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>
51 #include <asm-generic/sections.h>
52 #include <asm/cacheflush.h>
53 #include <asm/errno.h>
54 #include <asm/uaccess.h>
56 #define KPROBE_HASH_BITS 6
57 #define KPROBE_TABLE_SIZE (1 << KPROBE_HASH_BITS)
61 * Some oddball architectures like 64bit powerpc have function descriptors
62 * so this must be overridable.
64 #ifndef kprobe_lookup_name
65 #define kprobe_lookup_name(name, addr) \
66 addr = ((kprobe_opcode_t *)(kallsyms_lookup_name(name)))
69 static int kprobes_initialized;
70 static struct hlist_head kprobe_table[KPROBE_TABLE_SIZE];
71 static struct hlist_head kretprobe_inst_table[KPROBE_TABLE_SIZE];
73 /* NOTE: change this value only with kprobe_mutex held */
74 static bool kprobes_all_disarmed;
76 static DEFINE_MUTEX(kprobe_mutex); /* Protects kprobe_table */
77 static DEFINE_PER_CPU(struct kprobe *, kprobe_instance) = NULL;
79 spinlock_t lock ____cacheline_aligned_in_smp;
80 } kretprobe_table_locks[KPROBE_TABLE_SIZE];
82 static spinlock_t *kretprobe_table_lock_ptr(unsigned long hash)
84 return &(kretprobe_table_locks[hash].lock);
88 * Normally, functions that we'd want to prohibit kprobes in, are marked
89 * __kprobes. But, there are cases where such functions already belong to
90 * a different section (__sched for preempt_schedule)
92 * For such cases, we now have a blacklist
94 static struct kprobe_blackpoint kprobe_blacklist[] = {
95 {"preempt_schedule",},
96 {"native_get_debugreg",},
97 {"irq_entries_start",},
98 {"common_interrupt",},
99 {"mcount",}, /* mcount can be called from everywhere */
100 {NULL} /* Terminator */
103 #ifdef __ARCH_WANT_KPROBES_INSN_SLOT
105 * kprobe->ainsn.insn points to the copy of the instruction to be
106 * single-stepped. x86_64, POWER4 and above have no-exec support and
107 * stepping on the instruction on a vmalloced/kmalloced/data page
108 * is a recipe for disaster
110 struct kprobe_insn_page {
111 struct list_head list;
112 kprobe_opcode_t *insns; /* Page of instruction slots */
118 #define KPROBE_INSN_PAGE_SIZE(slots) \
119 (offsetof(struct kprobe_insn_page, slot_used) + \
120 (sizeof(char) * (slots)))
122 struct kprobe_insn_cache {
123 struct list_head pages; /* list of kprobe_insn_page */
124 size_t insn_size; /* size of instruction slot */
128 static int slots_per_page(struct kprobe_insn_cache *c)
130 return PAGE_SIZE/(c->insn_size * sizeof(kprobe_opcode_t));
133 enum kprobe_slot_state {
139 static DEFINE_MUTEX(kprobe_insn_mutex); /* Protects kprobe_insn_slots */
140 static struct kprobe_insn_cache kprobe_insn_slots = {
141 .pages = LIST_HEAD_INIT(kprobe_insn_slots.pages),
142 .insn_size = MAX_INSN_SIZE,
145 static int __kprobes collect_garbage_slots(struct kprobe_insn_cache *c);
148 * __get_insn_slot() - Find a slot on an executable page for an instruction.
149 * We allocate an executable page if there's no room on existing ones.
151 static kprobe_opcode_t __kprobes *__get_insn_slot(struct kprobe_insn_cache *c)
153 struct kprobe_insn_page *kip;
156 list_for_each_entry(kip, &c->pages, list) {
157 if (kip->nused < slots_per_page(c)) {
159 for (i = 0; i < slots_per_page(c); i++) {
160 if (kip->slot_used[i] == SLOT_CLEAN) {
161 kip->slot_used[i] = SLOT_USED;
163 return kip->insns + (i * c->insn_size);
166 /* kip->nused is broken. Fix it. */
167 kip->nused = slots_per_page(c);
172 /* If there are any garbage slots, collect it and try again. */
173 if (c->nr_garbage && collect_garbage_slots(c) == 0)
176 /* All out of space. Need to allocate a new page. */
177 kip = kmalloc(KPROBE_INSN_PAGE_SIZE(slots_per_page(c)), GFP_KERNEL);
182 * Use module_alloc so this page is within +/- 2GB of where the
183 * kernel image and loaded module images reside. This is required
184 * so x86_64 can correctly handle the %rip-relative fixups.
186 kip->insns = module_alloc(PAGE_SIZE);
191 INIT_LIST_HEAD(&kip->list);
192 memset(kip->slot_used, SLOT_CLEAN, slots_per_page(c));
193 kip->slot_used[0] = SLOT_USED;
196 list_add(&kip->list, &c->pages);
201 kprobe_opcode_t __kprobes *get_insn_slot(void)
203 kprobe_opcode_t *ret = NULL;
205 mutex_lock(&kprobe_insn_mutex);
206 ret = __get_insn_slot(&kprobe_insn_slots);
207 mutex_unlock(&kprobe_insn_mutex);
212 /* Return 1 if all garbages are collected, otherwise 0. */
213 static int __kprobes collect_one_slot(struct kprobe_insn_page *kip, int idx)
215 kip->slot_used[idx] = SLOT_CLEAN;
217 if (kip->nused == 0) {
219 * Page is no longer in use. Free it unless
220 * it's the last one. We keep the last one
221 * so as not to have to set it up again the
222 * next time somebody inserts a probe.
224 if (!list_is_singular(&kip->list)) {
225 list_del(&kip->list);
226 module_free(NULL, kip->insns);
234 static int __kprobes collect_garbage_slots(struct kprobe_insn_cache *c)
236 struct kprobe_insn_page *kip, *next;
238 /* Ensure no-one is interrupted on the garbages */
241 list_for_each_entry_safe(kip, next, &c->pages, list) {
243 if (kip->ngarbage == 0)
245 kip->ngarbage = 0; /* we will collect all garbages */
246 for (i = 0; i < slots_per_page(c); i++) {
247 if (kip->slot_used[i] == SLOT_DIRTY &&
248 collect_one_slot(kip, i))
256 static void __kprobes __free_insn_slot(struct kprobe_insn_cache *c,
257 kprobe_opcode_t *slot, int dirty)
259 struct kprobe_insn_page *kip;
261 list_for_each_entry(kip, &c->pages, list) {
262 long idx = ((long)slot - (long)kip->insns) /
263 (c->insn_size * sizeof(kprobe_opcode_t));
264 if (idx >= 0 && idx < slots_per_page(c)) {
265 WARN_ON(kip->slot_used[idx] != SLOT_USED);
267 kip->slot_used[idx] = SLOT_DIRTY;
269 if (++c->nr_garbage > slots_per_page(c))
270 collect_garbage_slots(c);
272 collect_one_slot(kip, idx);
276 /* Could not free this slot. */
280 void __kprobes free_insn_slot(kprobe_opcode_t * slot, int dirty)
282 mutex_lock(&kprobe_insn_mutex);
283 __free_insn_slot(&kprobe_insn_slots, slot, dirty);
284 mutex_unlock(&kprobe_insn_mutex);
286 #ifdef CONFIG_OPTPROBES
287 /* For optimized_kprobe buffer */
288 static DEFINE_MUTEX(kprobe_optinsn_mutex); /* Protects kprobe_optinsn_slots */
289 static struct kprobe_insn_cache kprobe_optinsn_slots = {
290 .pages = LIST_HEAD_INIT(kprobe_optinsn_slots.pages),
291 /* .insn_size is initialized later */
294 /* Get a slot for optimized_kprobe buffer */
295 kprobe_opcode_t __kprobes *get_optinsn_slot(void)
297 kprobe_opcode_t *ret = NULL;
299 mutex_lock(&kprobe_optinsn_mutex);
300 ret = __get_insn_slot(&kprobe_optinsn_slots);
301 mutex_unlock(&kprobe_optinsn_mutex);
306 void __kprobes free_optinsn_slot(kprobe_opcode_t * slot, int dirty)
308 mutex_lock(&kprobe_optinsn_mutex);
309 __free_insn_slot(&kprobe_optinsn_slots, slot, dirty);
310 mutex_unlock(&kprobe_optinsn_mutex);
315 /* We have preemption disabled.. so it is safe to use __ versions */
316 static inline void set_kprobe_instance(struct kprobe *kp)
318 __get_cpu_var(kprobe_instance) = kp;
321 static inline void reset_kprobe_instance(void)
323 __get_cpu_var(kprobe_instance) = NULL;
327 * This routine is called either:
328 * - under the kprobe_mutex - during kprobe_[un]register()
330 * - with preemption disabled - from arch/xxx/kernel/kprobes.c
332 struct kprobe __kprobes *get_kprobe(void *addr)
334 struct hlist_head *head;
335 struct hlist_node *node;
338 head = &kprobe_table[hash_ptr(addr, KPROBE_HASH_BITS)];
339 hlist_for_each_entry_rcu(p, node, head, hlist) {
347 static int __kprobes aggr_pre_handler(struct kprobe *p, struct pt_regs *regs);
349 /* Return true if the kprobe is an aggregator */
350 static inline int kprobe_aggrprobe(struct kprobe *p)
352 return p->pre_handler == aggr_pre_handler;
356 * Keep all fields in the kprobe consistent
358 static inline void copy_kprobe(struct kprobe *old_p, struct kprobe *p)
360 memcpy(&p->opcode, &old_p->opcode, sizeof(kprobe_opcode_t));
361 memcpy(&p->ainsn, &old_p->ainsn, sizeof(struct arch_specific_insn));
364 #ifdef CONFIG_OPTPROBES
365 /* NOTE: change this value only with kprobe_mutex held */
366 static bool kprobes_allow_optimization;
369 * Call all pre_handler on the list, but ignores its return value.
370 * This must be called from arch-dep optimized caller.
372 void __kprobes opt_pre_handler(struct kprobe *p, struct pt_regs *regs)
376 list_for_each_entry_rcu(kp, &p->list, list) {
377 if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
378 set_kprobe_instance(kp);
379 kp->pre_handler(kp, regs);
381 reset_kprobe_instance();
385 /* Return true(!0) if the kprobe is ready for optimization. */
386 static inline int kprobe_optready(struct kprobe *p)
388 struct optimized_kprobe *op;
390 if (kprobe_aggrprobe(p)) {
391 op = container_of(p, struct optimized_kprobe, kp);
392 return arch_prepared_optinsn(&op->optinsn);
399 * Return an optimized kprobe whose optimizing code replaces
400 * instructions including addr (exclude breakpoint).
402 static struct kprobe *__kprobes get_optimized_kprobe(unsigned long addr)
405 struct kprobe *p = NULL;
406 struct optimized_kprobe *op;
408 /* Don't check i == 0, since that is a breakpoint case. */
409 for (i = 1; !p && i < MAX_OPTIMIZED_LENGTH; i++)
410 p = get_kprobe((void *)(addr - i));
412 if (p && kprobe_optready(p)) {
413 op = container_of(p, struct optimized_kprobe, kp);
414 if (arch_within_optimized_kprobe(op, addr))
421 /* Optimization staging list, protected by kprobe_mutex */
422 static LIST_HEAD(optimizing_list);
424 static void kprobe_optimizer(struct work_struct *work);
425 static DECLARE_DELAYED_WORK(optimizing_work, kprobe_optimizer);
426 #define OPTIMIZE_DELAY 5
428 /* Kprobe jump optimizer */
429 static __kprobes void kprobe_optimizer(struct work_struct *work)
431 struct optimized_kprobe *op, *tmp;
433 /* Lock modules while optimizing kprobes */
434 mutex_lock(&module_mutex);
435 mutex_lock(&kprobe_mutex);
436 if (kprobes_all_disarmed || !kprobes_allow_optimization)
440 * Wait for quiesence period to ensure all running interrupts
441 * are done. Because optprobe may modify multiple instructions
442 * there is a chance that Nth instruction is interrupted. In that
443 * case, running interrupt can return to 2nd-Nth byte of jump
444 * instruction. This wait is for avoiding it.
449 * The optimization/unoptimization refers online_cpus via
450 * stop_machine() and cpu-hotplug modifies online_cpus.
451 * And same time, text_mutex will be held in cpu-hotplug and here.
452 * This combination can cause a deadlock (cpu-hotplug try to lock
453 * text_mutex but stop_machine can not be done because online_cpus
455 * To avoid this deadlock, we need to call get_online_cpus()
456 * for preventing cpu-hotplug outside of text_mutex locking.
459 mutex_lock(&text_mutex);
460 list_for_each_entry_safe(op, tmp, &optimizing_list, list) {
461 WARN_ON(kprobe_disabled(&op->kp));
462 if (arch_optimize_kprobe(op) < 0)
463 op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
464 list_del_init(&op->list);
466 mutex_unlock(&text_mutex);
469 mutex_unlock(&kprobe_mutex);
470 mutex_unlock(&module_mutex);
473 /* Optimize kprobe if p is ready to be optimized */
474 static __kprobes void optimize_kprobe(struct kprobe *p)
476 struct optimized_kprobe *op;
478 /* Check if the kprobe is disabled or not ready for optimization. */
479 if (!kprobe_optready(p) || !kprobes_allow_optimization ||
480 (kprobe_disabled(p) || kprobes_all_disarmed))
483 /* Both of break_handler and post_handler are not supported. */
484 if (p->break_handler || p->post_handler)
487 op = container_of(p, struct optimized_kprobe, kp);
489 /* Check there is no other kprobes at the optimized instructions */
490 if (arch_check_optimized_kprobe(op) < 0)
493 /* Check if it is already optimized. */
494 if (op->kp.flags & KPROBE_FLAG_OPTIMIZED)
497 op->kp.flags |= KPROBE_FLAG_OPTIMIZED;
498 list_add(&op->list, &optimizing_list);
499 if (!delayed_work_pending(&optimizing_work))
500 schedule_delayed_work(&optimizing_work, OPTIMIZE_DELAY);
503 /* Unoptimize a kprobe if p is optimized */
504 static __kprobes void unoptimize_kprobe(struct kprobe *p)
506 struct optimized_kprobe *op;
508 if ((p->flags & KPROBE_FLAG_OPTIMIZED) && kprobe_aggrprobe(p)) {
509 op = container_of(p, struct optimized_kprobe, kp);
510 if (!list_empty(&op->list))
511 /* Dequeue from the optimization queue */
512 list_del_init(&op->list);
514 /* Replace jump with break */
515 arch_unoptimize_kprobe(op);
516 op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
520 /* Remove optimized instructions */
521 static void __kprobes kill_optimized_kprobe(struct kprobe *p)
523 struct optimized_kprobe *op;
525 op = container_of(p, struct optimized_kprobe, kp);
526 if (!list_empty(&op->list)) {
527 /* Dequeue from the optimization queue */
528 list_del_init(&op->list);
529 op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
531 /* Don't unoptimize, because the target code will be freed. */
532 arch_remove_optimized_kprobe(op);
535 /* Try to prepare optimized instructions */
536 static __kprobes void prepare_optimized_kprobe(struct kprobe *p)
538 struct optimized_kprobe *op;
540 op = container_of(p, struct optimized_kprobe, kp);
541 arch_prepare_optimized_kprobe(op);
544 /* Free optimized instructions and optimized_kprobe */
545 static __kprobes void free_aggr_kprobe(struct kprobe *p)
547 struct optimized_kprobe *op;
549 op = container_of(p, struct optimized_kprobe, kp);
550 arch_remove_optimized_kprobe(op);
554 /* Allocate new optimized_kprobe and try to prepare optimized instructions */
555 static __kprobes struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
557 struct optimized_kprobe *op;
559 op = kzalloc(sizeof(struct optimized_kprobe), GFP_KERNEL);
563 INIT_LIST_HEAD(&op->list);
564 op->kp.addr = p->addr;
565 arch_prepare_optimized_kprobe(op);
570 static void __kprobes init_aggr_kprobe(struct kprobe *ap, struct kprobe *p);
573 * Prepare an optimized_kprobe and optimize it
574 * NOTE: p must be a normal registered kprobe
576 static __kprobes void try_to_optimize_kprobe(struct kprobe *p)
579 struct optimized_kprobe *op;
581 ap = alloc_aggr_kprobe(p);
585 op = container_of(ap, struct optimized_kprobe, kp);
586 if (!arch_prepared_optinsn(&op->optinsn)) {
587 /* If failed to setup optimizing, fallback to kprobe */
588 free_aggr_kprobe(ap);
592 init_aggr_kprobe(ap, p);
597 static void __kprobes optimize_all_kprobes(void)
599 struct hlist_head *head;
600 struct hlist_node *node;
604 /* If optimization is already allowed, just return */
605 if (kprobes_allow_optimization)
608 kprobes_allow_optimization = true;
609 mutex_lock(&text_mutex);
610 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
611 head = &kprobe_table[i];
612 hlist_for_each_entry_rcu(p, node, head, hlist)
613 if (!kprobe_disabled(p))
616 mutex_unlock(&text_mutex);
617 printk(KERN_INFO "Kprobes globally optimized\n");
620 static void __kprobes unoptimize_all_kprobes(void)
622 struct hlist_head *head;
623 struct hlist_node *node;
627 /* If optimization is already prohibited, just return */
628 if (!kprobes_allow_optimization)
631 kprobes_allow_optimization = false;
632 printk(KERN_INFO "Kprobes globally unoptimized\n");
633 get_online_cpus(); /* For avoiding text_mutex deadlock */
634 mutex_lock(&text_mutex);
635 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
636 head = &kprobe_table[i];
637 hlist_for_each_entry_rcu(p, node, head, hlist) {
638 if (!kprobe_disabled(p))
639 unoptimize_kprobe(p);
643 mutex_unlock(&text_mutex);
645 /* Allow all currently running kprobes to complete */
649 int sysctl_kprobes_optimization;
650 int proc_kprobes_optimization_handler(struct ctl_table *table, int write,
651 void __user *buffer, size_t *length,
656 mutex_lock(&kprobe_mutex);
657 sysctl_kprobes_optimization = kprobes_allow_optimization ? 1 : 0;
658 ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
660 if (sysctl_kprobes_optimization)
661 optimize_all_kprobes();
663 unoptimize_all_kprobes();
664 mutex_unlock(&kprobe_mutex);
668 #endif /* CONFIG_SYSCTL */
670 static void __kprobes __arm_kprobe(struct kprobe *p)
672 struct kprobe *old_p;
674 /* Check collision with other optimized kprobes */
675 old_p = get_optimized_kprobe((unsigned long)p->addr);
677 unoptimize_kprobe(old_p); /* Fallback to unoptimized kprobe */
680 optimize_kprobe(p); /* Try to optimize (add kprobe to a list) */
683 static void __kprobes __disarm_kprobe(struct kprobe *p)
685 struct kprobe *old_p;
687 unoptimize_kprobe(p); /* Try to unoptimize */
688 arch_disarm_kprobe(p);
690 /* If another kprobe was blocked, optimize it. */
691 old_p = get_optimized_kprobe((unsigned long)p->addr);
693 optimize_kprobe(old_p);
696 #else /* !CONFIG_OPTPROBES */
698 #define optimize_kprobe(p) do {} while (0)
699 #define unoptimize_kprobe(p) do {} while (0)
700 #define kill_optimized_kprobe(p) do {} while (0)
701 #define prepare_optimized_kprobe(p) do {} while (0)
702 #define try_to_optimize_kprobe(p) do {} while (0)
703 #define __arm_kprobe(p) arch_arm_kprobe(p)
704 #define __disarm_kprobe(p) arch_disarm_kprobe(p)
706 static __kprobes void free_aggr_kprobe(struct kprobe *p)
711 static __kprobes struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
713 return kzalloc(sizeof(struct kprobe), GFP_KERNEL);
715 #endif /* CONFIG_OPTPROBES */
717 /* Arm a kprobe with text_mutex */
718 static void __kprobes arm_kprobe(struct kprobe *kp)
721 * Here, since __arm_kprobe() doesn't use stop_machine(),
722 * this doesn't cause deadlock on text_mutex. So, we don't
723 * need get_online_cpus().
725 mutex_lock(&text_mutex);
727 mutex_unlock(&text_mutex);
730 /* Disarm a kprobe with text_mutex */
731 static void __kprobes disarm_kprobe(struct kprobe *kp)
733 get_online_cpus(); /* For avoiding text_mutex deadlock */
734 mutex_lock(&text_mutex);
736 mutex_unlock(&text_mutex);
741 * Aggregate handlers for multiple kprobes support - these handlers
742 * take care of invoking the individual kprobe handlers on p->list
744 static int __kprobes aggr_pre_handler(struct kprobe *p, struct pt_regs *regs)
748 list_for_each_entry_rcu(kp, &p->list, list) {
749 if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
750 set_kprobe_instance(kp);
751 if (kp->pre_handler(kp, regs))
754 reset_kprobe_instance();
759 static void __kprobes aggr_post_handler(struct kprobe *p, struct pt_regs *regs,
764 list_for_each_entry_rcu(kp, &p->list, list) {
765 if (kp->post_handler && likely(!kprobe_disabled(kp))) {
766 set_kprobe_instance(kp);
767 kp->post_handler(kp, regs, flags);
768 reset_kprobe_instance();
773 static int __kprobes aggr_fault_handler(struct kprobe *p, struct pt_regs *regs,
776 struct kprobe *cur = __get_cpu_var(kprobe_instance);
779 * if we faulted "during" the execution of a user specified
780 * probe handler, invoke just that probe's fault handler
782 if (cur && cur->fault_handler) {
783 if (cur->fault_handler(cur, regs, trapnr))
789 static int __kprobes aggr_break_handler(struct kprobe *p, struct pt_regs *regs)
791 struct kprobe *cur = __get_cpu_var(kprobe_instance);
794 if (cur && cur->break_handler) {
795 if (cur->break_handler(cur, regs))
798 reset_kprobe_instance();
802 /* Walks the list and increments nmissed count for multiprobe case */
803 void __kprobes kprobes_inc_nmissed_count(struct kprobe *p)
806 if (!kprobe_aggrprobe(p)) {
809 list_for_each_entry_rcu(kp, &p->list, list)
815 void __kprobes recycle_rp_inst(struct kretprobe_instance *ri,
816 struct hlist_head *head)
818 struct kretprobe *rp = ri->rp;
820 /* remove rp inst off the rprobe_inst_table */
821 hlist_del(&ri->hlist);
822 INIT_HLIST_NODE(&ri->hlist);
824 spin_lock(&rp->lock);
825 hlist_add_head(&ri->hlist, &rp->free_instances);
826 spin_unlock(&rp->lock);
829 hlist_add_head(&ri->hlist, head);
832 void __kprobes kretprobe_hash_lock(struct task_struct *tsk,
833 struct hlist_head **head, unsigned long *flags)
834 __acquires(hlist_lock)
836 unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
837 spinlock_t *hlist_lock;
839 *head = &kretprobe_inst_table[hash];
840 hlist_lock = kretprobe_table_lock_ptr(hash);
841 spin_lock_irqsave(hlist_lock, *flags);
844 static void __kprobes kretprobe_table_lock(unsigned long hash,
845 unsigned long *flags)
846 __acquires(hlist_lock)
848 spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
849 spin_lock_irqsave(hlist_lock, *flags);
852 void __kprobes kretprobe_hash_unlock(struct task_struct *tsk,
853 unsigned long *flags)
854 __releases(hlist_lock)
856 unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
857 spinlock_t *hlist_lock;
859 hlist_lock = kretprobe_table_lock_ptr(hash);
860 spin_unlock_irqrestore(hlist_lock, *flags);
863 static void __kprobes kretprobe_table_unlock(unsigned long hash,
864 unsigned long *flags)
865 __releases(hlist_lock)
867 spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
868 spin_unlock_irqrestore(hlist_lock, *flags);
872 * This function is called from finish_task_switch when task tk becomes dead,
873 * so that we can recycle any function-return probe instances associated
874 * with this task. These left over instances represent probed functions
875 * that have been called but will never return.
877 void __kprobes kprobe_flush_task(struct task_struct *tk)
879 struct kretprobe_instance *ri;
880 struct hlist_head *head, empty_rp;
881 struct hlist_node *node, *tmp;
882 unsigned long hash, flags = 0;
884 if (unlikely(!kprobes_initialized))
885 /* Early boot. kretprobe_table_locks not yet initialized. */
888 hash = hash_ptr(tk, KPROBE_HASH_BITS);
889 head = &kretprobe_inst_table[hash];
890 kretprobe_table_lock(hash, &flags);
891 hlist_for_each_entry_safe(ri, node, tmp, head, hlist) {
893 recycle_rp_inst(ri, &empty_rp);
895 kretprobe_table_unlock(hash, &flags);
896 INIT_HLIST_HEAD(&empty_rp);
897 hlist_for_each_entry_safe(ri, node, tmp, &empty_rp, hlist) {
898 hlist_del(&ri->hlist);
903 static inline void free_rp_inst(struct kretprobe *rp)
905 struct kretprobe_instance *ri;
906 struct hlist_node *pos, *next;
908 hlist_for_each_entry_safe(ri, pos, next, &rp->free_instances, hlist) {
909 hlist_del(&ri->hlist);
914 static void __kprobes cleanup_rp_inst(struct kretprobe *rp)
916 unsigned long flags, hash;
917 struct kretprobe_instance *ri;
918 struct hlist_node *pos, *next;
919 struct hlist_head *head;
922 for (hash = 0; hash < KPROBE_TABLE_SIZE; hash++) {
923 kretprobe_table_lock(hash, &flags);
924 head = &kretprobe_inst_table[hash];
925 hlist_for_each_entry_safe(ri, pos, next, head, hlist) {
929 kretprobe_table_unlock(hash, &flags);
935 * Add the new probe to ap->list. Fail if this is the
936 * second jprobe at the address - two jprobes can't coexist
938 static int __kprobes add_new_kprobe(struct kprobe *ap, struct kprobe *p)
940 BUG_ON(kprobe_gone(ap) || kprobe_gone(p));
942 if (p->break_handler || p->post_handler)
943 unoptimize_kprobe(ap); /* Fall back to normal kprobe */
945 if (p->break_handler) {
946 if (ap->break_handler)
948 list_add_tail_rcu(&p->list, &ap->list);
949 ap->break_handler = aggr_break_handler;
951 list_add_rcu(&p->list, &ap->list);
952 if (p->post_handler && !ap->post_handler)
953 ap->post_handler = aggr_post_handler;
955 if (kprobe_disabled(ap) && !kprobe_disabled(p)) {
956 ap->flags &= ~KPROBE_FLAG_DISABLED;
957 if (!kprobes_all_disarmed)
958 /* Arm the breakpoint again. */
965 * Fill in the required fields of the "manager kprobe". Replace the
966 * earlier kprobe in the hlist with the manager kprobe
968 static void __kprobes init_aggr_kprobe(struct kprobe *ap, struct kprobe *p)
970 /* Copy p's insn slot to ap */
974 ap->flags = p->flags & ~KPROBE_FLAG_OPTIMIZED;
975 ap->pre_handler = aggr_pre_handler;
976 ap->fault_handler = aggr_fault_handler;
977 /* We don't care the kprobe which has gone. */
978 if (p->post_handler && !kprobe_gone(p))
979 ap->post_handler = aggr_post_handler;
980 if (p->break_handler && !kprobe_gone(p))
981 ap->break_handler = aggr_break_handler;
983 INIT_LIST_HEAD(&ap->list);
984 INIT_HLIST_NODE(&ap->hlist);
986 list_add_rcu(&p->list, &ap->list);
987 hlist_replace_rcu(&p->hlist, &ap->hlist);
991 * This is the second or subsequent kprobe at the address - handle
994 static int __kprobes register_aggr_kprobe(struct kprobe *old_p,
998 struct kprobe *ap = old_p;
1000 if (!kprobe_aggrprobe(old_p)) {
1001 /* If old_p is not an aggr_kprobe, create new aggr_kprobe. */
1002 ap = alloc_aggr_kprobe(old_p);
1005 init_aggr_kprobe(ap, old_p);
1008 if (kprobe_gone(ap)) {
1010 * Attempting to insert new probe at the same location that
1011 * had a probe in the module vaddr area which already
1012 * freed. So, the instruction slot has already been
1013 * released. We need a new slot for the new probe.
1015 ret = arch_prepare_kprobe(ap);
1018 * Even if fail to allocate new slot, don't need to
1019 * free aggr_probe. It will be used next time, or
1020 * freed by unregister_kprobe.
1024 /* Prepare optimized instructions if possible. */
1025 prepare_optimized_kprobe(ap);
1028 * Clear gone flag to prevent allocating new slot again, and
1029 * set disabled flag because it is not armed yet.
1031 ap->flags = (ap->flags & ~KPROBE_FLAG_GONE)
1032 | KPROBE_FLAG_DISABLED;
1035 /* Copy ap's insn slot to p */
1037 return add_new_kprobe(ap, p);
1040 /* Try to disable aggr_kprobe, and return 1 if succeeded.*/
1041 static int __kprobes try_to_disable_aggr_kprobe(struct kprobe *p)
1045 list_for_each_entry_rcu(kp, &p->list, list) {
1046 if (!kprobe_disabled(kp))
1048 * There is an active probe on the list.
1049 * We can't disable aggr_kprobe.
1053 p->flags |= KPROBE_FLAG_DISABLED;
1057 static int __kprobes in_kprobes_functions(unsigned long addr)
1059 struct kprobe_blackpoint *kb;
1061 if (addr >= (unsigned long)__kprobes_text_start &&
1062 addr < (unsigned long)__kprobes_text_end)
1065 * If there exists a kprobe_blacklist, verify and
1066 * fail any probe registration in the prohibited area
1068 for (kb = kprobe_blacklist; kb->name != NULL; kb++) {
1069 if (kb->start_addr) {
1070 if (addr >= kb->start_addr &&
1071 addr < (kb->start_addr + kb->range))
1079 * If we have a symbol_name argument, look it up and add the offset field
1080 * to it. This way, we can specify a relative address to a symbol.
1082 static kprobe_opcode_t __kprobes *kprobe_addr(struct kprobe *p)
1084 kprobe_opcode_t *addr = p->addr;
1085 if (p->symbol_name) {
1088 kprobe_lookup_name(p->symbol_name, addr);
1093 return (kprobe_opcode_t *)(((char *)addr) + p->offset);
1096 /* Check passed kprobe is valid and return kprobe in kprobe_table. */
1097 static struct kprobe * __kprobes __get_valid_kprobe(struct kprobe *p)
1099 struct kprobe *old_p, *list_p;
1101 old_p = get_kprobe(p->addr);
1102 if (unlikely(!old_p))
1106 list_for_each_entry_rcu(list_p, &old_p->list, list)
1108 /* kprobe p is a valid probe */
1116 /* Return error if the kprobe is being re-registered */
1117 static inline int check_kprobe_rereg(struct kprobe *p)
1120 struct kprobe *old_p;
1122 mutex_lock(&kprobe_mutex);
1123 old_p = __get_valid_kprobe(p);
1126 mutex_unlock(&kprobe_mutex);
1130 int __kprobes register_kprobe(struct kprobe *p)
1133 struct kprobe *old_p;
1134 struct module *probed_mod;
1135 kprobe_opcode_t *addr;
1137 addr = kprobe_addr(p);
1142 ret = check_kprobe_rereg(p);
1147 if (!kernel_text_address((unsigned long) p->addr) ||
1148 in_kprobes_functions((unsigned long) p->addr) ||
1149 ftrace_text_reserved(p->addr, p->addr)) {
1154 /* User can pass only KPROBE_FLAG_DISABLED to register_kprobe */
1155 p->flags &= KPROBE_FLAG_DISABLED;
1158 * Check if are we probing a module.
1160 probed_mod = __module_text_address((unsigned long) p->addr);
1163 * We must hold a refcount of the probed module while updating
1164 * its code to prohibit unexpected unloading.
1166 if (unlikely(!try_module_get(probed_mod))) {
1171 * If the module freed .init.text, we couldn't insert
1174 if (within_module_init((unsigned long)p->addr, probed_mod) &&
1175 probed_mod->state != MODULE_STATE_COMING) {
1176 module_put(probed_mod);
1184 INIT_LIST_HEAD(&p->list);
1185 mutex_lock(&kprobe_mutex);
1187 get_online_cpus(); /* For avoiding text_mutex deadlock. */
1188 mutex_lock(&text_mutex);
1190 old_p = get_kprobe(p->addr);
1192 /* Since this may unoptimize old_p, locking text_mutex. */
1193 ret = register_aggr_kprobe(old_p, p);
1197 ret = arch_prepare_kprobe(p);
1201 INIT_HLIST_NODE(&p->hlist);
1202 hlist_add_head_rcu(&p->hlist,
1203 &kprobe_table[hash_ptr(p->addr, KPROBE_HASH_BITS)]);
1205 if (!kprobes_all_disarmed && !kprobe_disabled(p))
1208 /* Try to optimize kprobe */
1209 try_to_optimize_kprobe(p);
1212 mutex_unlock(&text_mutex);
1214 mutex_unlock(&kprobe_mutex);
1217 module_put(probed_mod);
1221 EXPORT_SYMBOL_GPL(register_kprobe);
1224 * Unregister a kprobe without a scheduler synchronization.
1226 static int __kprobes __unregister_kprobe_top(struct kprobe *p)
1228 struct kprobe *old_p, *list_p;
1230 old_p = __get_valid_kprobe(p);
1235 (kprobe_aggrprobe(old_p) &&
1236 list_is_singular(&old_p->list))) {
1238 * Only probe on the hash list. Disarm only if kprobes are
1239 * enabled and not gone - otherwise, the breakpoint would
1240 * already have been removed. We save on flushing icache.
1242 if (!kprobes_all_disarmed && !kprobe_disabled(old_p))
1243 disarm_kprobe(old_p);
1244 hlist_del_rcu(&old_p->hlist);
1246 if (p->break_handler && !kprobe_gone(p))
1247 old_p->break_handler = NULL;
1248 if (p->post_handler && !kprobe_gone(p)) {
1249 list_for_each_entry_rcu(list_p, &old_p->list, list) {
1250 if ((list_p != p) && (list_p->post_handler))
1253 old_p->post_handler = NULL;
1256 list_del_rcu(&p->list);
1257 if (!kprobe_disabled(old_p)) {
1258 try_to_disable_aggr_kprobe(old_p);
1259 if (!kprobes_all_disarmed) {
1260 if (kprobe_disabled(old_p))
1261 disarm_kprobe(old_p);
1263 /* Try to optimize this probe again */
1264 optimize_kprobe(old_p);
1271 static void __kprobes __unregister_kprobe_bottom(struct kprobe *p)
1273 struct kprobe *old_p;
1275 if (list_empty(&p->list))
1276 arch_remove_kprobe(p);
1277 else if (list_is_singular(&p->list)) {
1278 /* "p" is the last child of an aggr_kprobe */
1279 old_p = list_entry(p->list.next, struct kprobe, list);
1281 arch_remove_kprobe(old_p);
1282 free_aggr_kprobe(old_p);
1286 int __kprobes register_kprobes(struct kprobe **kps, int num)
1292 for (i = 0; i < num; i++) {
1293 ret = register_kprobe(kps[i]);
1296 unregister_kprobes(kps, i);
1302 EXPORT_SYMBOL_GPL(register_kprobes);
1304 void __kprobes unregister_kprobe(struct kprobe *p)
1306 unregister_kprobes(&p, 1);
1308 EXPORT_SYMBOL_GPL(unregister_kprobe);
1310 void __kprobes unregister_kprobes(struct kprobe **kps, int num)
1316 mutex_lock(&kprobe_mutex);
1317 for (i = 0; i < num; i++)
1318 if (__unregister_kprobe_top(kps[i]) < 0)
1319 kps[i]->addr = NULL;
1320 mutex_unlock(&kprobe_mutex);
1322 synchronize_sched();
1323 for (i = 0; i < num; i++)
1325 __unregister_kprobe_bottom(kps[i]);
1327 EXPORT_SYMBOL_GPL(unregister_kprobes);
1329 static struct notifier_block kprobe_exceptions_nb = {
1330 .notifier_call = kprobe_exceptions_notify,
1331 .priority = 0x7fffffff /* we need to be notified first */
1334 unsigned long __weak arch_deref_entry_point(void *entry)
1336 return (unsigned long)entry;
1339 int __kprobes register_jprobes(struct jprobe **jps, int num)
1346 for (i = 0; i < num; i++) {
1347 unsigned long addr, offset;
1349 addr = arch_deref_entry_point(jp->entry);
1351 /* Verify probepoint is a function entry point */
1352 if (kallsyms_lookup_size_offset(addr, NULL, &offset) &&
1354 jp->kp.pre_handler = setjmp_pre_handler;
1355 jp->kp.break_handler = longjmp_break_handler;
1356 ret = register_kprobe(&jp->kp);
1362 unregister_jprobes(jps, i);
1368 EXPORT_SYMBOL_GPL(register_jprobes);
1370 int __kprobes register_jprobe(struct jprobe *jp)
1372 return register_jprobes(&jp, 1);
1374 EXPORT_SYMBOL_GPL(register_jprobe);
1376 void __kprobes unregister_jprobe(struct jprobe *jp)
1378 unregister_jprobes(&jp, 1);
1380 EXPORT_SYMBOL_GPL(unregister_jprobe);
1382 void __kprobes unregister_jprobes(struct jprobe **jps, int num)
1388 mutex_lock(&kprobe_mutex);
1389 for (i = 0; i < num; i++)
1390 if (__unregister_kprobe_top(&jps[i]->kp) < 0)
1391 jps[i]->kp.addr = NULL;
1392 mutex_unlock(&kprobe_mutex);
1394 synchronize_sched();
1395 for (i = 0; i < num; i++) {
1396 if (jps[i]->kp.addr)
1397 __unregister_kprobe_bottom(&jps[i]->kp);
1400 EXPORT_SYMBOL_GPL(unregister_jprobes);
1402 #ifdef CONFIG_KRETPROBES
1404 * This kprobe pre_handler is registered with every kretprobe. When probe
1405 * hits it will set up the return probe.
1407 static int __kprobes pre_handler_kretprobe(struct kprobe *p,
1408 struct pt_regs *regs)
1410 struct kretprobe *rp = container_of(p, struct kretprobe, kp);
1411 unsigned long hash, flags = 0;
1412 struct kretprobe_instance *ri;
1414 /*TODO: consider to only swap the RA after the last pre_handler fired */
1415 hash = hash_ptr(current, KPROBE_HASH_BITS);
1416 spin_lock_irqsave(&rp->lock, flags);
1417 if (!hlist_empty(&rp->free_instances)) {
1418 ri = hlist_entry(rp->free_instances.first,
1419 struct kretprobe_instance, hlist);
1420 hlist_del(&ri->hlist);
1421 spin_unlock_irqrestore(&rp->lock, flags);
1426 if (rp->entry_handler && rp->entry_handler(ri, regs))
1429 arch_prepare_kretprobe(ri, regs);
1431 /* XXX(hch): why is there no hlist_move_head? */
1432 INIT_HLIST_NODE(&ri->hlist);
1433 kretprobe_table_lock(hash, &flags);
1434 hlist_add_head(&ri->hlist, &kretprobe_inst_table[hash]);
1435 kretprobe_table_unlock(hash, &flags);
1438 spin_unlock_irqrestore(&rp->lock, flags);
1443 int __kprobes register_kretprobe(struct kretprobe *rp)
1446 struct kretprobe_instance *inst;
1450 if (kretprobe_blacklist_size) {
1451 addr = kprobe_addr(&rp->kp);
1455 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
1456 if (kretprobe_blacklist[i].addr == addr)
1461 rp->kp.pre_handler = pre_handler_kretprobe;
1462 rp->kp.post_handler = NULL;
1463 rp->kp.fault_handler = NULL;
1464 rp->kp.break_handler = NULL;
1466 /* Pre-allocate memory for max kretprobe instances */
1467 if (rp->maxactive <= 0) {
1468 #ifdef CONFIG_PREEMPT
1469 rp->maxactive = max_t(unsigned int, 10, 2*num_possible_cpus());
1471 rp->maxactive = num_possible_cpus();
1474 spin_lock_init(&rp->lock);
1475 INIT_HLIST_HEAD(&rp->free_instances);
1476 for (i = 0; i < rp->maxactive; i++) {
1477 inst = kmalloc(sizeof(struct kretprobe_instance) +
1478 rp->data_size, GFP_KERNEL);
1483 INIT_HLIST_NODE(&inst->hlist);
1484 hlist_add_head(&inst->hlist, &rp->free_instances);
1488 /* Establish function entry probe point */
1489 ret = register_kprobe(&rp->kp);
1494 EXPORT_SYMBOL_GPL(register_kretprobe);
1496 int __kprobes register_kretprobes(struct kretprobe **rps, int num)
1502 for (i = 0; i < num; i++) {
1503 ret = register_kretprobe(rps[i]);
1506 unregister_kretprobes(rps, i);
1512 EXPORT_SYMBOL_GPL(register_kretprobes);
1514 void __kprobes unregister_kretprobe(struct kretprobe *rp)
1516 unregister_kretprobes(&rp, 1);
1518 EXPORT_SYMBOL_GPL(unregister_kretprobe);
1520 void __kprobes unregister_kretprobes(struct kretprobe **rps, int num)
1526 mutex_lock(&kprobe_mutex);
1527 for (i = 0; i < num; i++)
1528 if (__unregister_kprobe_top(&rps[i]->kp) < 0)
1529 rps[i]->kp.addr = NULL;
1530 mutex_unlock(&kprobe_mutex);
1532 synchronize_sched();
1533 for (i = 0; i < num; i++) {
1534 if (rps[i]->kp.addr) {
1535 __unregister_kprobe_bottom(&rps[i]->kp);
1536 cleanup_rp_inst(rps[i]);
1540 EXPORT_SYMBOL_GPL(unregister_kretprobes);
1542 #else /* CONFIG_KRETPROBES */
1543 int __kprobes register_kretprobe(struct kretprobe *rp)
1547 EXPORT_SYMBOL_GPL(register_kretprobe);
1549 int __kprobes register_kretprobes(struct kretprobe **rps, int num)
1553 EXPORT_SYMBOL_GPL(register_kretprobes);
1555 void __kprobes unregister_kretprobe(struct kretprobe *rp)
1558 EXPORT_SYMBOL_GPL(unregister_kretprobe);
1560 void __kprobes unregister_kretprobes(struct kretprobe **rps, int num)
1563 EXPORT_SYMBOL_GPL(unregister_kretprobes);
1565 static int __kprobes pre_handler_kretprobe(struct kprobe *p,
1566 struct pt_regs *regs)
1571 #endif /* CONFIG_KRETPROBES */
1573 /* Set the kprobe gone and remove its instruction buffer. */
1574 static void __kprobes kill_kprobe(struct kprobe *p)
1578 p->flags |= KPROBE_FLAG_GONE;
1579 if (kprobe_aggrprobe(p)) {
1581 * If this is an aggr_kprobe, we have to list all the
1582 * chained probes and mark them GONE.
1584 list_for_each_entry_rcu(kp, &p->list, list)
1585 kp->flags |= KPROBE_FLAG_GONE;
1586 p->post_handler = NULL;
1587 p->break_handler = NULL;
1588 kill_optimized_kprobe(p);
1591 * Here, we can remove insn_slot safely, because no thread calls
1592 * the original probed function (which will be freed soon) any more.
1594 arch_remove_kprobe(p);
1597 /* Disable one kprobe */
1598 int __kprobes disable_kprobe(struct kprobe *kp)
1603 mutex_lock(&kprobe_mutex);
1605 /* Check whether specified probe is valid. */
1606 p = __get_valid_kprobe(kp);
1607 if (unlikely(p == NULL)) {
1612 /* If the probe is already disabled (or gone), just return */
1613 if (kprobe_disabled(kp))
1616 kp->flags |= KPROBE_FLAG_DISABLED;
1618 /* When kp != p, p is always enabled. */
1619 try_to_disable_aggr_kprobe(p);
1621 if (!kprobes_all_disarmed && kprobe_disabled(p))
1624 mutex_unlock(&kprobe_mutex);
1627 EXPORT_SYMBOL_GPL(disable_kprobe);
1629 /* Enable one kprobe */
1630 int __kprobes enable_kprobe(struct kprobe *kp)
1635 mutex_lock(&kprobe_mutex);
1637 /* Check whether specified probe is valid. */
1638 p = __get_valid_kprobe(kp);
1639 if (unlikely(p == NULL)) {
1644 if (kprobe_gone(kp)) {
1645 /* This kprobe has gone, we couldn't enable it. */
1651 kp->flags &= ~KPROBE_FLAG_DISABLED;
1653 if (!kprobes_all_disarmed && kprobe_disabled(p)) {
1654 p->flags &= ~KPROBE_FLAG_DISABLED;
1658 mutex_unlock(&kprobe_mutex);
1661 EXPORT_SYMBOL_GPL(enable_kprobe);
1663 void __kprobes dump_kprobe(struct kprobe *kp)
1665 printk(KERN_WARNING "Dumping kprobe:\n");
1666 printk(KERN_WARNING "Name: %s\nAddress: %p\nOffset: %x\n",
1667 kp->symbol_name, kp->addr, kp->offset);
1670 /* Module notifier call back, checking kprobes on the module */
1671 static int __kprobes kprobes_module_callback(struct notifier_block *nb,
1672 unsigned long val, void *data)
1674 struct module *mod = data;
1675 struct hlist_head *head;
1676 struct hlist_node *node;
1679 int checkcore = (val == MODULE_STATE_GOING);
1681 if (val != MODULE_STATE_GOING && val != MODULE_STATE_LIVE)
1685 * When MODULE_STATE_GOING was notified, both of module .text and
1686 * .init.text sections would be freed. When MODULE_STATE_LIVE was
1687 * notified, only .init.text section would be freed. We need to
1688 * disable kprobes which have been inserted in the sections.
1690 mutex_lock(&kprobe_mutex);
1691 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
1692 head = &kprobe_table[i];
1693 hlist_for_each_entry_rcu(p, node, head, hlist)
1694 if (within_module_init((unsigned long)p->addr, mod) ||
1696 within_module_core((unsigned long)p->addr, mod))) {
1698 * The vaddr this probe is installed will soon
1699 * be vfreed buy not synced to disk. Hence,
1700 * disarming the breakpoint isn't needed.
1705 mutex_unlock(&kprobe_mutex);
1709 static struct notifier_block kprobe_module_nb = {
1710 .notifier_call = kprobes_module_callback,
1714 static int __init init_kprobes(void)
1717 unsigned long offset = 0, size = 0;
1718 char *modname, namebuf[128];
1719 const char *symbol_name;
1721 struct kprobe_blackpoint *kb;
1723 /* FIXME allocate the probe table, currently defined statically */
1724 /* initialize all list heads */
1725 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
1726 INIT_HLIST_HEAD(&kprobe_table[i]);
1727 INIT_HLIST_HEAD(&kretprobe_inst_table[i]);
1728 spin_lock_init(&(kretprobe_table_locks[i].lock));
1732 * Lookup and populate the kprobe_blacklist.
1734 * Unlike the kretprobe blacklist, we'll need to determine
1735 * the range of addresses that belong to the said functions,
1736 * since a kprobe need not necessarily be at the beginning
1739 for (kb = kprobe_blacklist; kb->name != NULL; kb++) {
1740 kprobe_lookup_name(kb->name, addr);
1744 kb->start_addr = (unsigned long)addr;
1745 symbol_name = kallsyms_lookup(kb->start_addr,
1746 &size, &offset, &modname, namebuf);
1753 if (kretprobe_blacklist_size) {
1754 /* lookup the function address from its name */
1755 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
1756 kprobe_lookup_name(kretprobe_blacklist[i].name,
1757 kretprobe_blacklist[i].addr);
1758 if (!kretprobe_blacklist[i].addr)
1759 printk("kretprobe: lookup failed: %s\n",
1760 kretprobe_blacklist[i].name);
1764 #if defined(CONFIG_OPTPROBES)
1765 #if defined(__ARCH_WANT_KPROBES_INSN_SLOT)
1766 /* Init kprobe_optinsn_slots */
1767 kprobe_optinsn_slots.insn_size = MAX_OPTINSN_SIZE;
1769 /* By default, kprobes can be optimized */
1770 kprobes_allow_optimization = true;
1773 /* By default, kprobes are armed */
1774 kprobes_all_disarmed = false;
1776 err = arch_init_kprobes();
1778 err = register_die_notifier(&kprobe_exceptions_nb);
1780 err = register_module_notifier(&kprobe_module_nb);
1782 kprobes_initialized = (err == 0);
1789 #ifdef CONFIG_DEBUG_FS
1790 static void __kprobes report_probe(struct seq_file *pi, struct kprobe *p,
1791 const char *sym, int offset, char *modname, struct kprobe *pp)
1795 if (p->pre_handler == pre_handler_kretprobe)
1797 else if (p->pre_handler == setjmp_pre_handler)
1803 seq_printf(pi, "%p %s %s+0x%x %s ",
1804 p->addr, kprobe_type, sym, offset,
1805 (modname ? modname : " "));
1807 seq_printf(pi, "%p %s %p ",
1808 p->addr, kprobe_type, p->addr);
1812 seq_printf(pi, "%s%s%s\n",
1813 (kprobe_gone(p) ? "[GONE]" : ""),
1814 ((kprobe_disabled(p) && !kprobe_gone(p)) ? "[DISABLED]" : ""),
1815 (kprobe_optimized(pp) ? "[OPTIMIZED]" : ""));
1818 static void __kprobes *kprobe_seq_start(struct seq_file *f, loff_t *pos)
1820 return (*pos < KPROBE_TABLE_SIZE) ? pos : NULL;
1823 static void __kprobes *kprobe_seq_next(struct seq_file *f, void *v, loff_t *pos)
1826 if (*pos >= KPROBE_TABLE_SIZE)
1831 static void __kprobes kprobe_seq_stop(struct seq_file *f, void *v)
1836 static int __kprobes show_kprobe_addr(struct seq_file *pi, void *v)
1838 struct hlist_head *head;
1839 struct hlist_node *node;
1840 struct kprobe *p, *kp;
1841 const char *sym = NULL;
1842 unsigned int i = *(loff_t *) v;
1843 unsigned long offset = 0;
1844 char *modname, namebuf[128];
1846 head = &kprobe_table[i];
1848 hlist_for_each_entry_rcu(p, node, head, hlist) {
1849 sym = kallsyms_lookup((unsigned long)p->addr, NULL,
1850 &offset, &modname, namebuf);
1851 if (kprobe_aggrprobe(p)) {
1852 list_for_each_entry_rcu(kp, &p->list, list)
1853 report_probe(pi, kp, sym, offset, modname, p);
1855 report_probe(pi, p, sym, offset, modname, NULL);
1861 static const struct seq_operations kprobes_seq_ops = {
1862 .start = kprobe_seq_start,
1863 .next = kprobe_seq_next,
1864 .stop = kprobe_seq_stop,
1865 .show = show_kprobe_addr
1868 static int __kprobes kprobes_open(struct inode *inode, struct file *filp)
1870 return seq_open(filp, &kprobes_seq_ops);
1873 static const struct file_operations debugfs_kprobes_operations = {
1874 .open = kprobes_open,
1876 .llseek = seq_lseek,
1877 .release = seq_release,
1880 static void __kprobes arm_all_kprobes(void)
1882 struct hlist_head *head;
1883 struct hlist_node *node;
1887 mutex_lock(&kprobe_mutex);
1889 /* If kprobes are armed, just return */
1890 if (!kprobes_all_disarmed)
1891 goto already_enabled;
1893 /* Arming kprobes doesn't optimize kprobe itself */
1894 mutex_lock(&text_mutex);
1895 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
1896 head = &kprobe_table[i];
1897 hlist_for_each_entry_rcu(p, node, head, hlist)
1898 if (!kprobe_disabled(p))
1901 mutex_unlock(&text_mutex);
1903 kprobes_all_disarmed = false;
1904 printk(KERN_INFO "Kprobes globally enabled\n");
1907 mutex_unlock(&kprobe_mutex);
1911 static void __kprobes disarm_all_kprobes(void)
1913 struct hlist_head *head;
1914 struct hlist_node *node;
1918 mutex_lock(&kprobe_mutex);
1920 /* If kprobes are already disarmed, just return */
1921 if (kprobes_all_disarmed)
1922 goto already_disabled;
1924 kprobes_all_disarmed = true;
1925 printk(KERN_INFO "Kprobes globally disabled\n");
1928 * Here we call get_online_cpus() for avoiding text_mutex deadlock,
1929 * because disarming may also unoptimize kprobes.
1932 mutex_lock(&text_mutex);
1933 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
1934 head = &kprobe_table[i];
1935 hlist_for_each_entry_rcu(p, node, head, hlist) {
1936 if (!arch_trampoline_kprobe(p) && !kprobe_disabled(p))
1941 mutex_unlock(&text_mutex);
1943 mutex_unlock(&kprobe_mutex);
1944 /* Allow all currently running kprobes to complete */
1945 synchronize_sched();
1949 mutex_unlock(&kprobe_mutex);
1954 * XXX: The debugfs bool file interface doesn't allow for callbacks
1955 * when the bool state is switched. We can reuse that facility when
1958 static ssize_t read_enabled_file_bool(struct file *file,
1959 char __user *user_buf, size_t count, loff_t *ppos)
1963 if (!kprobes_all_disarmed)
1969 return simple_read_from_buffer(user_buf, count, ppos, buf, 2);
1972 static ssize_t write_enabled_file_bool(struct file *file,
1973 const char __user *user_buf, size_t count, loff_t *ppos)
1978 buf_size = min(count, (sizeof(buf)-1));
1979 if (copy_from_user(buf, user_buf, buf_size))
1991 disarm_all_kprobes();
1998 static const struct file_operations fops_kp = {
1999 .read = read_enabled_file_bool,
2000 .write = write_enabled_file_bool,
2003 static int __kprobes debugfs_kprobe_init(void)
2005 struct dentry *dir, *file;
2006 unsigned int value = 1;
2008 dir = debugfs_create_dir("kprobes", NULL);
2012 file = debugfs_create_file("list", 0444, dir, NULL,
2013 &debugfs_kprobes_operations);
2015 debugfs_remove(dir);
2019 file = debugfs_create_file("enabled", 0600, dir,
2022 debugfs_remove(dir);
2029 late_initcall(debugfs_kprobe_init);
2030 #endif /* CONFIG_DEBUG_FS */
2032 module_init(init_kprobes);
2034 /* defined in arch/.../kernel/kprobes.c */
2035 EXPORT_SYMBOL_GPL(jprobe_return);