprobe(struct optimized_kprobe *op, struct kprobe *p)
804 if (!kprobe_ftrace(p))
805 arch_prepare_optimized_kprobe(op, p);
808 /* Try to prepare optimized instructions */
809 static void prepare_optimized_kprobe(struct kprobe *p)
811 struct optimized_kprobe *op;
813 op = container_of(p, struct optimized_kprobe, kp);
814 __prepare_optimized_kprobe(op, p);
817 /* Allocate new optimized_kprobe and try to prepare optimized instructions */
818 static struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
820 struct optimized_kprobe *op;
822 op = kzalloc(sizeof(struct optimized_kprobe), GFP_KERNEL);
826 INIT_LIST_HEAD(&op->list);
827 op->kp.addr = p->addr;
828 __prepare_optimized_kprobe(op, p);
833 static void init_aggr_kprobe(struct kprobe *ap, struct kprobe *p);
836 * Prepare an optimized_kprobe and optimize it
837 * NOTE: p must be a normal registered kprobe
839 static void try_to_optimize_kprobe(struct kprobe *p)
842 struct optimized_kprobe *op;
844 /* Impossible to optimize ftrace-based kprobe */
845 if (kprobe_ftrace(p))
848 /* For preparing optimization, jump_label_text_reserved() is called */
851 mutex_lock(&text_mutex);
853 ap = alloc_aggr_kprobe(p);
857 op = container_of(ap, struct optimized_kprobe, kp);
858 if (!arch_prepared_optinsn(&op->optinsn)) {
859 /* If failed to setup optimizing, fallback to kprobe */
860 arch_remove_optimized_kprobe(op);
865 init_aggr_kprobe(ap, p);
866 optimize_kprobe(ap); /* This just kicks optimizer thread */
869 mutex_unlock(&text_mutex);
874 static void optimize_all_kprobes(void)
876 struct hlist_head *head;
880 mutex_lock(&kprobe_mutex);
881 /* If optimization is already allowed, just return */
882 if (kprobes_allow_optimization)
886 kprobes_allow_optimization = true;
887 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
888 head = &kprobe_table[i];
889 hlist_for_each_entry(p, head, hlist)
890 if (!kprobe_disabled(p))
894 printk(KERN_INFO "Kprobes globally optimized\n");
896 mutex_unlock(&kprobe_mutex);
900 static void unoptimize_all_kprobes(void)
902 struct hlist_head *head;
906 mutex_lock(&kprobe_mutex);
907 /* If optimization is already prohibited, just return */
908 if (!kprobes_allow_optimization) {
909 mutex_unlock(&kprobe_mutex);
914 kprobes_allow_optimization = false;
915 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
916 head = &kprobe_table[i];
917 hlist_for_each_entry(p, head, hlist) {
918 if (!kprobe_disabled(p))
919 unoptimize_kprobe(p, false);
923 mutex_unlock(&kprobe_mutex);
925 /* Wait for unoptimizing completion */
926 wait_for_kprobe_optimizer();
927 printk(KERN_INFO "Kprobes globally unoptimized\n");
930 static DEFINE_MUTEX(kprobe_sysctl_mutex);
931 int sysctl_kprobes_optimization;
932 int proc_kprobes_optimization_handler(struct ctl_table *table, int write,
933 void *buffer, size_t *length,
938 mutex_lock(&kprobe_sysctl_mutex);
939 sysctl_kprobes_optimization = kprobes_allow_optimization ? 1 : 0;
940 ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
942 if (sysctl_kprobes_optimization)
943 optimize_all_kprobes();
945 unoptimize_all_kprobes();
946 mutex_unlock(&kprobe_sysctl_mutex);
950 #endif /* CONFIG_SYSCTL */
952 /* Put a breakpoint for a probe. Must be called with text_mutex locked */
953 static void __arm_kprobe(struct kprobe *p)
957 /* Check collision with other optimized kprobes */
958 _p = get_optimized_kprobe((unsigned long)p->addr);
960 /* Fallback to unoptimized kprobe */
961 unoptimize_kprobe(_p, true);
964 optimize_kprobe(p); /* Try to optimize (add kprobe to a list) */
967 /* Remove the breakpoint of a probe. Must be called with text_mutex locked */
968 static void __disarm_kprobe(struct kprobe *p, bool reopt)
972 /* Try to unoptimize */
973 unoptimize_kprobe(p, kprobes_all_disarmed);
975 if (!kprobe_queued(p)) {
976 arch_disarm_kprobe(p);
977 /* If another kprobe was blocked, optimize it. */
978 _p = get_optimized_kprobe((unsigned long)p->addr);
979 if (unlikely(_p) && reopt)
982 /* TODO: reoptimize others after unoptimized this probe */
985 #else /* !CONFIG_OPTPROBES */
987 #define optimize_kprobe(p) do {} while (0)
988 #define unoptimize_kprobe(p, f) do {} while (0)
989 #define kill_optimized_kprobe(p) do {} while (0)
990 #define prepare_optimized_kprobe(p) do {} while (0)
991 #define try_to_optimize_kprobe(p) do {} while (0)
992 #define __arm_kprobe(p) arch_arm_kprobe(p)
993 #define __disarm_kprobe(p, o) arch_disarm_kprobe(p)
994 #define kprobe_disarmed(p) kprobe_disabled(p)
995 #define wait_for_kprobe_optimizer() do {} while (0)
997 static int reuse_unused_kprobe(struct kprobe *ap)
1000 * If the optimized kprobe is NOT supported, the aggr kprobe is
1001 * released at the same time that the last aggregated kprobe is
1003 * Thus there should be no chance to reuse unused kprobe.
1005 printk(KERN_ERR "Error: There should be no unused kprobe here.\n");
1009 static void free_aggr_kprobe(struct kprobe *p)
1011 arch_remove_kprobe(p);
1015 static struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
1017 return kzalloc(sizeof(struct kprobe), GFP_KERNEL);
1019 #endif /* CONFIG_OPTPROBES */
1021 #ifdef CONFIG_KPROBES_ON_FTRACE
1022 static struct ftrace_ops kprobe_ftrace_ops __read_mostly = {
1023 .func = kprobe_ftrace_handler,
1024 .flags = FTRACE_OPS_FL_SAVE_REGS,
1027 static struct ftrace_ops kprobe_ipmodify_ops __read_mostly = {
1028 .func = kprobe_ftrace_handler,
1029 .flags = FTRACE_OPS_FL_SAVE_REGS | FTRACE_OPS_FL_IPMODIFY,
1032 static int kprobe_ipmodify_enabled;
1033 static int kprobe_ftrace_enabled;
1035 /* Must ensure p->addr is really on ftrace */
1036 static int prepare_kprobe(struct kprobe *p)
1038 if (!kprobe_ftrace(p))
1039 return arch_prepare_kprobe(p);
1041 return arch_prepare_kprobe_ftrace(p);
1044 /* Caller must lock kprobe_mutex */
1045 static int __arm_kprobe_ftrace(struct kprobe *p, struct ftrace_ops *ops,
1050 ret = ftrace_set_filter_ip(ops, (unsigned long)p->addr, 0, 0);
1052 pr_debug("Failed to arm kprobe-ftrace at %pS (%d)\n",
1058 ret = register_ftrace_function(ops);
1060 pr_debug("Failed to init kprobe-ftrace (%d)\n", ret);
1070 * At this point, sinec ops is not registered, we should be sefe from
1071 * registering empty filter.
1073 ftrace_set_filter_ip(ops, (unsigned long)p->addr, 1, 0);
1077 static int arm_kprobe_ftrace(struct kprobe *p)
1079 bool ipmodify = (p->post_handler != NULL);
1081 return __arm_kprobe_ftrace(p,
1082 ipmodify ? &kprobe_ipmodify_ops : &kprobe_ftrace_ops,
1083 ipmodify ? &kprobe_ipmodify_enabled : &kprobe_ftrace_enabled);
1086 /* Caller must lock kprobe_mutex */
1087 static int __disarm_kprobe_ftrace(struct kprobe *p, struct ftrace_ops *ops,
1093 ret = unregister_ftrace_function(ops);
1094 if (WARN(ret < 0, "Failed to unregister kprobe-ftrace (%d)\n", ret))
1100 ret = ftrace_set_filter_ip(ops, (unsigned long)p->addr, 1, 0);
1101 WARN_ONCE(ret < 0, "Failed to disarm kprobe-ftrace at %pS (%d)\n",
1106 static int disarm_kprobe_ftrace(struct kprobe *p)
1108 bool ipmodify = (p->post_handler != NULL);
1110 return __disarm_kprobe_ftrace(p,
1111 ipmodify ? &kprobe_ipmodify_ops : &kprobe_ftrace_ops,
1112 ipmodify ? &kprobe_ipmodify_enabled : &kprobe_ftrace_enabled);
1114 #else /* !CONFIG_KPROBES_ON_FTRACE */
1115 static inline int prepare_kprobe(struct kprobe *p)
1117 return arch_prepare_kprobe(p);
1120 static inline int arm_kprobe_ftrace(struct kprobe *p)
1125 static inline int disarm_kprobe_ftrace(struct kprobe *p)
1131 /* Arm a kprobe with text_mutex */
1132 static int arm_kprobe(struct kprobe *kp)
1134 if (unlikely(kprobe_ftrace(kp)))
1135 return arm_kprobe_ftrace(kp);
1138 mutex_lock(&text_mutex);
1140 mutex_unlock(&text_mutex);
1146 /* Disarm a kprobe with text_mutex */
1147 static int disarm_kprobe(struct kprobe *kp, bool reopt)
1149 if (unlikely(kprobe_ftrace(kp)))
1150 return disarm_kprobe_ftrace(kp);
1153 mutex_lock(&text_mutex);
1154 __disarm_kprobe(kp, reopt);
1155 mutex_unlock(&text_mutex);
1162 * Aggregate handlers for multiple kprobes support - these handlers
1163 * take care of invoking the individual kprobe handlers on p->list
1165 static int aggr_pre_handler(struct kprobe *p, struct pt_regs *regs)
1169 list_for_each_entry_rcu(kp, &p->list, list) {
1170 if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
1171 set_kprobe_instance(kp);
1172 if (kp->pre_handler(kp, regs))
1175 reset_kprobe_instance();
1179 NOKPROBE_SYMBOL(aggr_pre_handler);
1181 static void aggr_post_handler(struct kprobe *p, struct pt_regs *regs,
1182 unsigned long flags)
1186 list_for_each_entry_rcu(kp, &p->list, list) {
1187 if (kp->post_handler && likely(!kprobe_disabled(kp))) {
1188 set_kprobe_instance(kp);
1189 kp->post_handler(kp, regs, flags);
1190 reset_kprobe_instance();
1194 NOKPROBE_SYMBOL(aggr_post_handler);
1196 static int aggr_fault_handler(struct kprobe *p, struct pt_regs *regs,
1199 struct kprobe *cur = __this_cpu_read(kprobe_instance);
1202 * if we faulted "during" the execution of a user specified
1203 * probe handler, invoke just that probe's fault handler
1205 if (cur && cur->fault_handler) {
1206 if (cur->fault_handler(cur, regs, trapnr))
1211 NOKPROBE_SYMBOL(aggr_fault_handler);
1213 /* Walks the list and increments nmissed count for multiprobe case */
1214 void kprobes_inc_nmissed_count(struct kprobe *p)
1217 if (!kprobe_aggrprobe(p)) {
1220 list_for_each_entry_rcu(kp, &p->list, list)
1225 NOKPROBE_SYMBOL(kprobes_inc_nmissed_count);
1227 static void free_rp_inst_rcu(struct rcu_head *head)
1229 struct kretprobe_instance *ri = container_of(head, struct kretprobe_instance, rcu);
1231 if (refcount_dec_and_test(&ri->rph->ref))
1235 NOKPROBE_SYMBOL(free_rp_inst_rcu);
1237 static void recycle_rp_inst(struct kretprobe_instance *ri)
1239 struct kretprobe *rp = get_kretprobe(ri);
1242 freelist_add(&ri->freelist, &rp->freelist);
1244 call_rcu(&ri->rcu, free_rp_inst_rcu);
1246 NOKPROBE_SYMBOL(recycle_rp_inst);
1248 static struct kprobe kprobe_busy = {
1249 .addr = (void *) get_kprobe,
1252 void kprobe_busy_begin(void)
1254 struct kprobe_ctlblk *kcb;
1257 __this_cpu_write(current_kprobe, &kprobe_busy);
1258 kcb = get_kprobe_ctlblk();
1259 kcb->kprobe_status = KPROBE_HIT_ACTIVE;
1262 void kprobe_busy_end(void)
1264 __this_cpu_write(current_kprobe, NULL);
1269 * This function is called from finish_task_switch when task tk becomes dead,
1270 * so that we can recycle any function-return probe instances associated
1271 * with this task. These left over instances represent probed functions
1272 * that have been called but will never return.
1274 void kprobe_flush_task(struct task_struct *tk)
1276 struct kretprobe_instance *ri;
1277 struct llist_node *node;
1279 /* Early boot, not yet initialized. */
1280 if (unlikely(!kprobes_initialized))
1283 kprobe_busy_begin();
1285 node = __llist_del_all(&tk->kretprobe_instances);
1287 ri = container_of(node, struct kretprobe_instance, llist);
1290 recycle_rp_inst(ri);
1295 NOKPROBE_SYMBOL(kprobe_flush_task);
1297 static inline void free_rp_inst(struct kretprobe *rp)
1299 struct kretprobe_instance *ri;
1300 struct freelist_node *node;
1303 node = rp->freelist.head;
1305 ri = container_of(node, struct kretprobe_instance, freelist);
1312 if (refcount_sub_and_test(count, &rp->rph->ref)) {
1318 /* Add the new probe to ap->list */
1319 static int add_new_kprobe(struct kprobe *ap, struct kprobe *p)
1321 if (p->post_handler)
1322 unoptimize_kprobe(ap, true); /* Fall back to normal kprobe */
1324 list_add_rcu(&p->list, &ap->list);
1325 if (p->post_handler && !ap->post_handler)
1326 ap->post_handler = aggr_post_handler;
1332 * Fill in the required fields of the "manager kprobe". Replace the
1333 * earlier kprobe in the hlist with the manager kprobe
1335 static void init_aggr_kprobe(struct kprobe *ap, struct kprobe *p)
1337 /* Copy p's insn slot to ap */
1339 flush_insn_slot(ap);
1341 ap->flags = p->flags & ~KPROBE_FLAG_OPTIMIZED;
1342 ap->pre_handler = aggr_pre_handler;
1343 ap->fault_handler = aggr_fault_handler;
1344 /* We don't care the kprobe which has gone. */
1345 if (p->post_handler && !kprobe_gone(p))
1346 ap->post_handler = aggr_post_handler;
1348 INIT_LIST_HEAD(&ap->list);
1349 INIT_HLIST_NODE(&ap->hlist);
1351 list_add_rcu(&p->list, &ap->list);
1352 hlist_replace_rcu(&p->hlist, &ap->hlist);
1356 * This is the second or subsequent kprobe at the address - handle
1359 static int register_aggr_kprobe(struct kprobe *orig_p, struct kprobe *p)
1362 struct kprobe *ap = orig_p;
1366 /* For preparing optimization, jump_label_text_reserved() is called */
1368 mutex_lock(&text_mutex);
1370 if (!kprobe_aggrprobe(orig_p)) {
1371 /* If orig_p is not an aggr_kprobe, create new aggr_kprobe. */
1372 ap = alloc_aggr_kprobe(orig_p);
1377 init_aggr_kprobe(ap, orig_p);
1378 } else if (kprobe_unused(ap)) {
1379 /* This probe is going to die. Rescue it */
1380 ret = reuse_unused_kprobe(ap);
1385 if (kprobe_gone(ap)) {
1387 * Attempting to insert new probe at the same location that
1388 * had a probe in the module vaddr area which already
1389 * freed. So, the instruction slot has already been
1390 * released. We need a new slot for the new probe.
1392 ret = arch_prepare_kprobe(ap);
1395 * Even if fail to allocate new slot, don't need to
1396 * free aggr_probe. It will be used next time, or
1397 * freed by unregister_kprobe.
1401 /* Prepare optimized instructions if possible. */
1402 prepare_optimized_kprobe(ap);
1405 * Clear gone flag to prevent allocating new slot again, and
1406 * set disabled flag because it is not armed yet.
1408 ap->flags = (ap->flags & ~KPROBE_FLAG_GONE)
1409 | KPROBE_FLAG_DISABLED;
1412 /* Copy ap's insn slot to p */
1414 ret = add_new_kprobe(ap, p);
1417 mutex_unlock(&text_mutex);
1418 jump_label_unlock();
1421 if (ret == 0 && kprobe_disabled(ap) && !kprobe_disabled(p)) {
1422 ap->flags &= ~KPROBE_FLAG_DISABLED;
1423 if (!kprobes_all_disarmed) {
1424 /* Arm the breakpoint again. */
1425 ret = arm_kprobe(ap);
1427 ap->flags |= KPROBE_FLAG_DISABLED;
1428 list_del_rcu(&p->list);
1436 bool __weak arch_within_kprobe_blacklist(unsigned long addr)
1438 /* The __kprobes marked functions and entry code must not be probed */
1439 return addr >= (unsigned long)__kprobes_text_start &&
1440 addr < (unsigned long)__kprobes_text_end;
1443 static bool __within_kprobe_blacklist(unsigned long addr)
1445 struct kprobe_blacklist_entry *ent;
1447 if (arch_within_kprobe_blacklist(addr))
1450 * If there exists a kprobe_blacklist, verify and
1451 * fail any probe registration in the prohibited area
1453 list_for_each_entry(ent, &kprobe_blacklist, list) {
1454 if (addr >= ent->start_addr && addr < ent->end_addr)
1460 bool within_kprobe_blacklist(unsigned long addr)
1462 char symname[KSYM_NAME_LEN], *p;
1464 if (__within_kprobe_blacklist(addr))
1467 /* Check if the address is on a suffixed-symbol */
1468 if (!lookup_symbol_name(addr, symname)) {
1469 p = strchr(symname, '.');
1473 addr = (unsigned long)kprobe_lookup_name(symname, 0);
1475 return __within_kprobe_blacklist(addr);
1481 * If we have a symbol_name argument, look it up and add the offset field
1482 * to it. This way, we can specify a relative address to a symbol.
1483 * This returns encoded errors if it fails to look up symbol or invalid
1484 * combination of parameters.
1486 static kprobe_opcode_t *_kprobe_addr(kprobe_opcode_t *addr,
1487 const char *symbol_name, unsigned int offset)
1489 if ((symbol_name && addr) || (!symbol_name && !addr))
1493 addr = kprobe_lookup_name(symbol_name, offset);
1495 return ERR_PTR(-ENOENT);
1498 addr = (kprobe_opcode_t *)(((char *)addr) + offset);
1503 return ERR_PTR(-EINVAL);
1506 static kprobe_opcode_t *kprobe_addr(struct kprobe *p)
1508 return _kprobe_addr(p->addr, p->symbol_name, p->offset);
1511 /* Check passed kprobe is valid and return kprobe in kprobe_table. */
1512 static struct kprobe *__get_valid_kprobe(struct kprobe *p)
1514 struct kprobe *ap, *list_p;
1516 lockdep_assert_held(&kprobe_mutex);
1518 ap = get_kprobe(p->addr);
1523 list_for_each_entry(list_p, &ap->list, list)
1525 /* kprobe p is a valid probe */
1534 * Warn and return error if the kprobe is being re-registered since
1535 * there must be a software bug.
1537 static inline int warn_kprobe_rereg(struct kprobe *p)
1541 mutex_lock(&kprobe_mutex);
1542 if (WARN_ON_ONCE(__get_valid_kprobe(p)))
1544 mutex_unlock(&kprobe_mutex);
1549 int __weak arch_check_ftrace_location(struct kprobe *p)
1551 unsigned long ftrace_addr;
1553 ftrace_addr = ftrace_location((unsigned long)p->addr);
1555 #ifdef CONFIG_KPROBES_ON_FTRACE
1556 /* Given address is not on the instruction boundary */
1557 if ((unsigned long)p->addr != ftrace_addr)
1559 p->flags |= KPROBE_FLAG_FTRACE;
1560 #else /* !CONFIG_KPROBES_ON_FTRACE */
1567 static int check_kprobe_address_safe(struct kprobe *p,
1568 struct module **probed_mod)
1572 ret = arch_check_ftrace_location(p);
1578 /* Ensure it is not in reserved area nor out of text */
1579 if (!kernel_text_address((unsigned long) p->addr) ||
1580 within_kprobe_blacklist((unsigned long) p->addr) ||
1581 jump_label_text_reserved(p->addr, p->addr) ||
1582 find_bug((unsigned long)p->addr)) {
1587 /* Check if are we probing a module */
1588 *probed_mod = __module_text_address((unsigned long) p->addr);
1591 * We must hold a refcount of the probed module while updating
1592 * its code to prohibit unexpected unloading.
1594 if (unlikely(!try_module_get(*probed_mod))) {
1600 * If the module freed .init.text, we couldn't insert
1603 if (within_module_init((unsigned long)p->addr, *probed_mod) &&
1604 (*probed_mod)->state != MODULE_STATE_COMING) {
1605 module_put(*probed_mod);
1612 jump_label_unlock();
1617 int register_kprobe(struct kprobe *p)
1620 struct kprobe *old_p;
1621 struct module *probed_mod;
1622 kprobe_opcode_t *addr;
1624 /* Adjust probe address from symbol */
1625 addr = kprobe_addr(p);
1627 return PTR_ERR(addr);
1630 ret = warn_kprobe_rereg(p);
1634 /* User can pass only KPROBE_FLAG_DISABLED to register_kprobe */
1635 p->flags &= KPROBE_FLAG_DISABLED;
1637 INIT_LIST_HEAD(&p->list);
1639 ret = check_kprobe_address_safe(p, &probed_mod);
1643 mutex_lock(&kprobe_mutex);
1645 old_p = get_kprobe(p->addr);
1647 /* Since this may unoptimize old_p, locking text_mutex. */
1648 ret = register_aggr_kprobe(old_p, p);
1653 /* Prevent text modification */
1654 mutex_lock(&text_mutex);
1655 ret = prepare_kprobe(p);
1656 mutex_unlock(&text_mutex);
1661 INIT_HLIST_NODE(&p->hlist);
1662 hlist_add_head_rcu(&p->hlist,
1663 &kprobe_table[hash_ptr(p->addr, KPROBE_HASH_BITS)]);
1665 if (!kprobes_all_disarmed && !kprobe_disabled(p)) {
1666 ret = arm_kprobe(p);
1668 hlist_del_rcu(&p->hlist);
1674 /* Try to optimize kprobe */
1675 try_to_optimize_kprobe(p);
1677 mutex_unlock(&kprobe_mutex);
1680 module_put(probed_mod);
1684 EXPORT_SYMBOL_GPL(register_kprobe);
1686 /* Check if all probes on the aggrprobe are disabled */
1687 static int aggr_kprobe_disabled(struct kprobe *ap)
1691 lockdep_assert_held(&kprobe_mutex);
1693 list_for_each_entry(kp, &ap->list, list)
1694 if (!kprobe_disabled(kp))
1696 * There is an active probe on the list.
1697 * We can't disable this ap.
1704 /* Disable one kprobe: Make sure called under kprobe_mutex is locked */
1705 static struct kprobe *__disable_kprobe(struct kprobe *p)
1707 struct kprobe *orig_p;
1710 /* Get an original kprobe for return */
1711 orig_p = __get_valid_kprobe(p);
1712 if (unlikely(orig_p == NULL))
1713 return ERR_PTR(-EINVAL);
1715 if (!kprobe_disabled(p)) {
1716 /* Disable probe if it is a child probe */
1718 p->flags |= KPROBE_FLAG_DISABLED;
1720 /* Try to disarm and disable this/parent probe */
1721 if (p == orig_p || aggr_kprobe_disabled(orig_p)) {
1723 * If kprobes_all_disarmed is set, orig_p
1724 * should have already been disarmed, so
1725 * skip unneed disarming process.
1727 if (!kprobes_all_disarmed) {
1728 ret = disarm_kprobe(orig_p, true);
1730 p->flags &= ~KPROBE_FLAG_DISABLED;
1731 return ERR_PTR(ret);
1734 orig_p->flags |= KPROBE_FLAG_DISABLED;
1742 * Unregister a kprobe without a scheduler synchronization.
1744 static int __unregister_kprobe_top(struct kprobe *p)
1746 struct kprobe *ap, *list_p;
1748 /* Disable kprobe. This will disarm it if needed. */
1749 ap = __disable_kprobe(p);
1755 * This probe is an independent(and non-optimized) kprobe
1756 * (not an aggrprobe). Remove from the hash list.
1760 /* Following process expects this probe is an aggrprobe */
1761 WARN_ON(!kprobe_aggrprobe(ap));
1763 if (list_is_singular(&ap->list) && kprobe_disarmed(ap))
1765 * !disarmed could be happen if the probe is under delayed
1770 /* If disabling probe has special handlers, update aggrprobe */
1771 if (p->post_handler && !kprobe_gone(p)) {
1772 list_for_each_entry(list_p, &ap->list, list) {
1773 if ((list_p != p) && (list_p->post_handler))
1776 ap->post_handler = NULL;
1780 * Remove from the aggrprobe: this path will do nothing in
1781 * __unregister_kprobe_bottom().
1783 list_del_rcu(&p->list);
1784 if (!kprobe_disabled(ap) && !kprobes_all_disarmed)
1786 * Try to optimize this probe again, because post
1787 * handler may have been changed.
1789 optimize_kprobe(ap);
1794 hlist_del_rcu(&ap->hlist);
1798 static void __unregister_kprobe_bottom(struct kprobe *p)
1802 if (list_empty(&p->list))
1803 /* This is an independent kprobe */
1804 arch_remove_kprobe(p);
1805 else if (list_is_singular(&p->list)) {
1806 /* This is the last child of an aggrprobe */
1807 ap = list_entry(p->list.next, struct kprobe, list);
1809 free_aggr_kprobe(ap);
1811 /* Otherwise, do nothing. */
1814 int register_kprobes(struct kprobe **kps, int num)
1820 for (i = 0; i < num; i++) {
1821 ret = register_kprobe(kps[i]);
1824 unregister_kprobes(kps, i);
1830 EXPORT_SYMBOL_GPL(register_kprobes);
1832 void unregister_kprobe(struct kprobe *p)
1834 unregister_kprobes(&p, 1);
1836 EXPORT_SYMBOL_GPL(unregister_kprobe);
1838 void unregister_kprobes(struct kprobe **kps, int num)
1844 mutex_lock(&kprobe_mutex);
1845 for (i = 0; i < num; i++)
1846 if (__unregister_kprobe_top(kps[i]) < 0)
1847 kps[i]->addr = NULL;
1848 mutex_unlock(&kprobe_mutex);
1851 for (i = 0; i < num; i++)
1853 __unregister_kprobe_bottom(kps[i]);
1855 EXPORT_SYMBOL_GPL(unregister_kprobes);
1857 int __weak kprobe_exceptions_notify(struct notifier_block *self,
1858 unsigned long val, void *data)
1862 NOKPROBE_SYMBOL(kprobe_exceptions_notify);
1864 static struct notifier_block kprobe_exceptions_nb = {
1865 .notifier_call = kprobe_exceptions_notify,
1866 .priority = 0x7fffffff /* we need to be notified first */
1869 unsigned long __weak arch_deref_entry_point(void *entry)
1871 return (unsigned long)entry;
1874 #ifdef CONFIG_KRETPROBES
1876 unsigned long __kretprobe_trampoline_handler(struct pt_regs *regs,
1877 void *trampoline_address,
1878 void *frame_pointer)
1880 kprobe_opcode_t *correct_ret_addr = NULL;
1881 struct kretprobe_instance *ri = NULL;
1882 struct llist_node *first, *node;
1883 struct kretprobe *rp;
1885 /* Find all nodes for this frame. */
1886 first = node = current->kretprobe_instances.first;
1888 ri = container_of(node, struct kretprobe_instance, llist);
1890 BUG_ON(ri->fp != frame_pointer);
1892 if (ri->ret_addr != trampoline_address) {
1893 correct_ret_addr = ri->ret_addr;
1895 * This is the real return address. Any other
1896 * instances associated with this task are for
1897 * other calls deeper on the call stack
1904 pr_err("Oops! Kretprobe fails to find correct return address.\n");
1908 /* Unlink all nodes for this frame. */
1909 current->kretprobe_instances.first = node->next;
1914 ri = container_of(first, struct kretprobe_instance, llist);
1915 first = first->next;
1917 rp = get_kretprobe(ri);
1918 if (rp && rp->handler) {
1919 struct kprobe *prev = kprobe_running();
1921 __this_cpu_write(current_kprobe, &rp->kp);
1922 ri->ret_addr = correct_ret_addr;
1923 rp->handler(ri, regs);
1924 __this_cpu_write(current_kprobe, prev);
1927 recycle_rp_inst(ri);
1930 return (unsigned long)correct_ret_addr;
1932 NOKPROBE_SYMBOL(__kretprobe_trampoline_handler)
1935 * This kprobe pre_handler is registered with every kretprobe. When probe
1936 * hits it will set up the return probe.
1938 static int pre_handler_kretprobe(struct kprobe *p, struct pt_regs *regs)
1940 struct kretprobe *rp = container_of(p, struct kretprobe, kp);
1941 struct kretprobe_instance *ri;
1942 struct freelist_node *fn;
1944 fn = freelist_try_get(&rp->freelist);
1950 ri = container_of(fn, struct kretprobe_instance, freelist);
1952 if (rp->entry_handler && rp->entry_handler(ri, regs)) {
1953 freelist_add(&ri->freelist, &rp->freelist);
1957 arch_prepare_kretprobe(ri, regs);
1959 __llist_add(&ri->llist, ¤t->kretprobe_instances);
1963 NOKPROBE_SYMBOL(pre_handler_kretprobe);
1965 bool __weak arch_kprobe_on_func_entry(unsigned long offset)
1971 * kprobe_on_func_entry() -- check whether given address is function entry
1972 * @addr: Target address
1973 * @sym: Target symbol name
1974 * @offset: The offset from the symbol or the address
1976 * This checks whether the given @addr+@offset or @sym+@offset is on the
1977 * function entry address or not.
1978 * This returns 0 if it is the function entry, or -EINVAL if it is not.
1979 * And also it returns -ENOENT if it fails the symbol or address lookup.
1980 * Caller must pass @addr or @sym (either one must be NULL), or this
1983 int kprobe_on_func_entry(kprobe_opcode_t *addr, const char *sym, unsigned long offset)
1985 kprobe_opcode_t *kp_addr = _kprobe_addr(addr, sym, offset);
1987 if (IS_ERR(kp_addr))
1988 return PTR_ERR(kp_addr);
1990 if (!kallsyms_lookup_size_offset((unsigned long)kp_addr, NULL, &offset))
1993 if (!arch_kprobe_on_func_entry(offset))
1999 int register_kretprobe(struct kretprobe *rp)
2002 struct kretprobe_instance *inst;
2006 ret = kprobe_on_func_entry(rp->kp.addr, rp->kp.symbol_name, rp->kp.offset);
2010 /* If only rp->kp.addr is specified, check reregistering kprobes */
2011 if (rp->kp.addr && warn_kprobe_rereg(&rp->kp))
2014 if (kretprobe_blacklist_size) {
2015 addr = kprobe_addr(&rp->kp);
2017 return PTR_ERR(addr);
2019 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
2020 if (kretprobe_blacklist[i].addr == addr)
2025 rp->kp.pre_handler = pre_handler_kretprobe;
2026 rp->kp.post_handler = NULL;
2027 rp->kp.fault_handler = NULL;
2029 /* Pre-allocate memory for max kretprobe instances */
2030 if (rp->maxactive <= 0) {
2031 #ifdef CONFIG_PREEMPTION
2032 rp->maxactive = max_t(unsigned int, 10, 2*num_possible_cpus());
2034 rp->maxactive = num_possible_cpus();
2037 rp->freelist.head = NULL;
2038 rp->rph = kzalloc(sizeof(struct kretprobe_holder), GFP_KERNEL);
2043 for (i = 0; i < rp->maxactive; i++) {
2044 inst = kzalloc(sizeof(struct kretprobe_instance) +
2045 rp->data_size, GFP_KERNEL);
2047 refcount_set(&rp->rph->ref, i);
2051 inst->rph = rp->rph;
2052 freelist_add(&inst->freelist, &rp->freelist);
2054 refcount_set(&rp->rph->ref, i);
2057 /* Establish function entry probe point */
2058 ret = register_kprobe(&rp->kp);
2063 EXPORT_SYMBOL_GPL(register_kretprobe);
2065 int register_kretprobes(struct kretprobe **rps, int num)
2071 for (i = 0; i < num; i++) {
2072 ret = register_kretprobe(rps[i]);
2075 unregister_kretprobes(rps, i);
2081 EXPORT_SYMBOL_GPL(register_kretprobes);
2083 void unregister_kretprobe(struct kretprobe *rp)
2085 unregister_kretprobes(&rp, 1);
2087 EXPORT_SYMBOL_GPL(unregister_kretprobe);
2089 void unregister_kretprobes(struct kretprobe **rps, int num)
2095 mutex_lock(&kprobe_mutex);
2096 for (i = 0; i < num; i++) {
2097 if (__unregister_kprobe_top(&rps[i]->kp) < 0)
2098 rps[i]->kp.addr = NULL;
2099 rps[i]->rph->rp = NULL;
2101 mutex_unlock(&kprobe_mutex);
2104 for (i = 0; i < num; i++) {
2105 if (rps[i]->kp.addr) {
2106 __unregister_kprobe_bottom(&rps[i]->kp);
2107 free_rp_inst(rps[i]);
2111 EXPORT_SYMBOL_GPL(unregister_kretprobes);
2113 #else /* CONFIG_KRETPROBES */
2114 int register_kretprobe(struct kretprobe *rp)
2118 EXPORT_SYMBOL_GPL(register_kretprobe);
2120 int register_kretprobes(struct kretprobe **rps, int num)
2124 EXPORT_SYMBOL_GPL(register_kretprobes);
2126 void unregister_kretprobe(struct kretprobe *rp)
2129 EXPORT_SYMBOL_GPL(unregister_kretprobe);
2131 void unregister_kretprobes(struct kretprobe **rps, int num)
2134 EXPORT_SYMBOL_GPL(unregister_kretprobes);
2136 static int pre_handler_kretprobe(struct kprobe *p, struct pt_regs *regs)
2140 NOKPROBE_SYMBOL(pre_handler_kretprobe);
2142 #endif /* CONFIG_KRETPROBES */
2144 /* Set the kprobe gone and remove its instruction buffer. */
2145 static void kill_kprobe(struct kprobe *p)
2149 lockdep_assert_held(&kprobe_mutex);
2151 p->flags |= KPROBE_FLAG_GONE;
2152 if (kprobe_aggrprobe(p)) {
2154 * If this is an aggr_kprobe, we have to list all the
2155 * chained probes and mark them GONE.
2157 list_for_each_entry(kp, &p->list, list)
2158 kp->flags |= KPROBE_FLAG_GONE;
2159 p->post_handler = NULL;
2160 kill_optimized_kprobe(p);
2163 * Here, we can remove insn_slot safely, because no thread calls
2164 * the original probed function (which will be freed soon) any more.
2166 arch_remove_kprobe(p);
2169 * The module is going away. We should disarm the kprobe which
2170 * is using ftrace, because ftrace framework is still available at
2171 * MODULE_STATE_GOING notification.
2173 if (kprobe_ftrace(p) && !kprobe_disabled(p) && !kprobes_all_disarmed)
2174 disarm_kprobe_ftrace(p);
2177 /* Disable one kprobe */
2178 int disable_kprobe(struct kprobe *kp)
2183 mutex_lock(&kprobe_mutex);
2185 /* Disable this kprobe */
2186 p = __disable_kprobe(kp);
2190 mutex_unlock(&kprobe_mutex);
2193 EXPORT_SYMBOL_GPL(disable_kprobe);
2195 /* Enable one kprobe */
2196 int enable_kprobe(struct kprobe *kp)
2201 mutex_lock(&kprobe_mutex);
2203 /* Check whether specified probe is valid. */
2204 p = __get_valid_kprobe(kp);
2205 if (unlikely(p == NULL)) {
2210 if (kprobe_gone(kp)) {
2211 /* This kprobe has gone, we couldn't enable it. */
2217 kp->flags &= ~KPROBE_FLAG_DISABLED;
2219 if (!kprobes_all_disarmed && kprobe_disabled(p)) {
2220 p->flags &= ~KPROBE_FLAG_DISABLED;
2221 ret = arm_kprobe(p);
2223 p->flags |= KPROBE_FLAG_DISABLED;
2226 mutex_unlock(&kprobe_mutex);
2229 EXPORT_SYMBOL_GPL(enable_kprobe);
2231 /* Caller must NOT call this in usual path. This is only for critical case */
2232 void dump_kprobe(struct kprobe *kp)
2234 pr_err("Dumping kprobe:\n");
2235 pr_err("Name: %s\nOffset: %x\nAddress: %pS\n",
2236 kp->symbol_name, kp->offset, kp->addr);
2238 NOKPROBE_SYMBOL(dump_kprobe);
2240 int kprobe_add_ksym_blacklist(unsigned long entry)
2242 struct kprobe_blacklist_entry *ent;
2243 unsigned long offset = 0, size = 0;
2245 if (!kernel_text_address(entry) ||
2246 !kallsyms_lookup_size_offset(entry, &size, &offset))
2249 ent = kmalloc(sizeof(*ent), GFP_KERNEL);
2252 ent->start_addr = entry;
2253 ent->end_addr = entry + size;
2254 INIT_LIST_HEAD(&ent->list);
2255 list_add_tail(&ent->list, &kprobe_blacklist);
2260 /* Add all symbols in given area into kprobe blacklist */
2261 int kprobe_add_area_blacklist(unsigned long start, unsigned long end)
2263 unsigned long entry;
2266 for (entry = start; entry < end; entry += ret) {
2267 ret = kprobe_add_ksym_blacklist(entry);
2270 if (ret == 0) /* In case of alias symbol */
2276 /* Remove all symbols in given area from kprobe blacklist */
2277 static void kprobe_remove_area_blacklist(unsigned long start, unsigned long end)
2279 struct kprobe_blacklist_entry *ent, *n;
2281 list_for_each_entry_safe(ent, n, &kprobe_blacklist, list) {
2282 if (ent->start_addr < start || ent->start_addr >= end)
2284 list_del(&ent->list);
2289 static void kprobe_remove_ksym_blacklist(unsigned long entry)
2291 kprobe_remove_area_blacklist(entry, entry + 1);
2294 int __weak arch_kprobe_get_kallsym(unsigned int *symnum, unsigned long *value,
2295 char *type, char *sym)
2300 int kprobe_get_kallsym(unsigned int symnum, unsigned long *value, char *type,
2303 #ifdef __ARCH_WANT_KPROBES_INSN_SLOT
2304 if (!kprobe_cache_get_kallsym(&kprobe_insn_slots, &symnum, value, type, sym))
2306 #ifdef CONFIG_OPTPROBES
2307 if (!kprobe_cache_get_kallsym(&kprobe_optinsn_slots, &symnum, value, type, sym))
2311 if (!arch_kprobe_get_kallsym(&symnum, value, type, sym))
2316 int __init __weak arch_populate_kprobe_blacklist(void)
2322 * Lookup and populate the kprobe_blacklist.
2324 * Unlike the kretprobe blacklist, we'll need to determine
2325 * the range of addresses that belong to the said functions,
2326 * since a kprobe need not necessarily be at the beginning
2329 static int __init populate_kprobe_blacklist(unsigned long *start,
2332 unsigned long entry;
2333 unsigned long *iter;
2336 for (iter = start; iter < end; iter++) {
2337 entry = arch_deref_entry_point((void *)*iter);
2338 ret = kprobe_add_ksym_blacklist(entry);
2345 /* Symbols in __kprobes_text are blacklisted */
2346 ret = kprobe_add_area_blacklist((unsigned long)__kprobes_text_start,
2347 (unsigned long)__kprobes_text_end);
2351 /* Symbols in noinstr section are blacklisted */
2352 ret = kprobe_add_area_blacklist((unsigned long)__noinstr_text_start,
2353 (unsigned long)__noinstr_text_end);
2355 return ret ? : arch_populate_kprobe_blacklist();
2358 static void add_module_kprobe_blacklist(struct module *mod)
2360 unsigned long start, end;
2363 if (mod->kprobe_blacklist) {
2364 for (i = 0; i < mod->num_kprobe_blacklist; i++)
2365 kprobe_add_ksym_blacklist(mod->kprobe_blacklist[i]);
2368 start = (unsigned long)mod->kprobes_text_start;
2370 end = start + mod->kprobes_text_size;
2371 kprobe_add_area_blacklist(start, end);
2374 start = (unsigned long)mod->noinstr_text_start;
2376 end = start + mod->noinstr_text_size;
2377 kprobe_add_area_blacklist(start, end);
2381 static void remove_module_kprobe_blacklist(struct module *mod)
2383 unsigned long start, end;
2386 if (mod->kprobe_blacklist) {
2387 for (i = 0; i < mod->num_kprobe_blacklist; i++)
2388 kprobe_remove_ksym_blacklist(mod->kprobe_blacklist[i]);
2391 start = (unsigned long)mod->kprobes_text_start;
2393 end = start + mod->kprobes_text_size;
2394 kprobe_remove_area_blacklist(start, end);
2397 start = (unsigned long)mod->noinstr_text_start;
2399 end = start + mod->noinstr_text_size;
2400 kprobe_remove_area_blacklist(start, end);
2404 /* Module notifier call back, checking kprobes on the module */
2405 static int kprobes_module_callback(struct notifier_block *nb,
2406 unsigned long val, void *data)
2408 struct module *mod = data;
2409 struct hlist_head *head;
2412 int checkcore = (val == MODULE_STATE_GOING);
2414 if (val == MODULE_STATE_COMING) {
2415 mutex_lock(&kprobe_mutex);
2416 add_module_kprobe_blacklist(mod);
2417 mutex_unlock(&kprobe_mutex);
2419 if (val != MODULE_STATE_GOING && val != MODULE_STATE_LIVE)
2423 * When MODULE_STATE_GOING was notified, both of module .text and
2424 * .init.text sections would be freed. When MODULE_STATE_LIVE was
2425 * notified, only .init.text section would be freed. We need to
2426 * disable kprobes which have been inserted in the sections.
2428 mutex_lock(&kprobe_mutex);
2429 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2430 head = &kprobe_table[i];
2431 hlist_for_each_entry(p, head, hlist)
2432 if (within_module_init((unsigned long)p->addr, mod) ||
2434 within_module_core((unsigned long)p->addr, mod))) {
2436 * The vaddr this probe is installed will soon
2437 * be vfreed buy not synced to disk. Hence,
2438 * disarming the breakpoint isn't needed.
2440 * Note, this will also move any optimized probes
2441 * that are pending to be removed from their
2442 * corresponding lists to the freeing_list and
2443 * will not be touched by the delayed
2444 * kprobe_optimizer work handler.
2449 if (val == MODULE_STATE_GOING)
2450 remove_module_kprobe_blacklist(mod);
2451 mutex_unlock(&kprobe_mutex);
2455 static struct notifier_block kprobe_module_nb = {
2456 .notifier_call = kprobes_module_callback,
2460 /* Markers of _kprobe_blacklist section */
2461 extern unsigned long __start_kprobe_blacklist[];
2462 extern unsigned long __stop_kprobe_blacklist[];
2464 void kprobe_free_init_mem(void)
2466 void *start = (void *)(&__init_begin);
2467 void *end = (void *)(&__init_end);
2468 struct hlist_head *head;
2472 mutex_lock(&kprobe_mutex);
2474 /* Kill all kprobes on initmem */
2475 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2476 head = &kprobe_table[i];
2477 hlist_for_each_entry(p, head, hlist) {
2478 if (start <= (void *)p->addr && (void *)p->addr < end)
2483 mutex_unlock(&kprobe_mutex);
2486 static int __init init_kprobes(void)
2490 /* FIXME allocate the probe table, currently defined statically */
2491 /* initialize all list heads */
2492 for (i = 0; i < KPROBE_TABLE_SIZE; i++)
2493 INIT_HLIST_HEAD(&kprobe_table[i]);
2495 err = populate_kprobe_blacklist(__start_kprobe_blacklist,
2496 __stop_kprobe_blacklist);
2498 pr_err("kprobes: failed to populate blacklist: %d\n", err);
2499 pr_err("Please take care of using kprobes.\n");
2502 if (kretprobe_blacklist_size) {
2503 /* lookup the function address from its name */
2504 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
2505 kretprobe_blacklist[i].addr =
2506 kprobe_lookup_name(kretprobe_blacklist[i].name, 0);
2507 if (!kretprobe_blacklist[i].addr)
2508 printk("kretprobe: lookup failed: %s\n",
2509 kretprobe_blacklist[i].name);
2513 /* By default, kprobes are armed */
2514 kprobes_all_disarmed = false;
2516 #if defined(CONFIG_OPTPROBES) && defined(__ARCH_WANT_KPROBES_INSN_SLOT)
2517 /* Init kprobe_optinsn_slots for allocation */
2518 kprobe_optinsn_slots.insn_size = MAX_OPTINSN_SIZE;
2521 err = arch_init_kprobes();
2523 err = register_die_notifier(&kprobe_exceptions_nb);
2525 err = register_module_notifier(&kprobe_module_nb);
2527 kprobes_initialized = (err == 0);
2533 early_initcall(init_kprobes);
2535 #if defined(CONFIG_OPTPROBES)
2536 static int __init init_optprobes(void)
2539 * Enable kprobe optimization - this kicks the optimizer which
2540 * depends on synchronize_rcu_tasks() and ksoftirqd, that is
2541 * not spawned in early initcall. So delay the optimization.
2543 optimize_all_kprobes();
2547 subsys_initcall(init_optprobes);
2550 #ifdef CONFIG_DEBUG_FS
2551 static void report_probe(struct seq_file *pi, struct kprobe *p,
2552 const char *sym, int offset, char *modname, struct kprobe *pp)
2555 void *addr = p->addr;
2557 if (p->pre_handler == pre_handler_kretprobe)
2562 if (!kallsyms_show_value(pi->file->f_cred))
2566 seq_printf(pi, "%px %s %s+0x%x %s ",
2567 addr, kprobe_type, sym, offset,
2568 (modname ? modname : " "));
2569 else /* try to use %pS */
2570 seq_printf(pi, "%px %s %pS ",
2571 addr, kprobe_type, p->addr);
2575 seq_printf(pi, "%s%s%s%s\n",
2576 (kprobe_gone(p) ? "[GONE]" : ""),
2577 ((kprobe_disabled(p) && !kprobe_gone(p)) ? "[DISABLED]" : ""),
2578 (kprobe_optimized(pp) ? "[OPTIMIZED]" : ""),
2579 (kprobe_ftrace(pp) ? "[FTRACE]" : ""));
2582 static void *kprobe_seq_start(struct seq_file *f, loff_t *pos)
2584 return (*pos < KPROBE_TABLE_SIZE) ? pos : NULL;
2587 static void *kprobe_seq_next(struct seq_file *f, void *v, loff_t *pos)
2590 if (*pos >= KPROBE_TABLE_SIZE)
2595 static void kprobe_seq_stop(struct seq_file *f, void *v)
2600 static int show_kprobe_addr(struct seq_file *pi, void *v)
2602 struct hlist_head *head;
2603 struct kprobe *p, *kp;
2604 const char *sym = NULL;
2605 unsigned int i = *(loff_t *) v;
2606 unsigned long offset = 0;
2607 char *modname, namebuf[KSYM_NAME_LEN];
2609 head = &kprobe_table[i];
2611 hlist_for_each_entry_rcu(p, head, hlist) {
2612 sym = kallsyms_lookup((unsigned long)p->addr, NULL,
2613 &offset, &modname, namebuf);
2614 if (kprobe_aggrprobe(p)) {
2615 list_for_each_entry_rcu(kp, &p->list, list)
2616 report_probe(pi, kp, sym, offset, modname, p);
2618 report_probe(pi, p, sym, offset, modname, NULL);
2624 static const struct seq_operations kprobes_sops = {
2625 .start = kprobe_seq_start,
2626 .next = kprobe_seq_next,
2627 .stop = kprobe_seq_stop,
2628 .show = show_kprobe_addr
2631 DEFINE_SEQ_ATTRIBUTE(kprobes);
2633 /* kprobes/blacklist -- shows which functions can not be probed */
2634 static void *kprobe_blacklist_seq_start(struct seq_file *m, loff_t *pos)
2636 mutex_lock(&kprobe_mutex);
2637 return seq_list_start(&kprobe_blacklist, *pos);
2640 static void *kprobe_blacklist_seq_next(struct seq_file *m, void *v, loff_t *pos)
2642 return seq_list_next(v, &kprobe_blacklist, pos);
2645 static int kprobe_blacklist_seq_show(struct seq_file *m, void *v)
2647 struct kprobe_blacklist_entry *ent =
2648 list_entry(v, struct kprobe_blacklist_entry, list);
2651 * If /proc/kallsyms is not showing kernel address, we won't
2652 * show them here either.
2654 if (!kallsyms_show_value(m->file->f_cred))
2655 seq_printf(m, "0x%px-0x%px\t%ps\n", NULL, NULL,
2656 (void *)ent->start_addr);
2658 seq_printf(m, "0x%px-0x%px\t%ps\n", (void *)ent->start_addr,
2659 (void *)ent->end_addr, (void *)ent->start_addr);
2663 static void kprobe_blacklist_seq_stop(struct seq_file *f, void *v)
2665 mutex_unlock(&kprobe_mutex);
2668 static const struct seq_operations kprobe_blacklist_sops = {
2669 .start = kprobe_blacklist_seq_start,
2670 .next = kprobe_blacklist_seq_next,
2671 .stop = kprobe_blacklist_seq_stop,
2672 .show = kprobe_blacklist_seq_show,
2674 DEFINE_SEQ_ATTRIBUTE(kprobe_blacklist);
2676 static int arm_all_kprobes(void)
2678 struct hlist_head *head;
2680 unsigned int i, total = 0, errors = 0;
2683 mutex_lock(&kprobe_mutex);
2685 /* If kprobes are armed, just return */
2686 if (!kprobes_all_disarmed)
2687 goto already_enabled;
2690 * optimize_kprobe() called by arm_kprobe() checks
2691 * kprobes_all_disarmed, so set kprobes_all_disarmed before
2694 kprobes_all_disarmed = false;
2695 /* Arming kprobes doesn't optimize kprobe itself */
2696 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2697 head = &kprobe_table[i];
2698 /* Arm all kprobes on a best-effort basis */
2699 hlist_for_each_entry(p, head, hlist) {
2700 if (!kprobe_disabled(p)) {
2701 err = arm_kprobe(p);
2712 pr_warn("Kprobes globally enabled, but failed to arm %d out of %d probes\n",
2715 pr_info("Kprobes globally enabled\n");
2718 mutex_unlock(&kprobe_mutex);
2722 static int disarm_all_kprobes(void)
2724 struct hlist_head *head;
2726 unsigned int i, total = 0, errors = 0;
2729 mutex_lock(&kprobe_mutex);
2731 /* If kprobes are already disarmed, just return */
2732 if (kprobes_all_disarmed) {
2733 mutex_unlock(&kprobe_mutex);
2737 kprobes_all_disarmed = true;
2739 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2740 head = &kprobe_table[i];
2741 /* Disarm all kprobes on a best-effort basis */
2742 hlist_for_each_entry(p, head, hlist) {
2743 if (!arch_trampoline_kprobe(p) && !kprobe_disabled(p)) {
2744 err = disarm_kprobe(p, false);
2755 pr_warn("Kprobes globally disabled, but failed to disarm %d out of %d probes\n",
2758 pr_info("Kprobes globally disabled\n");
2760 mutex_unlock(&kprobe_mutex);
2762 /* Wait for disarming all kprobes by optimizer */
2763 wait_for_kprobe_optimizer();
2769 * XXX: The debugfs bool file interface doesn't allow for callbacks
2770 * when the bool state is switched. We can reuse that facility when
2773 static ssize_t read_enabled_file_bool(struct file *file,
2774 char __user *user_buf, size_t count, loff_t *ppos)
2778 if (!kprobes_all_disarmed)
2784 return simple_read_from_buffer(user_buf, count, ppos, buf, 2);
2787 static ssize_t write_enabled_file_bool(struct file *file,
2788 const char __user *user_buf, size_t count, loff_t *ppos)
2794 buf_size = min(count, (sizeof(buf)-1));
2795 if (copy_from_user(buf, user_buf, buf_size))
2798 buf[buf_size] = '\0';
2803 ret = arm_all_kprobes();
2808 ret = disarm_all_kprobes();
2820 static const struct file_operations fops_kp = {
2821 .read = read_enabled_file_bool,
2822 .write = write_enabled_file_bool,
2823 .llseek = default_llseek,
2826 static int __init debugfs_kprobe_init(void)
2829 unsigned int value = 1;
2831 dir = debugfs_create_dir("kprobes", NULL);
2833 debugfs_create_file("list", 0400, dir, NULL, &kprobes_fops);
2835 debugfs_create_file("enabled", 0600, dir, &value, &fops_kp);
2837 debugfs_create_file("blacklist", 0400, dir, NULL,
2838 &kprobe_blacklist_fops);
2843 late_initcall(debugfs_kprobe_init);
2844 #endif /* CONFIG_DEBUG_FS */