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 /* Optimization never be done when disarmed */
484 if (kprobes_all_disarmed || !kprobes_allow_optimization ||
485 list_empty(&optimizing_list))
489 * The optimization/unoptimization refers online_cpus via
490 * stop_machine() and cpu-hotplug modifies online_cpus.
491 * And same time, text_mutex will be held in cpu-hotplug and here.
492 * This combination can cause a deadlock (cpu-hotplug try to lock
493 * text_mutex but stop_machine can not be done because online_cpus
495 * To avoid this deadlock, we need to call get_online_cpus()
496 * for preventing cpu-hotplug outside of text_mutex locking.
499 mutex_lock(&text_mutex);
500 arch_optimize_kprobes(&optimizing_list);
501 mutex_unlock(&text_mutex);
506 * Unoptimize (replace a jump with a breakpoint and remove the breakpoint
507 * if need) kprobes listed on unoptimizing_list.
509 static __kprobes void do_unoptimize_kprobes(struct list_head *free_list)
511 struct optimized_kprobe *op, *tmp;
513 /* Unoptimization must be done anytime */
514 if (list_empty(&unoptimizing_list))
517 /* Ditto to do_optimize_kprobes */
519 mutex_lock(&text_mutex);
520 list_for_each_entry_safe(op, tmp, &unoptimizing_list, list) {
521 /* Unoptimize kprobes */
522 arch_unoptimize_kprobe(op);
523 /* Disarm probes if marked disabled */
524 if (kprobe_disabled(&op->kp))
525 arch_disarm_kprobe(&op->kp);
526 if (kprobe_unused(&op->kp)) {
528 * Remove unused probes from hash list. After waiting
529 * for synchronization, these probes are reclaimed.
530 * (reclaiming is done by do_free_cleaned_kprobes.)
532 hlist_del_rcu(&op->kp.hlist);
533 /* Move only unused probes on free_list */
534 list_move(&op->list, free_list);
536 list_del_init(&op->list);
538 mutex_unlock(&text_mutex);
542 /* Reclaim all kprobes on the free_list */
543 static __kprobes void do_free_cleaned_kprobes(struct list_head *free_list)
545 struct optimized_kprobe *op, *tmp;
547 list_for_each_entry_safe(op, tmp, free_list, list) {
548 BUG_ON(!kprobe_unused(&op->kp));
549 list_del_init(&op->list);
550 free_aggr_kprobe(&op->kp);
554 /* Start optimizer after OPTIMIZE_DELAY passed */
555 static __kprobes void kick_kprobe_optimizer(void)
557 if (!delayed_work_pending(&optimizing_work))
558 schedule_delayed_work(&optimizing_work, OPTIMIZE_DELAY);
561 /* Kprobe jump optimizer */
562 static __kprobes void kprobe_optimizer(struct work_struct *work)
564 LIST_HEAD(free_list);
566 /* Lock modules while optimizing kprobes */
567 mutex_lock(&module_mutex);
568 mutex_lock(&kprobe_mutex);
571 * Step 1: Unoptimize kprobes and collect cleaned (unused and disarmed)
572 * kprobes before waiting for quiesence period.
574 do_unoptimize_kprobes(&free_list);
577 * Step 2: Wait for quiesence period to ensure all running interrupts
578 * are done. Because optprobe may modify multiple instructions
579 * there is a chance that Nth instruction is interrupted. In that
580 * case, running interrupt can return to 2nd-Nth byte of jump
581 * instruction. This wait is for avoiding it.
585 /* Step 3: Optimize kprobes after quiesence period */
586 do_optimize_kprobes();
588 /* Step 4: Free cleaned kprobes after quiesence period */
589 do_free_cleaned_kprobes(&free_list);
591 mutex_unlock(&kprobe_mutex);
592 mutex_unlock(&module_mutex);
594 /* Step 5: Kick optimizer again if needed */
595 if (!list_empty(&optimizing_list))
596 kick_kprobe_optimizer();
598 /* Wake up all waiters */
599 complete_all(&optimizer_comp);
602 /* Wait for completing optimization and unoptimization */
603 static __kprobes void wait_for_kprobe_optimizer(void)
605 if (delayed_work_pending(&optimizing_work))
606 wait_for_completion(&optimizer_comp);
609 /* Optimize kprobe if p is ready to be optimized */
610 static __kprobes void optimize_kprobe(struct kprobe *p)
612 struct optimized_kprobe *op;
614 /* Check if the kprobe is disabled or not ready for optimization. */
615 if (!kprobe_optready(p) || !kprobes_allow_optimization ||
616 (kprobe_disabled(p) || kprobes_all_disarmed))
619 /* Both of break_handler and post_handler are not supported. */
620 if (p->break_handler || p->post_handler)
623 op = container_of(p, struct optimized_kprobe, kp);
625 /* Check there is no other kprobes at the optimized instructions */
626 if (arch_check_optimized_kprobe(op) < 0)
629 /* Check if it is already optimized. */
630 if (op->kp.flags & KPROBE_FLAG_OPTIMIZED)
632 op->kp.flags |= KPROBE_FLAG_OPTIMIZED;
634 if (!list_empty(&op->list))
635 /* This is under unoptimizing. Just dequeue the probe */
636 list_del_init(&op->list);
638 list_add(&op->list, &optimizing_list);
639 kick_kprobe_optimizer();
643 /* Short cut to direct unoptimizing */
644 static __kprobes void force_unoptimize_kprobe(struct optimized_kprobe *op)
647 arch_unoptimize_kprobe(op);
649 if (kprobe_disabled(&op->kp))
650 arch_disarm_kprobe(&op->kp);
653 /* Unoptimize a kprobe if p is optimized */
654 static __kprobes void unoptimize_kprobe(struct kprobe *p, bool force)
656 struct optimized_kprobe *op;
658 if (!kprobe_aggrprobe(p) || kprobe_disarmed(p))
659 return; /* This is not an optprobe nor optimized */
661 op = container_of(p, struct optimized_kprobe, kp);
662 if (!kprobe_optimized(p)) {
663 /* Unoptimized or unoptimizing case */
664 if (force && !list_empty(&op->list)) {
666 * Only if this is unoptimizing kprobe and forced,
667 * forcibly unoptimize it. (No need to unoptimize
668 * unoptimized kprobe again :)
670 list_del_init(&op->list);
671 force_unoptimize_kprobe(op);
676 op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
677 if (!list_empty(&op->list)) {
678 /* Dequeue from the optimization queue */
679 list_del_init(&op->list);
682 /* Optimized kprobe case */
684 /* Forcibly update the code: this is a special case */
685 force_unoptimize_kprobe(op);
687 list_add(&op->list, &unoptimizing_list);
688 kick_kprobe_optimizer();
692 /* Cancel unoptimizing for reusing */
693 static void reuse_unused_kprobe(struct kprobe *ap)
695 struct optimized_kprobe *op;
697 BUG_ON(!kprobe_unused(ap));
699 * Unused kprobe MUST be on the way of delayed unoptimizing (means
700 * there is still a relative jump) and disabled.
702 op = container_of(ap, struct optimized_kprobe, kp);
703 if (unlikely(list_empty(&op->list)))
704 printk(KERN_WARNING "Warning: found a stray unused "
705 "aggrprobe@%p\n", ap->addr);
706 /* Enable the probe again */
707 ap->flags &= ~KPROBE_FLAG_DISABLED;
708 /* Optimize it again (remove from op->list) */
709 BUG_ON(!kprobe_optready(ap));
713 /* Remove optimized instructions */
714 static void __kprobes kill_optimized_kprobe(struct kprobe *p)
716 struct optimized_kprobe *op;
718 op = container_of(p, struct optimized_kprobe, kp);
719 if (!list_empty(&op->list))
720 /* Dequeue from the (un)optimization queue */
721 list_del_init(&op->list);
723 op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
724 /* Don't touch the code, because it is already freed. */
725 arch_remove_optimized_kprobe(op);
728 /* Try to prepare optimized instructions */
729 static __kprobes void prepare_optimized_kprobe(struct kprobe *p)
731 struct optimized_kprobe *op;
733 op = container_of(p, struct optimized_kprobe, kp);
734 arch_prepare_optimized_kprobe(op);
737 /* Allocate new optimized_kprobe and try to prepare optimized instructions */
738 static __kprobes struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
740 struct optimized_kprobe *op;
742 op = kzalloc(sizeof(struct optimized_kprobe), GFP_KERNEL);
746 INIT_LIST_HEAD(&op->list);
747 op->kp.addr = p->addr;
748 arch_prepare_optimized_kprobe(op);
753 static void __kprobes init_aggr_kprobe(struct kprobe *ap, struct kprobe *p);
756 * Prepare an optimized_kprobe and optimize it
757 * NOTE: p must be a normal registered kprobe
759 static __kprobes void try_to_optimize_kprobe(struct kprobe *p)
762 struct optimized_kprobe *op;
764 ap = alloc_aggr_kprobe(p);
768 op = container_of(ap, struct optimized_kprobe, kp);
769 if (!arch_prepared_optinsn(&op->optinsn)) {
770 /* If failed to setup optimizing, fallback to kprobe */
771 arch_remove_optimized_kprobe(op);
776 init_aggr_kprobe(ap, p);
781 /* This should be called with kprobe_mutex locked */
782 static void __kprobes optimize_all_kprobes(void)
784 struct hlist_head *head;
785 struct hlist_node *node;
789 /* If optimization is already allowed, just return */
790 if (kprobes_allow_optimization)
793 kprobes_allow_optimization = true;
794 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
795 head = &kprobe_table[i];
796 hlist_for_each_entry_rcu(p, node, head, hlist)
797 if (!kprobe_disabled(p))
800 printk(KERN_INFO "Kprobes globally optimized\n");
803 /* This should be called with kprobe_mutex locked */
804 static void __kprobes unoptimize_all_kprobes(void)
806 struct hlist_head *head;
807 struct hlist_node *node;
811 /* If optimization is already prohibited, just return */
812 if (!kprobes_allow_optimization)
815 kprobes_allow_optimization = false;
816 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
817 head = &kprobe_table[i];
818 hlist_for_each_entry_rcu(p, node, head, hlist) {
819 if (!kprobe_disabled(p))
820 unoptimize_kprobe(p, false);
823 /* Wait for unoptimizing completion */
824 wait_for_kprobe_optimizer();
825 printk(KERN_INFO "Kprobes globally unoptimized\n");
828 int sysctl_kprobes_optimization;
829 int proc_kprobes_optimization_handler(struct ctl_table *table, int write,
830 void __user *buffer, size_t *length,
835 mutex_lock(&kprobe_mutex);
836 sysctl_kprobes_optimization = kprobes_allow_optimization ? 1 : 0;
837 ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
839 if (sysctl_kprobes_optimization)
840 optimize_all_kprobes();
842 unoptimize_all_kprobes();
843 mutex_unlock(&kprobe_mutex);
847 #endif /* CONFIG_SYSCTL */
849 /* Put a breakpoint for a probe. Must be called with text_mutex locked */
850 static void __kprobes __arm_kprobe(struct kprobe *p)
854 /* Check collision with other optimized kprobes */
855 _p = get_optimized_kprobe((unsigned long)p->addr);
857 /* Fallback to unoptimized kprobe */
858 unoptimize_kprobe(_p, true);
861 optimize_kprobe(p); /* Try to optimize (add kprobe to a list) */
864 /* Remove the breakpoint of a probe. Must be called with text_mutex locked */
865 static void __kprobes __disarm_kprobe(struct kprobe *p, bool reopt)
869 unoptimize_kprobe(p, false); /* Try to unoptimize */
871 if (!kprobe_queued(p)) {
872 arch_disarm_kprobe(p);
873 /* If another kprobe was blocked, optimize it. */
874 _p = get_optimized_kprobe((unsigned long)p->addr);
875 if (unlikely(_p) && reopt)
878 /* TODO: reoptimize others after unoptimized this probe */
881 #else /* !CONFIG_OPTPROBES */
883 #define optimize_kprobe(p) do {} while (0)
884 #define unoptimize_kprobe(p, f) do {} while (0)
885 #define kill_optimized_kprobe(p) do {} while (0)
886 #define prepare_optimized_kprobe(p) do {} while (0)
887 #define try_to_optimize_kprobe(p) do {} while (0)
888 #define __arm_kprobe(p) arch_arm_kprobe(p)
889 #define __disarm_kprobe(p, o) arch_disarm_kprobe(p)
890 #define kprobe_disarmed(p) kprobe_disabled(p)
891 #define wait_for_kprobe_optimizer() do {} while (0)
893 /* There should be no unused kprobes can be reused without optimization */
894 static void reuse_unused_kprobe(struct kprobe *ap)
896 printk(KERN_ERR "Error: There should be no unused kprobe here.\n");
897 BUG_ON(kprobe_unused(ap));
900 static __kprobes void free_aggr_kprobe(struct kprobe *p)
902 arch_remove_kprobe(p);
906 static __kprobes struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
908 return kzalloc(sizeof(struct kprobe), GFP_KERNEL);
910 #endif /* CONFIG_OPTPROBES */
912 /* Arm a kprobe with text_mutex */
913 static void __kprobes arm_kprobe(struct kprobe *kp)
916 * Here, since __arm_kprobe() doesn't use stop_machine(),
917 * this doesn't cause deadlock on text_mutex. So, we don't
918 * need get_online_cpus().
920 mutex_lock(&text_mutex);
922 mutex_unlock(&text_mutex);
925 /* Disarm a kprobe with text_mutex */
926 static void __kprobes disarm_kprobe(struct kprobe *kp)
929 mutex_lock(&text_mutex);
930 __disarm_kprobe(kp, true);
931 mutex_unlock(&text_mutex);
935 * Aggregate handlers for multiple kprobes support - these handlers
936 * take care of invoking the individual kprobe handlers on p->list
938 static int __kprobes aggr_pre_handler(struct kprobe *p, struct pt_regs *regs)
942 list_for_each_entry_rcu(kp, &p->list, list) {
943 if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
944 set_kprobe_instance(kp);
945 if (kp->pre_handler(kp, regs))
948 reset_kprobe_instance();
953 static void __kprobes aggr_post_handler(struct kprobe *p, struct pt_regs *regs,
958 list_for_each_entry_rcu(kp, &p->list, list) {
959 if (kp->post_handler && likely(!kprobe_disabled(kp))) {
960 set_kprobe_instance(kp);
961 kp->post_handler(kp, regs, flags);
962 reset_kprobe_instance();
967 static int __kprobes aggr_fault_handler(struct kprobe *p, struct pt_regs *regs,
970 struct kprobe *cur = __get_cpu_var(kprobe_instance);
973 * if we faulted "during" the execution of a user specified
974 * probe handler, invoke just that probe's fault handler
976 if (cur && cur->fault_handler) {
977 if (cur->fault_handler(cur, regs, trapnr))
983 static int __kprobes aggr_break_handler(struct kprobe *p, struct pt_regs *regs)
985 struct kprobe *cur = __get_cpu_var(kprobe_instance);
988 if (cur && cur->break_handler) {
989 if (cur->break_handler(cur, regs))
992 reset_kprobe_instance();
996 /* Walks the list and increments nmissed count for multiprobe case */
997 void __kprobes kprobes_inc_nmissed_count(struct kprobe *p)
1000 if (!kprobe_aggrprobe(p)) {
1003 list_for_each_entry_rcu(kp, &p->list, list)
1009 void __kprobes recycle_rp_inst(struct kretprobe_instance *ri,
1010 struct hlist_head *head)
1012 struct kretprobe *rp = ri->rp;
1014 /* remove rp inst off the rprobe_inst_table */
1015 hlist_del(&ri->hlist);
1016 INIT_HLIST_NODE(&ri->hlist);
1018 spin_lock(&rp->lock);
1019 hlist_add_head(&ri->hlist, &rp->free_instances);
1020 spin_unlock(&rp->lock);
1023 hlist_add_head(&ri->hlist, head);
1026 void __kprobes kretprobe_hash_lock(struct task_struct *tsk,
1027 struct hlist_head **head, unsigned long *flags)
1028 __acquires(hlist_lock)
1030 unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
1031 spinlock_t *hlist_lock;
1033 *head = &kretprobe_inst_table[hash];
1034 hlist_lock = kretprobe_table_lock_ptr(hash);
1035 spin_lock_irqsave(hlist_lock, *flags);
1038 static void __kprobes kretprobe_table_lock(unsigned long hash,
1039 unsigned long *flags)
1040 __acquires(hlist_lock)
1042 spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
1043 spin_lock_irqsave(hlist_lock, *flags);
1046 void __kprobes kretprobe_hash_unlock(struct task_struct *tsk,
1047 unsigned long *flags)
1048 __releases(hlist_lock)
1050 unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
1051 spinlock_t *hlist_lock;
1053 hlist_lock = kretprobe_table_lock_ptr(hash);
1054 spin_unlock_irqrestore(hlist_lock, *flags);
1057 static void __kprobes kretprobe_table_unlock(unsigned long hash,
1058 unsigned long *flags)
1059 __releases(hlist_lock)
1061 spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
1062 spin_unlock_irqrestore(hlist_lock, *flags);
1066 * This function is called from finish_task_switch when task tk becomes dead,
1067 * so that we can recycle any function-return probe instances associated
1068 * with this task. These left over instances represent probed functions
1069 * that have been called but will never return.
1071 void __kprobes kprobe_flush_task(struct task_struct *tk)
1073 struct kretprobe_instance *ri;
1074 struct hlist_head *head, empty_rp;
1075 struct hlist_node *node, *tmp;
1076 unsigned long hash, flags = 0;
1078 if (unlikely(!kprobes_initialized))
1079 /* Early boot. kretprobe_table_locks not yet initialized. */
1082 hash = hash_ptr(tk, KPROBE_HASH_BITS);
1083 head = &kretprobe_inst_table[hash];
1084 kretprobe_table_lock(hash, &flags);
1085 hlist_for_each_entry_safe(ri, node, tmp, head, hlist) {
1087 recycle_rp_inst(ri, &empty_rp);
1089 kretprobe_table_unlock(hash, &flags);
1090 INIT_HLIST_HEAD(&empty_rp);
1091 hlist_for_each_entry_safe(ri, node, tmp, &empty_rp, hlist) {
1092 hlist_del(&ri->hlist);
1097 static inline void free_rp_inst(struct kretprobe *rp)
1099 struct kretprobe_instance *ri;
1100 struct hlist_node *pos, *next;
1102 hlist_for_each_entry_safe(ri, pos, next, &rp->free_instances, hlist) {
1103 hlist_del(&ri->hlist);
1108 static void __kprobes cleanup_rp_inst(struct kretprobe *rp)
1110 unsigned long flags, hash;
1111 struct kretprobe_instance *ri;
1112 struct hlist_node *pos, *next;
1113 struct hlist_head *head;
1116 for (hash = 0; hash < KPROBE_TABLE_SIZE; hash++) {
1117 kretprobe_table_lock(hash, &flags);
1118 head = &kretprobe_inst_table[hash];
1119 hlist_for_each_entry_safe(ri, pos, next, head, hlist) {
1123 kretprobe_table_unlock(hash, &flags);
1129 * Add the new probe to ap->list. Fail if this is the
1130 * second jprobe at the address - two jprobes can't coexist
1132 static int __kprobes add_new_kprobe(struct kprobe *ap, struct kprobe *p)
1134 BUG_ON(kprobe_gone(ap) || kprobe_gone(p));
1136 if (p->break_handler || p->post_handler)
1137 unoptimize_kprobe(ap, true); /* Fall back to normal kprobe */
1139 if (p->break_handler) {
1140 if (ap->break_handler)
1142 list_add_tail_rcu(&p->list, &ap->list);
1143 ap->break_handler = aggr_break_handler;
1145 list_add_rcu(&p->list, &ap->list);
1146 if (p->post_handler && !ap->post_handler)
1147 ap->post_handler = aggr_post_handler;
1149 if (kprobe_disabled(ap) && !kprobe_disabled(p)) {
1150 ap->flags &= ~KPROBE_FLAG_DISABLED;
1151 if (!kprobes_all_disarmed)
1152 /* Arm the breakpoint again. */
1159 * Fill in the required fields of the "manager kprobe". Replace the
1160 * earlier kprobe in the hlist with the manager kprobe
1162 static void __kprobes init_aggr_kprobe(struct kprobe *ap, struct kprobe *p)
1164 /* Copy p's insn slot to ap */
1166 flush_insn_slot(ap);
1168 ap->flags = p->flags & ~KPROBE_FLAG_OPTIMIZED;
1169 ap->pre_handler = aggr_pre_handler;
1170 ap->fault_handler = aggr_fault_handler;
1171 /* We don't care the kprobe which has gone. */
1172 if (p->post_handler && !kprobe_gone(p))
1173 ap->post_handler = aggr_post_handler;
1174 if (p->break_handler && !kprobe_gone(p))
1175 ap->break_handler = aggr_break_handler;
1177 INIT_LIST_HEAD(&ap->list);
1178 INIT_HLIST_NODE(&ap->hlist);
1180 list_add_rcu(&p->list, &ap->list);
1181 hlist_replace_rcu(&p->hlist, &ap->hlist);
1185 * This is the second or subsequent kprobe at the address - handle
1188 static int __kprobes register_aggr_kprobe(struct kprobe *orig_p,
1192 struct kprobe *ap = orig_p;
1194 if (!kprobe_aggrprobe(orig_p)) {
1195 /* If orig_p is not an aggr_kprobe, create new aggr_kprobe. */
1196 ap = alloc_aggr_kprobe(orig_p);
1199 init_aggr_kprobe(ap, orig_p);
1200 } else if (kprobe_unused(ap))
1201 /* This probe is going to die. Rescue it */
1202 reuse_unused_kprobe(ap);
1204 if (kprobe_gone(ap)) {
1206 * Attempting to insert new probe at the same location that
1207 * had a probe in the module vaddr area which already
1208 * freed. So, the instruction slot has already been
1209 * released. We need a new slot for the new probe.
1211 ret = arch_prepare_kprobe(ap);
1214 * Even if fail to allocate new slot, don't need to
1215 * free aggr_probe. It will be used next time, or
1216 * freed by unregister_kprobe.
1220 /* Prepare optimized instructions if possible. */
1221 prepare_optimized_kprobe(ap);
1224 * Clear gone flag to prevent allocating new slot again, and
1225 * set disabled flag because it is not armed yet.
1227 ap->flags = (ap->flags & ~KPROBE_FLAG_GONE)
1228 | KPROBE_FLAG_DISABLED;
1231 /* Copy ap's insn slot to p */
1233 return add_new_kprobe(ap, p);
1236 static int __kprobes in_kprobes_functions(unsigned long addr)
1238 struct kprobe_blackpoint *kb;
1240 if (addr >= (unsigned long)__kprobes_text_start &&
1241 addr < (unsigned long)__kprobes_text_end)
1244 * If there exists a kprobe_blacklist, verify and
1245 * fail any probe registration in the prohibited area
1247 for (kb = kprobe_blacklist; kb->name != NULL; kb++) {
1248 if (kb->start_addr) {
1249 if (addr >= kb->start_addr &&
1250 addr < (kb->start_addr + kb->range))
1258 * If we have a symbol_name argument, look it up and add the offset field
1259 * to it. This way, we can specify a relative address to a symbol.
1261 static kprobe_opcode_t __kprobes *kprobe_addr(struct kprobe *p)
1263 kprobe_opcode_t *addr = p->addr;
1264 if (p->symbol_name) {
1267 kprobe_lookup_name(p->symbol_name, addr);
1272 return (kprobe_opcode_t *)(((char *)addr) + p->offset);
1275 /* Check passed kprobe is valid and return kprobe in kprobe_table. */
1276 static struct kprobe * __kprobes __get_valid_kprobe(struct kprobe *p)
1278 struct kprobe *ap, *list_p;
1280 ap = get_kprobe(p->addr);
1285 list_for_each_entry_rcu(list_p, &ap->list, list)
1287 /* kprobe p is a valid probe */
1295 /* Return error if the kprobe is being re-registered */
1296 static inline int check_kprobe_rereg(struct kprobe *p)
1300 mutex_lock(&kprobe_mutex);
1301 if (__get_valid_kprobe(p))
1303 mutex_unlock(&kprobe_mutex);
1308 int __kprobes register_kprobe(struct kprobe *p)
1311 struct kprobe *old_p;
1312 struct module *probed_mod;
1313 kprobe_opcode_t *addr;
1315 addr = kprobe_addr(p);
1320 ret = check_kprobe_rereg(p);
1326 if (!kernel_text_address((unsigned long) p->addr) ||
1327 in_kprobes_functions((unsigned long) p->addr) ||
1328 ftrace_text_reserved(p->addr, p->addr) ||
1329 jump_label_text_reserved(p->addr, p->addr))
1330 goto fail_with_jump_label;
1332 /* User can pass only KPROBE_FLAG_DISABLED to register_kprobe */
1333 p->flags &= KPROBE_FLAG_DISABLED;
1336 * Check if are we probing a module.
1338 probed_mod = __module_text_address((unsigned long) p->addr);
1341 * We must hold a refcount of the probed module while updating
1342 * its code to prohibit unexpected unloading.
1344 if (unlikely(!try_module_get(probed_mod)))
1345 goto fail_with_jump_label;
1348 * If the module freed .init.text, we couldn't insert
1351 if (within_module_init((unsigned long)p->addr, probed_mod) &&
1352 probed_mod->state != MODULE_STATE_COMING) {
1353 module_put(probed_mod);
1354 goto fail_with_jump_label;
1358 jump_label_unlock();
1361 INIT_LIST_HEAD(&p->list);
1362 mutex_lock(&kprobe_mutex);
1364 jump_label_lock(); /* needed to call jump_label_text_reserved() */
1366 get_online_cpus(); /* For avoiding text_mutex deadlock. */
1367 mutex_lock(&text_mutex);
1369 old_p = get_kprobe(p->addr);
1371 /* Since this may unoptimize old_p, locking text_mutex. */
1372 ret = register_aggr_kprobe(old_p, p);
1376 ret = arch_prepare_kprobe(p);
1380 INIT_HLIST_NODE(&p->hlist);
1381 hlist_add_head_rcu(&p->hlist,
1382 &kprobe_table[hash_ptr(p->addr, KPROBE_HASH_BITS)]);
1384 if (!kprobes_all_disarmed && !kprobe_disabled(p))
1387 /* Try to optimize kprobe */
1388 try_to_optimize_kprobe(p);
1391 mutex_unlock(&text_mutex);
1393 jump_label_unlock();
1394 mutex_unlock(&kprobe_mutex);
1397 module_put(probed_mod);
1401 fail_with_jump_label:
1403 jump_label_unlock();
1406 EXPORT_SYMBOL_GPL(register_kprobe);
1408 /* Check if all probes on the aggrprobe are disabled */
1409 static int __kprobes aggr_kprobe_disabled(struct kprobe *ap)
1413 list_for_each_entry_rcu(kp, &ap->list, list)
1414 if (!kprobe_disabled(kp))
1416 * There is an active probe on the list.
1417 * We can't disable this ap.
1424 /* Disable one kprobe: Make sure called under kprobe_mutex is locked */
1425 static struct kprobe *__kprobes __disable_kprobe(struct kprobe *p)
1427 struct kprobe *orig_p;
1429 /* Get an original kprobe for return */
1430 orig_p = __get_valid_kprobe(p);
1431 if (unlikely(orig_p == NULL))
1434 if (!kprobe_disabled(p)) {
1435 /* Disable probe if it is a child probe */
1437 p->flags |= KPROBE_FLAG_DISABLED;
1439 /* Try to disarm and disable this/parent probe */
1440 if (p == orig_p || aggr_kprobe_disabled(orig_p)) {
1441 disarm_kprobe(orig_p);
1442 orig_p->flags |= KPROBE_FLAG_DISABLED;
1450 * Unregister a kprobe without a scheduler synchronization.
1452 static int __kprobes __unregister_kprobe_top(struct kprobe *p)
1454 struct kprobe *ap, *list_p;
1456 /* Disable kprobe. This will disarm it if needed. */
1457 ap = __disable_kprobe(p);
1463 * This probe is an independent(and non-optimized) kprobe
1464 * (not an aggrprobe). Remove from the hash list.
1468 /* Following process expects this probe is an aggrprobe */
1469 WARN_ON(!kprobe_aggrprobe(ap));
1471 if (list_is_singular(&ap->list) && kprobe_disarmed(ap))
1473 * !disarmed could be happen if the probe is under delayed
1478 /* If disabling probe has special handlers, update aggrprobe */
1479 if (p->break_handler && !kprobe_gone(p))
1480 ap->break_handler = NULL;
1481 if (p->post_handler && !kprobe_gone(p)) {
1482 list_for_each_entry_rcu(list_p, &ap->list, list) {
1483 if ((list_p != p) && (list_p->post_handler))
1486 ap->post_handler = NULL;
1490 * Remove from the aggrprobe: this path will do nothing in
1491 * __unregister_kprobe_bottom().
1493 list_del_rcu(&p->list);
1494 if (!kprobe_disabled(ap) && !kprobes_all_disarmed)
1496 * Try to optimize this probe again, because post
1497 * handler may have been changed.
1499 optimize_kprobe(ap);
1504 BUG_ON(!kprobe_disarmed(ap));
1505 hlist_del_rcu(&ap->hlist);
1509 static void __kprobes __unregister_kprobe_bottom(struct kprobe *p)
1513 if (list_empty(&p->list))
1514 /* This is an independent kprobe */
1515 arch_remove_kprobe(p);
1516 else if (list_is_singular(&p->list)) {
1517 /* This is the last child of an aggrprobe */
1518 ap = list_entry(p->list.next, struct kprobe, list);
1520 free_aggr_kprobe(ap);
1522 /* Otherwise, do nothing. */
1525 int __kprobes register_kprobes(struct kprobe **kps, int num)
1531 for (i = 0; i < num; i++) {
1532 ret = register_kprobe(kps[i]);
1535 unregister_kprobes(kps, i);
1541 EXPORT_SYMBOL_GPL(register_kprobes);
1543 void __kprobes unregister_kprobe(struct kprobe *p)
1545 unregister_kprobes(&p, 1);
1547 EXPORT_SYMBOL_GPL(unregister_kprobe);
1549 void __kprobes unregister_kprobes(struct kprobe **kps, int num)
1555 mutex_lock(&kprobe_mutex);
1556 for (i = 0; i < num; i++)
1557 if (__unregister_kprobe_top(kps[i]) < 0)
1558 kps[i]->addr = NULL;
1559 mutex_unlock(&kprobe_mutex);
1561 synchronize_sched();
1562 for (i = 0; i < num; i++)
1564 __unregister_kprobe_bottom(kps[i]);
1566 EXPORT_SYMBOL_GPL(unregister_kprobes);
1568 static struct notifier_block kprobe_exceptions_nb = {
1569 .notifier_call = kprobe_exceptions_notify,
1570 .priority = 0x7fffffff /* we need to be notified first */
1573 unsigned long __weak arch_deref_entry_point(void *entry)
1575 return (unsigned long)entry;
1578 int __kprobes register_jprobes(struct jprobe **jps, int num)
1585 for (i = 0; i < num; i++) {
1586 unsigned long addr, offset;
1588 addr = arch_deref_entry_point(jp->entry);
1590 /* Verify probepoint is a function entry point */
1591 if (kallsyms_lookup_size_offset(addr, NULL, &offset) &&
1593 jp->kp.pre_handler = setjmp_pre_handler;
1594 jp->kp.break_handler = longjmp_break_handler;
1595 ret = register_kprobe(&jp->kp);
1601 unregister_jprobes(jps, i);
1607 EXPORT_SYMBOL_GPL(register_jprobes);
1609 int __kprobes register_jprobe(struct jprobe *jp)
1611 return register_jprobes(&jp, 1);
1613 EXPORT_SYMBOL_GPL(register_jprobe);
1615 void __kprobes unregister_jprobe(struct jprobe *jp)
1617 unregister_jprobes(&jp, 1);
1619 EXPORT_SYMBOL_GPL(unregister_jprobe);
1621 void __kprobes unregister_jprobes(struct jprobe **jps, int num)
1627 mutex_lock(&kprobe_mutex);
1628 for (i = 0; i < num; i++)
1629 if (__unregister_kprobe_top(&jps[i]->kp) < 0)
1630 jps[i]->kp.addr = NULL;
1631 mutex_unlock(&kprobe_mutex);
1633 synchronize_sched();
1634 for (i = 0; i < num; i++) {
1635 if (jps[i]->kp.addr)
1636 __unregister_kprobe_bottom(&jps[i]->kp);
1639 EXPORT_SYMBOL_GPL(unregister_jprobes);
1641 #ifdef CONFIG_KRETPROBES
1643 * This kprobe pre_handler is registered with every kretprobe. When probe
1644 * hits it will set up the return probe.
1646 static int __kprobes pre_handler_kretprobe(struct kprobe *p,
1647 struct pt_regs *regs)
1649 struct kretprobe *rp = container_of(p, struct kretprobe, kp);
1650 unsigned long hash, flags = 0;
1651 struct kretprobe_instance *ri;
1653 /*TODO: consider to only swap the RA after the last pre_handler fired */
1654 hash = hash_ptr(current, KPROBE_HASH_BITS);
1655 spin_lock_irqsave(&rp->lock, flags);
1656 if (!hlist_empty(&rp->free_instances)) {
1657 ri = hlist_entry(rp->free_instances.first,
1658 struct kretprobe_instance, hlist);
1659 hlist_del(&ri->hlist);
1660 spin_unlock_irqrestore(&rp->lock, flags);
1665 if (rp->entry_handler && rp->entry_handler(ri, regs))
1668 arch_prepare_kretprobe(ri, regs);
1670 /* XXX(hch): why is there no hlist_move_head? */
1671 INIT_HLIST_NODE(&ri->hlist);
1672 kretprobe_table_lock(hash, &flags);
1673 hlist_add_head(&ri->hlist, &kretprobe_inst_table[hash]);
1674 kretprobe_table_unlock(hash, &flags);
1677 spin_unlock_irqrestore(&rp->lock, flags);
1682 int __kprobes register_kretprobe(struct kretprobe *rp)
1685 struct kretprobe_instance *inst;
1689 if (kretprobe_blacklist_size) {
1690 addr = kprobe_addr(&rp->kp);
1694 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
1695 if (kretprobe_blacklist[i].addr == addr)
1700 rp->kp.pre_handler = pre_handler_kretprobe;
1701 rp->kp.post_handler = NULL;
1702 rp->kp.fault_handler = NULL;
1703 rp->kp.break_handler = NULL;
1705 /* Pre-allocate memory for max kretprobe instances */
1706 if (rp->maxactive <= 0) {
1707 #ifdef CONFIG_PREEMPT
1708 rp->maxactive = max_t(unsigned int, 10, 2*num_possible_cpus());
1710 rp->maxactive = num_possible_cpus();
1713 spin_lock_init(&rp->lock);
1714 INIT_HLIST_HEAD(&rp->free_instances);
1715 for (i = 0; i < rp->maxactive; i++) {
1716 inst = kmalloc(sizeof(struct kretprobe_instance) +
1717 rp->data_size, GFP_KERNEL);
1722 INIT_HLIST_NODE(&inst->hlist);
1723 hlist_add_head(&inst->hlist, &rp->free_instances);
1727 /* Establish function entry probe point */
1728 ret = register_kprobe(&rp->kp);
1733 EXPORT_SYMBOL_GPL(register_kretprobe);
1735 int __kprobes register_kretprobes(struct kretprobe **rps, int num)
1741 for (i = 0; i < num; i++) {
1742 ret = register_kretprobe(rps[i]);
1745 unregister_kretprobes(rps, i);
1751 EXPORT_SYMBOL_GPL(register_kretprobes);
1753 void __kprobes unregister_kretprobe(struct kretprobe *rp)
1755 unregister_kretprobes(&rp, 1);
1757 EXPORT_SYMBOL_GPL(unregister_kretprobe);
1759 void __kprobes unregister_kretprobes(struct kretprobe **rps, int num)
1765 mutex_lock(&kprobe_mutex);
1766 for (i = 0; i < num; i++)
1767 if (__unregister_kprobe_top(&rps[i]->kp) < 0)
1768 rps[i]->kp.addr = NULL;
1769 mutex_unlock(&kprobe_mutex);
1771 synchronize_sched();
1772 for (i = 0; i < num; i++) {
1773 if (rps[i]->kp.addr) {
1774 __unregister_kprobe_bottom(&rps[i]->kp);
1775 cleanup_rp_inst(rps[i]);
1779 EXPORT_SYMBOL_GPL(unregister_kretprobes);
1781 #else /* CONFIG_KRETPROBES */
1782 int __kprobes register_kretprobe(struct kretprobe *rp)
1786 EXPORT_SYMBOL_GPL(register_kretprobe);
1788 int __kprobes register_kretprobes(struct kretprobe **rps, int num)
1792 EXPORT_SYMBOL_GPL(register_kretprobes);
1794 void __kprobes unregister_kretprobe(struct kretprobe *rp)
1797 EXPORT_SYMBOL_GPL(unregister_kretprobe);
1799 void __kprobes unregister_kretprobes(struct kretprobe **rps, int num)
1802 EXPORT_SYMBOL_GPL(unregister_kretprobes);
1804 static int __kprobes pre_handler_kretprobe(struct kprobe *p,
1805 struct pt_regs *regs)
1810 #endif /* CONFIG_KRETPROBES */
1812 /* Set the kprobe gone and remove its instruction buffer. */
1813 static void __kprobes kill_kprobe(struct kprobe *p)
1817 p->flags |= KPROBE_FLAG_GONE;
1818 if (kprobe_aggrprobe(p)) {
1820 * If this is an aggr_kprobe, we have to list all the
1821 * chained probes and mark them GONE.
1823 list_for_each_entry_rcu(kp, &p->list, list)
1824 kp->flags |= KPROBE_FLAG_GONE;
1825 p->post_handler = NULL;
1826 p->break_handler = NULL;
1827 kill_optimized_kprobe(p);
1830 * Here, we can remove insn_slot safely, because no thread calls
1831 * the original probed function (which will be freed soon) any more.
1833 arch_remove_kprobe(p);
1836 /* Disable one kprobe */
1837 int __kprobes disable_kprobe(struct kprobe *kp)
1841 mutex_lock(&kprobe_mutex);
1843 /* Disable this kprobe */
1844 if (__disable_kprobe(kp) == NULL)
1847 mutex_unlock(&kprobe_mutex);
1850 EXPORT_SYMBOL_GPL(disable_kprobe);
1852 /* Enable one kprobe */
1853 int __kprobes enable_kprobe(struct kprobe *kp)
1858 mutex_lock(&kprobe_mutex);
1860 /* Check whether specified probe is valid. */
1861 p = __get_valid_kprobe(kp);
1862 if (unlikely(p == NULL)) {
1867 if (kprobe_gone(kp)) {
1868 /* This kprobe has gone, we couldn't enable it. */
1874 kp->flags &= ~KPROBE_FLAG_DISABLED;
1876 if (!kprobes_all_disarmed && kprobe_disabled(p)) {
1877 p->flags &= ~KPROBE_FLAG_DISABLED;
1881 mutex_unlock(&kprobe_mutex);
1884 EXPORT_SYMBOL_GPL(enable_kprobe);
1886 void __kprobes dump_kprobe(struct kprobe *kp)
1888 printk(KERN_WARNING "Dumping kprobe:\n");
1889 printk(KERN_WARNING "Name: %s\nAddress: %p\nOffset: %x\n",
1890 kp->symbol_name, kp->addr, kp->offset);
1893 /* Module notifier call back, checking kprobes on the module */
1894 static int __kprobes kprobes_module_callback(struct notifier_block *nb,
1895 unsigned long val, void *data)
1897 struct module *mod = data;
1898 struct hlist_head *head;
1899 struct hlist_node *node;
1902 int checkcore = (val == MODULE_STATE_GOING);
1904 if (val != MODULE_STATE_GOING && val != MODULE_STATE_LIVE)
1908 * When MODULE_STATE_GOING was notified, both of module .text and
1909 * .init.text sections would be freed. When MODULE_STATE_LIVE was
1910 * notified, only .init.text section would be freed. We need to
1911 * disable kprobes which have been inserted in the sections.
1913 mutex_lock(&kprobe_mutex);
1914 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
1915 head = &kprobe_table[i];
1916 hlist_for_each_entry_rcu(p, node, head, hlist)
1917 if (within_module_init((unsigned long)p->addr, mod) ||
1919 within_module_core((unsigned long)p->addr, mod))) {
1921 * The vaddr this probe is installed will soon
1922 * be vfreed buy not synced to disk. Hence,
1923 * disarming the breakpoint isn't needed.
1928 mutex_unlock(&kprobe_mutex);
1932 static struct notifier_block kprobe_module_nb = {
1933 .notifier_call = kprobes_module_callback,
1937 static int __init init_kprobes(void)
1940 unsigned long offset = 0, size = 0;
1941 char *modname, namebuf[128];
1942 const char *symbol_name;
1944 struct kprobe_blackpoint *kb;
1946 /* FIXME allocate the probe table, currently defined statically */
1947 /* initialize all list heads */
1948 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
1949 INIT_HLIST_HEAD(&kprobe_table[i]);
1950 INIT_HLIST_HEAD(&kretprobe_inst_table[i]);
1951 spin_lock_init(&(kretprobe_table_locks[i].lock));
1955 * Lookup and populate the kprobe_blacklist.
1957 * Unlike the kretprobe blacklist, we'll need to determine
1958 * the range of addresses that belong to the said functions,
1959 * since a kprobe need not necessarily be at the beginning
1962 for (kb = kprobe_blacklist; kb->name != NULL; kb++) {
1963 kprobe_lookup_name(kb->name, addr);
1967 kb->start_addr = (unsigned long)addr;
1968 symbol_name = kallsyms_lookup(kb->start_addr,
1969 &size, &offset, &modname, namebuf);
1976 if (kretprobe_blacklist_size) {
1977 /* lookup the function address from its name */
1978 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
1979 kprobe_lookup_name(kretprobe_blacklist[i].name,
1980 kretprobe_blacklist[i].addr);
1981 if (!kretprobe_blacklist[i].addr)
1982 printk("kretprobe: lookup failed: %s\n",
1983 kretprobe_blacklist[i].name);
1987 #if defined(CONFIG_OPTPROBES)
1988 #if defined(__ARCH_WANT_KPROBES_INSN_SLOT)
1989 /* Init kprobe_optinsn_slots */
1990 kprobe_optinsn_slots.insn_size = MAX_OPTINSN_SIZE;
1992 /* By default, kprobes can be optimized */
1993 kprobes_allow_optimization = true;
1996 /* By default, kprobes are armed */
1997 kprobes_all_disarmed = false;
1999 err = arch_init_kprobes();
2001 err = register_die_notifier(&kprobe_exceptions_nb);
2003 err = register_module_notifier(&kprobe_module_nb);
2005 kprobes_initialized = (err == 0);
2012 #ifdef CONFIG_DEBUG_FS
2013 static void __kprobes report_probe(struct seq_file *pi, struct kprobe *p,
2014 const char *sym, int offset, char *modname, struct kprobe *pp)
2018 if (p->pre_handler == pre_handler_kretprobe)
2020 else if (p->pre_handler == setjmp_pre_handler)
2026 seq_printf(pi, "%p %s %s+0x%x %s ",
2027 p->addr, kprobe_type, sym, offset,
2028 (modname ? modname : " "));
2030 seq_printf(pi, "%p %s %p ",
2031 p->addr, kprobe_type, p->addr);
2035 seq_printf(pi, "%s%s%s\n",
2036 (kprobe_gone(p) ? "[GONE]" : ""),
2037 ((kprobe_disabled(p) && !kprobe_gone(p)) ? "[DISABLED]" : ""),
2038 (kprobe_optimized(pp) ? "[OPTIMIZED]" : ""));
2041 static void __kprobes *kprobe_seq_start(struct seq_file *f, loff_t *pos)
2043 return (*pos < KPROBE_TABLE_SIZE) ? pos : NULL;
2046 static void __kprobes *kprobe_seq_next(struct seq_file *f, void *v, loff_t *pos)
2049 if (*pos >= KPROBE_TABLE_SIZE)
2054 static void __kprobes kprobe_seq_stop(struct seq_file *f, void *v)
2059 static int __kprobes show_kprobe_addr(struct seq_file *pi, void *v)
2061 struct hlist_head *head;
2062 struct hlist_node *node;
2063 struct kprobe *p, *kp;
2064 const char *sym = NULL;
2065 unsigned int i = *(loff_t *) v;
2066 unsigned long offset = 0;
2067 char *modname, namebuf[128];
2069 head = &kprobe_table[i];
2071 hlist_for_each_entry_rcu(p, node, head, hlist) {
2072 sym = kallsyms_lookup((unsigned long)p->addr, NULL,
2073 &offset, &modname, namebuf);
2074 if (kprobe_aggrprobe(p)) {
2075 list_for_each_entry_rcu(kp, &p->list, list)
2076 report_probe(pi, kp, sym, offset, modname, p);
2078 report_probe(pi, p, sym, offset, modname, NULL);
2084 static const struct seq_operations kprobes_seq_ops = {
2085 .start = kprobe_seq_start,
2086 .next = kprobe_seq_next,
2087 .stop = kprobe_seq_stop,
2088 .show = show_kprobe_addr
2091 static int __kprobes kprobes_open(struct inode *inode, struct file *filp)
2093 return seq_open(filp, &kprobes_seq_ops);
2096 static const struct file_operations debugfs_kprobes_operations = {
2097 .open = kprobes_open,
2099 .llseek = seq_lseek,
2100 .release = seq_release,
2103 static void __kprobes arm_all_kprobes(void)
2105 struct hlist_head *head;
2106 struct hlist_node *node;
2110 mutex_lock(&kprobe_mutex);
2112 /* If kprobes are armed, just return */
2113 if (!kprobes_all_disarmed)
2114 goto already_enabled;
2116 /* Arming kprobes doesn't optimize kprobe itself */
2117 mutex_lock(&text_mutex);
2118 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2119 head = &kprobe_table[i];
2120 hlist_for_each_entry_rcu(p, node, head, hlist)
2121 if (!kprobe_disabled(p))
2124 mutex_unlock(&text_mutex);
2126 kprobes_all_disarmed = false;
2127 printk(KERN_INFO "Kprobes globally enabled\n");
2130 mutex_unlock(&kprobe_mutex);
2134 static void __kprobes disarm_all_kprobes(void)
2136 struct hlist_head *head;
2137 struct hlist_node *node;
2141 mutex_lock(&kprobe_mutex);
2143 /* If kprobes are already disarmed, just return */
2144 if (kprobes_all_disarmed) {
2145 mutex_unlock(&kprobe_mutex);
2149 kprobes_all_disarmed = true;
2150 printk(KERN_INFO "Kprobes globally disabled\n");
2152 mutex_lock(&text_mutex);
2153 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2154 head = &kprobe_table[i];
2155 hlist_for_each_entry_rcu(p, node, head, hlist) {
2156 if (!arch_trampoline_kprobe(p) && !kprobe_disabled(p))
2157 __disarm_kprobe(p, false);
2160 mutex_unlock(&text_mutex);
2161 mutex_unlock(&kprobe_mutex);
2163 /* Wait for disarming all kprobes by optimizer */
2164 wait_for_kprobe_optimizer();
2168 * XXX: The debugfs bool file interface doesn't allow for callbacks
2169 * when the bool state is switched. We can reuse that facility when
2172 static ssize_t read_enabled_file_bool(struct file *file,
2173 char __user *user_buf, size_t count, loff_t *ppos)
2177 if (!kprobes_all_disarmed)
2183 return simple_read_from_buffer(user_buf, count, ppos, buf, 2);
2186 static ssize_t write_enabled_file_bool(struct file *file,
2187 const char __user *user_buf, size_t count, loff_t *ppos)
2192 buf_size = min(count, (sizeof(buf)-1));
2193 if (copy_from_user(buf, user_buf, buf_size))
2205 disarm_all_kprobes();
2212 static const struct file_operations fops_kp = {
2213 .read = read_enabled_file_bool,
2214 .write = write_enabled_file_bool,
2215 .llseek = default_llseek,
2218 static int __kprobes debugfs_kprobe_init(void)
2220 struct dentry *dir, *file;
2221 unsigned int value = 1;
2223 dir = debugfs_create_dir("kprobes", NULL);
2227 file = debugfs_create_file("list", 0444, dir, NULL,
2228 &debugfs_kprobes_operations);
2230 debugfs_remove(dir);
2234 file = debugfs_create_file("enabled", 0600, dir,
2237 debugfs_remove(dir);
2244 late_initcall(debugfs_kprobe_init);
2245 #endif /* CONFIG_DEBUG_FS */
2247 module_init(init_kprobes);
2249 /* defined in arch/.../kernel/kprobes.c */
2250 EXPORT_SYMBOL_GPL(jprobe_return);