2 * Kernel Probes (KProbes)
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write to the Free Software
17 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
19 * Copyright (C) IBM Corporation, 2002, 2004
21 * 2002-Oct Created by Vamsi Krishna S <vamsi_krishna@in.ibm.com> Kernel
22 * Probes initial implementation (includes suggestions from
24 * 2004-Aug Updated by Prasanna S Panchamukhi <prasanna@in.ibm.com> with
25 * hlists and exceptions notifier as suggested by Andi Kleen.
26 * 2004-July Suparna Bhattacharya <suparna@in.ibm.com> added jumper probes
27 * interface to access function arguments.
28 * 2004-Sep Prasanna S Panchamukhi <prasanna@in.ibm.com> Changed Kprobes
29 * exceptions notifier to be first on the priority list.
30 * 2005-May Hien Nguyen <hien@us.ibm.com>, Jim Keniston
31 * <jkenisto@us.ibm.com> and Prasanna S Panchamukhi
32 * <prasanna@in.ibm.com> added function-return probes.
34 #include <linux/kprobes.h>
35 #include <linux/hash.h>
36 #include <linux/init.h>
37 #include <linux/slab.h>
38 #include <linux/stddef.h>
39 #include <linux/export.h>
40 #include <linux/moduleloader.h>
41 #include <linux/kallsyms.h>
42 #include <linux/freezer.h>
43 #include <linux/seq_file.h>
44 #include <linux/debugfs.h>
45 #include <linux/sysctl.h>
46 #include <linux/kdebug.h>
47 #include <linux/memory.h>
48 #include <linux/ftrace.h>
49 #include <linux/cpu.h>
50 #include <linux/jump_label.h>
52 #include <asm-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 raw_spinlock_t lock ____cacheline_aligned_in_smp;
82 } kretprobe_table_locks[KPROBE_TABLE_SIZE];
84 static raw_spinlock_t *kretprobe_table_lock_ptr(unsigned long hash)
86 return &(kretprobe_table_locks[hash].lock);
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 __this_cpu_write(kprobe_instance, kp);
323 static inline void reset_kprobe_instance(void)
325 __this_cpu_write(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 arch_unoptimize_kprobes(&unoptimizing_list, free_list);
521 /* Loop free_list for disarming */
522 list_for_each_entry_safe(op, tmp, free_list, list) {
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);
534 list_del_init(&op->list);
536 mutex_unlock(&text_mutex);
540 /* Reclaim all kprobes on the free_list */
541 static __kprobes void do_free_cleaned_kprobes(struct list_head *free_list)
543 struct optimized_kprobe *op, *tmp;
545 list_for_each_entry_safe(op, tmp, free_list, list) {
546 BUG_ON(!kprobe_unused(&op->kp));
547 list_del_init(&op->list);
548 free_aggr_kprobe(&op->kp);
552 /* Start optimizer after OPTIMIZE_DELAY passed */
553 static __kprobes void kick_kprobe_optimizer(void)
555 if (!delayed_work_pending(&optimizing_work))
556 schedule_delayed_work(&optimizing_work, OPTIMIZE_DELAY);
559 /* Kprobe jump optimizer */
560 static __kprobes void kprobe_optimizer(struct work_struct *work)
562 LIST_HEAD(free_list);
564 /* Lock modules while optimizing kprobes */
565 mutex_lock(&module_mutex);
566 mutex_lock(&kprobe_mutex);
569 * Step 1: Unoptimize kprobes and collect cleaned (unused and disarmed)
570 * kprobes before waiting for quiesence period.
572 do_unoptimize_kprobes(&free_list);
575 * Step 2: Wait for quiesence period to ensure all running interrupts
576 * are done. Because optprobe may modify multiple instructions
577 * there is a chance that Nth instruction is interrupted. In that
578 * case, running interrupt can return to 2nd-Nth byte of jump
579 * instruction. This wait is for avoiding it.
583 /* Step 3: Optimize kprobes after quiesence period */
584 do_optimize_kprobes();
586 /* Step 4: Free cleaned kprobes after quiesence period */
587 do_free_cleaned_kprobes(&free_list);
589 mutex_unlock(&kprobe_mutex);
590 mutex_unlock(&module_mutex);
592 /* Step 5: Kick optimizer again if needed */
593 if (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list))
594 kick_kprobe_optimizer();
596 /* Wake up all waiters */
597 complete_all(&optimizer_comp);
600 /* Wait for completing optimization and unoptimization */
601 static __kprobes void wait_for_kprobe_optimizer(void)
603 if (delayed_work_pending(&optimizing_work))
604 wait_for_completion(&optimizer_comp);
607 /* Optimize kprobe if p is ready to be optimized */
608 static __kprobes void optimize_kprobe(struct kprobe *p)
610 struct optimized_kprobe *op;
612 /* Check if the kprobe is disabled or not ready for optimization. */
613 if (!kprobe_optready(p) || !kprobes_allow_optimization ||
614 (kprobe_disabled(p) || kprobes_all_disarmed))
617 /* Both of break_handler and post_handler are not supported. */
618 if (p->break_handler || p->post_handler)
621 op = container_of(p, struct optimized_kprobe, kp);
623 /* Check there is no other kprobes at the optimized instructions */
624 if (arch_check_optimized_kprobe(op) < 0)
627 /* Check if it is already optimized. */
628 if (op->kp.flags & KPROBE_FLAG_OPTIMIZED)
630 op->kp.flags |= KPROBE_FLAG_OPTIMIZED;
632 if (!list_empty(&op->list))
633 /* This is under unoptimizing. Just dequeue the probe */
634 list_del_init(&op->list);
636 list_add(&op->list, &optimizing_list);
637 kick_kprobe_optimizer();
641 /* Short cut to direct unoptimizing */
642 static __kprobes void force_unoptimize_kprobe(struct optimized_kprobe *op)
645 arch_unoptimize_kprobe(op);
647 if (kprobe_disabled(&op->kp))
648 arch_disarm_kprobe(&op->kp);
651 /* Unoptimize a kprobe if p is optimized */
652 static __kprobes void unoptimize_kprobe(struct kprobe *p, bool force)
654 struct optimized_kprobe *op;
656 if (!kprobe_aggrprobe(p) || kprobe_disarmed(p))
657 return; /* This is not an optprobe nor optimized */
659 op = container_of(p, struct optimized_kprobe, kp);
660 if (!kprobe_optimized(p)) {
661 /* Unoptimized or unoptimizing case */
662 if (force && !list_empty(&op->list)) {
664 * Only if this is unoptimizing kprobe and forced,
665 * forcibly unoptimize it. (No need to unoptimize
666 * unoptimized kprobe again :)
668 list_del_init(&op->list);
669 force_unoptimize_kprobe(op);
674 op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
675 if (!list_empty(&op->list)) {
676 /* Dequeue from the optimization queue */
677 list_del_init(&op->list);
680 /* Optimized kprobe case */
682 /* Forcibly update the code: this is a special case */
683 force_unoptimize_kprobe(op);
685 list_add(&op->list, &unoptimizing_list);
686 kick_kprobe_optimizer();
690 /* Cancel unoptimizing for reusing */
691 static void reuse_unused_kprobe(struct kprobe *ap)
693 struct optimized_kprobe *op;
695 BUG_ON(!kprobe_unused(ap));
697 * Unused kprobe MUST be on the way of delayed unoptimizing (means
698 * there is still a relative jump) and disabled.
700 op = container_of(ap, struct optimized_kprobe, kp);
701 if (unlikely(list_empty(&op->list)))
702 printk(KERN_WARNING "Warning: found a stray unused "
703 "aggrprobe@%p\n", ap->addr);
704 /* Enable the probe again */
705 ap->flags &= ~KPROBE_FLAG_DISABLED;
706 /* Optimize it again (remove from op->list) */
707 BUG_ON(!kprobe_optready(ap));
711 /* Remove optimized instructions */
712 static void __kprobes kill_optimized_kprobe(struct kprobe *p)
714 struct optimized_kprobe *op;
716 op = container_of(p, struct optimized_kprobe, kp);
717 if (!list_empty(&op->list))
718 /* Dequeue from the (un)optimization queue */
719 list_del_init(&op->list);
721 op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
722 /* Don't touch the code, because it is already freed. */
723 arch_remove_optimized_kprobe(op);
726 /* Try to prepare optimized instructions */
727 static __kprobes void prepare_optimized_kprobe(struct kprobe *p)
729 struct optimized_kprobe *op;
731 op = container_of(p, struct optimized_kprobe, kp);
732 arch_prepare_optimized_kprobe(op);
735 /* Allocate new optimized_kprobe and try to prepare optimized instructions */
736 static __kprobes struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
738 struct optimized_kprobe *op;
740 op = kzalloc(sizeof(struct optimized_kprobe), GFP_KERNEL);
744 INIT_LIST_HEAD(&op->list);
745 op->kp.addr = p->addr;
746 arch_prepare_optimized_kprobe(op);
751 static void __kprobes init_aggr_kprobe(struct kprobe *ap, struct kprobe *p);
754 * Prepare an optimized_kprobe and optimize it
755 * NOTE: p must be a normal registered kprobe
757 static __kprobes void try_to_optimize_kprobe(struct kprobe *p)
760 struct optimized_kprobe *op;
762 ap = alloc_aggr_kprobe(p);
766 op = container_of(ap, struct optimized_kprobe, kp);
767 if (!arch_prepared_optinsn(&op->optinsn)) {
768 /* If failed to setup optimizing, fallback to kprobe */
769 arch_remove_optimized_kprobe(op);
774 init_aggr_kprobe(ap, p);
779 /* This should be called with kprobe_mutex locked */
780 static void __kprobes optimize_all_kprobes(void)
782 struct hlist_head *head;
783 struct hlist_node *node;
787 /* If optimization is already allowed, just return */
788 if (kprobes_allow_optimization)
791 kprobes_allow_optimization = true;
792 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
793 head = &kprobe_table[i];
794 hlist_for_each_entry_rcu(p, node, head, hlist)
795 if (!kprobe_disabled(p))
798 printk(KERN_INFO "Kprobes globally optimized\n");
801 /* This should be called with kprobe_mutex locked */
802 static void __kprobes unoptimize_all_kprobes(void)
804 struct hlist_head *head;
805 struct hlist_node *node;
809 /* If optimization is already prohibited, just return */
810 if (!kprobes_allow_optimization)
813 kprobes_allow_optimization = false;
814 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
815 head = &kprobe_table[i];
816 hlist_for_each_entry_rcu(p, node, head, hlist) {
817 if (!kprobe_disabled(p))
818 unoptimize_kprobe(p, false);
821 /* Wait for unoptimizing completion */
822 wait_for_kprobe_optimizer();
823 printk(KERN_INFO "Kprobes globally unoptimized\n");
826 int sysctl_kprobes_optimization;
827 int proc_kprobes_optimization_handler(struct ctl_table *table, int write,
828 void __user *buffer, size_t *length,
833 mutex_lock(&kprobe_mutex);
834 sysctl_kprobes_optimization = kprobes_allow_optimization ? 1 : 0;
835 ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
837 if (sysctl_kprobes_optimization)
838 optimize_all_kprobes();
840 unoptimize_all_kprobes();
841 mutex_unlock(&kprobe_mutex);
845 #endif /* CONFIG_SYSCTL */
847 /* Put a breakpoint for a probe. Must be called with text_mutex locked */
848 static void __kprobes __arm_kprobe(struct kprobe *p)
852 /* Check collision with other optimized kprobes */
853 _p = get_optimized_kprobe((unsigned long)p->addr);
855 /* Fallback to unoptimized kprobe */
856 unoptimize_kprobe(_p, true);
859 optimize_kprobe(p); /* Try to optimize (add kprobe to a list) */
862 /* Remove the breakpoint of a probe. Must be called with text_mutex locked */
863 static void __kprobes __disarm_kprobe(struct kprobe *p, bool reopt)
867 unoptimize_kprobe(p, false); /* Try to unoptimize */
869 if (!kprobe_queued(p)) {
870 arch_disarm_kprobe(p);
871 /* If another kprobe was blocked, optimize it. */
872 _p = get_optimized_kprobe((unsigned long)p->addr);
873 if (unlikely(_p) && reopt)
876 /* TODO: reoptimize others after unoptimized this probe */
879 #else /* !CONFIG_OPTPROBES */
881 #define optimize_kprobe(p) do {} while (0)
882 #define unoptimize_kprobe(p, f) do {} while (0)
883 #define kill_optimized_kprobe(p) do {} while (0)
884 #define prepare_optimized_kprobe(p) do {} while (0)
885 #define try_to_optimize_kprobe(p) do {} while (0)
886 #define __arm_kprobe(p) arch_arm_kprobe(p)
887 #define __disarm_kprobe(p, o) arch_disarm_kprobe(p)
888 #define kprobe_disarmed(p) kprobe_disabled(p)
889 #define wait_for_kprobe_optimizer() do {} while (0)
891 /* There should be no unused kprobes can be reused without optimization */
892 static void reuse_unused_kprobe(struct kprobe *ap)
894 printk(KERN_ERR "Error: There should be no unused kprobe here.\n");
895 BUG_ON(kprobe_unused(ap));
898 static __kprobes void free_aggr_kprobe(struct kprobe *p)
900 arch_remove_kprobe(p);
904 static __kprobes struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
906 return kzalloc(sizeof(struct kprobe), GFP_KERNEL);
908 #endif /* CONFIG_OPTPROBES */
910 /* Arm a kprobe with text_mutex */
911 static void __kprobes arm_kprobe(struct kprobe *kp)
914 * Here, since __arm_kprobe() doesn't use stop_machine(),
915 * this doesn't cause deadlock on text_mutex. So, we don't
916 * need get_online_cpus().
918 mutex_lock(&text_mutex);
920 mutex_unlock(&text_mutex);
923 /* Disarm a kprobe with text_mutex */
924 static void __kprobes disarm_kprobe(struct kprobe *kp)
927 mutex_lock(&text_mutex);
928 __disarm_kprobe(kp, true);
929 mutex_unlock(&text_mutex);
933 * Aggregate handlers for multiple kprobes support - these handlers
934 * take care of invoking the individual kprobe handlers on p->list
936 static int __kprobes aggr_pre_handler(struct kprobe *p, struct pt_regs *regs)
940 list_for_each_entry_rcu(kp, &p->list, list) {
941 if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
942 set_kprobe_instance(kp);
943 if (kp->pre_handler(kp, regs))
946 reset_kprobe_instance();
951 static void __kprobes aggr_post_handler(struct kprobe *p, struct pt_regs *regs,
956 list_for_each_entry_rcu(kp, &p->list, list) {
957 if (kp->post_handler && likely(!kprobe_disabled(kp))) {
958 set_kprobe_instance(kp);
959 kp->post_handler(kp, regs, flags);
960 reset_kprobe_instance();
965 static int __kprobes aggr_fault_handler(struct kprobe *p, struct pt_regs *regs,
968 struct kprobe *cur = __this_cpu_read(kprobe_instance);
971 * if we faulted "during" the execution of a user specified
972 * probe handler, invoke just that probe's fault handler
974 if (cur && cur->fault_handler) {
975 if (cur->fault_handler(cur, regs, trapnr))
981 static int __kprobes aggr_break_handler(struct kprobe *p, struct pt_regs *regs)
983 struct kprobe *cur = __this_cpu_read(kprobe_instance);
986 if (cur && cur->break_handler) {
987 if (cur->break_handler(cur, regs))
990 reset_kprobe_instance();
994 /* Walks the list and increments nmissed count for multiprobe case */
995 void __kprobes kprobes_inc_nmissed_count(struct kprobe *p)
998 if (!kprobe_aggrprobe(p)) {
1001 list_for_each_entry_rcu(kp, &p->list, list)
1007 void __kprobes recycle_rp_inst(struct kretprobe_instance *ri,
1008 struct hlist_head *head)
1010 struct kretprobe *rp = ri->rp;
1012 /* remove rp inst off the rprobe_inst_table */
1013 hlist_del(&ri->hlist);
1014 INIT_HLIST_NODE(&ri->hlist);
1016 raw_spin_lock(&rp->lock);
1017 hlist_add_head(&ri->hlist, &rp->free_instances);
1018 raw_spin_unlock(&rp->lock);
1021 hlist_add_head(&ri->hlist, head);
1024 void __kprobes kretprobe_hash_lock(struct task_struct *tsk,
1025 struct hlist_head **head, unsigned long *flags)
1026 __acquires(hlist_lock)
1028 unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
1029 raw_spinlock_t *hlist_lock;
1031 *head = &kretprobe_inst_table[hash];
1032 hlist_lock = kretprobe_table_lock_ptr(hash);
1033 raw_spin_lock_irqsave(hlist_lock, *flags);
1036 static void __kprobes kretprobe_table_lock(unsigned long hash,
1037 unsigned long *flags)
1038 __acquires(hlist_lock)
1040 raw_spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
1041 raw_spin_lock_irqsave(hlist_lock, *flags);
1044 void __kprobes kretprobe_hash_unlock(struct task_struct *tsk,
1045 unsigned long *flags)
1046 __releases(hlist_lock)
1048 unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
1049 raw_spinlock_t *hlist_lock;
1051 hlist_lock = kretprobe_table_lock_ptr(hash);
1052 raw_spin_unlock_irqrestore(hlist_lock, *flags);
1055 static void __kprobes kretprobe_table_unlock(unsigned long hash,
1056 unsigned long *flags)
1057 __releases(hlist_lock)
1059 raw_spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
1060 raw_spin_unlock_irqrestore(hlist_lock, *flags);
1064 * This function is called from finish_task_switch when task tk becomes dead,
1065 * so that we can recycle any function-return probe instances associated
1066 * with this task. These left over instances represent probed functions
1067 * that have been called but will never return.
1069 void __kprobes kprobe_flush_task(struct task_struct *tk)
1071 struct kretprobe_instance *ri;
1072 struct hlist_head *head, empty_rp;
1073 struct hlist_node *node, *tmp;
1074 unsigned long hash, flags = 0;
1076 if (unlikely(!kprobes_initialized))
1077 /* Early boot. kretprobe_table_locks not yet initialized. */
1080 INIT_HLIST_HEAD(&empty_rp);
1081 hash = hash_ptr(tk, KPROBE_HASH_BITS);
1082 head = &kretprobe_inst_table[hash];
1083 kretprobe_table_lock(hash, &flags);
1084 hlist_for_each_entry_safe(ri, node, tmp, head, hlist) {
1086 recycle_rp_inst(ri, &empty_rp);
1088 kretprobe_table_unlock(hash, &flags);
1089 hlist_for_each_entry_safe(ri, node, tmp, &empty_rp, hlist) {
1090 hlist_del(&ri->hlist);
1095 static inline void free_rp_inst(struct kretprobe *rp)
1097 struct kretprobe_instance *ri;
1098 struct hlist_node *pos, *next;
1100 hlist_for_each_entry_safe(ri, pos, next, &rp->free_instances, hlist) {
1101 hlist_del(&ri->hlist);
1106 static void __kprobes cleanup_rp_inst(struct kretprobe *rp)
1108 unsigned long flags, hash;
1109 struct kretprobe_instance *ri;
1110 struct hlist_node *pos, *next;
1111 struct hlist_head *head;
1114 for (hash = 0; hash < KPROBE_TABLE_SIZE; hash++) {
1115 kretprobe_table_lock(hash, &flags);
1116 head = &kretprobe_inst_table[hash];
1117 hlist_for_each_entry_safe(ri, pos, next, head, hlist) {
1121 kretprobe_table_unlock(hash, &flags);
1127 * Add the new probe to ap->list. Fail if this is the
1128 * second jprobe at the address - two jprobes can't coexist
1130 static int __kprobes add_new_kprobe(struct kprobe *ap, struct kprobe *p)
1132 BUG_ON(kprobe_gone(ap) || kprobe_gone(p));
1134 if (p->break_handler || p->post_handler)
1135 unoptimize_kprobe(ap, true); /* Fall back to normal kprobe */
1137 if (p->break_handler) {
1138 if (ap->break_handler)
1140 list_add_tail_rcu(&p->list, &ap->list);
1141 ap->break_handler = aggr_break_handler;
1143 list_add_rcu(&p->list, &ap->list);
1144 if (p->post_handler && !ap->post_handler)
1145 ap->post_handler = aggr_post_handler;
1147 if (kprobe_disabled(ap) && !kprobe_disabled(p)) {
1148 ap->flags &= ~KPROBE_FLAG_DISABLED;
1149 if (!kprobes_all_disarmed)
1150 /* Arm the breakpoint again. */
1157 * Fill in the required fields of the "manager kprobe". Replace the
1158 * earlier kprobe in the hlist with the manager kprobe
1160 static void __kprobes init_aggr_kprobe(struct kprobe *ap, struct kprobe *p)
1162 /* Copy p's insn slot to ap */
1164 flush_insn_slot(ap);
1166 ap->flags = p->flags & ~KPROBE_FLAG_OPTIMIZED;
1167 ap->pre_handler = aggr_pre_handler;
1168 ap->fault_handler = aggr_fault_handler;
1169 /* We don't care the kprobe which has gone. */
1170 if (p->post_handler && !kprobe_gone(p))
1171 ap->post_handler = aggr_post_handler;
1172 if (p->break_handler && !kprobe_gone(p))
1173 ap->break_handler = aggr_break_handler;
1175 INIT_LIST_HEAD(&ap->list);
1176 INIT_HLIST_NODE(&ap->hlist);
1178 list_add_rcu(&p->list, &ap->list);
1179 hlist_replace_rcu(&p->hlist, &ap->hlist);
1183 * This is the second or subsequent kprobe at the address - handle
1186 static int __kprobes register_aggr_kprobe(struct kprobe *orig_p,
1190 struct kprobe *ap = orig_p;
1192 if (!kprobe_aggrprobe(orig_p)) {
1193 /* If orig_p is not an aggr_kprobe, create new aggr_kprobe. */
1194 ap = alloc_aggr_kprobe(orig_p);
1197 init_aggr_kprobe(ap, orig_p);
1198 } else if (kprobe_unused(ap))
1199 /* This probe is going to die. Rescue it */
1200 reuse_unused_kprobe(ap);
1202 if (kprobe_gone(ap)) {
1204 * Attempting to insert new probe at the same location that
1205 * had a probe in the module vaddr area which already
1206 * freed. So, the instruction slot has already been
1207 * released. We need a new slot for the new probe.
1209 ret = arch_prepare_kprobe(ap);
1212 * Even if fail to allocate new slot, don't need to
1213 * free aggr_probe. It will be used next time, or
1214 * freed by unregister_kprobe.
1218 /* Prepare optimized instructions if possible. */
1219 prepare_optimized_kprobe(ap);
1222 * Clear gone flag to prevent allocating new slot again, and
1223 * set disabled flag because it is not armed yet.
1225 ap->flags = (ap->flags & ~KPROBE_FLAG_GONE)
1226 | KPROBE_FLAG_DISABLED;
1229 /* Copy ap's insn slot to p */
1231 return add_new_kprobe(ap, p);
1234 static int __kprobes in_kprobes_functions(unsigned long addr)
1236 struct kprobe_blackpoint *kb;
1238 if (addr >= (unsigned long)__kprobes_text_start &&
1239 addr < (unsigned long)__kprobes_text_end)
1242 * If there exists a kprobe_blacklist, verify and
1243 * fail any probe registration in the prohibited area
1245 for (kb = kprobe_blacklist; kb->name != NULL; kb++) {
1246 if (kb->start_addr) {
1247 if (addr >= kb->start_addr &&
1248 addr < (kb->start_addr + kb->range))
1256 * If we have a symbol_name argument, look it up and add the offset field
1257 * to it. This way, we can specify a relative address to a symbol.
1258 * This returns encoded errors if it fails to look up symbol or invalid
1259 * combination of parameters.
1261 static kprobe_opcode_t __kprobes *kprobe_addr(struct kprobe *p)
1263 kprobe_opcode_t *addr = p->addr;
1265 if ((p->symbol_name && p->addr) ||
1266 (!p->symbol_name && !p->addr))
1269 if (p->symbol_name) {
1270 kprobe_lookup_name(p->symbol_name, addr);
1272 return ERR_PTR(-ENOENT);
1275 addr = (kprobe_opcode_t *)(((char *)addr) + p->offset);
1280 return ERR_PTR(-EINVAL);
1283 /* Check passed kprobe is valid and return kprobe in kprobe_table. */
1284 static struct kprobe * __kprobes __get_valid_kprobe(struct kprobe *p)
1286 struct kprobe *ap, *list_p;
1288 ap = get_kprobe(p->addr);
1293 list_for_each_entry_rcu(list_p, &ap->list, list)
1295 /* kprobe p is a valid probe */
1303 /* Return error if the kprobe is being re-registered */
1304 static inline int check_kprobe_rereg(struct kprobe *p)
1308 mutex_lock(&kprobe_mutex);
1309 if (__get_valid_kprobe(p))
1311 mutex_unlock(&kprobe_mutex);
1316 int __kprobes register_kprobe(struct kprobe *p)
1319 struct kprobe *old_p;
1320 struct module *probed_mod;
1321 kprobe_opcode_t *addr;
1323 addr = kprobe_addr(p);
1325 return PTR_ERR(addr);
1328 ret = check_kprobe_rereg(p);
1334 if (!kernel_text_address((unsigned long) p->addr) ||
1335 in_kprobes_functions((unsigned long) p->addr) ||
1336 ftrace_text_reserved(p->addr, p->addr) ||
1337 jump_label_text_reserved(p->addr, p->addr)) {
1342 /* User can pass only KPROBE_FLAG_DISABLED to register_kprobe */
1343 p->flags &= KPROBE_FLAG_DISABLED;
1346 * Check if are we probing a module.
1348 probed_mod = __module_text_address((unsigned long) p->addr);
1350 /* Return -ENOENT if fail. */
1353 * We must hold a refcount of the probed module while updating
1354 * its code to prohibit unexpected unloading.
1356 if (unlikely(!try_module_get(probed_mod)))
1360 * If the module freed .init.text, we couldn't insert
1363 if (within_module_init((unsigned long)p->addr, probed_mod) &&
1364 probed_mod->state != MODULE_STATE_COMING) {
1365 module_put(probed_mod);
1368 /* ret will be updated by following code */
1371 jump_label_unlock();
1374 INIT_LIST_HEAD(&p->list);
1375 mutex_lock(&kprobe_mutex);
1377 jump_label_lock(); /* needed to call jump_label_text_reserved() */
1379 get_online_cpus(); /* For avoiding text_mutex deadlock. */
1380 mutex_lock(&text_mutex);
1382 old_p = get_kprobe(p->addr);
1384 /* Since this may unoptimize old_p, locking text_mutex. */
1385 ret = register_aggr_kprobe(old_p, p);
1389 ret = arch_prepare_kprobe(p);
1393 INIT_HLIST_NODE(&p->hlist);
1394 hlist_add_head_rcu(&p->hlist,
1395 &kprobe_table[hash_ptr(p->addr, KPROBE_HASH_BITS)]);
1397 if (!kprobes_all_disarmed && !kprobe_disabled(p))
1400 /* Try to optimize kprobe */
1401 try_to_optimize_kprobe(p);
1404 mutex_unlock(&text_mutex);
1406 jump_label_unlock();
1407 mutex_unlock(&kprobe_mutex);
1410 module_put(probed_mod);
1416 jump_label_unlock();
1419 EXPORT_SYMBOL_GPL(register_kprobe);
1421 /* Check if all probes on the aggrprobe are disabled */
1422 static int __kprobes aggr_kprobe_disabled(struct kprobe *ap)
1426 list_for_each_entry_rcu(kp, &ap->list, list)
1427 if (!kprobe_disabled(kp))
1429 * There is an active probe on the list.
1430 * We can't disable this ap.
1437 /* Disable one kprobe: Make sure called under kprobe_mutex is locked */
1438 static struct kprobe *__kprobes __disable_kprobe(struct kprobe *p)
1440 struct kprobe *orig_p;
1442 /* Get an original kprobe for return */
1443 orig_p = __get_valid_kprobe(p);
1444 if (unlikely(orig_p == NULL))
1447 if (!kprobe_disabled(p)) {
1448 /* Disable probe if it is a child probe */
1450 p->flags |= KPROBE_FLAG_DISABLED;
1452 /* Try to disarm and disable this/parent probe */
1453 if (p == orig_p || aggr_kprobe_disabled(orig_p)) {
1454 disarm_kprobe(orig_p);
1455 orig_p->flags |= KPROBE_FLAG_DISABLED;
1463 * Unregister a kprobe without a scheduler synchronization.
1465 static int __kprobes __unregister_kprobe_top(struct kprobe *p)
1467 struct kprobe *ap, *list_p;
1469 /* Disable kprobe. This will disarm it if needed. */
1470 ap = __disable_kprobe(p);
1476 * This probe is an independent(and non-optimized) kprobe
1477 * (not an aggrprobe). Remove from the hash list.
1481 /* Following process expects this probe is an aggrprobe */
1482 WARN_ON(!kprobe_aggrprobe(ap));
1484 if (list_is_singular(&ap->list) && kprobe_disarmed(ap))
1486 * !disarmed could be happen if the probe is under delayed
1491 /* If disabling probe has special handlers, update aggrprobe */
1492 if (p->break_handler && !kprobe_gone(p))
1493 ap->break_handler = NULL;
1494 if (p->post_handler && !kprobe_gone(p)) {
1495 list_for_each_entry_rcu(list_p, &ap->list, list) {
1496 if ((list_p != p) && (list_p->post_handler))
1499 ap->post_handler = NULL;
1503 * Remove from the aggrprobe: this path will do nothing in
1504 * __unregister_kprobe_bottom().
1506 list_del_rcu(&p->list);
1507 if (!kprobe_disabled(ap) && !kprobes_all_disarmed)
1509 * Try to optimize this probe again, because post
1510 * handler may have been changed.
1512 optimize_kprobe(ap);
1517 BUG_ON(!kprobe_disarmed(ap));
1518 hlist_del_rcu(&ap->hlist);
1522 static void __kprobes __unregister_kprobe_bottom(struct kprobe *p)
1526 if (list_empty(&p->list))
1527 /* This is an independent kprobe */
1528 arch_remove_kprobe(p);
1529 else if (list_is_singular(&p->list)) {
1530 /* This is the last child of an aggrprobe */
1531 ap = list_entry(p->list.next, struct kprobe, list);
1533 free_aggr_kprobe(ap);
1535 /* Otherwise, do nothing. */
1538 int __kprobes register_kprobes(struct kprobe **kps, int num)
1544 for (i = 0; i < num; i++) {
1545 ret = register_kprobe(kps[i]);
1548 unregister_kprobes(kps, i);
1554 EXPORT_SYMBOL_GPL(register_kprobes);
1556 void __kprobes unregister_kprobe(struct kprobe *p)
1558 unregister_kprobes(&p, 1);
1560 EXPORT_SYMBOL_GPL(unregister_kprobe);
1562 void __kprobes unregister_kprobes(struct kprobe **kps, int num)
1568 mutex_lock(&kprobe_mutex);
1569 for (i = 0; i < num; i++)
1570 if (__unregister_kprobe_top(kps[i]) < 0)
1571 kps[i]->addr = NULL;
1572 mutex_unlock(&kprobe_mutex);
1574 synchronize_sched();
1575 for (i = 0; i < num; i++)
1577 __unregister_kprobe_bottom(kps[i]);
1579 EXPORT_SYMBOL_GPL(unregister_kprobes);
1581 static struct notifier_block kprobe_exceptions_nb = {
1582 .notifier_call = kprobe_exceptions_notify,
1583 .priority = 0x7fffffff /* we need to be notified first */
1586 unsigned long __weak arch_deref_entry_point(void *entry)
1588 return (unsigned long)entry;
1591 int __kprobes register_jprobes(struct jprobe **jps, int num)
1598 for (i = 0; i < num; i++) {
1599 unsigned long addr, offset;
1601 addr = arch_deref_entry_point(jp->entry);
1603 /* Verify probepoint is a function entry point */
1604 if (kallsyms_lookup_size_offset(addr, NULL, &offset) &&
1606 jp->kp.pre_handler = setjmp_pre_handler;
1607 jp->kp.break_handler = longjmp_break_handler;
1608 ret = register_kprobe(&jp->kp);
1614 unregister_jprobes(jps, i);
1620 EXPORT_SYMBOL_GPL(register_jprobes);
1622 int __kprobes register_jprobe(struct jprobe *jp)
1624 return register_jprobes(&jp, 1);
1626 EXPORT_SYMBOL_GPL(register_jprobe);
1628 void __kprobes unregister_jprobe(struct jprobe *jp)
1630 unregister_jprobes(&jp, 1);
1632 EXPORT_SYMBOL_GPL(unregister_jprobe);
1634 void __kprobes unregister_jprobes(struct jprobe **jps, int num)
1640 mutex_lock(&kprobe_mutex);
1641 for (i = 0; i < num; i++)
1642 if (__unregister_kprobe_top(&jps[i]->kp) < 0)
1643 jps[i]->kp.addr = NULL;
1644 mutex_unlock(&kprobe_mutex);
1646 synchronize_sched();
1647 for (i = 0; i < num; i++) {
1648 if (jps[i]->kp.addr)
1649 __unregister_kprobe_bottom(&jps[i]->kp);
1652 EXPORT_SYMBOL_GPL(unregister_jprobes);
1654 #ifdef CONFIG_KRETPROBES
1656 * This kprobe pre_handler is registered with every kretprobe. When probe
1657 * hits it will set up the return probe.
1659 static int __kprobes pre_handler_kretprobe(struct kprobe *p,
1660 struct pt_regs *regs)
1662 struct kretprobe *rp = container_of(p, struct kretprobe, kp);
1663 unsigned long hash, flags = 0;
1664 struct kretprobe_instance *ri;
1666 /*TODO: consider to only swap the RA after the last pre_handler fired */
1667 hash = hash_ptr(current, KPROBE_HASH_BITS);
1668 raw_spin_lock_irqsave(&rp->lock, flags);
1669 if (!hlist_empty(&rp->free_instances)) {
1670 ri = hlist_entry(rp->free_instances.first,
1671 struct kretprobe_instance, hlist);
1672 hlist_del(&ri->hlist);
1673 raw_spin_unlock_irqrestore(&rp->lock, flags);
1678 if (rp->entry_handler && rp->entry_handler(ri, regs)) {
1679 raw_spin_lock_irqsave(&rp->lock, flags);
1680 hlist_add_head(&ri->hlist, &rp->free_instances);
1681 raw_spin_unlock_irqrestore(&rp->lock, flags);
1685 arch_prepare_kretprobe(ri, regs);
1687 /* XXX(hch): why is there no hlist_move_head? */
1688 INIT_HLIST_NODE(&ri->hlist);
1689 kretprobe_table_lock(hash, &flags);
1690 hlist_add_head(&ri->hlist, &kretprobe_inst_table[hash]);
1691 kretprobe_table_unlock(hash, &flags);
1694 raw_spin_unlock_irqrestore(&rp->lock, flags);
1699 int __kprobes register_kretprobe(struct kretprobe *rp)
1702 struct kretprobe_instance *inst;
1706 if (kretprobe_blacklist_size) {
1707 addr = kprobe_addr(&rp->kp);
1709 return PTR_ERR(addr);
1711 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
1712 if (kretprobe_blacklist[i].addr == addr)
1717 rp->kp.pre_handler = pre_handler_kretprobe;
1718 rp->kp.post_handler = NULL;
1719 rp->kp.fault_handler = NULL;
1720 rp->kp.break_handler = NULL;
1722 /* Pre-allocate memory for max kretprobe instances */
1723 if (rp->maxactive <= 0) {
1724 #ifdef CONFIG_PREEMPT
1725 rp->maxactive = max_t(unsigned int, 10, 2*num_possible_cpus());
1727 rp->maxactive = num_possible_cpus();
1730 raw_spin_lock_init(&rp->lock);
1731 INIT_HLIST_HEAD(&rp->free_instances);
1732 for (i = 0; i < rp->maxactive; i++) {
1733 inst = kmalloc(sizeof(struct kretprobe_instance) +
1734 rp->data_size, GFP_KERNEL);
1739 INIT_HLIST_NODE(&inst->hlist);
1740 hlist_add_head(&inst->hlist, &rp->free_instances);
1744 /* Establish function entry probe point */
1745 ret = register_kprobe(&rp->kp);
1750 EXPORT_SYMBOL_GPL(register_kretprobe);
1752 int __kprobes register_kretprobes(struct kretprobe **rps, int num)
1758 for (i = 0; i < num; i++) {
1759 ret = register_kretprobe(rps[i]);
1762 unregister_kretprobes(rps, i);
1768 EXPORT_SYMBOL_GPL(register_kretprobes);
1770 void __kprobes unregister_kretprobe(struct kretprobe *rp)
1772 unregister_kretprobes(&rp, 1);
1774 EXPORT_SYMBOL_GPL(unregister_kretprobe);
1776 void __kprobes unregister_kretprobes(struct kretprobe **rps, int num)
1782 mutex_lock(&kprobe_mutex);
1783 for (i = 0; i < num; i++)
1784 if (__unregister_kprobe_top(&rps[i]->kp) < 0)
1785 rps[i]->kp.addr = NULL;
1786 mutex_unlock(&kprobe_mutex);
1788 synchronize_sched();
1789 for (i = 0; i < num; i++) {
1790 if (rps[i]->kp.addr) {
1791 __unregister_kprobe_bottom(&rps[i]->kp);
1792 cleanup_rp_inst(rps[i]);
1796 EXPORT_SYMBOL_GPL(unregister_kretprobes);
1798 #else /* CONFIG_KRETPROBES */
1799 int __kprobes register_kretprobe(struct kretprobe *rp)
1803 EXPORT_SYMBOL_GPL(register_kretprobe);
1805 int __kprobes register_kretprobes(struct kretprobe **rps, int num)
1809 EXPORT_SYMBOL_GPL(register_kretprobes);
1811 void __kprobes unregister_kretprobe(struct kretprobe *rp)
1814 EXPORT_SYMBOL_GPL(unregister_kretprobe);
1816 void __kprobes unregister_kretprobes(struct kretprobe **rps, int num)
1819 EXPORT_SYMBOL_GPL(unregister_kretprobes);
1821 static int __kprobes pre_handler_kretprobe(struct kprobe *p,
1822 struct pt_regs *regs)
1827 #endif /* CONFIG_KRETPROBES */
1829 /* Set the kprobe gone and remove its instruction buffer. */
1830 static void __kprobes kill_kprobe(struct kprobe *p)
1834 p->flags |= KPROBE_FLAG_GONE;
1835 if (kprobe_aggrprobe(p)) {
1837 * If this is an aggr_kprobe, we have to list all the
1838 * chained probes and mark them GONE.
1840 list_for_each_entry_rcu(kp, &p->list, list)
1841 kp->flags |= KPROBE_FLAG_GONE;
1842 p->post_handler = NULL;
1843 p->break_handler = NULL;
1844 kill_optimized_kprobe(p);
1847 * Here, we can remove insn_slot safely, because no thread calls
1848 * the original probed function (which will be freed soon) any more.
1850 arch_remove_kprobe(p);
1853 /* Disable one kprobe */
1854 int __kprobes disable_kprobe(struct kprobe *kp)
1858 mutex_lock(&kprobe_mutex);
1860 /* Disable this kprobe */
1861 if (__disable_kprobe(kp) == NULL)
1864 mutex_unlock(&kprobe_mutex);
1867 EXPORT_SYMBOL_GPL(disable_kprobe);
1869 /* Enable one kprobe */
1870 int __kprobes enable_kprobe(struct kprobe *kp)
1875 mutex_lock(&kprobe_mutex);
1877 /* Check whether specified probe is valid. */
1878 p = __get_valid_kprobe(kp);
1879 if (unlikely(p == NULL)) {
1884 if (kprobe_gone(kp)) {
1885 /* This kprobe has gone, we couldn't enable it. */
1891 kp->flags &= ~KPROBE_FLAG_DISABLED;
1893 if (!kprobes_all_disarmed && kprobe_disabled(p)) {
1894 p->flags &= ~KPROBE_FLAG_DISABLED;
1898 mutex_unlock(&kprobe_mutex);
1901 EXPORT_SYMBOL_GPL(enable_kprobe);
1903 void __kprobes dump_kprobe(struct kprobe *kp)
1905 printk(KERN_WARNING "Dumping kprobe:\n");
1906 printk(KERN_WARNING "Name: %s\nAddress: %p\nOffset: %x\n",
1907 kp->symbol_name, kp->addr, kp->offset);
1910 /* Module notifier call back, checking kprobes on the module */
1911 static int __kprobes kprobes_module_callback(struct notifier_block *nb,
1912 unsigned long val, void *data)
1914 struct module *mod = data;
1915 struct hlist_head *head;
1916 struct hlist_node *node;
1919 int checkcore = (val == MODULE_STATE_GOING);
1921 if (val != MODULE_STATE_GOING && val != MODULE_STATE_LIVE)
1925 * When MODULE_STATE_GOING was notified, both of module .text and
1926 * .init.text sections would be freed. When MODULE_STATE_LIVE was
1927 * notified, only .init.text section would be freed. We need to
1928 * disable kprobes which have been inserted in the sections.
1930 mutex_lock(&kprobe_mutex);
1931 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
1932 head = &kprobe_table[i];
1933 hlist_for_each_entry_rcu(p, node, head, hlist)
1934 if (within_module_init((unsigned long)p->addr, mod) ||
1936 within_module_core((unsigned long)p->addr, mod))) {
1938 * The vaddr this probe is installed will soon
1939 * be vfreed buy not synced to disk. Hence,
1940 * disarming the breakpoint isn't needed.
1945 mutex_unlock(&kprobe_mutex);
1949 static struct notifier_block kprobe_module_nb = {
1950 .notifier_call = kprobes_module_callback,
1954 static int __init init_kprobes(void)
1957 unsigned long offset = 0, size = 0;
1958 char *modname, namebuf[128];
1959 const char *symbol_name;
1961 struct kprobe_blackpoint *kb;
1963 /* FIXME allocate the probe table, currently defined statically */
1964 /* initialize all list heads */
1965 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
1966 INIT_HLIST_HEAD(&kprobe_table[i]);
1967 INIT_HLIST_HEAD(&kretprobe_inst_table[i]);
1968 raw_spin_lock_init(&(kretprobe_table_locks[i].lock));
1972 * Lookup and populate the kprobe_blacklist.
1974 * Unlike the kretprobe blacklist, we'll need to determine
1975 * the range of addresses that belong to the said functions,
1976 * since a kprobe need not necessarily be at the beginning
1979 for (kb = kprobe_blacklist; kb->name != NULL; kb++) {
1980 kprobe_lookup_name(kb->name, addr);
1984 kb->start_addr = (unsigned long)addr;
1985 symbol_name = kallsyms_lookup(kb->start_addr,
1986 &size, &offset, &modname, namebuf);
1993 if (kretprobe_blacklist_size) {
1994 /* lookup the function address from its name */
1995 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
1996 kprobe_lookup_name(kretprobe_blacklist[i].name,
1997 kretprobe_blacklist[i].addr);
1998 if (!kretprobe_blacklist[i].addr)
1999 printk("kretprobe: lookup failed: %s\n",
2000 kretprobe_blacklist[i].name);
2004 #if defined(CONFIG_OPTPROBES)
2005 #if defined(__ARCH_WANT_KPROBES_INSN_SLOT)
2006 /* Init kprobe_optinsn_slots */
2007 kprobe_optinsn_slots.insn_size = MAX_OPTINSN_SIZE;
2009 /* By default, kprobes can be optimized */
2010 kprobes_allow_optimization = true;
2013 /* By default, kprobes are armed */
2014 kprobes_all_disarmed = false;
2016 err = arch_init_kprobes();
2018 err = register_die_notifier(&kprobe_exceptions_nb);
2020 err = register_module_notifier(&kprobe_module_nb);
2022 kprobes_initialized = (err == 0);
2029 #ifdef CONFIG_DEBUG_FS
2030 static void __kprobes report_probe(struct seq_file *pi, struct kprobe *p,
2031 const char *sym, int offset, char *modname, struct kprobe *pp)
2035 if (p->pre_handler == pre_handler_kretprobe)
2037 else if (p->pre_handler == setjmp_pre_handler)
2043 seq_printf(pi, "%p %s %s+0x%x %s ",
2044 p->addr, kprobe_type, sym, offset,
2045 (modname ? modname : " "));
2047 seq_printf(pi, "%p %s %p ",
2048 p->addr, kprobe_type, p->addr);
2052 seq_printf(pi, "%s%s%s\n",
2053 (kprobe_gone(p) ? "[GONE]" : ""),
2054 ((kprobe_disabled(p) && !kprobe_gone(p)) ? "[DISABLED]" : ""),
2055 (kprobe_optimized(pp) ? "[OPTIMIZED]" : ""));
2058 static void __kprobes *kprobe_seq_start(struct seq_file *f, loff_t *pos)
2060 return (*pos < KPROBE_TABLE_SIZE) ? pos : NULL;
2063 static void __kprobes *kprobe_seq_next(struct seq_file *f, void *v, loff_t *pos)
2066 if (*pos >= KPROBE_TABLE_SIZE)
2071 static void __kprobes kprobe_seq_stop(struct seq_file *f, void *v)
2076 static int __kprobes show_kprobe_addr(struct seq_file *pi, void *v)
2078 struct hlist_head *head;
2079 struct hlist_node *node;
2080 struct kprobe *p, *kp;
2081 const char *sym = NULL;
2082 unsigned int i = *(loff_t *) v;
2083 unsigned long offset = 0;
2084 char *modname, namebuf[128];
2086 head = &kprobe_table[i];
2088 hlist_for_each_entry_rcu(p, node, head, hlist) {
2089 sym = kallsyms_lookup((unsigned long)p->addr, NULL,
2090 &offset, &modname, namebuf);
2091 if (kprobe_aggrprobe(p)) {
2092 list_for_each_entry_rcu(kp, &p->list, list)
2093 report_probe(pi, kp, sym, offset, modname, p);
2095 report_probe(pi, p, sym, offset, modname, NULL);
2101 static const struct seq_operations kprobes_seq_ops = {
2102 .start = kprobe_seq_start,
2103 .next = kprobe_seq_next,
2104 .stop = kprobe_seq_stop,
2105 .show = show_kprobe_addr
2108 static int __kprobes kprobes_open(struct inode *inode, struct file *filp)
2110 return seq_open(filp, &kprobes_seq_ops);
2113 static const struct file_operations debugfs_kprobes_operations = {
2114 .open = kprobes_open,
2116 .llseek = seq_lseek,
2117 .release = seq_release,
2120 static void __kprobes arm_all_kprobes(void)
2122 struct hlist_head *head;
2123 struct hlist_node *node;
2127 mutex_lock(&kprobe_mutex);
2129 /* If kprobes are armed, just return */
2130 if (!kprobes_all_disarmed)
2131 goto already_enabled;
2133 /* Arming kprobes doesn't optimize kprobe itself */
2134 mutex_lock(&text_mutex);
2135 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2136 head = &kprobe_table[i];
2137 hlist_for_each_entry_rcu(p, node, head, hlist)
2138 if (!kprobe_disabled(p))
2141 mutex_unlock(&text_mutex);
2143 kprobes_all_disarmed = false;
2144 printk(KERN_INFO "Kprobes globally enabled\n");
2147 mutex_unlock(&kprobe_mutex);
2151 static void __kprobes disarm_all_kprobes(void)
2153 struct hlist_head *head;
2154 struct hlist_node *node;
2158 mutex_lock(&kprobe_mutex);
2160 /* If kprobes are already disarmed, just return */
2161 if (kprobes_all_disarmed) {
2162 mutex_unlock(&kprobe_mutex);
2166 kprobes_all_disarmed = true;
2167 printk(KERN_INFO "Kprobes globally disabled\n");
2169 mutex_lock(&text_mutex);
2170 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2171 head = &kprobe_table[i];
2172 hlist_for_each_entry_rcu(p, node, head, hlist) {
2173 if (!arch_trampoline_kprobe(p) && !kprobe_disabled(p))
2174 __disarm_kprobe(p, false);
2177 mutex_unlock(&text_mutex);
2178 mutex_unlock(&kprobe_mutex);
2180 /* Wait for disarming all kprobes by optimizer */
2181 wait_for_kprobe_optimizer();
2185 * XXX: The debugfs bool file interface doesn't allow for callbacks
2186 * when the bool state is switched. We can reuse that facility when
2189 static ssize_t read_enabled_file_bool(struct file *file,
2190 char __user *user_buf, size_t count, loff_t *ppos)
2194 if (!kprobes_all_disarmed)
2200 return simple_read_from_buffer(user_buf, count, ppos, buf, 2);
2203 static ssize_t write_enabled_file_bool(struct file *file,
2204 const char __user *user_buf, size_t count, loff_t *ppos)
2209 buf_size = min(count, (sizeof(buf)-1));
2210 if (copy_from_user(buf, user_buf, buf_size))
2222 disarm_all_kprobes();
2229 static const struct file_operations fops_kp = {
2230 .read = read_enabled_file_bool,
2231 .write = write_enabled_file_bool,
2232 .llseek = default_llseek,
2235 static int __kprobes debugfs_kprobe_init(void)
2237 struct dentry *dir, *file;
2238 unsigned int value = 1;
2240 dir = debugfs_create_dir("kprobes", NULL);
2244 file = debugfs_create_file("list", 0444, dir, NULL,
2245 &debugfs_kprobes_operations);
2247 debugfs_remove(dir);
2251 file = debugfs_create_file("enabled", 0600, dir,
2254 debugfs_remove(dir);
2261 late_initcall(debugfs_kprobe_init);
2262 #endif /* CONFIG_DEBUG_FS */
2264 module_init(init_kprobes);
2266 /* defined in arch/.../kernel/kprobes.c */
2267 EXPORT_SYMBOL_GPL(jprobe_return);