1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * Kernel Probes (KProbes)
6 * Copyright (C) IBM Corporation, 2002, 2004
8 * 2002-Oct Created by Vamsi Krishna S <vamsi_krishna@in.ibm.com> Kernel
9 * Probes initial implementation (includes suggestions from
11 * 2004-Aug Updated by Prasanna S Panchamukhi <prasanna@in.ibm.com> with
12 * hlists and exceptions notifier as suggested by Andi Kleen.
13 * 2004-July Suparna Bhattacharya <suparna@in.ibm.com> added jumper probes
14 * interface to access function arguments.
15 * 2004-Sep Prasanna S Panchamukhi <prasanna@in.ibm.com> Changed Kprobes
16 * exceptions notifier to be first on the priority list.
17 * 2005-May Hien Nguyen <hien@us.ibm.com>, Jim Keniston
18 * <jkenisto@us.ibm.com> and Prasanna S Panchamukhi
19 * <prasanna@in.ibm.com> added function-return probes.
21 #include <linux/kprobes.h>
22 #include <linux/hash.h>
23 #include <linux/init.h>
24 #include <linux/slab.h>
25 #include <linux/stddef.h>
26 #include <linux/export.h>
27 #include <linux/moduleloader.h>
28 #include <linux/kallsyms.h>
29 #include <linux/freezer.h>
30 #include <linux/seq_file.h>
31 #include <linux/debugfs.h>
32 #include <linux/sysctl.h>
33 #include <linux/kdebug.h>
34 #include <linux/memory.h>
35 #include <linux/ftrace.h>
36 #include <linux/cpu.h>
37 #include <linux/jump_label.h>
38 #include <linux/static_call.h>
39 #include <linux/perf_event.h>
41 #include <asm/sections.h>
42 #include <asm/cacheflush.h>
43 #include <asm/errno.h>
44 #include <linux/uaccess.h>
46 #define KPROBE_HASH_BITS 6
47 #define KPROBE_TABLE_SIZE (1 << KPROBE_HASH_BITS)
50 static int kprobes_initialized;
51 /* kprobe_table can be accessed by
52 * - Normal hlist traversal and RCU add/del under kprobe_mutex is held.
54 * - RCU hlist traversal under disabling preempt (breakpoint handlers)
56 static struct hlist_head kprobe_table[KPROBE_TABLE_SIZE];
58 /* NOTE: change this value only with kprobe_mutex held */
59 static bool kprobes_all_disarmed;
61 /* This protects kprobe_table and optimizing_list */
62 static DEFINE_MUTEX(kprobe_mutex);
63 static DEFINE_PER_CPU(struct kprobe *, kprobe_instance) = NULL;
65 kprobe_opcode_t * __weak kprobe_lookup_name(const char *name,
66 unsigned int __unused)
68 return ((kprobe_opcode_t *)(kallsyms_lookup_name(name)));
71 /* Blacklist -- list of struct kprobe_blacklist_entry */
72 static LIST_HEAD(kprobe_blacklist);
74 #ifdef __ARCH_WANT_KPROBES_INSN_SLOT
76 * kprobe->ainsn.insn points to the copy of the instruction to be
77 * single-stepped. x86_64, POWER4 and above have no-exec support and
78 * stepping on the instruction on a vmalloced/kmalloced/data page
79 * is a recipe for disaster
81 struct kprobe_insn_page {
82 struct list_head list;
83 kprobe_opcode_t *insns; /* Page of instruction slots */
84 struct kprobe_insn_cache *cache;
90 #define KPROBE_INSN_PAGE_SIZE(slots) \
91 (offsetof(struct kprobe_insn_page, slot_used) + \
92 (sizeof(char) * (slots)))
94 static int slots_per_page(struct kprobe_insn_cache *c)
96 return PAGE_SIZE/(c->insn_size * sizeof(kprobe_opcode_t));
99 enum kprobe_slot_state {
105 void __weak *alloc_insn_page(void)
107 return module_alloc(PAGE_SIZE);
110 static void free_insn_page(void *page)
112 module_memfree(page);
115 struct kprobe_insn_cache kprobe_insn_slots = {
116 .mutex = __MUTEX_INITIALIZER(kprobe_insn_slots.mutex),
117 .alloc = alloc_insn_page,
118 .free = free_insn_page,
119 .sym = KPROBE_INSN_PAGE_SYM,
120 .pages = LIST_HEAD_INIT(kprobe_insn_slots.pages),
121 .insn_size = MAX_INSN_SIZE,
124 static int collect_garbage_slots(struct kprobe_insn_cache *c);
127 * __get_insn_slot() - Find a slot on an executable page for an instruction.
128 * We allocate an executable page if there's no room on existing ones.
130 kprobe_opcode_t *__get_insn_slot(struct kprobe_insn_cache *c)
132 struct kprobe_insn_page *kip;
133 kprobe_opcode_t *slot = NULL;
135 /* Since the slot array is not protected by rcu, we need a mutex */
136 mutex_lock(&c->mutex);
139 list_for_each_entry_rcu(kip, &c->pages, list) {
140 if (kip->nused < slots_per_page(c)) {
142 for (i = 0; i < slots_per_page(c); i++) {
143 if (kip->slot_used[i] == SLOT_CLEAN) {
144 kip->slot_used[i] = SLOT_USED;
146 slot = kip->insns + (i * c->insn_size);
151 /* kip->nused is broken. Fix it. */
152 kip->nused = slots_per_page(c);
158 /* If there are any garbage slots, collect it and try again. */
159 if (c->nr_garbage && collect_garbage_slots(c) == 0)
162 /* All out of space. Need to allocate a new page. */
163 kip = kmalloc(KPROBE_INSN_PAGE_SIZE(slots_per_page(c)), GFP_KERNEL);
168 * Use module_alloc so this page is within +/- 2GB of where the
169 * kernel image and loaded module images reside. This is required
170 * so x86_64 can correctly handle the %rip-relative fixups.
172 kip->insns = c->alloc();
177 INIT_LIST_HEAD(&kip->list);
178 memset(kip->slot_used, SLOT_CLEAN, slots_per_page(c));
179 kip->slot_used[0] = SLOT_USED;
183 list_add_rcu(&kip->list, &c->pages);
186 /* Record the perf ksymbol register event after adding the page */
187 perf_event_ksymbol(PERF_RECORD_KSYMBOL_TYPE_OOL, (unsigned long)kip->insns,
188 PAGE_SIZE, false, c->sym);
190 mutex_unlock(&c->mutex);
194 /* Return 1 if all garbages are collected, otherwise 0. */
195 static int collect_one_slot(struct kprobe_insn_page *kip, int idx)
197 kip->slot_used[idx] = SLOT_CLEAN;
199 if (kip->nused == 0) {
201 * Page is no longer in use. Free it unless
202 * it's the last one. We keep the last one
203 * so as not to have to set it up again the
204 * next time somebody inserts a probe.
206 if (!list_is_singular(&kip->list)) {
208 * Record perf ksymbol unregister event before removing
211 perf_event_ksymbol(PERF_RECORD_KSYMBOL_TYPE_OOL,
212 (unsigned long)kip->insns, PAGE_SIZE, true,
214 list_del_rcu(&kip->list);
216 kip->cache->free(kip->insns);
224 static int collect_garbage_slots(struct kprobe_insn_cache *c)
226 struct kprobe_insn_page *kip, *next;
228 /* Ensure no-one is interrupted on the garbages */
231 list_for_each_entry_safe(kip, next, &c->pages, list) {
233 if (kip->ngarbage == 0)
235 kip->ngarbage = 0; /* we will collect all garbages */
236 for (i = 0; i < slots_per_page(c); i++) {
237 if (kip->slot_used[i] == SLOT_DIRTY && collect_one_slot(kip, i))
245 void __free_insn_slot(struct kprobe_insn_cache *c,
246 kprobe_opcode_t *slot, int dirty)
248 struct kprobe_insn_page *kip;
251 mutex_lock(&c->mutex);
253 list_for_each_entry_rcu(kip, &c->pages, list) {
254 idx = ((long)slot - (long)kip->insns) /
255 (c->insn_size * sizeof(kprobe_opcode_t));
256 if (idx >= 0 && idx < slots_per_page(c))
259 /* Could not find this slot. */
264 /* Mark and sweep: this may sleep */
266 /* Check double free */
267 WARN_ON(kip->slot_used[idx] != SLOT_USED);
269 kip->slot_used[idx] = SLOT_DIRTY;
271 if (++c->nr_garbage > slots_per_page(c))
272 collect_garbage_slots(c);
274 collect_one_slot(kip, idx);
277 mutex_unlock(&c->mutex);
281 * Check given address is on the page of kprobe instruction slots.
282 * This will be used for checking whether the address on a stack
283 * is on a text area or not.
285 bool __is_insn_slot_addr(struct kprobe_insn_cache *c, unsigned long addr)
287 struct kprobe_insn_page *kip;
291 list_for_each_entry_rcu(kip, &c->pages, list) {
292 if (addr >= (unsigned long)kip->insns &&
293 addr < (unsigned long)kip->insns + PAGE_SIZE) {
303 int kprobe_cache_get_kallsym(struct kprobe_insn_cache *c, unsigned int *symnum,
304 unsigned long *value, char *type, char *sym)
306 struct kprobe_insn_page *kip;
310 list_for_each_entry_rcu(kip, &c->pages, list) {
313 strlcpy(sym, c->sym, KSYM_NAME_LEN);
315 *value = (unsigned long)kip->insns;
324 #ifdef CONFIG_OPTPROBES
325 void __weak *alloc_optinsn_page(void)
327 return alloc_insn_page();
330 void __weak free_optinsn_page(void *page)
332 free_insn_page(page);
335 /* For optimized_kprobe buffer */
336 struct kprobe_insn_cache kprobe_optinsn_slots = {
337 .mutex = __MUTEX_INITIALIZER(kprobe_optinsn_slots.mutex),
338 .alloc = alloc_optinsn_page,
339 .free = free_optinsn_page,
340 .sym = KPROBE_OPTINSN_PAGE_SYM,
341 .pages = LIST_HEAD_INIT(kprobe_optinsn_slots.pages),
342 /* .insn_size is initialized later */
348 /* We have preemption disabled.. so it is safe to use __ versions */
349 static inline void set_kprobe_instance(struct kprobe *kp)
351 __this_cpu_write(kprobe_instance, kp);
354 static inline void reset_kprobe_instance(void)
356 __this_cpu_write(kprobe_instance, NULL);
360 * This routine is called either:
361 * - under the kprobe_mutex - during kprobe_[un]register()
363 * - with preemption disabled - from arch/xxx/kernel/kprobes.c
365 struct kprobe *get_kprobe(void *addr)
367 struct hlist_head *head;
370 head = &kprobe_table[hash_ptr(addr, KPROBE_HASH_BITS)];
371 hlist_for_each_entry_rcu(p, head, hlist,
372 lockdep_is_held(&kprobe_mutex)) {
379 NOKPROBE_SYMBOL(get_kprobe);
381 static int aggr_pre_handler(struct kprobe *p, struct pt_regs *regs);
383 /* Return true if the kprobe is an aggregator */
384 static inline int kprobe_aggrprobe(struct kprobe *p)
386 return p->pre_handler == aggr_pre_handler;
389 /* Return true(!0) if the kprobe is unused */
390 static inline int kprobe_unused(struct kprobe *p)
392 return kprobe_aggrprobe(p) && kprobe_disabled(p) &&
393 list_empty(&p->list);
397 * Keep all fields in the kprobe consistent
399 static inline void copy_kprobe(struct kprobe *ap, struct kprobe *p)
401 memcpy(&p->opcode, &ap->opcode, sizeof(kprobe_opcode_t));
402 memcpy(&p->ainsn, &ap->ainsn, sizeof(struct arch_specific_insn));
405 #ifdef CONFIG_OPTPROBES
406 /* NOTE: change this value only with kprobe_mutex held */
407 static bool kprobes_allow_optimization;
410 * Call all pre_handler on the list, but ignores its return value.
411 * This must be called from arch-dep optimized caller.
413 void opt_pre_handler(struct kprobe *p, struct pt_regs *regs)
417 list_for_each_entry_rcu(kp, &p->list, list) {
418 if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
419 set_kprobe_instance(kp);
420 kp->pre_handler(kp, regs);
422 reset_kprobe_instance();
425 NOKPROBE_SYMBOL(opt_pre_handler);
427 /* Free optimized instructions and optimized_kprobe */
428 static void free_aggr_kprobe(struct kprobe *p)
430 struct optimized_kprobe *op;
432 op = container_of(p, struct optimized_kprobe, kp);
433 arch_remove_optimized_kprobe(op);
434 arch_remove_kprobe(p);
438 /* Return true(!0) if the kprobe is ready for optimization. */
439 static inline int kprobe_optready(struct kprobe *p)
441 struct optimized_kprobe *op;
443 if (kprobe_aggrprobe(p)) {
444 op = container_of(p, struct optimized_kprobe, kp);
445 return arch_prepared_optinsn(&op->optinsn);
451 /* Return true(!0) if the kprobe is disarmed. Note: p must be on hash list */
452 static inline int kprobe_disarmed(struct kprobe *p)
454 struct optimized_kprobe *op;
456 /* If kprobe is not aggr/opt probe, just return kprobe is disabled */
457 if (!kprobe_aggrprobe(p))
458 return kprobe_disabled(p);
460 op = container_of(p, struct optimized_kprobe, kp);
462 return kprobe_disabled(p) && list_empty(&op->list);
465 /* Return true(!0) if the probe is queued on (un)optimizing lists */
466 static int kprobe_queued(struct kprobe *p)
468 struct optimized_kprobe *op;
470 if (kprobe_aggrprobe(p)) {
471 op = container_of(p, struct optimized_kprobe, kp);
472 if (!list_empty(&op->list))
479 * Return an optimized kprobe whose optimizing code replaces
480 * instructions including addr (exclude breakpoint).
482 static struct kprobe *get_optimized_kprobe(unsigned long addr)
485 struct kprobe *p = NULL;
486 struct optimized_kprobe *op;
488 /* Don't check i == 0, since that is a breakpoint case. */
489 for (i = 1; !p && i < MAX_OPTIMIZED_LENGTH; i++)
490 p = get_kprobe((void *)(addr - i));
492 if (p && kprobe_optready(p)) {
493 op = container_of(p, struct optimized_kprobe, kp);
494 if (arch_within_optimized_kprobe(op, addr))
501 /* Optimization staging list, protected by kprobe_mutex */
502 static LIST_HEAD(optimizing_list);
503 static LIST_HEAD(unoptimizing_list);
504 static LIST_HEAD(freeing_list);
506 static void kprobe_optimizer(struct work_struct *work);
507 static DECLARE_DELAYED_WORK(optimizing_work, kprobe_optimizer);
508 #define OPTIMIZE_DELAY 5
511 * Optimize (replace a breakpoint with a jump) kprobes listed on
514 static void do_optimize_kprobes(void)
516 lockdep_assert_held(&text_mutex);
518 * The optimization/unoptimization refers online_cpus via
519 * stop_machine() and cpu-hotplug modifies online_cpus.
520 * And same time, text_mutex will be held in cpu-hotplug and here.
521 * This combination can cause a deadlock (cpu-hotplug try to lock
522 * text_mutex but stop_machine can not be done because online_cpus
524 * To avoid this deadlock, caller must have locked cpu hotplug
525 * for preventing cpu-hotplug outside of text_mutex locking.
527 lockdep_assert_cpus_held();
529 /* Optimization never be done when disarmed */
530 if (kprobes_all_disarmed || !kprobes_allow_optimization ||
531 list_empty(&optimizing_list))
534 arch_optimize_kprobes(&optimizing_list);
538 * Unoptimize (replace a jump with a breakpoint and remove the breakpoint
539 * if need) kprobes listed on unoptimizing_list.
541 static void do_unoptimize_kprobes(void)
543 struct optimized_kprobe *op, *tmp;
545 lockdep_assert_held(&text_mutex);
546 /* See comment in do_optimize_kprobes() */
547 lockdep_assert_cpus_held();
549 /* Unoptimization must be done anytime */
550 if (list_empty(&unoptimizing_list))
553 arch_unoptimize_kprobes(&unoptimizing_list, &freeing_list);
554 /* Loop free_list for disarming */
555 list_for_each_entry_safe(op, tmp, &freeing_list, list) {
556 /* Switching from detour code to origin */
557 op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
558 /* Disarm probes if marked disabled */
559 if (kprobe_disabled(&op->kp))
560 arch_disarm_kprobe(&op->kp);
561 if (kprobe_unused(&op->kp)) {
563 * Remove unused probes from hash list. After waiting
564 * for synchronization, these probes are reclaimed.
565 * (reclaiming is done by do_free_cleaned_kprobes.)
567 hlist_del_rcu(&op->kp.hlist);
569 list_del_init(&op->list);
573 /* Reclaim all kprobes on the free_list */
574 static void do_free_cleaned_kprobes(void)
576 struct optimized_kprobe *op, *tmp;
578 list_for_each_entry_safe(op, tmp, &freeing_list, list) {
579 list_del_init(&op->list);
580 if (WARN_ON_ONCE(!kprobe_unused(&op->kp))) {
582 * This must not happen, but if there is a kprobe
583 * still in use, keep it on kprobes hash list.
587 free_aggr_kprobe(&op->kp);
591 /* Start optimizer after OPTIMIZE_DELAY passed */
592 static void kick_kprobe_optimizer(void)
594 schedule_delayed_work(&optimizing_work, OPTIMIZE_DELAY);
597 /* Kprobe jump optimizer */
598 static void kprobe_optimizer(struct work_struct *work)
600 mutex_lock(&kprobe_mutex);
602 mutex_lock(&text_mutex);
605 * Step 1: Unoptimize kprobes and collect cleaned (unused and disarmed)
606 * kprobes before waiting for quiesence period.
608 do_unoptimize_kprobes();
611 * Step 2: Wait for quiesence period to ensure all potentially
612 * preempted tasks to have normally scheduled. Because optprobe
613 * may modify multiple instructions, there is a chance that Nth
614 * instruction is preempted. In that case, such tasks can return
615 * to 2nd-Nth byte of jump instruction. This wait is for avoiding it.
616 * Note that on non-preemptive kernel, this is transparently converted
617 * to synchronoze_sched() to wait for all interrupts to have completed.
619 synchronize_rcu_tasks();
621 /* Step 3: Optimize kprobes after quiesence period */
622 do_optimize_kprobes();
624 /* Step 4: Free cleaned kprobes after quiesence period */
625 do_free_cleaned_kprobes();
627 mutex_unlock(&text_mutex);
630 /* Step 5: Kick optimizer again if needed */
631 if (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list))
632 kick_kprobe_optimizer();
634 mutex_unlock(&kprobe_mutex);
637 /* Wait for completing optimization and unoptimization */
638 void wait_for_kprobe_optimizer(void)
640 mutex_lock(&kprobe_mutex);
642 while (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list)) {
643 mutex_unlock(&kprobe_mutex);
645 /* this will also make optimizing_work execute immmediately */
646 flush_delayed_work(&optimizing_work);
647 /* @optimizing_work might not have been queued yet, relax */
650 mutex_lock(&kprobe_mutex);
653 mutex_unlock(&kprobe_mutex);
656 static bool optprobe_queued_unopt(struct optimized_kprobe *op)
658 struct optimized_kprobe *_op;
660 list_for_each_entry(_op, &unoptimizing_list, list) {
668 /* Optimize kprobe if p is ready to be optimized */
669 static void optimize_kprobe(struct kprobe *p)
671 struct optimized_kprobe *op;
673 /* Check if the kprobe is disabled or not ready for optimization. */
674 if (!kprobe_optready(p) || !kprobes_allow_optimization ||
675 (kprobe_disabled(p) || kprobes_all_disarmed))
678 /* kprobes with post_handler can not be optimized */
682 op = container_of(p, struct optimized_kprobe, kp);
684 /* Check there is no other kprobes at the optimized instructions */
685 if (arch_check_optimized_kprobe(op) < 0)
688 /* Check if it is already optimized. */
689 if (op->kp.flags & KPROBE_FLAG_OPTIMIZED) {
690 if (optprobe_queued_unopt(op)) {
691 /* This is under unoptimizing. Just dequeue the probe */
692 list_del_init(&op->list);
696 op->kp.flags |= KPROBE_FLAG_OPTIMIZED;
698 /* On unoptimizing/optimizing_list, op must have OPTIMIZED flag */
699 if (WARN_ON_ONCE(!list_empty(&op->list)))
702 list_add(&op->list, &optimizing_list);
703 kick_kprobe_optimizer();
706 /* Short cut to direct unoptimizing */
707 static void force_unoptimize_kprobe(struct optimized_kprobe *op)
709 lockdep_assert_cpus_held();
710 arch_unoptimize_kprobe(op);
711 op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
714 /* Unoptimize a kprobe if p is optimized */
715 static void unoptimize_kprobe(struct kprobe *p, bool force)
717 struct optimized_kprobe *op;
719 if (!kprobe_aggrprobe(p) || kprobe_disarmed(p))
720 return; /* This is not an optprobe nor optimized */
722 op = container_of(p, struct optimized_kprobe, kp);
723 if (!kprobe_optimized(p))
726 if (!list_empty(&op->list)) {
727 if (optprobe_queued_unopt(op)) {
728 /* Queued in unoptimizing queue */
731 * Forcibly unoptimize the kprobe here, and queue it
732 * in the freeing list for release afterwards.
734 force_unoptimize_kprobe(op);
735 list_move(&op->list, &freeing_list);
738 /* Dequeue from the optimizing queue */
739 list_del_init(&op->list);
740 op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
745 /* Optimized kprobe case */
747 /* Forcibly update the code: this is a special case */
748 force_unoptimize_kprobe(op);
750 list_add(&op->list, &unoptimizing_list);
751 kick_kprobe_optimizer();
755 /* Cancel unoptimizing for reusing */
756 static int reuse_unused_kprobe(struct kprobe *ap)
758 struct optimized_kprobe *op;
761 * Unused kprobe MUST be on the way of delayed unoptimizing (means
762 * there is still a relative jump) and disabled.
764 op = container_of(ap, struct optimized_kprobe, kp);
765 WARN_ON_ONCE(list_empty(&op->list));
766 /* Enable the probe again */
767 ap->flags &= ~KPROBE_FLAG_DISABLED;
768 /* Optimize it again (remove from op->list) */
769 if (!kprobe_optready(ap))
776 /* Remove optimized instructions */
777 static void kill_optimized_kprobe(struct kprobe *p)
779 struct optimized_kprobe *op;
781 op = container_of(p, struct optimized_kprobe, kp);
782 if (!list_empty(&op->list))
783 /* Dequeue from the (un)optimization queue */
784 list_del_init(&op->list);
785 op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
787 if (kprobe_unused(p)) {
788 /* Enqueue if it is unused */
789 list_add(&op->list, &freeing_list);
791 * Remove unused probes from the hash list. After waiting
792 * for synchronization, this probe is reclaimed.
793 * (reclaiming is done by do_free_cleaned_kprobes().)
795 hlist_del_rcu(&op->kp.hlist);
798 /* Don't touch the code, because it is already freed. */
799 arch_remove_optimized_kprobe(op);
803 void __prepare_optimized_kprobe(struct optimized_kprobe *op, struct kprobe *p)
805 if (!kprobe_ftrace(p))
806 arch_prepare_optimized_kprobe(op, p);
809 /* Try to prepare optimized instructions */
810 static void prepare_optimized_kprobe(struct kprobe *p)
812 struct optimized_kprobe *op;
814 op = container_of(p, struct optimized_kprobe, kp);
815 __prepare_optimized_kprobe(op, p);
818 /* Allocate new optimized_kprobe and try to prepare optimized instructions */
819 static struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
821 struct optimized_kprobe *op;
823 op = kzalloc(sizeof(struct optimized_kprobe), GFP_KERNEL);
827 INIT_LIST_HEAD(&op->list);
828 op->kp.addr = p->addr;
829 __prepare_optimized_kprobe(op, p);
834 static void init_aggr_kprobe(struct kprobe *ap, struct kprobe *p);
837 * Prepare an optimized_kprobe and optimize it
838 * NOTE: p must be a normal registered kprobe
840 static void try_to_optimize_kprobe(struct kprobe *p)
843 struct optimized_kprobe *op;
845 /* Impossible to optimize ftrace-based kprobe */
846 if (kprobe_ftrace(p))
849 /* For preparing optimization, jump_label_text_reserved() is called */
852 mutex_lock(&text_mutex);
854 ap = alloc_aggr_kprobe(p);
858 op = container_of(ap, struct optimized_kprobe, kp);
859 if (!arch_prepared_optinsn(&op->optinsn)) {
860 /* If failed to setup optimizing, fallback to kprobe */
861 arch_remove_optimized_kprobe(op);
866 init_aggr_kprobe(ap, p);
867 optimize_kprobe(ap); /* This just kicks optimizer thread */
870 mutex_unlock(&text_mutex);
875 static void optimize_all_kprobes(void)
877 struct hlist_head *head;
881 mutex_lock(&kprobe_mutex);
882 /* If optimization is already allowed, just return */
883 if (kprobes_allow_optimization)
887 kprobes_allow_optimization = true;
888 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
889 head = &kprobe_table[i];
890 hlist_for_each_entry(p, head, hlist)
891 if (!kprobe_disabled(p))
895 printk(KERN_INFO "Kprobes globally optimized\n");
897 mutex_unlock(&kprobe_mutex);
901 static void unoptimize_all_kprobes(void)
903 struct hlist_head *head;
907 mutex_lock(&kprobe_mutex);
908 /* If optimization is already prohibited, just return */
909 if (!kprobes_allow_optimization) {
910 mutex_unlock(&kprobe_mutex);
915 kprobes_allow_optimization = false;
916 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
917 head = &kprobe_table[i];
918 hlist_for_each_entry(p, head, hlist) {
919 if (!kprobe_disabled(p))
920 unoptimize_kprobe(p, false);
924 mutex_unlock(&kprobe_mutex);
926 /* Wait for unoptimizing completion */
927 wait_for_kprobe_optimizer();
928 printk(KERN_INFO "Kprobes globally unoptimized\n");
931 static DEFINE_MUTEX(kprobe_sysctl_mutex);
932 int sysctl_kprobes_optimization;
933 int proc_kprobes_optimization_handler(struct ctl_table *table, int write,
934 void *buffer, size_t *length,
939 mutex_lock(&kprobe_sysctl_mutex);
940 sysctl_kprobes_optimization = kprobes_allow_optimization ? 1 : 0;
941 ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
943 if (sysctl_kprobes_optimization)
944 optimize_all_kprobes();
946 unoptimize_all_kprobes();
947 mutex_unlock(&kprobe_sysctl_mutex);
951 #endif /* CONFIG_SYSCTL */
953 /* Put a breakpoint for a probe. Must be called with text_mutex locked */
954 static void __arm_kprobe(struct kprobe *p)
958 /* Check collision with other optimized kprobes */
959 _p = get_optimized_kprobe((unsigned long)p->addr);
961 /* Fallback to unoptimized kprobe */
962 unoptimize_kprobe(_p, true);
965 optimize_kprobe(p); /* Try to optimize (add kprobe to a list) */
968 /* Remove the breakpoint of a probe. Must be called with text_mutex locked */
969 static void __disarm_kprobe(struct kprobe *p, bool reopt)
973 /* Try to unoptimize */
974 unoptimize_kprobe(p, kprobes_all_disarmed);
976 if (!kprobe_queued(p)) {
977 arch_disarm_kprobe(p);
978 /* If another kprobe was blocked, optimize it. */
979 _p = get_optimized_kprobe((unsigned long)p->addr);
980 if (unlikely(_p) && reopt)
983 /* TODO: reoptimize others after unoptimized this probe */
986 #else /* !CONFIG_OPTPROBES */
988 #define optimize_kprobe(p) do {} while (0)
989 #define unoptimize_kprobe(p, f) do {} while (0)
990 #define kill_optimized_kprobe(p) do {} while (0)
991 #define prepare_optimized_kprobe(p) do {} while (0)
992 #define try_to_optimize_kprobe(p) do {} while (0)
993 #define __arm_kprobe(p) arch_arm_kprobe(p)
994 #define __disarm_kprobe(p, o) arch_disarm_kprobe(p)
995 #define kprobe_disarmed(p) kprobe_disabled(p)
996 #define wait_for_kprobe_optimizer() do {} while (0)
998 static int reuse_unused_kprobe(struct kprobe *ap)
1001 * If the optimized kprobe is NOT supported, the aggr kprobe is
1002 * released at the same time that the last aggregated kprobe is
1004 * Thus there should be no chance to reuse unused kprobe.
1006 printk(KERN_ERR "Error: There should be no unused kprobe here.\n");
1010 static void free_aggr_kprobe(struct kprobe *p)
1012 arch_remove_kprobe(p);
1016 static struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
1018 return kzalloc(sizeof(struct kprobe), GFP_KERNEL);
1020 #endif /* CONFIG_OPTPROBES */
1022 #ifdef CONFIG_KPROBES_ON_FTRACE
1023 static struct ftrace_ops kprobe_ftrace_ops __read_mostly = {
1024 .func = kprobe_ftrace_handler,
1025 .flags = FTRACE_OPS_FL_SAVE_REGS,
1028 static struct ftrace_ops kprobe_ipmodify_ops __read_mostly = {
1029 .func = kprobe_ftrace_handler,
1030 .flags = FTRACE_OPS_FL_SAVE_REGS | FTRACE_OPS_FL_IPMODIFY,
1033 static int kprobe_ipmodify_enabled;
1034 static int kprobe_ftrace_enabled;
1036 /* Caller must lock kprobe_mutex */
1037 static int __arm_kprobe_ftrace(struct kprobe *p, struct ftrace_ops *ops,
1042 ret = ftrace_set_filter_ip(ops, (unsigned long)p->addr, 0, 0);
1044 pr_debug("Failed to arm kprobe-ftrace at %pS (%d)\n",
1050 ret = register_ftrace_function(ops);
1052 pr_debug("Failed to init kprobe-ftrace (%d)\n", ret);
1062 * At this point, sinec ops is not registered, we should be sefe from
1063 * registering empty filter.
1065 ftrace_set_filter_ip(ops, (unsigned long)p->addr, 1, 0);
1069 static int arm_kprobe_ftrace(struct kprobe *p)
1071 bool ipmodify = (p->post_handler != NULL);
1073 return __arm_kprobe_ftrace(p,
1074 ipmodify ? &kprobe_ipmodify_ops : &kprobe_ftrace_ops,
1075 ipmodify ? &kprobe_ipmodify_enabled : &kprobe_ftrace_enabled);
1078 /* Caller must lock kprobe_mutex */
1079 static int __disarm_kprobe_ftrace(struct kprobe *p, struct ftrace_ops *ops,
1085 ret = unregister_ftrace_function(ops);
1086 if (WARN(ret < 0, "Failed to unregister kprobe-ftrace (%d)\n", ret))
1092 ret = ftrace_set_filter_ip(ops, (unsigned long)p->addr, 1, 0);
1093 WARN_ONCE(ret < 0, "Failed to disarm kprobe-ftrace at %pS (%d)\n",
1098 static int disarm_kprobe_ftrace(struct kprobe *p)
1100 bool ipmodify = (p->post_handler != NULL);
1102 return __disarm_kprobe_ftrace(p,
1103 ipmodify ? &kprobe_ipmodify_ops : &kprobe_ftrace_ops,
1104 ipmodify ? &kprobe_ipmodify_enabled : &kprobe_ftrace_enabled);
1106 #else /* !CONFIG_KPROBES_ON_FTRACE */
1107 static inline int arm_kprobe_ftrace(struct kprobe *p)
1112 static inline int disarm_kprobe_ftrace(struct kprobe *p)
1118 static int prepare_kprobe(struct kprobe *p)
1120 /* Must ensure p->addr is really on ftrace */
1121 if (kprobe_ftrace(p))
1122 return arch_prepare_kprobe_ftrace(p);
1124 return arch_prepare_kprobe(p);
1127 /* Arm a kprobe with text_mutex */
1128 static int arm_kprobe(struct kprobe *kp)
1130 if (unlikely(kprobe_ftrace(kp)))
1131 return arm_kprobe_ftrace(kp);
1134 mutex_lock(&text_mutex);
1136 mutex_unlock(&text_mutex);
1142 /* Disarm a kprobe with text_mutex */
1143 static int disarm_kprobe(struct kprobe *kp, bool reopt)
1145 if (unlikely(kprobe_ftrace(kp)))
1146 return disarm_kprobe_ftrace(kp);
1149 mutex_lock(&text_mutex);
1150 __disarm_kprobe(kp, reopt);
1151 mutex_unlock(&text_mutex);
1158 * Aggregate handlers for multiple kprobes support - these handlers
1159 * take care of invoking the individual kprobe handlers on p->list
1161 static int aggr_pre_handler(struct kprobe *p, struct pt_regs *regs)
1165 list_for_each_entry_rcu(kp, &p->list, list) {
1166 if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
1167 set_kprobe_instance(kp);
1168 if (kp->pre_handler(kp, regs))
1171 reset_kprobe_instance();
1175 NOKPROBE_SYMBOL(aggr_pre_handler);
1177 static void aggr_post_handler(struct kprobe *p, struct pt_regs *regs,
1178 unsigned long flags)
1182 list_for_each_entry_rcu(kp, &p->list, list) {
1183 if (kp->post_handler && likely(!kprobe_disabled(kp))) {
1184 set_kprobe_instance(kp);
1185 kp->post_handler(kp, regs, flags);
1186 reset_kprobe_instance();
1190 NOKPROBE_SYMBOL(aggr_post_handler);
1192 /* Walks the list and increments nmissed count for multiprobe case */
1193 void kprobes_inc_nmissed_count(struct kprobe *p)
1196 if (!kprobe_aggrprobe(p)) {
1199 list_for_each_entry_rcu(kp, &p->list, list)
1204 NOKPROBE_SYMBOL(kprobes_inc_nmissed_count);
1206 static void free_rp_inst_rcu(struct rcu_head *head)
1208 struct kretprobe_instance *ri = container_of(head, struct kretprobe_instance, rcu);
1210 if (refcount_dec_and_test(&ri->rph->ref))
1214 NOKPROBE_SYMBOL(free_rp_inst_rcu);
1216 static void recycle_rp_inst(struct kretprobe_instance *ri)
1218 struct kretprobe *rp = get_kretprobe(ri);
1221 freelist_add(&ri->freelist, &rp->freelist);
1223 call_rcu(&ri->rcu, free_rp_inst_rcu);
1225 NOKPROBE_SYMBOL(recycle_rp_inst);
1227 static struct kprobe kprobe_busy = {
1228 .addr = (void *) get_kprobe,
1231 void kprobe_busy_begin(void)
1233 struct kprobe_ctlblk *kcb;
1236 __this_cpu_write(current_kprobe, &kprobe_busy);
1237 kcb = get_kprobe_ctlblk();
1238 kcb->kprobe_status = KPROBE_HIT_ACTIVE;
1241 void kprobe_busy_end(void)
1243 __this_cpu_write(current_kprobe, NULL);
1248 * This function is called from finish_task_switch when task tk becomes dead,
1249 * so that we can recycle any function-return probe instances associated
1250 * with this task. These left over instances represent probed functions
1251 * that have been called but will never return.
1253 void kprobe_flush_task(struct task_struct *tk)
1255 struct kretprobe_instance *ri;
1256 struct llist_node *node;
1258 /* Early boot, not yet initialized. */
1259 if (unlikely(!kprobes_initialized))
1262 kprobe_busy_begin();
1264 node = __llist_del_all(&tk->kretprobe_instances);
1266 ri = container_of(node, struct kretprobe_instance, llist);
1269 recycle_rp_inst(ri);
1274 NOKPROBE_SYMBOL(kprobe_flush_task);
1276 static inline void free_rp_inst(struct kretprobe *rp)
1278 struct kretprobe_instance *ri;
1279 struct freelist_node *node;
1282 node = rp->freelist.head;
1284 ri = container_of(node, struct kretprobe_instance, freelist);
1291 if (refcount_sub_and_test(count, &rp->rph->ref)) {
1297 /* Add the new probe to ap->list */
1298 static int add_new_kprobe(struct kprobe *ap, struct kprobe *p)
1300 if (p->post_handler)
1301 unoptimize_kprobe(ap, true); /* Fall back to normal kprobe */
1303 list_add_rcu(&p->list, &ap->list);
1304 if (p->post_handler && !ap->post_handler)
1305 ap->post_handler = aggr_post_handler;
1311 * Fill in the required fields of the "manager kprobe". Replace the
1312 * earlier kprobe in the hlist with the manager kprobe
1314 static void init_aggr_kprobe(struct kprobe *ap, struct kprobe *p)
1316 /* Copy p's insn slot to ap */
1318 flush_insn_slot(ap);
1320 ap->flags = p->flags & ~KPROBE_FLAG_OPTIMIZED;
1321 ap->pre_handler = aggr_pre_handler;
1322 /* We don't care the kprobe which has gone. */
1323 if (p->post_handler && !kprobe_gone(p))
1324 ap->post_handler = aggr_post_handler;
1326 INIT_LIST_HEAD(&ap->list);
1327 INIT_HLIST_NODE(&ap->hlist);
1329 list_add_rcu(&p->list, &ap->list);
1330 hlist_replace_rcu(&p->hlist, &ap->hlist);
1334 * This is the second or subsequent kprobe at the address - handle
1337 static int register_aggr_kprobe(struct kprobe *orig_p, struct kprobe *p)
1340 struct kprobe *ap = orig_p;
1344 /* For preparing optimization, jump_label_text_reserved() is called */
1346 mutex_lock(&text_mutex);
1348 if (!kprobe_aggrprobe(orig_p)) {
1349 /* If orig_p is not an aggr_kprobe, create new aggr_kprobe. */
1350 ap = alloc_aggr_kprobe(orig_p);
1355 init_aggr_kprobe(ap, orig_p);
1356 } else if (kprobe_unused(ap)) {
1357 /* This probe is going to die. Rescue it */
1358 ret = reuse_unused_kprobe(ap);
1363 if (kprobe_gone(ap)) {
1365 * Attempting to insert new probe at the same location that
1366 * had a probe in the module vaddr area which already
1367 * freed. So, the instruction slot has already been
1368 * released. We need a new slot for the new probe.
1370 ret = arch_prepare_kprobe(ap);
1373 * Even if fail to allocate new slot, don't need to
1374 * free aggr_probe. It will be used next time, or
1375 * freed by unregister_kprobe.
1379 /* Prepare optimized instructions if possible. */
1380 prepare_optimized_kprobe(ap);
1383 * Clear gone flag to prevent allocating new slot again, and
1384 * set disabled flag because it is not armed yet.
1386 ap->flags = (ap->flags & ~KPROBE_FLAG_GONE)
1387 | KPROBE_FLAG_DISABLED;
1390 /* Copy ap's insn slot to p */
1392 ret = add_new_kprobe(ap, p);
1395 mutex_unlock(&text_mutex);
1396 jump_label_unlock();
1399 if (ret == 0 && kprobe_disabled(ap) && !kprobe_disabled(p)) {
1400 ap->flags &= ~KPROBE_FLAG_DISABLED;
1401 if (!kprobes_all_disarmed) {
1402 /* Arm the breakpoint again. */
1403 ret = arm_kprobe(ap);
1405 ap->flags |= KPROBE_FLAG_DISABLED;
1406 list_del_rcu(&p->list);
1414 bool __weak arch_within_kprobe_blacklist(unsigned long addr)
1416 /* The __kprobes marked functions and entry code must not be probed */
1417 return addr >= (unsigned long)__kprobes_text_start &&
1418 addr < (unsigned long)__kprobes_text_end;
1421 static bool __within_kprobe_blacklist(unsigned long addr)
1423 struct kprobe_blacklist_entry *ent;
1425 if (arch_within_kprobe_blacklist(addr))
1428 * If there exists a kprobe_blacklist, verify and
1429 * fail any probe registration in the prohibited area
1431 list_for_each_entry(ent, &kprobe_blacklist, list) {
1432 if (addr >= ent->start_addr && addr < ent->end_addr)
1438 bool within_kprobe_blacklist(unsigned long addr)
1440 char symname[KSYM_NAME_LEN], *p;
1442 if (__within_kprobe_blacklist(addr))
1445 /* Check if the address is on a suffixed-symbol */
1446 if (!lookup_symbol_name(addr, symname)) {
1447 p = strchr(symname, '.');
1451 addr = (unsigned long)kprobe_lookup_name(symname, 0);
1453 return __within_kprobe_blacklist(addr);
1459 * If we have a symbol_name argument, look it up and add the offset field
1460 * to it. This way, we can specify a relative address to a symbol.
1461 * This returns encoded errors if it fails to look up symbol or invalid
1462 * combination of parameters.
1464 static kprobe_opcode_t *_kprobe_addr(kprobe_opcode_t *addr,
1465 const char *symbol_name, unsigned int offset)
1467 if ((symbol_name && addr) || (!symbol_name && !addr))
1471 addr = kprobe_lookup_name(symbol_name, offset);
1473 return ERR_PTR(-ENOENT);
1476 addr = (kprobe_opcode_t *)(((char *)addr) + offset);
1481 return ERR_PTR(-EINVAL);
1484 static kprobe_opcode_t *kprobe_addr(struct kprobe *p)
1486 return _kprobe_addr(p->addr, p->symbol_name, p->offset);
1489 /* Check passed kprobe is valid and return kprobe in kprobe_table. */
1490 static struct kprobe *__get_valid_kprobe(struct kprobe *p)
1492 struct kprobe *ap, *list_p;
1494 lockdep_assert_held(&kprobe_mutex);
1496 ap = get_kprobe(p->addr);
1501 list_for_each_entry(list_p, &ap->list, list)
1503 /* kprobe p is a valid probe */
1512 * Warn and return error if the kprobe is being re-registered since
1513 * there must be a software bug.
1515 static inline int warn_kprobe_rereg(struct kprobe *p)
1519 mutex_lock(&kprobe_mutex);
1520 if (WARN_ON_ONCE(__get_valid_kprobe(p)))
1522 mutex_unlock(&kprobe_mutex);
1527 static int check_ftrace_location(struct kprobe *p)
1529 unsigned long ftrace_addr;
1531 ftrace_addr = ftrace_location((unsigned long)p->addr);
1533 #ifdef CONFIG_KPROBES_ON_FTRACE
1534 /* Given address is not on the instruction boundary */
1535 if ((unsigned long)p->addr != ftrace_addr)
1537 p->flags |= KPROBE_FLAG_FTRACE;
1538 #else /* !CONFIG_KPROBES_ON_FTRACE */
1545 static int check_kprobe_address_safe(struct kprobe *p,
1546 struct module **probed_mod)
1550 ret = check_ftrace_location(p);
1556 /* Ensure it is not in reserved area nor out of text */
1557 if (!kernel_text_address((unsigned long) p->addr) ||
1558 within_kprobe_blacklist((unsigned long) p->addr) ||
1559 jump_label_text_reserved(p->addr, p->addr) ||
1560 static_call_text_reserved(p->addr, p->addr) ||
1561 find_bug((unsigned long)p->addr)) {
1566 /* Check if are we probing a module */
1567 *probed_mod = __module_text_address((unsigned long) p->addr);
1570 * We must hold a refcount of the probed module while updating
1571 * its code to prohibit unexpected unloading.
1573 if (unlikely(!try_module_get(*probed_mod))) {
1579 * If the module freed .init.text, we couldn't insert
1582 if (within_module_init((unsigned long)p->addr, *probed_mod) &&
1583 (*probed_mod)->state != MODULE_STATE_COMING) {
1584 module_put(*probed_mod);
1591 jump_label_unlock();
1596 int register_kprobe(struct kprobe *p)
1599 struct kprobe *old_p;
1600 struct module *probed_mod;
1601 kprobe_opcode_t *addr;
1603 /* Adjust probe address from symbol */
1604 addr = kprobe_addr(p);
1606 return PTR_ERR(addr);
1609 ret = warn_kprobe_rereg(p);
1613 /* User can pass only KPROBE_FLAG_DISABLED to register_kprobe */
1614 p->flags &= KPROBE_FLAG_DISABLED;
1616 INIT_LIST_HEAD(&p->list);
1618 ret = check_kprobe_address_safe(p, &probed_mod);
1622 mutex_lock(&kprobe_mutex);
1624 old_p = get_kprobe(p->addr);
1626 /* Since this may unoptimize old_p, locking text_mutex. */
1627 ret = register_aggr_kprobe(old_p, p);
1632 /* Prevent text modification */
1633 mutex_lock(&text_mutex);
1634 ret = prepare_kprobe(p);
1635 mutex_unlock(&text_mutex);
1640 INIT_HLIST_NODE(&p->hlist);
1641 hlist_add_head_rcu(&p->hlist,
1642 &kprobe_table[hash_ptr(p->addr, KPROBE_HASH_BITS)]);
1644 if (!kprobes_all_disarmed && !kprobe_disabled(p)) {
1645 ret = arm_kprobe(p);
1647 hlist_del_rcu(&p->hlist);
1653 /* Try to optimize kprobe */
1654 try_to_optimize_kprobe(p);
1656 mutex_unlock(&kprobe_mutex);
1659 module_put(probed_mod);
1663 EXPORT_SYMBOL_GPL(register_kprobe);
1665 /* Check if all probes on the aggrprobe are disabled */
1666 static int aggr_kprobe_disabled(struct kprobe *ap)
1670 lockdep_assert_held(&kprobe_mutex);
1672 list_for_each_entry(kp, &ap->list, list)
1673 if (!kprobe_disabled(kp))
1675 * There is an active probe on the list.
1676 * We can't disable this ap.
1683 /* Disable one kprobe: Make sure called under kprobe_mutex is locked */
1684 static struct kprobe *__disable_kprobe(struct kprobe *p)
1686 struct kprobe *orig_p;
1689 /* Get an original kprobe for return */
1690 orig_p = __get_valid_kprobe(p);
1691 if (unlikely(orig_p == NULL))
1692 return ERR_PTR(-EINVAL);
1694 if (!kprobe_disabled(p)) {
1695 /* Disable probe if it is a child probe */
1697 p->flags |= KPROBE_FLAG_DISABLED;
1699 /* Try to disarm and disable this/parent probe */
1700 if (p == orig_p || aggr_kprobe_disabled(orig_p)) {
1702 * If kprobes_all_disarmed is set, orig_p
1703 * should have already been disarmed, so
1704 * skip unneed disarming process.
1706 if (!kprobes_all_disarmed) {
1707 ret = disarm_kprobe(orig_p, true);
1709 p->flags &= ~KPROBE_FLAG_DISABLED;
1710 return ERR_PTR(ret);
1713 orig_p->flags |= KPROBE_FLAG_DISABLED;
1721 * Unregister a kprobe without a scheduler synchronization.
1723 static int __unregister_kprobe_top(struct kprobe *p)
1725 struct kprobe *ap, *list_p;
1727 /* Disable kprobe. This will disarm it if needed. */
1728 ap = __disable_kprobe(p);
1734 * This probe is an independent(and non-optimized) kprobe
1735 * (not an aggrprobe). Remove from the hash list.
1739 /* Following process expects this probe is an aggrprobe */
1740 WARN_ON(!kprobe_aggrprobe(ap));
1742 if (list_is_singular(&ap->list) && kprobe_disarmed(ap))
1744 * !disarmed could be happen if the probe is under delayed
1749 /* If disabling probe has special handlers, update aggrprobe */
1750 if (p->post_handler && !kprobe_gone(p)) {
1751 list_for_each_entry(list_p, &ap->list, list) {
1752 if ((list_p != p) && (list_p->post_handler))
1755 ap->post_handler = NULL;
1759 * Remove from the aggrprobe: this path will do nothing in
1760 * __unregister_kprobe_bottom().
1762 list_del_rcu(&p->list);
1763 if (!kprobe_disabled(ap) && !kprobes_all_disarmed)
1765 * Try to optimize this probe again, because post
1766 * handler may have been changed.
1768 optimize_kprobe(ap);
1773 hlist_del_rcu(&ap->hlist);
1777 static void __unregister_kprobe_bottom(struct kprobe *p)
1781 if (list_empty(&p->list))
1782 /* This is an independent kprobe */
1783 arch_remove_kprobe(p);
1784 else if (list_is_singular(&p->list)) {
1785 /* This is the last child of an aggrprobe */
1786 ap = list_entry(p->list.next, struct kprobe, list);
1788 free_aggr_kprobe(ap);
1790 /* Otherwise, do nothing. */
1793 int register_kprobes(struct kprobe **kps, int num)
1799 for (i = 0; i < num; i++) {
1800 ret = register_kprobe(kps[i]);
1803 unregister_kprobes(kps, i);
1809 EXPORT_SYMBOL_GPL(register_kprobes);
1811 void unregister_kprobe(struct kprobe *p)
1813 unregister_kprobes(&p, 1);
1815 EXPORT_SYMBOL_GPL(unregister_kprobe);
1817 void unregister_kprobes(struct kprobe **kps, int num)
1823 mutex_lock(&kprobe_mutex);
1824 for (i = 0; i < num; i++)
1825 if (__unregister_kprobe_top(kps[i]) < 0)
1826 kps[i]->addr = NULL;
1827 mutex_unlock(&kprobe_mutex);
1830 for (i = 0; i < num; i++)
1832 __unregister_kprobe_bottom(kps[i]);
1834 EXPORT_SYMBOL_GPL(unregister_kprobes);
1836 int __weak kprobe_exceptions_notify(struct notifier_block *self,
1837 unsigned long val, void *data)
1841 NOKPROBE_SYMBOL(kprobe_exceptions_notify);
1843 static struct notifier_block kprobe_exceptions_nb = {
1844 .notifier_call = kprobe_exceptions_notify,
1845 .priority = 0x7fffffff /* we need to be notified first */
1848 unsigned long __weak arch_deref_entry_point(void *entry)
1850 return (unsigned long)entry;
1853 #ifdef CONFIG_KRETPROBES
1855 unsigned long __kretprobe_trampoline_handler(struct pt_regs *regs,
1856 void *trampoline_address,
1857 void *frame_pointer)
1859 kprobe_opcode_t *correct_ret_addr = NULL;
1860 struct kretprobe_instance *ri = NULL;
1861 struct llist_node *first, *node;
1862 struct kretprobe *rp;
1864 /* Find all nodes for this frame. */
1865 first = node = current->kretprobe_instances.first;
1867 ri = container_of(node, struct kretprobe_instance, llist);
1869 BUG_ON(ri->fp != frame_pointer);
1871 if (ri->ret_addr != trampoline_address) {
1872 correct_ret_addr = ri->ret_addr;
1874 * This is the real return address. Any other
1875 * instances associated with this task are for
1876 * other calls deeper on the call stack
1883 pr_err("Oops! Kretprobe fails to find correct return address.\n");
1887 /* Unlink all nodes for this frame. */
1888 current->kretprobe_instances.first = node->next;
1893 ri = container_of(first, struct kretprobe_instance, llist);
1894 first = first->next;
1896 rp = get_kretprobe(ri);
1897 if (rp && rp->handler) {
1898 struct kprobe *prev = kprobe_running();
1900 __this_cpu_write(current_kprobe, &rp->kp);
1901 ri->ret_addr = correct_ret_addr;
1902 rp->handler(ri, regs);
1903 __this_cpu_write(current_kprobe, prev);
1906 recycle_rp_inst(ri);
1909 return (unsigned long)correct_ret_addr;
1911 NOKPROBE_SYMBOL(__kretprobe_trampoline_handler)
1914 * This kprobe pre_handler is registered with every kretprobe. When probe
1915 * hits it will set up the return probe.
1917 static int pre_handler_kretprobe(struct kprobe *p, struct pt_regs *regs)
1919 struct kretprobe *rp = container_of(p, struct kretprobe, kp);
1920 struct kretprobe_instance *ri;
1921 struct freelist_node *fn;
1923 fn = freelist_try_get(&rp->freelist);
1929 ri = container_of(fn, struct kretprobe_instance, freelist);
1931 if (rp->entry_handler && rp->entry_handler(ri, regs)) {
1932 freelist_add(&ri->freelist, &rp->freelist);
1936 arch_prepare_kretprobe(ri, regs);
1938 __llist_add(&ri->llist, ¤t->kretprobe_instances);
1942 NOKPROBE_SYMBOL(pre_handler_kretprobe);
1944 bool __weak arch_kprobe_on_func_entry(unsigned long offset)
1950 * kprobe_on_func_entry() -- check whether given address is function entry
1951 * @addr: Target address
1952 * @sym: Target symbol name
1953 * @offset: The offset from the symbol or the address
1955 * This checks whether the given @addr+@offset or @sym+@offset is on the
1956 * function entry address or not.
1957 * This returns 0 if it is the function entry, or -EINVAL if it is not.
1958 * And also it returns -ENOENT if it fails the symbol or address lookup.
1959 * Caller must pass @addr or @sym (either one must be NULL), or this
1962 int kprobe_on_func_entry(kprobe_opcode_t *addr, const char *sym, unsigned long offset)
1964 kprobe_opcode_t *kp_addr = _kprobe_addr(addr, sym, offset);
1966 if (IS_ERR(kp_addr))
1967 return PTR_ERR(kp_addr);
1969 if (!kallsyms_lookup_size_offset((unsigned long)kp_addr, NULL, &offset))
1972 if (!arch_kprobe_on_func_entry(offset))
1978 int register_kretprobe(struct kretprobe *rp)
1981 struct kretprobe_instance *inst;
1985 ret = kprobe_on_func_entry(rp->kp.addr, rp->kp.symbol_name, rp->kp.offset);
1989 /* If only rp->kp.addr is specified, check reregistering kprobes */
1990 if (rp->kp.addr && warn_kprobe_rereg(&rp->kp))
1993 if (kretprobe_blacklist_size) {
1994 addr = kprobe_addr(&rp->kp);
1996 return PTR_ERR(addr);
1998 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
1999 if (kretprobe_blacklist[i].addr == addr)
2004 rp->kp.pre_handler = pre_handler_kretprobe;
2005 rp->kp.post_handler = NULL;
2007 /* Pre-allocate memory for max kretprobe instances */
2008 if (rp->maxactive <= 0) {
2009 #ifdef CONFIG_PREEMPTION
2010 rp->maxactive = max_t(unsigned int, 10, 2*num_possible_cpus());
2012 rp->maxactive = num_possible_cpus();
2015 rp->freelist.head = NULL;
2016 rp->rph = kzalloc(sizeof(struct kretprobe_holder), GFP_KERNEL);
2021 for (i = 0; i < rp->maxactive; i++) {
2022 inst = kzalloc(sizeof(struct kretprobe_instance) +
2023 rp->data_size, GFP_KERNEL);
2025 refcount_set(&rp->rph->ref, i);
2029 inst->rph = rp->rph;
2030 freelist_add(&inst->freelist, &rp->freelist);
2032 refcount_set(&rp->rph->ref, i);
2035 /* Establish function entry probe point */
2036 ret = register_kprobe(&rp->kp);
2041 EXPORT_SYMBOL_GPL(register_kretprobe);
2043 int register_kretprobes(struct kretprobe **rps, int num)
2049 for (i = 0; i < num; i++) {
2050 ret = register_kretprobe(rps[i]);
2053 unregister_kretprobes(rps, i);
2059 EXPORT_SYMBOL_GPL(register_kretprobes);
2061 void unregister_kretprobe(struct kretprobe *rp)
2063 unregister_kretprobes(&rp, 1);
2065 EXPORT_SYMBOL_GPL(unregister_kretprobe);
2067 void unregister_kretprobes(struct kretprobe **rps, int num)
2073 mutex_lock(&kprobe_mutex);
2074 for (i = 0; i < num; i++) {
2075 if (__unregister_kprobe_top(&rps[i]->kp) < 0)
2076 rps[i]->kp.addr = NULL;
2077 rps[i]->rph->rp = NULL;
2079 mutex_unlock(&kprobe_mutex);
2082 for (i = 0; i < num; i++) {
2083 if (rps[i]->kp.addr) {
2084 __unregister_kprobe_bottom(&rps[i]->kp);
2085 free_rp_inst(rps[i]);
2089 EXPORT_SYMBOL_GPL(unregister_kretprobes);
2091 #else /* CONFIG_KRETPROBES */
2092 int register_kretprobe(struct kretprobe *rp)
2096 EXPORT_SYMBOL_GPL(register_kretprobe);
2098 int register_kretprobes(struct kretprobe **rps, int num)
2102 EXPORT_SYMBOL_GPL(register_kretprobes);
2104 void unregister_kretprobe(struct kretprobe *rp)
2107 EXPORT_SYMBOL_GPL(unregister_kretprobe);
2109 void unregister_kretprobes(struct kretprobe **rps, int num)
2112 EXPORT_SYMBOL_GPL(unregister_kretprobes);
2114 static int pre_handler_kretprobe(struct kprobe *p, struct pt_regs *regs)
2118 NOKPROBE_SYMBOL(pre_handler_kretprobe);
2120 #endif /* CONFIG_KRETPROBES */
2122 /* Set the kprobe gone and remove its instruction buffer. */
2123 static void kill_kprobe(struct kprobe *p)
2127 lockdep_assert_held(&kprobe_mutex);
2129 p->flags |= KPROBE_FLAG_GONE;
2130 if (kprobe_aggrprobe(p)) {
2132 * If this is an aggr_kprobe, we have to list all the
2133 * chained probes and mark them GONE.
2135 list_for_each_entry(kp, &p->list, list)
2136 kp->flags |= KPROBE_FLAG_GONE;
2137 p->post_handler = NULL;
2138 kill_optimized_kprobe(p);
2141 * Here, we can remove insn_slot safely, because no thread calls
2142 * the original probed function (which will be freed soon) any more.
2144 arch_remove_kprobe(p);
2147 * The module is going away. We should disarm the kprobe which
2148 * is using ftrace, because ftrace framework is still available at
2149 * MODULE_STATE_GOING notification.
2151 if (kprobe_ftrace(p) && !kprobe_disabled(p) && !kprobes_all_disarmed)
2152 disarm_kprobe_ftrace(p);
2155 /* Disable one kprobe */
2156 int disable_kprobe(struct kprobe *kp)
2161 mutex_lock(&kprobe_mutex);
2163 /* Disable this kprobe */
2164 p = __disable_kprobe(kp);
2168 mutex_unlock(&kprobe_mutex);
2171 EXPORT_SYMBOL_GPL(disable_kprobe);
2173 /* Enable one kprobe */
2174 int enable_kprobe(struct kprobe *kp)
2179 mutex_lock(&kprobe_mutex);
2181 /* Check whether specified probe is valid. */
2182 p = __get_valid_kprobe(kp);
2183 if (unlikely(p == NULL)) {
2188 if (kprobe_gone(kp)) {
2189 /* This kprobe has gone, we couldn't enable it. */
2195 kp->flags &= ~KPROBE_FLAG_DISABLED;
2197 if (!kprobes_all_disarmed && kprobe_disabled(p)) {
2198 p->flags &= ~KPROBE_FLAG_DISABLED;
2199 ret = arm_kprobe(p);
2201 p->flags |= KPROBE_FLAG_DISABLED;
2204 mutex_unlock(&kprobe_mutex);
2207 EXPORT_SYMBOL_GPL(enable_kprobe);
2209 /* Caller must NOT call this in usual path. This is only for critical case */
2210 void dump_kprobe(struct kprobe *kp)
2212 pr_err("Dumping kprobe:\n");
2213 pr_err("Name: %s\nOffset: %x\nAddress: %pS\n",
2214 kp->symbol_name, kp->offset, kp->addr);
2216 NOKPROBE_SYMBOL(dump_kprobe);
2218 int kprobe_add_ksym_blacklist(unsigned long entry)
2220 struct kprobe_blacklist_entry *ent;
2221 unsigned long offset = 0, size = 0;
2223 if (!kernel_text_address(entry) ||
2224 !kallsyms_lookup_size_offset(entry, &size, &offset))
2227 ent = kmalloc(sizeof(*ent), GFP_KERNEL);
2230 ent->start_addr = entry;
2231 ent->end_addr = entry + size;
2232 INIT_LIST_HEAD(&ent->list);
2233 list_add_tail(&ent->list, &kprobe_blacklist);
2238 /* Add all symbols in given area into kprobe blacklist */
2239 int kprobe_add_area_blacklist(unsigned long start, unsigned long end)
2241 unsigned long entry;
2244 for (entry = start; entry < end; entry += ret) {
2245 ret = kprobe_add_ksym_blacklist(entry);
2248 if (ret == 0) /* In case of alias symbol */
2254 /* Remove all symbols in given area from kprobe blacklist */
2255 static void kprobe_remove_area_blacklist(unsigned long start, unsigned long end)
2257 struct kprobe_blacklist_entry *ent, *n;
2259 list_for_each_entry_safe(ent, n, &kprobe_blacklist, list) {
2260 if (ent->start_addr < start || ent->start_addr >= end)
2262 list_del(&ent->list);
2267 static void kprobe_remove_ksym_blacklist(unsigned long entry)
2269 kprobe_remove_area_blacklist(entry, entry + 1);
2272 int __weak arch_kprobe_get_kallsym(unsigned int *symnum, unsigned long *value,
2273 char *type, char *sym)
2278 int kprobe_get_kallsym(unsigned int symnum, unsigned long *value, char *type,
2281 #ifdef __ARCH_WANT_KPROBES_INSN_SLOT
2282 if (!kprobe_cache_get_kallsym(&kprobe_insn_slots, &symnum, value, type, sym))
2284 #ifdef CONFIG_OPTPROBES
2285 if (!kprobe_cache_get_kallsym(&kprobe_optinsn_slots, &symnum, value, type, sym))
2289 if (!arch_kprobe_get_kallsym(&symnum, value, type, sym))
2294 int __init __weak arch_populate_kprobe_blacklist(void)
2300 * Lookup and populate the kprobe_blacklist.
2302 * Unlike the kretprobe blacklist, we'll need to determine
2303 * the range of addresses that belong to the said functions,
2304 * since a kprobe need not necessarily be at the beginning
2307 static int __init populate_kprobe_blacklist(unsigned long *start,
2310 unsigned long entry;
2311 unsigned long *iter;
2314 for (iter = start; iter < end; iter++) {
2315 entry = arch_deref_entry_point((void *)*iter);
2316 ret = kprobe_add_ksym_blacklist(entry);
2323 /* Symbols in __kprobes_text are blacklisted */
2324 ret = kprobe_add_area_blacklist((unsigned long)__kprobes_text_start,
2325 (unsigned long)__kprobes_text_end);
2329 /* Symbols in noinstr section are blacklisted */
2330 ret = kprobe_add_area_blacklist((unsigned long)__noinstr_text_start,
2331 (unsigned long)__noinstr_text_end);
2333 return ret ? : arch_populate_kprobe_blacklist();
2336 static void add_module_kprobe_blacklist(struct module *mod)
2338 unsigned long start, end;
2341 if (mod->kprobe_blacklist) {
2342 for (i = 0; i < mod->num_kprobe_blacklist; i++)
2343 kprobe_add_ksym_blacklist(mod->kprobe_blacklist[i]);
2346 start = (unsigned long)mod->kprobes_text_start;
2348 end = start + mod->kprobes_text_size;
2349 kprobe_add_area_blacklist(start, end);
2352 start = (unsigned long)mod->noinstr_text_start;
2354 end = start + mod->noinstr_text_size;
2355 kprobe_add_area_blacklist(start, end);
2359 static void remove_module_kprobe_blacklist(struct module *mod)
2361 unsigned long start, end;
2364 if (mod->kprobe_blacklist) {
2365 for (i = 0; i < mod->num_kprobe_blacklist; i++)
2366 kprobe_remove_ksym_blacklist(mod->kprobe_blacklist[i]);
2369 start = (unsigned long)mod->kprobes_text_start;
2371 end = start + mod->kprobes_text_size;
2372 kprobe_remove_area_blacklist(start, end);
2375 start = (unsigned long)mod->noinstr_text_start;
2377 end = start + mod->noinstr_text_size;
2378 kprobe_remove_area_blacklist(start, end);
2382 /* Module notifier call back, checking kprobes on the module */
2383 static int kprobes_module_callback(struct notifier_block *nb,
2384 unsigned long val, void *data)
2386 struct module *mod = data;
2387 struct hlist_head *head;
2390 int checkcore = (val == MODULE_STATE_GOING);
2392 if (val == MODULE_STATE_COMING) {
2393 mutex_lock(&kprobe_mutex);
2394 add_module_kprobe_blacklist(mod);
2395 mutex_unlock(&kprobe_mutex);
2397 if (val != MODULE_STATE_GOING && val != MODULE_STATE_LIVE)
2401 * When MODULE_STATE_GOING was notified, both of module .text and
2402 * .init.text sections would be freed. When MODULE_STATE_LIVE was
2403 * notified, only .init.text section would be freed. We need to
2404 * disable kprobes which have been inserted in the sections.
2406 mutex_lock(&kprobe_mutex);
2407 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2408 head = &kprobe_table[i];
2409 hlist_for_each_entry(p, head, hlist)
2410 if (within_module_init((unsigned long)p->addr, mod) ||
2412 within_module_core((unsigned long)p->addr, mod))) {
2414 * The vaddr this probe is installed will soon
2415 * be vfreed buy not synced to disk. Hence,
2416 * disarming the breakpoint isn't needed.
2418 * Note, this will also move any optimized probes
2419 * that are pending to be removed from their
2420 * corresponding lists to the freeing_list and
2421 * will not be touched by the delayed
2422 * kprobe_optimizer work handler.
2427 if (val == MODULE_STATE_GOING)
2428 remove_module_kprobe_blacklist(mod);
2429 mutex_unlock(&kprobe_mutex);
2433 static struct notifier_block kprobe_module_nb = {
2434 .notifier_call = kprobes_module_callback,
2438 /* Markers of _kprobe_blacklist section */
2439 extern unsigned long __start_kprobe_blacklist[];
2440 extern unsigned long __stop_kprobe_blacklist[];
2442 void kprobe_free_init_mem(void)
2444 void *start = (void *)(&__init_begin);
2445 void *end = (void *)(&__init_end);
2446 struct hlist_head *head;
2450 mutex_lock(&kprobe_mutex);
2452 /* Kill all kprobes on initmem */
2453 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2454 head = &kprobe_table[i];
2455 hlist_for_each_entry(p, head, hlist) {
2456 if (start <= (void *)p->addr && (void *)p->addr < end)
2461 mutex_unlock(&kprobe_mutex);
2464 static int __init init_kprobes(void)
2468 /* FIXME allocate the probe table, currently defined statically */
2469 /* initialize all list heads */
2470 for (i = 0; i < KPROBE_TABLE_SIZE; i++)
2471 INIT_HLIST_HEAD(&kprobe_table[i]);
2473 err = populate_kprobe_blacklist(__start_kprobe_blacklist,
2474 __stop_kprobe_blacklist);
2476 pr_err("kprobes: failed to populate blacklist: %d\n", err);
2477 pr_err("Please take care of using kprobes.\n");
2480 if (kretprobe_blacklist_size) {
2481 /* lookup the function address from its name */
2482 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
2483 kretprobe_blacklist[i].addr =
2484 kprobe_lookup_name(kretprobe_blacklist[i].name, 0);
2485 if (!kretprobe_blacklist[i].addr)
2486 printk("kretprobe: lookup failed: %s\n",
2487 kretprobe_blacklist[i].name);
2491 /* By default, kprobes are armed */
2492 kprobes_all_disarmed = false;
2494 #if defined(CONFIG_OPTPROBES) && defined(__ARCH_WANT_KPROBES_INSN_SLOT)
2495 /* Init kprobe_optinsn_slots for allocation */
2496 kprobe_optinsn_slots.insn_size = MAX_OPTINSN_SIZE;
2499 err = arch_init_kprobes();
2501 err = register_die_notifier(&kprobe_exceptions_nb);
2503 err = register_module_notifier(&kprobe_module_nb);
2505 kprobes_initialized = (err == 0);
2511 early_initcall(init_kprobes);
2513 #if defined(CONFIG_OPTPROBES)
2514 static int __init init_optprobes(void)
2517 * Enable kprobe optimization - this kicks the optimizer which
2518 * depends on synchronize_rcu_tasks() and ksoftirqd, that is
2519 * not spawned in early initcall. So delay the optimization.
2521 optimize_all_kprobes();
2525 subsys_initcall(init_optprobes);
2528 #ifdef CONFIG_DEBUG_FS
2529 static void report_probe(struct seq_file *pi, struct kprobe *p,
2530 const char *sym, int offset, char *modname, struct kprobe *pp)
2533 void *addr = p->addr;
2535 if (p->pre_handler == pre_handler_kretprobe)
2540 if (!kallsyms_show_value(pi->file->f_cred))
2544 seq_printf(pi, "%px %s %s+0x%x %s ",
2545 addr, kprobe_type, sym, offset,
2546 (modname ? modname : " "));
2547 else /* try to use %pS */
2548 seq_printf(pi, "%px %s %pS ",
2549 addr, kprobe_type, p->addr);
2553 seq_printf(pi, "%s%s%s%s\n",
2554 (kprobe_gone(p) ? "[GONE]" : ""),
2555 ((kprobe_disabled(p) && !kprobe_gone(p)) ? "[DISABLED]" : ""),
2556 (kprobe_optimized(pp) ? "[OPTIMIZED]" : ""),
2557 (kprobe_ftrace(pp) ? "[FTRACE]" : ""));
2560 static void *kprobe_seq_start(struct seq_file *f, loff_t *pos)
2562 return (*pos < KPROBE_TABLE_SIZE) ? pos : NULL;
2565 static void *kprobe_seq_next(struct seq_file *f, void *v, loff_t *pos)
2568 if (*pos >= KPROBE_TABLE_SIZE)
2573 static void kprobe_seq_stop(struct seq_file *f, void *v)
2578 static int show_kprobe_addr(struct seq_file *pi, void *v)
2580 struct hlist_head *head;
2581 struct kprobe *p, *kp;
2582 const char *sym = NULL;
2583 unsigned int i = *(loff_t *) v;
2584 unsigned long offset = 0;
2585 char *modname, namebuf[KSYM_NAME_LEN];
2587 head = &kprobe_table[i];
2589 hlist_for_each_entry_rcu(p, head, hlist) {
2590 sym = kallsyms_lookup((unsigned long)p->addr, NULL,
2591 &offset, &modname, namebuf);
2592 if (kprobe_aggrprobe(p)) {
2593 list_for_each_entry_rcu(kp, &p->list, list)
2594 report_probe(pi, kp, sym, offset, modname, p);
2596 report_probe(pi, p, sym, offset, modname, NULL);
2602 static const struct seq_operations kprobes_sops = {
2603 .start = kprobe_seq_start,
2604 .next = kprobe_seq_next,
2605 .stop = kprobe_seq_stop,
2606 .show = show_kprobe_addr
2609 DEFINE_SEQ_ATTRIBUTE(kprobes);
2611 /* kprobes/blacklist -- shows which functions can not be probed */
2612 static void *kprobe_blacklist_seq_start(struct seq_file *m, loff_t *pos)
2614 mutex_lock(&kprobe_mutex);
2615 return seq_list_start(&kprobe_blacklist, *pos);
2618 static void *kprobe_blacklist_seq_next(struct seq_file *m, void *v, loff_t *pos)
2620 return seq_list_next(v, &kprobe_blacklist, pos);
2623 static int kprobe_blacklist_seq_show(struct seq_file *m, void *v)
2625 struct kprobe_blacklist_entry *ent =
2626 list_entry(v, struct kprobe_blacklist_entry, list);
2629 * If /proc/kallsyms is not showing kernel address, we won't
2630 * show them here either.
2632 if (!kallsyms_show_value(m->file->f_cred))
2633 seq_printf(m, "0x%px-0x%px\t%ps\n", NULL, NULL,
2634 (void *)ent->start_addr);
2636 seq_printf(m, "0x%px-0x%px\t%ps\n", (void *)ent->start_addr,
2637 (void *)ent->end_addr, (void *)ent->start_addr);
2641 static void kprobe_blacklist_seq_stop(struct seq_file *f, void *v)
2643 mutex_unlock(&kprobe_mutex);
2646 static const struct seq_operations kprobe_blacklist_sops = {
2647 .start = kprobe_blacklist_seq_start,
2648 .next = kprobe_blacklist_seq_next,
2649 .stop = kprobe_blacklist_seq_stop,
2650 .show = kprobe_blacklist_seq_show,
2652 DEFINE_SEQ_ATTRIBUTE(kprobe_blacklist);
2654 static int arm_all_kprobes(void)
2656 struct hlist_head *head;
2658 unsigned int i, total = 0, errors = 0;
2661 mutex_lock(&kprobe_mutex);
2663 /* If kprobes are armed, just return */
2664 if (!kprobes_all_disarmed)
2665 goto already_enabled;
2668 * optimize_kprobe() called by arm_kprobe() checks
2669 * kprobes_all_disarmed, so set kprobes_all_disarmed before
2672 kprobes_all_disarmed = false;
2673 /* Arming kprobes doesn't optimize kprobe itself */
2674 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2675 head = &kprobe_table[i];
2676 /* Arm all kprobes on a best-effort basis */
2677 hlist_for_each_entry(p, head, hlist) {
2678 if (!kprobe_disabled(p)) {
2679 err = arm_kprobe(p);
2690 pr_warn("Kprobes globally enabled, but failed to arm %d out of %d probes\n",
2693 pr_info("Kprobes globally enabled\n");
2696 mutex_unlock(&kprobe_mutex);
2700 static int disarm_all_kprobes(void)
2702 struct hlist_head *head;
2704 unsigned int i, total = 0, errors = 0;
2707 mutex_lock(&kprobe_mutex);
2709 /* If kprobes are already disarmed, just return */
2710 if (kprobes_all_disarmed) {
2711 mutex_unlock(&kprobe_mutex);
2715 kprobes_all_disarmed = true;
2717 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2718 head = &kprobe_table[i];
2719 /* Disarm all kprobes on a best-effort basis */
2720 hlist_for_each_entry(p, head, hlist) {
2721 if (!arch_trampoline_kprobe(p) && !kprobe_disabled(p)) {
2722 err = disarm_kprobe(p, false);
2733 pr_warn("Kprobes globally disabled, but failed to disarm %d out of %d probes\n",
2736 pr_info("Kprobes globally disabled\n");
2738 mutex_unlock(&kprobe_mutex);
2740 /* Wait for disarming all kprobes by optimizer */
2741 wait_for_kprobe_optimizer();
2747 * XXX: The debugfs bool file interface doesn't allow for callbacks
2748 * when the bool state is switched. We can reuse that facility when
2751 static ssize_t read_enabled_file_bool(struct file *file,
2752 char __user *user_buf, size_t count, loff_t *ppos)
2756 if (!kprobes_all_disarmed)
2762 return simple_read_from_buffer(user_buf, count, ppos, buf, 2);
2765 static ssize_t write_enabled_file_bool(struct file *file,
2766 const char __user *user_buf, size_t count, loff_t *ppos)
2771 ret = kstrtobool_from_user(user_buf, count, &enable);
2775 ret = enable ? arm_all_kprobes() : disarm_all_kprobes();
2782 static const struct file_operations fops_kp = {
2783 .read = read_enabled_file_bool,
2784 .write = write_enabled_file_bool,
2785 .llseek = default_llseek,
2788 static int __init debugfs_kprobe_init(void)
2792 dir = debugfs_create_dir("kprobes", NULL);
2794 debugfs_create_file("list", 0400, dir, NULL, &kprobes_fops);
2796 debugfs_create_file("enabled", 0600, dir, NULL, &fops_kp);
2798 debugfs_create_file("blacklist", 0400, dir, NULL,
2799 &kprobe_blacklist_fops);
2804 late_initcall(debugfs_kprobe_init);
2805 #endif /* CONFIG_DEBUG_FS */