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 kprobe_enabled;
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 {NULL} /* Terminator */
96 #ifdef __ARCH_WANT_KPROBES_INSN_SLOT
98 * kprobe->ainsn.insn points to the copy of the instruction to be
99 * single-stepped. x86_64, POWER4 and above have no-exec support and
100 * stepping on the instruction on a vmalloced/kmalloced/data page
101 * is a recipe for disaster
103 #define INSNS_PER_PAGE (PAGE_SIZE/(MAX_INSN_SIZE * sizeof(kprobe_opcode_t)))
105 struct kprobe_insn_page {
106 struct hlist_node hlist;
107 kprobe_opcode_t *insns; /* Page of instruction slots */
108 char slot_used[INSNS_PER_PAGE];
113 enum kprobe_slot_state {
119 static DEFINE_MUTEX(kprobe_insn_mutex); /* Protects kprobe_insn_pages */
120 static struct hlist_head kprobe_insn_pages;
121 static int kprobe_garbage_slots;
122 static int collect_garbage_slots(void);
124 static int __kprobes check_safety(void)
127 #if defined(CONFIG_PREEMPT) && defined(CONFIG_FREEZER)
128 ret = freeze_processes();
130 struct task_struct *p, *q;
131 do_each_thread(p, q) {
132 if (p != current && p->state == TASK_RUNNING &&
134 printk("Check failed: %s is running\n",p->comm);
138 } while_each_thread(p, q);
149 * __get_insn_slot() - Find a slot on an executable page for an instruction.
150 * We allocate an executable page if there's no room on existing ones.
152 static kprobe_opcode_t __kprobes *__get_insn_slot(void)
154 struct kprobe_insn_page *kip;
155 struct hlist_node *pos;
158 hlist_for_each_entry(kip, pos, &kprobe_insn_pages, hlist) {
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_HLIST_NODE(&kip->hlist);
193 hlist_add_head(&kip->hlist, &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 hlist_del(&kip->hlist);
223 if (hlist_empty(&kprobe_insn_pages)) {
224 INIT_HLIST_NODE(&kip->hlist);
225 hlist_add_head(&kip->hlist,
228 module_free(NULL, kip->insns);
236 static int __kprobes collect_garbage_slots(void)
238 struct kprobe_insn_page *kip;
239 struct hlist_node *pos, *next;
242 /* Ensure no-one is preepmted on the garbages */
243 mutex_unlock(&kprobe_insn_mutex);
244 safety = check_safety();
245 mutex_lock(&kprobe_insn_mutex);
249 hlist_for_each_entry_safe(kip, pos, next, &kprobe_insn_pages, hlist) {
251 if (kip->ngarbage == 0)
253 kip->ngarbage = 0; /* we will collect all garbages */
254 for (i = 0; i < INSNS_PER_PAGE; i++) {
255 if (kip->slot_used[i] == SLOT_DIRTY &&
256 collect_one_slot(kip, i))
260 kprobe_garbage_slots = 0;
264 void __kprobes free_insn_slot(kprobe_opcode_t * slot, int dirty)
266 struct kprobe_insn_page *kip;
267 struct hlist_node *pos;
269 mutex_lock(&kprobe_insn_mutex);
270 hlist_for_each_entry(kip, pos, &kprobe_insn_pages, hlist) {
271 if (kip->insns <= slot &&
272 slot < kip->insns + (INSNS_PER_PAGE * MAX_INSN_SIZE)) {
273 int i = (slot - kip->insns) / MAX_INSN_SIZE;
275 kip->slot_used[i] = SLOT_DIRTY;
278 collect_one_slot(kip, i);
284 if (dirty && ++kprobe_garbage_slots > INSNS_PER_PAGE)
285 collect_garbage_slots();
287 mutex_unlock(&kprobe_insn_mutex);
291 /* We have preemption disabled.. so it is safe to use __ versions */
292 static inline void set_kprobe_instance(struct kprobe *kp)
294 __get_cpu_var(kprobe_instance) = kp;
297 static inline void reset_kprobe_instance(void)
299 __get_cpu_var(kprobe_instance) = NULL;
303 * This routine is called either:
304 * - under the kprobe_mutex - during kprobe_[un]register()
306 * - with preemption disabled - from arch/xxx/kernel/kprobes.c
308 struct kprobe __kprobes *get_kprobe(void *addr)
310 struct hlist_head *head;
311 struct hlist_node *node;
314 head = &kprobe_table[hash_ptr(addr, KPROBE_HASH_BITS)];
315 hlist_for_each_entry_rcu(p, node, head, hlist) {
323 * Aggregate handlers for multiple kprobes support - these handlers
324 * take care of invoking the individual kprobe handlers on p->list
326 static int __kprobes aggr_pre_handler(struct kprobe *p, struct pt_regs *regs)
330 list_for_each_entry_rcu(kp, &p->list, list) {
331 if (kp->pre_handler && !kprobe_gone(kp)) {
332 set_kprobe_instance(kp);
333 if (kp->pre_handler(kp, regs))
336 reset_kprobe_instance();
341 static void __kprobes aggr_post_handler(struct kprobe *p, struct pt_regs *regs,
346 list_for_each_entry_rcu(kp, &p->list, list) {
347 if (kp->post_handler && !kprobe_gone(kp)) {
348 set_kprobe_instance(kp);
349 kp->post_handler(kp, regs, flags);
350 reset_kprobe_instance();
355 static int __kprobes aggr_fault_handler(struct kprobe *p, struct pt_regs *regs,
358 struct kprobe *cur = __get_cpu_var(kprobe_instance);
361 * if we faulted "during" the execution of a user specified
362 * probe handler, invoke just that probe's fault handler
364 if (cur && cur->fault_handler) {
365 if (cur->fault_handler(cur, regs, trapnr))
371 static int __kprobes aggr_break_handler(struct kprobe *p, struct pt_regs *regs)
373 struct kprobe *cur = __get_cpu_var(kprobe_instance);
376 if (cur && cur->break_handler) {
377 if (cur->break_handler(cur, regs))
380 reset_kprobe_instance();
384 /* Walks the list and increments nmissed count for multiprobe case */
385 void __kprobes kprobes_inc_nmissed_count(struct kprobe *p)
388 if (p->pre_handler != aggr_pre_handler) {
391 list_for_each_entry_rcu(kp, &p->list, list)
397 void __kprobes recycle_rp_inst(struct kretprobe_instance *ri,
398 struct hlist_head *head)
400 struct kretprobe *rp = ri->rp;
402 /* remove rp inst off the rprobe_inst_table */
403 hlist_del(&ri->hlist);
404 INIT_HLIST_NODE(&ri->hlist);
406 spin_lock(&rp->lock);
407 hlist_add_head(&ri->hlist, &rp->free_instances);
408 spin_unlock(&rp->lock);
411 hlist_add_head(&ri->hlist, head);
414 void __kprobes kretprobe_hash_lock(struct task_struct *tsk,
415 struct hlist_head **head, unsigned long *flags)
417 unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
418 spinlock_t *hlist_lock;
420 *head = &kretprobe_inst_table[hash];
421 hlist_lock = kretprobe_table_lock_ptr(hash);
422 spin_lock_irqsave(hlist_lock, *flags);
425 static void __kprobes kretprobe_table_lock(unsigned long hash,
426 unsigned long *flags)
428 spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
429 spin_lock_irqsave(hlist_lock, *flags);
432 void __kprobes kretprobe_hash_unlock(struct task_struct *tsk,
433 unsigned long *flags)
435 unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
436 spinlock_t *hlist_lock;
438 hlist_lock = kretprobe_table_lock_ptr(hash);
439 spin_unlock_irqrestore(hlist_lock, *flags);
442 void __kprobes kretprobe_table_unlock(unsigned long hash, unsigned long *flags)
444 spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
445 spin_unlock_irqrestore(hlist_lock, *flags);
449 * This function is called from finish_task_switch when task tk becomes dead,
450 * so that we can recycle any function-return probe instances associated
451 * with this task. These left over instances represent probed functions
452 * that have been called but will never return.
454 void __kprobes kprobe_flush_task(struct task_struct *tk)
456 struct kretprobe_instance *ri;
457 struct hlist_head *head, empty_rp;
458 struct hlist_node *node, *tmp;
459 unsigned long hash, flags = 0;
461 if (unlikely(!kprobes_initialized))
462 /* Early boot. kretprobe_table_locks not yet initialized. */
465 hash = hash_ptr(tk, KPROBE_HASH_BITS);
466 head = &kretprobe_inst_table[hash];
467 kretprobe_table_lock(hash, &flags);
468 hlist_for_each_entry_safe(ri, node, tmp, head, hlist) {
470 recycle_rp_inst(ri, &empty_rp);
472 kretprobe_table_unlock(hash, &flags);
473 INIT_HLIST_HEAD(&empty_rp);
474 hlist_for_each_entry_safe(ri, node, tmp, &empty_rp, hlist) {
475 hlist_del(&ri->hlist);
480 static inline void free_rp_inst(struct kretprobe *rp)
482 struct kretprobe_instance *ri;
483 struct hlist_node *pos, *next;
485 hlist_for_each_entry_safe(ri, pos, next, &rp->free_instances, hlist) {
486 hlist_del(&ri->hlist);
491 static void __kprobes cleanup_rp_inst(struct kretprobe *rp)
493 unsigned long flags, hash;
494 struct kretprobe_instance *ri;
495 struct hlist_node *pos, *next;
496 struct hlist_head *head;
499 for (hash = 0; hash < KPROBE_TABLE_SIZE; hash++) {
500 kretprobe_table_lock(hash, &flags);
501 head = &kretprobe_inst_table[hash];
502 hlist_for_each_entry_safe(ri, pos, next, head, hlist) {
506 kretprobe_table_unlock(hash, &flags);
512 * Keep all fields in the kprobe consistent
514 static inline void copy_kprobe(struct kprobe *old_p, struct kprobe *p)
516 memcpy(&p->opcode, &old_p->opcode, sizeof(kprobe_opcode_t));
517 memcpy(&p->ainsn, &old_p->ainsn, sizeof(struct arch_specific_insn));
521 * Add the new probe to old_p->list. Fail if this is the
522 * second jprobe at the address - two jprobes can't coexist
524 static int __kprobes add_new_kprobe(struct kprobe *old_p, struct kprobe *p)
526 if (p->break_handler) {
527 if (old_p->break_handler)
529 list_add_tail_rcu(&p->list, &old_p->list);
530 old_p->break_handler = aggr_break_handler;
532 list_add_rcu(&p->list, &old_p->list);
533 if (p->post_handler && !old_p->post_handler)
534 old_p->post_handler = aggr_post_handler;
539 * Fill in the required fields of the "manager kprobe". Replace the
540 * earlier kprobe in the hlist with the manager kprobe
542 static inline void add_aggr_kprobe(struct kprobe *ap, struct kprobe *p)
547 ap->pre_handler = aggr_pre_handler;
548 ap->fault_handler = aggr_fault_handler;
549 /* We don't care the kprobe which has gone. */
550 if (p->post_handler && !kprobe_gone(p))
551 ap->post_handler = aggr_post_handler;
552 if (p->break_handler && !kprobe_gone(p))
553 ap->break_handler = aggr_break_handler;
555 INIT_LIST_HEAD(&ap->list);
556 list_add_rcu(&p->list, &ap->list);
558 hlist_replace_rcu(&p->hlist, &ap->hlist);
562 * This is the second or subsequent kprobe at the address - handle
565 static int __kprobes register_aggr_kprobe(struct kprobe *old_p,
571 if (kprobe_gone(old_p)) {
573 * Attempting to insert new probe at the same location that
574 * had a probe in the module vaddr area which already
575 * freed. So, the instruction slot has already been
576 * released. We need a new slot for the new probe.
578 ret = arch_prepare_kprobe(old_p);
582 if (old_p->pre_handler == aggr_pre_handler) {
583 copy_kprobe(old_p, p);
584 ret = add_new_kprobe(old_p, p);
587 ap = kzalloc(sizeof(struct kprobe), GFP_KERNEL);
589 if (kprobe_gone(old_p))
590 arch_remove_kprobe(old_p);
593 add_aggr_kprobe(ap, old_p);
595 ret = add_new_kprobe(ap, p);
597 if (kprobe_gone(old_p)) {
599 * If the old_p has gone, its breakpoint has been disarmed.
600 * We have to arm it again after preparing real kprobes.
602 ap->flags &= ~KPROBE_FLAG_GONE;
609 static int __kprobes in_kprobes_functions(unsigned long addr)
611 struct kprobe_blackpoint *kb;
613 if (addr >= (unsigned long)__kprobes_text_start &&
614 addr < (unsigned long)__kprobes_text_end)
617 * If there exists a kprobe_blacklist, verify and
618 * fail any probe registration in the prohibited area
620 for (kb = kprobe_blacklist; kb->name != NULL; kb++) {
621 if (kb->start_addr) {
622 if (addr >= kb->start_addr &&
623 addr < (kb->start_addr + kb->range))
631 * If we have a symbol_name argument, look it up and add the offset field
632 * to it. This way, we can specify a relative address to a symbol.
634 static kprobe_opcode_t __kprobes *kprobe_addr(struct kprobe *p)
636 kprobe_opcode_t *addr = p->addr;
637 if (p->symbol_name) {
640 kprobe_lookup_name(p->symbol_name, addr);
645 return (kprobe_opcode_t *)(((char *)addr) + p->offset);
648 int __kprobes register_kprobe(struct kprobe *p)
651 struct kprobe *old_p;
652 struct module *probed_mod;
653 kprobe_opcode_t *addr;
655 addr = kprobe_addr(p);
661 if (!__kernel_text_address((unsigned long) p->addr) ||
662 in_kprobes_functions((unsigned long) p->addr)) {
669 * Check if are we probing a module.
671 probed_mod = __module_text_address((unsigned long) p->addr);
674 * We must hold a refcount of the probed module while updating
675 * its code to prohibit unexpected unloading.
677 if (unlikely(!try_module_get(probed_mod))) {
682 * If the module freed .init.text, we couldn't insert
685 if (within_module_init((unsigned long)p->addr, probed_mod) &&
686 probed_mod->state != MODULE_STATE_COMING) {
687 module_put(probed_mod);
695 INIT_LIST_HEAD(&p->list);
696 mutex_lock(&kprobe_mutex);
697 old_p = get_kprobe(p->addr);
699 ret = register_aggr_kprobe(old_p, p);
703 mutex_lock(&text_mutex);
704 ret = arch_prepare_kprobe(p);
706 goto out_unlock_text;
708 INIT_HLIST_NODE(&p->hlist);
709 hlist_add_head_rcu(&p->hlist,
710 &kprobe_table[hash_ptr(p->addr, KPROBE_HASH_BITS)]);
716 mutex_unlock(&text_mutex);
718 mutex_unlock(&kprobe_mutex);
721 module_put(probed_mod);
727 * Unregister a kprobe without a scheduler synchronization.
729 static int __kprobes __unregister_kprobe_top(struct kprobe *p)
731 struct kprobe *old_p, *list_p;
733 old_p = get_kprobe(p->addr);
734 if (unlikely(!old_p))
738 list_for_each_entry_rcu(list_p, &old_p->list, list)
740 /* kprobe p is a valid probe */
746 (old_p->pre_handler == aggr_pre_handler &&
747 list_is_singular(&old_p->list))) {
749 * Only probe on the hash list. Disarm only if kprobes are
750 * enabled and not gone - otherwise, the breakpoint would
751 * already have been removed. We save on flushing icache.
753 if (kprobe_enabled && !kprobe_gone(old_p)) {
754 mutex_lock(&text_mutex);
755 arch_disarm_kprobe(p);
756 mutex_unlock(&text_mutex);
758 hlist_del_rcu(&old_p->hlist);
760 if (p->break_handler && !kprobe_gone(p))
761 old_p->break_handler = NULL;
762 if (p->post_handler && !kprobe_gone(p)) {
763 list_for_each_entry_rcu(list_p, &old_p->list, list) {
764 if ((list_p != p) && (list_p->post_handler))
767 old_p->post_handler = NULL;
770 list_del_rcu(&p->list);
775 static void __kprobes __unregister_kprobe_bottom(struct kprobe *p)
777 struct kprobe *old_p;
779 if (list_empty(&p->list))
780 arch_remove_kprobe(p);
781 else if (list_is_singular(&p->list)) {
782 /* "p" is the last child of an aggr_kprobe */
783 old_p = list_entry(p->list.next, struct kprobe, list);
785 arch_remove_kprobe(old_p);
790 int __kprobes register_kprobes(struct kprobe **kps, int num)
796 for (i = 0; i < num; i++) {
797 ret = register_kprobe(kps[i]);
800 unregister_kprobes(kps, i);
807 void __kprobes unregister_kprobe(struct kprobe *p)
809 unregister_kprobes(&p, 1);
812 void __kprobes unregister_kprobes(struct kprobe **kps, int num)
818 mutex_lock(&kprobe_mutex);
819 for (i = 0; i < num; i++)
820 if (__unregister_kprobe_top(kps[i]) < 0)
822 mutex_unlock(&kprobe_mutex);
825 for (i = 0; i < num; i++)
827 __unregister_kprobe_bottom(kps[i]);
830 static struct notifier_block kprobe_exceptions_nb = {
831 .notifier_call = kprobe_exceptions_notify,
832 .priority = 0x7fffffff /* we need to be notified first */
835 unsigned long __weak arch_deref_entry_point(void *entry)
837 return (unsigned long)entry;
840 int __kprobes register_jprobes(struct jprobe **jps, int num)
847 for (i = 0; i < num; i++) {
850 addr = arch_deref_entry_point(jp->entry);
852 if (!kernel_text_address(addr))
855 /* Todo: Verify probepoint is a function entry point */
856 jp->kp.pre_handler = setjmp_pre_handler;
857 jp->kp.break_handler = longjmp_break_handler;
858 ret = register_kprobe(&jp->kp);
862 unregister_jprobes(jps, i);
869 int __kprobes register_jprobe(struct jprobe *jp)
871 return register_jprobes(&jp, 1);
874 void __kprobes unregister_jprobe(struct jprobe *jp)
876 unregister_jprobes(&jp, 1);
879 void __kprobes unregister_jprobes(struct jprobe **jps, int num)
885 mutex_lock(&kprobe_mutex);
886 for (i = 0; i < num; i++)
887 if (__unregister_kprobe_top(&jps[i]->kp) < 0)
888 jps[i]->kp.addr = NULL;
889 mutex_unlock(&kprobe_mutex);
892 for (i = 0; i < num; i++) {
894 __unregister_kprobe_bottom(&jps[i]->kp);
898 #ifdef CONFIG_KRETPROBES
900 * This kprobe pre_handler is registered with every kretprobe. When probe
901 * hits it will set up the return probe.
903 static int __kprobes pre_handler_kretprobe(struct kprobe *p,
904 struct pt_regs *regs)
906 struct kretprobe *rp = container_of(p, struct kretprobe, kp);
907 unsigned long hash, flags = 0;
908 struct kretprobe_instance *ri;
910 /*TODO: consider to only swap the RA after the last pre_handler fired */
911 hash = hash_ptr(current, KPROBE_HASH_BITS);
912 spin_lock_irqsave(&rp->lock, flags);
913 if (!hlist_empty(&rp->free_instances)) {
914 ri = hlist_entry(rp->free_instances.first,
915 struct kretprobe_instance, hlist);
916 hlist_del(&ri->hlist);
917 spin_unlock_irqrestore(&rp->lock, flags);
922 if (rp->entry_handler && rp->entry_handler(ri, regs))
925 arch_prepare_kretprobe(ri, regs);
927 /* XXX(hch): why is there no hlist_move_head? */
928 INIT_HLIST_NODE(&ri->hlist);
929 kretprobe_table_lock(hash, &flags);
930 hlist_add_head(&ri->hlist, &kretprobe_inst_table[hash]);
931 kretprobe_table_unlock(hash, &flags);
934 spin_unlock_irqrestore(&rp->lock, flags);
939 int __kprobes register_kretprobe(struct kretprobe *rp)
942 struct kretprobe_instance *inst;
946 if (kretprobe_blacklist_size) {
947 addr = kprobe_addr(&rp->kp);
951 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
952 if (kretprobe_blacklist[i].addr == addr)
957 rp->kp.pre_handler = pre_handler_kretprobe;
958 rp->kp.post_handler = NULL;
959 rp->kp.fault_handler = NULL;
960 rp->kp.break_handler = NULL;
962 /* Pre-allocate memory for max kretprobe instances */
963 if (rp->maxactive <= 0) {
964 #ifdef CONFIG_PREEMPT
965 rp->maxactive = max(10, 2 * NR_CPUS);
967 rp->maxactive = NR_CPUS;
970 spin_lock_init(&rp->lock);
971 INIT_HLIST_HEAD(&rp->free_instances);
972 for (i = 0; i < rp->maxactive; i++) {
973 inst = kmalloc(sizeof(struct kretprobe_instance) +
974 rp->data_size, GFP_KERNEL);
979 INIT_HLIST_NODE(&inst->hlist);
980 hlist_add_head(&inst->hlist, &rp->free_instances);
984 /* Establish function entry probe point */
985 ret = register_kprobe(&rp->kp);
991 int __kprobes register_kretprobes(struct kretprobe **rps, int num)
997 for (i = 0; i < num; i++) {
998 ret = register_kretprobe(rps[i]);
1001 unregister_kretprobes(rps, i);
1008 void __kprobes unregister_kretprobe(struct kretprobe *rp)
1010 unregister_kretprobes(&rp, 1);
1013 void __kprobes unregister_kretprobes(struct kretprobe **rps, int num)
1019 mutex_lock(&kprobe_mutex);
1020 for (i = 0; i < num; i++)
1021 if (__unregister_kprobe_top(&rps[i]->kp) < 0)
1022 rps[i]->kp.addr = NULL;
1023 mutex_unlock(&kprobe_mutex);
1025 synchronize_sched();
1026 for (i = 0; i < num; i++) {
1027 if (rps[i]->kp.addr) {
1028 __unregister_kprobe_bottom(&rps[i]->kp);
1029 cleanup_rp_inst(rps[i]);
1034 #else /* CONFIG_KRETPROBES */
1035 int __kprobes register_kretprobe(struct kretprobe *rp)
1040 int __kprobes register_kretprobes(struct kretprobe **rps, int num)
1044 void __kprobes unregister_kretprobe(struct kretprobe *rp)
1048 void __kprobes unregister_kretprobes(struct kretprobe **rps, int num)
1052 static int __kprobes pre_handler_kretprobe(struct kprobe *p,
1053 struct pt_regs *regs)
1058 #endif /* CONFIG_KRETPROBES */
1060 /* Set the kprobe gone and remove its instruction buffer. */
1061 static void __kprobes kill_kprobe(struct kprobe *p)
1064 p->flags |= KPROBE_FLAG_GONE;
1065 if (p->pre_handler == aggr_pre_handler) {
1067 * If this is an aggr_kprobe, we have to list all the
1068 * chained probes and mark them GONE.
1070 list_for_each_entry_rcu(kp, &p->list, list)
1071 kp->flags |= KPROBE_FLAG_GONE;
1072 p->post_handler = NULL;
1073 p->break_handler = NULL;
1076 * Here, we can remove insn_slot safely, because no thread calls
1077 * the original probed function (which will be freed soon) any more.
1079 arch_remove_kprobe(p);
1082 /* Module notifier call back, checking kprobes on the module */
1083 static int __kprobes kprobes_module_callback(struct notifier_block *nb,
1084 unsigned long val, void *data)
1086 struct module *mod = data;
1087 struct hlist_head *head;
1088 struct hlist_node *node;
1091 int checkcore = (val == MODULE_STATE_GOING);
1093 if (val != MODULE_STATE_GOING && val != MODULE_STATE_LIVE)
1097 * When MODULE_STATE_GOING was notified, both of module .text and
1098 * .init.text sections would be freed. When MODULE_STATE_LIVE was
1099 * notified, only .init.text section would be freed. We need to
1100 * disable kprobes which have been inserted in the sections.
1102 mutex_lock(&kprobe_mutex);
1103 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
1104 head = &kprobe_table[i];
1105 hlist_for_each_entry_rcu(p, node, head, hlist)
1106 if (within_module_init((unsigned long)p->addr, mod) ||
1108 within_module_core((unsigned long)p->addr, mod))) {
1110 * The vaddr this probe is installed will soon
1111 * be vfreed buy not synced to disk. Hence,
1112 * disarming the breakpoint isn't needed.
1117 mutex_unlock(&kprobe_mutex);
1121 static struct notifier_block kprobe_module_nb = {
1122 .notifier_call = kprobes_module_callback,
1126 static int __init init_kprobes(void)
1129 unsigned long offset = 0, size = 0;
1130 char *modname, namebuf[128];
1131 const char *symbol_name;
1133 struct kprobe_blackpoint *kb;
1135 /* FIXME allocate the probe table, currently defined statically */
1136 /* initialize all list heads */
1137 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
1138 INIT_HLIST_HEAD(&kprobe_table[i]);
1139 INIT_HLIST_HEAD(&kretprobe_inst_table[i]);
1140 spin_lock_init(&(kretprobe_table_locks[i].lock));
1144 * Lookup and populate the kprobe_blacklist.
1146 * Unlike the kretprobe blacklist, we'll need to determine
1147 * the range of addresses that belong to the said functions,
1148 * since a kprobe need not necessarily be at the beginning
1151 for (kb = kprobe_blacklist; kb->name != NULL; kb++) {
1152 kprobe_lookup_name(kb->name, addr);
1156 kb->start_addr = (unsigned long)addr;
1157 symbol_name = kallsyms_lookup(kb->start_addr,
1158 &size, &offset, &modname, namebuf);
1165 if (kretprobe_blacklist_size) {
1166 /* lookup the function address from its name */
1167 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
1168 kprobe_lookup_name(kretprobe_blacklist[i].name,
1169 kretprobe_blacklist[i].addr);
1170 if (!kretprobe_blacklist[i].addr)
1171 printk("kretprobe: lookup failed: %s\n",
1172 kretprobe_blacklist[i].name);
1176 /* By default, kprobes are enabled */
1177 kprobe_enabled = true;
1179 err = arch_init_kprobes();
1181 err = register_die_notifier(&kprobe_exceptions_nb);
1183 err = register_module_notifier(&kprobe_module_nb);
1185 kprobes_initialized = (err == 0);
1192 #ifdef CONFIG_DEBUG_FS
1193 static void __kprobes report_probe(struct seq_file *pi, struct kprobe *p,
1194 const char *sym, int offset,char *modname)
1198 if (p->pre_handler == pre_handler_kretprobe)
1200 else if (p->pre_handler == setjmp_pre_handler)
1205 seq_printf(pi, "%p %s %s+0x%x %s %s\n", p->addr, kprobe_type,
1206 sym, offset, (modname ? modname : " "),
1207 (kprobe_gone(p) ? "[GONE]" : ""));
1209 seq_printf(pi, "%p %s %p %s\n", p->addr, kprobe_type, p->addr,
1210 (kprobe_gone(p) ? "[GONE]" : ""));
1213 static void __kprobes *kprobe_seq_start(struct seq_file *f, loff_t *pos)
1215 return (*pos < KPROBE_TABLE_SIZE) ? pos : NULL;
1218 static void __kprobes *kprobe_seq_next(struct seq_file *f, void *v, loff_t *pos)
1221 if (*pos >= KPROBE_TABLE_SIZE)
1226 static void __kprobes kprobe_seq_stop(struct seq_file *f, void *v)
1231 static int __kprobes show_kprobe_addr(struct seq_file *pi, void *v)
1233 struct hlist_head *head;
1234 struct hlist_node *node;
1235 struct kprobe *p, *kp;
1236 const char *sym = NULL;
1237 unsigned int i = *(loff_t *) v;
1238 unsigned long offset = 0;
1239 char *modname, namebuf[128];
1241 head = &kprobe_table[i];
1243 hlist_for_each_entry_rcu(p, node, head, hlist) {
1244 sym = kallsyms_lookup((unsigned long)p->addr, NULL,
1245 &offset, &modname, namebuf);
1246 if (p->pre_handler == aggr_pre_handler) {
1247 list_for_each_entry_rcu(kp, &p->list, list)
1248 report_probe(pi, kp, sym, offset, modname);
1250 report_probe(pi, p, sym, offset, modname);
1256 static struct seq_operations kprobes_seq_ops = {
1257 .start = kprobe_seq_start,
1258 .next = kprobe_seq_next,
1259 .stop = kprobe_seq_stop,
1260 .show = show_kprobe_addr
1263 static int __kprobes kprobes_open(struct inode *inode, struct file *filp)
1265 return seq_open(filp, &kprobes_seq_ops);
1268 static struct file_operations debugfs_kprobes_operations = {
1269 .open = kprobes_open,
1271 .llseek = seq_lseek,
1272 .release = seq_release,
1275 static void __kprobes enable_all_kprobes(void)
1277 struct hlist_head *head;
1278 struct hlist_node *node;
1282 mutex_lock(&kprobe_mutex);
1284 /* If kprobes are already enabled, just return */
1286 goto already_enabled;
1288 mutex_lock(&text_mutex);
1289 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
1290 head = &kprobe_table[i];
1291 hlist_for_each_entry_rcu(p, node, head, hlist)
1292 if (!kprobe_gone(p))
1295 mutex_unlock(&text_mutex);
1297 kprobe_enabled = true;
1298 printk(KERN_INFO "Kprobes globally enabled\n");
1301 mutex_unlock(&kprobe_mutex);
1305 static void __kprobes disable_all_kprobes(void)
1307 struct hlist_head *head;
1308 struct hlist_node *node;
1312 mutex_lock(&kprobe_mutex);
1314 /* If kprobes are already disabled, just return */
1315 if (!kprobe_enabled)
1316 goto already_disabled;
1318 kprobe_enabled = false;
1319 printk(KERN_INFO "Kprobes globally disabled\n");
1320 mutex_lock(&text_mutex);
1321 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
1322 head = &kprobe_table[i];
1323 hlist_for_each_entry_rcu(p, node, head, hlist) {
1324 if (!arch_trampoline_kprobe(p) && !kprobe_gone(p))
1325 arch_disarm_kprobe(p);
1329 mutex_unlock(&text_mutex);
1330 mutex_unlock(&kprobe_mutex);
1331 /* Allow all currently running kprobes to complete */
1332 synchronize_sched();
1336 mutex_unlock(&kprobe_mutex);
1341 * XXX: The debugfs bool file interface doesn't allow for callbacks
1342 * when the bool state is switched. We can reuse that facility when
1345 static ssize_t read_enabled_file_bool(struct file *file,
1346 char __user *user_buf, size_t count, loff_t *ppos)
1356 return simple_read_from_buffer(user_buf, count, ppos, buf, 2);
1359 static ssize_t write_enabled_file_bool(struct file *file,
1360 const char __user *user_buf, size_t count, loff_t *ppos)
1365 buf_size = min(count, (sizeof(buf)-1));
1366 if (copy_from_user(buf, user_buf, buf_size))
1373 enable_all_kprobes();
1378 disable_all_kprobes();
1385 static struct file_operations fops_kp = {
1386 .read = read_enabled_file_bool,
1387 .write = write_enabled_file_bool,
1390 static int __kprobes debugfs_kprobe_init(void)
1392 struct dentry *dir, *file;
1393 unsigned int value = 1;
1395 dir = debugfs_create_dir("kprobes", NULL);
1399 file = debugfs_create_file("list", 0444, dir, NULL,
1400 &debugfs_kprobes_operations);
1402 debugfs_remove(dir);
1406 file = debugfs_create_file("enabled", 0600, dir,
1409 debugfs_remove(dir);
1416 late_initcall(debugfs_kprobe_init);
1417 #endif /* CONFIG_DEBUG_FS */
1419 module_init(init_kprobes);
1421 EXPORT_SYMBOL_GPL(register_kprobe);
1422 EXPORT_SYMBOL_GPL(unregister_kprobe);
1423 EXPORT_SYMBOL_GPL(register_kprobes);
1424 EXPORT_SYMBOL_GPL(unregister_kprobes);
1425 EXPORT_SYMBOL_GPL(register_jprobe);
1426 EXPORT_SYMBOL_GPL(unregister_jprobe);
1427 EXPORT_SYMBOL_GPL(register_jprobes);
1428 EXPORT_SYMBOL_GPL(unregister_jprobes);
1429 EXPORT_SYMBOL_GPL(jprobe_return);
1430 EXPORT_SYMBOL_GPL(register_kretprobe);
1431 EXPORT_SYMBOL_GPL(unregister_kretprobe);
1432 EXPORT_SYMBOL_GPL(register_kretprobes);
1433 EXPORT_SYMBOL_GPL(unregister_kretprobes);