2 * Kernel Probes (KProbes)
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write to the Free Software
17 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
19 * Copyright (C) IBM Corporation, 2002, 2004
21 * 2002-Oct Created by Vamsi Krishna S <vamsi_krishna@in.ibm.com> Kernel
22 * Probes initial implementation (includes suggestions from
24 * 2004-Aug Updated by Prasanna S Panchamukhi <prasanna@in.ibm.com> with
25 * hlists and exceptions notifier as suggested by Andi Kleen.
26 * 2004-July Suparna Bhattacharya <suparna@in.ibm.com> added jumper probes
27 * interface to access function arguments.
28 * 2004-Sep Prasanna S Panchamukhi <prasanna@in.ibm.com> Changed Kprobes
29 * exceptions notifier to be first on the priority list.
30 * 2005-May Hien Nguyen <hien@us.ibm.com>, Jim Keniston
31 * <jkenisto@us.ibm.com> and Prasanna S Panchamukhi
32 * <prasanna@in.ibm.com> added function-return probes.
34 #include <linux/kprobes.h>
35 #include <linux/hash.h>
36 #include <linux/init.h>
37 #include <linux/slab.h>
38 #include <linux/stddef.h>
39 #include <linux/export.h>
40 #include <linux/moduleloader.h>
41 #include <linux/kallsyms.h>
42 #include <linux/freezer.h>
43 #include <linux/seq_file.h>
44 #include <linux/debugfs.h>
45 #include <linux/sysctl.h>
46 #include <linux/kdebug.h>
47 #include <linux/memory.h>
48 #include <linux/ftrace.h>
49 #include <linux/cpu.h>
50 #include <linux/jump_label.h>
52 #include <asm-generic/sections.h>
53 #include <asm/cacheflush.h>
54 #include <asm/errno.h>
55 #include <asm/uaccess.h>
57 #define KPROBE_HASH_BITS 6
58 #define KPROBE_TABLE_SIZE (1 << KPROBE_HASH_BITS)
62 * Some oddball architectures like 64bit powerpc have function descriptors
63 * so this must be overridable.
65 #ifndef kprobe_lookup_name
66 #define kprobe_lookup_name(name, addr) \
67 addr = ((kprobe_opcode_t *)(kallsyms_lookup_name(name)))
70 static int kprobes_initialized;
71 static struct hlist_head kprobe_table[KPROBE_TABLE_SIZE];
72 static struct hlist_head kretprobe_inst_table[KPROBE_TABLE_SIZE];
74 /* NOTE: change this value only with kprobe_mutex held */
75 static bool kprobes_all_disarmed;
77 /* This protects kprobe_table and optimizing_list */
78 static DEFINE_MUTEX(kprobe_mutex);
79 static DEFINE_PER_CPU(struct kprobe *, kprobe_instance) = NULL;
81 raw_spinlock_t lock ____cacheline_aligned_in_smp;
82 } kretprobe_table_locks[KPROBE_TABLE_SIZE];
84 static raw_spinlock_t *kretprobe_table_lock_ptr(unsigned long hash)
86 return &(kretprobe_table_locks[hash].lock);
89 /* Blacklist -- list of struct kprobe_blacklist_entry */
90 static LIST_HEAD(kprobe_blacklist);
92 #ifdef __ARCH_WANT_KPROBES_INSN_SLOT
94 * kprobe->ainsn.insn points to the copy of the instruction to be
95 * single-stepped. x86_64, POWER4 and above have no-exec support and
96 * stepping on the instruction on a vmalloced/kmalloced/data page
97 * is a recipe for disaster
99 struct kprobe_insn_page {
100 struct list_head list;
101 kprobe_opcode_t *insns; /* Page of instruction slots */
102 struct kprobe_insn_cache *cache;
108 #define KPROBE_INSN_PAGE_SIZE(slots) \
109 (offsetof(struct kprobe_insn_page, slot_used) + \
110 (sizeof(char) * (slots)))
112 static int slots_per_page(struct kprobe_insn_cache *c)
114 return PAGE_SIZE/(c->insn_size * sizeof(kprobe_opcode_t));
117 enum kprobe_slot_state {
123 static void *alloc_insn_page(void)
125 return module_alloc(PAGE_SIZE);
128 static void free_insn_page(void *page)
130 module_free(NULL, page);
133 struct kprobe_insn_cache kprobe_insn_slots = {
134 .mutex = __MUTEX_INITIALIZER(kprobe_insn_slots.mutex),
135 .alloc = alloc_insn_page,
136 .free = free_insn_page,
137 .pages = LIST_HEAD_INIT(kprobe_insn_slots.pages),
138 .insn_size = MAX_INSN_SIZE,
141 static int collect_garbage_slots(struct kprobe_insn_cache *c);
144 * __get_insn_slot() - Find a slot on an executable page for an instruction.
145 * We allocate an executable page if there's no room on existing ones.
147 kprobe_opcode_t *__get_insn_slot(struct kprobe_insn_cache *c)
149 struct kprobe_insn_page *kip;
150 kprobe_opcode_t *slot = NULL;
152 mutex_lock(&c->mutex);
154 list_for_each_entry(kip, &c->pages, list) {
155 if (kip->nused < slots_per_page(c)) {
157 for (i = 0; i < slots_per_page(c); i++) {
158 if (kip->slot_used[i] == SLOT_CLEAN) {
159 kip->slot_used[i] = SLOT_USED;
161 slot = kip->insns + (i * c->insn_size);
165 /* kip->nused is broken. Fix it. */
166 kip->nused = slots_per_page(c);
171 /* If there are any garbage slots, collect it and try again. */
172 if (c->nr_garbage && collect_garbage_slots(c) == 0)
175 /* All out of space. Need to allocate a new page. */
176 kip = kmalloc(KPROBE_INSN_PAGE_SIZE(slots_per_page(c)), GFP_KERNEL);
181 * Use module_alloc so this page is within +/- 2GB of where the
182 * kernel image and loaded module images reside. This is required
183 * so x86_64 can correctly handle the %rip-relative fixups.
185 kip->insns = c->alloc();
190 INIT_LIST_HEAD(&kip->list);
191 memset(kip->slot_used, SLOT_CLEAN, slots_per_page(c));
192 kip->slot_used[0] = SLOT_USED;
196 list_add(&kip->list, &c->pages);
199 mutex_unlock(&c->mutex);
203 /* Return 1 if all garbages are collected, otherwise 0. */
204 static int collect_one_slot(struct kprobe_insn_page *kip, int idx)
206 kip->slot_used[idx] = SLOT_CLEAN;
208 if (kip->nused == 0) {
210 * Page is no longer in use. Free it unless
211 * it's the last one. We keep the last one
212 * so as not to have to set it up again the
213 * next time somebody inserts a probe.
215 if (!list_is_singular(&kip->list)) {
216 list_del(&kip->list);
217 kip->cache->free(kip->insns);
225 static int collect_garbage_slots(struct kprobe_insn_cache *c)
227 struct kprobe_insn_page *kip, *next;
229 /* Ensure no-one is interrupted on the garbages */
232 list_for_each_entry_safe(kip, next, &c->pages, list) {
234 if (kip->ngarbage == 0)
236 kip->ngarbage = 0; /* we will collect all garbages */
237 for (i = 0; i < slots_per_page(c); i++) {
238 if (kip->slot_used[i] == SLOT_DIRTY &&
239 collect_one_slot(kip, i))
247 void __free_insn_slot(struct kprobe_insn_cache *c,
248 kprobe_opcode_t *slot, int dirty)
250 struct kprobe_insn_page *kip;
252 mutex_lock(&c->mutex);
253 list_for_each_entry(kip, &c->pages, list) {
254 long idx = ((long)slot - (long)kip->insns) /
255 (c->insn_size * sizeof(kprobe_opcode_t));
256 if (idx >= 0 && idx < slots_per_page(c)) {
257 WARN_ON(kip->slot_used[idx] != SLOT_USED);
259 kip->slot_used[idx] = SLOT_DIRTY;
261 if (++c->nr_garbage > slots_per_page(c))
262 collect_garbage_slots(c);
264 collect_one_slot(kip, idx);
268 /* Could not free this slot. */
271 mutex_unlock(&c->mutex);
274 #ifdef CONFIG_OPTPROBES
275 /* For optimized_kprobe buffer */
276 struct kprobe_insn_cache kprobe_optinsn_slots = {
277 .mutex = __MUTEX_INITIALIZER(kprobe_optinsn_slots.mutex),
278 .alloc = alloc_insn_page,
279 .free = free_insn_page,
280 .pages = LIST_HEAD_INIT(kprobe_optinsn_slots.pages),
281 /* .insn_size is initialized later */
287 /* We have preemption disabled.. so it is safe to use __ versions */
288 static inline void set_kprobe_instance(struct kprobe *kp)
290 __this_cpu_write(kprobe_instance, kp);
293 static inline void reset_kprobe_instance(void)
295 __this_cpu_write(kprobe_instance, NULL);
299 * This routine is called either:
300 * - under the kprobe_mutex - during kprobe_[un]register()
302 * - with preemption disabled - from arch/xxx/kernel/kprobes.c
304 struct kprobe __kprobes *get_kprobe(void *addr)
306 struct hlist_head *head;
309 head = &kprobe_table[hash_ptr(addr, KPROBE_HASH_BITS)];
310 hlist_for_each_entry_rcu(p, head, hlist) {
318 static int __kprobes aggr_pre_handler(struct kprobe *p, struct pt_regs *regs);
320 /* Return true if the kprobe is an aggregator */
321 static inline int kprobe_aggrprobe(struct kprobe *p)
323 return p->pre_handler == aggr_pre_handler;
326 /* Return true(!0) if the kprobe is unused */
327 static inline int kprobe_unused(struct kprobe *p)
329 return kprobe_aggrprobe(p) && kprobe_disabled(p) &&
330 list_empty(&p->list);
334 * Keep all fields in the kprobe consistent
336 static inline void copy_kprobe(struct kprobe *ap, struct kprobe *p)
338 memcpy(&p->opcode, &ap->opcode, sizeof(kprobe_opcode_t));
339 memcpy(&p->ainsn, &ap->ainsn, sizeof(struct arch_specific_insn));
342 #ifdef CONFIG_OPTPROBES
343 /* NOTE: change this value only with kprobe_mutex held */
344 static bool kprobes_allow_optimization;
347 * Call all pre_handler on the list, but ignores its return value.
348 * This must be called from arch-dep optimized caller.
350 void __kprobes opt_pre_handler(struct kprobe *p, struct pt_regs *regs)
354 list_for_each_entry_rcu(kp, &p->list, list) {
355 if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
356 set_kprobe_instance(kp);
357 kp->pre_handler(kp, regs);
359 reset_kprobe_instance();
363 /* Free optimized instructions and optimized_kprobe */
364 static void free_aggr_kprobe(struct kprobe *p)
366 struct optimized_kprobe *op;
368 op = container_of(p, struct optimized_kprobe, kp);
369 arch_remove_optimized_kprobe(op);
370 arch_remove_kprobe(p);
374 /* Return true(!0) if the kprobe is ready for optimization. */
375 static inline int kprobe_optready(struct kprobe *p)
377 struct optimized_kprobe *op;
379 if (kprobe_aggrprobe(p)) {
380 op = container_of(p, struct optimized_kprobe, kp);
381 return arch_prepared_optinsn(&op->optinsn);
387 /* Return true(!0) if the kprobe is disarmed. Note: p must be on hash list */
388 static inline int kprobe_disarmed(struct kprobe *p)
390 struct optimized_kprobe *op;
392 /* If kprobe is not aggr/opt probe, just return kprobe is disabled */
393 if (!kprobe_aggrprobe(p))
394 return kprobe_disabled(p);
396 op = container_of(p, struct optimized_kprobe, kp);
398 return kprobe_disabled(p) && list_empty(&op->list);
401 /* Return true(!0) if the probe is queued on (un)optimizing lists */
402 static int kprobe_queued(struct kprobe *p)
404 struct optimized_kprobe *op;
406 if (kprobe_aggrprobe(p)) {
407 op = container_of(p, struct optimized_kprobe, kp);
408 if (!list_empty(&op->list))
415 * Return an optimized kprobe whose optimizing code replaces
416 * instructions including addr (exclude breakpoint).
418 static struct kprobe *get_optimized_kprobe(unsigned long addr)
421 struct kprobe *p = NULL;
422 struct optimized_kprobe *op;
424 /* Don't check i == 0, since that is a breakpoint case. */
425 for (i = 1; !p && i < MAX_OPTIMIZED_LENGTH; i++)
426 p = get_kprobe((void *)(addr - i));
428 if (p && kprobe_optready(p)) {
429 op = container_of(p, struct optimized_kprobe, kp);
430 if (arch_within_optimized_kprobe(op, addr))
437 /* Optimization staging list, protected by kprobe_mutex */
438 static LIST_HEAD(optimizing_list);
439 static LIST_HEAD(unoptimizing_list);
440 static LIST_HEAD(freeing_list);
442 static void kprobe_optimizer(struct work_struct *work);
443 static DECLARE_DELAYED_WORK(optimizing_work, kprobe_optimizer);
444 #define OPTIMIZE_DELAY 5
447 * Optimize (replace a breakpoint with a jump) kprobes listed on
450 static void do_optimize_kprobes(void)
452 /* Optimization never be done when disarmed */
453 if (kprobes_all_disarmed || !kprobes_allow_optimization ||
454 list_empty(&optimizing_list))
458 * The optimization/unoptimization refers online_cpus via
459 * stop_machine() and cpu-hotplug modifies online_cpus.
460 * And same time, text_mutex will be held in cpu-hotplug and here.
461 * This combination can cause a deadlock (cpu-hotplug try to lock
462 * text_mutex but stop_machine can not be done because online_cpus
464 * To avoid this deadlock, we need to call get_online_cpus()
465 * for preventing cpu-hotplug outside of text_mutex locking.
468 mutex_lock(&text_mutex);
469 arch_optimize_kprobes(&optimizing_list);
470 mutex_unlock(&text_mutex);
475 * Unoptimize (replace a jump with a breakpoint and remove the breakpoint
476 * if need) kprobes listed on unoptimizing_list.
478 static void do_unoptimize_kprobes(void)
480 struct optimized_kprobe *op, *tmp;
482 /* Unoptimization must be done anytime */
483 if (list_empty(&unoptimizing_list))
486 /* Ditto to do_optimize_kprobes */
488 mutex_lock(&text_mutex);
489 arch_unoptimize_kprobes(&unoptimizing_list, &freeing_list);
490 /* Loop free_list for disarming */
491 list_for_each_entry_safe(op, tmp, &freeing_list, list) {
492 /* Disarm probes if marked disabled */
493 if (kprobe_disabled(&op->kp))
494 arch_disarm_kprobe(&op->kp);
495 if (kprobe_unused(&op->kp)) {
497 * Remove unused probes from hash list. After waiting
498 * for synchronization, these probes are reclaimed.
499 * (reclaiming is done by do_free_cleaned_kprobes.)
501 hlist_del_rcu(&op->kp.hlist);
503 list_del_init(&op->list);
505 mutex_unlock(&text_mutex);
509 /* Reclaim all kprobes on the free_list */
510 static void do_free_cleaned_kprobes(void)
512 struct optimized_kprobe *op, *tmp;
514 list_for_each_entry_safe(op, tmp, &freeing_list, list) {
515 BUG_ON(!kprobe_unused(&op->kp));
516 list_del_init(&op->list);
517 free_aggr_kprobe(&op->kp);
521 /* Start optimizer after OPTIMIZE_DELAY passed */
522 static void kick_kprobe_optimizer(void)
524 schedule_delayed_work(&optimizing_work, OPTIMIZE_DELAY);
527 /* Kprobe jump optimizer */
528 static void kprobe_optimizer(struct work_struct *work)
530 mutex_lock(&kprobe_mutex);
531 /* Lock modules while optimizing kprobes */
532 mutex_lock(&module_mutex);
535 * Step 1: Unoptimize kprobes and collect cleaned (unused and disarmed)
536 * kprobes before waiting for quiesence period.
538 do_unoptimize_kprobes();
541 * Step 2: Wait for quiesence period to ensure all running interrupts
542 * are done. Because optprobe may modify multiple instructions
543 * there is a chance that Nth instruction is interrupted. In that
544 * case, running interrupt can return to 2nd-Nth byte of jump
545 * instruction. This wait is for avoiding it.
549 /* Step 3: Optimize kprobes after quiesence period */
550 do_optimize_kprobes();
552 /* Step 4: Free cleaned kprobes after quiesence period */
553 do_free_cleaned_kprobes();
555 mutex_unlock(&module_mutex);
556 mutex_unlock(&kprobe_mutex);
558 /* Step 5: Kick optimizer again if needed */
559 if (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list))
560 kick_kprobe_optimizer();
563 /* Wait for completing optimization and unoptimization */
564 static void wait_for_kprobe_optimizer(void)
566 mutex_lock(&kprobe_mutex);
568 while (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list)) {
569 mutex_unlock(&kprobe_mutex);
571 /* this will also make optimizing_work execute immmediately */
572 flush_delayed_work(&optimizing_work);
573 /* @optimizing_work might not have been queued yet, relax */
576 mutex_lock(&kprobe_mutex);
579 mutex_unlock(&kprobe_mutex);
582 /* Optimize kprobe if p is ready to be optimized */
583 static void optimize_kprobe(struct kprobe *p)
585 struct optimized_kprobe *op;
587 /* Check if the kprobe is disabled or not ready for optimization. */
588 if (!kprobe_optready(p) || !kprobes_allow_optimization ||
589 (kprobe_disabled(p) || kprobes_all_disarmed))
592 /* Both of break_handler and post_handler are not supported. */
593 if (p->break_handler || p->post_handler)
596 op = container_of(p, struct optimized_kprobe, kp);
598 /* Check there is no other kprobes at the optimized instructions */
599 if (arch_check_optimized_kprobe(op) < 0)
602 /* Check if it is already optimized. */
603 if (op->kp.flags & KPROBE_FLAG_OPTIMIZED)
605 op->kp.flags |= KPROBE_FLAG_OPTIMIZED;
607 if (!list_empty(&op->list))
608 /* This is under unoptimizing. Just dequeue the probe */
609 list_del_init(&op->list);
611 list_add(&op->list, &optimizing_list);
612 kick_kprobe_optimizer();
616 /* Short cut to direct unoptimizing */
617 static void force_unoptimize_kprobe(struct optimized_kprobe *op)
620 arch_unoptimize_kprobe(op);
622 if (kprobe_disabled(&op->kp))
623 arch_disarm_kprobe(&op->kp);
626 /* Unoptimize a kprobe if p is optimized */
627 static void unoptimize_kprobe(struct kprobe *p, bool force)
629 struct optimized_kprobe *op;
631 if (!kprobe_aggrprobe(p) || kprobe_disarmed(p))
632 return; /* This is not an optprobe nor optimized */
634 op = container_of(p, struct optimized_kprobe, kp);
635 if (!kprobe_optimized(p)) {
636 /* Unoptimized or unoptimizing case */
637 if (force && !list_empty(&op->list)) {
639 * Only if this is unoptimizing kprobe and forced,
640 * forcibly unoptimize it. (No need to unoptimize
641 * unoptimized kprobe again :)
643 list_del_init(&op->list);
644 force_unoptimize_kprobe(op);
649 op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
650 if (!list_empty(&op->list)) {
651 /* Dequeue from the optimization queue */
652 list_del_init(&op->list);
655 /* Optimized kprobe case */
657 /* Forcibly update the code: this is a special case */
658 force_unoptimize_kprobe(op);
660 list_add(&op->list, &unoptimizing_list);
661 kick_kprobe_optimizer();
665 /* Cancel unoptimizing for reusing */
666 static void reuse_unused_kprobe(struct kprobe *ap)
668 struct optimized_kprobe *op;
670 BUG_ON(!kprobe_unused(ap));
672 * Unused kprobe MUST be on the way of delayed unoptimizing (means
673 * there is still a relative jump) and disabled.
675 op = container_of(ap, struct optimized_kprobe, kp);
676 if (unlikely(list_empty(&op->list)))
677 printk(KERN_WARNING "Warning: found a stray unused "
678 "aggrprobe@%p\n", ap->addr);
679 /* Enable the probe again */
680 ap->flags &= ~KPROBE_FLAG_DISABLED;
681 /* Optimize it again (remove from op->list) */
682 BUG_ON(!kprobe_optready(ap));
686 /* Remove optimized instructions */
687 static void kill_optimized_kprobe(struct kprobe *p)
689 struct optimized_kprobe *op;
691 op = container_of(p, struct optimized_kprobe, kp);
692 if (!list_empty(&op->list))
693 /* Dequeue from the (un)optimization queue */
694 list_del_init(&op->list);
695 op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
697 if (kprobe_unused(p)) {
698 /* Enqueue if it is unused */
699 list_add(&op->list, &freeing_list);
701 * Remove unused probes from the hash list. After waiting
702 * for synchronization, this probe is reclaimed.
703 * (reclaiming is done by do_free_cleaned_kprobes().)
705 hlist_del_rcu(&op->kp.hlist);
708 /* Don't touch the code, because it is already freed. */
709 arch_remove_optimized_kprobe(op);
712 /* Try to prepare optimized instructions */
713 static void prepare_optimized_kprobe(struct kprobe *p)
715 struct optimized_kprobe *op;
717 op = container_of(p, struct optimized_kprobe, kp);
718 arch_prepare_optimized_kprobe(op);
721 /* Allocate new optimized_kprobe and try to prepare optimized instructions */
722 static struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
724 struct optimized_kprobe *op;
726 op = kzalloc(sizeof(struct optimized_kprobe), GFP_KERNEL);
730 INIT_LIST_HEAD(&op->list);
731 op->kp.addr = p->addr;
732 arch_prepare_optimized_kprobe(op);
737 static void init_aggr_kprobe(struct kprobe *ap, struct kprobe *p);
740 * Prepare an optimized_kprobe and optimize it
741 * NOTE: p must be a normal registered kprobe
743 static void try_to_optimize_kprobe(struct kprobe *p)
746 struct optimized_kprobe *op;
748 /* Impossible to optimize ftrace-based kprobe */
749 if (kprobe_ftrace(p))
752 /* For preparing optimization, jump_label_text_reserved() is called */
754 mutex_lock(&text_mutex);
756 ap = alloc_aggr_kprobe(p);
760 op = container_of(ap, struct optimized_kprobe, kp);
761 if (!arch_prepared_optinsn(&op->optinsn)) {
762 /* If failed to setup optimizing, fallback to kprobe */
763 arch_remove_optimized_kprobe(op);
768 init_aggr_kprobe(ap, p);
769 optimize_kprobe(ap); /* This just kicks optimizer thread */
772 mutex_unlock(&text_mutex);
777 static void optimize_all_kprobes(void)
779 struct hlist_head *head;
783 mutex_lock(&kprobe_mutex);
784 /* If optimization is already allowed, just return */
785 if (kprobes_allow_optimization)
788 kprobes_allow_optimization = true;
789 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
790 head = &kprobe_table[i];
791 hlist_for_each_entry_rcu(p, head, hlist)
792 if (!kprobe_disabled(p))
795 printk(KERN_INFO "Kprobes globally optimized\n");
797 mutex_unlock(&kprobe_mutex);
800 static void unoptimize_all_kprobes(void)
802 struct hlist_head *head;
806 mutex_lock(&kprobe_mutex);
807 /* If optimization is already prohibited, just return */
808 if (!kprobes_allow_optimization) {
809 mutex_unlock(&kprobe_mutex);
813 kprobes_allow_optimization = false;
814 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
815 head = &kprobe_table[i];
816 hlist_for_each_entry_rcu(p, head, hlist) {
817 if (!kprobe_disabled(p))
818 unoptimize_kprobe(p, false);
821 mutex_unlock(&kprobe_mutex);
823 /* Wait for unoptimizing completion */
824 wait_for_kprobe_optimizer();
825 printk(KERN_INFO "Kprobes globally unoptimized\n");
828 static DEFINE_MUTEX(kprobe_sysctl_mutex);
829 int sysctl_kprobes_optimization;
830 int proc_kprobes_optimization_handler(struct ctl_table *table, int write,
831 void __user *buffer, size_t *length,
836 mutex_lock(&kprobe_sysctl_mutex);
837 sysctl_kprobes_optimization = kprobes_allow_optimization ? 1 : 0;
838 ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
840 if (sysctl_kprobes_optimization)
841 optimize_all_kprobes();
843 unoptimize_all_kprobes();
844 mutex_unlock(&kprobe_sysctl_mutex);
848 #endif /* CONFIG_SYSCTL */
850 /* Put a breakpoint for a probe. Must be called with text_mutex locked */
851 static void __arm_kprobe(struct kprobe *p)
855 /* Check collision with other optimized kprobes */
856 _p = get_optimized_kprobe((unsigned long)p->addr);
858 /* Fallback to unoptimized kprobe */
859 unoptimize_kprobe(_p, true);
862 optimize_kprobe(p); /* Try to optimize (add kprobe to a list) */
865 /* Remove the breakpoint of a probe. Must be called with text_mutex locked */
866 static void __disarm_kprobe(struct kprobe *p, bool reopt)
870 unoptimize_kprobe(p, false); /* Try to unoptimize */
872 if (!kprobe_queued(p)) {
873 arch_disarm_kprobe(p);
874 /* If another kprobe was blocked, optimize it. */
875 _p = get_optimized_kprobe((unsigned long)p->addr);
876 if (unlikely(_p) && reopt)
879 /* TODO: reoptimize others after unoptimized this probe */
882 #else /* !CONFIG_OPTPROBES */
884 #define optimize_kprobe(p) do {} while (0)
885 #define unoptimize_kprobe(p, f) do {} while (0)
886 #define kill_optimized_kprobe(p) do {} while (0)
887 #define prepare_optimized_kprobe(p) do {} while (0)
888 #define try_to_optimize_kprobe(p) do {} while (0)
889 #define __arm_kprobe(p) arch_arm_kprobe(p)
890 #define __disarm_kprobe(p, o) arch_disarm_kprobe(p)
891 #define kprobe_disarmed(p) kprobe_disabled(p)
892 #define wait_for_kprobe_optimizer() do {} while (0)
894 /* There should be no unused kprobes can be reused without optimization */
895 static void reuse_unused_kprobe(struct kprobe *ap)
897 printk(KERN_ERR "Error: There should be no unused kprobe here.\n");
898 BUG_ON(kprobe_unused(ap));
901 static void free_aggr_kprobe(struct kprobe *p)
903 arch_remove_kprobe(p);
907 static struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
909 return kzalloc(sizeof(struct kprobe), GFP_KERNEL);
911 #endif /* CONFIG_OPTPROBES */
913 #ifdef CONFIG_KPROBES_ON_FTRACE
914 static struct ftrace_ops kprobe_ftrace_ops __read_mostly = {
915 .func = kprobe_ftrace_handler,
916 .flags = FTRACE_OPS_FL_SAVE_REGS,
918 static int kprobe_ftrace_enabled;
920 /* Must ensure p->addr is really on ftrace */
921 static int prepare_kprobe(struct kprobe *p)
923 if (!kprobe_ftrace(p))
924 return arch_prepare_kprobe(p);
926 return arch_prepare_kprobe_ftrace(p);
929 /* Caller must lock kprobe_mutex */
930 static void arm_kprobe_ftrace(struct kprobe *p)
934 ret = ftrace_set_filter_ip(&kprobe_ftrace_ops,
935 (unsigned long)p->addr, 0, 0);
936 WARN(ret < 0, "Failed to arm kprobe-ftrace at %p (%d)\n", p->addr, ret);
937 kprobe_ftrace_enabled++;
938 if (kprobe_ftrace_enabled == 1) {
939 ret = register_ftrace_function(&kprobe_ftrace_ops);
940 WARN(ret < 0, "Failed to init kprobe-ftrace (%d)\n", ret);
944 /* Caller must lock kprobe_mutex */
945 static void disarm_kprobe_ftrace(struct kprobe *p)
949 kprobe_ftrace_enabled--;
950 if (kprobe_ftrace_enabled == 0) {
951 ret = unregister_ftrace_function(&kprobe_ftrace_ops);
952 WARN(ret < 0, "Failed to init kprobe-ftrace (%d)\n", ret);
954 ret = ftrace_set_filter_ip(&kprobe_ftrace_ops,
955 (unsigned long)p->addr, 1, 0);
956 WARN(ret < 0, "Failed to disarm kprobe-ftrace at %p (%d)\n", p->addr, ret);
958 #else /* !CONFIG_KPROBES_ON_FTRACE */
959 #define prepare_kprobe(p) arch_prepare_kprobe(p)
960 #define arm_kprobe_ftrace(p) do {} while (0)
961 #define disarm_kprobe_ftrace(p) do {} while (0)
964 /* Arm a kprobe with text_mutex */
965 static void arm_kprobe(struct kprobe *kp)
967 if (unlikely(kprobe_ftrace(kp))) {
968 arm_kprobe_ftrace(kp);
972 * Here, since __arm_kprobe() doesn't use stop_machine(),
973 * this doesn't cause deadlock on text_mutex. So, we don't
974 * need get_online_cpus().
976 mutex_lock(&text_mutex);
978 mutex_unlock(&text_mutex);
981 /* Disarm a kprobe with text_mutex */
982 static void disarm_kprobe(struct kprobe *kp, bool reopt)
984 if (unlikely(kprobe_ftrace(kp))) {
985 disarm_kprobe_ftrace(kp);
989 mutex_lock(&text_mutex);
990 __disarm_kprobe(kp, reopt);
991 mutex_unlock(&text_mutex);
995 * Aggregate handlers for multiple kprobes support - these handlers
996 * take care of invoking the individual kprobe handlers on p->list
998 static int __kprobes aggr_pre_handler(struct kprobe *p, struct pt_regs *regs)
1002 list_for_each_entry_rcu(kp, &p->list, list) {
1003 if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
1004 set_kprobe_instance(kp);
1005 if (kp->pre_handler(kp, regs))
1008 reset_kprobe_instance();
1013 static void __kprobes aggr_post_handler(struct kprobe *p, struct pt_regs *regs,
1014 unsigned long flags)
1018 list_for_each_entry_rcu(kp, &p->list, list) {
1019 if (kp->post_handler && likely(!kprobe_disabled(kp))) {
1020 set_kprobe_instance(kp);
1021 kp->post_handler(kp, regs, flags);
1022 reset_kprobe_instance();
1027 static int __kprobes aggr_fault_handler(struct kprobe *p, struct pt_regs *regs,
1030 struct kprobe *cur = __this_cpu_read(kprobe_instance);
1033 * if we faulted "during" the execution of a user specified
1034 * probe handler, invoke just that probe's fault handler
1036 if (cur && cur->fault_handler) {
1037 if (cur->fault_handler(cur, regs, trapnr))
1043 static int __kprobes aggr_break_handler(struct kprobe *p, struct pt_regs *regs)
1045 struct kprobe *cur = __this_cpu_read(kprobe_instance);
1048 if (cur && cur->break_handler) {
1049 if (cur->break_handler(cur, regs))
1052 reset_kprobe_instance();
1056 /* Walks the list and increments nmissed count for multiprobe case */
1057 void __kprobes kprobes_inc_nmissed_count(struct kprobe *p)
1060 if (!kprobe_aggrprobe(p)) {
1063 list_for_each_entry_rcu(kp, &p->list, list)
1069 void __kprobes recycle_rp_inst(struct kretprobe_instance *ri,
1070 struct hlist_head *head)
1072 struct kretprobe *rp = ri->rp;
1074 /* remove rp inst off the rprobe_inst_table */
1075 hlist_del(&ri->hlist);
1076 INIT_HLIST_NODE(&ri->hlist);
1078 raw_spin_lock(&rp->lock);
1079 hlist_add_head(&ri->hlist, &rp->free_instances);
1080 raw_spin_unlock(&rp->lock);
1083 hlist_add_head(&ri->hlist, head);
1086 void __kprobes kretprobe_hash_lock(struct task_struct *tsk,
1087 struct hlist_head **head, unsigned long *flags)
1088 __acquires(hlist_lock)
1090 unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
1091 raw_spinlock_t *hlist_lock;
1093 *head = &kretprobe_inst_table[hash];
1094 hlist_lock = kretprobe_table_lock_ptr(hash);
1095 raw_spin_lock_irqsave(hlist_lock, *flags);
1098 static void __kprobes kretprobe_table_lock(unsigned long hash,
1099 unsigned long *flags)
1100 __acquires(hlist_lock)
1102 raw_spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
1103 raw_spin_lock_irqsave(hlist_lock, *flags);
1106 void __kprobes kretprobe_hash_unlock(struct task_struct *tsk,
1107 unsigned long *flags)
1108 __releases(hlist_lock)
1110 unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
1111 raw_spinlock_t *hlist_lock;
1113 hlist_lock = kretprobe_table_lock_ptr(hash);
1114 raw_spin_unlock_irqrestore(hlist_lock, *flags);
1117 static void __kprobes kretprobe_table_unlock(unsigned long hash,
1118 unsigned long *flags)
1119 __releases(hlist_lock)
1121 raw_spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
1122 raw_spin_unlock_irqrestore(hlist_lock, *flags);
1126 * This function is called from finish_task_switch when task tk becomes dead,
1127 * so that we can recycle any function-return probe instances associated
1128 * with this task. These left over instances represent probed functions
1129 * that have been called but will never return.
1131 void __kprobes kprobe_flush_task(struct task_struct *tk)
1133 struct kretprobe_instance *ri;
1134 struct hlist_head *head, empty_rp;
1135 struct hlist_node *tmp;
1136 unsigned long hash, flags = 0;
1138 if (unlikely(!kprobes_initialized))
1139 /* Early boot. kretprobe_table_locks not yet initialized. */
1142 INIT_HLIST_HEAD(&empty_rp);
1143 hash = hash_ptr(tk, KPROBE_HASH_BITS);
1144 head = &kretprobe_inst_table[hash];
1145 kretprobe_table_lock(hash, &flags);
1146 hlist_for_each_entry_safe(ri, tmp, head, hlist) {
1148 recycle_rp_inst(ri, &empty_rp);
1150 kretprobe_table_unlock(hash, &flags);
1151 hlist_for_each_entry_safe(ri, tmp, &empty_rp, hlist) {
1152 hlist_del(&ri->hlist);
1157 static inline void free_rp_inst(struct kretprobe *rp)
1159 struct kretprobe_instance *ri;
1160 struct hlist_node *next;
1162 hlist_for_each_entry_safe(ri, next, &rp->free_instances, hlist) {
1163 hlist_del(&ri->hlist);
1168 static void __kprobes cleanup_rp_inst(struct kretprobe *rp)
1170 unsigned long flags, hash;
1171 struct kretprobe_instance *ri;
1172 struct hlist_node *next;
1173 struct hlist_head *head;
1176 for (hash = 0; hash < KPROBE_TABLE_SIZE; hash++) {
1177 kretprobe_table_lock(hash, &flags);
1178 head = &kretprobe_inst_table[hash];
1179 hlist_for_each_entry_safe(ri, next, head, hlist) {
1183 kretprobe_table_unlock(hash, &flags);
1189 * Add the new probe to ap->list. Fail if this is the
1190 * second jprobe at the address - two jprobes can't coexist
1192 static int add_new_kprobe(struct kprobe *ap, struct kprobe *p)
1194 BUG_ON(kprobe_gone(ap) || kprobe_gone(p));
1196 if (p->break_handler || p->post_handler)
1197 unoptimize_kprobe(ap, true); /* Fall back to normal kprobe */
1199 if (p->break_handler) {
1200 if (ap->break_handler)
1202 list_add_tail_rcu(&p->list, &ap->list);
1203 ap->break_handler = aggr_break_handler;
1205 list_add_rcu(&p->list, &ap->list);
1206 if (p->post_handler && !ap->post_handler)
1207 ap->post_handler = aggr_post_handler;
1213 * Fill in the required fields of the "manager kprobe". Replace the
1214 * earlier kprobe in the hlist with the manager kprobe
1216 static void init_aggr_kprobe(struct kprobe *ap, struct kprobe *p)
1218 /* Copy p's insn slot to ap */
1220 flush_insn_slot(ap);
1222 ap->flags = p->flags & ~KPROBE_FLAG_OPTIMIZED;
1223 ap->pre_handler = aggr_pre_handler;
1224 ap->fault_handler = aggr_fault_handler;
1225 /* We don't care the kprobe which has gone. */
1226 if (p->post_handler && !kprobe_gone(p))
1227 ap->post_handler = aggr_post_handler;
1228 if (p->break_handler && !kprobe_gone(p))
1229 ap->break_handler = aggr_break_handler;
1231 INIT_LIST_HEAD(&ap->list);
1232 INIT_HLIST_NODE(&ap->hlist);
1234 list_add_rcu(&p->list, &ap->list);
1235 hlist_replace_rcu(&p->hlist, &ap->hlist);
1239 * This is the second or subsequent kprobe at the address - handle
1242 static int register_aggr_kprobe(struct kprobe *orig_p, struct kprobe *p)
1245 struct kprobe *ap = orig_p;
1247 /* For preparing optimization, jump_label_text_reserved() is called */
1250 * Get online CPUs to avoid text_mutex deadlock.with stop machine,
1251 * which is invoked by unoptimize_kprobe() in add_new_kprobe()
1254 mutex_lock(&text_mutex);
1256 if (!kprobe_aggrprobe(orig_p)) {
1257 /* If orig_p is not an aggr_kprobe, create new aggr_kprobe. */
1258 ap = alloc_aggr_kprobe(orig_p);
1263 init_aggr_kprobe(ap, orig_p);
1264 } else if (kprobe_unused(ap))
1265 /* This probe is going to die. Rescue it */
1266 reuse_unused_kprobe(ap);
1268 if (kprobe_gone(ap)) {
1270 * Attempting to insert new probe at the same location that
1271 * had a probe in the module vaddr area which already
1272 * freed. So, the instruction slot has already been
1273 * released. We need a new slot for the new probe.
1275 ret = arch_prepare_kprobe(ap);
1278 * Even if fail to allocate new slot, don't need to
1279 * free aggr_probe. It will be used next time, or
1280 * freed by unregister_kprobe.
1284 /* Prepare optimized instructions if possible. */
1285 prepare_optimized_kprobe(ap);
1288 * Clear gone flag to prevent allocating new slot again, and
1289 * set disabled flag because it is not armed yet.
1291 ap->flags = (ap->flags & ~KPROBE_FLAG_GONE)
1292 | KPROBE_FLAG_DISABLED;
1295 /* Copy ap's insn slot to p */
1297 ret = add_new_kprobe(ap, p);
1300 mutex_unlock(&text_mutex);
1302 jump_label_unlock();
1304 if (ret == 0 && kprobe_disabled(ap) && !kprobe_disabled(p)) {
1305 ap->flags &= ~KPROBE_FLAG_DISABLED;
1306 if (!kprobes_all_disarmed)
1307 /* Arm the breakpoint again. */
1313 bool __weak arch_within_kprobe_blacklist(unsigned long addr)
1315 /* The __kprobes marked functions and entry code must not be probed */
1316 return addr >= (unsigned long)__kprobes_text_start &&
1317 addr < (unsigned long)__kprobes_text_end;
1320 static bool within_kprobe_blacklist(unsigned long addr)
1322 struct kprobe_blacklist_entry *ent;
1324 if (arch_within_kprobe_blacklist(addr))
1327 * If there exists a kprobe_blacklist, verify and
1328 * fail any probe registration in the prohibited area
1330 list_for_each_entry(ent, &kprobe_blacklist, list) {
1331 if (addr >= ent->start_addr && addr < ent->end_addr)
1339 * If we have a symbol_name argument, look it up and add the offset field
1340 * to it. This way, we can specify a relative address to a symbol.
1341 * This returns encoded errors if it fails to look up symbol or invalid
1342 * combination of parameters.
1344 static kprobe_opcode_t *kprobe_addr(struct kprobe *p)
1346 kprobe_opcode_t *addr = p->addr;
1348 if ((p->symbol_name && p->addr) ||
1349 (!p->symbol_name && !p->addr))
1352 if (p->symbol_name) {
1353 kprobe_lookup_name(p->symbol_name, addr);
1355 return ERR_PTR(-ENOENT);
1358 addr = (kprobe_opcode_t *)(((char *)addr) + p->offset);
1363 return ERR_PTR(-EINVAL);
1366 /* Check passed kprobe is valid and return kprobe in kprobe_table. */
1367 static struct kprobe *__get_valid_kprobe(struct kprobe *p)
1369 struct kprobe *ap, *list_p;
1371 ap = get_kprobe(p->addr);
1376 list_for_each_entry_rcu(list_p, &ap->list, list)
1378 /* kprobe p is a valid probe */
1386 /* Return error if the kprobe is being re-registered */
1387 static inline int check_kprobe_rereg(struct kprobe *p)
1391 mutex_lock(&kprobe_mutex);
1392 if (__get_valid_kprobe(p))
1394 mutex_unlock(&kprobe_mutex);
1399 static int check_kprobe_address_safe(struct kprobe *p,
1400 struct module **probed_mod)
1403 unsigned long ftrace_addr;
1406 * If the address is located on a ftrace nop, set the
1407 * breakpoint to the following instruction.
1409 ftrace_addr = ftrace_location((unsigned long)p->addr);
1411 #ifdef CONFIG_KPROBES_ON_FTRACE
1412 /* Given address is not on the instruction boundary */
1413 if ((unsigned long)p->addr != ftrace_addr)
1415 p->flags |= KPROBE_FLAG_FTRACE;
1416 #else /* !CONFIG_KPROBES_ON_FTRACE */
1424 /* Ensure it is not in reserved area nor out of text */
1425 if (!kernel_text_address((unsigned long) p->addr) ||
1426 within_kprobe_blacklist((unsigned long) p->addr) ||
1427 jump_label_text_reserved(p->addr, p->addr)) {
1432 /* Check if are we probing a module */
1433 *probed_mod = __module_text_address((unsigned long) p->addr);
1436 * We must hold a refcount of the probed module while updating
1437 * its code to prohibit unexpected unloading.
1439 if (unlikely(!try_module_get(*probed_mod))) {
1445 * If the module freed .init.text, we couldn't insert
1448 if (within_module_init((unsigned long)p->addr, *probed_mod) &&
1449 (*probed_mod)->state != MODULE_STATE_COMING) {
1450 module_put(*probed_mod);
1457 jump_label_unlock();
1462 int register_kprobe(struct kprobe *p)
1465 struct kprobe *old_p;
1466 struct module *probed_mod;
1467 kprobe_opcode_t *addr;
1469 /* Adjust probe address from symbol */
1470 addr = kprobe_addr(p);
1472 return PTR_ERR(addr);
1475 ret = check_kprobe_rereg(p);
1479 /* User can pass only KPROBE_FLAG_DISABLED to register_kprobe */
1480 p->flags &= KPROBE_FLAG_DISABLED;
1482 INIT_LIST_HEAD(&p->list);
1484 ret = check_kprobe_address_safe(p, &probed_mod);
1488 mutex_lock(&kprobe_mutex);
1490 old_p = get_kprobe(p->addr);
1492 /* Since this may unoptimize old_p, locking text_mutex. */
1493 ret = register_aggr_kprobe(old_p, p);
1497 mutex_lock(&text_mutex); /* Avoiding text modification */
1498 ret = prepare_kprobe(p);
1499 mutex_unlock(&text_mutex);
1503 INIT_HLIST_NODE(&p->hlist);
1504 hlist_add_head_rcu(&p->hlist,
1505 &kprobe_table[hash_ptr(p->addr, KPROBE_HASH_BITS)]);
1507 if (!kprobes_all_disarmed && !kprobe_disabled(p))
1510 /* Try to optimize kprobe */
1511 try_to_optimize_kprobe(p);
1514 mutex_unlock(&kprobe_mutex);
1517 module_put(probed_mod);
1521 EXPORT_SYMBOL_GPL(register_kprobe);
1523 /* Check if all probes on the aggrprobe are disabled */
1524 static int aggr_kprobe_disabled(struct kprobe *ap)
1528 list_for_each_entry_rcu(kp, &ap->list, list)
1529 if (!kprobe_disabled(kp))
1531 * There is an active probe on the list.
1532 * We can't disable this ap.
1539 /* Disable one kprobe: Make sure called under kprobe_mutex is locked */
1540 static struct kprobe *__disable_kprobe(struct kprobe *p)
1542 struct kprobe *orig_p;
1544 /* Get an original kprobe for return */
1545 orig_p = __get_valid_kprobe(p);
1546 if (unlikely(orig_p == NULL))
1549 if (!kprobe_disabled(p)) {
1550 /* Disable probe if it is a child probe */
1552 p->flags |= KPROBE_FLAG_DISABLED;
1554 /* Try to disarm and disable this/parent probe */
1555 if (p == orig_p || aggr_kprobe_disabled(orig_p)) {
1556 disarm_kprobe(orig_p, true);
1557 orig_p->flags |= KPROBE_FLAG_DISABLED;
1565 * Unregister a kprobe without a scheduler synchronization.
1567 static int __unregister_kprobe_top(struct kprobe *p)
1569 struct kprobe *ap, *list_p;
1571 /* Disable kprobe. This will disarm it if needed. */
1572 ap = __disable_kprobe(p);
1578 * This probe is an independent(and non-optimized) kprobe
1579 * (not an aggrprobe). Remove from the hash list.
1583 /* Following process expects this probe is an aggrprobe */
1584 WARN_ON(!kprobe_aggrprobe(ap));
1586 if (list_is_singular(&ap->list) && kprobe_disarmed(ap))
1588 * !disarmed could be happen if the probe is under delayed
1593 /* If disabling probe has special handlers, update aggrprobe */
1594 if (p->break_handler && !kprobe_gone(p))
1595 ap->break_handler = NULL;
1596 if (p->post_handler && !kprobe_gone(p)) {
1597 list_for_each_entry_rcu(list_p, &ap->list, list) {
1598 if ((list_p != p) && (list_p->post_handler))
1601 ap->post_handler = NULL;
1605 * Remove from the aggrprobe: this path will do nothing in
1606 * __unregister_kprobe_bottom().
1608 list_del_rcu(&p->list);
1609 if (!kprobe_disabled(ap) && !kprobes_all_disarmed)
1611 * Try to optimize this probe again, because post
1612 * handler may have been changed.
1614 optimize_kprobe(ap);
1619 BUG_ON(!kprobe_disarmed(ap));
1620 hlist_del_rcu(&ap->hlist);
1624 static void __unregister_kprobe_bottom(struct kprobe *p)
1628 if (list_empty(&p->list))
1629 /* This is an independent kprobe */
1630 arch_remove_kprobe(p);
1631 else if (list_is_singular(&p->list)) {
1632 /* This is the last child of an aggrprobe */
1633 ap = list_entry(p->list.next, struct kprobe, list);
1635 free_aggr_kprobe(ap);
1637 /* Otherwise, do nothing. */
1640 int register_kprobes(struct kprobe **kps, int num)
1646 for (i = 0; i < num; i++) {
1647 ret = register_kprobe(kps[i]);
1650 unregister_kprobes(kps, i);
1656 EXPORT_SYMBOL_GPL(register_kprobes);
1658 void unregister_kprobe(struct kprobe *p)
1660 unregister_kprobes(&p, 1);
1662 EXPORT_SYMBOL_GPL(unregister_kprobe);
1664 void unregister_kprobes(struct kprobe **kps, int num)
1670 mutex_lock(&kprobe_mutex);
1671 for (i = 0; i < num; i++)
1672 if (__unregister_kprobe_top(kps[i]) < 0)
1673 kps[i]->addr = NULL;
1674 mutex_unlock(&kprobe_mutex);
1676 synchronize_sched();
1677 for (i = 0; i < num; i++)
1679 __unregister_kprobe_bottom(kps[i]);
1681 EXPORT_SYMBOL_GPL(unregister_kprobes);
1683 static struct notifier_block kprobe_exceptions_nb = {
1684 .notifier_call = kprobe_exceptions_notify,
1685 .priority = 0x7fffffff /* we need to be notified first */
1688 unsigned long __weak arch_deref_entry_point(void *entry)
1690 return (unsigned long)entry;
1693 int register_jprobes(struct jprobe **jps, int num)
1700 for (i = 0; i < num; i++) {
1701 unsigned long addr, offset;
1703 addr = arch_deref_entry_point(jp->entry);
1705 /* Verify probepoint is a function entry point */
1706 if (kallsyms_lookup_size_offset(addr, NULL, &offset) &&
1708 jp->kp.pre_handler = setjmp_pre_handler;
1709 jp->kp.break_handler = longjmp_break_handler;
1710 ret = register_kprobe(&jp->kp);
1716 unregister_jprobes(jps, i);
1722 EXPORT_SYMBOL_GPL(register_jprobes);
1724 int register_jprobe(struct jprobe *jp)
1726 return register_jprobes(&jp, 1);
1728 EXPORT_SYMBOL_GPL(register_jprobe);
1730 void unregister_jprobe(struct jprobe *jp)
1732 unregister_jprobes(&jp, 1);
1734 EXPORT_SYMBOL_GPL(unregister_jprobe);
1736 void unregister_jprobes(struct jprobe **jps, int num)
1742 mutex_lock(&kprobe_mutex);
1743 for (i = 0; i < num; i++)
1744 if (__unregister_kprobe_top(&jps[i]->kp) < 0)
1745 jps[i]->kp.addr = NULL;
1746 mutex_unlock(&kprobe_mutex);
1748 synchronize_sched();
1749 for (i = 0; i < num; i++) {
1750 if (jps[i]->kp.addr)
1751 __unregister_kprobe_bottom(&jps[i]->kp);
1754 EXPORT_SYMBOL_GPL(unregister_jprobes);
1756 #ifdef CONFIG_KRETPROBES
1758 * This kprobe pre_handler is registered with every kretprobe. When probe
1759 * hits it will set up the return probe.
1761 static int __kprobes pre_handler_kretprobe(struct kprobe *p,
1762 struct pt_regs *regs)
1764 struct kretprobe *rp = container_of(p, struct kretprobe, kp);
1765 unsigned long hash, flags = 0;
1766 struct kretprobe_instance *ri;
1768 /*TODO: consider to only swap the RA after the last pre_handler fired */
1769 hash = hash_ptr(current, KPROBE_HASH_BITS);
1770 raw_spin_lock_irqsave(&rp->lock, flags);
1771 if (!hlist_empty(&rp->free_instances)) {
1772 ri = hlist_entry(rp->free_instances.first,
1773 struct kretprobe_instance, hlist);
1774 hlist_del(&ri->hlist);
1775 raw_spin_unlock_irqrestore(&rp->lock, flags);
1780 if (rp->entry_handler && rp->entry_handler(ri, regs)) {
1781 raw_spin_lock_irqsave(&rp->lock, flags);
1782 hlist_add_head(&ri->hlist, &rp->free_instances);
1783 raw_spin_unlock_irqrestore(&rp->lock, flags);
1787 arch_prepare_kretprobe(ri, regs);
1789 /* XXX(hch): why is there no hlist_move_head? */
1790 INIT_HLIST_NODE(&ri->hlist);
1791 kretprobe_table_lock(hash, &flags);
1792 hlist_add_head(&ri->hlist, &kretprobe_inst_table[hash]);
1793 kretprobe_table_unlock(hash, &flags);
1796 raw_spin_unlock_irqrestore(&rp->lock, flags);
1801 int register_kretprobe(struct kretprobe *rp)
1804 struct kretprobe_instance *inst;
1808 if (kretprobe_blacklist_size) {
1809 addr = kprobe_addr(&rp->kp);
1811 return PTR_ERR(addr);
1813 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
1814 if (kretprobe_blacklist[i].addr == addr)
1819 rp->kp.pre_handler = pre_handler_kretprobe;
1820 rp->kp.post_handler = NULL;
1821 rp->kp.fault_handler = NULL;
1822 rp->kp.break_handler = NULL;
1824 /* Pre-allocate memory for max kretprobe instances */
1825 if (rp->maxactive <= 0) {
1826 #ifdef CONFIG_PREEMPT
1827 rp->maxactive = max_t(unsigned int, 10, 2*num_possible_cpus());
1829 rp->maxactive = num_possible_cpus();
1832 raw_spin_lock_init(&rp->lock);
1833 INIT_HLIST_HEAD(&rp->free_instances);
1834 for (i = 0; i < rp->maxactive; i++) {
1835 inst = kmalloc(sizeof(struct kretprobe_instance) +
1836 rp->data_size, GFP_KERNEL);
1841 INIT_HLIST_NODE(&inst->hlist);
1842 hlist_add_head(&inst->hlist, &rp->free_instances);
1846 /* Establish function entry probe point */
1847 ret = register_kprobe(&rp->kp);
1852 EXPORT_SYMBOL_GPL(register_kretprobe);
1854 int register_kretprobes(struct kretprobe **rps, int num)
1860 for (i = 0; i < num; i++) {
1861 ret = register_kretprobe(rps[i]);
1864 unregister_kretprobes(rps, i);
1870 EXPORT_SYMBOL_GPL(register_kretprobes);
1872 void unregister_kretprobe(struct kretprobe *rp)
1874 unregister_kretprobes(&rp, 1);
1876 EXPORT_SYMBOL_GPL(unregister_kretprobe);
1878 void unregister_kretprobes(struct kretprobe **rps, int num)
1884 mutex_lock(&kprobe_mutex);
1885 for (i = 0; i < num; i++)
1886 if (__unregister_kprobe_top(&rps[i]->kp) < 0)
1887 rps[i]->kp.addr = NULL;
1888 mutex_unlock(&kprobe_mutex);
1890 synchronize_sched();
1891 for (i = 0; i < num; i++) {
1892 if (rps[i]->kp.addr) {
1893 __unregister_kprobe_bottom(&rps[i]->kp);
1894 cleanup_rp_inst(rps[i]);
1898 EXPORT_SYMBOL_GPL(unregister_kretprobes);
1900 #else /* CONFIG_KRETPROBES */
1901 int register_kretprobe(struct kretprobe *rp)
1905 EXPORT_SYMBOL_GPL(register_kretprobe);
1907 int register_kretprobes(struct kretprobe **rps, int num)
1911 EXPORT_SYMBOL_GPL(register_kretprobes);
1913 void unregister_kretprobe(struct kretprobe *rp)
1916 EXPORT_SYMBOL_GPL(unregister_kretprobe);
1918 void unregister_kretprobes(struct kretprobe **rps, int num)
1921 EXPORT_SYMBOL_GPL(unregister_kretprobes);
1923 static int __kprobes pre_handler_kretprobe(struct kprobe *p,
1924 struct pt_regs *regs)
1929 #endif /* CONFIG_KRETPROBES */
1931 /* Set the kprobe gone and remove its instruction buffer. */
1932 static void kill_kprobe(struct kprobe *p)
1936 p->flags |= KPROBE_FLAG_GONE;
1937 if (kprobe_aggrprobe(p)) {
1939 * If this is an aggr_kprobe, we have to list all the
1940 * chained probes and mark them GONE.
1942 list_for_each_entry_rcu(kp, &p->list, list)
1943 kp->flags |= KPROBE_FLAG_GONE;
1944 p->post_handler = NULL;
1945 p->break_handler = NULL;
1946 kill_optimized_kprobe(p);
1949 * Here, we can remove insn_slot safely, because no thread calls
1950 * the original probed function (which will be freed soon) any more.
1952 arch_remove_kprobe(p);
1955 /* Disable one kprobe */
1956 int disable_kprobe(struct kprobe *kp)
1960 mutex_lock(&kprobe_mutex);
1962 /* Disable this kprobe */
1963 if (__disable_kprobe(kp) == NULL)
1966 mutex_unlock(&kprobe_mutex);
1969 EXPORT_SYMBOL_GPL(disable_kprobe);
1971 /* Enable one kprobe */
1972 int enable_kprobe(struct kprobe *kp)
1977 mutex_lock(&kprobe_mutex);
1979 /* Check whether specified probe is valid. */
1980 p = __get_valid_kprobe(kp);
1981 if (unlikely(p == NULL)) {
1986 if (kprobe_gone(kp)) {
1987 /* This kprobe has gone, we couldn't enable it. */
1993 kp->flags &= ~KPROBE_FLAG_DISABLED;
1995 if (!kprobes_all_disarmed && kprobe_disabled(p)) {
1996 p->flags &= ~KPROBE_FLAG_DISABLED;
2000 mutex_unlock(&kprobe_mutex);
2003 EXPORT_SYMBOL_GPL(enable_kprobe);
2005 void __kprobes dump_kprobe(struct kprobe *kp)
2007 printk(KERN_WARNING "Dumping kprobe:\n");
2008 printk(KERN_WARNING "Name: %s\nAddress: %p\nOffset: %x\n",
2009 kp->symbol_name, kp->addr, kp->offset);
2013 * Lookup and populate the kprobe_blacklist.
2015 * Unlike the kretprobe blacklist, we'll need to determine
2016 * the range of addresses that belong to the said functions,
2017 * since a kprobe need not necessarily be at the beginning
2020 static int __init populate_kprobe_blacklist(unsigned long *start,
2023 unsigned long *iter;
2024 struct kprobe_blacklist_entry *ent;
2025 unsigned long offset = 0, size = 0;
2027 for (iter = start; iter < end; iter++) {
2028 if (!kallsyms_lookup_size_offset(*iter, &size, &offset)) {
2029 pr_err("Failed to find blacklist %p\n", (void *)*iter);
2033 ent = kmalloc(sizeof(*ent), GFP_KERNEL);
2036 ent->start_addr = *iter;
2037 ent->end_addr = *iter + size;
2038 INIT_LIST_HEAD(&ent->list);
2039 list_add_tail(&ent->list, &kprobe_blacklist);
2044 /* Module notifier call back, checking kprobes on the module */
2045 static int kprobes_module_callback(struct notifier_block *nb,
2046 unsigned long val, void *data)
2048 struct module *mod = data;
2049 struct hlist_head *head;
2052 int checkcore = (val == MODULE_STATE_GOING);
2054 if (val != MODULE_STATE_GOING && val != MODULE_STATE_LIVE)
2058 * When MODULE_STATE_GOING was notified, both of module .text and
2059 * .init.text sections would be freed. When MODULE_STATE_LIVE was
2060 * notified, only .init.text section would be freed. We need to
2061 * disable kprobes which have been inserted in the sections.
2063 mutex_lock(&kprobe_mutex);
2064 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2065 head = &kprobe_table[i];
2066 hlist_for_each_entry_rcu(p, head, hlist)
2067 if (within_module_init((unsigned long)p->addr, mod) ||
2069 within_module_core((unsigned long)p->addr, mod))) {
2071 * The vaddr this probe is installed will soon
2072 * be vfreed buy not synced to disk. Hence,
2073 * disarming the breakpoint isn't needed.
2078 mutex_unlock(&kprobe_mutex);
2082 static struct notifier_block kprobe_module_nb = {
2083 .notifier_call = kprobes_module_callback,
2087 /* Markers of _kprobe_blacklist section */
2088 extern unsigned long __start_kprobe_blacklist[];
2089 extern unsigned long __stop_kprobe_blacklist[];
2091 static int __init init_kprobes(void)
2095 /* FIXME allocate the probe table, currently defined statically */
2096 /* initialize all list heads */
2097 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2098 INIT_HLIST_HEAD(&kprobe_table[i]);
2099 INIT_HLIST_HEAD(&kretprobe_inst_table[i]);
2100 raw_spin_lock_init(&(kretprobe_table_locks[i].lock));
2103 err = populate_kprobe_blacklist(__start_kprobe_blacklist,
2104 __stop_kprobe_blacklist);
2106 pr_err("kprobes: failed to populate blacklist: %d\n", err);
2107 pr_err("Please take care of using kprobes.\n");
2110 if (kretprobe_blacklist_size) {
2111 /* lookup the function address from its name */
2112 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
2113 kprobe_lookup_name(kretprobe_blacklist[i].name,
2114 kretprobe_blacklist[i].addr);
2115 if (!kretprobe_blacklist[i].addr)
2116 printk("kretprobe: lookup failed: %s\n",
2117 kretprobe_blacklist[i].name);
2121 #if defined(CONFIG_OPTPROBES)
2122 #if defined(__ARCH_WANT_KPROBES_INSN_SLOT)
2123 /* Init kprobe_optinsn_slots */
2124 kprobe_optinsn_slots.insn_size = MAX_OPTINSN_SIZE;
2126 /* By default, kprobes can be optimized */
2127 kprobes_allow_optimization = true;
2130 /* By default, kprobes are armed */
2131 kprobes_all_disarmed = false;
2133 err = arch_init_kprobes();
2135 err = register_die_notifier(&kprobe_exceptions_nb);
2137 err = register_module_notifier(&kprobe_module_nb);
2139 kprobes_initialized = (err == 0);
2146 #ifdef CONFIG_DEBUG_FS
2147 static void report_probe(struct seq_file *pi, struct kprobe *p,
2148 const char *sym, int offset, char *modname, struct kprobe *pp)
2152 if (p->pre_handler == pre_handler_kretprobe)
2154 else if (p->pre_handler == setjmp_pre_handler)
2160 seq_printf(pi, "%p %s %s+0x%x %s ",
2161 p->addr, kprobe_type, sym, offset,
2162 (modname ? modname : " "));
2164 seq_printf(pi, "%p %s %p ",
2165 p->addr, kprobe_type, p->addr);
2169 seq_printf(pi, "%s%s%s%s\n",
2170 (kprobe_gone(p) ? "[GONE]" : ""),
2171 ((kprobe_disabled(p) && !kprobe_gone(p)) ? "[DISABLED]" : ""),
2172 (kprobe_optimized(pp) ? "[OPTIMIZED]" : ""),
2173 (kprobe_ftrace(pp) ? "[FTRACE]" : ""));
2176 static void *kprobe_seq_start(struct seq_file *f, loff_t *pos)
2178 return (*pos < KPROBE_TABLE_SIZE) ? pos : NULL;
2181 static void *kprobe_seq_next(struct seq_file *f, void *v, loff_t *pos)
2184 if (*pos >= KPROBE_TABLE_SIZE)
2189 static void kprobe_seq_stop(struct seq_file *f, void *v)
2194 static int show_kprobe_addr(struct seq_file *pi, void *v)
2196 struct hlist_head *head;
2197 struct kprobe *p, *kp;
2198 const char *sym = NULL;
2199 unsigned int i = *(loff_t *) v;
2200 unsigned long offset = 0;
2201 char *modname, namebuf[KSYM_NAME_LEN];
2203 head = &kprobe_table[i];
2205 hlist_for_each_entry_rcu(p, head, hlist) {
2206 sym = kallsyms_lookup((unsigned long)p->addr, NULL,
2207 &offset, &modname, namebuf);
2208 if (kprobe_aggrprobe(p)) {
2209 list_for_each_entry_rcu(kp, &p->list, list)
2210 report_probe(pi, kp, sym, offset, modname, p);
2212 report_probe(pi, p, sym, offset, modname, NULL);
2218 static const struct seq_operations kprobes_seq_ops = {
2219 .start = kprobe_seq_start,
2220 .next = kprobe_seq_next,
2221 .stop = kprobe_seq_stop,
2222 .show = show_kprobe_addr
2225 static int kprobes_open(struct inode *inode, struct file *filp)
2227 return seq_open(filp, &kprobes_seq_ops);
2230 static const struct file_operations debugfs_kprobes_operations = {
2231 .open = kprobes_open,
2233 .llseek = seq_lseek,
2234 .release = seq_release,
2237 static void arm_all_kprobes(void)
2239 struct hlist_head *head;
2243 mutex_lock(&kprobe_mutex);
2245 /* If kprobes are armed, just return */
2246 if (!kprobes_all_disarmed)
2247 goto already_enabled;
2249 /* Arming kprobes doesn't optimize kprobe itself */
2250 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2251 head = &kprobe_table[i];
2252 hlist_for_each_entry_rcu(p, head, hlist)
2253 if (!kprobe_disabled(p))
2257 kprobes_all_disarmed = false;
2258 printk(KERN_INFO "Kprobes globally enabled\n");
2261 mutex_unlock(&kprobe_mutex);
2265 static void disarm_all_kprobes(void)
2267 struct hlist_head *head;
2271 mutex_lock(&kprobe_mutex);
2273 /* If kprobes are already disarmed, just return */
2274 if (kprobes_all_disarmed) {
2275 mutex_unlock(&kprobe_mutex);
2279 kprobes_all_disarmed = true;
2280 printk(KERN_INFO "Kprobes globally disabled\n");
2282 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2283 head = &kprobe_table[i];
2284 hlist_for_each_entry_rcu(p, head, hlist) {
2285 if (!arch_trampoline_kprobe(p) && !kprobe_disabled(p))
2286 disarm_kprobe(p, false);
2289 mutex_unlock(&kprobe_mutex);
2291 /* Wait for disarming all kprobes by optimizer */
2292 wait_for_kprobe_optimizer();
2296 * XXX: The debugfs bool file interface doesn't allow for callbacks
2297 * when the bool state is switched. We can reuse that facility when
2300 static ssize_t read_enabled_file_bool(struct file *file,
2301 char __user *user_buf, size_t count, loff_t *ppos)
2305 if (!kprobes_all_disarmed)
2311 return simple_read_from_buffer(user_buf, count, ppos, buf, 2);
2314 static ssize_t write_enabled_file_bool(struct file *file,
2315 const char __user *user_buf, size_t count, loff_t *ppos)
2320 buf_size = min(count, (sizeof(buf)-1));
2321 if (copy_from_user(buf, user_buf, buf_size))
2324 buf[buf_size] = '\0';
2334 disarm_all_kprobes();
2343 static const struct file_operations fops_kp = {
2344 .read = read_enabled_file_bool,
2345 .write = write_enabled_file_bool,
2346 .llseek = default_llseek,
2349 static int __init debugfs_kprobe_init(void)
2351 struct dentry *dir, *file;
2352 unsigned int value = 1;
2354 dir = debugfs_create_dir("kprobes", NULL);
2358 file = debugfs_create_file("list", 0444, dir, NULL,
2359 &debugfs_kprobes_operations);
2361 debugfs_remove(dir);
2365 file = debugfs_create_file("enabled", 0600, dir,
2368 debugfs_remove(dir);
2375 late_initcall(debugfs_kprobe_init);
2376 #endif /* CONFIG_DEBUG_FS */
2378 module_init(init_kprobes);
2380 /* defined in arch/.../kernel/kprobes.c */
2381 EXPORT_SYMBOL_GPL(jprobe_return);