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 void __weak 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 /* For optimized_kprobe buffer */
326 struct kprobe_insn_cache kprobe_optinsn_slots = {
327 .mutex = __MUTEX_INITIALIZER(kprobe_optinsn_slots.mutex),
328 .alloc = alloc_insn_page,
329 .free = free_insn_page,
330 .sym = KPROBE_OPTINSN_PAGE_SYM,
331 .pages = LIST_HEAD_INIT(kprobe_optinsn_slots.pages),
332 /* .insn_size is initialized later */
338 /* We have preemption disabled.. so it is safe to use __ versions */
339 static inline void set_kprobe_instance(struct kprobe *kp)
341 __this_cpu_write(kprobe_instance, kp);
344 static inline void reset_kprobe_instance(void)
346 __this_cpu_write(kprobe_instance, NULL);
350 * This routine is called either:
351 * - under the kprobe_mutex - during kprobe_[un]register()
353 * - with preemption disabled - from arch/xxx/kernel/kprobes.c
355 struct kprobe *get_kprobe(void *addr)
357 struct hlist_head *head;
360 head = &kprobe_table[hash_ptr(addr, KPROBE_HASH_BITS)];
361 hlist_for_each_entry_rcu(p, head, hlist,
362 lockdep_is_held(&kprobe_mutex)) {
369 NOKPROBE_SYMBOL(get_kprobe);
371 static int aggr_pre_handler(struct kprobe *p, struct pt_regs *regs);
373 /* Return true if the kprobe is an aggregator */
374 static inline int kprobe_aggrprobe(struct kprobe *p)
376 return p->pre_handler == aggr_pre_handler;
379 /* Return true(!0) if the kprobe is unused */
380 static inline int kprobe_unused(struct kprobe *p)
382 return kprobe_aggrprobe(p) && kprobe_disabled(p) &&
383 list_empty(&p->list);
387 * Keep all fields in the kprobe consistent
389 static inline void copy_kprobe(struct kprobe *ap, struct kprobe *p)
391 memcpy(&p->opcode, &ap->opcode, sizeof(kprobe_opcode_t));
392 memcpy(&p->ainsn, &ap->ainsn, sizeof(struct arch_specific_insn));
395 #ifdef CONFIG_OPTPROBES
396 /* NOTE: change this value only with kprobe_mutex held */
397 static bool kprobes_allow_optimization;
400 * Call all pre_handler on the list, but ignores its return value.
401 * This must be called from arch-dep optimized caller.
403 void opt_pre_handler(struct kprobe *p, struct pt_regs *regs)
407 list_for_each_entry_rcu(kp, &p->list, list) {
408 if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
409 set_kprobe_instance(kp);
410 kp->pre_handler(kp, regs);
412 reset_kprobe_instance();
415 NOKPROBE_SYMBOL(opt_pre_handler);
417 /* Free optimized instructions and optimized_kprobe */
418 static void free_aggr_kprobe(struct kprobe *p)
420 struct optimized_kprobe *op;
422 op = container_of(p, struct optimized_kprobe, kp);
423 arch_remove_optimized_kprobe(op);
424 arch_remove_kprobe(p);
428 /* Return true(!0) if the kprobe is ready for optimization. */
429 static inline int kprobe_optready(struct kprobe *p)
431 struct optimized_kprobe *op;
433 if (kprobe_aggrprobe(p)) {
434 op = container_of(p, struct optimized_kprobe, kp);
435 return arch_prepared_optinsn(&op->optinsn);
441 /* Return true(!0) if the kprobe is disarmed. Note: p must be on hash list */
442 static inline int kprobe_disarmed(struct kprobe *p)
444 struct optimized_kprobe *op;
446 /* If kprobe is not aggr/opt probe, just return kprobe is disabled */
447 if (!kprobe_aggrprobe(p))
448 return kprobe_disabled(p);
450 op = container_of(p, struct optimized_kprobe, kp);
452 return kprobe_disabled(p) && list_empty(&op->list);
455 /* Return true(!0) if the probe is queued on (un)optimizing lists */
456 static int kprobe_queued(struct kprobe *p)
458 struct optimized_kprobe *op;
460 if (kprobe_aggrprobe(p)) {
461 op = container_of(p, struct optimized_kprobe, kp);
462 if (!list_empty(&op->list))
469 * Return an optimized kprobe whose optimizing code replaces
470 * instructions including addr (exclude breakpoint).
472 static struct kprobe *get_optimized_kprobe(unsigned long addr)
475 struct kprobe *p = NULL;
476 struct optimized_kprobe *op;
478 /* Don't check i == 0, since that is a breakpoint case. */
479 for (i = 1; !p && i < MAX_OPTIMIZED_LENGTH; i++)
480 p = get_kprobe((void *)(addr - i));
482 if (p && kprobe_optready(p)) {
483 op = container_of(p, struct optimized_kprobe, kp);
484 if (arch_within_optimized_kprobe(op, addr))
491 /* Optimization staging list, protected by kprobe_mutex */
492 static LIST_HEAD(optimizing_list);
493 static LIST_HEAD(unoptimizing_list);
494 static LIST_HEAD(freeing_list);
496 static void kprobe_optimizer(struct work_struct *work);
497 static DECLARE_DELAYED_WORK(optimizing_work, kprobe_optimizer);
498 #define OPTIMIZE_DELAY 5
501 * Optimize (replace a breakpoint with a jump) kprobes listed on
504 static void do_optimize_kprobes(void)
506 lockdep_assert_held(&text_mutex);
508 * The optimization/unoptimization refers online_cpus via
509 * stop_machine() and cpu-hotplug modifies online_cpus.
510 * And same time, text_mutex will be held in cpu-hotplug and here.
511 * This combination can cause a deadlock (cpu-hotplug try to lock
512 * text_mutex but stop_machine can not be done because online_cpus
514 * To avoid this deadlock, caller must have locked cpu hotplug
515 * for preventing cpu-hotplug outside of text_mutex locking.
517 lockdep_assert_cpus_held();
519 /* Optimization never be done when disarmed */
520 if (kprobes_all_disarmed || !kprobes_allow_optimization ||
521 list_empty(&optimizing_list))
524 arch_optimize_kprobes(&optimizing_list);
528 * Unoptimize (replace a jump with a breakpoint and remove the breakpoint
529 * if need) kprobes listed on unoptimizing_list.
531 static void do_unoptimize_kprobes(void)
533 struct optimized_kprobe *op, *tmp;
535 lockdep_assert_held(&text_mutex);
536 /* See comment in do_optimize_kprobes() */
537 lockdep_assert_cpus_held();
539 /* Unoptimization must be done anytime */
540 if (list_empty(&unoptimizing_list))
543 arch_unoptimize_kprobes(&unoptimizing_list, &freeing_list);
544 /* Loop free_list for disarming */
545 list_for_each_entry_safe(op, tmp, &freeing_list, list) {
546 /* Switching from detour code to origin */
547 op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
548 /* Disarm probes if marked disabled */
549 if (kprobe_disabled(&op->kp))
550 arch_disarm_kprobe(&op->kp);
551 if (kprobe_unused(&op->kp)) {
553 * Remove unused probes from hash list. After waiting
554 * for synchronization, these probes are reclaimed.
555 * (reclaiming is done by do_free_cleaned_kprobes.)
557 hlist_del_rcu(&op->kp.hlist);
559 list_del_init(&op->list);
563 /* Reclaim all kprobes on the free_list */
564 static void do_free_cleaned_kprobes(void)
566 struct optimized_kprobe *op, *tmp;
568 list_for_each_entry_safe(op, tmp, &freeing_list, list) {
569 list_del_init(&op->list);
570 if (WARN_ON_ONCE(!kprobe_unused(&op->kp))) {
572 * This must not happen, but if there is a kprobe
573 * still in use, keep it on kprobes hash list.
577 free_aggr_kprobe(&op->kp);
581 /* Start optimizer after OPTIMIZE_DELAY passed */
582 static void kick_kprobe_optimizer(void)
584 schedule_delayed_work(&optimizing_work, OPTIMIZE_DELAY);
587 /* Kprobe jump optimizer */
588 static void kprobe_optimizer(struct work_struct *work)
590 mutex_lock(&kprobe_mutex);
592 mutex_lock(&text_mutex);
595 * Step 1: Unoptimize kprobes and collect cleaned (unused and disarmed)
596 * kprobes before waiting for quiesence period.
598 do_unoptimize_kprobes();
601 * Step 2: Wait for quiesence period to ensure all potentially
602 * preempted tasks to have normally scheduled. Because optprobe
603 * may modify multiple instructions, there is a chance that Nth
604 * instruction is preempted. In that case, such tasks can return
605 * to 2nd-Nth byte of jump instruction. This wait is for avoiding it.
606 * Note that on non-preemptive kernel, this is transparently converted
607 * to synchronoze_sched() to wait for all interrupts to have completed.
609 synchronize_rcu_tasks();
611 /* Step 3: Optimize kprobes after quiesence period */
612 do_optimize_kprobes();
614 /* Step 4: Free cleaned kprobes after quiesence period */
615 do_free_cleaned_kprobes();
617 mutex_unlock(&text_mutex);
620 /* Step 5: Kick optimizer again if needed */
621 if (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list))
622 kick_kprobe_optimizer();
624 mutex_unlock(&kprobe_mutex);
627 /* Wait for completing optimization and unoptimization */
628 void wait_for_kprobe_optimizer(void)
630 mutex_lock(&kprobe_mutex);
632 while (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list)) {
633 mutex_unlock(&kprobe_mutex);
635 /* this will also make optimizing_work execute immmediately */
636 flush_delayed_work(&optimizing_work);
637 /* @optimizing_work might not have been queued yet, relax */
640 mutex_lock(&kprobe_mutex);
643 mutex_unlock(&kprobe_mutex);
646 static bool optprobe_queued_unopt(struct optimized_kprobe *op)
648 struct optimized_kprobe *_op;
650 list_for_each_entry(_op, &unoptimizing_list, list) {
658 /* Optimize kprobe if p is ready to be optimized */
659 static void optimize_kprobe(struct kprobe *p)
661 struct optimized_kprobe *op;
663 /* Check if the kprobe is disabled or not ready for optimization. */
664 if (!kprobe_optready(p) || !kprobes_allow_optimization ||
665 (kprobe_disabled(p) || kprobes_all_disarmed))
668 /* kprobes with post_handler can not be optimized */
672 op = container_of(p, struct optimized_kprobe, kp);
674 /* Check there is no other kprobes at the optimized instructions */
675 if (arch_check_optimized_kprobe(op) < 0)
678 /* Check if it is already optimized. */
679 if (op->kp.flags & KPROBE_FLAG_OPTIMIZED) {
680 if (optprobe_queued_unopt(op)) {
681 /* This is under unoptimizing. Just dequeue the probe */
682 list_del_init(&op->list);
686 op->kp.flags |= KPROBE_FLAG_OPTIMIZED;
688 /* On unoptimizing/optimizing_list, op must have OPTIMIZED flag */
689 if (WARN_ON_ONCE(!list_empty(&op->list)))
692 list_add(&op->list, &optimizing_list);
693 kick_kprobe_optimizer();
696 /* Short cut to direct unoptimizing */
697 static void force_unoptimize_kprobe(struct optimized_kprobe *op)
699 lockdep_assert_cpus_held();
700 arch_unoptimize_kprobe(op);
701 op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
704 /* Unoptimize a kprobe if p is optimized */
705 static void unoptimize_kprobe(struct kprobe *p, bool force)
707 struct optimized_kprobe *op;
709 if (!kprobe_aggrprobe(p) || kprobe_disarmed(p))
710 return; /* This is not an optprobe nor optimized */
712 op = container_of(p, struct optimized_kprobe, kp);
713 if (!kprobe_optimized(p))
716 if (!list_empty(&op->list)) {
717 if (optprobe_queued_unopt(op)) {
718 /* Queued in unoptimizing queue */
721 * Forcibly unoptimize the kprobe here, and queue it
722 * in the freeing list for release afterwards.
724 force_unoptimize_kprobe(op);
725 list_move(&op->list, &freeing_list);
728 /* Dequeue from the optimizing queue */
729 list_del_init(&op->list);
730 op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
735 /* Optimized kprobe case */
737 /* Forcibly update the code: this is a special case */
738 force_unoptimize_kprobe(op);
740 list_add(&op->list, &unoptimizing_list);
741 kick_kprobe_optimizer();
745 /* Cancel unoptimizing for reusing */
746 static int reuse_unused_kprobe(struct kprobe *ap)
748 struct optimized_kprobe *op;
751 * Unused kprobe MUST be on the way of delayed unoptimizing (means
752 * there is still a relative jump) and disabled.
754 op = container_of(ap, struct optimized_kprobe, kp);
755 WARN_ON_ONCE(list_empty(&op->list));
756 /* Enable the probe again */
757 ap->flags &= ~KPROBE_FLAG_DISABLED;
758 /* Optimize it again (remove from op->list) */
759 if (!kprobe_optready(ap))
766 /* Remove optimized instructions */
767 static void kill_optimized_kprobe(struct kprobe *p)
769 struct optimized_kprobe *op;
771 op = container_of(p, struct optimized_kprobe, kp);
772 if (!list_empty(&op->list))
773 /* Dequeue from the (un)optimization queue */
774 list_del_init(&op->list);
775 op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
777 if (kprobe_unused(p)) {
778 /* Enqueue if it is unused */
779 list_add(&op->list, &freeing_list);
781 * Remove unused probes from the hash list. After waiting
782 * for synchronization, this probe is reclaimed.
783 * (reclaiming is done by do_free_cleaned_kprobes().)
785 hlist_del_rcu(&op->kp.hlist);
788 /* Don't touch the code, because it is already freed. */
789 arch_remove_optimized_kprobe(op);
793 void __prepare_optimized_kprobe(struct optimized_kprobe *op, struct kprobe *p)
795 if (!kprobe_ftrace(p))
796 arch_prepare_optimized_kprobe(op, p);
799 /* Try to prepare optimized instructions */
800 static void prepare_optimized_kprobe(struct kprobe *p)
802 struct optimized_kprobe *op;
804 op = container_of(p, struct optimized_kprobe, kp);
805 __prepare_optimized_kprobe(op, p);
808 /* Allocate new optimized_kprobe and try to prepare optimized instructions */
809 static struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
811 struct optimized_kprobe *op;
813 op = kzalloc(sizeof(struct optimized_kprobe), GFP_KERNEL);
817 INIT_LIST_HEAD(&op->list);
818 op->kp.addr = p->addr;
819 __prepare_optimized_kprobe(op, p);
824 static void init_aggr_kprobe(struct kprobe *ap, struct kprobe *p);
827 * Prepare an optimized_kprobe and optimize it
828 * NOTE: p must be a normal registered kprobe
830 static void try_to_optimize_kprobe(struct kprobe *p)
833 struct optimized_kprobe *op;
835 /* Impossible to optimize ftrace-based kprobe */
836 if (kprobe_ftrace(p))
839 /* For preparing optimization, jump_label_text_reserved() is called */
842 mutex_lock(&text_mutex);
844 ap = alloc_aggr_kprobe(p);
848 op = container_of(ap, struct optimized_kprobe, kp);
849 if (!arch_prepared_optinsn(&op->optinsn)) {
850 /* If failed to setup optimizing, fallback to kprobe */
851 arch_remove_optimized_kprobe(op);
856 init_aggr_kprobe(ap, p);
857 optimize_kprobe(ap); /* This just kicks optimizer thread */
860 mutex_unlock(&text_mutex);
865 static void optimize_all_kprobes(void)
867 struct hlist_head *head;
871 mutex_lock(&kprobe_mutex);
872 /* If optimization is already allowed, just return */
873 if (kprobes_allow_optimization)
877 kprobes_allow_optimization = true;
878 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
879 head = &kprobe_table[i];
880 hlist_for_each_entry(p, head, hlist)
881 if (!kprobe_disabled(p))
885 printk(KERN_INFO "Kprobes globally optimized\n");
887 mutex_unlock(&kprobe_mutex);
891 static void unoptimize_all_kprobes(void)
893 struct hlist_head *head;
897 mutex_lock(&kprobe_mutex);
898 /* If optimization is already prohibited, just return */
899 if (!kprobes_allow_optimization) {
900 mutex_unlock(&kprobe_mutex);
905 kprobes_allow_optimization = false;
906 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
907 head = &kprobe_table[i];
908 hlist_for_each_entry(p, head, hlist) {
909 if (!kprobe_disabled(p))
910 unoptimize_kprobe(p, false);
914 mutex_unlock(&kprobe_mutex);
916 /* Wait for unoptimizing completion */
917 wait_for_kprobe_optimizer();
918 printk(KERN_INFO "Kprobes globally unoptimized\n");
921 static DEFINE_MUTEX(kprobe_sysctl_mutex);
922 int sysctl_kprobes_optimization;
923 int proc_kprobes_optimization_handler(struct ctl_table *table, int write,
924 void *buffer, size_t *length,
929 mutex_lock(&kprobe_sysctl_mutex);
930 sysctl_kprobes_optimization = kprobes_allow_optimization ? 1 : 0;
931 ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
933 if (sysctl_kprobes_optimization)
934 optimize_all_kprobes();
936 unoptimize_all_kprobes();
937 mutex_unlock(&kprobe_sysctl_mutex);
941 #endif /* CONFIG_SYSCTL */
943 /* Put a breakpoint for a probe. Must be called with text_mutex locked */
944 static void __arm_kprobe(struct kprobe *p)
948 /* Check collision with other optimized kprobes */
949 _p = get_optimized_kprobe((unsigned long)p->addr);
951 /* Fallback to unoptimized kprobe */
952 unoptimize_kprobe(_p, true);
955 optimize_kprobe(p); /* Try to optimize (add kprobe to a list) */
958 /* Remove the breakpoint of a probe. Must be called with text_mutex locked */
959 static void __disarm_kprobe(struct kprobe *p, bool reopt)
963 /* Try to unoptimize */
964 unoptimize_kprobe(p, kprobes_all_disarmed);
966 if (!kprobe_queued(p)) {
967 arch_disarm_kprobe(p);
968 /* If another kprobe was blocked, optimize it. */
969 _p = get_optimized_kprobe((unsigned long)p->addr);
970 if (unlikely(_p) && reopt)
973 /* TODO: reoptimize others after unoptimized this probe */
976 #else /* !CONFIG_OPTPROBES */
978 #define optimize_kprobe(p) do {} while (0)
979 #define unoptimize_kprobe(p, f) do {} while (0)
980 #define kill_optimized_kprobe(p) do {} while (0)
981 #define prepare_optimized_kprobe(p) do {} while (0)
982 #define try_to_optimize_kprobe(p) do {} while (0)
983 #define __arm_kprobe(p) arch_arm_kprobe(p)
984 #define __disarm_kprobe(p, o) arch_disarm_kprobe(p)
985 #define kprobe_disarmed(p) kprobe_disabled(p)
986 #define wait_for_kprobe_optimizer() do {} while (0)
988 static int reuse_unused_kprobe(struct kprobe *ap)
991 * If the optimized kprobe is NOT supported, the aggr kprobe is
992 * released at the same time that the last aggregated kprobe is
994 * Thus there should be no chance to reuse unused kprobe.
996 printk(KERN_ERR "Error: There should be no unused kprobe here.\n");
1000 static void free_aggr_kprobe(struct kprobe *p)
1002 arch_remove_kprobe(p);
1006 static struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
1008 return kzalloc(sizeof(struct kprobe), GFP_KERNEL);
1010 #endif /* CONFIG_OPTPROBES */
1012 #ifdef CONFIG_KPROBES_ON_FTRACE
1013 static struct ftrace_ops kprobe_ftrace_ops __read_mostly = {
1014 .func = kprobe_ftrace_handler,
1015 .flags = FTRACE_OPS_FL_SAVE_REGS,
1018 static struct ftrace_ops kprobe_ipmodify_ops __read_mostly = {
1019 .func = kprobe_ftrace_handler,
1020 .flags = FTRACE_OPS_FL_SAVE_REGS | FTRACE_OPS_FL_IPMODIFY,
1023 static int kprobe_ipmodify_enabled;
1024 static int kprobe_ftrace_enabled;
1026 /* Must ensure p->addr is really on ftrace */
1027 static int prepare_kprobe(struct kprobe *p)
1029 if (!kprobe_ftrace(p))
1030 return arch_prepare_kprobe(p);
1032 return arch_prepare_kprobe_ftrace(p);
1035 /* Caller must lock kprobe_mutex */
1036 static int __arm_kprobe_ftrace(struct kprobe *p, struct ftrace_ops *ops,
1041 ret = ftrace_set_filter_ip(ops, (unsigned long)p->addr, 0, 0);
1043 pr_debug("Failed to arm kprobe-ftrace at %pS (%d)\n",
1049 ret = register_ftrace_function(ops);
1051 pr_debug("Failed to init kprobe-ftrace (%d)\n", ret);
1061 * At this point, sinec ops is not registered, we should be sefe from
1062 * registering empty filter.
1064 ftrace_set_filter_ip(ops, (unsigned long)p->addr, 1, 0);
1068 static int arm_kprobe_ftrace(struct kprobe *p)
1070 bool ipmodify = (p->post_handler != NULL);
1072 return __arm_kprobe_ftrace(p,
1073 ipmodify ? &kprobe_ipmodify_ops : &kprobe_ftrace_ops,
1074 ipmodify ? &kprobe_ipmodify_enabled : &kprobe_ftrace_enabled);
1077 /* Caller must lock kprobe_mutex */
1078 static int __disarm_kprobe_ftrace(struct kprobe *p, struct ftrace_ops *ops,
1084 ret = unregister_ftrace_function(ops);
1085 if (WARN(ret < 0, "Failed to unregister kprobe-ftrace (%d)\n", ret))
1091 ret = ftrace_set_filter_ip(ops, (unsigned long)p->addr, 1, 0);
1092 WARN_ONCE(ret < 0, "Failed to disarm kprobe-ftrace at %pS (%d)\n",
1097 static int disarm_kprobe_ftrace(struct kprobe *p)
1099 bool ipmodify = (p->post_handler != NULL);
1101 return __disarm_kprobe_ftrace(p,
1102 ipmodify ? &kprobe_ipmodify_ops : &kprobe_ftrace_ops,
1103 ipmodify ? &kprobe_ipmodify_enabled : &kprobe_ftrace_enabled);
1105 #else /* !CONFIG_KPROBES_ON_FTRACE */
1106 static inline int prepare_kprobe(struct kprobe *p)
1108 return arch_prepare_kprobe(p);
1111 static inline int arm_kprobe_ftrace(struct kprobe *p)
1116 static inline int disarm_kprobe_ftrace(struct kprobe *p)
1122 /* Arm a kprobe with text_mutex */
1123 static int arm_kprobe(struct kprobe *kp)
1125 if (unlikely(kprobe_ftrace(kp)))
1126 return arm_kprobe_ftrace(kp);
1129 mutex_lock(&text_mutex);
1131 mutex_unlock(&text_mutex);
1137 /* Disarm a kprobe with text_mutex */
1138 static int disarm_kprobe(struct kprobe *kp, bool reopt)
1140 if (unlikely(kprobe_ftrace(kp)))
1141 return disarm_kprobe_ftrace(kp);
1144 mutex_lock(&text_mutex);
1145 __disarm_kprobe(kp, reopt);
1146 mutex_unlock(&text_mutex);
1153 * Aggregate handlers for multiple kprobes support - these handlers
1154 * take care of invoking the individual kprobe handlers on p->list
1156 static int aggr_pre_handler(struct kprobe *p, struct pt_regs *regs)
1160 list_for_each_entry_rcu(kp, &p->list, list) {
1161 if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
1162 set_kprobe_instance(kp);
1163 if (kp->pre_handler(kp, regs))
1166 reset_kprobe_instance();
1170 NOKPROBE_SYMBOL(aggr_pre_handler);
1172 static void aggr_post_handler(struct kprobe *p, struct pt_regs *regs,
1173 unsigned long flags)
1177 list_for_each_entry_rcu(kp, &p->list, list) {
1178 if (kp->post_handler && likely(!kprobe_disabled(kp))) {
1179 set_kprobe_instance(kp);
1180 kp->post_handler(kp, regs, flags);
1181 reset_kprobe_instance();
1185 NOKPROBE_SYMBOL(aggr_post_handler);
1187 static int aggr_fault_handler(struct kprobe *p, struct pt_regs *regs,
1190 struct kprobe *cur = __this_cpu_read(kprobe_instance);
1193 * if we faulted "during" the execution of a user specified
1194 * probe handler, invoke just that probe's fault handler
1196 if (cur && cur->fault_handler) {
1197 if (cur->fault_handler(cur, regs, trapnr))
1202 NOKPROBE_SYMBOL(aggr_fault_handler);
1204 /* Walks the list and increments nmissed count for multiprobe case */
1205 void kprobes_inc_nmissed_count(struct kprobe *p)
1208 if (!kprobe_aggrprobe(p)) {
1211 list_for_each_entry_rcu(kp, &p->list, list)
1216 NOKPROBE_SYMBOL(kprobes_inc_nmissed_count);
1218 static void free_rp_inst_rcu(struct rcu_head *head)
1220 struct kretprobe_instance *ri = container_of(head, struct kretprobe_instance, rcu);
1222 if (refcount_dec_and_test(&ri->rph->ref))
1226 NOKPROBE_SYMBOL(free_rp_inst_rcu);
1228 static void recycle_rp_inst(struct kretprobe_instance *ri)
1230 struct kretprobe *rp = get_kretprobe(ri);
1233 freelist_add(&ri->freelist, &rp->freelist);
1235 call_rcu(&ri->rcu, free_rp_inst_rcu);
1237 NOKPROBE_SYMBOL(recycle_rp_inst);
1239 static struct kprobe kprobe_busy = {
1240 .addr = (void *) get_kprobe,
1243 void kprobe_busy_begin(void)
1245 struct kprobe_ctlblk *kcb;
1248 __this_cpu_write(current_kprobe, &kprobe_busy);
1249 kcb = get_kprobe_ctlblk();
1250 kcb->kprobe_status = KPROBE_HIT_ACTIVE;
1253 void kprobe_busy_end(void)
1255 __this_cpu_write(current_kprobe, NULL);
1260 * This function is called from finish_task_switch when task tk becomes dead,
1261 * so that we can recycle any function-return probe instances associated
1262 * with this task. These left over instances represent probed functions
1263 * that have been called but will never return.
1265 void kprobe_flush_task(struct task_struct *tk)
1267 struct kretprobe_instance *ri;
1268 struct llist_node *node;
1270 /* Early boot, not yet initialized. */
1271 if (unlikely(!kprobes_initialized))
1274 kprobe_busy_begin();
1276 node = __llist_del_all(&tk->kretprobe_instances);
1278 ri = container_of(node, struct kretprobe_instance, llist);
1281 recycle_rp_inst(ri);
1286 NOKPROBE_SYMBOL(kprobe_flush_task);
1288 static inline void free_rp_inst(struct kretprobe *rp)
1290 struct kretprobe_instance *ri;
1291 struct freelist_node *node;
1294 node = rp->freelist.head;
1296 ri = container_of(node, struct kretprobe_instance, freelist);
1303 if (refcount_sub_and_test(count, &rp->rph->ref)) {
1309 /* Add the new probe to ap->list */
1310 static int add_new_kprobe(struct kprobe *ap, struct kprobe *p)
1312 if (p->post_handler)
1313 unoptimize_kprobe(ap, true); /* Fall back to normal kprobe */
1315 list_add_rcu(&p->list, &ap->list);
1316 if (p->post_handler && !ap->post_handler)
1317 ap->post_handler = aggr_post_handler;
1323 * Fill in the required fields of the "manager kprobe". Replace the
1324 * earlier kprobe in the hlist with the manager kprobe
1326 static void init_aggr_kprobe(struct kprobe *ap, struct kprobe *p)
1328 /* Copy p's insn slot to ap */
1330 flush_insn_slot(ap);
1332 ap->flags = p->flags & ~KPROBE_FLAG_OPTIMIZED;
1333 ap->pre_handler = aggr_pre_handler;
1334 ap->fault_handler = aggr_fault_handler;
1335 /* We don't care the kprobe which has gone. */
1336 if (p->post_handler && !kprobe_gone(p))
1337 ap->post_handler = aggr_post_handler;
1339 INIT_LIST_HEAD(&ap->list);
1340 INIT_HLIST_NODE(&ap->hlist);
1342 list_add_rcu(&p->list, &ap->list);
1343 hlist_replace_rcu(&p->hlist, &ap->hlist);
1347 * This is the second or subsequent kprobe at the address - handle
1350 static int register_aggr_kprobe(struct kprobe *orig_p, struct kprobe *p)
1353 struct kprobe *ap = orig_p;
1357 /* For preparing optimization, jump_label_text_reserved() is called */
1359 mutex_lock(&text_mutex);
1361 if (!kprobe_aggrprobe(orig_p)) {
1362 /* If orig_p is not an aggr_kprobe, create new aggr_kprobe. */
1363 ap = alloc_aggr_kprobe(orig_p);
1368 init_aggr_kprobe(ap, orig_p);
1369 } else if (kprobe_unused(ap)) {
1370 /* This probe is going to die. Rescue it */
1371 ret = reuse_unused_kprobe(ap);
1376 if (kprobe_gone(ap)) {
1378 * Attempting to insert new probe at the same location that
1379 * had a probe in the module vaddr area which already
1380 * freed. So, the instruction slot has already been
1381 * released. We need a new slot for the new probe.
1383 ret = arch_prepare_kprobe(ap);
1386 * Even if fail to allocate new slot, don't need to
1387 * free aggr_probe. It will be used next time, or
1388 * freed by unregister_kprobe.
1392 /* Prepare optimized instructions if possible. */
1393 prepare_optimized_kprobe(ap);
1396 * Clear gone flag to prevent allocating new slot again, and
1397 * set disabled flag because it is not armed yet.
1399 ap->flags = (ap->flags & ~KPROBE_FLAG_GONE)
1400 | KPROBE_FLAG_DISABLED;
1403 /* Copy ap's insn slot to p */
1405 ret = add_new_kprobe(ap, p);
1408 mutex_unlock(&text_mutex);
1409 jump_label_unlock();
1412 if (ret == 0 && kprobe_disabled(ap) && !kprobe_disabled(p)) {
1413 ap->flags &= ~KPROBE_FLAG_DISABLED;
1414 if (!kprobes_all_disarmed) {
1415 /* Arm the breakpoint again. */
1416 ret = arm_kprobe(ap);
1418 ap->flags |= KPROBE_FLAG_DISABLED;
1419 list_del_rcu(&p->list);
1427 bool __weak arch_within_kprobe_blacklist(unsigned long addr)
1429 /* The __kprobes marked functions and entry code must not be probed */
1430 return addr >= (unsigned long)__kprobes_text_start &&
1431 addr < (unsigned long)__kprobes_text_end;
1434 static bool __within_kprobe_blacklist(unsigned long addr)
1436 struct kprobe_blacklist_entry *ent;
1438 if (arch_within_kprobe_blacklist(addr))
1441 * If there exists a kprobe_blacklist, verify and
1442 * fail any probe registration in the prohibited area
1444 list_for_each_entry(ent, &kprobe_blacklist, list) {
1445 if (addr >= ent->start_addr && addr < ent->end_addr)
1451 bool within_kprobe_blacklist(unsigned long addr)
1453 char symname[KSYM_NAME_LEN], *p;
1455 if (__within_kprobe_blacklist(addr))
1458 /* Check if the address is on a suffixed-symbol */
1459 if (!lookup_symbol_name(addr, symname)) {
1460 p = strchr(symname, '.');
1464 addr = (unsigned long)kprobe_lookup_name(symname, 0);
1466 return __within_kprobe_blacklist(addr);
1472 * If we have a symbol_name argument, look it up and add the offset field
1473 * to it. This way, we can specify a relative address to a symbol.
1474 * This returns encoded errors if it fails to look up symbol or invalid
1475 * combination of parameters.
1477 static kprobe_opcode_t *_kprobe_addr(kprobe_opcode_t *addr,
1478 const char *symbol_name, unsigned int offset)
1480 if ((symbol_name && addr) || (!symbol_name && !addr))
1484 addr = kprobe_lookup_name(symbol_name, offset);
1486 return ERR_PTR(-ENOENT);
1489 addr = (kprobe_opcode_t *)(((char *)addr) + offset);
1494 return ERR_PTR(-EINVAL);
1497 static kprobe_opcode_t *kprobe_addr(struct kprobe *p)
1499 return _kprobe_addr(p->addr, p->symbol_name, p->offset);
1502 /* Check passed kprobe is valid and return kprobe in kprobe_table. */
1503 static struct kprobe *__get_valid_kprobe(struct kprobe *p)
1505 struct kprobe *ap, *list_p;
1507 lockdep_assert_held(&kprobe_mutex);
1509 ap = get_kprobe(p->addr);
1514 list_for_each_entry(list_p, &ap->list, list)
1516 /* kprobe p is a valid probe */
1525 * Warn and return error if the kprobe is being re-registered since
1526 * there must be a software bug.
1528 static inline int warn_kprobe_rereg(struct kprobe *p)
1532 mutex_lock(&kprobe_mutex);
1533 if (WARN_ON_ONCE(__get_valid_kprobe(p)))
1535 mutex_unlock(&kprobe_mutex);
1540 int __weak arch_check_ftrace_location(struct kprobe *p)
1542 unsigned long ftrace_addr;
1544 ftrace_addr = ftrace_location((unsigned long)p->addr);
1546 #ifdef CONFIG_KPROBES_ON_FTRACE
1547 /* Given address is not on the instruction boundary */
1548 if ((unsigned long)p->addr != ftrace_addr)
1550 p->flags |= KPROBE_FLAG_FTRACE;
1551 #else /* !CONFIG_KPROBES_ON_FTRACE */
1558 static int check_kprobe_address_safe(struct kprobe *p,
1559 struct module **probed_mod)
1563 ret = arch_check_ftrace_location(p);
1569 /* Ensure it is not in reserved area nor out of text */
1570 if (!kernel_text_address((unsigned long) p->addr) ||
1571 within_kprobe_blacklist((unsigned long) p->addr) ||
1572 jump_label_text_reserved(p->addr, p->addr) ||
1573 static_call_text_reserved(p->addr, p->addr) ||
1574 find_bug((unsigned long)p->addr)) {
1579 /* Check if are we probing a module */
1580 *probed_mod = __module_text_address((unsigned long) p->addr);
1583 * We must hold a refcount of the probed module while updating
1584 * its code to prohibit unexpected unloading.
1586 if (unlikely(!try_module_get(*probed_mod))) {
1592 * If the module freed .init.text, we couldn't insert
1595 if (within_module_init((unsigned long)p->addr, *probed_mod) &&
1596 (*probed_mod)->state != MODULE_STATE_COMING) {
1597 module_put(*probed_mod);
1604 jump_label_unlock();
1609 int register_kprobe(struct kprobe *p)
1612 struct kprobe *old_p;
1613 struct module *probed_mod;
1614 kprobe_opcode_t *addr;
1616 /* Adjust probe address from symbol */
1617 addr = kprobe_addr(p);
1619 return PTR_ERR(addr);
1622 ret = warn_kprobe_rereg(p);
1626 /* User can pass only KPROBE_FLAG_DISABLED to register_kprobe */
1627 p->flags &= KPROBE_FLAG_DISABLED;
1629 INIT_LIST_HEAD(&p->list);
1631 ret = check_kprobe_address_safe(p, &probed_mod);
1635 mutex_lock(&kprobe_mutex);
1637 old_p = get_kprobe(p->addr);
1639 /* Since this may unoptimize old_p, locking text_mutex. */
1640 ret = register_aggr_kprobe(old_p, p);
1645 /* Prevent text modification */
1646 mutex_lock(&text_mutex);
1647 ret = prepare_kprobe(p);
1648 mutex_unlock(&text_mutex);
1653 INIT_HLIST_NODE(&p->hlist);
1654 hlist_add_head_rcu(&p->hlist,
1655 &kprobe_table[hash_ptr(p->addr, KPROBE_HASH_BITS)]);
1657 if (!kprobes_all_disarmed && !kprobe_disabled(p)) {
1658 ret = arm_kprobe(p);
1660 hlist_del_rcu(&p->hlist);
1666 /* Try to optimize kprobe */
1667 try_to_optimize_kprobe(p);
1669 mutex_unlock(&kprobe_mutex);
1672 module_put(probed_mod);
1676 EXPORT_SYMBOL_GPL(register_kprobe);
1678 /* Check if all probes on the aggrprobe are disabled */
1679 static int aggr_kprobe_disabled(struct kprobe *ap)
1683 lockdep_assert_held(&kprobe_mutex);
1685 list_for_each_entry(kp, &ap->list, list)
1686 if (!kprobe_disabled(kp))
1688 * There is an active probe on the list.
1689 * We can't disable this ap.
1696 /* Disable one kprobe: Make sure called under kprobe_mutex is locked */
1697 static struct kprobe *__disable_kprobe(struct kprobe *p)
1699 struct kprobe *orig_p;
1702 /* Get an original kprobe for return */
1703 orig_p = __get_valid_kprobe(p);
1704 if (unlikely(orig_p == NULL))
1705 return ERR_PTR(-EINVAL);
1707 if (!kprobe_disabled(p)) {
1708 /* Disable probe if it is a child probe */
1710 p->flags |= KPROBE_FLAG_DISABLED;
1712 /* Try to disarm and disable this/parent probe */
1713 if (p == orig_p || aggr_kprobe_disabled(orig_p)) {
1715 * If kprobes_all_disarmed is set, orig_p
1716 * should have already been disarmed, so
1717 * skip unneed disarming process.
1719 if (!kprobes_all_disarmed) {
1720 ret = disarm_kprobe(orig_p, true);
1722 p->flags &= ~KPROBE_FLAG_DISABLED;
1723 return ERR_PTR(ret);
1726 orig_p->flags |= KPROBE_FLAG_DISABLED;
1734 * Unregister a kprobe without a scheduler synchronization.
1736 static int __unregister_kprobe_top(struct kprobe *p)
1738 struct kprobe *ap, *list_p;
1740 /* Disable kprobe. This will disarm it if needed. */
1741 ap = __disable_kprobe(p);
1747 * This probe is an independent(and non-optimized) kprobe
1748 * (not an aggrprobe). Remove from the hash list.
1752 /* Following process expects this probe is an aggrprobe */
1753 WARN_ON(!kprobe_aggrprobe(ap));
1755 if (list_is_singular(&ap->list) && kprobe_disarmed(ap))
1757 * !disarmed could be happen if the probe is under delayed
1762 /* If disabling probe has special handlers, update aggrprobe */
1763 if (p->post_handler && !kprobe_gone(p)) {
1764 list_for_each_entry(list_p, &ap->list, list) {
1765 if ((list_p != p) && (list_p->post_handler))
1768 ap->post_handler = NULL;
1772 * Remove from the aggrprobe: this path will do nothing in
1773 * __unregister_kprobe_bottom().
1775 list_del_rcu(&p->list);
1776 if (!kprobe_disabled(ap) && !kprobes_all_disarmed)
1778 * Try to optimize this probe again, because post
1779 * handler may have been changed.
1781 optimize_kprobe(ap);
1786 hlist_del_rcu(&ap->hlist);
1790 static void __unregister_kprobe_bottom(struct kprobe *p)
1794 if (list_empty(&p->list))
1795 /* This is an independent kprobe */
1796 arch_remove_kprobe(p);
1797 else if (list_is_singular(&p->list)) {
1798 /* This is the last child of an aggrprobe */
1799 ap = list_entry(p->list.next, struct kprobe, list);
1801 free_aggr_kprobe(ap);
1803 /* Otherwise, do nothing. */
1806 int register_kprobes(struct kprobe **kps, int num)
1812 for (i = 0; i < num; i++) {
1813 ret = register_kprobe(kps[i]);
1816 unregister_kprobes(kps, i);
1822 EXPORT_SYMBOL_GPL(register_kprobes);
1824 void unregister_kprobe(struct kprobe *p)
1826 unregister_kprobes(&p, 1);
1828 EXPORT_SYMBOL_GPL(unregister_kprobe);
1830 void unregister_kprobes(struct kprobe **kps, int num)
1836 mutex_lock(&kprobe_mutex);
1837 for (i = 0; i < num; i++)
1838 if (__unregister_kprobe_top(kps[i]) < 0)
1839 kps[i]->addr = NULL;
1840 mutex_unlock(&kprobe_mutex);
1843 for (i = 0; i < num; i++)
1845 __unregister_kprobe_bottom(kps[i]);
1847 EXPORT_SYMBOL_GPL(unregister_kprobes);
1849 int __weak kprobe_exceptions_notify(struct notifier_block *self,
1850 unsigned long val, void *data)
1854 NOKPROBE_SYMBOL(kprobe_exceptions_notify);
1856 static struct notifier_block kprobe_exceptions_nb = {
1857 .notifier_call = kprobe_exceptions_notify,
1858 .priority = 0x7fffffff /* we need to be notified first */
1861 unsigned long __weak arch_deref_entry_point(void *entry)
1863 return (unsigned long)entry;
1866 #ifdef CONFIG_KRETPROBES
1868 unsigned long __kretprobe_trampoline_handler(struct pt_regs *regs,
1869 void *trampoline_address,
1870 void *frame_pointer)
1872 kprobe_opcode_t *correct_ret_addr = NULL;
1873 struct kretprobe_instance *ri = NULL;
1874 struct llist_node *first, *node;
1875 struct kretprobe *rp;
1877 /* Find all nodes for this frame. */
1878 first = node = current->kretprobe_instances.first;
1880 ri = container_of(node, struct kretprobe_instance, llist);
1882 BUG_ON(ri->fp != frame_pointer);
1884 if (ri->ret_addr != trampoline_address) {
1885 correct_ret_addr = ri->ret_addr;
1887 * This is the real return address. Any other
1888 * instances associated with this task are for
1889 * other calls deeper on the call stack
1896 pr_err("Oops! Kretprobe fails to find correct return address.\n");
1900 /* Unlink all nodes for this frame. */
1901 current->kretprobe_instances.first = node->next;
1906 ri = container_of(first, struct kretprobe_instance, llist);
1907 first = first->next;
1909 rp = get_kretprobe(ri);
1910 if (rp && rp->handler) {
1911 struct kprobe *prev = kprobe_running();
1913 __this_cpu_write(current_kprobe, &rp->kp);
1914 ri->ret_addr = correct_ret_addr;
1915 rp->handler(ri, regs);
1916 __this_cpu_write(current_kprobe, prev);
1919 recycle_rp_inst(ri);
1922 return (unsigned long)correct_ret_addr;
1924 NOKPROBE_SYMBOL(__kretprobe_trampoline_handler)
1927 * This kprobe pre_handler is registered with every kretprobe. When probe
1928 * hits it will set up the return probe.
1930 static int pre_handler_kretprobe(struct kprobe *p, struct pt_regs *regs)
1932 struct kretprobe *rp = container_of(p, struct kretprobe, kp);
1933 struct kretprobe_instance *ri;
1934 struct freelist_node *fn;
1936 fn = freelist_try_get(&rp->freelist);
1942 ri = container_of(fn, struct kretprobe_instance, freelist);
1944 if (rp->entry_handler && rp->entry_handler(ri, regs)) {
1945 freelist_add(&ri->freelist, &rp->freelist);
1949 arch_prepare_kretprobe(ri, regs);
1951 __llist_add(&ri->llist, ¤t->kretprobe_instances);
1955 NOKPROBE_SYMBOL(pre_handler_kretprobe);
1957 bool __weak arch_kprobe_on_func_entry(unsigned long offset)
1963 * kprobe_on_func_entry() -- check whether given address is function entry
1964 * @addr: Target address
1965 * @sym: Target symbol name
1966 * @offset: The offset from the symbol or the address
1968 * This checks whether the given @addr+@offset or @sym+@offset is on the
1969 * function entry address or not.
1970 * This returns 0 if it is the function entry, or -EINVAL if it is not.
1971 * And also it returns -ENOENT if it fails the symbol or address lookup.
1972 * Caller must pass @addr or @sym (either one must be NULL), or this
1975 int kprobe_on_func_entry(kprobe_opcode_t *addr, const char *sym, unsigned long offset)
1977 kprobe_opcode_t *kp_addr = _kprobe_addr(addr, sym, offset);
1979 if (IS_ERR(kp_addr))
1980 return PTR_ERR(kp_addr);
1982 if (!kallsyms_lookup_size_offset((unsigned long)kp_addr, NULL, &offset))
1985 if (!arch_kprobe_on_func_entry(offset))
1991 int register_kretprobe(struct kretprobe *rp)
1994 struct kretprobe_instance *inst;
1998 ret = kprobe_on_func_entry(rp->kp.addr, rp->kp.symbol_name, rp->kp.offset);
2002 /* If only rp->kp.addr is specified, check reregistering kprobes */
2003 if (rp->kp.addr && warn_kprobe_rereg(&rp->kp))
2006 if (kretprobe_blacklist_size) {
2007 addr = kprobe_addr(&rp->kp);
2009 return PTR_ERR(addr);
2011 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
2012 if (kretprobe_blacklist[i].addr == addr)
2017 rp->kp.pre_handler = pre_handler_kretprobe;
2018 rp->kp.post_handler = NULL;
2019 rp->kp.fault_handler = NULL;
2021 /* Pre-allocate memory for max kretprobe instances */
2022 if (rp->maxactive <= 0) {
2023 #ifdef CONFIG_PREEMPTION
2024 rp->maxactive = max_t(unsigned int, 10, 2*num_possible_cpus());
2026 rp->maxactive = num_possible_cpus();
2029 rp->freelist.head = NULL;
2030 rp->rph = kzalloc(sizeof(struct kretprobe_holder), GFP_KERNEL);
2035 for (i = 0; i < rp->maxactive; i++) {
2036 inst = kzalloc(sizeof(struct kretprobe_instance) +
2037 rp->data_size, GFP_KERNEL);
2039 refcount_set(&rp->rph->ref, i);
2043 inst->rph = rp->rph;
2044 freelist_add(&inst->freelist, &rp->freelist);
2046 refcount_set(&rp->rph->ref, i);
2049 /* Establish function entry probe point */
2050 ret = register_kprobe(&rp->kp);
2055 EXPORT_SYMBOL_GPL(register_kretprobe);
2057 int register_kretprobes(struct kretprobe **rps, int num)
2063 for (i = 0; i < num; i++) {
2064 ret = register_kretprobe(rps[i]);
2067 unregister_kretprobes(rps, i);
2073 EXPORT_SYMBOL_GPL(register_kretprobes);
2075 void unregister_kretprobe(struct kretprobe *rp)
2077 unregister_kretprobes(&rp, 1);
2079 EXPORT_SYMBOL_GPL(unregister_kretprobe);
2081 void unregister_kretprobes(struct kretprobe **rps, int num)
2087 mutex_lock(&kprobe_mutex);
2088 for (i = 0; i < num; i++) {
2089 if (__unregister_kprobe_top(&rps[i]->kp) < 0)
2090 rps[i]->kp.addr = NULL;
2091 rps[i]->rph->rp = NULL;
2093 mutex_unlock(&kprobe_mutex);
2096 for (i = 0; i < num; i++) {
2097 if (rps[i]->kp.addr) {
2098 __unregister_kprobe_bottom(&rps[i]->kp);
2099 free_rp_inst(rps[i]);
2103 EXPORT_SYMBOL_GPL(unregister_kretprobes);
2105 #else /* CONFIG_KRETPROBES */
2106 int register_kretprobe(struct kretprobe *rp)
2110 EXPORT_SYMBOL_GPL(register_kretprobe);
2112 int register_kretprobes(struct kretprobe **rps, int num)
2116 EXPORT_SYMBOL_GPL(register_kretprobes);
2118 void unregister_kretprobe(struct kretprobe *rp)
2121 EXPORT_SYMBOL_GPL(unregister_kretprobe);
2123 void unregister_kretprobes(struct kretprobe **rps, int num)
2126 EXPORT_SYMBOL_GPL(unregister_kretprobes);
2128 static int pre_handler_kretprobe(struct kprobe *p, struct pt_regs *regs)
2132 NOKPROBE_SYMBOL(pre_handler_kretprobe);
2134 #endif /* CONFIG_KRETPROBES */
2136 /* Set the kprobe gone and remove its instruction buffer. */
2137 static void kill_kprobe(struct kprobe *p)
2141 lockdep_assert_held(&kprobe_mutex);
2143 p->flags |= KPROBE_FLAG_GONE;
2144 if (kprobe_aggrprobe(p)) {
2146 * If this is an aggr_kprobe, we have to list all the
2147 * chained probes and mark them GONE.
2149 list_for_each_entry(kp, &p->list, list)
2150 kp->flags |= KPROBE_FLAG_GONE;
2151 p->post_handler = NULL;
2152 kill_optimized_kprobe(p);
2155 * Here, we can remove insn_slot safely, because no thread calls
2156 * the original probed function (which will be freed soon) any more.
2158 arch_remove_kprobe(p);
2161 * The module is going away. We should disarm the kprobe which
2162 * is using ftrace, because ftrace framework is still available at
2163 * MODULE_STATE_GOING notification.
2165 if (kprobe_ftrace(p) && !kprobe_disabled(p) && !kprobes_all_disarmed)
2166 disarm_kprobe_ftrace(p);
2169 /* Disable one kprobe */
2170 int disable_kprobe(struct kprobe *kp)
2175 mutex_lock(&kprobe_mutex);
2177 /* Disable this kprobe */
2178 p = __disable_kprobe(kp);
2182 mutex_unlock(&kprobe_mutex);
2185 EXPORT_SYMBOL_GPL(disable_kprobe);
2187 /* Enable one kprobe */
2188 int enable_kprobe(struct kprobe *kp)
2193 mutex_lock(&kprobe_mutex);
2195 /* Check whether specified probe is valid. */
2196 p = __get_valid_kprobe(kp);
2197 if (unlikely(p == NULL)) {
2202 if (kprobe_gone(kp)) {
2203 /* This kprobe has gone, we couldn't enable it. */
2209 kp->flags &= ~KPROBE_FLAG_DISABLED;
2211 if (!kprobes_all_disarmed && kprobe_disabled(p)) {
2212 p->flags &= ~KPROBE_FLAG_DISABLED;
2213 ret = arm_kprobe(p);
2215 p->flags |= KPROBE_FLAG_DISABLED;
2218 mutex_unlock(&kprobe_mutex);
2221 EXPORT_SYMBOL_GPL(enable_kprobe);
2223 /* Caller must NOT call this in usual path. This is only for critical case */
2224 void dump_kprobe(struct kprobe *kp)
2226 pr_err("Dumping kprobe:\n");
2227 pr_err("Name: %s\nOffset: %x\nAddress: %pS\n",
2228 kp->symbol_name, kp->offset, kp->addr);
2230 NOKPROBE_SYMBOL(dump_kprobe);
2232 int kprobe_add_ksym_blacklist(unsigned long entry)
2234 struct kprobe_blacklist_entry *ent;
2235 unsigned long offset = 0, size = 0;
2237 if (!kernel_text_address(entry) ||
2238 !kallsyms_lookup_size_offset(entry, &size, &offset))
2241 ent = kmalloc(sizeof(*ent), GFP_KERNEL);
2244 ent->start_addr = entry;
2245 ent->end_addr = entry + size;
2246 INIT_LIST_HEAD(&ent->list);
2247 list_add_tail(&ent->list, &kprobe_blacklist);
2252 /* Add all symbols in given area into kprobe blacklist */
2253 int kprobe_add_area_blacklist(unsigned long start, unsigned long end)
2255 unsigned long entry;
2258 for (entry = start; entry < end; entry += ret) {
2259 ret = kprobe_add_ksym_blacklist(entry);
2262 if (ret == 0) /* In case of alias symbol */
2268 /* Remove all symbols in given area from kprobe blacklist */
2269 static void kprobe_remove_area_blacklist(unsigned long start, unsigned long end)
2271 struct kprobe_blacklist_entry *ent, *n;
2273 list_for_each_entry_safe(ent, n, &kprobe_blacklist, list) {
2274 if (ent->start_addr < start || ent->start_addr >= end)
2276 list_del(&ent->list);
2281 static void kprobe_remove_ksym_blacklist(unsigned long entry)
2283 kprobe_remove_area_blacklist(entry, entry + 1);
2286 int __weak arch_kprobe_get_kallsym(unsigned int *symnum, unsigned long *value,
2287 char *type, char *sym)
2292 int kprobe_get_kallsym(unsigned int symnum, unsigned long *value, char *type,
2295 #ifdef __ARCH_WANT_KPROBES_INSN_SLOT
2296 if (!kprobe_cache_get_kallsym(&kprobe_insn_slots, &symnum, value, type, sym))
2298 #ifdef CONFIG_OPTPROBES
2299 if (!kprobe_cache_get_kallsym(&kprobe_optinsn_slots, &symnum, value, type, sym))
2303 if (!arch_kprobe_get_kallsym(&symnum, value, type, sym))
2308 int __init __weak arch_populate_kprobe_blacklist(void)
2314 * Lookup and populate the kprobe_blacklist.
2316 * Unlike the kretprobe blacklist, we'll need to determine
2317 * the range of addresses that belong to the said functions,
2318 * since a kprobe need not necessarily be at the beginning
2321 static int __init populate_kprobe_blacklist(unsigned long *start,
2324 unsigned long entry;
2325 unsigned long *iter;
2328 for (iter = start; iter < end; iter++) {
2329 entry = arch_deref_entry_point((void *)*iter);
2330 ret = kprobe_add_ksym_blacklist(entry);
2337 /* Symbols in __kprobes_text are blacklisted */
2338 ret = kprobe_add_area_blacklist((unsigned long)__kprobes_text_start,
2339 (unsigned long)__kprobes_text_end);
2343 /* Symbols in noinstr section are blacklisted */
2344 ret = kprobe_add_area_blacklist((unsigned long)__noinstr_text_start,
2345 (unsigned long)__noinstr_text_end);
2347 return ret ? : arch_populate_kprobe_blacklist();
2350 static void add_module_kprobe_blacklist(struct module *mod)
2352 unsigned long start, end;
2355 if (mod->kprobe_blacklist) {
2356 for (i = 0; i < mod->num_kprobe_blacklist; i++)
2357 kprobe_add_ksym_blacklist(mod->kprobe_blacklist[i]);
2360 start = (unsigned long)mod->kprobes_text_start;
2362 end = start + mod->kprobes_text_size;
2363 kprobe_add_area_blacklist(start, end);
2366 start = (unsigned long)mod->noinstr_text_start;
2368 end = start + mod->noinstr_text_size;
2369 kprobe_add_area_blacklist(start, end);
2373 static void remove_module_kprobe_blacklist(struct module *mod)
2375 unsigned long start, end;
2378 if (mod->kprobe_blacklist) {
2379 for (i = 0; i < mod->num_kprobe_blacklist; i++)
2380 kprobe_remove_ksym_blacklist(mod->kprobe_blacklist[i]);
2383 start = (unsigned long)mod->kprobes_text_start;
2385 end = start + mod->kprobes_text_size;
2386 kprobe_remove_area_blacklist(start, end);
2389 start = (unsigned long)mod->noinstr_text_start;
2391 end = start + mod->noinstr_text_size;
2392 kprobe_remove_area_blacklist(start, end);
2396 /* Module notifier call back, checking kprobes on the module */
2397 static int kprobes_module_callback(struct notifier_block *nb,
2398 unsigned long val, void *data)
2400 struct module *mod = data;
2401 struct hlist_head *head;
2404 int checkcore = (val == MODULE_STATE_GOING);
2406 if (val == MODULE_STATE_COMING) {
2407 mutex_lock(&kprobe_mutex);
2408 add_module_kprobe_blacklist(mod);
2409 mutex_unlock(&kprobe_mutex);
2411 if (val != MODULE_STATE_GOING && val != MODULE_STATE_LIVE)
2415 * When MODULE_STATE_GOING was notified, both of module .text and
2416 * .init.text sections would be freed. When MODULE_STATE_LIVE was
2417 * notified, only .init.text section would be freed. We need to
2418 * disable kprobes which have been inserted in the sections.
2420 mutex_lock(&kprobe_mutex);
2421 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2422 head = &kprobe_table[i];
2423 hlist_for_each_entry(p, head, hlist)
2424 if (within_module_init((unsigned long)p->addr, mod) ||
2426 within_module_core((unsigned long)p->addr, mod))) {
2428 * The vaddr this probe is installed will soon
2429 * be vfreed buy not synced to disk. Hence,
2430 * disarming the breakpoint isn't needed.
2432 * Note, this will also move any optimized probes
2433 * that are pending to be removed from their
2434 * corresponding lists to the freeing_list and
2435 * will not be touched by the delayed
2436 * kprobe_optimizer work handler.
2441 if (val == MODULE_STATE_GOING)
2442 remove_module_kprobe_blacklist(mod);
2443 mutex_unlock(&kprobe_mutex);
2447 static struct notifier_block kprobe_module_nb = {
2448 .notifier_call = kprobes_module_callback,
2452 /* Markers of _kprobe_blacklist section */
2453 extern unsigned long __start_kprobe_blacklist[];
2454 extern unsigned long __stop_kprobe_blacklist[];
2456 void kprobe_free_init_mem(void)
2458 void *start = (void *)(&__init_begin);
2459 void *end = (void *)(&__init_end);
2460 struct hlist_head *head;
2464 mutex_lock(&kprobe_mutex);
2466 /* Kill all kprobes on initmem */
2467 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2468 head = &kprobe_table[i];
2469 hlist_for_each_entry(p, head, hlist) {
2470 if (start <= (void *)p->addr && (void *)p->addr < end)
2475 mutex_unlock(&kprobe_mutex);
2478 static int __init init_kprobes(void)
2482 /* FIXME allocate the probe table, currently defined statically */
2483 /* initialize all list heads */
2484 for (i = 0; i < KPROBE_TABLE_SIZE; i++)
2485 INIT_HLIST_HEAD(&kprobe_table[i]);
2487 err = populate_kprobe_blacklist(__start_kprobe_blacklist,
2488 __stop_kprobe_blacklist);
2490 pr_err("kprobes: failed to populate blacklist: %d\n", err);
2491 pr_err("Please take care of using kprobes.\n");
2494 if (kretprobe_blacklist_size) {
2495 /* lookup the function address from its name */
2496 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
2497 kretprobe_blacklist[i].addr =
2498 kprobe_lookup_name(kretprobe_blacklist[i].name, 0);
2499 if (!kretprobe_blacklist[i].addr)
2500 printk("kretprobe: lookup failed: %s\n",
2501 kretprobe_blacklist[i].name);
2505 /* By default, kprobes are armed */
2506 kprobes_all_disarmed = false;
2508 #if defined(CONFIG_OPTPROBES) && defined(__ARCH_WANT_KPROBES_INSN_SLOT)
2509 /* Init kprobe_optinsn_slots for allocation */
2510 kprobe_optinsn_slots.insn_size = MAX_OPTINSN_SIZE;
2513 err = arch_init_kprobes();
2515 err = register_die_notifier(&kprobe_exceptions_nb);
2517 err = register_module_notifier(&kprobe_module_nb);
2519 kprobes_initialized = (err == 0);
2525 early_initcall(init_kprobes);
2527 #if defined(CONFIG_OPTPROBES)
2528 static int __init init_optprobes(void)
2531 * Enable kprobe optimization - this kicks the optimizer which
2532 * depends on synchronize_rcu_tasks() and ksoftirqd, that is
2533 * not spawned in early initcall. So delay the optimization.
2535 optimize_all_kprobes();
2539 subsys_initcall(init_optprobes);
2542 #ifdef CONFIG_DEBUG_FS
2543 static void report_probe(struct seq_file *pi, struct kprobe *p,
2544 const char *sym, int offset, char *modname, struct kprobe *pp)
2547 void *addr = p->addr;
2549 if (p->pre_handler == pre_handler_kretprobe)
2554 if (!kallsyms_show_value(pi->file->f_cred))
2558 seq_printf(pi, "%px %s %s+0x%x %s ",
2559 addr, kprobe_type, sym, offset,
2560 (modname ? modname : " "));
2561 else /* try to use %pS */
2562 seq_printf(pi, "%px %s %pS ",
2563 addr, kprobe_type, p->addr);
2567 seq_printf(pi, "%s%s%s%s\n",
2568 (kprobe_gone(p) ? "[GONE]" : ""),
2569 ((kprobe_disabled(p) && !kprobe_gone(p)) ? "[DISABLED]" : ""),
2570 (kprobe_optimized(pp) ? "[OPTIMIZED]" : ""),
2571 (kprobe_ftrace(pp) ? "[FTRACE]" : ""));
2574 static void *kprobe_seq_start(struct seq_file *f, loff_t *pos)
2576 return (*pos < KPROBE_TABLE_SIZE) ? pos : NULL;
2579 static void *kprobe_seq_next(struct seq_file *f, void *v, loff_t *pos)
2582 if (*pos >= KPROBE_TABLE_SIZE)
2587 static void kprobe_seq_stop(struct seq_file *f, void *v)
2592 static int show_kprobe_addr(struct seq_file *pi, void *v)
2594 struct hlist_head *head;
2595 struct kprobe *p, *kp;
2596 const char *sym = NULL;
2597 unsigned int i = *(loff_t *) v;
2598 unsigned long offset = 0;
2599 char *modname, namebuf[KSYM_NAME_LEN];
2601 head = &kprobe_table[i];
2603 hlist_for_each_entry_rcu(p, head, hlist) {
2604 sym = kallsyms_lookup((unsigned long)p->addr, NULL,
2605 &offset, &modname, namebuf);
2606 if (kprobe_aggrprobe(p)) {
2607 list_for_each_entry_rcu(kp, &p->list, list)
2608 report_probe(pi, kp, sym, offset, modname, p);
2610 report_probe(pi, p, sym, offset, modname, NULL);
2616 static const struct seq_operations kprobes_sops = {
2617 .start = kprobe_seq_start,
2618 .next = kprobe_seq_next,
2619 .stop = kprobe_seq_stop,
2620 .show = show_kprobe_addr
2623 DEFINE_SEQ_ATTRIBUTE(kprobes);
2625 /* kprobes/blacklist -- shows which functions can not be probed */
2626 static void *kprobe_blacklist_seq_start(struct seq_file *m, loff_t *pos)
2628 mutex_lock(&kprobe_mutex);
2629 return seq_list_start(&kprobe_blacklist, *pos);
2632 static void *kprobe_blacklist_seq_next(struct seq_file *m, void *v, loff_t *pos)
2634 return seq_list_next(v, &kprobe_blacklist, pos);
2637 static int kprobe_blacklist_seq_show(struct seq_file *m, void *v)
2639 struct kprobe_blacklist_entry *ent =
2640 list_entry(v, struct kprobe_blacklist_entry, list);
2643 * If /proc/kallsyms is not showing kernel address, we won't
2644 * show them here either.
2646 if (!kallsyms_show_value(m->file->f_cred))
2647 seq_printf(m, "0x%px-0x%px\t%ps\n", NULL, NULL,
2648 (void *)ent->start_addr);
2650 seq_printf(m, "0x%px-0x%px\t%ps\n", (void *)ent->start_addr,
2651 (void *)ent->end_addr, (void *)ent->start_addr);
2655 static void kprobe_blacklist_seq_stop(struct seq_file *f, void *v)
2657 mutex_unlock(&kprobe_mutex);
2660 static const struct seq_operations kprobe_blacklist_sops = {
2661 .start = kprobe_blacklist_seq_start,
2662 .next = kprobe_blacklist_seq_next,
2663 .stop = kprobe_blacklist_seq_stop,
2664 .show = kprobe_blacklist_seq_show,
2666 DEFINE_SEQ_ATTRIBUTE(kprobe_blacklist);
2668 static int arm_all_kprobes(void)
2670 struct hlist_head *head;
2672 unsigned int i, total = 0, errors = 0;
2675 mutex_lock(&kprobe_mutex);
2677 /* If kprobes are armed, just return */
2678 if (!kprobes_all_disarmed)
2679 goto already_enabled;
2682 * optimize_kprobe() called by arm_kprobe() checks
2683 * kprobes_all_disarmed, so set kprobes_all_disarmed before
2686 kprobes_all_disarmed = false;
2687 /* Arming kprobes doesn't optimize kprobe itself */
2688 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2689 head = &kprobe_table[i];
2690 /* Arm all kprobes on a best-effort basis */
2691 hlist_for_each_entry(p, head, hlist) {
2692 if (!kprobe_disabled(p)) {
2693 err = arm_kprobe(p);
2704 pr_warn("Kprobes globally enabled, but failed to arm %d out of %d probes\n",
2707 pr_info("Kprobes globally enabled\n");
2710 mutex_unlock(&kprobe_mutex);
2714 static int disarm_all_kprobes(void)
2716 struct hlist_head *head;
2718 unsigned int i, total = 0, errors = 0;
2721 mutex_lock(&kprobe_mutex);
2723 /* If kprobes are already disarmed, just return */
2724 if (kprobes_all_disarmed) {
2725 mutex_unlock(&kprobe_mutex);
2729 kprobes_all_disarmed = true;
2731 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2732 head = &kprobe_table[i];
2733 /* Disarm all kprobes on a best-effort basis */
2734 hlist_for_each_entry(p, head, hlist) {
2735 if (!arch_trampoline_kprobe(p) && !kprobe_disabled(p)) {
2736 err = disarm_kprobe(p, false);
2747 pr_warn("Kprobes globally disabled, but failed to disarm %d out of %d probes\n",
2750 pr_info("Kprobes globally disabled\n");
2752 mutex_unlock(&kprobe_mutex);
2754 /* Wait for disarming all kprobes by optimizer */
2755 wait_for_kprobe_optimizer();
2761 * XXX: The debugfs bool file interface doesn't allow for callbacks
2762 * when the bool state is switched. We can reuse that facility when
2765 static ssize_t read_enabled_file_bool(struct file *file,
2766 char __user *user_buf, size_t count, loff_t *ppos)
2770 if (!kprobes_all_disarmed)
2776 return simple_read_from_buffer(user_buf, count, ppos, buf, 2);
2779 static ssize_t write_enabled_file_bool(struct file *file,
2780 const char __user *user_buf, size_t count, loff_t *ppos)
2786 buf_size = min(count, (sizeof(buf)-1));
2787 if (copy_from_user(buf, user_buf, buf_size))
2790 buf[buf_size] = '\0';
2795 ret = arm_all_kprobes();
2800 ret = disarm_all_kprobes();
2812 static const struct file_operations fops_kp = {
2813 .read = read_enabled_file_bool,
2814 .write = write_enabled_file_bool,
2815 .llseek = default_llseek,
2818 static int __init debugfs_kprobe_init(void)
2821 unsigned int value = 1;
2823 dir = debugfs_create_dir("kprobes", NULL);
2825 debugfs_create_file("list", 0400, dir, NULL, &kprobes_fops);
2827 debugfs_create_file("enabled", 0600, dir, &value, &fops_kp);
2829 debugfs_create_file("blacklist", 0400, dir, NULL,
2830 &kprobe_blacklist_fops);
2835 late_initcall(debugfs_kprobe_init);
2836 #endif /* CONFIG_DEBUG_FS */