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/kdebug.h>
46 #include <linux/memory.h>
48 #include <asm-generic/sections.h>
49 #include <asm/cacheflush.h>
50 #include <asm/errno.h>
51 #include <asm/uaccess.h>
53 #define KPROBE_HASH_BITS 6
54 #define KPROBE_TABLE_SIZE (1 << KPROBE_HASH_BITS)
58 * Some oddball architectures like 64bit powerpc have function descriptors
59 * so this must be overridable.
61 #ifndef kprobe_lookup_name
62 #define kprobe_lookup_name(name, addr) \
63 addr = ((kprobe_opcode_t *)(kallsyms_lookup_name(name)))
66 static int kprobes_initialized;
67 static struct hlist_head kprobe_table[KPROBE_TABLE_SIZE];
68 static struct hlist_head kretprobe_inst_table[KPROBE_TABLE_SIZE];
70 /* NOTE: change this value only with kprobe_mutex held */
71 static bool kprobes_all_disarmed;
73 static DEFINE_MUTEX(kprobe_mutex); /* Protects kprobe_table */
74 static DEFINE_PER_CPU(struct kprobe *, kprobe_instance) = NULL;
76 spinlock_t lock ____cacheline_aligned_in_smp;
77 } kretprobe_table_locks[KPROBE_TABLE_SIZE];
79 static spinlock_t *kretprobe_table_lock_ptr(unsigned long hash)
81 return &(kretprobe_table_locks[hash].lock);
85 * Normally, functions that we'd want to prohibit kprobes in, are marked
86 * __kprobes. But, there are cases where such functions already belong to
87 * a different section (__sched for preempt_schedule)
89 * For such cases, we now have a blacklist
91 static struct kprobe_blackpoint kprobe_blacklist[] = {
92 {"preempt_schedule",},
93 {"native_get_debugreg",},
94 {NULL} /* Terminator */
97 #ifdef __ARCH_WANT_KPROBES_INSN_SLOT
99 * kprobe->ainsn.insn points to the copy of the instruction to be
100 * single-stepped. x86_64, POWER4 and above have no-exec support and
101 * stepping on the instruction on a vmalloced/kmalloced/data page
102 * is a recipe for disaster
104 #define INSNS_PER_PAGE (PAGE_SIZE/(MAX_INSN_SIZE * sizeof(kprobe_opcode_t)))
106 struct kprobe_insn_page {
107 struct list_head list;
108 kprobe_opcode_t *insns; /* Page of instruction slots */
109 char slot_used[INSNS_PER_PAGE];
114 enum kprobe_slot_state {
120 static DEFINE_MUTEX(kprobe_insn_mutex); /* Protects kprobe_insn_pages */
121 static LIST_HEAD(kprobe_insn_pages);
122 static int kprobe_garbage_slots;
123 static int collect_garbage_slots(void);
125 static int __kprobes check_safety(void)
128 #if defined(CONFIG_PREEMPT) && defined(CONFIG_FREEZER)
129 ret = freeze_processes();
131 struct task_struct *p, *q;
132 do_each_thread(p, q) {
133 if (p != current && p->state == TASK_RUNNING &&
135 printk("Check failed: %s is running\n",p->comm);
139 } while_each_thread(p, q);
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(void)
155 struct kprobe_insn_page *kip;
158 list_for_each_entry(kip, &kprobe_insn_pages, list) {
159 if (kip->nused < INSNS_PER_PAGE) {
161 for (i = 0; i < INSNS_PER_PAGE; i++) {
162 if (kip->slot_used[i] == SLOT_CLEAN) {
163 kip->slot_used[i] = SLOT_USED;
165 return kip->insns + (i * MAX_INSN_SIZE);
168 /* Surprise! No unused slots. Fix kip->nused. */
169 kip->nused = INSNS_PER_PAGE;
173 /* If there are any garbage slots, collect it and try again. */
174 if (kprobe_garbage_slots && collect_garbage_slots() == 0) {
177 /* All out of space. Need to allocate a new page. Use slot 0. */
178 kip = kmalloc(sizeof(struct kprobe_insn_page), GFP_KERNEL);
183 * Use module_alloc so this page is within +/- 2GB of where the
184 * kernel image and loaded module images reside. This is required
185 * so x86_64 can correctly handle the %rip-relative fixups.
187 kip->insns = module_alloc(PAGE_SIZE);
192 INIT_LIST_HEAD(&kip->list);
193 list_add(&kip->list, &kprobe_insn_pages);
194 memset(kip->slot_used, SLOT_CLEAN, INSNS_PER_PAGE);
195 kip->slot_used[0] = SLOT_USED;
201 kprobe_opcode_t __kprobes *get_insn_slot(void)
203 kprobe_opcode_t *ret;
204 mutex_lock(&kprobe_insn_mutex);
205 ret = __get_insn_slot();
206 mutex_unlock(&kprobe_insn_mutex);
210 /* Return 1 if all garbages are collected, otherwise 0. */
211 static int __kprobes collect_one_slot(struct kprobe_insn_page *kip, int idx)
213 kip->slot_used[idx] = SLOT_CLEAN;
215 if (kip->nused == 0) {
217 * Page is no longer in use. Free it unless
218 * it's the last one. We keep the last one
219 * so as not to have to set it up again the
220 * next time somebody inserts a probe.
222 if (!list_is_singular(&kprobe_insn_pages)) {
223 list_del(&kip->list);
224 module_free(NULL, kip->insns);
232 static int __kprobes collect_garbage_slots(void)
234 struct kprobe_insn_page *kip, *next;
236 /* Ensure no-one is preepmted on the garbages */
240 list_for_each_entry_safe(kip, next, &kprobe_insn_pages, list) {
242 if (kip->ngarbage == 0)
244 kip->ngarbage = 0; /* we will collect all garbages */
245 for (i = 0; i < INSNS_PER_PAGE; i++) {
246 if (kip->slot_used[i] == SLOT_DIRTY &&
247 collect_one_slot(kip, i))
251 kprobe_garbage_slots = 0;
255 void __kprobes free_insn_slot(kprobe_opcode_t * slot, int dirty)
257 struct kprobe_insn_page *kip;
259 mutex_lock(&kprobe_insn_mutex);
260 list_for_each_entry(kip, &kprobe_insn_pages, list) {
261 if (kip->insns <= slot &&
262 slot < kip->insns + (INSNS_PER_PAGE * MAX_INSN_SIZE)) {
263 int i = (slot - kip->insns) / MAX_INSN_SIZE;
265 kip->slot_used[i] = SLOT_DIRTY;
268 collect_one_slot(kip, i);
273 if (dirty && ++kprobe_garbage_slots > INSNS_PER_PAGE)
274 collect_garbage_slots();
276 mutex_unlock(&kprobe_insn_mutex);
280 /* We have preemption disabled.. so it is safe to use __ versions */
281 static inline void set_kprobe_instance(struct kprobe *kp)
283 __get_cpu_var(kprobe_instance) = kp;
286 static inline void reset_kprobe_instance(void)
288 __get_cpu_var(kprobe_instance) = NULL;
292 * This routine is called either:
293 * - under the kprobe_mutex - during kprobe_[un]register()
295 * - with preemption disabled - from arch/xxx/kernel/kprobes.c
297 struct kprobe __kprobes *get_kprobe(void *addr)
299 struct hlist_head *head;
300 struct hlist_node *node;
303 head = &kprobe_table[hash_ptr(addr, KPROBE_HASH_BITS)];
304 hlist_for_each_entry_rcu(p, node, head, hlist) {
311 /* Arm a kprobe with text_mutex */
312 static void __kprobes arm_kprobe(struct kprobe *kp)
314 mutex_lock(&text_mutex);
316 mutex_unlock(&text_mutex);
319 /* Disarm a kprobe with text_mutex */
320 static void __kprobes disarm_kprobe(struct kprobe *kp)
322 mutex_lock(&text_mutex);
323 arch_disarm_kprobe(kp);
324 mutex_unlock(&text_mutex);
328 * Aggregate handlers for multiple kprobes support - these handlers
329 * take care of invoking the individual kprobe handlers on p->list
331 static int __kprobes aggr_pre_handler(struct kprobe *p, struct pt_regs *regs)
335 list_for_each_entry_rcu(kp, &p->list, list) {
336 if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
337 set_kprobe_instance(kp);
338 if (kp->pre_handler(kp, regs))
341 reset_kprobe_instance();
346 static void __kprobes aggr_post_handler(struct kprobe *p, struct pt_regs *regs,
351 list_for_each_entry_rcu(kp, &p->list, list) {
352 if (kp->post_handler && likely(!kprobe_disabled(kp))) {
353 set_kprobe_instance(kp);
354 kp->post_handler(kp, regs, flags);
355 reset_kprobe_instance();
360 static int __kprobes aggr_fault_handler(struct kprobe *p, struct pt_regs *regs,
363 struct kprobe *cur = __get_cpu_var(kprobe_instance);
366 * if we faulted "during" the execution of a user specified
367 * probe handler, invoke just that probe's fault handler
369 if (cur && cur->fault_handler) {
370 if (cur->fault_handler(cur, regs, trapnr))
376 static int __kprobes aggr_break_handler(struct kprobe *p, struct pt_regs *regs)
378 struct kprobe *cur = __get_cpu_var(kprobe_instance);
381 if (cur && cur->break_handler) {
382 if (cur->break_handler(cur, regs))
385 reset_kprobe_instance();
389 /* Walks the list and increments nmissed count for multiprobe case */
390 void __kprobes kprobes_inc_nmissed_count(struct kprobe *p)
393 if (p->pre_handler != aggr_pre_handler) {
396 list_for_each_entry_rcu(kp, &p->list, list)
402 void __kprobes recycle_rp_inst(struct kretprobe_instance *ri,
403 struct hlist_head *head)
405 struct kretprobe *rp = ri->rp;
407 /* remove rp inst off the rprobe_inst_table */
408 hlist_del(&ri->hlist);
409 INIT_HLIST_NODE(&ri->hlist);
411 spin_lock(&rp->lock);
412 hlist_add_head(&ri->hlist, &rp->free_instances);
413 spin_unlock(&rp->lock);
416 hlist_add_head(&ri->hlist, head);
419 void __kprobes kretprobe_hash_lock(struct task_struct *tsk,
420 struct hlist_head **head, unsigned long *flags)
422 unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
423 spinlock_t *hlist_lock;
425 *head = &kretprobe_inst_table[hash];
426 hlist_lock = kretprobe_table_lock_ptr(hash);
427 spin_lock_irqsave(hlist_lock, *flags);
430 static void __kprobes kretprobe_table_lock(unsigned long hash,
431 unsigned long *flags)
433 spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
434 spin_lock_irqsave(hlist_lock, *flags);
437 void __kprobes kretprobe_hash_unlock(struct task_struct *tsk,
438 unsigned long *flags)
440 unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
441 spinlock_t *hlist_lock;
443 hlist_lock = kretprobe_table_lock_ptr(hash);
444 spin_unlock_irqrestore(hlist_lock, *flags);
447 void __kprobes kretprobe_table_unlock(unsigned long hash, unsigned long *flags)
449 spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
450 spin_unlock_irqrestore(hlist_lock, *flags);
454 * This function is called from finish_task_switch when task tk becomes dead,
455 * so that we can recycle any function-return probe instances associated
456 * with this task. These left over instances represent probed functions
457 * that have been called but will never return.
459 void __kprobes kprobe_flush_task(struct task_struct *tk)
461 struct kretprobe_instance *ri;
462 struct hlist_head *head, empty_rp;
463 struct hlist_node *node, *tmp;
464 unsigned long hash, flags = 0;
466 if (unlikely(!kprobes_initialized))
467 /* Early boot. kretprobe_table_locks not yet initialized. */
470 hash = hash_ptr(tk, KPROBE_HASH_BITS);
471 head = &kretprobe_inst_table[hash];
472 kretprobe_table_lock(hash, &flags);
473 hlist_for_each_entry_safe(ri, node, tmp, head, hlist) {
475 recycle_rp_inst(ri, &empty_rp);
477 kretprobe_table_unlock(hash, &flags);
478 INIT_HLIST_HEAD(&empty_rp);
479 hlist_for_each_entry_safe(ri, node, tmp, &empty_rp, hlist) {
480 hlist_del(&ri->hlist);
485 static inline void free_rp_inst(struct kretprobe *rp)
487 struct kretprobe_instance *ri;
488 struct hlist_node *pos, *next;
490 hlist_for_each_entry_safe(ri, pos, next, &rp->free_instances, hlist) {
491 hlist_del(&ri->hlist);
496 static void __kprobes cleanup_rp_inst(struct kretprobe *rp)
498 unsigned long flags, hash;
499 struct kretprobe_instance *ri;
500 struct hlist_node *pos, *next;
501 struct hlist_head *head;
504 for (hash = 0; hash < KPROBE_TABLE_SIZE; hash++) {
505 kretprobe_table_lock(hash, &flags);
506 head = &kretprobe_inst_table[hash];
507 hlist_for_each_entry_safe(ri, pos, next, head, hlist) {
511 kretprobe_table_unlock(hash, &flags);
517 * Keep all fields in the kprobe consistent
519 static inline void copy_kprobe(struct kprobe *old_p, struct kprobe *p)
521 memcpy(&p->opcode, &old_p->opcode, sizeof(kprobe_opcode_t));
522 memcpy(&p->ainsn, &old_p->ainsn, sizeof(struct arch_specific_insn));
526 * Add the new probe to ap->list. Fail if this is the
527 * second jprobe at the address - two jprobes can't coexist
529 static int __kprobes add_new_kprobe(struct kprobe *ap, struct kprobe *p)
531 BUG_ON(kprobe_gone(ap) || kprobe_gone(p));
532 if (p->break_handler) {
533 if (ap->break_handler)
535 list_add_tail_rcu(&p->list, &ap->list);
536 ap->break_handler = aggr_break_handler;
538 list_add_rcu(&p->list, &ap->list);
539 if (p->post_handler && !ap->post_handler)
540 ap->post_handler = aggr_post_handler;
542 if (kprobe_disabled(ap) && !kprobe_disabled(p)) {
543 ap->flags &= ~KPROBE_FLAG_DISABLED;
544 if (!kprobes_all_disarmed)
545 /* Arm the breakpoint again. */
552 * Fill in the required fields of the "manager kprobe". Replace the
553 * earlier kprobe in the hlist with the manager kprobe
555 static inline void add_aggr_kprobe(struct kprobe *ap, struct kprobe *p)
560 ap->flags = p->flags;
561 ap->pre_handler = aggr_pre_handler;
562 ap->fault_handler = aggr_fault_handler;
563 /* We don't care the kprobe which has gone. */
564 if (p->post_handler && !kprobe_gone(p))
565 ap->post_handler = aggr_post_handler;
566 if (p->break_handler && !kprobe_gone(p))
567 ap->break_handler = aggr_break_handler;
569 INIT_LIST_HEAD(&ap->list);
570 list_add_rcu(&p->list, &ap->list);
572 hlist_replace_rcu(&p->hlist, &ap->hlist);
576 * This is the second or subsequent kprobe at the address - handle
579 static int __kprobes register_aggr_kprobe(struct kprobe *old_p,
583 struct kprobe *ap = old_p;
585 if (old_p->pre_handler != aggr_pre_handler) {
586 /* If old_p is not an aggr_probe, create new aggr_kprobe. */
587 ap = kzalloc(sizeof(struct kprobe), GFP_KERNEL);
590 add_aggr_kprobe(ap, old_p);
593 if (kprobe_gone(ap)) {
595 * Attempting to insert new probe at the same location that
596 * had a probe in the module vaddr area which already
597 * freed. So, the instruction slot has already been
598 * released. We need a new slot for the new probe.
600 ret = arch_prepare_kprobe(ap);
603 * Even if fail to allocate new slot, don't need to
604 * free aggr_probe. It will be used next time, or
605 * freed by unregister_kprobe.
610 * Clear gone flag to prevent allocating new slot again, and
611 * set disabled flag because it is not armed yet.
613 ap->flags = (ap->flags & ~KPROBE_FLAG_GONE)
614 | KPROBE_FLAG_DISABLED;
618 return add_new_kprobe(ap, p);
621 /* Try to disable aggr_kprobe, and return 1 if succeeded.*/
622 static int __kprobes try_to_disable_aggr_kprobe(struct kprobe *p)
626 list_for_each_entry_rcu(kp, &p->list, list) {
627 if (!kprobe_disabled(kp))
629 * There is an active probe on the list.
630 * We can't disable aggr_kprobe.
634 p->flags |= KPROBE_FLAG_DISABLED;
638 static int __kprobes in_kprobes_functions(unsigned long addr)
640 struct kprobe_blackpoint *kb;
642 if (addr >= (unsigned long)__kprobes_text_start &&
643 addr < (unsigned long)__kprobes_text_end)
646 * If there exists a kprobe_blacklist, verify and
647 * fail any probe registration in the prohibited area
649 for (kb = kprobe_blacklist; kb->name != NULL; kb++) {
650 if (kb->start_addr) {
651 if (addr >= kb->start_addr &&
652 addr < (kb->start_addr + kb->range))
660 * If we have a symbol_name argument, look it up and add the offset field
661 * to it. This way, we can specify a relative address to a symbol.
663 static kprobe_opcode_t __kprobes *kprobe_addr(struct kprobe *p)
665 kprobe_opcode_t *addr = p->addr;
666 if (p->symbol_name) {
669 kprobe_lookup_name(p->symbol_name, addr);
674 return (kprobe_opcode_t *)(((char *)addr) + p->offset);
677 int __kprobes register_kprobe(struct kprobe *p)
680 struct kprobe *old_p;
681 struct module *probed_mod;
682 kprobe_opcode_t *addr;
684 addr = kprobe_addr(p);
690 if (!kernel_text_address((unsigned long) p->addr) ||
691 in_kprobes_functions((unsigned long) p->addr)) {
696 /* User can pass only KPROBE_FLAG_DISABLED to register_kprobe */
697 p->flags &= KPROBE_FLAG_DISABLED;
700 * Check if are we probing a module.
702 probed_mod = __module_text_address((unsigned long) p->addr);
705 * We must hold a refcount of the probed module while updating
706 * its code to prohibit unexpected unloading.
708 if (unlikely(!try_module_get(probed_mod))) {
713 * If the module freed .init.text, we couldn't insert
716 if (within_module_init((unsigned long)p->addr, probed_mod) &&
717 probed_mod->state != MODULE_STATE_COMING) {
718 module_put(probed_mod);
726 INIT_LIST_HEAD(&p->list);
727 mutex_lock(&kprobe_mutex);
728 old_p = get_kprobe(p->addr);
730 ret = register_aggr_kprobe(old_p, p);
734 mutex_lock(&text_mutex);
735 ret = arch_prepare_kprobe(p);
737 goto out_unlock_text;
739 INIT_HLIST_NODE(&p->hlist);
740 hlist_add_head_rcu(&p->hlist,
741 &kprobe_table[hash_ptr(p->addr, KPROBE_HASH_BITS)]);
743 if (!kprobes_all_disarmed && !kprobe_disabled(p))
747 mutex_unlock(&text_mutex);
749 mutex_unlock(&kprobe_mutex);
752 module_put(probed_mod);
756 EXPORT_SYMBOL_GPL(register_kprobe);
758 /* Check passed kprobe is valid and return kprobe in kprobe_table. */
759 static struct kprobe * __kprobes __get_valid_kprobe(struct kprobe *p)
761 struct kprobe *old_p, *list_p;
763 old_p = get_kprobe(p->addr);
764 if (unlikely(!old_p))
768 list_for_each_entry_rcu(list_p, &old_p->list, list)
770 /* kprobe p is a valid probe */
779 * Unregister a kprobe without a scheduler synchronization.
781 static int __kprobes __unregister_kprobe_top(struct kprobe *p)
783 struct kprobe *old_p, *list_p;
785 old_p = __get_valid_kprobe(p);
790 (old_p->pre_handler == aggr_pre_handler &&
791 list_is_singular(&old_p->list))) {
793 * Only probe on the hash list. Disarm only if kprobes are
794 * enabled and not gone - otherwise, the breakpoint would
795 * already have been removed. We save on flushing icache.
797 if (!kprobes_all_disarmed && !kprobe_disabled(old_p))
799 hlist_del_rcu(&old_p->hlist);
801 if (p->break_handler && !kprobe_gone(p))
802 old_p->break_handler = NULL;
803 if (p->post_handler && !kprobe_gone(p)) {
804 list_for_each_entry_rcu(list_p, &old_p->list, list) {
805 if ((list_p != p) && (list_p->post_handler))
808 old_p->post_handler = NULL;
811 list_del_rcu(&p->list);
812 if (!kprobe_disabled(old_p)) {
813 try_to_disable_aggr_kprobe(old_p);
814 if (!kprobes_all_disarmed && kprobe_disabled(old_p))
815 disarm_kprobe(old_p);
821 static void __kprobes __unregister_kprobe_bottom(struct kprobe *p)
823 struct kprobe *old_p;
825 if (list_empty(&p->list))
826 arch_remove_kprobe(p);
827 else if (list_is_singular(&p->list)) {
828 /* "p" is the last child of an aggr_kprobe */
829 old_p = list_entry(p->list.next, struct kprobe, list);
831 arch_remove_kprobe(old_p);
836 int __kprobes register_kprobes(struct kprobe **kps, int num)
842 for (i = 0; i < num; i++) {
843 ret = register_kprobe(kps[i]);
846 unregister_kprobes(kps, i);
852 EXPORT_SYMBOL_GPL(register_kprobes);
854 void __kprobes unregister_kprobe(struct kprobe *p)
856 unregister_kprobes(&p, 1);
858 EXPORT_SYMBOL_GPL(unregister_kprobe);
860 void __kprobes unregister_kprobes(struct kprobe **kps, int num)
866 mutex_lock(&kprobe_mutex);
867 for (i = 0; i < num; i++)
868 if (__unregister_kprobe_top(kps[i]) < 0)
870 mutex_unlock(&kprobe_mutex);
873 for (i = 0; i < num; i++)
875 __unregister_kprobe_bottom(kps[i]);
877 EXPORT_SYMBOL_GPL(unregister_kprobes);
879 static struct notifier_block kprobe_exceptions_nb = {
880 .notifier_call = kprobe_exceptions_notify,
881 .priority = 0x7fffffff /* we need to be notified first */
884 unsigned long __weak arch_deref_entry_point(void *entry)
886 return (unsigned long)entry;
889 int __kprobes register_jprobes(struct jprobe **jps, int num)
896 for (i = 0; i < num; i++) {
899 addr = arch_deref_entry_point(jp->entry);
901 if (!kernel_text_address(addr))
904 /* Todo: Verify probepoint is a function entry point */
905 jp->kp.pre_handler = setjmp_pre_handler;
906 jp->kp.break_handler = longjmp_break_handler;
907 ret = register_kprobe(&jp->kp);
911 unregister_jprobes(jps, i);
917 EXPORT_SYMBOL_GPL(register_jprobes);
919 int __kprobes register_jprobe(struct jprobe *jp)
921 return register_jprobes(&jp, 1);
923 EXPORT_SYMBOL_GPL(register_jprobe);
925 void __kprobes unregister_jprobe(struct jprobe *jp)
927 unregister_jprobes(&jp, 1);
929 EXPORT_SYMBOL_GPL(unregister_jprobe);
931 void __kprobes unregister_jprobes(struct jprobe **jps, int num)
937 mutex_lock(&kprobe_mutex);
938 for (i = 0; i < num; i++)
939 if (__unregister_kprobe_top(&jps[i]->kp) < 0)
940 jps[i]->kp.addr = NULL;
941 mutex_unlock(&kprobe_mutex);
944 for (i = 0; i < num; i++) {
946 __unregister_kprobe_bottom(&jps[i]->kp);
949 EXPORT_SYMBOL_GPL(unregister_jprobes);
951 #ifdef CONFIG_KRETPROBES
953 * This kprobe pre_handler is registered with every kretprobe. When probe
954 * hits it will set up the return probe.
956 static int __kprobes pre_handler_kretprobe(struct kprobe *p,
957 struct pt_regs *regs)
959 struct kretprobe *rp = container_of(p, struct kretprobe, kp);
960 unsigned long hash, flags = 0;
961 struct kretprobe_instance *ri;
963 /*TODO: consider to only swap the RA after the last pre_handler fired */
964 hash = hash_ptr(current, KPROBE_HASH_BITS);
965 spin_lock_irqsave(&rp->lock, flags);
966 if (!hlist_empty(&rp->free_instances)) {
967 ri = hlist_entry(rp->free_instances.first,
968 struct kretprobe_instance, hlist);
969 hlist_del(&ri->hlist);
970 spin_unlock_irqrestore(&rp->lock, flags);
975 if (rp->entry_handler && rp->entry_handler(ri, regs))
978 arch_prepare_kretprobe(ri, regs);
980 /* XXX(hch): why is there no hlist_move_head? */
981 INIT_HLIST_NODE(&ri->hlist);
982 kretprobe_table_lock(hash, &flags);
983 hlist_add_head(&ri->hlist, &kretprobe_inst_table[hash]);
984 kretprobe_table_unlock(hash, &flags);
987 spin_unlock_irqrestore(&rp->lock, flags);
992 int __kprobes register_kretprobe(struct kretprobe *rp)
995 struct kretprobe_instance *inst;
999 if (kretprobe_blacklist_size) {
1000 addr = kprobe_addr(&rp->kp);
1004 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
1005 if (kretprobe_blacklist[i].addr == addr)
1010 rp->kp.pre_handler = pre_handler_kretprobe;
1011 rp->kp.post_handler = NULL;
1012 rp->kp.fault_handler = NULL;
1013 rp->kp.break_handler = NULL;
1015 /* Pre-allocate memory for max kretprobe instances */
1016 if (rp->maxactive <= 0) {
1017 #ifdef CONFIG_PREEMPT
1018 rp->maxactive = max(10, 2 * NR_CPUS);
1020 rp->maxactive = NR_CPUS;
1023 spin_lock_init(&rp->lock);
1024 INIT_HLIST_HEAD(&rp->free_instances);
1025 for (i = 0; i < rp->maxactive; i++) {
1026 inst = kmalloc(sizeof(struct kretprobe_instance) +
1027 rp->data_size, GFP_KERNEL);
1032 INIT_HLIST_NODE(&inst->hlist);
1033 hlist_add_head(&inst->hlist, &rp->free_instances);
1037 /* Establish function entry probe point */
1038 ret = register_kprobe(&rp->kp);
1043 EXPORT_SYMBOL_GPL(register_kretprobe);
1045 int __kprobes register_kretprobes(struct kretprobe **rps, int num)
1051 for (i = 0; i < num; i++) {
1052 ret = register_kretprobe(rps[i]);
1055 unregister_kretprobes(rps, i);
1061 EXPORT_SYMBOL_GPL(register_kretprobes);
1063 void __kprobes unregister_kretprobe(struct kretprobe *rp)
1065 unregister_kretprobes(&rp, 1);
1067 EXPORT_SYMBOL_GPL(unregister_kretprobe);
1069 void __kprobes unregister_kretprobes(struct kretprobe **rps, int num)
1075 mutex_lock(&kprobe_mutex);
1076 for (i = 0; i < num; i++)
1077 if (__unregister_kprobe_top(&rps[i]->kp) < 0)
1078 rps[i]->kp.addr = NULL;
1079 mutex_unlock(&kprobe_mutex);
1081 synchronize_sched();
1082 for (i = 0; i < num; i++) {
1083 if (rps[i]->kp.addr) {
1084 __unregister_kprobe_bottom(&rps[i]->kp);
1085 cleanup_rp_inst(rps[i]);
1089 EXPORT_SYMBOL_GPL(unregister_kretprobes);
1091 #else /* CONFIG_KRETPROBES */
1092 int __kprobes register_kretprobe(struct kretprobe *rp)
1096 EXPORT_SYMBOL_GPL(register_kretprobe);
1098 int __kprobes register_kretprobes(struct kretprobe **rps, int num)
1102 EXPORT_SYMBOL_GPL(register_kretprobes);
1104 void __kprobes unregister_kretprobe(struct kretprobe *rp)
1107 EXPORT_SYMBOL_GPL(unregister_kretprobe);
1109 void __kprobes unregister_kretprobes(struct kretprobe **rps, int num)
1112 EXPORT_SYMBOL_GPL(unregister_kretprobes);
1114 static int __kprobes pre_handler_kretprobe(struct kprobe *p,
1115 struct pt_regs *regs)
1120 #endif /* CONFIG_KRETPROBES */
1122 /* Set the kprobe gone and remove its instruction buffer. */
1123 static void __kprobes kill_kprobe(struct kprobe *p)
1127 p->flags |= KPROBE_FLAG_GONE;
1128 if (p->pre_handler == aggr_pre_handler) {
1130 * If this is an aggr_kprobe, we have to list all the
1131 * chained probes and mark them GONE.
1133 list_for_each_entry_rcu(kp, &p->list, list)
1134 kp->flags |= KPROBE_FLAG_GONE;
1135 p->post_handler = NULL;
1136 p->break_handler = NULL;
1139 * Here, we can remove insn_slot safely, because no thread calls
1140 * the original probed function (which will be freed soon) any more.
1142 arch_remove_kprobe(p);
1145 void __kprobes dump_kprobe(struct kprobe *kp)
1147 printk(KERN_WARNING "Dumping kprobe:\n");
1148 printk(KERN_WARNING "Name: %s\nAddress: %p\nOffset: %x\n",
1149 kp->symbol_name, kp->addr, kp->offset);
1152 /* Module notifier call back, checking kprobes on the module */
1153 static int __kprobes kprobes_module_callback(struct notifier_block *nb,
1154 unsigned long val, void *data)
1156 struct module *mod = data;
1157 struct hlist_head *head;
1158 struct hlist_node *node;
1161 int checkcore = (val == MODULE_STATE_GOING);
1163 if (val != MODULE_STATE_GOING && val != MODULE_STATE_LIVE)
1167 * When MODULE_STATE_GOING was notified, both of module .text and
1168 * .init.text sections would be freed. When MODULE_STATE_LIVE was
1169 * notified, only .init.text section would be freed. We need to
1170 * disable kprobes which have been inserted in the sections.
1172 mutex_lock(&kprobe_mutex);
1173 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
1174 head = &kprobe_table[i];
1175 hlist_for_each_entry_rcu(p, node, head, hlist)
1176 if (within_module_init((unsigned long)p->addr, mod) ||
1178 within_module_core((unsigned long)p->addr, mod))) {
1180 * The vaddr this probe is installed will soon
1181 * be vfreed buy not synced to disk. Hence,
1182 * disarming the breakpoint isn't needed.
1187 mutex_unlock(&kprobe_mutex);
1191 static struct notifier_block kprobe_module_nb = {
1192 .notifier_call = kprobes_module_callback,
1196 static int __init init_kprobes(void)
1199 unsigned long offset = 0, size = 0;
1200 char *modname, namebuf[128];
1201 const char *symbol_name;
1203 struct kprobe_blackpoint *kb;
1205 /* FIXME allocate the probe table, currently defined statically */
1206 /* initialize all list heads */
1207 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
1208 INIT_HLIST_HEAD(&kprobe_table[i]);
1209 INIT_HLIST_HEAD(&kretprobe_inst_table[i]);
1210 spin_lock_init(&(kretprobe_table_locks[i].lock));
1214 * Lookup and populate the kprobe_blacklist.
1216 * Unlike the kretprobe blacklist, we'll need to determine
1217 * the range of addresses that belong to the said functions,
1218 * since a kprobe need not necessarily be at the beginning
1221 for (kb = kprobe_blacklist; kb->name != NULL; kb++) {
1222 kprobe_lookup_name(kb->name, addr);
1226 kb->start_addr = (unsigned long)addr;
1227 symbol_name = kallsyms_lookup(kb->start_addr,
1228 &size, &offset, &modname, namebuf);
1235 if (kretprobe_blacklist_size) {
1236 /* lookup the function address from its name */
1237 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
1238 kprobe_lookup_name(kretprobe_blacklist[i].name,
1239 kretprobe_blacklist[i].addr);
1240 if (!kretprobe_blacklist[i].addr)
1241 printk("kretprobe: lookup failed: %s\n",
1242 kretprobe_blacklist[i].name);
1246 /* By default, kprobes are armed */
1247 kprobes_all_disarmed = false;
1249 err = arch_init_kprobes();
1251 err = register_die_notifier(&kprobe_exceptions_nb);
1253 err = register_module_notifier(&kprobe_module_nb);
1255 kprobes_initialized = (err == 0);
1262 #ifdef CONFIG_DEBUG_FS
1263 static void __kprobes report_probe(struct seq_file *pi, struct kprobe *p,
1264 const char *sym, int offset,char *modname)
1268 if (p->pre_handler == pre_handler_kretprobe)
1270 else if (p->pre_handler == setjmp_pre_handler)
1275 seq_printf(pi, "%p %s %s+0x%x %s %s%s\n",
1276 p->addr, kprobe_type, sym, offset,
1277 (modname ? modname : " "),
1278 (kprobe_gone(p) ? "[GONE]" : ""),
1279 ((kprobe_disabled(p) && !kprobe_gone(p)) ?
1280 "[DISABLED]" : ""));
1282 seq_printf(pi, "%p %s %p %s%s\n",
1283 p->addr, kprobe_type, p->addr,
1284 (kprobe_gone(p) ? "[GONE]" : ""),
1285 ((kprobe_disabled(p) && !kprobe_gone(p)) ?
1286 "[DISABLED]" : ""));
1289 static void __kprobes *kprobe_seq_start(struct seq_file *f, loff_t *pos)
1291 return (*pos < KPROBE_TABLE_SIZE) ? pos : NULL;
1294 static void __kprobes *kprobe_seq_next(struct seq_file *f, void *v, loff_t *pos)
1297 if (*pos >= KPROBE_TABLE_SIZE)
1302 static void __kprobes kprobe_seq_stop(struct seq_file *f, void *v)
1307 static int __kprobes show_kprobe_addr(struct seq_file *pi, void *v)
1309 struct hlist_head *head;
1310 struct hlist_node *node;
1311 struct kprobe *p, *kp;
1312 const char *sym = NULL;
1313 unsigned int i = *(loff_t *) v;
1314 unsigned long offset = 0;
1315 char *modname, namebuf[128];
1317 head = &kprobe_table[i];
1319 hlist_for_each_entry_rcu(p, node, head, hlist) {
1320 sym = kallsyms_lookup((unsigned long)p->addr, NULL,
1321 &offset, &modname, namebuf);
1322 if (p->pre_handler == aggr_pre_handler) {
1323 list_for_each_entry_rcu(kp, &p->list, list)
1324 report_probe(pi, kp, sym, offset, modname);
1326 report_probe(pi, p, sym, offset, modname);
1332 static struct seq_operations kprobes_seq_ops = {
1333 .start = kprobe_seq_start,
1334 .next = kprobe_seq_next,
1335 .stop = kprobe_seq_stop,
1336 .show = show_kprobe_addr
1339 static int __kprobes kprobes_open(struct inode *inode, struct file *filp)
1341 return seq_open(filp, &kprobes_seq_ops);
1344 static struct file_operations debugfs_kprobes_operations = {
1345 .open = kprobes_open,
1347 .llseek = seq_lseek,
1348 .release = seq_release,
1351 /* Disable one kprobe */
1352 int __kprobes disable_kprobe(struct kprobe *kp)
1357 mutex_lock(&kprobe_mutex);
1359 /* Check whether specified probe is valid. */
1360 p = __get_valid_kprobe(kp);
1361 if (unlikely(p == NULL)) {
1366 /* If the probe is already disabled (or gone), just return */
1367 if (kprobe_disabled(kp))
1370 kp->flags |= KPROBE_FLAG_DISABLED;
1372 /* When kp != p, p is always enabled. */
1373 try_to_disable_aggr_kprobe(p);
1375 if (!kprobes_all_disarmed && kprobe_disabled(p))
1378 mutex_unlock(&kprobe_mutex);
1381 EXPORT_SYMBOL_GPL(disable_kprobe);
1383 /* Enable one kprobe */
1384 int __kprobes enable_kprobe(struct kprobe *kp)
1389 mutex_lock(&kprobe_mutex);
1391 /* Check whether specified probe is valid. */
1392 p = __get_valid_kprobe(kp);
1393 if (unlikely(p == NULL)) {
1398 if (kprobe_gone(kp)) {
1399 /* This kprobe has gone, we couldn't enable it. */
1404 if (!kprobes_all_disarmed && kprobe_disabled(p))
1407 p->flags &= ~KPROBE_FLAG_DISABLED;
1409 kp->flags &= ~KPROBE_FLAG_DISABLED;
1411 mutex_unlock(&kprobe_mutex);
1414 EXPORT_SYMBOL_GPL(enable_kprobe);
1416 static void __kprobes arm_all_kprobes(void)
1418 struct hlist_head *head;
1419 struct hlist_node *node;
1423 mutex_lock(&kprobe_mutex);
1425 /* If kprobes are armed, just return */
1426 if (!kprobes_all_disarmed)
1427 goto already_enabled;
1429 mutex_lock(&text_mutex);
1430 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
1431 head = &kprobe_table[i];
1432 hlist_for_each_entry_rcu(p, node, head, hlist)
1433 if (!kprobe_disabled(p))
1436 mutex_unlock(&text_mutex);
1438 kprobes_all_disarmed = false;
1439 printk(KERN_INFO "Kprobes globally enabled\n");
1442 mutex_unlock(&kprobe_mutex);
1446 static void __kprobes disarm_all_kprobes(void)
1448 struct hlist_head *head;
1449 struct hlist_node *node;
1453 mutex_lock(&kprobe_mutex);
1455 /* If kprobes are already disarmed, just return */
1456 if (kprobes_all_disarmed)
1457 goto already_disabled;
1459 kprobes_all_disarmed = true;
1460 printk(KERN_INFO "Kprobes globally disabled\n");
1461 mutex_lock(&text_mutex);
1462 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
1463 head = &kprobe_table[i];
1464 hlist_for_each_entry_rcu(p, node, head, hlist) {
1465 if (!arch_trampoline_kprobe(p) && !kprobe_disabled(p))
1466 arch_disarm_kprobe(p);
1470 mutex_unlock(&text_mutex);
1471 mutex_unlock(&kprobe_mutex);
1472 /* Allow all currently running kprobes to complete */
1473 synchronize_sched();
1477 mutex_unlock(&kprobe_mutex);
1482 * XXX: The debugfs bool file interface doesn't allow for callbacks
1483 * when the bool state is switched. We can reuse that facility when
1486 static ssize_t read_enabled_file_bool(struct file *file,
1487 char __user *user_buf, size_t count, loff_t *ppos)
1491 if (!kprobes_all_disarmed)
1497 return simple_read_from_buffer(user_buf, count, ppos, buf, 2);
1500 static ssize_t write_enabled_file_bool(struct file *file,
1501 const char __user *user_buf, size_t count, loff_t *ppos)
1506 buf_size = min(count, (sizeof(buf)-1));
1507 if (copy_from_user(buf, user_buf, buf_size))
1519 disarm_all_kprobes();
1526 static struct file_operations fops_kp = {
1527 .read = read_enabled_file_bool,
1528 .write = write_enabled_file_bool,
1531 static int __kprobes debugfs_kprobe_init(void)
1533 struct dentry *dir, *file;
1534 unsigned int value = 1;
1536 dir = debugfs_create_dir("kprobes", NULL);
1540 file = debugfs_create_file("list", 0444, dir, NULL,
1541 &debugfs_kprobes_operations);
1543 debugfs_remove(dir);
1547 file = debugfs_create_file("enabled", 0600, dir,
1550 debugfs_remove(dir);
1557 late_initcall(debugfs_kprobe_init);
1558 #endif /* CONFIG_DEBUG_FS */
1560 module_init(init_kprobes);
1562 /* defined in arch/.../kernel/kprobes.c */
1563 EXPORT_SYMBOL_GPL(jprobe_return);