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/module.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 spinlock_t lock ____cacheline_aligned_in_smp;
82 } kretprobe_table_locks[KPROBE_TABLE_SIZE];
84 static spinlock_t *kretprobe_table_lock_ptr(unsigned long hash)
86 return &(kretprobe_table_locks[hash].lock);
90 * Normally, functions that we'd want to prohibit kprobes in, are marked
91 * __kprobes. But, there are cases where such functions already belong to
92 * a different section (__sched for preempt_schedule)
94 * For such cases, we now have a blacklist
96 static struct kprobe_blackpoint kprobe_blacklist[] = {
97 {"preempt_schedule",},
98 {"native_get_debugreg",},
99 {"irq_entries_start",},
100 {"common_interrupt",},
101 {"mcount",}, /* mcount can be called from everywhere */
102 {NULL} /* Terminator */
105 #ifdef __ARCH_WANT_KPROBES_INSN_SLOT
107 * kprobe->ainsn.insn points to the copy of the instruction to be
108 * single-stepped. x86_64, POWER4 and above have no-exec support and
109 * stepping on the instruction on a vmalloced/kmalloced/data page
110 * is a recipe for disaster
112 struct kprobe_insn_page {
113 struct list_head list;
114 kprobe_opcode_t *insns; /* Page of instruction slots */
120 #define KPROBE_INSN_PAGE_SIZE(slots) \
121 (offsetof(struct kprobe_insn_page, slot_used) + \
122 (sizeof(char) * (slots)))
124 struct kprobe_insn_cache {
125 struct list_head pages; /* list of kprobe_insn_page */
126 size_t insn_size; /* size of instruction slot */
130 static int slots_per_page(struct kprobe_insn_cache *c)
132 return PAGE_SIZE/(c->insn_size * sizeof(kprobe_opcode_t));
135 enum kprobe_slot_state {
141 static DEFINE_MUTEX(kprobe_insn_mutex); /* Protects kprobe_insn_slots */
142 static struct kprobe_insn_cache kprobe_insn_slots = {
143 .pages = LIST_HEAD_INIT(kprobe_insn_slots.pages),
144 .insn_size = MAX_INSN_SIZE,
147 static int __kprobes collect_garbage_slots(struct kprobe_insn_cache *c);
150 * __get_insn_slot() - Find a slot on an executable page for an instruction.
151 * We allocate an executable page if there's no room on existing ones.
153 static kprobe_opcode_t __kprobes *__get_insn_slot(struct kprobe_insn_cache *c)
155 struct kprobe_insn_page *kip;
158 list_for_each_entry(kip, &c->pages, list) {
159 if (kip->nused < slots_per_page(c)) {
161 for (i = 0; i < slots_per_page(c); i++) {
162 if (kip->slot_used[i] == SLOT_CLEAN) {
163 kip->slot_used[i] = SLOT_USED;
165 return kip->insns + (i * c->insn_size);
168 /* kip->nused is broken. Fix it. */
169 kip->nused = slots_per_page(c);
174 /* If there are any garbage slots, collect it and try again. */
175 if (c->nr_garbage && collect_garbage_slots(c) == 0)
178 /* All out of space. Need to allocate a new page. */
179 kip = kmalloc(KPROBE_INSN_PAGE_SIZE(slots_per_page(c)), GFP_KERNEL);
184 * Use module_alloc so this page is within +/- 2GB of where the
185 * kernel image and loaded module images reside. This is required
186 * so x86_64 can correctly handle the %rip-relative fixups.
188 kip->insns = module_alloc(PAGE_SIZE);
193 INIT_LIST_HEAD(&kip->list);
194 memset(kip->slot_used, SLOT_CLEAN, slots_per_page(c));
195 kip->slot_used[0] = SLOT_USED;
198 list_add(&kip->list, &c->pages);
203 kprobe_opcode_t __kprobes *get_insn_slot(void)
205 kprobe_opcode_t *ret = NULL;
207 mutex_lock(&kprobe_insn_mutex);
208 ret = __get_insn_slot(&kprobe_insn_slots);
209 mutex_unlock(&kprobe_insn_mutex);
214 /* Return 1 if all garbages are collected, otherwise 0. */
215 static int __kprobes collect_one_slot(struct kprobe_insn_page *kip, int idx)
217 kip->slot_used[idx] = SLOT_CLEAN;
219 if (kip->nused == 0) {
221 * Page is no longer in use. Free it unless
222 * it's the last one. We keep the last one
223 * so as not to have to set it up again the
224 * next time somebody inserts a probe.
226 if (!list_is_singular(&kip->list)) {
227 list_del(&kip->list);
228 module_free(NULL, kip->insns);
236 static int __kprobes collect_garbage_slots(struct kprobe_insn_cache *c)
238 struct kprobe_insn_page *kip, *next;
240 /* Ensure no-one is interrupted on the garbages */
243 list_for_each_entry_safe(kip, next, &c->pages, list) {
245 if (kip->ngarbage == 0)
247 kip->ngarbage = 0; /* we will collect all garbages */
248 for (i = 0; i < slots_per_page(c); i++) {
249 if (kip->slot_used[i] == SLOT_DIRTY &&
250 collect_one_slot(kip, i))
258 static void __kprobes __free_insn_slot(struct kprobe_insn_cache *c,
259 kprobe_opcode_t *slot, int dirty)
261 struct kprobe_insn_page *kip;
263 list_for_each_entry(kip, &c->pages, list) {
264 long idx = ((long)slot - (long)kip->insns) /
265 (c->insn_size * sizeof(kprobe_opcode_t));
266 if (idx >= 0 && idx < slots_per_page(c)) {
267 WARN_ON(kip->slot_used[idx] != SLOT_USED);
269 kip->slot_used[idx] = SLOT_DIRTY;
271 if (++c->nr_garbage > slots_per_page(c))
272 collect_garbage_slots(c);
274 collect_one_slot(kip, idx);
278 /* Could not free this slot. */
282 void __kprobes free_insn_slot(kprobe_opcode_t * slot, int dirty)
284 mutex_lock(&kprobe_insn_mutex);
285 __free_insn_slot(&kprobe_insn_slots, slot, dirty);
286 mutex_unlock(&kprobe_insn_mutex);
288 #ifdef CONFIG_OPTPROBES
289 /* For optimized_kprobe buffer */
290 static DEFINE_MUTEX(kprobe_optinsn_mutex); /* Protects kprobe_optinsn_slots */
291 static struct kprobe_insn_cache kprobe_optinsn_slots = {
292 .pages = LIST_HEAD_INIT(kprobe_optinsn_slots.pages),
293 /* .insn_size is initialized later */
296 /* Get a slot for optimized_kprobe buffer */
297 kprobe_opcode_t __kprobes *get_optinsn_slot(void)
299 kprobe_opcode_t *ret = NULL;
301 mutex_lock(&kprobe_optinsn_mutex);
302 ret = __get_insn_slot(&kprobe_optinsn_slots);
303 mutex_unlock(&kprobe_optinsn_mutex);
308 void __kprobes free_optinsn_slot(kprobe_opcode_t * slot, int dirty)
310 mutex_lock(&kprobe_optinsn_mutex);
311 __free_insn_slot(&kprobe_optinsn_slots, slot, dirty);
312 mutex_unlock(&kprobe_optinsn_mutex);
317 /* We have preemption disabled.. so it is safe to use __ versions */
318 static inline void set_kprobe_instance(struct kprobe *kp)
320 __get_cpu_var(kprobe_instance) = kp;
323 static inline void reset_kprobe_instance(void)
325 __get_cpu_var(kprobe_instance) = NULL;
329 * This routine is called either:
330 * - under the kprobe_mutex - during kprobe_[un]register()
332 * - with preemption disabled - from arch/xxx/kernel/kprobes.c
334 struct kprobe __kprobes *get_kprobe(void *addr)
336 struct hlist_head *head;
337 struct hlist_node *node;
340 head = &kprobe_table[hash_ptr(addr, KPROBE_HASH_BITS)];
341 hlist_for_each_entry_rcu(p, node, head, hlist) {
349 static int __kprobes aggr_pre_handler(struct kprobe *p, struct pt_regs *regs);
351 /* Return true if the kprobe is an aggregator */
352 static inline int kprobe_aggrprobe(struct kprobe *p)
354 return p->pre_handler == aggr_pre_handler;
357 /* Return true(!0) if the kprobe is unused */
358 static inline int kprobe_unused(struct kprobe *p)
360 return kprobe_aggrprobe(p) && kprobe_disabled(p) &&
361 list_empty(&p->list);
365 * Keep all fields in the kprobe consistent
367 static inline void copy_kprobe(struct kprobe *ap, struct kprobe *p)
369 memcpy(&p->opcode, &ap->opcode, sizeof(kprobe_opcode_t));
370 memcpy(&p->ainsn, &ap->ainsn, sizeof(struct arch_specific_insn));
373 #ifdef CONFIG_OPTPROBES
374 /* NOTE: change this value only with kprobe_mutex held */
375 static bool kprobes_allow_optimization;
378 * Call all pre_handler on the list, but ignores its return value.
379 * This must be called from arch-dep optimized caller.
381 void __kprobes opt_pre_handler(struct kprobe *p, struct pt_regs *regs)
385 list_for_each_entry_rcu(kp, &p->list, list) {
386 if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
387 set_kprobe_instance(kp);
388 kp->pre_handler(kp, regs);
390 reset_kprobe_instance();
394 /* Free optimized instructions and optimized_kprobe */
395 static __kprobes void free_aggr_kprobe(struct kprobe *p)
397 struct optimized_kprobe *op;
399 op = container_of(p, struct optimized_kprobe, kp);
400 arch_remove_optimized_kprobe(op);
401 arch_remove_kprobe(p);
405 /* Return true(!0) if the kprobe is ready for optimization. */
406 static inline int kprobe_optready(struct kprobe *p)
408 struct optimized_kprobe *op;
410 if (kprobe_aggrprobe(p)) {
411 op = container_of(p, struct optimized_kprobe, kp);
412 return arch_prepared_optinsn(&op->optinsn);
418 /* Return true(!0) if the kprobe is disarmed. Note: p must be on hash list */
419 static inline int kprobe_disarmed(struct kprobe *p)
421 struct optimized_kprobe *op;
423 /* If kprobe is not aggr/opt probe, just return kprobe is disabled */
424 if (!kprobe_aggrprobe(p))
425 return kprobe_disabled(p);
427 op = container_of(p, struct optimized_kprobe, kp);
429 return kprobe_disabled(p) && list_empty(&op->list);
432 /* Return true(!0) if the probe is queued on (un)optimizing lists */
433 static int __kprobes kprobe_queued(struct kprobe *p)
435 struct optimized_kprobe *op;
437 if (kprobe_aggrprobe(p)) {
438 op = container_of(p, struct optimized_kprobe, kp);
439 if (!list_empty(&op->list))
446 * Return an optimized kprobe whose optimizing code replaces
447 * instructions including addr (exclude breakpoint).
449 static struct kprobe *__kprobes get_optimized_kprobe(unsigned long addr)
452 struct kprobe *p = NULL;
453 struct optimized_kprobe *op;
455 /* Don't check i == 0, since that is a breakpoint case. */
456 for (i = 1; !p && i < MAX_OPTIMIZED_LENGTH; i++)
457 p = get_kprobe((void *)(addr - i));
459 if (p && kprobe_optready(p)) {
460 op = container_of(p, struct optimized_kprobe, kp);
461 if (arch_within_optimized_kprobe(op, addr))
468 /* Optimization staging list, protected by kprobe_mutex */
469 static LIST_HEAD(optimizing_list);
470 static LIST_HEAD(unoptimizing_list);
472 static void kprobe_optimizer(struct work_struct *work);
473 static DECLARE_DELAYED_WORK(optimizing_work, kprobe_optimizer);
474 static DECLARE_COMPLETION(optimizer_comp);
475 #define OPTIMIZE_DELAY 5
478 * Optimize (replace a breakpoint with a jump) kprobes listed on
481 static __kprobes void do_optimize_kprobes(void)
483 struct optimized_kprobe *op, *tmp;
485 /* Optimization never be done when disarmed */
486 if (kprobes_all_disarmed || !kprobes_allow_optimization ||
487 list_empty(&optimizing_list))
491 * The optimization/unoptimization refers online_cpus via
492 * stop_machine() and cpu-hotplug modifies online_cpus.
493 * And same time, text_mutex will be held in cpu-hotplug and here.
494 * This combination can cause a deadlock (cpu-hotplug try to lock
495 * text_mutex but stop_machine can not be done because online_cpus
497 * To avoid this deadlock, we need to call get_online_cpus()
498 * for preventing cpu-hotplug outside of text_mutex locking.
501 mutex_lock(&text_mutex);
502 list_for_each_entry_safe(op, tmp, &optimizing_list, list) {
503 WARN_ON(kprobe_disabled(&op->kp));
504 if (arch_optimize_kprobe(op) < 0)
505 op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
506 list_del_init(&op->list);
508 mutex_unlock(&text_mutex);
513 * Unoptimize (replace a jump with a breakpoint and remove the breakpoint
514 * if need) kprobes listed on unoptimizing_list.
516 static __kprobes void do_unoptimize_kprobes(struct list_head *free_list)
518 struct optimized_kprobe *op, *tmp;
520 /* Unoptimization must be done anytime */
521 if (list_empty(&unoptimizing_list))
524 /* Ditto to do_optimize_kprobes */
526 mutex_lock(&text_mutex);
527 list_for_each_entry_safe(op, tmp, &unoptimizing_list, list) {
528 /* Unoptimize kprobes */
529 arch_unoptimize_kprobe(op);
530 /* Disarm probes if marked disabled */
531 if (kprobe_disabled(&op->kp))
532 arch_disarm_kprobe(&op->kp);
533 if (kprobe_unused(&op->kp)) {
535 * Remove unused probes from hash list. After waiting
536 * for synchronization, these probes are reclaimed.
537 * (reclaiming is done by do_free_cleaned_kprobes.)
539 hlist_del_rcu(&op->kp.hlist);
540 /* Move only unused probes on free_list */
541 list_move(&op->list, free_list);
543 list_del_init(&op->list);
545 mutex_unlock(&text_mutex);
549 /* Reclaim all kprobes on the free_list */
550 static __kprobes void do_free_cleaned_kprobes(struct list_head *free_list)
552 struct optimized_kprobe *op, *tmp;
554 list_for_each_entry_safe(op, tmp, free_list, list) {
555 BUG_ON(!kprobe_unused(&op->kp));
556 list_del_init(&op->list);
557 free_aggr_kprobe(&op->kp);
561 /* Start optimizer after OPTIMIZE_DELAY passed */
562 static __kprobes void kick_kprobe_optimizer(void)
564 if (!delayed_work_pending(&optimizing_work))
565 schedule_delayed_work(&optimizing_work, OPTIMIZE_DELAY);
568 /* Kprobe jump optimizer */
569 static __kprobes void kprobe_optimizer(struct work_struct *work)
571 LIST_HEAD(free_list);
573 /* Lock modules while optimizing kprobes */
574 mutex_lock(&module_mutex);
575 mutex_lock(&kprobe_mutex);
578 * Step 1: Unoptimize kprobes and collect cleaned (unused and disarmed)
579 * kprobes before waiting for quiesence period.
581 do_unoptimize_kprobes(&free_list);
584 * Step 2: Wait for quiesence period to ensure all running interrupts
585 * are done. Because optprobe may modify multiple instructions
586 * there is a chance that Nth instruction is interrupted. In that
587 * case, running interrupt can return to 2nd-Nth byte of jump
588 * instruction. This wait is for avoiding it.
592 /* Step 3: Optimize kprobes after quiesence period */
593 do_optimize_kprobes();
595 /* Step 4: Free cleaned kprobes after quiesence period */
596 do_free_cleaned_kprobes(&free_list);
598 mutex_unlock(&kprobe_mutex);
599 mutex_unlock(&module_mutex);
601 /* Wake up all waiters */
602 complete_all(&optimizer_comp);
605 /* Wait for completing optimization and unoptimization */
606 static __kprobes void wait_for_kprobe_optimizer(void)
608 if (delayed_work_pending(&optimizing_work))
609 wait_for_completion(&optimizer_comp);
612 /* Optimize kprobe if p is ready to be optimized */
613 static __kprobes void optimize_kprobe(struct kprobe *p)
615 struct optimized_kprobe *op;
617 /* Check if the kprobe is disabled or not ready for optimization. */
618 if (!kprobe_optready(p) || !kprobes_allow_optimization ||
619 (kprobe_disabled(p) || kprobes_all_disarmed))
622 /* Both of break_handler and post_handler are not supported. */
623 if (p->break_handler || p->post_handler)
626 op = container_of(p, struct optimized_kprobe, kp);
628 /* Check there is no other kprobes at the optimized instructions */
629 if (arch_check_optimized_kprobe(op) < 0)
632 /* Check if it is already optimized. */
633 if (op->kp.flags & KPROBE_FLAG_OPTIMIZED)
635 op->kp.flags |= KPROBE_FLAG_OPTIMIZED;
637 if (!list_empty(&op->list))
638 /* This is under unoptimizing. Just dequeue the probe */
639 list_del_init(&op->list);
641 list_add(&op->list, &optimizing_list);
642 kick_kprobe_optimizer();
646 /* Short cut to direct unoptimizing */
647 static __kprobes void force_unoptimize_kprobe(struct optimized_kprobe *op)
650 arch_unoptimize_kprobe(op);
652 if (kprobe_disabled(&op->kp))
653 arch_disarm_kprobe(&op->kp);
656 /* Unoptimize a kprobe if p is optimized */
657 static __kprobes void unoptimize_kprobe(struct kprobe *p, bool force)
659 struct optimized_kprobe *op;
661 if (!kprobe_aggrprobe(p) || kprobe_disarmed(p))
662 return; /* This is not an optprobe nor optimized */
664 op = container_of(p, struct optimized_kprobe, kp);
665 if (!kprobe_optimized(p)) {
666 /* Unoptimized or unoptimizing case */
667 if (force && !list_empty(&op->list)) {
669 * Only if this is unoptimizing kprobe and forced,
670 * forcibly unoptimize it. (No need to unoptimize
671 * unoptimized kprobe again :)
673 list_del_init(&op->list);
674 force_unoptimize_kprobe(op);
679 op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
680 if (!list_empty(&op->list)) {
681 /* Dequeue from the optimization queue */
682 list_del_init(&op->list);
685 /* Optimized kprobe case */
687 /* Forcibly update the code: this is a special case */
688 force_unoptimize_kprobe(op);
690 list_add(&op->list, &unoptimizing_list);
691 kick_kprobe_optimizer();
695 /* Cancel unoptimizing for reusing */
696 static void reuse_unused_kprobe(struct kprobe *ap)
698 struct optimized_kprobe *op;
700 BUG_ON(!kprobe_unused(ap));
702 * Unused kprobe MUST be on the way of delayed unoptimizing (means
703 * there is still a relative jump) and disabled.
705 op = container_of(ap, struct optimized_kprobe, kp);
706 if (unlikely(list_empty(&op->list)))
707 printk(KERN_WARNING "Warning: found a stray unused "
708 "aggrprobe@%p\n", ap->addr);
709 /* Enable the probe again */
710 ap->flags &= ~KPROBE_FLAG_DISABLED;
711 /* Optimize it again (remove from op->list) */
712 BUG_ON(!kprobe_optready(ap));
716 /* Remove optimized instructions */
717 static void __kprobes kill_optimized_kprobe(struct kprobe *p)
719 struct optimized_kprobe *op;
721 op = container_of(p, struct optimized_kprobe, kp);
722 if (!list_empty(&op->list))
723 /* Dequeue from the (un)optimization queue */
724 list_del_init(&op->list);
726 op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
727 /* Don't touch the code, because it is already freed. */
728 arch_remove_optimized_kprobe(op);
731 /* Try to prepare optimized instructions */
732 static __kprobes void prepare_optimized_kprobe(struct kprobe *p)
734 struct optimized_kprobe *op;
736 op = container_of(p, struct optimized_kprobe, kp);
737 arch_prepare_optimized_kprobe(op);
740 /* Allocate new optimized_kprobe and try to prepare optimized instructions */
741 static __kprobes struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
743 struct optimized_kprobe *op;
745 op = kzalloc(sizeof(struct optimized_kprobe), GFP_KERNEL);
749 INIT_LIST_HEAD(&op->list);
750 op->kp.addr = p->addr;
751 arch_prepare_optimized_kprobe(op);
756 static void __kprobes init_aggr_kprobe(struct kprobe *ap, struct kprobe *p);
759 * Prepare an optimized_kprobe and optimize it
760 * NOTE: p must be a normal registered kprobe
762 static __kprobes void try_to_optimize_kprobe(struct kprobe *p)
765 struct optimized_kprobe *op;
767 ap = alloc_aggr_kprobe(p);
771 op = container_of(ap, struct optimized_kprobe, kp);
772 if (!arch_prepared_optinsn(&op->optinsn)) {
773 /* If failed to setup optimizing, fallback to kprobe */
774 arch_remove_optimized_kprobe(op);
779 init_aggr_kprobe(ap, p);
784 /* This should be called with kprobe_mutex locked */
785 static void __kprobes optimize_all_kprobes(void)
787 struct hlist_head *head;
788 struct hlist_node *node;
792 /* If optimization is already allowed, just return */
793 if (kprobes_allow_optimization)
796 kprobes_allow_optimization = true;
797 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
798 head = &kprobe_table[i];
799 hlist_for_each_entry_rcu(p, node, head, hlist)
800 if (!kprobe_disabled(p))
803 printk(KERN_INFO "Kprobes globally optimized\n");
806 /* This should be called with kprobe_mutex locked */
807 static void __kprobes unoptimize_all_kprobes(void)
809 struct hlist_head *head;
810 struct hlist_node *node;
814 /* If optimization is already prohibited, just return */
815 if (!kprobes_allow_optimization)
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, node, head, hlist) {
822 if (!kprobe_disabled(p))
823 unoptimize_kprobe(p, false);
826 /* Wait for unoptimizing completion */
827 wait_for_kprobe_optimizer();
828 printk(KERN_INFO "Kprobes globally unoptimized\n");
831 int sysctl_kprobes_optimization;
832 int proc_kprobes_optimization_handler(struct ctl_table *table, int write,
833 void __user *buffer, size_t *length,
838 mutex_lock(&kprobe_mutex);
839 sysctl_kprobes_optimization = kprobes_allow_optimization ? 1 : 0;
840 ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
842 if (sysctl_kprobes_optimization)
843 optimize_all_kprobes();
845 unoptimize_all_kprobes();
846 mutex_unlock(&kprobe_mutex);
850 #endif /* CONFIG_SYSCTL */
852 /* Put a breakpoint for a probe. Must be called with text_mutex locked */
853 static void __kprobes __arm_kprobe(struct kprobe *p)
857 /* Check collision with other optimized kprobes */
858 _p = get_optimized_kprobe((unsigned long)p->addr);
860 /* Fallback to unoptimized kprobe */
861 unoptimize_kprobe(_p, true);
864 optimize_kprobe(p); /* Try to optimize (add kprobe to a list) */
867 /* Remove the breakpoint of a probe. Must be called with text_mutex locked */
868 static void __kprobes __disarm_kprobe(struct kprobe *p, bool reopt)
872 unoptimize_kprobe(p, false); /* Try to unoptimize */
874 if (!kprobe_queued(p)) {
875 arch_disarm_kprobe(p);
876 /* If another kprobe was blocked, optimize it. */
877 _p = get_optimized_kprobe((unsigned long)p->addr);
878 if (unlikely(_p) && reopt)
881 /* TODO: reoptimize others after unoptimized this probe */
884 #else /* !CONFIG_OPTPROBES */
886 #define optimize_kprobe(p) do {} while (0)
887 #define unoptimize_kprobe(p, f) do {} while (0)
888 #define kill_optimized_kprobe(p) do {} while (0)
889 #define prepare_optimized_kprobe(p) do {} while (0)
890 #define try_to_optimize_kprobe(p) do {} while (0)
891 #define __arm_kprobe(p) arch_arm_kprobe(p)
892 #define __disarm_kprobe(p, o) arch_disarm_kprobe(p)
893 #define kprobe_disarmed(p) kprobe_disabled(p)
894 #define wait_for_kprobe_optimizer() do {} while (0)
896 /* There should be no unused kprobes can be reused without optimization */
897 static void reuse_unused_kprobe(struct kprobe *ap)
899 printk(KERN_ERR "Error: There should be no unused kprobe here.\n");
900 BUG_ON(kprobe_unused(ap));
903 static __kprobes void free_aggr_kprobe(struct kprobe *p)
905 arch_remove_kprobe(p);
909 static __kprobes struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
911 return kzalloc(sizeof(struct kprobe), GFP_KERNEL);
913 #endif /* CONFIG_OPTPROBES */
915 /* Arm a kprobe with text_mutex */
916 static void __kprobes arm_kprobe(struct kprobe *kp)
919 * Here, since __arm_kprobe() doesn't use stop_machine(),
920 * this doesn't cause deadlock on text_mutex. So, we don't
921 * need get_online_cpus().
923 mutex_lock(&text_mutex);
925 mutex_unlock(&text_mutex);
928 /* Disarm a kprobe with text_mutex */
929 static void __kprobes disarm_kprobe(struct kprobe *kp)
932 mutex_lock(&text_mutex);
933 __disarm_kprobe(kp, true);
934 mutex_unlock(&text_mutex);
938 * Aggregate handlers for multiple kprobes support - these handlers
939 * take care of invoking the individual kprobe handlers on p->list
941 static int __kprobes aggr_pre_handler(struct kprobe *p, struct pt_regs *regs)
945 list_for_each_entry_rcu(kp, &p->list, list) {
946 if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
947 set_kprobe_instance(kp);
948 if (kp->pre_handler(kp, regs))
951 reset_kprobe_instance();
956 static void __kprobes aggr_post_handler(struct kprobe *p, struct pt_regs *regs,
961 list_for_each_entry_rcu(kp, &p->list, list) {
962 if (kp->post_handler && likely(!kprobe_disabled(kp))) {
963 set_kprobe_instance(kp);
964 kp->post_handler(kp, regs, flags);
965 reset_kprobe_instance();
970 static int __kprobes aggr_fault_handler(struct kprobe *p, struct pt_regs *regs,
973 struct kprobe *cur = __get_cpu_var(kprobe_instance);
976 * if we faulted "during" the execution of a user specified
977 * probe handler, invoke just that probe's fault handler
979 if (cur && cur->fault_handler) {
980 if (cur->fault_handler(cur, regs, trapnr))
986 static int __kprobes aggr_break_handler(struct kprobe *p, struct pt_regs *regs)
988 struct kprobe *cur = __get_cpu_var(kprobe_instance);
991 if (cur && cur->break_handler) {
992 if (cur->break_handler(cur, regs))
995 reset_kprobe_instance();
999 /* Walks the list and increments nmissed count for multiprobe case */
1000 void __kprobes kprobes_inc_nmissed_count(struct kprobe *p)
1003 if (!kprobe_aggrprobe(p)) {
1006 list_for_each_entry_rcu(kp, &p->list, list)
1012 void __kprobes recycle_rp_inst(struct kretprobe_instance *ri,
1013 struct hlist_head *head)
1015 struct kretprobe *rp = ri->rp;
1017 /* remove rp inst off the rprobe_inst_table */
1018 hlist_del(&ri->hlist);
1019 INIT_HLIST_NODE(&ri->hlist);
1021 spin_lock(&rp->lock);
1022 hlist_add_head(&ri->hlist, &rp->free_instances);
1023 spin_unlock(&rp->lock);
1026 hlist_add_head(&ri->hlist, head);
1029 void __kprobes kretprobe_hash_lock(struct task_struct *tsk,
1030 struct hlist_head **head, unsigned long *flags)
1031 __acquires(hlist_lock)
1033 unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
1034 spinlock_t *hlist_lock;
1036 *head = &kretprobe_inst_table[hash];
1037 hlist_lock = kretprobe_table_lock_ptr(hash);
1038 spin_lock_irqsave(hlist_lock, *flags);
1041 static void __kprobes kretprobe_table_lock(unsigned long hash,
1042 unsigned long *flags)
1043 __acquires(hlist_lock)
1045 spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
1046 spin_lock_irqsave(hlist_lock, *flags);
1049 void __kprobes kretprobe_hash_unlock(struct task_struct *tsk,
1050 unsigned long *flags)
1051 __releases(hlist_lock)
1053 unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
1054 spinlock_t *hlist_lock;
1056 hlist_lock = kretprobe_table_lock_ptr(hash);
1057 spin_unlock_irqrestore(hlist_lock, *flags);
1060 static void __kprobes kretprobe_table_unlock(unsigned long hash,
1061 unsigned long *flags)
1062 __releases(hlist_lock)
1064 spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
1065 spin_unlock_irqrestore(hlist_lock, *flags);
1069 * This function is called from finish_task_switch when task tk becomes dead,
1070 * so that we can recycle any function-return probe instances associated
1071 * with this task. These left over instances represent probed functions
1072 * that have been called but will never return.
1074 void __kprobes kprobe_flush_task(struct task_struct *tk)
1076 struct kretprobe_instance *ri;
1077 struct hlist_head *head, empty_rp;
1078 struct hlist_node *node, *tmp;
1079 unsigned long hash, flags = 0;
1081 if (unlikely(!kprobes_initialized))
1082 /* Early boot. kretprobe_table_locks not yet initialized. */
1085 hash = hash_ptr(tk, KPROBE_HASH_BITS);
1086 head = &kretprobe_inst_table[hash];
1087 kretprobe_table_lock(hash, &flags);
1088 hlist_for_each_entry_safe(ri, node, tmp, head, hlist) {
1090 recycle_rp_inst(ri, &empty_rp);
1092 kretprobe_table_unlock(hash, &flags);
1093 INIT_HLIST_HEAD(&empty_rp);
1094 hlist_for_each_entry_safe(ri, node, tmp, &empty_rp, hlist) {
1095 hlist_del(&ri->hlist);
1100 static inline void free_rp_inst(struct kretprobe *rp)
1102 struct kretprobe_instance *ri;
1103 struct hlist_node *pos, *next;
1105 hlist_for_each_entry_safe(ri, pos, next, &rp->free_instances, hlist) {
1106 hlist_del(&ri->hlist);
1111 static void __kprobes cleanup_rp_inst(struct kretprobe *rp)
1113 unsigned long flags, hash;
1114 struct kretprobe_instance *ri;
1115 struct hlist_node *pos, *next;
1116 struct hlist_head *head;
1119 for (hash = 0; hash < KPROBE_TABLE_SIZE; hash++) {
1120 kretprobe_table_lock(hash, &flags);
1121 head = &kretprobe_inst_table[hash];
1122 hlist_for_each_entry_safe(ri, pos, next, head, hlist) {
1126 kretprobe_table_unlock(hash, &flags);
1132 * Add the new probe to ap->list. Fail if this is the
1133 * second jprobe at the address - two jprobes can't coexist
1135 static int __kprobes add_new_kprobe(struct kprobe *ap, struct kprobe *p)
1137 BUG_ON(kprobe_gone(ap) || kprobe_gone(p));
1139 if (p->break_handler || p->post_handler)
1140 unoptimize_kprobe(ap, true); /* Fall back to normal kprobe */
1142 if (p->break_handler) {
1143 if (ap->break_handler)
1145 list_add_tail_rcu(&p->list, &ap->list);
1146 ap->break_handler = aggr_break_handler;
1148 list_add_rcu(&p->list, &ap->list);
1149 if (p->post_handler && !ap->post_handler)
1150 ap->post_handler = aggr_post_handler;
1152 if (kprobe_disabled(ap) && !kprobe_disabled(p)) {
1153 ap->flags &= ~KPROBE_FLAG_DISABLED;
1154 if (!kprobes_all_disarmed)
1155 /* Arm the breakpoint again. */
1162 * Fill in the required fields of the "manager kprobe". Replace the
1163 * earlier kprobe in the hlist with the manager kprobe
1165 static void __kprobes init_aggr_kprobe(struct kprobe *ap, struct kprobe *p)
1167 /* Copy p's insn slot to ap */
1169 flush_insn_slot(ap);
1171 ap->flags = p->flags & ~KPROBE_FLAG_OPTIMIZED;
1172 ap->pre_handler = aggr_pre_handler;
1173 ap->fault_handler = aggr_fault_handler;
1174 /* We don't care the kprobe which has gone. */
1175 if (p->post_handler && !kprobe_gone(p))
1176 ap->post_handler = aggr_post_handler;
1177 if (p->break_handler && !kprobe_gone(p))
1178 ap->break_handler = aggr_break_handler;
1180 INIT_LIST_HEAD(&ap->list);
1181 INIT_HLIST_NODE(&ap->hlist);
1183 list_add_rcu(&p->list, &ap->list);
1184 hlist_replace_rcu(&p->hlist, &ap->hlist);
1188 * This is the second or subsequent kprobe at the address - handle
1191 static int __kprobes register_aggr_kprobe(struct kprobe *orig_p,
1195 struct kprobe *ap = orig_p;
1197 if (!kprobe_aggrprobe(orig_p)) {
1198 /* If orig_p is not an aggr_kprobe, create new aggr_kprobe. */
1199 ap = alloc_aggr_kprobe(orig_p);
1202 init_aggr_kprobe(ap, orig_p);
1203 } else if (kprobe_unused(ap))
1204 /* This probe is going to die. Rescue it */
1205 reuse_unused_kprobe(ap);
1207 if (kprobe_gone(ap)) {
1209 * Attempting to insert new probe at the same location that
1210 * had a probe in the module vaddr area which already
1211 * freed. So, the instruction slot has already been
1212 * released. We need a new slot for the new probe.
1214 ret = arch_prepare_kprobe(ap);
1217 * Even if fail to allocate new slot, don't need to
1218 * free aggr_probe. It will be used next time, or
1219 * freed by unregister_kprobe.
1223 /* Prepare optimized instructions if possible. */
1224 prepare_optimized_kprobe(ap);
1227 * Clear gone flag to prevent allocating new slot again, and
1228 * set disabled flag because it is not armed yet.
1230 ap->flags = (ap->flags & ~KPROBE_FLAG_GONE)
1231 | KPROBE_FLAG_DISABLED;
1234 /* Copy ap's insn slot to p */
1236 return add_new_kprobe(ap, p);
1239 static int __kprobes in_kprobes_functions(unsigned long addr)
1241 struct kprobe_blackpoint *kb;
1243 if (addr >= (unsigned long)__kprobes_text_start &&
1244 addr < (unsigned long)__kprobes_text_end)
1247 * If there exists a kprobe_blacklist, verify and
1248 * fail any probe registration in the prohibited area
1250 for (kb = kprobe_blacklist; kb->name != NULL; kb++) {
1251 if (kb->start_addr) {
1252 if (addr >= kb->start_addr &&
1253 addr < (kb->start_addr + kb->range))
1261 * If we have a symbol_name argument, look it up and add the offset field
1262 * to it. This way, we can specify a relative address to a symbol.
1264 static kprobe_opcode_t __kprobes *kprobe_addr(struct kprobe *p)
1266 kprobe_opcode_t *addr = p->addr;
1267 if (p->symbol_name) {
1270 kprobe_lookup_name(p->symbol_name, addr);
1275 return (kprobe_opcode_t *)(((char *)addr) + p->offset);
1278 /* Check passed kprobe is valid and return kprobe in kprobe_table. */
1279 static struct kprobe * __kprobes __get_valid_kprobe(struct kprobe *p)
1281 struct kprobe *ap, *list_p;
1283 ap = get_kprobe(p->addr);
1288 list_for_each_entry_rcu(list_p, &ap->list, list)
1290 /* kprobe p is a valid probe */
1298 /* Return error if the kprobe is being re-registered */
1299 static inline int check_kprobe_rereg(struct kprobe *p)
1303 mutex_lock(&kprobe_mutex);
1304 if (__get_valid_kprobe(p))
1306 mutex_unlock(&kprobe_mutex);
1311 int __kprobes register_kprobe(struct kprobe *p)
1314 struct kprobe *old_p;
1315 struct module *probed_mod;
1316 kprobe_opcode_t *addr;
1318 addr = kprobe_addr(p);
1323 ret = check_kprobe_rereg(p);
1329 if (!kernel_text_address((unsigned long) p->addr) ||
1330 in_kprobes_functions((unsigned long) p->addr) ||
1331 ftrace_text_reserved(p->addr, p->addr) ||
1332 jump_label_text_reserved(p->addr, p->addr))
1333 goto fail_with_jump_label;
1335 /* User can pass only KPROBE_FLAG_DISABLED to register_kprobe */
1336 p->flags &= KPROBE_FLAG_DISABLED;
1339 * Check if are we probing a module.
1341 probed_mod = __module_text_address((unsigned long) p->addr);
1344 * We must hold a refcount of the probed module while updating
1345 * its code to prohibit unexpected unloading.
1347 if (unlikely(!try_module_get(probed_mod)))
1348 goto fail_with_jump_label;
1351 * If the module freed .init.text, we couldn't insert
1354 if (within_module_init((unsigned long)p->addr, probed_mod) &&
1355 probed_mod->state != MODULE_STATE_COMING) {
1356 module_put(probed_mod);
1357 goto fail_with_jump_label;
1361 jump_label_unlock();
1364 INIT_LIST_HEAD(&p->list);
1365 mutex_lock(&kprobe_mutex);
1367 jump_label_lock(); /* needed to call jump_label_text_reserved() */
1369 get_online_cpus(); /* For avoiding text_mutex deadlock. */
1370 mutex_lock(&text_mutex);
1372 old_p = get_kprobe(p->addr);
1374 /* Since this may unoptimize old_p, locking text_mutex. */
1375 ret = register_aggr_kprobe(old_p, p);
1379 ret = arch_prepare_kprobe(p);
1383 INIT_HLIST_NODE(&p->hlist);
1384 hlist_add_head_rcu(&p->hlist,
1385 &kprobe_table[hash_ptr(p->addr, KPROBE_HASH_BITS)]);
1387 if (!kprobes_all_disarmed && !kprobe_disabled(p))
1390 /* Try to optimize kprobe */
1391 try_to_optimize_kprobe(p);
1394 mutex_unlock(&text_mutex);
1396 jump_label_unlock();
1397 mutex_unlock(&kprobe_mutex);
1400 module_put(probed_mod);
1404 fail_with_jump_label:
1406 jump_label_unlock();
1409 EXPORT_SYMBOL_GPL(register_kprobe);
1411 /* Check if all probes on the aggrprobe are disabled */
1412 static int __kprobes aggr_kprobe_disabled(struct kprobe *ap)
1416 list_for_each_entry_rcu(kp, &ap->list, list)
1417 if (!kprobe_disabled(kp))
1419 * There is an active probe on the list.
1420 * We can't disable this ap.
1427 /* Disable one kprobe: Make sure called under kprobe_mutex is locked */
1428 static struct kprobe *__kprobes __disable_kprobe(struct kprobe *p)
1430 struct kprobe *orig_p;
1432 /* Get an original kprobe for return */
1433 orig_p = __get_valid_kprobe(p);
1434 if (unlikely(orig_p == NULL))
1437 if (!kprobe_disabled(p)) {
1438 /* Disable probe if it is a child probe */
1440 p->flags |= KPROBE_FLAG_DISABLED;
1442 /* Try to disarm and disable this/parent probe */
1443 if (p == orig_p || aggr_kprobe_disabled(orig_p)) {
1444 disarm_kprobe(orig_p);
1445 orig_p->flags |= KPROBE_FLAG_DISABLED;
1453 * Unregister a kprobe without a scheduler synchronization.
1455 static int __kprobes __unregister_kprobe_top(struct kprobe *p)
1457 struct kprobe *ap, *list_p;
1459 /* Disable kprobe. This will disarm it if needed. */
1460 ap = __disable_kprobe(p);
1466 * This probe is an independent(and non-optimized) kprobe
1467 * (not an aggrprobe). Remove from the hash list.
1471 /* Following process expects this probe is an aggrprobe */
1472 WARN_ON(!kprobe_aggrprobe(ap));
1474 if (list_is_singular(&ap->list) && kprobe_disarmed(ap))
1476 * !disarmed could be happen if the probe is under delayed
1481 /* If disabling probe has special handlers, update aggrprobe */
1482 if (p->break_handler && !kprobe_gone(p))
1483 ap->break_handler = NULL;
1484 if (p->post_handler && !kprobe_gone(p)) {
1485 list_for_each_entry_rcu(list_p, &ap->list, list) {
1486 if ((list_p != p) && (list_p->post_handler))
1489 ap->post_handler = NULL;
1493 * Remove from the aggrprobe: this path will do nothing in
1494 * __unregister_kprobe_bottom().
1496 list_del_rcu(&p->list);
1497 if (!kprobe_disabled(ap) && !kprobes_all_disarmed)
1499 * Try to optimize this probe again, because post
1500 * handler may have been changed.
1502 optimize_kprobe(ap);
1507 BUG_ON(!kprobe_disarmed(ap));
1508 hlist_del_rcu(&ap->hlist);
1512 static void __kprobes __unregister_kprobe_bottom(struct kprobe *p)
1516 if (list_empty(&p->list))
1517 /* This is an independent kprobe */
1518 arch_remove_kprobe(p);
1519 else if (list_is_singular(&p->list)) {
1520 /* This is the last child of an aggrprobe */
1521 ap = list_entry(p->list.next, struct kprobe, list);
1523 free_aggr_kprobe(ap);
1525 /* Otherwise, do nothing. */
1528 int __kprobes register_kprobes(struct kprobe **kps, int num)
1534 for (i = 0; i < num; i++) {
1535 ret = register_kprobe(kps[i]);
1538 unregister_kprobes(kps, i);
1544 EXPORT_SYMBOL_GPL(register_kprobes);
1546 void __kprobes unregister_kprobe(struct kprobe *p)
1548 unregister_kprobes(&p, 1);
1550 EXPORT_SYMBOL_GPL(unregister_kprobe);
1552 void __kprobes unregister_kprobes(struct kprobe **kps, int num)
1558 mutex_lock(&kprobe_mutex);
1559 for (i = 0; i < num; i++)
1560 if (__unregister_kprobe_top(kps[i]) < 0)
1561 kps[i]->addr = NULL;
1562 mutex_unlock(&kprobe_mutex);
1564 synchronize_sched();
1565 for (i = 0; i < num; i++)
1567 __unregister_kprobe_bottom(kps[i]);
1569 EXPORT_SYMBOL_GPL(unregister_kprobes);
1571 static struct notifier_block kprobe_exceptions_nb = {
1572 .notifier_call = kprobe_exceptions_notify,
1573 .priority = 0x7fffffff /* we need to be notified first */
1576 unsigned long __weak arch_deref_entry_point(void *entry)
1578 return (unsigned long)entry;
1581 int __kprobes register_jprobes(struct jprobe **jps, int num)
1588 for (i = 0; i < num; i++) {
1589 unsigned long addr, offset;
1591 addr = arch_deref_entry_point(jp->entry);
1593 /* Verify probepoint is a function entry point */
1594 if (kallsyms_lookup_size_offset(addr, NULL, &offset) &&
1596 jp->kp.pre_handler = setjmp_pre_handler;
1597 jp->kp.break_handler = longjmp_break_handler;
1598 ret = register_kprobe(&jp->kp);
1604 unregister_jprobes(jps, i);
1610 EXPORT_SYMBOL_GPL(register_jprobes);
1612 int __kprobes register_jprobe(struct jprobe *jp)
1614 return register_jprobes(&jp, 1);
1616 EXPORT_SYMBOL_GPL(register_jprobe);
1618 void __kprobes unregister_jprobe(struct jprobe *jp)
1620 unregister_jprobes(&jp, 1);
1622 EXPORT_SYMBOL_GPL(unregister_jprobe);
1624 void __kprobes unregister_jprobes(struct jprobe **jps, int num)
1630 mutex_lock(&kprobe_mutex);
1631 for (i = 0; i < num; i++)
1632 if (__unregister_kprobe_top(&jps[i]->kp) < 0)
1633 jps[i]->kp.addr = NULL;
1634 mutex_unlock(&kprobe_mutex);
1636 synchronize_sched();
1637 for (i = 0; i < num; i++) {
1638 if (jps[i]->kp.addr)
1639 __unregister_kprobe_bottom(&jps[i]->kp);
1642 EXPORT_SYMBOL_GPL(unregister_jprobes);
1644 #ifdef CONFIG_KRETPROBES
1646 * This kprobe pre_handler is registered with every kretprobe. When probe
1647 * hits it will set up the return probe.
1649 static int __kprobes pre_handler_kretprobe(struct kprobe *p,
1650 struct pt_regs *regs)
1652 struct kretprobe *rp = container_of(p, struct kretprobe, kp);
1653 unsigned long hash, flags = 0;
1654 struct kretprobe_instance *ri;
1656 /*TODO: consider to only swap the RA after the last pre_handler fired */
1657 hash = hash_ptr(current, KPROBE_HASH_BITS);
1658 spin_lock_irqsave(&rp->lock, flags);
1659 if (!hlist_empty(&rp->free_instances)) {
1660 ri = hlist_entry(rp->free_instances.first,
1661 struct kretprobe_instance, hlist);
1662 hlist_del(&ri->hlist);
1663 spin_unlock_irqrestore(&rp->lock, flags);
1668 if (rp->entry_handler && rp->entry_handler(ri, regs))
1671 arch_prepare_kretprobe(ri, regs);
1673 /* XXX(hch): why is there no hlist_move_head? */
1674 INIT_HLIST_NODE(&ri->hlist);
1675 kretprobe_table_lock(hash, &flags);
1676 hlist_add_head(&ri->hlist, &kretprobe_inst_table[hash]);
1677 kretprobe_table_unlock(hash, &flags);
1680 spin_unlock_irqrestore(&rp->lock, flags);
1685 int __kprobes register_kretprobe(struct kretprobe *rp)
1688 struct kretprobe_instance *inst;
1692 if (kretprobe_blacklist_size) {
1693 addr = kprobe_addr(&rp->kp);
1697 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
1698 if (kretprobe_blacklist[i].addr == addr)
1703 rp->kp.pre_handler = pre_handler_kretprobe;
1704 rp->kp.post_handler = NULL;
1705 rp->kp.fault_handler = NULL;
1706 rp->kp.break_handler = NULL;
1708 /* Pre-allocate memory for max kretprobe instances */
1709 if (rp->maxactive <= 0) {
1710 #ifdef CONFIG_PREEMPT
1711 rp->maxactive = max_t(unsigned int, 10, 2*num_possible_cpus());
1713 rp->maxactive = num_possible_cpus();
1716 spin_lock_init(&rp->lock);
1717 INIT_HLIST_HEAD(&rp->free_instances);
1718 for (i = 0; i < rp->maxactive; i++) {
1719 inst = kmalloc(sizeof(struct kretprobe_instance) +
1720 rp->data_size, GFP_KERNEL);
1725 INIT_HLIST_NODE(&inst->hlist);
1726 hlist_add_head(&inst->hlist, &rp->free_instances);
1730 /* Establish function entry probe point */
1731 ret = register_kprobe(&rp->kp);
1736 EXPORT_SYMBOL_GPL(register_kretprobe);
1738 int __kprobes register_kretprobes(struct kretprobe **rps, int num)
1744 for (i = 0; i < num; i++) {
1745 ret = register_kretprobe(rps[i]);
1748 unregister_kretprobes(rps, i);
1754 EXPORT_SYMBOL_GPL(register_kretprobes);
1756 void __kprobes unregister_kretprobe(struct kretprobe *rp)
1758 unregister_kretprobes(&rp, 1);
1760 EXPORT_SYMBOL_GPL(unregister_kretprobe);
1762 void __kprobes unregister_kretprobes(struct kretprobe **rps, int num)
1768 mutex_lock(&kprobe_mutex);
1769 for (i = 0; i < num; i++)
1770 if (__unregister_kprobe_top(&rps[i]->kp) < 0)
1771 rps[i]->kp.addr = NULL;
1772 mutex_unlock(&kprobe_mutex);
1774 synchronize_sched();
1775 for (i = 0; i < num; i++) {
1776 if (rps[i]->kp.addr) {
1777 __unregister_kprobe_bottom(&rps[i]->kp);
1778 cleanup_rp_inst(rps[i]);
1782 EXPORT_SYMBOL_GPL(unregister_kretprobes);
1784 #else /* CONFIG_KRETPROBES */
1785 int __kprobes register_kretprobe(struct kretprobe *rp)
1789 EXPORT_SYMBOL_GPL(register_kretprobe);
1791 int __kprobes register_kretprobes(struct kretprobe **rps, int num)
1795 EXPORT_SYMBOL_GPL(register_kretprobes);
1797 void __kprobes unregister_kretprobe(struct kretprobe *rp)
1800 EXPORT_SYMBOL_GPL(unregister_kretprobe);
1802 void __kprobes unregister_kretprobes(struct kretprobe **rps, int num)
1805 EXPORT_SYMBOL_GPL(unregister_kretprobes);
1807 static int __kprobes pre_handler_kretprobe(struct kprobe *p,
1808 struct pt_regs *regs)
1813 #endif /* CONFIG_KRETPROBES */
1815 /* Set the kprobe gone and remove its instruction buffer. */
1816 static void __kprobes kill_kprobe(struct kprobe *p)
1820 p->flags |= KPROBE_FLAG_GONE;
1821 if (kprobe_aggrprobe(p)) {
1823 * If this is an aggr_kprobe, we have to list all the
1824 * chained probes and mark them GONE.
1826 list_for_each_entry_rcu(kp, &p->list, list)
1827 kp->flags |= KPROBE_FLAG_GONE;
1828 p->post_handler = NULL;
1829 p->break_handler = NULL;
1830 kill_optimized_kprobe(p);
1833 * Here, we can remove insn_slot safely, because no thread calls
1834 * the original probed function (which will be freed soon) any more.
1836 arch_remove_kprobe(p);
1839 /* Disable one kprobe */
1840 int __kprobes disable_kprobe(struct kprobe *kp)
1844 mutex_lock(&kprobe_mutex);
1846 /* Disable this kprobe */
1847 if (__disable_kprobe(kp) == NULL)
1850 mutex_unlock(&kprobe_mutex);
1853 EXPORT_SYMBOL_GPL(disable_kprobe);
1855 /* Enable one kprobe */
1856 int __kprobes enable_kprobe(struct kprobe *kp)
1861 mutex_lock(&kprobe_mutex);
1863 /* Check whether specified probe is valid. */
1864 p = __get_valid_kprobe(kp);
1865 if (unlikely(p == NULL)) {
1870 if (kprobe_gone(kp)) {
1871 /* This kprobe has gone, we couldn't enable it. */
1877 kp->flags &= ~KPROBE_FLAG_DISABLED;
1879 if (!kprobes_all_disarmed && kprobe_disabled(p)) {
1880 p->flags &= ~KPROBE_FLAG_DISABLED;
1884 mutex_unlock(&kprobe_mutex);
1887 EXPORT_SYMBOL_GPL(enable_kprobe);
1889 void __kprobes dump_kprobe(struct kprobe *kp)
1891 printk(KERN_WARNING "Dumping kprobe:\n");
1892 printk(KERN_WARNING "Name: %s\nAddress: %p\nOffset: %x\n",
1893 kp->symbol_name, kp->addr, kp->offset);
1896 /* Module notifier call back, checking kprobes on the module */
1897 static int __kprobes kprobes_module_callback(struct notifier_block *nb,
1898 unsigned long val, void *data)
1900 struct module *mod = data;
1901 struct hlist_head *head;
1902 struct hlist_node *node;
1905 int checkcore = (val == MODULE_STATE_GOING);
1907 if (val != MODULE_STATE_GOING && val != MODULE_STATE_LIVE)
1911 * When MODULE_STATE_GOING was notified, both of module .text and
1912 * .init.text sections would be freed. When MODULE_STATE_LIVE was
1913 * notified, only .init.text section would be freed. We need to
1914 * disable kprobes which have been inserted in the sections.
1916 mutex_lock(&kprobe_mutex);
1917 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
1918 head = &kprobe_table[i];
1919 hlist_for_each_entry_rcu(p, node, head, hlist)
1920 if (within_module_init((unsigned long)p->addr, mod) ||
1922 within_module_core((unsigned long)p->addr, mod))) {
1924 * The vaddr this probe is installed will soon
1925 * be vfreed buy not synced to disk. Hence,
1926 * disarming the breakpoint isn't needed.
1931 mutex_unlock(&kprobe_mutex);
1935 static struct notifier_block kprobe_module_nb = {
1936 .notifier_call = kprobes_module_callback,
1940 static int __init init_kprobes(void)
1943 unsigned long offset = 0, size = 0;
1944 char *modname, namebuf[128];
1945 const char *symbol_name;
1947 struct kprobe_blackpoint *kb;
1949 /* FIXME allocate the probe table, currently defined statically */
1950 /* initialize all list heads */
1951 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
1952 INIT_HLIST_HEAD(&kprobe_table[i]);
1953 INIT_HLIST_HEAD(&kretprobe_inst_table[i]);
1954 spin_lock_init(&(kretprobe_table_locks[i].lock));
1958 * Lookup and populate the kprobe_blacklist.
1960 * Unlike the kretprobe blacklist, we'll need to determine
1961 * the range of addresses that belong to the said functions,
1962 * since a kprobe need not necessarily be at the beginning
1965 for (kb = kprobe_blacklist; kb->name != NULL; kb++) {
1966 kprobe_lookup_name(kb->name, addr);
1970 kb->start_addr = (unsigned long)addr;
1971 symbol_name = kallsyms_lookup(kb->start_addr,
1972 &size, &offset, &modname, namebuf);
1979 if (kretprobe_blacklist_size) {
1980 /* lookup the function address from its name */
1981 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
1982 kprobe_lookup_name(kretprobe_blacklist[i].name,
1983 kretprobe_blacklist[i].addr);
1984 if (!kretprobe_blacklist[i].addr)
1985 printk("kretprobe: lookup failed: %s\n",
1986 kretprobe_blacklist[i].name);
1990 #if defined(CONFIG_OPTPROBES)
1991 #if defined(__ARCH_WANT_KPROBES_INSN_SLOT)
1992 /* Init kprobe_optinsn_slots */
1993 kprobe_optinsn_slots.insn_size = MAX_OPTINSN_SIZE;
1995 /* By default, kprobes can be optimized */
1996 kprobes_allow_optimization = true;
1999 /* By default, kprobes are armed */
2000 kprobes_all_disarmed = false;
2002 err = arch_init_kprobes();
2004 err = register_die_notifier(&kprobe_exceptions_nb);
2006 err = register_module_notifier(&kprobe_module_nb);
2008 kprobes_initialized = (err == 0);
2015 #ifdef CONFIG_DEBUG_FS
2016 static void __kprobes report_probe(struct seq_file *pi, struct kprobe *p,
2017 const char *sym, int offset, char *modname, struct kprobe *pp)
2021 if (p->pre_handler == pre_handler_kretprobe)
2023 else if (p->pre_handler == setjmp_pre_handler)
2029 seq_printf(pi, "%p %s %s+0x%x %s ",
2030 p->addr, kprobe_type, sym, offset,
2031 (modname ? modname : " "));
2033 seq_printf(pi, "%p %s %p ",
2034 p->addr, kprobe_type, p->addr);
2038 seq_printf(pi, "%s%s%s\n",
2039 (kprobe_gone(p) ? "[GONE]" : ""),
2040 ((kprobe_disabled(p) && !kprobe_gone(p)) ? "[DISABLED]" : ""),
2041 (kprobe_optimized(pp) ? "[OPTIMIZED]" : ""));
2044 static void __kprobes *kprobe_seq_start(struct seq_file *f, loff_t *pos)
2046 return (*pos < KPROBE_TABLE_SIZE) ? pos : NULL;
2049 static void __kprobes *kprobe_seq_next(struct seq_file *f, void *v, loff_t *pos)
2052 if (*pos >= KPROBE_TABLE_SIZE)
2057 static void __kprobes kprobe_seq_stop(struct seq_file *f, void *v)
2062 static int __kprobes show_kprobe_addr(struct seq_file *pi, void *v)
2064 struct hlist_head *head;
2065 struct hlist_node *node;
2066 struct kprobe *p, *kp;
2067 const char *sym = NULL;
2068 unsigned int i = *(loff_t *) v;
2069 unsigned long offset = 0;
2070 char *modname, namebuf[128];
2072 head = &kprobe_table[i];
2074 hlist_for_each_entry_rcu(p, node, head, hlist) {
2075 sym = kallsyms_lookup((unsigned long)p->addr, NULL,
2076 &offset, &modname, namebuf);
2077 if (kprobe_aggrprobe(p)) {
2078 list_for_each_entry_rcu(kp, &p->list, list)
2079 report_probe(pi, kp, sym, offset, modname, p);
2081 report_probe(pi, p, sym, offset, modname, NULL);
2087 static const struct seq_operations kprobes_seq_ops = {
2088 .start = kprobe_seq_start,
2089 .next = kprobe_seq_next,
2090 .stop = kprobe_seq_stop,
2091 .show = show_kprobe_addr
2094 static int __kprobes kprobes_open(struct inode *inode, struct file *filp)
2096 return seq_open(filp, &kprobes_seq_ops);
2099 static const struct file_operations debugfs_kprobes_operations = {
2100 .open = kprobes_open,
2102 .llseek = seq_lseek,
2103 .release = seq_release,
2106 static void __kprobes arm_all_kprobes(void)
2108 struct hlist_head *head;
2109 struct hlist_node *node;
2113 mutex_lock(&kprobe_mutex);
2115 /* If kprobes are armed, just return */
2116 if (!kprobes_all_disarmed)
2117 goto already_enabled;
2119 /* Arming kprobes doesn't optimize kprobe itself */
2120 mutex_lock(&text_mutex);
2121 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2122 head = &kprobe_table[i];
2123 hlist_for_each_entry_rcu(p, node, head, hlist)
2124 if (!kprobe_disabled(p))
2127 mutex_unlock(&text_mutex);
2129 kprobes_all_disarmed = false;
2130 printk(KERN_INFO "Kprobes globally enabled\n");
2133 mutex_unlock(&kprobe_mutex);
2137 static void __kprobes disarm_all_kprobes(void)
2139 struct hlist_head *head;
2140 struct hlist_node *node;
2144 mutex_lock(&kprobe_mutex);
2146 /* If kprobes are already disarmed, just return */
2147 if (kprobes_all_disarmed) {
2148 mutex_unlock(&kprobe_mutex);
2152 kprobes_all_disarmed = true;
2153 printk(KERN_INFO "Kprobes globally disabled\n");
2155 mutex_lock(&text_mutex);
2156 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2157 head = &kprobe_table[i];
2158 hlist_for_each_entry_rcu(p, node, head, hlist) {
2159 if (!arch_trampoline_kprobe(p) && !kprobe_disabled(p))
2160 __disarm_kprobe(p, false);
2163 mutex_unlock(&text_mutex);
2164 mutex_unlock(&kprobe_mutex);
2166 /* Wait for disarming all kprobes by optimizer */
2167 wait_for_kprobe_optimizer();
2171 * XXX: The debugfs bool file interface doesn't allow for callbacks
2172 * when the bool state is switched. We can reuse that facility when
2175 static ssize_t read_enabled_file_bool(struct file *file,
2176 char __user *user_buf, size_t count, loff_t *ppos)
2180 if (!kprobes_all_disarmed)
2186 return simple_read_from_buffer(user_buf, count, ppos, buf, 2);
2189 static ssize_t write_enabled_file_bool(struct file *file,
2190 const char __user *user_buf, size_t count, loff_t *ppos)
2195 buf_size = min(count, (sizeof(buf)-1));
2196 if (copy_from_user(buf, user_buf, buf_size))
2208 disarm_all_kprobes();
2215 static const struct file_operations fops_kp = {
2216 .read = read_enabled_file_bool,
2217 .write = write_enabled_file_bool,
2218 .llseek = default_llseek,
2221 static int __kprobes debugfs_kprobe_init(void)
2223 struct dentry *dir, *file;
2224 unsigned int value = 1;
2226 dir = debugfs_create_dir("kprobes", NULL);
2230 file = debugfs_create_file("list", 0444, dir, NULL,
2231 &debugfs_kprobes_operations);
2233 debugfs_remove(dir);
2237 file = debugfs_create_file("enabled", 0600, dir,
2240 debugfs_remove(dir);
2247 late_initcall(debugfs_kprobe_init);
2248 #endif /* CONFIG_DEBUG_FS */
2250 module_init(init_kprobes);
2252 /* defined in arch/.../kernel/kprobes.c */
2253 EXPORT_SYMBOL_GPL(jprobe_return);