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/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 void __weak free_insn_page(void *page)
130 module_memfree(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 *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) {
317 NOKPROBE_SYMBOL(get_kprobe);
319 static int aggr_pre_handler(struct kprobe *p, struct pt_regs *regs);
321 /* Return true if the kprobe is an aggregator */
322 static inline int kprobe_aggrprobe(struct kprobe *p)
324 return p->pre_handler == aggr_pre_handler;
327 /* Return true(!0) if the kprobe is unused */
328 static inline int kprobe_unused(struct kprobe *p)
330 return kprobe_aggrprobe(p) && kprobe_disabled(p) &&
331 list_empty(&p->list);
335 * Keep all fields in the kprobe consistent
337 static inline void copy_kprobe(struct kprobe *ap, struct kprobe *p)
339 memcpy(&p->opcode, &ap->opcode, sizeof(kprobe_opcode_t));
340 memcpy(&p->ainsn, &ap->ainsn, sizeof(struct arch_specific_insn));
343 #ifdef CONFIG_OPTPROBES
344 /* NOTE: change this value only with kprobe_mutex held */
345 static bool kprobes_allow_optimization;
348 * Call all pre_handler on the list, but ignores its return value.
349 * This must be called from arch-dep optimized caller.
351 void opt_pre_handler(struct kprobe *p, struct pt_regs *regs)
355 list_for_each_entry_rcu(kp, &p->list, list) {
356 if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
357 set_kprobe_instance(kp);
358 kp->pre_handler(kp, regs);
360 reset_kprobe_instance();
363 NOKPROBE_SYMBOL(opt_pre_handler);
365 /* Free optimized instructions and optimized_kprobe */
366 static void free_aggr_kprobe(struct kprobe *p)
368 struct optimized_kprobe *op;
370 op = container_of(p, struct optimized_kprobe, kp);
371 arch_remove_optimized_kprobe(op);
372 arch_remove_kprobe(p);
376 /* Return true(!0) if the kprobe is ready for optimization. */
377 static inline int kprobe_optready(struct kprobe *p)
379 struct optimized_kprobe *op;
381 if (kprobe_aggrprobe(p)) {
382 op = container_of(p, struct optimized_kprobe, kp);
383 return arch_prepared_optinsn(&op->optinsn);
389 /* Return true(!0) if the kprobe is disarmed. Note: p must be on hash list */
390 static inline int kprobe_disarmed(struct kprobe *p)
392 struct optimized_kprobe *op;
394 /* If kprobe is not aggr/opt probe, just return kprobe is disabled */
395 if (!kprobe_aggrprobe(p))
396 return kprobe_disabled(p);
398 op = container_of(p, struct optimized_kprobe, kp);
400 return kprobe_disabled(p) && list_empty(&op->list);
403 /* Return true(!0) if the probe is queued on (un)optimizing lists */
404 static int kprobe_queued(struct kprobe *p)
406 struct optimized_kprobe *op;
408 if (kprobe_aggrprobe(p)) {
409 op = container_of(p, struct optimized_kprobe, kp);
410 if (!list_empty(&op->list))
417 * Return an optimized kprobe whose optimizing code replaces
418 * instructions including addr (exclude breakpoint).
420 static struct kprobe *get_optimized_kprobe(unsigned long addr)
423 struct kprobe *p = NULL;
424 struct optimized_kprobe *op;
426 /* Don't check i == 0, since that is a breakpoint case. */
427 for (i = 1; !p && i < MAX_OPTIMIZED_LENGTH; i++)
428 p = get_kprobe((void *)(addr - i));
430 if (p && kprobe_optready(p)) {
431 op = container_of(p, struct optimized_kprobe, kp);
432 if (arch_within_optimized_kprobe(op, addr))
439 /* Optimization staging list, protected by kprobe_mutex */
440 static LIST_HEAD(optimizing_list);
441 static LIST_HEAD(unoptimizing_list);
442 static LIST_HEAD(freeing_list);
444 static void kprobe_optimizer(struct work_struct *work);
445 static DECLARE_DELAYED_WORK(optimizing_work, kprobe_optimizer);
446 #define OPTIMIZE_DELAY 5
449 * Optimize (replace a breakpoint with a jump) kprobes listed on
452 static void do_optimize_kprobes(void)
454 /* Optimization never be done when disarmed */
455 if (kprobes_all_disarmed || !kprobes_allow_optimization ||
456 list_empty(&optimizing_list))
460 * The optimization/unoptimization refers online_cpus via
461 * stop_machine() and cpu-hotplug modifies online_cpus.
462 * And same time, text_mutex will be held in cpu-hotplug and here.
463 * This combination can cause a deadlock (cpu-hotplug try to lock
464 * text_mutex but stop_machine can not be done because online_cpus
466 * To avoid this deadlock, we need to call get_online_cpus()
467 * for preventing cpu-hotplug outside of text_mutex locking.
470 mutex_lock(&text_mutex);
471 arch_optimize_kprobes(&optimizing_list);
472 mutex_unlock(&text_mutex);
477 * Unoptimize (replace a jump with a breakpoint and remove the breakpoint
478 * if need) kprobes listed on unoptimizing_list.
480 static void do_unoptimize_kprobes(void)
482 struct optimized_kprobe *op, *tmp;
484 /* Unoptimization must be done anytime */
485 if (list_empty(&unoptimizing_list))
488 /* Ditto to do_optimize_kprobes */
490 mutex_lock(&text_mutex);
491 arch_unoptimize_kprobes(&unoptimizing_list, &freeing_list);
492 /* Loop free_list for disarming */
493 list_for_each_entry_safe(op, tmp, &freeing_list, list) {
494 /* Disarm probes if marked disabled */
495 if (kprobe_disabled(&op->kp))
496 arch_disarm_kprobe(&op->kp);
497 if (kprobe_unused(&op->kp)) {
499 * Remove unused probes from hash list. After waiting
500 * for synchronization, these probes are reclaimed.
501 * (reclaiming is done by do_free_cleaned_kprobes.)
503 hlist_del_rcu(&op->kp.hlist);
505 list_del_init(&op->list);
507 mutex_unlock(&text_mutex);
511 /* Reclaim all kprobes on the free_list */
512 static void do_free_cleaned_kprobes(void)
514 struct optimized_kprobe *op, *tmp;
516 list_for_each_entry_safe(op, tmp, &freeing_list, list) {
517 BUG_ON(!kprobe_unused(&op->kp));
518 list_del_init(&op->list);
519 free_aggr_kprobe(&op->kp);
523 /* Start optimizer after OPTIMIZE_DELAY passed */
524 static void kick_kprobe_optimizer(void)
526 schedule_delayed_work(&optimizing_work, OPTIMIZE_DELAY);
529 /* Kprobe jump optimizer */
530 static void kprobe_optimizer(struct work_struct *work)
532 mutex_lock(&kprobe_mutex);
533 /* Lock modules while optimizing kprobes */
534 mutex_lock(&module_mutex);
537 * Step 1: Unoptimize kprobes and collect cleaned (unused and disarmed)
538 * kprobes before waiting for quiesence period.
540 do_unoptimize_kprobes();
543 * Step 2: Wait for quiesence period to ensure all running interrupts
544 * are done. Because optprobe may modify multiple instructions
545 * there is a chance that Nth instruction is interrupted. In that
546 * case, running interrupt can return to 2nd-Nth byte of jump
547 * instruction. This wait is for avoiding it.
551 /* Step 3: Optimize kprobes after quiesence period */
552 do_optimize_kprobes();
554 /* Step 4: Free cleaned kprobes after quiesence period */
555 do_free_cleaned_kprobes();
557 mutex_unlock(&module_mutex);
558 mutex_unlock(&kprobe_mutex);
560 /* Step 5: Kick optimizer again if needed */
561 if (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list))
562 kick_kprobe_optimizer();
565 /* Wait for completing optimization and unoptimization */
566 void wait_for_kprobe_optimizer(void)
568 mutex_lock(&kprobe_mutex);
570 while (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list)) {
571 mutex_unlock(&kprobe_mutex);
573 /* this will also make optimizing_work execute immmediately */
574 flush_delayed_work(&optimizing_work);
575 /* @optimizing_work might not have been queued yet, relax */
578 mutex_lock(&kprobe_mutex);
581 mutex_unlock(&kprobe_mutex);
584 /* Optimize kprobe if p is ready to be optimized */
585 static void optimize_kprobe(struct kprobe *p)
587 struct optimized_kprobe *op;
589 /* Check if the kprobe is disabled or not ready for optimization. */
590 if (!kprobe_optready(p) || !kprobes_allow_optimization ||
591 (kprobe_disabled(p) || kprobes_all_disarmed))
594 /* Both of break_handler and post_handler are not supported. */
595 if (p->break_handler || p->post_handler)
598 op = container_of(p, struct optimized_kprobe, kp);
600 /* Check there is no other kprobes at the optimized instructions */
601 if (arch_check_optimized_kprobe(op) < 0)
604 /* Check if it is already optimized. */
605 if (op->kp.flags & KPROBE_FLAG_OPTIMIZED)
607 op->kp.flags |= KPROBE_FLAG_OPTIMIZED;
609 if (!list_empty(&op->list))
610 /* This is under unoptimizing. Just dequeue the probe */
611 list_del_init(&op->list);
613 list_add(&op->list, &optimizing_list);
614 kick_kprobe_optimizer();
618 /* Short cut to direct unoptimizing */
619 static void force_unoptimize_kprobe(struct optimized_kprobe *op)
622 arch_unoptimize_kprobe(op);
624 if (kprobe_disabled(&op->kp))
625 arch_disarm_kprobe(&op->kp);
628 /* Unoptimize a kprobe if p is optimized */
629 static void unoptimize_kprobe(struct kprobe *p, bool force)
631 struct optimized_kprobe *op;
633 if (!kprobe_aggrprobe(p) || kprobe_disarmed(p))
634 return; /* This is not an optprobe nor optimized */
636 op = container_of(p, struct optimized_kprobe, kp);
637 if (!kprobe_optimized(p)) {
638 /* Unoptimized or unoptimizing case */
639 if (force && !list_empty(&op->list)) {
641 * Only if this is unoptimizing kprobe and forced,
642 * forcibly unoptimize it. (No need to unoptimize
643 * unoptimized kprobe again :)
645 list_del_init(&op->list);
646 force_unoptimize_kprobe(op);
651 op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
652 if (!list_empty(&op->list)) {
653 /* Dequeue from the optimization queue */
654 list_del_init(&op->list);
657 /* Optimized kprobe case */
659 /* Forcibly update the code: this is a special case */
660 force_unoptimize_kprobe(op);
662 list_add(&op->list, &unoptimizing_list);
663 kick_kprobe_optimizer();
667 /* Cancel unoptimizing for reusing */
668 static int reuse_unused_kprobe(struct kprobe *ap)
670 struct optimized_kprobe *op;
672 BUG_ON(!kprobe_unused(ap));
674 * Unused kprobe MUST be on the way of delayed unoptimizing (means
675 * there is still a relative jump) and disabled.
677 op = container_of(ap, struct optimized_kprobe, kp);
678 if (unlikely(list_empty(&op->list)))
679 printk(KERN_WARNING "Warning: found a stray unused "
680 "aggrprobe@%p\n", ap->addr);
681 /* Enable the probe again */
682 ap->flags &= ~KPROBE_FLAG_DISABLED;
683 /* Optimize it again (remove from op->list) */
684 if (!kprobe_optready(ap))
691 /* Remove optimized instructions */
692 static void kill_optimized_kprobe(struct kprobe *p)
694 struct optimized_kprobe *op;
696 op = container_of(p, struct optimized_kprobe, kp);
697 if (!list_empty(&op->list))
698 /* Dequeue from the (un)optimization queue */
699 list_del_init(&op->list);
700 op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
702 if (kprobe_unused(p)) {
703 /* Enqueue if it is unused */
704 list_add(&op->list, &freeing_list);
706 * Remove unused probes from the hash list. After waiting
707 * for synchronization, this probe is reclaimed.
708 * (reclaiming is done by do_free_cleaned_kprobes().)
710 hlist_del_rcu(&op->kp.hlist);
713 /* Don't touch the code, because it is already freed. */
714 arch_remove_optimized_kprobe(op);
717 /* Try to prepare optimized instructions */
718 static void prepare_optimized_kprobe(struct kprobe *p)
720 struct optimized_kprobe *op;
722 op = container_of(p, struct optimized_kprobe, kp);
723 arch_prepare_optimized_kprobe(op, p);
726 /* Allocate new optimized_kprobe and try to prepare optimized instructions */
727 static struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
729 struct optimized_kprobe *op;
731 op = kzalloc(sizeof(struct optimized_kprobe), GFP_KERNEL);
735 INIT_LIST_HEAD(&op->list);
736 op->kp.addr = p->addr;
737 arch_prepare_optimized_kprobe(op, p);
742 static void init_aggr_kprobe(struct kprobe *ap, struct kprobe *p);
745 * Prepare an optimized_kprobe and optimize it
746 * NOTE: p must be a normal registered kprobe
748 static void try_to_optimize_kprobe(struct kprobe *p)
751 struct optimized_kprobe *op;
753 /* Impossible to optimize ftrace-based kprobe */
754 if (kprobe_ftrace(p))
757 /* For preparing optimization, jump_label_text_reserved() is called */
759 mutex_lock(&text_mutex);
761 ap = alloc_aggr_kprobe(p);
765 op = container_of(ap, struct optimized_kprobe, kp);
766 if (!arch_prepared_optinsn(&op->optinsn)) {
767 /* If failed to setup optimizing, fallback to kprobe */
768 arch_remove_optimized_kprobe(op);
773 init_aggr_kprobe(ap, p);
774 optimize_kprobe(ap); /* This just kicks optimizer thread */
777 mutex_unlock(&text_mutex);
782 static void optimize_all_kprobes(void)
784 struct hlist_head *head;
788 mutex_lock(&kprobe_mutex);
789 /* If optimization is already allowed, just return */
790 if (kprobes_allow_optimization)
793 kprobes_allow_optimization = true;
794 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
795 head = &kprobe_table[i];
796 hlist_for_each_entry_rcu(p, head, hlist)
797 if (!kprobe_disabled(p))
800 printk(KERN_INFO "Kprobes globally optimized\n");
802 mutex_unlock(&kprobe_mutex);
805 static void unoptimize_all_kprobes(void)
807 struct hlist_head *head;
811 mutex_lock(&kprobe_mutex);
812 /* If optimization is already prohibited, just return */
813 if (!kprobes_allow_optimization) {
814 mutex_unlock(&kprobe_mutex);
818 kprobes_allow_optimization = false;
819 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
820 head = &kprobe_table[i];
821 hlist_for_each_entry_rcu(p, head, hlist) {
822 if (!kprobe_disabled(p))
823 unoptimize_kprobe(p, false);
826 mutex_unlock(&kprobe_mutex);
828 /* Wait for unoptimizing completion */
829 wait_for_kprobe_optimizer();
830 printk(KERN_INFO "Kprobes globally unoptimized\n");
833 static DEFINE_MUTEX(kprobe_sysctl_mutex);
834 int sysctl_kprobes_optimization;
835 int proc_kprobes_optimization_handler(struct ctl_table *table, int write,
836 void __user *buffer, size_t *length,
841 mutex_lock(&kprobe_sysctl_mutex);
842 sysctl_kprobes_optimization = kprobes_allow_optimization ? 1 : 0;
843 ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
845 if (sysctl_kprobes_optimization)
846 optimize_all_kprobes();
848 unoptimize_all_kprobes();
849 mutex_unlock(&kprobe_sysctl_mutex);
853 #endif /* CONFIG_SYSCTL */
855 /* Put a breakpoint for a probe. Must be called with text_mutex locked */
856 static void __arm_kprobe(struct kprobe *p)
860 /* Check collision with other optimized kprobes */
861 _p = get_optimized_kprobe((unsigned long)p->addr);
863 /* Fallback to unoptimized kprobe */
864 unoptimize_kprobe(_p, true);
867 optimize_kprobe(p); /* Try to optimize (add kprobe to a list) */
870 /* Remove the breakpoint of a probe. Must be called with text_mutex locked */
871 static void __disarm_kprobe(struct kprobe *p, bool reopt)
875 /* Try to unoptimize */
876 unoptimize_kprobe(p, kprobes_all_disarmed);
878 if (!kprobe_queued(p)) {
879 arch_disarm_kprobe(p);
880 /* If another kprobe was blocked, optimize it. */
881 _p = get_optimized_kprobe((unsigned long)p->addr);
882 if (unlikely(_p) && reopt)
885 /* TODO: reoptimize others after unoptimized this probe */
888 #else /* !CONFIG_OPTPROBES */
890 #define optimize_kprobe(p) do {} while (0)
891 #define unoptimize_kprobe(p, f) do {} while (0)
892 #define kill_optimized_kprobe(p) do {} while (0)
893 #define prepare_optimized_kprobe(p) do {} while (0)
894 #define try_to_optimize_kprobe(p) do {} while (0)
895 #define __arm_kprobe(p) arch_arm_kprobe(p)
896 #define __disarm_kprobe(p, o) arch_disarm_kprobe(p)
897 #define kprobe_disarmed(p) kprobe_disabled(p)
898 #define wait_for_kprobe_optimizer() do {} while (0)
900 static int reuse_unused_kprobe(struct kprobe *ap)
903 * If the optimized kprobe is NOT supported, the aggr kprobe is
904 * released at the same time that the last aggregated kprobe is
906 * Thus there should be no chance to reuse unused kprobe.
908 printk(KERN_ERR "Error: There should be no unused kprobe here.\n");
912 static void free_aggr_kprobe(struct kprobe *p)
914 arch_remove_kprobe(p);
918 static struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
920 return kzalloc(sizeof(struct kprobe), GFP_KERNEL);
922 #endif /* CONFIG_OPTPROBES */
924 #ifdef CONFIG_KPROBES_ON_FTRACE
925 static struct ftrace_ops kprobe_ftrace_ops __read_mostly = {
926 .func = kprobe_ftrace_handler,
927 .flags = FTRACE_OPS_FL_SAVE_REGS | FTRACE_OPS_FL_IPMODIFY,
929 static int kprobe_ftrace_enabled;
931 /* Must ensure p->addr is really on ftrace */
932 static int prepare_kprobe(struct kprobe *p)
934 if (!kprobe_ftrace(p))
935 return arch_prepare_kprobe(p);
937 return arch_prepare_kprobe_ftrace(p);
940 /* Caller must lock kprobe_mutex */
941 static void arm_kprobe_ftrace(struct kprobe *p)
945 ret = ftrace_set_filter_ip(&kprobe_ftrace_ops,
946 (unsigned long)p->addr, 0, 0);
947 WARN(ret < 0, "Failed to arm kprobe-ftrace at %p (%d)\n", p->addr, ret);
948 kprobe_ftrace_enabled++;
949 if (kprobe_ftrace_enabled == 1) {
950 ret = register_ftrace_function(&kprobe_ftrace_ops);
951 WARN(ret < 0, "Failed to init kprobe-ftrace (%d)\n", ret);
955 /* Caller must lock kprobe_mutex */
956 static void disarm_kprobe_ftrace(struct kprobe *p)
960 kprobe_ftrace_enabled--;
961 if (kprobe_ftrace_enabled == 0) {
962 ret = unregister_ftrace_function(&kprobe_ftrace_ops);
963 WARN(ret < 0, "Failed to init kprobe-ftrace (%d)\n", ret);
965 ret = ftrace_set_filter_ip(&kprobe_ftrace_ops,
966 (unsigned long)p->addr, 1, 0);
967 WARN(ret < 0, "Failed to disarm kprobe-ftrace at %p (%d)\n", p->addr, ret);
969 #else /* !CONFIG_KPROBES_ON_FTRACE */
970 #define prepare_kprobe(p) arch_prepare_kprobe(p)
971 #define arm_kprobe_ftrace(p) do {} while (0)
972 #define disarm_kprobe_ftrace(p) do {} while (0)
975 /* Arm a kprobe with text_mutex */
976 static void arm_kprobe(struct kprobe *kp)
978 if (unlikely(kprobe_ftrace(kp))) {
979 arm_kprobe_ftrace(kp);
983 * Here, since __arm_kprobe() doesn't use stop_machine(),
984 * this doesn't cause deadlock on text_mutex. So, we don't
985 * need get_online_cpus().
987 mutex_lock(&text_mutex);
989 mutex_unlock(&text_mutex);
992 /* Disarm a kprobe with text_mutex */
993 static void disarm_kprobe(struct kprobe *kp, bool reopt)
995 if (unlikely(kprobe_ftrace(kp))) {
996 disarm_kprobe_ftrace(kp);
1000 mutex_lock(&text_mutex);
1001 __disarm_kprobe(kp, reopt);
1002 mutex_unlock(&text_mutex);
1006 * Aggregate handlers for multiple kprobes support - these handlers
1007 * take care of invoking the individual kprobe handlers on p->list
1009 static int aggr_pre_handler(struct kprobe *p, struct pt_regs *regs)
1013 list_for_each_entry_rcu(kp, &p->list, list) {
1014 if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
1015 set_kprobe_instance(kp);
1016 if (kp->pre_handler(kp, regs))
1019 reset_kprobe_instance();
1023 NOKPROBE_SYMBOL(aggr_pre_handler);
1025 static void aggr_post_handler(struct kprobe *p, struct pt_regs *regs,
1026 unsigned long flags)
1030 list_for_each_entry_rcu(kp, &p->list, list) {
1031 if (kp->post_handler && likely(!kprobe_disabled(kp))) {
1032 set_kprobe_instance(kp);
1033 kp->post_handler(kp, regs, flags);
1034 reset_kprobe_instance();
1038 NOKPROBE_SYMBOL(aggr_post_handler);
1040 static int aggr_fault_handler(struct kprobe *p, struct pt_regs *regs,
1043 struct kprobe *cur = __this_cpu_read(kprobe_instance);
1046 * if we faulted "during" the execution of a user specified
1047 * probe handler, invoke just that probe's fault handler
1049 if (cur && cur->fault_handler) {
1050 if (cur->fault_handler(cur, regs, trapnr))
1055 NOKPROBE_SYMBOL(aggr_fault_handler);
1057 static int aggr_break_handler(struct kprobe *p, struct pt_regs *regs)
1059 struct kprobe *cur = __this_cpu_read(kprobe_instance);
1062 if (cur && cur->break_handler) {
1063 if (cur->break_handler(cur, regs))
1066 reset_kprobe_instance();
1069 NOKPROBE_SYMBOL(aggr_break_handler);
1071 /* Walks the list and increments nmissed count for multiprobe case */
1072 void kprobes_inc_nmissed_count(struct kprobe *p)
1075 if (!kprobe_aggrprobe(p)) {
1078 list_for_each_entry_rcu(kp, &p->list, list)
1083 NOKPROBE_SYMBOL(kprobes_inc_nmissed_count);
1085 void recycle_rp_inst(struct kretprobe_instance *ri,
1086 struct hlist_head *head)
1088 struct kretprobe *rp = ri->rp;
1090 /* remove rp inst off the rprobe_inst_table */
1091 hlist_del(&ri->hlist);
1092 INIT_HLIST_NODE(&ri->hlist);
1094 raw_spin_lock(&rp->lock);
1095 hlist_add_head(&ri->hlist, &rp->free_instances);
1096 raw_spin_unlock(&rp->lock);
1099 hlist_add_head(&ri->hlist, head);
1101 NOKPROBE_SYMBOL(recycle_rp_inst);
1103 void kretprobe_hash_lock(struct task_struct *tsk,
1104 struct hlist_head **head, unsigned long *flags)
1105 __acquires(hlist_lock)
1107 unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
1108 raw_spinlock_t *hlist_lock;
1110 *head = &kretprobe_inst_table[hash];
1111 hlist_lock = kretprobe_table_lock_ptr(hash);
1112 raw_spin_lock_irqsave(hlist_lock, *flags);
1114 NOKPROBE_SYMBOL(kretprobe_hash_lock);
1116 static void kretprobe_table_lock(unsigned long hash,
1117 unsigned long *flags)
1118 __acquires(hlist_lock)
1120 raw_spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
1121 raw_spin_lock_irqsave(hlist_lock, *flags);
1123 NOKPROBE_SYMBOL(kretprobe_table_lock);
1125 void kretprobe_hash_unlock(struct task_struct *tsk,
1126 unsigned long *flags)
1127 __releases(hlist_lock)
1129 unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
1130 raw_spinlock_t *hlist_lock;
1132 hlist_lock = kretprobe_table_lock_ptr(hash);
1133 raw_spin_unlock_irqrestore(hlist_lock, *flags);
1135 NOKPROBE_SYMBOL(kretprobe_hash_unlock);
1137 static void kretprobe_table_unlock(unsigned long hash,
1138 unsigned long *flags)
1139 __releases(hlist_lock)
1141 raw_spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
1142 raw_spin_unlock_irqrestore(hlist_lock, *flags);
1144 NOKPROBE_SYMBOL(kretprobe_table_unlock);
1147 * This function is called from finish_task_switch when task tk becomes dead,
1148 * so that we can recycle any function-return probe instances associated
1149 * with this task. These left over instances represent probed functions
1150 * that have been called but will never return.
1152 void kprobe_flush_task(struct task_struct *tk)
1154 struct kretprobe_instance *ri;
1155 struct hlist_head *head, empty_rp;
1156 struct hlist_node *tmp;
1157 unsigned long hash, flags = 0;
1159 if (unlikely(!kprobes_initialized))
1160 /* Early boot. kretprobe_table_locks not yet initialized. */
1163 INIT_HLIST_HEAD(&empty_rp);
1164 hash = hash_ptr(tk, KPROBE_HASH_BITS);
1165 head = &kretprobe_inst_table[hash];
1166 kretprobe_table_lock(hash, &flags);
1167 hlist_for_each_entry_safe(ri, tmp, head, hlist) {
1169 recycle_rp_inst(ri, &empty_rp);
1171 kretprobe_table_unlock(hash, &flags);
1172 hlist_for_each_entry_safe(ri, tmp, &empty_rp, hlist) {
1173 hlist_del(&ri->hlist);
1177 NOKPROBE_SYMBOL(kprobe_flush_task);
1179 static inline void free_rp_inst(struct kretprobe *rp)
1181 struct kretprobe_instance *ri;
1182 struct hlist_node *next;
1184 hlist_for_each_entry_safe(ri, next, &rp->free_instances, hlist) {
1185 hlist_del(&ri->hlist);
1190 static void cleanup_rp_inst(struct kretprobe *rp)
1192 unsigned long flags, hash;
1193 struct kretprobe_instance *ri;
1194 struct hlist_node *next;
1195 struct hlist_head *head;
1198 for (hash = 0; hash < KPROBE_TABLE_SIZE; hash++) {
1199 kretprobe_table_lock(hash, &flags);
1200 head = &kretprobe_inst_table[hash];
1201 hlist_for_each_entry_safe(ri, next, head, hlist) {
1205 kretprobe_table_unlock(hash, &flags);
1209 NOKPROBE_SYMBOL(cleanup_rp_inst);
1212 * Add the new probe to ap->list. Fail if this is the
1213 * second jprobe at the address - two jprobes can't coexist
1215 static int add_new_kprobe(struct kprobe *ap, struct kprobe *p)
1217 BUG_ON(kprobe_gone(ap) || kprobe_gone(p));
1219 if (p->break_handler || p->post_handler)
1220 unoptimize_kprobe(ap, true); /* Fall back to normal kprobe */
1222 if (p->break_handler) {
1223 if (ap->break_handler)
1225 list_add_tail_rcu(&p->list, &ap->list);
1226 ap->break_handler = aggr_break_handler;
1228 list_add_rcu(&p->list, &ap->list);
1229 if (p->post_handler && !ap->post_handler)
1230 ap->post_handler = aggr_post_handler;
1236 * Fill in the required fields of the "manager kprobe". Replace the
1237 * earlier kprobe in the hlist with the manager kprobe
1239 static void init_aggr_kprobe(struct kprobe *ap, struct kprobe *p)
1241 /* Copy p's insn slot to ap */
1243 flush_insn_slot(ap);
1245 ap->flags = p->flags & ~KPROBE_FLAG_OPTIMIZED;
1246 ap->pre_handler = aggr_pre_handler;
1247 ap->fault_handler = aggr_fault_handler;
1248 /* We don't care the kprobe which has gone. */
1249 if (p->post_handler && !kprobe_gone(p))
1250 ap->post_handler = aggr_post_handler;
1251 if (p->break_handler && !kprobe_gone(p))
1252 ap->break_handler = aggr_break_handler;
1254 INIT_LIST_HEAD(&ap->list);
1255 INIT_HLIST_NODE(&ap->hlist);
1257 list_add_rcu(&p->list, &ap->list);
1258 hlist_replace_rcu(&p->hlist, &ap->hlist);
1262 * This is the second or subsequent kprobe at the address - handle
1265 static int register_aggr_kprobe(struct kprobe *orig_p, struct kprobe *p)
1268 struct kprobe *ap = orig_p;
1270 /* For preparing optimization, jump_label_text_reserved() is called */
1273 * Get online CPUs to avoid text_mutex deadlock.with stop machine,
1274 * which is invoked by unoptimize_kprobe() in add_new_kprobe()
1277 mutex_lock(&text_mutex);
1279 if (!kprobe_aggrprobe(orig_p)) {
1280 /* If orig_p is not an aggr_kprobe, create new aggr_kprobe. */
1281 ap = alloc_aggr_kprobe(orig_p);
1286 init_aggr_kprobe(ap, orig_p);
1287 } else if (kprobe_unused(ap)) {
1288 /* This probe is going to die. Rescue it */
1289 ret = reuse_unused_kprobe(ap);
1294 if (kprobe_gone(ap)) {
1296 * Attempting to insert new probe at the same location that
1297 * had a probe in the module vaddr area which already
1298 * freed. So, the instruction slot has already been
1299 * released. We need a new slot for the new probe.
1301 ret = arch_prepare_kprobe(ap);
1304 * Even if fail to allocate new slot, don't need to
1305 * free aggr_probe. It will be used next time, or
1306 * freed by unregister_kprobe.
1310 /* Prepare optimized instructions if possible. */
1311 prepare_optimized_kprobe(ap);
1314 * Clear gone flag to prevent allocating new slot again, and
1315 * set disabled flag because it is not armed yet.
1317 ap->flags = (ap->flags & ~KPROBE_FLAG_GONE)
1318 | KPROBE_FLAG_DISABLED;
1321 /* Copy ap's insn slot to p */
1323 ret = add_new_kprobe(ap, p);
1326 mutex_unlock(&text_mutex);
1328 jump_label_unlock();
1330 if (ret == 0 && kprobe_disabled(ap) && !kprobe_disabled(p)) {
1331 ap->flags &= ~KPROBE_FLAG_DISABLED;
1332 if (!kprobes_all_disarmed)
1333 /* Arm the breakpoint again. */
1339 bool __weak arch_within_kprobe_blacklist(unsigned long addr)
1341 /* The __kprobes marked functions and entry code must not be probed */
1342 return addr >= (unsigned long)__kprobes_text_start &&
1343 addr < (unsigned long)__kprobes_text_end;
1346 bool within_kprobe_blacklist(unsigned long addr)
1348 struct kprobe_blacklist_entry *ent;
1350 if (arch_within_kprobe_blacklist(addr))
1353 * If there exists a kprobe_blacklist, verify and
1354 * fail any probe registration in the prohibited area
1356 list_for_each_entry(ent, &kprobe_blacklist, list) {
1357 if (addr >= ent->start_addr && addr < ent->end_addr)
1365 * If we have a symbol_name argument, look it up and add the offset field
1366 * to it. This way, we can specify a relative address to a symbol.
1367 * This returns encoded errors if it fails to look up symbol or invalid
1368 * combination of parameters.
1370 static kprobe_opcode_t *kprobe_addr(struct kprobe *p)
1372 kprobe_opcode_t *addr = p->addr;
1374 if ((p->symbol_name && p->addr) ||
1375 (!p->symbol_name && !p->addr))
1378 if (p->symbol_name) {
1379 kprobe_lookup_name(p->symbol_name, addr);
1381 return ERR_PTR(-ENOENT);
1384 addr = (kprobe_opcode_t *)(((char *)addr) + p->offset);
1389 return ERR_PTR(-EINVAL);
1392 /* Check passed kprobe is valid and return kprobe in kprobe_table. */
1393 static struct kprobe *__get_valid_kprobe(struct kprobe *p)
1395 struct kprobe *ap, *list_p;
1397 ap = get_kprobe(p->addr);
1402 list_for_each_entry_rcu(list_p, &ap->list, list)
1404 /* kprobe p is a valid probe */
1412 /* Return error if the kprobe is being re-registered */
1413 static inline int check_kprobe_rereg(struct kprobe *p)
1417 mutex_lock(&kprobe_mutex);
1418 if (__get_valid_kprobe(p))
1420 mutex_unlock(&kprobe_mutex);
1425 int __weak arch_check_ftrace_location(struct kprobe *p)
1427 unsigned long ftrace_addr;
1429 ftrace_addr = ftrace_location((unsigned long)p->addr);
1431 #ifdef CONFIG_KPROBES_ON_FTRACE
1432 /* Given address is not on the instruction boundary */
1433 if ((unsigned long)p->addr != ftrace_addr)
1435 p->flags |= KPROBE_FLAG_FTRACE;
1436 #else /* !CONFIG_KPROBES_ON_FTRACE */
1443 static int check_kprobe_address_safe(struct kprobe *p,
1444 struct module **probed_mod)
1448 ret = arch_check_ftrace_location(p);
1454 /* Ensure it is not in reserved area nor out of text */
1455 if (!kernel_text_address((unsigned long) p->addr) ||
1456 within_kprobe_blacklist((unsigned long) p->addr) ||
1457 jump_label_text_reserved(p->addr, p->addr)) {
1462 /* Check if are we probing a module */
1463 *probed_mod = __module_text_address((unsigned long) p->addr);
1466 * We must hold a refcount of the probed module while updating
1467 * its code to prohibit unexpected unloading.
1469 if (unlikely(!try_module_get(*probed_mod))) {
1475 * If the module freed .init.text, we couldn't insert
1478 if (within_module_init((unsigned long)p->addr, *probed_mod) &&
1479 (*probed_mod)->state != MODULE_STATE_COMING) {
1480 module_put(*probed_mod);
1487 jump_label_unlock();
1492 int register_kprobe(struct kprobe *p)
1495 struct kprobe *old_p;
1496 struct module *probed_mod;
1497 kprobe_opcode_t *addr;
1499 /* Adjust probe address from symbol */
1500 addr = kprobe_addr(p);
1502 return PTR_ERR(addr);
1505 ret = check_kprobe_rereg(p);
1509 /* User can pass only KPROBE_FLAG_DISABLED to register_kprobe */
1510 p->flags &= KPROBE_FLAG_DISABLED;
1512 INIT_LIST_HEAD(&p->list);
1514 ret = check_kprobe_address_safe(p, &probed_mod);
1518 mutex_lock(&kprobe_mutex);
1520 old_p = get_kprobe(p->addr);
1522 /* Since this may unoptimize old_p, locking text_mutex. */
1523 ret = register_aggr_kprobe(old_p, p);
1527 mutex_lock(&text_mutex); /* Avoiding text modification */
1528 ret = prepare_kprobe(p);
1529 mutex_unlock(&text_mutex);
1533 INIT_HLIST_NODE(&p->hlist);
1534 hlist_add_head_rcu(&p->hlist,
1535 &kprobe_table[hash_ptr(p->addr, KPROBE_HASH_BITS)]);
1537 if (!kprobes_all_disarmed && !kprobe_disabled(p))
1540 /* Try to optimize kprobe */
1541 try_to_optimize_kprobe(p);
1544 mutex_unlock(&kprobe_mutex);
1547 module_put(probed_mod);
1551 EXPORT_SYMBOL_GPL(register_kprobe);
1553 /* Check if all probes on the aggrprobe are disabled */
1554 static int aggr_kprobe_disabled(struct kprobe *ap)
1558 list_for_each_entry_rcu(kp, &ap->list, list)
1559 if (!kprobe_disabled(kp))
1561 * There is an active probe on the list.
1562 * We can't disable this ap.
1569 /* Disable one kprobe: Make sure called under kprobe_mutex is locked */
1570 static struct kprobe *__disable_kprobe(struct kprobe *p)
1572 struct kprobe *orig_p;
1574 /* Get an original kprobe for return */
1575 orig_p = __get_valid_kprobe(p);
1576 if (unlikely(orig_p == NULL))
1579 if (!kprobe_disabled(p)) {
1580 /* Disable probe if it is a child probe */
1582 p->flags |= KPROBE_FLAG_DISABLED;
1584 /* Try to disarm and disable this/parent probe */
1585 if (p == orig_p || aggr_kprobe_disabled(orig_p)) {
1587 * If kprobes_all_disarmed is set, orig_p
1588 * should have already been disarmed, so
1589 * skip unneed disarming process.
1591 if (!kprobes_all_disarmed)
1592 disarm_kprobe(orig_p, true);
1593 orig_p->flags |= KPROBE_FLAG_DISABLED;
1601 * Unregister a kprobe without a scheduler synchronization.
1603 static int __unregister_kprobe_top(struct kprobe *p)
1605 struct kprobe *ap, *list_p;
1607 /* Disable kprobe. This will disarm it if needed. */
1608 ap = __disable_kprobe(p);
1614 * This probe is an independent(and non-optimized) kprobe
1615 * (not an aggrprobe). Remove from the hash list.
1619 /* Following process expects this probe is an aggrprobe */
1620 WARN_ON(!kprobe_aggrprobe(ap));
1622 if (list_is_singular(&ap->list) && kprobe_disarmed(ap))
1624 * !disarmed could be happen if the probe is under delayed
1629 /* If disabling probe has special handlers, update aggrprobe */
1630 if (p->break_handler && !kprobe_gone(p))
1631 ap->break_handler = NULL;
1632 if (p->post_handler && !kprobe_gone(p)) {
1633 list_for_each_entry_rcu(list_p, &ap->list, list) {
1634 if ((list_p != p) && (list_p->post_handler))
1637 ap->post_handler = NULL;
1641 * Remove from the aggrprobe: this path will do nothing in
1642 * __unregister_kprobe_bottom().
1644 list_del_rcu(&p->list);
1645 if (!kprobe_disabled(ap) && !kprobes_all_disarmed)
1647 * Try to optimize this probe again, because post
1648 * handler may have been changed.
1650 optimize_kprobe(ap);
1655 BUG_ON(!kprobe_disarmed(ap));
1656 hlist_del_rcu(&ap->hlist);
1660 static void __unregister_kprobe_bottom(struct kprobe *p)
1664 if (list_empty(&p->list))
1665 /* This is an independent kprobe */
1666 arch_remove_kprobe(p);
1667 else if (list_is_singular(&p->list)) {
1668 /* This is the last child of an aggrprobe */
1669 ap = list_entry(p->list.next, struct kprobe, list);
1671 free_aggr_kprobe(ap);
1673 /* Otherwise, do nothing. */
1676 int register_kprobes(struct kprobe **kps, int num)
1682 for (i = 0; i < num; i++) {
1683 ret = register_kprobe(kps[i]);
1686 unregister_kprobes(kps, i);
1692 EXPORT_SYMBOL_GPL(register_kprobes);
1694 void unregister_kprobe(struct kprobe *p)
1696 unregister_kprobes(&p, 1);
1698 EXPORT_SYMBOL_GPL(unregister_kprobe);
1700 void unregister_kprobes(struct kprobe **kps, int num)
1706 mutex_lock(&kprobe_mutex);
1707 for (i = 0; i < num; i++)
1708 if (__unregister_kprobe_top(kps[i]) < 0)
1709 kps[i]->addr = NULL;
1710 mutex_unlock(&kprobe_mutex);
1712 synchronize_sched();
1713 for (i = 0; i < num; i++)
1715 __unregister_kprobe_bottom(kps[i]);
1717 EXPORT_SYMBOL_GPL(unregister_kprobes);
1719 static struct notifier_block kprobe_exceptions_nb = {
1720 .notifier_call = kprobe_exceptions_notify,
1721 .priority = 0x7fffffff /* we need to be notified first */
1724 unsigned long __weak arch_deref_entry_point(void *entry)
1726 return (unsigned long)entry;
1729 int register_jprobes(struct jprobe **jps, int num)
1736 for (i = 0; i < num; i++) {
1737 unsigned long addr, offset;
1739 addr = arch_deref_entry_point(jp->entry);
1741 /* Verify probepoint is a function entry point */
1742 if (kallsyms_lookup_size_offset(addr, NULL, &offset) &&
1744 jp->kp.pre_handler = setjmp_pre_handler;
1745 jp->kp.break_handler = longjmp_break_handler;
1746 ret = register_kprobe(&jp->kp);
1752 unregister_jprobes(jps, i);
1758 EXPORT_SYMBOL_GPL(register_jprobes);
1760 int register_jprobe(struct jprobe *jp)
1762 return register_jprobes(&jp, 1);
1764 EXPORT_SYMBOL_GPL(register_jprobe);
1766 void unregister_jprobe(struct jprobe *jp)
1768 unregister_jprobes(&jp, 1);
1770 EXPORT_SYMBOL_GPL(unregister_jprobe);
1772 void unregister_jprobes(struct jprobe **jps, int num)
1778 mutex_lock(&kprobe_mutex);
1779 for (i = 0; i < num; i++)
1780 if (__unregister_kprobe_top(&jps[i]->kp) < 0)
1781 jps[i]->kp.addr = NULL;
1782 mutex_unlock(&kprobe_mutex);
1784 synchronize_sched();
1785 for (i = 0; i < num; i++) {
1786 if (jps[i]->kp.addr)
1787 __unregister_kprobe_bottom(&jps[i]->kp);
1790 EXPORT_SYMBOL_GPL(unregister_jprobes);
1792 #ifdef CONFIG_KRETPROBES
1794 * This kprobe pre_handler is registered with every kretprobe. When probe
1795 * hits it will set up the return probe.
1797 static int pre_handler_kretprobe(struct kprobe *p, struct pt_regs *regs)
1799 struct kretprobe *rp = container_of(p, struct kretprobe, kp);
1800 unsigned long hash, flags = 0;
1801 struct kretprobe_instance *ri;
1804 * To avoid deadlocks, prohibit return probing in NMI contexts,
1805 * just skip the probe and increase the (inexact) 'nmissed'
1806 * statistical counter, so that the user is informed that
1807 * something happened:
1809 if (unlikely(in_nmi())) {
1814 /* TODO: consider to only swap the RA after the last pre_handler fired */
1815 hash = hash_ptr(current, KPROBE_HASH_BITS);
1816 raw_spin_lock_irqsave(&rp->lock, flags);
1817 if (!hlist_empty(&rp->free_instances)) {
1818 ri = hlist_entry(rp->free_instances.first,
1819 struct kretprobe_instance, hlist);
1820 hlist_del(&ri->hlist);
1821 raw_spin_unlock_irqrestore(&rp->lock, flags);
1826 if (rp->entry_handler && rp->entry_handler(ri, regs)) {
1827 raw_spin_lock_irqsave(&rp->lock, flags);
1828 hlist_add_head(&ri->hlist, &rp->free_instances);
1829 raw_spin_unlock_irqrestore(&rp->lock, flags);
1833 arch_prepare_kretprobe(ri, regs);
1835 /* XXX(hch): why is there no hlist_move_head? */
1836 INIT_HLIST_NODE(&ri->hlist);
1837 kretprobe_table_lock(hash, &flags);
1838 hlist_add_head(&ri->hlist, &kretprobe_inst_table[hash]);
1839 kretprobe_table_unlock(hash, &flags);
1842 raw_spin_unlock_irqrestore(&rp->lock, flags);
1846 NOKPROBE_SYMBOL(pre_handler_kretprobe);
1848 int register_kretprobe(struct kretprobe *rp)
1851 struct kretprobe_instance *inst;
1855 if (kretprobe_blacklist_size) {
1856 addr = kprobe_addr(&rp->kp);
1858 return PTR_ERR(addr);
1860 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
1861 if (kretprobe_blacklist[i].addr == addr)
1866 rp->kp.pre_handler = pre_handler_kretprobe;
1867 rp->kp.post_handler = NULL;
1868 rp->kp.fault_handler = NULL;
1869 rp->kp.break_handler = NULL;
1871 /* Pre-allocate memory for max kretprobe instances */
1872 if (rp->maxactive <= 0) {
1873 #ifdef CONFIG_PREEMPT
1874 rp->maxactive = max_t(unsigned int, 10, 2*num_possible_cpus());
1876 rp->maxactive = num_possible_cpus();
1879 raw_spin_lock_init(&rp->lock);
1880 INIT_HLIST_HEAD(&rp->free_instances);
1881 for (i = 0; i < rp->maxactive; i++) {
1882 inst = kmalloc(sizeof(struct kretprobe_instance) +
1883 rp->data_size, GFP_KERNEL);
1888 INIT_HLIST_NODE(&inst->hlist);
1889 hlist_add_head(&inst->hlist, &rp->free_instances);
1893 /* Establish function entry probe point */
1894 ret = register_kprobe(&rp->kp);
1899 EXPORT_SYMBOL_GPL(register_kretprobe);
1901 int register_kretprobes(struct kretprobe **rps, int num)
1907 for (i = 0; i < num; i++) {
1908 ret = register_kretprobe(rps[i]);
1911 unregister_kretprobes(rps, i);
1917 EXPORT_SYMBOL_GPL(register_kretprobes);
1919 void unregister_kretprobe(struct kretprobe *rp)
1921 unregister_kretprobes(&rp, 1);
1923 EXPORT_SYMBOL_GPL(unregister_kretprobe);
1925 void unregister_kretprobes(struct kretprobe **rps, int num)
1931 mutex_lock(&kprobe_mutex);
1932 for (i = 0; i < num; i++)
1933 if (__unregister_kprobe_top(&rps[i]->kp) < 0)
1934 rps[i]->kp.addr = NULL;
1935 mutex_unlock(&kprobe_mutex);
1937 synchronize_sched();
1938 for (i = 0; i < num; i++) {
1939 if (rps[i]->kp.addr) {
1940 __unregister_kprobe_bottom(&rps[i]->kp);
1941 cleanup_rp_inst(rps[i]);
1945 EXPORT_SYMBOL_GPL(unregister_kretprobes);
1947 #else /* CONFIG_KRETPROBES */
1948 int register_kretprobe(struct kretprobe *rp)
1952 EXPORT_SYMBOL_GPL(register_kretprobe);
1954 int register_kretprobes(struct kretprobe **rps, int num)
1958 EXPORT_SYMBOL_GPL(register_kretprobes);
1960 void unregister_kretprobe(struct kretprobe *rp)
1963 EXPORT_SYMBOL_GPL(unregister_kretprobe);
1965 void unregister_kretprobes(struct kretprobe **rps, int num)
1968 EXPORT_SYMBOL_GPL(unregister_kretprobes);
1970 static int pre_handler_kretprobe(struct kprobe *p, struct pt_regs *regs)
1974 NOKPROBE_SYMBOL(pre_handler_kretprobe);
1976 #endif /* CONFIG_KRETPROBES */
1978 /* Set the kprobe gone and remove its instruction buffer. */
1979 static void kill_kprobe(struct kprobe *p)
1983 p->flags |= KPROBE_FLAG_GONE;
1984 if (kprobe_aggrprobe(p)) {
1986 * If this is an aggr_kprobe, we have to list all the
1987 * chained probes and mark them GONE.
1989 list_for_each_entry_rcu(kp, &p->list, list)
1990 kp->flags |= KPROBE_FLAG_GONE;
1991 p->post_handler = NULL;
1992 p->break_handler = NULL;
1993 kill_optimized_kprobe(p);
1996 * Here, we can remove insn_slot safely, because no thread calls
1997 * the original probed function (which will be freed soon) any more.
1999 arch_remove_kprobe(p);
2002 /* Disable one kprobe */
2003 int disable_kprobe(struct kprobe *kp)
2007 mutex_lock(&kprobe_mutex);
2009 /* Disable this kprobe */
2010 if (__disable_kprobe(kp) == NULL)
2013 mutex_unlock(&kprobe_mutex);
2016 EXPORT_SYMBOL_GPL(disable_kprobe);
2018 /* Enable one kprobe */
2019 int enable_kprobe(struct kprobe *kp)
2024 mutex_lock(&kprobe_mutex);
2026 /* Check whether specified probe is valid. */
2027 p = __get_valid_kprobe(kp);
2028 if (unlikely(p == NULL)) {
2033 if (kprobe_gone(kp)) {
2034 /* This kprobe has gone, we couldn't enable it. */
2040 kp->flags &= ~KPROBE_FLAG_DISABLED;
2042 if (!kprobes_all_disarmed && kprobe_disabled(p)) {
2043 p->flags &= ~KPROBE_FLAG_DISABLED;
2047 mutex_unlock(&kprobe_mutex);
2050 EXPORT_SYMBOL_GPL(enable_kprobe);
2052 void dump_kprobe(struct kprobe *kp)
2054 printk(KERN_WARNING "Dumping kprobe:\n");
2055 printk(KERN_WARNING "Name: %s\nAddress: %p\nOffset: %x\n",
2056 kp->symbol_name, kp->addr, kp->offset);
2058 NOKPROBE_SYMBOL(dump_kprobe);
2061 * Lookup and populate the kprobe_blacklist.
2063 * Unlike the kretprobe blacklist, we'll need to determine
2064 * the range of addresses that belong to the said functions,
2065 * since a kprobe need not necessarily be at the beginning
2068 static int __init populate_kprobe_blacklist(unsigned long *start,
2071 unsigned long *iter;
2072 struct kprobe_blacklist_entry *ent;
2073 unsigned long entry, offset = 0, size = 0;
2075 for (iter = start; iter < end; iter++) {
2076 entry = arch_deref_entry_point((void *)*iter);
2078 if (!kernel_text_address(entry) ||
2079 !kallsyms_lookup_size_offset(entry, &size, &offset)) {
2080 pr_err("Failed to find blacklist at %p\n",
2085 ent = kmalloc(sizeof(*ent), GFP_KERNEL);
2088 ent->start_addr = entry;
2089 ent->end_addr = entry + size;
2090 INIT_LIST_HEAD(&ent->list);
2091 list_add_tail(&ent->list, &kprobe_blacklist);
2096 /* Module notifier call back, checking kprobes on the module */
2097 static int kprobes_module_callback(struct notifier_block *nb,
2098 unsigned long val, void *data)
2100 struct module *mod = data;
2101 struct hlist_head *head;
2104 int checkcore = (val == MODULE_STATE_GOING);
2106 if (val != MODULE_STATE_GOING && val != MODULE_STATE_LIVE)
2110 * When MODULE_STATE_GOING was notified, both of module .text and
2111 * .init.text sections would be freed. When MODULE_STATE_LIVE was
2112 * notified, only .init.text section would be freed. We need to
2113 * disable kprobes which have been inserted in the sections.
2115 mutex_lock(&kprobe_mutex);
2116 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2117 head = &kprobe_table[i];
2118 hlist_for_each_entry_rcu(p, head, hlist)
2119 if (within_module_init((unsigned long)p->addr, mod) ||
2121 within_module_core((unsigned long)p->addr, mod))) {
2123 * The vaddr this probe is installed will soon
2124 * be vfreed buy not synced to disk. Hence,
2125 * disarming the breakpoint isn't needed.
2130 mutex_unlock(&kprobe_mutex);
2134 static struct notifier_block kprobe_module_nb = {
2135 .notifier_call = kprobes_module_callback,
2139 /* Markers of _kprobe_blacklist section */
2140 extern unsigned long __start_kprobe_blacklist[];
2141 extern unsigned long __stop_kprobe_blacklist[];
2143 static int __init init_kprobes(void)
2147 /* FIXME allocate the probe table, currently defined statically */
2148 /* initialize all list heads */
2149 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2150 INIT_HLIST_HEAD(&kprobe_table[i]);
2151 INIT_HLIST_HEAD(&kretprobe_inst_table[i]);
2152 raw_spin_lock_init(&(kretprobe_table_locks[i].lock));
2155 err = populate_kprobe_blacklist(__start_kprobe_blacklist,
2156 __stop_kprobe_blacklist);
2158 pr_err("kprobes: failed to populate blacklist: %d\n", err);
2159 pr_err("Please take care of using kprobes.\n");
2162 if (kretprobe_blacklist_size) {
2163 /* lookup the function address from its name */
2164 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
2165 kprobe_lookup_name(kretprobe_blacklist[i].name,
2166 kretprobe_blacklist[i].addr);
2167 if (!kretprobe_blacklist[i].addr)
2168 printk("kretprobe: lookup failed: %s\n",
2169 kretprobe_blacklist[i].name);
2173 #if defined(CONFIG_OPTPROBES)
2174 #if defined(__ARCH_WANT_KPROBES_INSN_SLOT)
2175 /* Init kprobe_optinsn_slots */
2176 kprobe_optinsn_slots.insn_size = MAX_OPTINSN_SIZE;
2178 /* By default, kprobes can be optimized */
2179 kprobes_allow_optimization = true;
2182 /* By default, kprobes are armed */
2183 kprobes_all_disarmed = false;
2185 err = arch_init_kprobes();
2187 err = register_die_notifier(&kprobe_exceptions_nb);
2189 err = register_module_notifier(&kprobe_module_nb);
2191 kprobes_initialized = (err == 0);
2198 #ifdef CONFIG_DEBUG_FS
2199 static void report_probe(struct seq_file *pi, struct kprobe *p,
2200 const char *sym, int offset, char *modname, struct kprobe *pp)
2204 if (p->pre_handler == pre_handler_kretprobe)
2206 else if (p->pre_handler == setjmp_pre_handler)
2212 seq_printf(pi, "%p %s %s+0x%x %s ",
2213 p->addr, kprobe_type, sym, offset,
2214 (modname ? modname : " "));
2216 seq_printf(pi, "%p %s %p ",
2217 p->addr, kprobe_type, p->addr);
2221 seq_printf(pi, "%s%s%s%s\n",
2222 (kprobe_gone(p) ? "[GONE]" : ""),
2223 ((kprobe_disabled(p) && !kprobe_gone(p)) ? "[DISABLED]" : ""),
2224 (kprobe_optimized(pp) ? "[OPTIMIZED]" : ""),
2225 (kprobe_ftrace(pp) ? "[FTRACE]" : ""));
2228 static void *kprobe_seq_start(struct seq_file *f, loff_t *pos)
2230 return (*pos < KPROBE_TABLE_SIZE) ? pos : NULL;
2233 static void *kprobe_seq_next(struct seq_file *f, void *v, loff_t *pos)
2236 if (*pos >= KPROBE_TABLE_SIZE)
2241 static void kprobe_seq_stop(struct seq_file *f, void *v)
2246 static int show_kprobe_addr(struct seq_file *pi, void *v)
2248 struct hlist_head *head;
2249 struct kprobe *p, *kp;
2250 const char *sym = NULL;
2251 unsigned int i = *(loff_t *) v;
2252 unsigned long offset = 0;
2253 char *modname, namebuf[KSYM_NAME_LEN];
2255 head = &kprobe_table[i];
2257 hlist_for_each_entry_rcu(p, head, hlist) {
2258 sym = kallsyms_lookup((unsigned long)p->addr, NULL,
2259 &offset, &modname, namebuf);
2260 if (kprobe_aggrprobe(p)) {
2261 list_for_each_entry_rcu(kp, &p->list, list)
2262 report_probe(pi, kp, sym, offset, modname, p);
2264 report_probe(pi, p, sym, offset, modname, NULL);
2270 static const struct seq_operations kprobes_seq_ops = {
2271 .start = kprobe_seq_start,
2272 .next = kprobe_seq_next,
2273 .stop = kprobe_seq_stop,
2274 .show = show_kprobe_addr
2277 static int kprobes_open(struct inode *inode, struct file *filp)
2279 return seq_open(filp, &kprobes_seq_ops);
2282 static const struct file_operations debugfs_kprobes_operations = {
2283 .open = kprobes_open,
2285 .llseek = seq_lseek,
2286 .release = seq_release,
2289 /* kprobes/blacklist -- shows which functions can not be probed */
2290 static void *kprobe_blacklist_seq_start(struct seq_file *m, loff_t *pos)
2292 return seq_list_start(&kprobe_blacklist, *pos);
2295 static void *kprobe_blacklist_seq_next(struct seq_file *m, void *v, loff_t *pos)
2297 return seq_list_next(v, &kprobe_blacklist, pos);
2300 static int kprobe_blacklist_seq_show(struct seq_file *m, void *v)
2302 struct kprobe_blacklist_entry *ent =
2303 list_entry(v, struct kprobe_blacklist_entry, list);
2305 seq_printf(m, "0x%p-0x%p\t%ps\n", (void *)ent->start_addr,
2306 (void *)ent->end_addr, (void *)ent->start_addr);
2310 static const struct seq_operations kprobe_blacklist_seq_ops = {
2311 .start = kprobe_blacklist_seq_start,
2312 .next = kprobe_blacklist_seq_next,
2313 .stop = kprobe_seq_stop, /* Reuse void function */
2314 .show = kprobe_blacklist_seq_show,
2317 static int kprobe_blacklist_open(struct inode *inode, struct file *filp)
2319 return seq_open(filp, &kprobe_blacklist_seq_ops);
2322 static const struct file_operations debugfs_kprobe_blacklist_ops = {
2323 .open = kprobe_blacklist_open,
2325 .llseek = seq_lseek,
2326 .release = seq_release,
2329 static void arm_all_kprobes(void)
2331 struct hlist_head *head;
2335 mutex_lock(&kprobe_mutex);
2337 /* If kprobes are armed, just return */
2338 if (!kprobes_all_disarmed)
2339 goto already_enabled;
2342 * optimize_kprobe() called by arm_kprobe() checks
2343 * kprobes_all_disarmed, so set kprobes_all_disarmed before
2346 kprobes_all_disarmed = false;
2347 /* Arming kprobes doesn't optimize kprobe itself */
2348 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2349 head = &kprobe_table[i];
2350 hlist_for_each_entry_rcu(p, head, hlist)
2351 if (!kprobe_disabled(p))
2355 printk(KERN_INFO "Kprobes globally enabled\n");
2358 mutex_unlock(&kprobe_mutex);
2362 static void disarm_all_kprobes(void)
2364 struct hlist_head *head;
2368 mutex_lock(&kprobe_mutex);
2370 /* If kprobes are already disarmed, just return */
2371 if (kprobes_all_disarmed) {
2372 mutex_unlock(&kprobe_mutex);
2376 kprobes_all_disarmed = true;
2377 printk(KERN_INFO "Kprobes globally disabled\n");
2379 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2380 head = &kprobe_table[i];
2381 hlist_for_each_entry_rcu(p, head, hlist) {
2382 if (!arch_trampoline_kprobe(p) && !kprobe_disabled(p))
2383 disarm_kprobe(p, false);
2386 mutex_unlock(&kprobe_mutex);
2388 /* Wait for disarming all kprobes by optimizer */
2389 wait_for_kprobe_optimizer();
2393 * XXX: The debugfs bool file interface doesn't allow for callbacks
2394 * when the bool state is switched. We can reuse that facility when
2397 static ssize_t read_enabled_file_bool(struct file *file,
2398 char __user *user_buf, size_t count, loff_t *ppos)
2402 if (!kprobes_all_disarmed)
2408 return simple_read_from_buffer(user_buf, count, ppos, buf, 2);
2411 static ssize_t write_enabled_file_bool(struct file *file,
2412 const char __user *user_buf, size_t count, loff_t *ppos)
2417 buf_size = min(count, (sizeof(buf)-1));
2418 if (copy_from_user(buf, user_buf, buf_size))
2421 buf[buf_size] = '\0';
2431 disarm_all_kprobes();
2440 static const struct file_operations fops_kp = {
2441 .read = read_enabled_file_bool,
2442 .write = write_enabled_file_bool,
2443 .llseek = default_llseek,
2446 static int __init debugfs_kprobe_init(void)
2448 struct dentry *dir, *file;
2449 unsigned int value = 1;
2451 dir = debugfs_create_dir("kprobes", NULL);
2455 file = debugfs_create_file("list", 0400, dir, NULL,
2456 &debugfs_kprobes_operations);
2460 file = debugfs_create_file("enabled", 0600, dir,
2465 file = debugfs_create_file("blacklist", 0400, dir, NULL,
2466 &debugfs_kprobe_blacklist_ops);
2473 debugfs_remove(dir);
2477 late_initcall(debugfs_kprobe_init);
2478 #endif /* CONFIG_DEBUG_FS */
2480 module_init(init_kprobes);
2482 /* defined in arch/.../kernel/kprobes.c */
2483 EXPORT_SYMBOL_GPL(jprobe_return);