1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * Kernel Probes (KProbes)
6 * Copyright (C) IBM Corporation, 2002, 2004
8 * 2002-Oct Created by Vamsi Krishna S <vamsi_krishna@in.ibm.com> Kernel
9 * Probes initial implementation (includes suggestions from
11 * 2004-Aug Updated by Prasanna S Panchamukhi <prasanna@in.ibm.com> with
12 * hlists and exceptions notifier as suggested by Andi Kleen.
13 * 2004-July Suparna Bhattacharya <suparna@in.ibm.com> added jumper probes
14 * interface to access function arguments.
15 * 2004-Sep Prasanna S Panchamukhi <prasanna@in.ibm.com> Changed Kprobes
16 * exceptions notifier to be first on the priority list.
17 * 2005-May Hien Nguyen <hien@us.ibm.com>, Jim Keniston
18 * <jkenisto@us.ibm.com> and Prasanna S Panchamukhi
19 * <prasanna@in.ibm.com> added function-return probes.
21 #include <linux/kprobes.h>
22 #include <linux/hash.h>
23 #include <linux/init.h>
24 #include <linux/slab.h>
25 #include <linux/stddef.h>
26 #include <linux/export.h>
27 #include <linux/moduleloader.h>
28 #include <linux/kallsyms.h>
29 #include <linux/freezer.h>
30 #include <linux/seq_file.h>
31 #include <linux/debugfs.h>
32 #include <linux/sysctl.h>
33 #include <linux/kdebug.h>
34 #include <linux/memory.h>
35 #include <linux/ftrace.h>
36 #include <linux/cpu.h>
37 #include <linux/jump_label.h>
39 #include <asm/sections.h>
40 #include <asm/cacheflush.h>
41 #include <asm/errno.h>
42 #include <linux/uaccess.h>
44 #define KPROBE_HASH_BITS 6
45 #define KPROBE_TABLE_SIZE (1 << KPROBE_HASH_BITS)
48 static int kprobes_initialized;
49 static struct hlist_head kprobe_table[KPROBE_TABLE_SIZE];
50 static struct hlist_head kretprobe_inst_table[KPROBE_TABLE_SIZE];
52 /* NOTE: change this value only with kprobe_mutex held */
53 static bool kprobes_all_disarmed;
55 /* This protects kprobe_table and optimizing_list */
56 static DEFINE_MUTEX(kprobe_mutex);
57 static DEFINE_PER_CPU(struct kprobe *, kprobe_instance) = NULL;
59 raw_spinlock_t lock ____cacheline_aligned_in_smp;
60 } kretprobe_table_locks[KPROBE_TABLE_SIZE];
62 kprobe_opcode_t * __weak kprobe_lookup_name(const char *name,
63 unsigned int __unused)
65 return ((kprobe_opcode_t *)(kallsyms_lookup_name(name)));
68 static raw_spinlock_t *kretprobe_table_lock_ptr(unsigned long hash)
70 return &(kretprobe_table_locks[hash].lock);
73 /* Blacklist -- list of struct kprobe_blacklist_entry */
74 static LIST_HEAD(kprobe_blacklist);
76 #ifdef __ARCH_WANT_KPROBES_INSN_SLOT
78 * kprobe->ainsn.insn points to the copy of the instruction to be
79 * single-stepped. x86_64, POWER4 and above have no-exec support and
80 * stepping on the instruction on a vmalloced/kmalloced/data page
81 * is a recipe for disaster
83 struct kprobe_insn_page {
84 struct list_head list;
85 kprobe_opcode_t *insns; /* Page of instruction slots */
86 struct kprobe_insn_cache *cache;
92 #define KPROBE_INSN_PAGE_SIZE(slots) \
93 (offsetof(struct kprobe_insn_page, slot_used) + \
94 (sizeof(char) * (slots)))
96 static int slots_per_page(struct kprobe_insn_cache *c)
98 return PAGE_SIZE/(c->insn_size * sizeof(kprobe_opcode_t));
101 enum kprobe_slot_state {
107 void __weak *alloc_insn_page(void)
109 return module_alloc(PAGE_SIZE);
112 void __weak free_insn_page(void *page)
114 module_memfree(page);
117 struct kprobe_insn_cache kprobe_insn_slots = {
118 .mutex = __MUTEX_INITIALIZER(kprobe_insn_slots.mutex),
119 .alloc = alloc_insn_page,
120 .free = free_insn_page,
121 .pages = LIST_HEAD_INIT(kprobe_insn_slots.pages),
122 .insn_size = MAX_INSN_SIZE,
125 static int collect_garbage_slots(struct kprobe_insn_cache *c);
128 * __get_insn_slot() - Find a slot on an executable page for an instruction.
129 * We allocate an executable page if there's no room on existing ones.
131 kprobe_opcode_t *__get_insn_slot(struct kprobe_insn_cache *c)
133 struct kprobe_insn_page *kip;
134 kprobe_opcode_t *slot = NULL;
136 /* Since the slot array is not protected by rcu, we need a mutex */
137 mutex_lock(&c->mutex);
140 list_for_each_entry_rcu(kip, &c->pages, list) {
141 if (kip->nused < slots_per_page(c)) {
143 for (i = 0; i < slots_per_page(c); i++) {
144 if (kip->slot_used[i] == SLOT_CLEAN) {
145 kip->slot_used[i] = SLOT_USED;
147 slot = kip->insns + (i * c->insn_size);
152 /* kip->nused is broken. Fix it. */
153 kip->nused = slots_per_page(c);
159 /* If there are any garbage slots, collect it and try again. */
160 if (c->nr_garbage && collect_garbage_slots(c) == 0)
163 /* All out of space. Need to allocate a new page. */
164 kip = kmalloc(KPROBE_INSN_PAGE_SIZE(slots_per_page(c)), GFP_KERNEL);
169 * Use module_alloc so this page is within +/- 2GB of where the
170 * kernel image and loaded module images reside. This is required
171 * so x86_64 can correctly handle the %rip-relative fixups.
173 kip->insns = c->alloc();
178 INIT_LIST_HEAD(&kip->list);
179 memset(kip->slot_used, SLOT_CLEAN, slots_per_page(c));
180 kip->slot_used[0] = SLOT_USED;
184 list_add_rcu(&kip->list, &c->pages);
187 mutex_unlock(&c->mutex);
191 /* Return 1 if all garbages are collected, otherwise 0. */
192 static int collect_one_slot(struct kprobe_insn_page *kip, int idx)
194 kip->slot_used[idx] = SLOT_CLEAN;
196 if (kip->nused == 0) {
198 * Page is no longer in use. Free it unless
199 * it's the last one. We keep the last one
200 * so as not to have to set it up again the
201 * next time somebody inserts a probe.
203 if (!list_is_singular(&kip->list)) {
204 list_del_rcu(&kip->list);
206 kip->cache->free(kip->insns);
214 static int collect_garbage_slots(struct kprobe_insn_cache *c)
216 struct kprobe_insn_page *kip, *next;
218 /* Ensure no-one is interrupted on the garbages */
221 list_for_each_entry_safe(kip, next, &c->pages, list) {
223 if (kip->ngarbage == 0)
225 kip->ngarbage = 0; /* we will collect all garbages */
226 for (i = 0; i < slots_per_page(c); i++) {
227 if (kip->slot_used[i] == SLOT_DIRTY && collect_one_slot(kip, i))
235 void __free_insn_slot(struct kprobe_insn_cache *c,
236 kprobe_opcode_t *slot, int dirty)
238 struct kprobe_insn_page *kip;
241 mutex_lock(&c->mutex);
243 list_for_each_entry_rcu(kip, &c->pages, list) {
244 idx = ((long)slot - (long)kip->insns) /
245 (c->insn_size * sizeof(kprobe_opcode_t));
246 if (idx >= 0 && idx < slots_per_page(c))
249 /* Could not find this slot. */
254 /* Mark and sweep: this may sleep */
256 /* Check double free */
257 WARN_ON(kip->slot_used[idx] != SLOT_USED);
259 kip->slot_used[idx] = SLOT_DIRTY;
261 if (++c->nr_garbage > slots_per_page(c))
262 collect_garbage_slots(c);
264 collect_one_slot(kip, idx);
267 mutex_unlock(&c->mutex);
271 * Check given address is on the page of kprobe instruction slots.
272 * This will be used for checking whether the address on a stack
273 * is on a text area or not.
275 bool __is_insn_slot_addr(struct kprobe_insn_cache *c, unsigned long addr)
277 struct kprobe_insn_page *kip;
281 list_for_each_entry_rcu(kip, &c->pages, list) {
282 if (addr >= (unsigned long)kip->insns &&
283 addr < (unsigned long)kip->insns + PAGE_SIZE) {
293 #ifdef CONFIG_OPTPROBES
294 /* For optimized_kprobe buffer */
295 struct kprobe_insn_cache kprobe_optinsn_slots = {
296 .mutex = __MUTEX_INITIALIZER(kprobe_optinsn_slots.mutex),
297 .alloc = alloc_insn_page,
298 .free = free_insn_page,
299 .pages = LIST_HEAD_INIT(kprobe_optinsn_slots.pages),
300 /* .insn_size is initialized later */
306 /* We have preemption disabled.. so it is safe to use __ versions */
307 static inline void set_kprobe_instance(struct kprobe *kp)
309 __this_cpu_write(kprobe_instance, kp);
312 static inline void reset_kprobe_instance(void)
314 __this_cpu_write(kprobe_instance, NULL);
318 * This routine is called either:
319 * - under the kprobe_mutex - during kprobe_[un]register()
321 * - with preemption disabled - from arch/xxx/kernel/kprobes.c
323 struct kprobe *get_kprobe(void *addr)
325 struct hlist_head *head;
328 head = &kprobe_table[hash_ptr(addr, KPROBE_HASH_BITS)];
329 hlist_for_each_entry_rcu(p, head, hlist,
330 lockdep_is_held(&kprobe_mutex)) {
337 NOKPROBE_SYMBOL(get_kprobe);
339 static int aggr_pre_handler(struct kprobe *p, struct pt_regs *regs);
341 /* Return true if the kprobe is an aggregator */
342 static inline int kprobe_aggrprobe(struct kprobe *p)
344 return p->pre_handler == aggr_pre_handler;
347 /* Return true(!0) if the kprobe is unused */
348 static inline int kprobe_unused(struct kprobe *p)
350 return kprobe_aggrprobe(p) && kprobe_disabled(p) &&
351 list_empty(&p->list);
355 * Keep all fields in the kprobe consistent
357 static inline void copy_kprobe(struct kprobe *ap, struct kprobe *p)
359 memcpy(&p->opcode, &ap->opcode, sizeof(kprobe_opcode_t));
360 memcpy(&p->ainsn, &ap->ainsn, sizeof(struct arch_specific_insn));
363 #ifdef CONFIG_OPTPROBES
364 /* NOTE: change this value only with kprobe_mutex held */
365 static bool kprobes_allow_optimization;
368 * Call all pre_handler on the list, but ignores its return value.
369 * This must be called from arch-dep optimized caller.
371 void opt_pre_handler(struct kprobe *p, struct pt_regs *regs)
375 list_for_each_entry_rcu(kp, &p->list, list) {
376 if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
377 set_kprobe_instance(kp);
378 kp->pre_handler(kp, regs);
380 reset_kprobe_instance();
383 NOKPROBE_SYMBOL(opt_pre_handler);
385 /* Free optimized instructions and optimized_kprobe */
386 static void free_aggr_kprobe(struct kprobe *p)
388 struct optimized_kprobe *op;
390 op = container_of(p, struct optimized_kprobe, kp);
391 arch_remove_optimized_kprobe(op);
392 arch_remove_kprobe(p);
396 /* Return true(!0) if the kprobe is ready for optimization. */
397 static inline int kprobe_optready(struct kprobe *p)
399 struct optimized_kprobe *op;
401 if (kprobe_aggrprobe(p)) {
402 op = container_of(p, struct optimized_kprobe, kp);
403 return arch_prepared_optinsn(&op->optinsn);
409 /* Return true if the kprobe is disarmed. Note: p must be on hash list */
410 bool kprobe_disarmed(struct kprobe *p)
412 struct optimized_kprobe *op;
414 /* If kprobe is not aggr/opt probe, just return kprobe is disabled */
415 if (!kprobe_aggrprobe(p))
416 return kprobe_disabled(p);
418 op = container_of(p, struct optimized_kprobe, kp);
420 return kprobe_disabled(p) && list_empty(&op->list);
423 /* Return true(!0) if the probe is queued on (un)optimizing lists */
424 static int kprobe_queued(struct kprobe *p)
426 struct optimized_kprobe *op;
428 if (kprobe_aggrprobe(p)) {
429 op = container_of(p, struct optimized_kprobe, kp);
430 if (!list_empty(&op->list))
437 * Return an optimized kprobe whose optimizing code replaces
438 * instructions including addr (exclude breakpoint).
440 static struct kprobe *get_optimized_kprobe(unsigned long addr)
443 struct kprobe *p = NULL;
444 struct optimized_kprobe *op;
446 /* Don't check i == 0, since that is a breakpoint case. */
447 for (i = 1; !p && i < MAX_OPTIMIZED_LENGTH; i++)
448 p = get_kprobe((void *)(addr - i));
450 if (p && kprobe_optready(p)) {
451 op = container_of(p, struct optimized_kprobe, kp);
452 if (arch_within_optimized_kprobe(op, addr))
459 /* Optimization staging list, protected by kprobe_mutex */
460 static LIST_HEAD(optimizing_list);
461 static LIST_HEAD(unoptimizing_list);
462 static LIST_HEAD(freeing_list);
464 static void kprobe_optimizer(struct work_struct *work);
465 static DECLARE_DELAYED_WORK(optimizing_work, kprobe_optimizer);
466 #define OPTIMIZE_DELAY 5
469 * Optimize (replace a breakpoint with a jump) kprobes listed on
472 static void do_optimize_kprobes(void)
474 lockdep_assert_held(&text_mutex);
476 * The optimization/unoptimization refers online_cpus via
477 * stop_machine() and cpu-hotplug modifies online_cpus.
478 * And same time, text_mutex will be held in cpu-hotplug and here.
479 * This combination can cause a deadlock (cpu-hotplug try to lock
480 * text_mutex but stop_machine can not be done because online_cpus
482 * To avoid this deadlock, caller must have locked cpu hotplug
483 * for preventing cpu-hotplug outside of text_mutex locking.
485 lockdep_assert_cpus_held();
487 /* Optimization never be done when disarmed */
488 if (kprobes_all_disarmed || !kprobes_allow_optimization ||
489 list_empty(&optimizing_list))
492 arch_optimize_kprobes(&optimizing_list);
496 * Unoptimize (replace a jump with a breakpoint and remove the breakpoint
497 * if need) kprobes listed on unoptimizing_list.
499 static void do_unoptimize_kprobes(void)
501 struct optimized_kprobe *op, *tmp;
503 lockdep_assert_held(&text_mutex);
504 /* See comment in do_optimize_kprobes() */
505 lockdep_assert_cpus_held();
507 /* Unoptimization must be done anytime */
508 if (list_empty(&unoptimizing_list))
511 arch_unoptimize_kprobes(&unoptimizing_list, &freeing_list);
512 /* Loop free_list for disarming */
513 list_for_each_entry_safe(op, tmp, &freeing_list, list) {
514 /* Switching from detour code to origin */
515 op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
516 /* Disarm probes if marked disabled */
517 if (kprobe_disabled(&op->kp))
518 arch_disarm_kprobe(&op->kp);
519 if (kprobe_unused(&op->kp)) {
521 * Remove unused probes from hash list. After waiting
522 * for synchronization, these probes are reclaimed.
523 * (reclaiming is done by do_free_cleaned_kprobes.)
525 hlist_del_rcu(&op->kp.hlist);
527 list_del_init(&op->list);
531 /* Reclaim all kprobes on the free_list */
532 static void do_free_cleaned_kprobes(void)
534 struct optimized_kprobe *op, *tmp;
536 list_for_each_entry_safe(op, tmp, &freeing_list, list) {
537 list_del_init(&op->list);
538 if (WARN_ON_ONCE(!kprobe_unused(&op->kp))) {
540 * This must not happen, but if there is a kprobe
541 * still in use, keep it on kprobes hash list.
545 free_aggr_kprobe(&op->kp);
549 /* Start optimizer after OPTIMIZE_DELAY passed */
550 static void kick_kprobe_optimizer(void)
552 schedule_delayed_work(&optimizing_work, OPTIMIZE_DELAY);
555 /* Kprobe jump optimizer */
556 static void kprobe_optimizer(struct work_struct *work)
558 mutex_lock(&kprobe_mutex);
560 mutex_lock(&text_mutex);
561 /* Lock modules while optimizing kprobes */
562 mutex_lock(&module_mutex);
565 * Step 1: Unoptimize kprobes and collect cleaned (unused and disarmed)
566 * kprobes before waiting for quiesence period.
568 do_unoptimize_kprobes();
571 * Step 2: Wait for quiesence period to ensure all potentially
572 * preempted tasks to have normally scheduled. Because optprobe
573 * may modify multiple instructions, there is a chance that Nth
574 * instruction is preempted. In that case, such tasks can return
575 * to 2nd-Nth byte of jump instruction. This wait is for avoiding it.
576 * Note that on non-preemptive kernel, this is transparently converted
577 * to synchronoze_sched() to wait for all interrupts to have completed.
579 synchronize_rcu_tasks();
581 /* Step 3: Optimize kprobes after quiesence period */
582 do_optimize_kprobes();
584 /* Step 4: Free cleaned kprobes after quiesence period */
585 do_free_cleaned_kprobes();
587 mutex_unlock(&module_mutex);
588 mutex_unlock(&text_mutex);
591 /* Step 5: Kick optimizer again if needed */
592 if (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list))
593 kick_kprobe_optimizer();
595 mutex_unlock(&kprobe_mutex);
598 /* Wait for completing optimization and unoptimization */
599 void wait_for_kprobe_optimizer(void)
601 mutex_lock(&kprobe_mutex);
603 while (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list)) {
604 mutex_unlock(&kprobe_mutex);
606 /* this will also make optimizing_work execute immmediately */
607 flush_delayed_work(&optimizing_work);
608 /* @optimizing_work might not have been queued yet, relax */
611 mutex_lock(&kprobe_mutex);
614 mutex_unlock(&kprobe_mutex);
617 bool optprobe_queued_unopt(struct optimized_kprobe *op)
619 struct optimized_kprobe *_op;
621 list_for_each_entry(_op, &unoptimizing_list, list) {
629 /* Optimize kprobe if p is ready to be optimized */
630 static void optimize_kprobe(struct kprobe *p)
632 struct optimized_kprobe *op;
634 /* Check if the kprobe is disabled or not ready for optimization. */
635 if (!kprobe_optready(p) || !kprobes_allow_optimization ||
636 (kprobe_disabled(p) || kprobes_all_disarmed))
639 /* kprobes with post_handler can not be optimized */
643 op = container_of(p, struct optimized_kprobe, kp);
645 /* Check there is no other kprobes at the optimized instructions */
646 if (arch_check_optimized_kprobe(op) < 0)
649 /* Check if it is already optimized. */
650 if (op->kp.flags & KPROBE_FLAG_OPTIMIZED) {
651 if (optprobe_queued_unopt(op)) {
652 /* This is under unoptimizing. Just dequeue the probe */
653 list_del_init(&op->list);
657 op->kp.flags |= KPROBE_FLAG_OPTIMIZED;
659 /* On unoptimizing/optimizing_list, op must have OPTIMIZED flag */
660 if (WARN_ON_ONCE(!list_empty(&op->list)))
663 list_add(&op->list, &optimizing_list);
664 kick_kprobe_optimizer();
667 /* Short cut to direct unoptimizing */
668 static void force_unoptimize_kprobe(struct optimized_kprobe *op)
670 lockdep_assert_cpus_held();
671 arch_unoptimize_kprobe(op);
672 op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
673 if (kprobe_disabled(&op->kp))
674 arch_disarm_kprobe(&op->kp);
677 /* Unoptimize a kprobe if p is optimized */
678 static void unoptimize_kprobe(struct kprobe *p, bool force)
680 struct optimized_kprobe *op;
682 if (!kprobe_aggrprobe(p) || kprobe_disarmed(p))
683 return; /* This is not an optprobe nor optimized */
685 op = container_of(p, struct optimized_kprobe, kp);
686 if (!kprobe_optimized(p))
689 if (!list_empty(&op->list)) {
690 if (optprobe_queued_unopt(op)) {
691 /* Queued in unoptimizing queue */
694 * Forcibly unoptimize the kprobe here, and queue it
695 * in the freeing list for release afterwards.
697 force_unoptimize_kprobe(op);
698 list_move(&op->list, &freeing_list);
701 /* Dequeue from the optimizing queue */
702 list_del_init(&op->list);
703 op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
708 /* Optimized kprobe case */
710 /* Forcibly update the code: this is a special case */
711 force_unoptimize_kprobe(op);
713 list_add(&op->list, &unoptimizing_list);
714 kick_kprobe_optimizer();
718 /* Cancel unoptimizing for reusing */
719 static int reuse_unused_kprobe(struct kprobe *ap)
721 struct optimized_kprobe *op;
724 * Unused kprobe MUST be on the way of delayed unoptimizing (means
725 * there is still a relative jump) and disabled.
727 op = container_of(ap, struct optimized_kprobe, kp);
728 WARN_ON_ONCE(list_empty(&op->list));
729 /* Enable the probe again */
730 ap->flags &= ~KPROBE_FLAG_DISABLED;
731 /* Optimize it again (remove from op->list) */
732 if (!kprobe_optready(ap))
739 /* Remove optimized instructions */
740 static void kill_optimized_kprobe(struct kprobe *p)
742 struct optimized_kprobe *op;
744 op = container_of(p, struct optimized_kprobe, kp);
745 if (!list_empty(&op->list))
746 /* Dequeue from the (un)optimization queue */
747 list_del_init(&op->list);
748 op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
750 if (kprobe_unused(p)) {
751 /* Enqueue if it is unused */
752 list_add(&op->list, &freeing_list);
754 * Remove unused probes from the hash list. After waiting
755 * for synchronization, this probe is reclaimed.
756 * (reclaiming is done by do_free_cleaned_kprobes().)
758 hlist_del_rcu(&op->kp.hlist);
761 /* Don't touch the code, because it is already freed. */
762 arch_remove_optimized_kprobe(op);
766 void __prepare_optimized_kprobe(struct optimized_kprobe *op, struct kprobe *p)
768 if (!kprobe_ftrace(p))
769 arch_prepare_optimized_kprobe(op, p);
772 /* Try to prepare optimized instructions */
773 static void prepare_optimized_kprobe(struct kprobe *p)
775 struct optimized_kprobe *op;
777 op = container_of(p, struct optimized_kprobe, kp);
778 __prepare_optimized_kprobe(op, p);
781 /* Allocate new optimized_kprobe and try to prepare optimized instructions */
782 static struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
784 struct optimized_kprobe *op;
786 op = kzalloc(sizeof(struct optimized_kprobe), GFP_KERNEL);
790 INIT_LIST_HEAD(&op->list);
791 op->kp.addr = p->addr;
792 __prepare_optimized_kprobe(op, p);
797 static void init_aggr_kprobe(struct kprobe *ap, struct kprobe *p);
800 * Prepare an optimized_kprobe and optimize it
801 * NOTE: p must be a normal registered kprobe
803 static void try_to_optimize_kprobe(struct kprobe *p)
806 struct optimized_kprobe *op;
808 /* Impossible to optimize ftrace-based kprobe */
809 if (kprobe_ftrace(p))
812 /* For preparing optimization, jump_label_text_reserved() is called */
815 mutex_lock(&text_mutex);
817 ap = alloc_aggr_kprobe(p);
821 op = container_of(ap, struct optimized_kprobe, kp);
822 if (!arch_prepared_optinsn(&op->optinsn)) {
823 /* If failed to setup optimizing, fallback to kprobe */
824 arch_remove_optimized_kprobe(op);
829 init_aggr_kprobe(ap, p);
830 optimize_kprobe(ap); /* This just kicks optimizer thread */
833 mutex_unlock(&text_mutex);
839 static void optimize_all_kprobes(void)
841 struct hlist_head *head;
845 mutex_lock(&kprobe_mutex);
846 /* If optimization is already allowed, just return */
847 if (kprobes_allow_optimization)
851 kprobes_allow_optimization = true;
852 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
853 head = &kprobe_table[i];
854 hlist_for_each_entry_rcu(p, head, hlist)
855 if (!kprobe_disabled(p))
859 printk(KERN_INFO "Kprobes globally optimized\n");
861 mutex_unlock(&kprobe_mutex);
864 static void unoptimize_all_kprobes(void)
866 struct hlist_head *head;
870 mutex_lock(&kprobe_mutex);
871 /* If optimization is already prohibited, just return */
872 if (!kprobes_allow_optimization) {
873 mutex_unlock(&kprobe_mutex);
878 kprobes_allow_optimization = false;
879 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
880 head = &kprobe_table[i];
881 hlist_for_each_entry_rcu(p, head, hlist) {
882 if (!kprobe_disabled(p))
883 unoptimize_kprobe(p, false);
887 mutex_unlock(&kprobe_mutex);
889 /* Wait for unoptimizing completion */
890 wait_for_kprobe_optimizer();
891 printk(KERN_INFO "Kprobes globally unoptimized\n");
894 static DEFINE_MUTEX(kprobe_sysctl_mutex);
895 int sysctl_kprobes_optimization;
896 int proc_kprobes_optimization_handler(struct ctl_table *table, int write,
897 void __user *buffer, size_t *length,
902 mutex_lock(&kprobe_sysctl_mutex);
903 sysctl_kprobes_optimization = kprobes_allow_optimization ? 1 : 0;
904 ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
906 if (sysctl_kprobes_optimization)
907 optimize_all_kprobes();
909 unoptimize_all_kprobes();
910 mutex_unlock(&kprobe_sysctl_mutex);
914 #endif /* CONFIG_SYSCTL */
916 /* Put a breakpoint for a probe. Must be called with text_mutex locked */
917 static void __arm_kprobe(struct kprobe *p)
921 /* Check collision with other optimized kprobes */
922 _p = get_optimized_kprobe((unsigned long)p->addr);
924 /* Fallback to unoptimized kprobe */
925 unoptimize_kprobe(_p, true);
928 optimize_kprobe(p); /* Try to optimize (add kprobe to a list) */
931 /* Remove the breakpoint of a probe. Must be called with text_mutex locked */
932 static void __disarm_kprobe(struct kprobe *p, bool reopt)
936 /* Try to unoptimize */
937 unoptimize_kprobe(p, kprobes_all_disarmed);
939 if (!kprobe_queued(p)) {
940 arch_disarm_kprobe(p);
941 /* If another kprobe was blocked, optimize it. */
942 _p = get_optimized_kprobe((unsigned long)p->addr);
943 if (unlikely(_p) && reopt)
946 /* TODO: reoptimize others after unoptimized this probe */
949 #else /* !CONFIG_OPTPROBES */
951 #define optimize_kprobe(p) do {} while (0)
952 #define unoptimize_kprobe(p, f) do {} while (0)
953 #define kill_optimized_kprobe(p) do {} while (0)
954 #define prepare_optimized_kprobe(p) do {} while (0)
955 #define try_to_optimize_kprobe(p) do {} while (0)
956 #define __arm_kprobe(p) arch_arm_kprobe(p)
957 #define __disarm_kprobe(p, o) arch_disarm_kprobe(p)
958 #define kprobe_disarmed(p) kprobe_disabled(p)
959 #define wait_for_kprobe_optimizer() do {} while (0)
961 static int reuse_unused_kprobe(struct kprobe *ap)
964 * If the optimized kprobe is NOT supported, the aggr kprobe is
965 * released at the same time that the last aggregated kprobe is
967 * Thus there should be no chance to reuse unused kprobe.
969 printk(KERN_ERR "Error: There should be no unused kprobe here.\n");
973 static void free_aggr_kprobe(struct kprobe *p)
975 arch_remove_kprobe(p);
979 static struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
981 return kzalloc(sizeof(struct kprobe), GFP_KERNEL);
983 #endif /* CONFIG_OPTPROBES */
985 #ifdef CONFIG_KPROBES_ON_FTRACE
986 static struct ftrace_ops kprobe_ftrace_ops __read_mostly = {
987 .func = kprobe_ftrace_handler,
988 .flags = FTRACE_OPS_FL_SAVE_REGS,
991 static struct ftrace_ops kprobe_ipmodify_ops __read_mostly = {
992 .func = kprobe_ftrace_handler,
993 .flags = FTRACE_OPS_FL_SAVE_REGS | FTRACE_OPS_FL_IPMODIFY,
996 static int kprobe_ipmodify_enabled;
997 static int kprobe_ftrace_enabled;
999 /* Must ensure p->addr is really on ftrace */
1000 static int prepare_kprobe(struct kprobe *p)
1002 if (!kprobe_ftrace(p))
1003 return arch_prepare_kprobe(p);
1005 return arch_prepare_kprobe_ftrace(p);
1008 /* Caller must lock kprobe_mutex */
1009 static int __arm_kprobe_ftrace(struct kprobe *p, struct ftrace_ops *ops,
1014 ret = ftrace_set_filter_ip(ops, (unsigned long)p->addr, 0, 0);
1016 pr_debug("Failed to arm kprobe-ftrace at %pS (%d)\n",
1022 ret = register_ftrace_function(ops);
1024 pr_debug("Failed to init kprobe-ftrace (%d)\n", ret);
1034 * At this point, sinec ops is not registered, we should be sefe from
1035 * registering empty filter.
1037 ftrace_set_filter_ip(ops, (unsigned long)p->addr, 1, 0);
1041 static int arm_kprobe_ftrace(struct kprobe *p)
1043 bool ipmodify = (p->post_handler != NULL);
1045 return __arm_kprobe_ftrace(p,
1046 ipmodify ? &kprobe_ipmodify_ops : &kprobe_ftrace_ops,
1047 ipmodify ? &kprobe_ipmodify_enabled : &kprobe_ftrace_enabled);
1050 /* Caller must lock kprobe_mutex */
1051 static int __disarm_kprobe_ftrace(struct kprobe *p, struct ftrace_ops *ops,
1057 ret = unregister_ftrace_function(ops);
1058 if (WARN(ret < 0, "Failed to unregister kprobe-ftrace (%d)\n", ret))
1064 ret = ftrace_set_filter_ip(ops, (unsigned long)p->addr, 1, 0);
1065 WARN_ONCE(ret < 0, "Failed to disarm kprobe-ftrace at %pS (%d)\n",
1070 static int disarm_kprobe_ftrace(struct kprobe *p)
1072 bool ipmodify = (p->post_handler != NULL);
1074 return __disarm_kprobe_ftrace(p,
1075 ipmodify ? &kprobe_ipmodify_ops : &kprobe_ftrace_ops,
1076 ipmodify ? &kprobe_ipmodify_enabled : &kprobe_ftrace_enabled);
1078 #else /* !CONFIG_KPROBES_ON_FTRACE */
1079 static inline int prepare_kprobe(struct kprobe *p)
1081 return arch_prepare_kprobe(p);
1084 static inline int arm_kprobe_ftrace(struct kprobe *p)
1089 static inline int disarm_kprobe_ftrace(struct kprobe *p)
1095 /* Arm a kprobe with text_mutex */
1096 static int arm_kprobe(struct kprobe *kp)
1098 if (unlikely(kprobe_ftrace(kp)))
1099 return arm_kprobe_ftrace(kp);
1102 mutex_lock(&text_mutex);
1104 mutex_unlock(&text_mutex);
1110 /* Disarm a kprobe with text_mutex */
1111 static int disarm_kprobe(struct kprobe *kp, bool reopt)
1113 if (unlikely(kprobe_ftrace(kp)))
1114 return disarm_kprobe_ftrace(kp);
1117 mutex_lock(&text_mutex);
1118 __disarm_kprobe(kp, reopt);
1119 mutex_unlock(&text_mutex);
1126 * Aggregate handlers for multiple kprobes support - these handlers
1127 * take care of invoking the individual kprobe handlers on p->list
1129 static int aggr_pre_handler(struct kprobe *p, struct pt_regs *regs)
1133 list_for_each_entry_rcu(kp, &p->list, list) {
1134 if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
1135 set_kprobe_instance(kp);
1136 if (kp->pre_handler(kp, regs))
1139 reset_kprobe_instance();
1143 NOKPROBE_SYMBOL(aggr_pre_handler);
1145 static void aggr_post_handler(struct kprobe *p, struct pt_regs *regs,
1146 unsigned long flags)
1150 list_for_each_entry_rcu(kp, &p->list, list) {
1151 if (kp->post_handler && likely(!kprobe_disabled(kp))) {
1152 set_kprobe_instance(kp);
1153 kp->post_handler(kp, regs, flags);
1154 reset_kprobe_instance();
1158 NOKPROBE_SYMBOL(aggr_post_handler);
1160 static int aggr_fault_handler(struct kprobe *p, struct pt_regs *regs,
1163 struct kprobe *cur = __this_cpu_read(kprobe_instance);
1166 * if we faulted "during" the execution of a user specified
1167 * probe handler, invoke just that probe's fault handler
1169 if (cur && cur->fault_handler) {
1170 if (cur->fault_handler(cur, regs, trapnr))
1175 NOKPROBE_SYMBOL(aggr_fault_handler);
1177 /* Walks the list and increments nmissed count for multiprobe case */
1178 void kprobes_inc_nmissed_count(struct kprobe *p)
1181 if (!kprobe_aggrprobe(p)) {
1184 list_for_each_entry_rcu(kp, &p->list, list)
1189 NOKPROBE_SYMBOL(kprobes_inc_nmissed_count);
1191 void recycle_rp_inst(struct kretprobe_instance *ri,
1192 struct hlist_head *head)
1194 struct kretprobe *rp = ri->rp;
1196 /* remove rp inst off the rprobe_inst_table */
1197 hlist_del(&ri->hlist);
1198 INIT_HLIST_NODE(&ri->hlist);
1200 raw_spin_lock(&rp->lock);
1201 hlist_add_head(&ri->hlist, &rp->free_instances);
1202 raw_spin_unlock(&rp->lock);
1205 hlist_add_head(&ri->hlist, head);
1207 NOKPROBE_SYMBOL(recycle_rp_inst);
1209 void kretprobe_hash_lock(struct task_struct *tsk,
1210 struct hlist_head **head, unsigned long *flags)
1211 __acquires(hlist_lock)
1213 unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
1214 raw_spinlock_t *hlist_lock;
1216 *head = &kretprobe_inst_table[hash];
1217 hlist_lock = kretprobe_table_lock_ptr(hash);
1218 raw_spin_lock_irqsave(hlist_lock, *flags);
1220 NOKPROBE_SYMBOL(kretprobe_hash_lock);
1222 static void kretprobe_table_lock(unsigned long hash,
1223 unsigned long *flags)
1224 __acquires(hlist_lock)
1226 raw_spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
1227 raw_spin_lock_irqsave(hlist_lock, *flags);
1229 NOKPROBE_SYMBOL(kretprobe_table_lock);
1231 void kretprobe_hash_unlock(struct task_struct *tsk,
1232 unsigned long *flags)
1233 __releases(hlist_lock)
1235 unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
1236 raw_spinlock_t *hlist_lock;
1238 hlist_lock = kretprobe_table_lock_ptr(hash);
1239 raw_spin_unlock_irqrestore(hlist_lock, *flags);
1241 NOKPROBE_SYMBOL(kretprobe_hash_unlock);
1243 static void kretprobe_table_unlock(unsigned long hash,
1244 unsigned long *flags)
1245 __releases(hlist_lock)
1247 raw_spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
1248 raw_spin_unlock_irqrestore(hlist_lock, *flags);
1250 NOKPROBE_SYMBOL(kretprobe_table_unlock);
1252 struct kprobe kprobe_busy = {
1253 .addr = (void *) get_kprobe,
1256 void kprobe_busy_begin(void)
1258 struct kprobe_ctlblk *kcb;
1261 __this_cpu_write(current_kprobe, &kprobe_busy);
1262 kcb = get_kprobe_ctlblk();
1263 kcb->kprobe_status = KPROBE_HIT_ACTIVE;
1266 void kprobe_busy_end(void)
1268 __this_cpu_write(current_kprobe, NULL);
1273 * This function is called from finish_task_switch when task tk becomes dead,
1274 * so that we can recycle any function-return probe instances associated
1275 * with this task. These left over instances represent probed functions
1276 * that have been called but will never return.
1278 void kprobe_flush_task(struct task_struct *tk)
1280 struct kretprobe_instance *ri;
1281 struct hlist_head *head, empty_rp;
1282 struct hlist_node *tmp;
1283 unsigned long hash, flags = 0;
1285 if (unlikely(!kprobes_initialized))
1286 /* Early boot. kretprobe_table_locks not yet initialized. */
1289 kprobe_busy_begin();
1291 INIT_HLIST_HEAD(&empty_rp);
1292 hash = hash_ptr(tk, KPROBE_HASH_BITS);
1293 head = &kretprobe_inst_table[hash];
1294 kretprobe_table_lock(hash, &flags);
1295 hlist_for_each_entry_safe(ri, tmp, head, hlist) {
1297 recycle_rp_inst(ri, &empty_rp);
1299 kretprobe_table_unlock(hash, &flags);
1300 hlist_for_each_entry_safe(ri, tmp, &empty_rp, hlist) {
1301 hlist_del(&ri->hlist);
1307 NOKPROBE_SYMBOL(kprobe_flush_task);
1309 static inline void free_rp_inst(struct kretprobe *rp)
1311 struct kretprobe_instance *ri;
1312 struct hlist_node *next;
1314 hlist_for_each_entry_safe(ri, next, &rp->free_instances, hlist) {
1315 hlist_del(&ri->hlist);
1320 static void cleanup_rp_inst(struct kretprobe *rp)
1322 unsigned long flags, hash;
1323 struct kretprobe_instance *ri;
1324 struct hlist_node *next;
1325 struct hlist_head *head;
1328 for (hash = 0; hash < KPROBE_TABLE_SIZE; hash++) {
1329 kretprobe_table_lock(hash, &flags);
1330 head = &kretprobe_inst_table[hash];
1331 hlist_for_each_entry_safe(ri, next, head, hlist) {
1335 kretprobe_table_unlock(hash, &flags);
1339 NOKPROBE_SYMBOL(cleanup_rp_inst);
1341 /* Add the new probe to ap->list */
1342 static int add_new_kprobe(struct kprobe *ap, struct kprobe *p)
1344 if (p->post_handler)
1345 unoptimize_kprobe(ap, true); /* Fall back to normal kprobe */
1347 list_add_rcu(&p->list, &ap->list);
1348 if (p->post_handler && !ap->post_handler)
1349 ap->post_handler = aggr_post_handler;
1355 * Fill in the required fields of the "manager kprobe". Replace the
1356 * earlier kprobe in the hlist with the manager kprobe
1358 static void init_aggr_kprobe(struct kprobe *ap, struct kprobe *p)
1360 /* Copy p's insn slot to ap */
1362 flush_insn_slot(ap);
1364 ap->flags = p->flags & ~KPROBE_FLAG_OPTIMIZED;
1365 ap->pre_handler = aggr_pre_handler;
1366 ap->fault_handler = aggr_fault_handler;
1367 /* We don't care the kprobe which has gone. */
1368 if (p->post_handler && !kprobe_gone(p))
1369 ap->post_handler = aggr_post_handler;
1371 INIT_LIST_HEAD(&ap->list);
1372 INIT_HLIST_NODE(&ap->hlist);
1374 list_add_rcu(&p->list, &ap->list);
1375 hlist_replace_rcu(&p->hlist, &ap->hlist);
1379 * This is the second or subsequent kprobe at the address - handle
1382 static int register_aggr_kprobe(struct kprobe *orig_p, struct kprobe *p)
1385 struct kprobe *ap = orig_p;
1389 /* For preparing optimization, jump_label_text_reserved() is called */
1391 mutex_lock(&text_mutex);
1393 if (!kprobe_aggrprobe(orig_p)) {
1394 /* If orig_p is not an aggr_kprobe, create new aggr_kprobe. */
1395 ap = alloc_aggr_kprobe(orig_p);
1400 init_aggr_kprobe(ap, orig_p);
1401 } else if (kprobe_unused(ap)) {
1402 /* This probe is going to die. Rescue it */
1403 ret = reuse_unused_kprobe(ap);
1408 if (kprobe_gone(ap)) {
1410 * Attempting to insert new probe at the same location that
1411 * had a probe in the module vaddr area which already
1412 * freed. So, the instruction slot has already been
1413 * released. We need a new slot for the new probe.
1415 ret = arch_prepare_kprobe(ap);
1418 * Even if fail to allocate new slot, don't need to
1419 * free aggr_probe. It will be used next time, or
1420 * freed by unregister_kprobe.
1424 /* Prepare optimized instructions if possible. */
1425 prepare_optimized_kprobe(ap);
1428 * Clear gone flag to prevent allocating new slot again, and
1429 * set disabled flag because it is not armed yet.
1431 ap->flags = (ap->flags & ~KPROBE_FLAG_GONE)
1432 | KPROBE_FLAG_DISABLED;
1435 /* Copy ap's insn slot to p */
1437 ret = add_new_kprobe(ap, p);
1440 mutex_unlock(&text_mutex);
1441 jump_label_unlock();
1444 if (ret == 0 && kprobe_disabled(ap) && !kprobe_disabled(p)) {
1445 ap->flags &= ~KPROBE_FLAG_DISABLED;
1446 if (!kprobes_all_disarmed) {
1447 /* Arm the breakpoint again. */
1448 ret = arm_kprobe(ap);
1450 ap->flags |= KPROBE_FLAG_DISABLED;
1451 list_del_rcu(&p->list);
1459 bool __weak arch_within_kprobe_blacklist(unsigned long addr)
1461 /* The __kprobes marked functions and entry code must not be probed */
1462 return addr >= (unsigned long)__kprobes_text_start &&
1463 addr < (unsigned long)__kprobes_text_end;
1466 static bool __within_kprobe_blacklist(unsigned long addr)
1468 struct kprobe_blacklist_entry *ent;
1470 if (arch_within_kprobe_blacklist(addr))
1473 * If there exists a kprobe_blacklist, verify and
1474 * fail any probe registration in the prohibited area
1476 list_for_each_entry(ent, &kprobe_blacklist, list) {
1477 if (addr >= ent->start_addr && addr < ent->end_addr)
1483 bool within_kprobe_blacklist(unsigned long addr)
1485 char symname[KSYM_NAME_LEN], *p;
1487 if (__within_kprobe_blacklist(addr))
1490 /* Check if the address is on a suffixed-symbol */
1491 if (!lookup_symbol_name(addr, symname)) {
1492 p = strchr(symname, '.');
1496 addr = (unsigned long)kprobe_lookup_name(symname, 0);
1498 return __within_kprobe_blacklist(addr);
1504 * If we have a symbol_name argument, look it up and add the offset field
1505 * to it. This way, we can specify a relative address to a symbol.
1506 * This returns encoded errors if it fails to look up symbol or invalid
1507 * combination of parameters.
1509 static kprobe_opcode_t *_kprobe_addr(kprobe_opcode_t *addr,
1510 const char *symbol_name, unsigned int offset)
1512 if ((symbol_name && addr) || (!symbol_name && !addr))
1516 addr = kprobe_lookup_name(symbol_name, offset);
1518 return ERR_PTR(-ENOENT);
1521 addr = (kprobe_opcode_t *)(((char *)addr) + offset);
1526 return ERR_PTR(-EINVAL);
1529 static kprobe_opcode_t *kprobe_addr(struct kprobe *p)
1531 return _kprobe_addr(p->addr, p->symbol_name, p->offset);
1534 /* Check passed kprobe is valid and return kprobe in kprobe_table. */
1535 static struct kprobe *__get_valid_kprobe(struct kprobe *p)
1537 struct kprobe *ap, *list_p;
1539 ap = get_kprobe(p->addr);
1544 list_for_each_entry_rcu(list_p, &ap->list, list)
1546 /* kprobe p is a valid probe */
1554 /* Return error if the kprobe is being re-registered */
1555 static inline int check_kprobe_rereg(struct kprobe *p)
1559 mutex_lock(&kprobe_mutex);
1560 if (__get_valid_kprobe(p))
1562 mutex_unlock(&kprobe_mutex);
1567 int __weak arch_check_ftrace_location(struct kprobe *p)
1569 unsigned long ftrace_addr;
1571 ftrace_addr = ftrace_location((unsigned long)p->addr);
1573 #ifdef CONFIG_KPROBES_ON_FTRACE
1574 /* Given address is not on the instruction boundary */
1575 if ((unsigned long)p->addr != ftrace_addr)
1577 p->flags |= KPROBE_FLAG_FTRACE;
1578 #else /* !CONFIG_KPROBES_ON_FTRACE */
1585 static int check_kprobe_address_safe(struct kprobe *p,
1586 struct module **probed_mod)
1590 ret = arch_check_ftrace_location(p);
1596 /* Ensure it is not in reserved area nor out of text */
1597 if (!(core_kernel_text((unsigned long) p->addr) ||
1598 is_module_text_address((unsigned long) p->addr)) ||
1599 in_gate_area_no_mm((unsigned long) p->addr) ||
1600 within_kprobe_blacklist((unsigned long) p->addr) ||
1601 jump_label_text_reserved(p->addr, p->addr) ||
1602 find_bug((unsigned long)p->addr)) {
1607 /* Check if are we probing a module */
1608 *probed_mod = __module_text_address((unsigned long) p->addr);
1611 * We must hold a refcount of the probed module while updating
1612 * its code to prohibit unexpected unloading.
1614 if (unlikely(!try_module_get(*probed_mod))) {
1620 * If the module freed .init.text, we couldn't insert
1623 if (within_module_init((unsigned long)p->addr, *probed_mod) &&
1624 (*probed_mod)->state != MODULE_STATE_COMING) {
1625 module_put(*probed_mod);
1632 jump_label_unlock();
1637 int register_kprobe(struct kprobe *p)
1640 struct kprobe *old_p;
1641 struct module *probed_mod;
1642 kprobe_opcode_t *addr;
1644 /* Adjust probe address from symbol */
1645 addr = kprobe_addr(p);
1647 return PTR_ERR(addr);
1650 ret = check_kprobe_rereg(p);
1654 /* User can pass only KPROBE_FLAG_DISABLED to register_kprobe */
1655 p->flags &= KPROBE_FLAG_DISABLED;
1657 INIT_LIST_HEAD(&p->list);
1659 ret = check_kprobe_address_safe(p, &probed_mod);
1663 mutex_lock(&kprobe_mutex);
1665 old_p = get_kprobe(p->addr);
1667 /* Since this may unoptimize old_p, locking text_mutex. */
1668 ret = register_aggr_kprobe(old_p, p);
1673 /* Prevent text modification */
1674 mutex_lock(&text_mutex);
1675 ret = prepare_kprobe(p);
1676 mutex_unlock(&text_mutex);
1681 INIT_HLIST_NODE(&p->hlist);
1682 hlist_add_head_rcu(&p->hlist,
1683 &kprobe_table[hash_ptr(p->addr, KPROBE_HASH_BITS)]);
1685 if (!kprobes_all_disarmed && !kprobe_disabled(p)) {
1686 ret = arm_kprobe(p);
1688 hlist_del_rcu(&p->hlist);
1694 /* Try to optimize kprobe */
1695 try_to_optimize_kprobe(p);
1697 mutex_unlock(&kprobe_mutex);
1700 module_put(probed_mod);
1704 EXPORT_SYMBOL_GPL(register_kprobe);
1706 /* Check if all probes on the aggrprobe are disabled */
1707 static int aggr_kprobe_disabled(struct kprobe *ap)
1711 list_for_each_entry_rcu(kp, &ap->list, list)
1712 if (!kprobe_disabled(kp))
1714 * There is an active probe on the list.
1715 * We can't disable this ap.
1722 /* Disable one kprobe: Make sure called under kprobe_mutex is locked */
1723 static struct kprobe *__disable_kprobe(struct kprobe *p)
1725 struct kprobe *orig_p;
1728 /* Get an original kprobe for return */
1729 orig_p = __get_valid_kprobe(p);
1730 if (unlikely(orig_p == NULL))
1731 return ERR_PTR(-EINVAL);
1733 if (!kprobe_disabled(p)) {
1734 /* Disable probe if it is a child probe */
1736 p->flags |= KPROBE_FLAG_DISABLED;
1738 /* Try to disarm and disable this/parent probe */
1739 if (p == orig_p || aggr_kprobe_disabled(orig_p)) {
1741 * Don't be lazy here. Even if 'kprobes_all_disarmed'
1742 * is false, 'orig_p' might not have been armed yet.
1743 * Note arm_all_kprobes() __tries__ to arm all kprobes
1744 * on the best effort basis.
1746 if (!kprobes_all_disarmed && !kprobe_disabled(orig_p)) {
1747 ret = disarm_kprobe(orig_p, true);
1749 p->flags &= ~KPROBE_FLAG_DISABLED;
1750 return ERR_PTR(ret);
1753 orig_p->flags |= KPROBE_FLAG_DISABLED;
1761 * Unregister a kprobe without a scheduler synchronization.
1763 static int __unregister_kprobe_top(struct kprobe *p)
1765 struct kprobe *ap, *list_p;
1767 /* Disable kprobe. This will disarm it if needed. */
1768 ap = __disable_kprobe(p);
1774 * This probe is an independent(and non-optimized) kprobe
1775 * (not an aggrprobe). Remove from the hash list.
1779 /* Following process expects this probe is an aggrprobe */
1780 WARN_ON(!kprobe_aggrprobe(ap));
1782 if (list_is_singular(&ap->list) && kprobe_disarmed(ap))
1784 * !disarmed could be happen if the probe is under delayed
1789 /* If disabling probe has special handlers, update aggrprobe */
1790 if (p->post_handler && !kprobe_gone(p)) {
1791 list_for_each_entry_rcu(list_p, &ap->list, list) {
1792 if ((list_p != p) && (list_p->post_handler))
1796 * For the kprobe-on-ftrace case, we keep the
1797 * post_handler setting to identify this aggrprobe
1798 * armed with kprobe_ipmodify_ops.
1800 if (!kprobe_ftrace(ap))
1801 ap->post_handler = NULL;
1805 * Remove from the aggrprobe: this path will do nothing in
1806 * __unregister_kprobe_bottom().
1808 list_del_rcu(&p->list);
1809 if (!kprobe_disabled(ap) && !kprobes_all_disarmed)
1811 * Try to optimize this probe again, because post
1812 * handler may have been changed.
1814 optimize_kprobe(ap);
1819 hlist_del_rcu(&ap->hlist);
1823 static void __unregister_kprobe_bottom(struct kprobe *p)
1827 if (list_empty(&p->list))
1828 /* This is an independent kprobe */
1829 arch_remove_kprobe(p);
1830 else if (list_is_singular(&p->list)) {
1831 /* This is the last child of an aggrprobe */
1832 ap = list_entry(p->list.next, struct kprobe, list);
1834 free_aggr_kprobe(ap);
1836 /* Otherwise, do nothing. */
1839 int register_kprobes(struct kprobe **kps, int num)
1845 for (i = 0; i < num; i++) {
1846 ret = register_kprobe(kps[i]);
1849 unregister_kprobes(kps, i);
1855 EXPORT_SYMBOL_GPL(register_kprobes);
1857 void unregister_kprobe(struct kprobe *p)
1859 unregister_kprobes(&p, 1);
1861 EXPORT_SYMBOL_GPL(unregister_kprobe);
1863 void unregister_kprobes(struct kprobe **kps, int num)
1869 mutex_lock(&kprobe_mutex);
1870 for (i = 0; i < num; i++)
1871 if (__unregister_kprobe_top(kps[i]) < 0)
1872 kps[i]->addr = NULL;
1873 mutex_unlock(&kprobe_mutex);
1876 for (i = 0; i < num; i++)
1878 __unregister_kprobe_bottom(kps[i]);
1880 EXPORT_SYMBOL_GPL(unregister_kprobes);
1882 int __weak kprobe_exceptions_notify(struct notifier_block *self,
1883 unsigned long val, void *data)
1887 NOKPROBE_SYMBOL(kprobe_exceptions_notify);
1889 static struct notifier_block kprobe_exceptions_nb = {
1890 .notifier_call = kprobe_exceptions_notify,
1891 .priority = 0x7fffffff /* we need to be notified first */
1894 unsigned long __weak arch_deref_entry_point(void *entry)
1896 return (unsigned long)entry;
1899 #ifdef CONFIG_KRETPROBES
1901 * This kprobe pre_handler is registered with every kretprobe. When probe
1902 * hits it will set up the return probe.
1904 static int pre_handler_kretprobe(struct kprobe *p, struct pt_regs *regs)
1906 struct kretprobe *rp = container_of(p, struct kretprobe, kp);
1907 unsigned long hash, flags = 0;
1908 struct kretprobe_instance *ri;
1911 * To avoid deadlocks, prohibit return probing in NMI contexts,
1912 * just skip the probe and increase the (inexact) 'nmissed'
1913 * statistical counter, so that the user is informed that
1914 * something happened:
1916 if (unlikely(in_nmi())) {
1921 /* TODO: consider to only swap the RA after the last pre_handler fired */
1922 hash = hash_ptr(current, KPROBE_HASH_BITS);
1923 raw_spin_lock_irqsave(&rp->lock, flags);
1924 if (!hlist_empty(&rp->free_instances)) {
1925 ri = hlist_entry(rp->free_instances.first,
1926 struct kretprobe_instance, hlist);
1927 hlist_del(&ri->hlist);
1928 raw_spin_unlock_irqrestore(&rp->lock, flags);
1933 if (rp->entry_handler && rp->entry_handler(ri, regs)) {
1934 raw_spin_lock_irqsave(&rp->lock, flags);
1935 hlist_add_head(&ri->hlist, &rp->free_instances);
1936 raw_spin_unlock_irqrestore(&rp->lock, flags);
1940 arch_prepare_kretprobe(ri, regs);
1942 /* XXX(hch): why is there no hlist_move_head? */
1943 INIT_HLIST_NODE(&ri->hlist);
1944 kretprobe_table_lock(hash, &flags);
1945 hlist_add_head(&ri->hlist, &kretprobe_inst_table[hash]);
1946 kretprobe_table_unlock(hash, &flags);
1949 raw_spin_unlock_irqrestore(&rp->lock, flags);
1953 NOKPROBE_SYMBOL(pre_handler_kretprobe);
1955 bool __weak arch_kprobe_on_func_entry(unsigned long offset)
1961 * kprobe_on_func_entry() -- check whether given address is function entry
1962 * @addr: Target address
1963 * @sym: Target symbol name
1964 * @offset: The offset from the symbol or the address
1966 * This checks whether the given @addr+@offset or @sym+@offset is on the
1967 * function entry address or not.
1968 * This returns 0 if it is the function entry, or -EINVAL if it is not.
1969 * And also it returns -ENOENT if it fails the symbol or address lookup.
1970 * Caller must pass @addr or @sym (either one must be NULL), or this
1973 int kprobe_on_func_entry(kprobe_opcode_t *addr, const char *sym, unsigned long offset)
1975 kprobe_opcode_t *kp_addr = _kprobe_addr(addr, sym, offset);
1977 if (IS_ERR(kp_addr))
1978 return PTR_ERR(kp_addr);
1980 if (!kallsyms_lookup_size_offset((unsigned long)kp_addr, NULL, &offset))
1983 if (!arch_kprobe_on_func_entry(offset))
1989 int register_kretprobe(struct kretprobe *rp)
1992 struct kretprobe_instance *inst;
1996 ret = kprobe_on_func_entry(rp->kp.addr, rp->kp.symbol_name, rp->kp.offset);
2000 /* If only rp->kp.addr is specified, check reregistering kprobes */
2001 if (rp->kp.addr && check_kprobe_rereg(&rp->kp))
2004 if (kretprobe_blacklist_size) {
2005 addr = kprobe_addr(&rp->kp);
2007 return PTR_ERR(addr);
2009 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
2010 if (kretprobe_blacklist[i].addr == addr)
2015 if (rp->data_size > KRETPROBE_MAX_DATA_SIZE)
2018 rp->kp.pre_handler = pre_handler_kretprobe;
2019 rp->kp.post_handler = NULL;
2020 rp->kp.fault_handler = NULL;
2022 /* Pre-allocate memory for max kretprobe instances */
2023 if (rp->maxactive <= 0) {
2024 #ifdef CONFIG_PREEMPTION
2025 rp->maxactive = max_t(unsigned int, 10, 2*num_possible_cpus());
2027 rp->maxactive = num_possible_cpus();
2030 raw_spin_lock_init(&rp->lock);
2031 INIT_HLIST_HEAD(&rp->free_instances);
2032 for (i = 0; i < rp->maxactive; i++) {
2033 inst = kmalloc(sizeof(struct kretprobe_instance) +
2034 rp->data_size, GFP_KERNEL);
2039 INIT_HLIST_NODE(&inst->hlist);
2040 hlist_add_head(&inst->hlist, &rp->free_instances);
2044 /* Establish function entry probe point */
2045 ret = register_kprobe(&rp->kp);
2050 EXPORT_SYMBOL_GPL(register_kretprobe);
2052 int register_kretprobes(struct kretprobe **rps, int num)
2058 for (i = 0; i < num; i++) {
2059 ret = register_kretprobe(rps[i]);
2062 unregister_kretprobes(rps, i);
2068 EXPORT_SYMBOL_GPL(register_kretprobes);
2070 void unregister_kretprobe(struct kretprobe *rp)
2072 unregister_kretprobes(&rp, 1);
2074 EXPORT_SYMBOL_GPL(unregister_kretprobe);
2076 void unregister_kretprobes(struct kretprobe **rps, int num)
2082 mutex_lock(&kprobe_mutex);
2083 for (i = 0; i < num; i++)
2084 if (__unregister_kprobe_top(&rps[i]->kp) < 0)
2085 rps[i]->kp.addr = NULL;
2086 mutex_unlock(&kprobe_mutex);
2089 for (i = 0; i < num; i++) {
2090 if (rps[i]->kp.addr) {
2091 __unregister_kprobe_bottom(&rps[i]->kp);
2092 cleanup_rp_inst(rps[i]);
2096 EXPORT_SYMBOL_GPL(unregister_kretprobes);
2098 #else /* CONFIG_KRETPROBES */
2099 int register_kretprobe(struct kretprobe *rp)
2103 EXPORT_SYMBOL_GPL(register_kretprobe);
2105 int register_kretprobes(struct kretprobe **rps, int num)
2109 EXPORT_SYMBOL_GPL(register_kretprobes);
2111 void unregister_kretprobe(struct kretprobe *rp)
2114 EXPORT_SYMBOL_GPL(unregister_kretprobe);
2116 void unregister_kretprobes(struct kretprobe **rps, int num)
2119 EXPORT_SYMBOL_GPL(unregister_kretprobes);
2121 static int pre_handler_kretprobe(struct kprobe *p, struct pt_regs *regs)
2125 NOKPROBE_SYMBOL(pre_handler_kretprobe);
2127 #endif /* CONFIG_KRETPROBES */
2129 /* Set the kprobe gone and remove its instruction buffer. */
2130 static void kill_kprobe(struct kprobe *p)
2134 if (WARN_ON_ONCE(kprobe_gone(p)))
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_rcu(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;
2211 kp->flags |= KPROBE_FLAG_DISABLED;
2215 mutex_unlock(&kprobe_mutex);
2218 EXPORT_SYMBOL_GPL(enable_kprobe);
2220 /* Caller must NOT call this in usual path. This is only for critical case */
2221 void dump_kprobe(struct kprobe *kp)
2223 pr_err("Dumping kprobe:\n");
2224 pr_err("Name: %s\nOffset: %x\nAddress: %pS\n",
2225 kp->symbol_name, kp->offset, kp->addr);
2227 NOKPROBE_SYMBOL(dump_kprobe);
2229 int kprobe_add_ksym_blacklist(unsigned long entry)
2231 struct kprobe_blacklist_entry *ent;
2232 unsigned long offset = 0, size = 0;
2234 if (!kernel_text_address(entry) ||
2235 !kallsyms_lookup_size_offset(entry, &size, &offset))
2238 ent = kmalloc(sizeof(*ent), GFP_KERNEL);
2241 ent->start_addr = entry;
2242 ent->end_addr = entry + size;
2243 INIT_LIST_HEAD(&ent->list);
2244 list_add_tail(&ent->list, &kprobe_blacklist);
2249 /* Add all symbols in given area into kprobe blacklist */
2250 int kprobe_add_area_blacklist(unsigned long start, unsigned long end)
2252 unsigned long entry;
2255 for (entry = start; entry < end; entry += ret) {
2256 ret = kprobe_add_ksym_blacklist(entry);
2259 if (ret == 0) /* In case of alias symbol */
2265 int __init __weak arch_populate_kprobe_blacklist(void)
2271 * Lookup and populate the kprobe_blacklist.
2273 * Unlike the kretprobe blacklist, we'll need to determine
2274 * the range of addresses that belong to the said functions,
2275 * since a kprobe need not necessarily be at the beginning
2278 static int __init populate_kprobe_blacklist(unsigned long *start,
2281 unsigned long entry;
2282 unsigned long *iter;
2285 for (iter = start; iter < end; iter++) {
2286 entry = arch_deref_entry_point((void *)*iter);
2287 ret = kprobe_add_ksym_blacklist(entry);
2294 /* Symbols in __kprobes_text are blacklisted */
2295 ret = kprobe_add_area_blacklist((unsigned long)__kprobes_text_start,
2296 (unsigned long)__kprobes_text_end);
2298 return ret ? : arch_populate_kprobe_blacklist();
2301 /* Module notifier call back, checking kprobes on the module */
2302 static int kprobes_module_callback(struct notifier_block *nb,
2303 unsigned long val, void *data)
2305 struct module *mod = data;
2306 struct hlist_head *head;
2309 int checkcore = (val == MODULE_STATE_GOING);
2311 if (val != MODULE_STATE_GOING && val != MODULE_STATE_LIVE)
2315 * When MODULE_STATE_GOING was notified, both of module .text and
2316 * .init.text sections would be freed. When MODULE_STATE_LIVE was
2317 * notified, only .init.text section would be freed. We need to
2318 * disable kprobes which have been inserted in the sections.
2320 mutex_lock(&kprobe_mutex);
2321 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2322 head = &kprobe_table[i];
2323 hlist_for_each_entry_rcu(p, head, hlist) {
2327 if (within_module_init((unsigned long)p->addr, mod) ||
2329 within_module_core((unsigned long)p->addr, mod))) {
2331 * The vaddr this probe is installed will soon
2332 * be vfreed buy not synced to disk. Hence,
2333 * disarming the breakpoint isn't needed.
2335 * Note, this will also move any optimized probes
2336 * that are pending to be removed from their
2337 * corresponding lists to the freeing_list and
2338 * will not be touched by the delayed
2339 * kprobe_optimizer work handler.
2345 mutex_unlock(&kprobe_mutex);
2349 static struct notifier_block kprobe_module_nb = {
2350 .notifier_call = kprobes_module_callback,
2354 /* Markers of _kprobe_blacklist section */
2355 extern unsigned long __start_kprobe_blacklist[];
2356 extern unsigned long __stop_kprobe_blacklist[];
2358 void kprobe_free_init_mem(void)
2360 void *start = (void *)(&__init_begin);
2361 void *end = (void *)(&__init_end);
2362 struct hlist_head *head;
2366 mutex_lock(&kprobe_mutex);
2368 /* Kill all kprobes on initmem */
2369 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2370 head = &kprobe_table[i];
2371 hlist_for_each_entry(p, head, hlist) {
2372 if (start <= (void *)p->addr && (void *)p->addr < end)
2377 mutex_unlock(&kprobe_mutex);
2380 static int __init init_kprobes(void)
2384 /* FIXME allocate the probe table, currently defined statically */
2385 /* initialize all list heads */
2386 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2387 INIT_HLIST_HEAD(&kprobe_table[i]);
2388 INIT_HLIST_HEAD(&kretprobe_inst_table[i]);
2389 raw_spin_lock_init(&(kretprobe_table_locks[i].lock));
2392 err = populate_kprobe_blacklist(__start_kprobe_blacklist,
2393 __stop_kprobe_blacklist);
2395 pr_err("kprobes: failed to populate blacklist: %d\n", err);
2396 pr_err("Please take care of using kprobes.\n");
2399 if (kretprobe_blacklist_size) {
2400 /* lookup the function address from its name */
2401 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
2402 kretprobe_blacklist[i].addr =
2403 kprobe_lookup_name(kretprobe_blacklist[i].name, 0);
2404 if (!kretprobe_blacklist[i].addr)
2405 printk("kretprobe: lookup failed: %s\n",
2406 kretprobe_blacklist[i].name);
2410 #if defined(CONFIG_OPTPROBES)
2411 #if defined(__ARCH_WANT_KPROBES_INSN_SLOT)
2412 /* Init kprobe_optinsn_slots */
2413 kprobe_optinsn_slots.insn_size = MAX_OPTINSN_SIZE;
2415 /* By default, kprobes can be optimized */
2416 kprobes_allow_optimization = true;
2419 /* By default, kprobes are armed */
2420 kprobes_all_disarmed = false;
2422 err = arch_init_kprobes();
2424 err = register_die_notifier(&kprobe_exceptions_nb);
2426 err = register_module_notifier(&kprobe_module_nb);
2428 kprobes_initialized = (err == 0);
2434 subsys_initcall(init_kprobes);
2436 #ifdef CONFIG_DEBUG_FS
2437 static void report_probe(struct seq_file *pi, struct kprobe *p,
2438 const char *sym, int offset, char *modname, struct kprobe *pp)
2441 void *addr = p->addr;
2443 if (p->pre_handler == pre_handler_kretprobe)
2448 if (!kallsyms_show_value(pi->file->f_cred))
2452 seq_printf(pi, "%px %s %s+0x%x %s ",
2453 addr, kprobe_type, sym, offset,
2454 (modname ? modname : " "));
2455 else /* try to use %pS */
2456 seq_printf(pi, "%px %s %pS ",
2457 addr, kprobe_type, p->addr);
2461 seq_printf(pi, "%s%s%s%s\n",
2462 (kprobe_gone(p) ? "[GONE]" : ""),
2463 ((kprobe_disabled(p) && !kprobe_gone(p)) ? "[DISABLED]" : ""),
2464 (kprobe_optimized(pp) ? "[OPTIMIZED]" : ""),
2465 (kprobe_ftrace(pp) ? "[FTRACE]" : ""));
2468 static void *kprobe_seq_start(struct seq_file *f, loff_t *pos)
2470 return (*pos < KPROBE_TABLE_SIZE) ? pos : NULL;
2473 static void *kprobe_seq_next(struct seq_file *f, void *v, loff_t *pos)
2476 if (*pos >= KPROBE_TABLE_SIZE)
2481 static void kprobe_seq_stop(struct seq_file *f, void *v)
2486 static int show_kprobe_addr(struct seq_file *pi, void *v)
2488 struct hlist_head *head;
2489 struct kprobe *p, *kp;
2490 const char *sym = NULL;
2491 unsigned int i = *(loff_t *) v;
2492 unsigned long offset = 0;
2493 char *modname, namebuf[KSYM_NAME_LEN];
2495 head = &kprobe_table[i];
2497 hlist_for_each_entry_rcu(p, head, hlist) {
2498 sym = kallsyms_lookup((unsigned long)p->addr, NULL,
2499 &offset, &modname, namebuf);
2500 if (kprobe_aggrprobe(p)) {
2501 list_for_each_entry_rcu(kp, &p->list, list)
2502 report_probe(pi, kp, sym, offset, modname, p);
2504 report_probe(pi, p, sym, offset, modname, NULL);
2510 static const struct seq_operations kprobes_seq_ops = {
2511 .start = kprobe_seq_start,
2512 .next = kprobe_seq_next,
2513 .stop = kprobe_seq_stop,
2514 .show = show_kprobe_addr
2517 static int kprobes_open(struct inode *inode, struct file *filp)
2519 return seq_open(filp, &kprobes_seq_ops);
2522 static const struct file_operations debugfs_kprobes_operations = {
2523 .open = kprobes_open,
2525 .llseek = seq_lseek,
2526 .release = seq_release,
2529 /* kprobes/blacklist -- shows which functions can not be probed */
2530 static void *kprobe_blacklist_seq_start(struct seq_file *m, loff_t *pos)
2532 return seq_list_start(&kprobe_blacklist, *pos);
2535 static void *kprobe_blacklist_seq_next(struct seq_file *m, void *v, loff_t *pos)
2537 return seq_list_next(v, &kprobe_blacklist, pos);
2540 static int kprobe_blacklist_seq_show(struct seq_file *m, void *v)
2542 struct kprobe_blacklist_entry *ent =
2543 list_entry(v, struct kprobe_blacklist_entry, list);
2546 * If /proc/kallsyms is not showing kernel address, we won't
2547 * show them here either.
2549 if (!kallsyms_show_value(m->file->f_cred))
2550 seq_printf(m, "0x%px-0x%px\t%ps\n", NULL, NULL,
2551 (void *)ent->start_addr);
2553 seq_printf(m, "0x%px-0x%px\t%ps\n", (void *)ent->start_addr,
2554 (void *)ent->end_addr, (void *)ent->start_addr);
2558 static const struct seq_operations kprobe_blacklist_seq_ops = {
2559 .start = kprobe_blacklist_seq_start,
2560 .next = kprobe_blacklist_seq_next,
2561 .stop = kprobe_seq_stop, /* Reuse void function */
2562 .show = kprobe_blacklist_seq_show,
2565 static int kprobe_blacklist_open(struct inode *inode, struct file *filp)
2567 return seq_open(filp, &kprobe_blacklist_seq_ops);
2570 static const struct file_operations debugfs_kprobe_blacklist_ops = {
2571 .open = kprobe_blacklist_open,
2573 .llseek = seq_lseek,
2574 .release = seq_release,
2577 static int arm_all_kprobes(void)
2579 struct hlist_head *head;
2581 unsigned int i, total = 0, errors = 0;
2584 mutex_lock(&kprobe_mutex);
2586 /* If kprobes are armed, just return */
2587 if (!kprobes_all_disarmed)
2588 goto already_enabled;
2591 * optimize_kprobe() called by arm_kprobe() checks
2592 * kprobes_all_disarmed, so set kprobes_all_disarmed before
2595 kprobes_all_disarmed = false;
2596 /* Arming kprobes doesn't optimize kprobe itself */
2597 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2598 head = &kprobe_table[i];
2599 /* Arm all kprobes on a best-effort basis */
2600 hlist_for_each_entry_rcu(p, head, hlist) {
2601 if (!kprobe_disabled(p)) {
2602 err = arm_kprobe(p);
2613 pr_warn("Kprobes globally enabled, but failed to arm %d out of %d probes\n",
2616 pr_info("Kprobes globally enabled\n");
2619 mutex_unlock(&kprobe_mutex);
2623 static int disarm_all_kprobes(void)
2625 struct hlist_head *head;
2627 unsigned int i, total = 0, errors = 0;
2630 mutex_lock(&kprobe_mutex);
2632 /* If kprobes are already disarmed, just return */
2633 if (kprobes_all_disarmed) {
2634 mutex_unlock(&kprobe_mutex);
2638 kprobes_all_disarmed = true;
2640 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2641 head = &kprobe_table[i];
2642 /* Disarm all kprobes on a best-effort basis */
2643 hlist_for_each_entry_rcu(p, head, hlist) {
2644 if (!arch_trampoline_kprobe(p) && !kprobe_disabled(p)) {
2645 err = disarm_kprobe(p, false);
2656 pr_warn("Kprobes globally disabled, but failed to disarm %d out of %d probes\n",
2659 pr_info("Kprobes globally disabled\n");
2661 mutex_unlock(&kprobe_mutex);
2663 /* Wait for disarming all kprobes by optimizer */
2664 wait_for_kprobe_optimizer();
2670 * XXX: The debugfs bool file interface doesn't allow for callbacks
2671 * when the bool state is switched. We can reuse that facility when
2674 static ssize_t read_enabled_file_bool(struct file *file,
2675 char __user *user_buf, size_t count, loff_t *ppos)
2679 if (!kprobes_all_disarmed)
2685 return simple_read_from_buffer(user_buf, count, ppos, buf, 2);
2688 static ssize_t write_enabled_file_bool(struct file *file,
2689 const char __user *user_buf, size_t count, loff_t *ppos)
2695 buf_size = min(count, (sizeof(buf)-1));
2696 if (copy_from_user(buf, user_buf, buf_size))
2699 buf[buf_size] = '\0';
2704 ret = arm_all_kprobes();
2709 ret = disarm_all_kprobes();
2721 static const struct file_operations fops_kp = {
2722 .read = read_enabled_file_bool,
2723 .write = write_enabled_file_bool,
2724 .llseek = default_llseek,
2727 static int __init debugfs_kprobe_init(void)
2731 dir = debugfs_create_dir("kprobes", NULL);
2733 debugfs_create_file("list", 0400, dir, NULL,
2734 &debugfs_kprobes_operations);
2736 debugfs_create_file("enabled", 0600, dir, NULL, &fops_kp);
2738 debugfs_create_file("blacklist", 0400, dir, NULL,
2739 &debugfs_kprobe_blacklist_ops);
2744 late_initcall(debugfs_kprobe_init);
2745 #endif /* CONFIG_DEBUG_FS */