nick piggin: change email address
[linux-2.6.git] / kernel / kprobes.c
index 6524252..282035f 100644 (file)
  *             <prasanna@in.ibm.com> added function-return probes.
  */
 #include <linux/kprobes.h>
-#include <linux/spinlock.h>
 #include <linux/hash.h>
 #include <linux/init.h>
+#include <linux/slab.h>
+#include <linux/stddef.h>
 #include <linux/module.h>
 #include <linux/moduleloader.h>
+#include <linux/kallsyms.h>
+#include <linux/freezer.h>
+#include <linux/seq_file.h>
+#include <linux/debugfs.h>
+#include <linux/sysctl.h>
+#include <linux/kdebug.h>
+#include <linux/memory.h>
+#include <linux/ftrace.h>
+#include <linux/cpu.h>
+
+#include <asm-generic/sections.h>
 #include <asm/cacheflush.h>
 #include <asm/errno.h>
-#include <asm/kdebug.h>
+#include <asm/uaccess.h>
 
 #define KPROBE_HASH_BITS 6
 #define KPROBE_TABLE_SIZE (1 << KPROBE_HASH_BITS)
 
+
+/*
+ * Some oddball architectures like 64bit powerpc have function descriptors
+ * so this must be overridable.
+ */
+#ifndef kprobe_lookup_name
+#define kprobe_lookup_name(name, addr) \
+       addr = ((kprobe_opcode_t *)(kallsyms_lookup_name(name)))
+#endif
+
+static int kprobes_initialized;
 static struct hlist_head kprobe_table[KPROBE_TABLE_SIZE];
 static struct hlist_head kretprobe_inst_table[KPROBE_TABLE_SIZE];
 
-unsigned int kprobe_cpu = NR_CPUS;
-static DEFINE_SPINLOCK(kprobe_lock);
-static struct kprobe *curr_kprobe;
+/* NOTE: change this value only with kprobe_mutex held */
+static bool kprobes_all_disarmed;
+
+static DEFINE_MUTEX(kprobe_mutex);     /* Protects kprobe_table */
+static DEFINE_PER_CPU(struct kprobe *, kprobe_instance) = NULL;
+static struct {
+       spinlock_t lock ____cacheline_aligned_in_smp;
+} kretprobe_table_locks[KPROBE_TABLE_SIZE];
+
+static spinlock_t *kretprobe_table_lock_ptr(unsigned long hash)
+{
+       return &(kretprobe_table_locks[hash].lock);
+}
 
 /*
+ * Normally, functions that we'd want to prohibit kprobes in, are marked
+ * __kprobes. But, there are cases where such functions already belong to
+ * a different section (__sched for preempt_schedule)
+ *
+ * For such cases, we now have a blacklist
+ */
+static struct kprobe_blackpoint kprobe_blacklist[] = {
+       {"preempt_schedule",},
+       {"native_get_debugreg",},
+       {"irq_entries_start",},
+       {"common_interrupt",},
+       {"mcount",},    /* mcount can be called from everywhere */
+       {NULL}    /* Terminator */
+};
+
+#ifdef __ARCH_WANT_KPROBES_INSN_SLOT
+/*
  * kprobe->ainsn.insn points to the copy of the instruction to be
  * single-stepped. x86_64, POWER4 and above have no-exec support and
  * stepping on the instruction on a vmalloced/kmalloced/data page
  * is a recipe for disaster
  */
-#define INSNS_PER_PAGE (PAGE_SIZE/(MAX_INSN_SIZE * sizeof(kprobe_opcode_t)))
-
 struct kprobe_insn_page {
-       struct hlist_node hlist;
+       struct list_head list;
        kprobe_opcode_t *insns;         /* Page of instruction slots */
-       char slot_used[INSNS_PER_PAGE];
        int nused;
+       int ngarbage;
+       char slot_used[];
+};
+
+#define KPROBE_INSN_PAGE_SIZE(slots)                   \
+       (offsetof(struct kprobe_insn_page, slot_used) + \
+        (sizeof(char) * (slots)))
+
+struct kprobe_insn_cache {
+       struct list_head pages; /* list of kprobe_insn_page */
+       size_t insn_size;       /* size of instruction slot */
+       int nr_garbage;
+};
+
+static int slots_per_page(struct kprobe_insn_cache *c)
+{
+       return PAGE_SIZE/(c->insn_size * sizeof(kprobe_opcode_t));
+}
+
+enum kprobe_slot_state {
+       SLOT_CLEAN = 0,
+       SLOT_DIRTY = 1,
+       SLOT_USED = 2,
 };
 
-static struct hlist_head kprobe_insn_pages;
+static DEFINE_MUTEX(kprobe_insn_mutex);        /* Protects kprobe_insn_slots */
+static struct kprobe_insn_cache kprobe_insn_slots = {
+       .pages = LIST_HEAD_INIT(kprobe_insn_slots.pages),
+       .insn_size = MAX_INSN_SIZE,
+       .nr_garbage = 0,
+};
+static int __kprobes collect_garbage_slots(struct kprobe_insn_cache *c);
 
 /**
- * get_insn_slot() - Find a slot on an executable page for an instruction.
+ * __get_insn_slot() - Find a slot on an executable page for an instruction.
  * We allocate an executable page if there's no room on existing ones.
  */
-kprobe_opcode_t *get_insn_slot(void)
+static kprobe_opcode_t __kprobes *__get_insn_slot(struct kprobe_insn_cache *c)
 {
        struct kprobe_insn_page *kip;
-       struct hlist_node *pos;
 
-       hlist_for_each(pos, &kprobe_insn_pages) {
-               kip = hlist_entry(pos, struct kprobe_insn_page, hlist);
-               if (kip->nused < INSNS_PER_PAGE) {
+ retry:
+       list_for_each_entry(kip, &c->pages, list) {
+               if (kip->nused < slots_per_page(c)) {
                        int i;
-                       for (i = 0; i < INSNS_PER_PAGE; i++) {
-                               if (!kip->slot_used[i]) {
-                                       kip->slot_used[i] = 1;
+                       for (i = 0; i < slots_per_page(c); i++) {
+                               if (kip->slot_used[i] == SLOT_CLEAN) {
+                                       kip->slot_used[i] = SLOT_USED;
                                        kip->nused++;
-                                       return kip->insns + (i * MAX_INSN_SIZE);
+                                       return kip->insns + (i * c->insn_size);
                                }
                        }
-                       /* Surprise!  No unused slots.  Fix kip->nused. */
-                       kip->nused = INSNS_PER_PAGE;
+                       /* kip->nused is broken. Fix it. */
+                       kip->nused = slots_per_page(c);
+                       WARN_ON(1);
                }
        }
 
-       /* All out of space.  Need to allocate a new page. Use slot 0.*/
-       kip = kmalloc(sizeof(struct kprobe_insn_page), GFP_KERNEL);
-       if (!kip) {
+       /* If there are any garbage slots, collect it and try again. */
+       if (c->nr_garbage && collect_garbage_slots(c) == 0)
+               goto retry;
+
+       /* All out of space.  Need to allocate a new page. */
+       kip = kmalloc(KPROBE_INSN_PAGE_SIZE(slots_per_page(c)), GFP_KERNEL);
+       if (!kip)
                return NULL;
-       }
 
        /*
         * Use module_alloc so this page is within +/- 2GB of where the
@@ -109,294 +188,771 @@ kprobe_opcode_t *get_insn_slot(void)
                kfree(kip);
                return NULL;
        }
-       INIT_HLIST_NODE(&kip->hlist);
-       hlist_add_head(&kip->hlist, &kprobe_insn_pages);
-       memset(kip->slot_used, 0, INSNS_PER_PAGE);
-       kip->slot_used[0] = 1;
+       INIT_LIST_HEAD(&kip->list);
+       memset(kip->slot_used, SLOT_CLEAN, slots_per_page(c));
+       kip->slot_used[0] = SLOT_USED;
        kip->nused = 1;
+       kip->ngarbage = 0;
+       list_add(&kip->list, &c->pages);
        return kip->insns;
 }
 
-void free_insn_slot(kprobe_opcode_t *slot)
+
+kprobe_opcode_t __kprobes *get_insn_slot(void)
+{
+       kprobe_opcode_t *ret = NULL;
+
+       mutex_lock(&kprobe_insn_mutex);
+       ret = __get_insn_slot(&kprobe_insn_slots);
+       mutex_unlock(&kprobe_insn_mutex);
+
+       return ret;
+}
+
+/* Return 1 if all garbages are collected, otherwise 0. */
+static int __kprobes collect_one_slot(struct kprobe_insn_page *kip, int idx)
+{
+       kip->slot_used[idx] = SLOT_CLEAN;
+       kip->nused--;
+       if (kip->nused == 0) {
+               /*
+                * Page is no longer in use.  Free it unless
+                * it's the last one.  We keep the last one
+                * so as not to have to set it up again the
+                * next time somebody inserts a probe.
+                */
+               if (!list_is_singular(&kip->list)) {
+                       list_del(&kip->list);
+                       module_free(NULL, kip->insns);
+                       kfree(kip);
+               }
+               return 1;
+       }
+       return 0;
+}
+
+static int __kprobes collect_garbage_slots(struct kprobe_insn_cache *c)
+{
+       struct kprobe_insn_page *kip, *next;
+
+       /* Ensure no-one is interrupted on the garbages */
+       synchronize_sched();
+
+       list_for_each_entry_safe(kip, next, &c->pages, list) {
+               int i;
+               if (kip->ngarbage == 0)
+                       continue;
+               kip->ngarbage = 0;      /* we will collect all garbages */
+               for (i = 0; i < slots_per_page(c); i++) {
+                       if (kip->slot_used[i] == SLOT_DIRTY &&
+                           collect_one_slot(kip, i))
+                               break;
+               }
+       }
+       c->nr_garbage = 0;
+       return 0;
+}
+
+static void __kprobes __free_insn_slot(struct kprobe_insn_cache *c,
+                                      kprobe_opcode_t *slot, int dirty)
 {
        struct kprobe_insn_page *kip;
-       struct hlist_node *pos;
-
-       hlist_for_each(pos, &kprobe_insn_pages) {
-               kip = hlist_entry(pos, struct kprobe_insn_page, hlist);
-               if (kip->insns <= slot &&
-                   slot < kip->insns + (INSNS_PER_PAGE * MAX_INSN_SIZE)) {
-                       int i = (slot - kip->insns) / MAX_INSN_SIZE;
-                       kip->slot_used[i] = 0;
-                       kip->nused--;
-                       if (kip->nused == 0) {
-                               /*
-                                * Page is no longer in use.  Free it unless
-                                * it's the last one.  We keep the last one
-                                * so as not to have to set it up again the
-                                * next time somebody inserts a probe.
-                                */
-                               hlist_del(&kip->hlist);
-                               if (hlist_empty(&kprobe_insn_pages)) {
-                                       INIT_HLIST_NODE(&kip->hlist);
-                                       hlist_add_head(&kip->hlist,
-                                               &kprobe_insn_pages);
-                               } else {
-                                       module_free(NULL, kip->insns);
-                                       kfree(kip);
-                               }
-                       }
+
+       list_for_each_entry(kip, &c->pages, list) {
+               long idx = ((long)slot - (long)kip->insns) /
+                               (c->insn_size * sizeof(kprobe_opcode_t));
+               if (idx >= 0 && idx < slots_per_page(c)) {
+                       WARN_ON(kip->slot_used[idx] != SLOT_USED);
+                       if (dirty) {
+                               kip->slot_used[idx] = SLOT_DIRTY;
+                               kip->ngarbage++;
+                               if (++c->nr_garbage > slots_per_page(c))
+                                       collect_garbage_slots(c);
+                       } else
+                               collect_one_slot(kip, idx);
                        return;
                }
        }
+       /* Could not free this slot. */
+       WARN_ON(1);
+}
+
+void __kprobes free_insn_slot(kprobe_opcode_t * slot, int dirty)
+{
+       mutex_lock(&kprobe_insn_mutex);
+       __free_insn_slot(&kprobe_insn_slots, slot, dirty);
+       mutex_unlock(&kprobe_insn_mutex);
+}
+#ifdef CONFIG_OPTPROBES
+/* For optimized_kprobe buffer */
+static DEFINE_MUTEX(kprobe_optinsn_mutex); /* Protects kprobe_optinsn_slots */
+static struct kprobe_insn_cache kprobe_optinsn_slots = {
+       .pages = LIST_HEAD_INIT(kprobe_optinsn_slots.pages),
+       /* .insn_size is initialized later */
+       .nr_garbage = 0,
+};
+/* Get a slot for optimized_kprobe buffer */
+kprobe_opcode_t __kprobes *get_optinsn_slot(void)
+{
+       kprobe_opcode_t *ret = NULL;
+
+       mutex_lock(&kprobe_optinsn_mutex);
+       ret = __get_insn_slot(&kprobe_optinsn_slots);
+       mutex_unlock(&kprobe_optinsn_mutex);
+
+       return ret;
+}
+
+void __kprobes free_optinsn_slot(kprobe_opcode_t * slot, int dirty)
+{
+       mutex_lock(&kprobe_optinsn_mutex);
+       __free_insn_slot(&kprobe_optinsn_slots, slot, dirty);
+       mutex_unlock(&kprobe_optinsn_mutex);
 }
+#endif
+#endif
 
-/* Locks kprobe: irqs must be disabled */
-void lock_kprobes(void)
+/* We have preemption disabled.. so it is safe to use __ versions */
+static inline void set_kprobe_instance(struct kprobe *kp)
 {
-       spin_lock(&kprobe_lock);
-       kprobe_cpu = smp_processor_id();
+       __get_cpu_var(kprobe_instance) = kp;
 }
 
-void unlock_kprobes(void)
+static inline void reset_kprobe_instance(void)
 {
-       kprobe_cpu = NR_CPUS;
-       spin_unlock(&kprobe_lock);
+       __get_cpu_var(kprobe_instance) = NULL;
 }
 
-/* You have to be holding the kprobe_lock */
-struct kprobe *get_kprobe(void *addr)
+/*
+ * This routine is called either:
+ *     - under the kprobe_mutex - during kprobe_[un]register()
+ *                             OR
+ *     - with preemption disabled - from arch/xxx/kernel/kprobes.c
+ */
+struct kprobe __kprobes *get_kprobe(void *addr)
 {
        struct hlist_head *head;
        struct hlist_node *node;
+       struct kprobe *p;
 
        head = &kprobe_table[hash_ptr(addr, KPROBE_HASH_BITS)];
-       hlist_for_each(node, head) {
-               struct kprobe *p = hlist_entry(node, struct kprobe, hlist);
+       hlist_for_each_entry_rcu(p, node, head, hlist) {
                if (p->addr == addr)
                        return p;
        }
+
        return NULL;
 }
 
+static int __kprobes aggr_pre_handler(struct kprobe *p, struct pt_regs *regs);
+
+/* Return true if the kprobe is an aggregator */
+static inline int kprobe_aggrprobe(struct kprobe *p)
+{
+       return p->pre_handler == aggr_pre_handler;
+}
+
 /*
- * Aggregate handlers for multiple kprobes support - these handlers
- * take care of invoking the individual kprobe handlers on p->list
+ * Keep all fields in the kprobe consistent
+ */
+static inline void copy_kprobe(struct kprobe *old_p, struct kprobe *p)
+{
+       memcpy(&p->opcode, &old_p->opcode, sizeof(kprobe_opcode_t));
+       memcpy(&p->ainsn, &old_p->ainsn, sizeof(struct arch_specific_insn));
+}
+
+#ifdef CONFIG_OPTPROBES
+/* NOTE: change this value only with kprobe_mutex held */
+static bool kprobes_allow_optimization;
+
+/*
+ * Call all pre_handler on the list, but ignores its return value.
+ * This must be called from arch-dep optimized caller.
  */
-static int aggr_pre_handler(struct kprobe *p, struct pt_regs *regs)
+void __kprobes opt_pre_handler(struct kprobe *p, struct pt_regs *regs)
 {
        struct kprobe *kp;
 
-       list_for_each_entry(kp, &p->list, list) {
-               if (kp->pre_handler) {
-                       curr_kprobe = kp;
-                       if (kp->pre_handler(kp, regs))
-                               return 1;
+       list_for_each_entry_rcu(kp, &p->list, list) {
+               if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
+                       set_kprobe_instance(kp);
+                       kp->pre_handler(kp, regs);
                }
-               curr_kprobe = NULL;
+               reset_kprobe_instance();
        }
+}
+
+/* Return true(!0) if the kprobe is ready for optimization. */
+static inline int kprobe_optready(struct kprobe *p)
+{
+       struct optimized_kprobe *op;
+
+       if (kprobe_aggrprobe(p)) {
+               op = container_of(p, struct optimized_kprobe, kp);
+               return arch_prepared_optinsn(&op->optinsn);
+       }
+
        return 0;
 }
 
-static void aggr_post_handler(struct kprobe *p, struct pt_regs *regs,
-                             unsigned long flags)
+/*
+ * Return an optimized kprobe whose optimizing code replaces
+ * instructions including addr (exclude breakpoint).
+ */
+struct kprobe *__kprobes get_optimized_kprobe(unsigned long addr)
 {
-       struct kprobe *kp;
+       int i;
+       struct kprobe *p = NULL;
+       struct optimized_kprobe *op;
 
-       list_for_each_entry(kp, &p->list, list) {
-               if (kp->post_handler) {
-                       curr_kprobe = kp;
-                       kp->post_handler(kp, regs, flags);
-                       curr_kprobe = NULL;
-               }
+       /* Don't check i == 0, since that is a breakpoint case. */
+       for (i = 1; !p && i < MAX_OPTIMIZED_LENGTH; i++)
+               p = get_kprobe((void *)(addr - i));
+
+       if (p && kprobe_optready(p)) {
+               op = container_of(p, struct optimized_kprobe, kp);
+               if (arch_within_optimized_kprobe(op, addr))
+                       return p;
        }
-       return;
+
+       return NULL;
 }
 
-static int aggr_fault_handler(struct kprobe *p, struct pt_regs *regs,
-                             int trapnr)
+/* Optimization staging list, protected by kprobe_mutex */
+static LIST_HEAD(optimizing_list);
+
+static void kprobe_optimizer(struct work_struct *work);
+static DECLARE_DELAYED_WORK(optimizing_work, kprobe_optimizer);
+#define OPTIMIZE_DELAY 5
+
+/* Kprobe jump optimizer */
+static __kprobes void kprobe_optimizer(struct work_struct *work)
 {
+       struct optimized_kprobe *op, *tmp;
+
+       /* Lock modules while optimizing kprobes */
+       mutex_lock(&module_mutex);
+       mutex_lock(&kprobe_mutex);
+       if (kprobes_all_disarmed || !kprobes_allow_optimization)
+               goto end;
+
        /*
-        * if we faulted "during" the execution of a user specified
-        * probe handler, invoke just that probe's fault handler
+        * Wait for quiesence period to ensure all running interrupts
+        * are done. Because optprobe may modify multiple instructions
+        * there is a chance that Nth instruction is interrupted. In that
+        * case, running interrupt can return to 2nd-Nth byte of jump
+        * instruction. This wait is for avoiding it.
         */
-       if (curr_kprobe && curr_kprobe->fault_handler) {
-               if (curr_kprobe->fault_handler(curr_kprobe, regs, trapnr))
-                       return 1;
+       synchronize_sched();
+
+       /*
+        * The optimization/unoptimization refers online_cpus via
+        * stop_machine() and cpu-hotplug modifies online_cpus.
+        * And same time, text_mutex will be held in cpu-hotplug and here.
+        * This combination can cause a deadlock (cpu-hotplug try to lock
+        * text_mutex but stop_machine can not be done because online_cpus
+        * has been changed)
+        * To avoid this deadlock, we need to call get_online_cpus()
+        * for preventing cpu-hotplug outside of text_mutex locking.
+        */
+       get_online_cpus();
+       mutex_lock(&text_mutex);
+       list_for_each_entry_safe(op, tmp, &optimizing_list, list) {
+               WARN_ON(kprobe_disabled(&op->kp));
+               if (arch_optimize_kprobe(op) < 0)
+                       op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
+               list_del_init(&op->list);
        }
-       return 0;
+       mutex_unlock(&text_mutex);
+       put_online_cpus();
+end:
+       mutex_unlock(&kprobe_mutex);
+       mutex_unlock(&module_mutex);
 }
 
-static int aggr_break_handler(struct kprobe *p, struct pt_regs *regs)
+/* Optimize kprobe if p is ready to be optimized */
+static __kprobes void optimize_kprobe(struct kprobe *p)
 {
-       struct kprobe *kp = curr_kprobe;
-       if (curr_kprobe && kp->break_handler) {
-               if (kp->break_handler(kp, regs)) {
-                       curr_kprobe = NULL;
-                       return 1;
-               }
+       struct optimized_kprobe *op;
+
+       /* Check if the kprobe is disabled or not ready for optimization. */
+       if (!kprobe_optready(p) || !kprobes_allow_optimization ||
+           (kprobe_disabled(p) || kprobes_all_disarmed))
+               return;
+
+       /* Both of break_handler and post_handler are not supported. */
+       if (p->break_handler || p->post_handler)
+               return;
+
+       op = container_of(p, struct optimized_kprobe, kp);
+
+       /* Check there is no other kprobes at the optimized instructions */
+       if (arch_check_optimized_kprobe(op) < 0)
+               return;
+
+       /* Check if it is already optimized. */
+       if (op->kp.flags & KPROBE_FLAG_OPTIMIZED)
+               return;
+
+       op->kp.flags |= KPROBE_FLAG_OPTIMIZED;
+       list_add(&op->list, &optimizing_list);
+       if (!delayed_work_pending(&optimizing_work))
+               schedule_delayed_work(&optimizing_work, OPTIMIZE_DELAY);
+}
+
+/* Unoptimize a kprobe if p is optimized */
+static __kprobes void unoptimize_kprobe(struct kprobe *p)
+{
+       struct optimized_kprobe *op;
+
+       if ((p->flags & KPROBE_FLAG_OPTIMIZED) && kprobe_aggrprobe(p)) {
+               op = container_of(p, struct optimized_kprobe, kp);
+               if (!list_empty(&op->list))
+                       /* Dequeue from the optimization queue */
+                       list_del_init(&op->list);
+               else
+                       /* Replace jump with break */
+                       arch_unoptimize_kprobe(op);
+               op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
        }
-       curr_kprobe = NULL;
-       return 0;
 }
 
-struct kprobe trampoline_p = {
-               .addr = (kprobe_opcode_t *) &kretprobe_trampoline,
-               .pre_handler = trampoline_probe_handler,
-               .post_handler = trampoline_post_handler
-};
+/* Remove optimized instructions */
+static void __kprobes kill_optimized_kprobe(struct kprobe *p)
+{
+       struct optimized_kprobe *op;
 
-struct kretprobe_instance *get_free_rp_inst(struct kretprobe *rp)
+       op = container_of(p, struct optimized_kprobe, kp);
+       if (!list_empty(&op->list)) {
+               /* Dequeue from the optimization queue */
+               list_del_init(&op->list);
+               op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
+       }
+       /* Don't unoptimize, because the target code will be freed. */
+       arch_remove_optimized_kprobe(op);
+}
+
+/* Try to prepare optimized instructions */
+static __kprobes void prepare_optimized_kprobe(struct kprobe *p)
 {
-       struct hlist_node *node;
-       struct kretprobe_instance *ri;
-       hlist_for_each_entry(ri, node, &rp->free_instances, uflist)
-               return ri;
-       return NULL;
+       struct optimized_kprobe *op;
+
+       op = container_of(p, struct optimized_kprobe, kp);
+       arch_prepare_optimized_kprobe(op);
+}
+
+/* Free optimized instructions and optimized_kprobe */
+static __kprobes void free_aggr_kprobe(struct kprobe *p)
+{
+       struct optimized_kprobe *op;
+
+       op = container_of(p, struct optimized_kprobe, kp);
+       arch_remove_optimized_kprobe(op);
+       kfree(op);
+}
+
+/* Allocate new optimized_kprobe and try to prepare optimized instructions */
+static __kprobes struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
+{
+       struct optimized_kprobe *op;
+
+       op = kzalloc(sizeof(struct optimized_kprobe), GFP_KERNEL);
+       if (!op)
+               return NULL;
+
+       INIT_LIST_HEAD(&op->list);
+       op->kp.addr = p->addr;
+       arch_prepare_optimized_kprobe(op);
+
+       return &op->kp;
+}
+
+static void __kprobes init_aggr_kprobe(struct kprobe *ap, struct kprobe *p);
+
+/*
+ * Prepare an optimized_kprobe and optimize it
+ * NOTE: p must be a normal registered kprobe
+ */
+static __kprobes void try_to_optimize_kprobe(struct kprobe *p)
+{
+       struct kprobe *ap;
+       struct optimized_kprobe *op;
+
+       ap = alloc_aggr_kprobe(p);
+       if (!ap)
+               return;
+
+       op = container_of(ap, struct optimized_kprobe, kp);
+       if (!arch_prepared_optinsn(&op->optinsn)) {
+               /* If failed to setup optimizing, fallback to kprobe */
+               free_aggr_kprobe(ap);
+               return;
+       }
+
+       init_aggr_kprobe(ap, p);
+       optimize_kprobe(ap);
 }
 
-static struct kretprobe_instance *get_used_rp_inst(struct kretprobe *rp)
+#ifdef CONFIG_SYSCTL
+static void __kprobes optimize_all_kprobes(void)
 {
+       struct hlist_head *head;
        struct hlist_node *node;
-       struct kretprobe_instance *ri;
-       hlist_for_each_entry(ri, node, &rp->used_instances, uflist)
-               return ri;
-       return NULL;
+       struct kprobe *p;
+       unsigned int i;
+
+       /* If optimization is already allowed, just return */
+       if (kprobes_allow_optimization)
+               return;
+
+       kprobes_allow_optimization = true;
+       mutex_lock(&text_mutex);
+       for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
+               head = &kprobe_table[i];
+               hlist_for_each_entry_rcu(p, node, head, hlist)
+                       if (!kprobe_disabled(p))
+                               optimize_kprobe(p);
+       }
+       mutex_unlock(&text_mutex);
+       printk(KERN_INFO "Kprobes globally optimized\n");
 }
 
-struct kretprobe_instance *get_rp_inst(void *sara)
+static void __kprobes unoptimize_all_kprobes(void)
 {
        struct hlist_head *head;
        struct hlist_node *node;
-       struct task_struct *tsk;
-       struct kretprobe_instance *ri;
+       struct kprobe *p;
+       unsigned int i;
+
+       /* If optimization is already prohibited, just return */
+       if (!kprobes_allow_optimization)
+               return;
 
-       tsk = arch_get_kprobe_task(sara);
-       head = &kretprobe_inst_table[hash_ptr(tsk, KPROBE_HASH_BITS)];
-       hlist_for_each_entry(ri, node, head, hlist) {
-               if (ri->stack_addr == sara)
-                       return ri;
+       kprobes_allow_optimization = false;
+       printk(KERN_INFO "Kprobes globally unoptimized\n");
+       get_online_cpus();      /* For avoiding text_mutex deadlock */
+       mutex_lock(&text_mutex);
+       for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
+               head = &kprobe_table[i];
+               hlist_for_each_entry_rcu(p, node, head, hlist) {
+                       if (!kprobe_disabled(p))
+                               unoptimize_kprobe(p);
+               }
        }
-       return NULL;
+
+       mutex_unlock(&text_mutex);
+       put_online_cpus();
+       /* Allow all currently running kprobes to complete */
+       synchronize_sched();
+}
+
+int sysctl_kprobes_optimization;
+int proc_kprobes_optimization_handler(struct ctl_table *table, int write,
+                                     void __user *buffer, size_t *length,
+                                     loff_t *ppos)
+{
+       int ret;
+
+       mutex_lock(&kprobe_mutex);
+       sysctl_kprobes_optimization = kprobes_allow_optimization ? 1 : 0;
+       ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
+
+       if (sysctl_kprobes_optimization)
+               optimize_all_kprobes();
+       else
+               unoptimize_all_kprobes();
+       mutex_unlock(&kprobe_mutex);
+
+       return ret;
+}
+#endif /* CONFIG_SYSCTL */
+
+static void __kprobes __arm_kprobe(struct kprobe *p)
+{
+       struct kprobe *old_p;
+
+       /* Check collision with other optimized kprobes */
+       old_p = get_optimized_kprobe((unsigned long)p->addr);
+       if (unlikely(old_p))
+               unoptimize_kprobe(old_p); /* Fallback to unoptimized kprobe */
+
+       arch_arm_kprobe(p);
+       optimize_kprobe(p);     /* Try to optimize (add kprobe to a list) */
+}
+
+static void __kprobes __disarm_kprobe(struct kprobe *p)
+{
+       struct kprobe *old_p;
+
+       unoptimize_kprobe(p);   /* Try to unoptimize */
+       arch_disarm_kprobe(p);
+
+       /* If another kprobe was blocked, optimize it. */
+       old_p = get_optimized_kprobe((unsigned long)p->addr);
+       if (unlikely(old_p))
+               optimize_kprobe(old_p);
+}
+
+#else /* !CONFIG_OPTPROBES */
+
+#define optimize_kprobe(p)                     do {} while (0)
+#define unoptimize_kprobe(p)                   do {} while (0)
+#define kill_optimized_kprobe(p)               do {} while (0)
+#define prepare_optimized_kprobe(p)            do {} while (0)
+#define try_to_optimize_kprobe(p)              do {} while (0)
+#define __arm_kprobe(p)                                arch_arm_kprobe(p)
+#define __disarm_kprobe(p)                     arch_disarm_kprobe(p)
+
+static __kprobes void free_aggr_kprobe(struct kprobe *p)
+{
+       kfree(p);
 }
 
-void add_rp_inst(struct kretprobe_instance *ri)
+static __kprobes struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
+{
+       return kzalloc(sizeof(struct kprobe), GFP_KERNEL);
+}
+#endif /* CONFIG_OPTPROBES */
+
+/* Arm a kprobe with text_mutex */
+static void __kprobes arm_kprobe(struct kprobe *kp)
 {
-       struct task_struct *tsk;
        /*
-        * Remove rp inst off the free list -
-        * Add it back when probed function returns
+        * Here, since __arm_kprobe() doesn't use stop_machine(),
+        * this doesn't cause deadlock on text_mutex. So, we don't
+        * need get_online_cpus().
         */
-       hlist_del(&ri->uflist);
-       tsk = arch_get_kprobe_task(ri->stack_addr);
-       /* Add rp inst onto table */
-       INIT_HLIST_NODE(&ri->hlist);
-       hlist_add_head(&ri->hlist,
-                       &kretprobe_inst_table[hash_ptr(tsk, KPROBE_HASH_BITS)]);
+       mutex_lock(&text_mutex);
+       __arm_kprobe(kp);
+       mutex_unlock(&text_mutex);
+}
+
+/* Disarm a kprobe with text_mutex */
+static void __kprobes disarm_kprobe(struct kprobe *kp)
+{
+       get_online_cpus();      /* For avoiding text_mutex deadlock */
+       mutex_lock(&text_mutex);
+       __disarm_kprobe(kp);
+       mutex_unlock(&text_mutex);
+       put_online_cpus();
+}
+
+/*
+ * Aggregate handlers for multiple kprobes support - these handlers
+ * take care of invoking the individual kprobe handlers on p->list
+ */
+static int __kprobes aggr_pre_handler(struct kprobe *p, struct pt_regs *regs)
+{
+       struct kprobe *kp;
+
+       list_for_each_entry_rcu(kp, &p->list, list) {
+               if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
+                       set_kprobe_instance(kp);
+                       if (kp->pre_handler(kp, regs))
+                               return 1;
+               }
+               reset_kprobe_instance();
+       }
+       return 0;
+}
+
+static void __kprobes aggr_post_handler(struct kprobe *p, struct pt_regs *regs,
+                                       unsigned long flags)
+{
+       struct kprobe *kp;
+
+       list_for_each_entry_rcu(kp, &p->list, list) {
+               if (kp->post_handler && likely(!kprobe_disabled(kp))) {
+                       set_kprobe_instance(kp);
+                       kp->post_handler(kp, regs, flags);
+                       reset_kprobe_instance();
+               }
+       }
+}
+
+static int __kprobes aggr_fault_handler(struct kprobe *p, struct pt_regs *regs,
+                                       int trapnr)
+{
+       struct kprobe *cur = __get_cpu_var(kprobe_instance);
 
-       /* Also add this rp inst to the used list. */
-       INIT_HLIST_NODE(&ri->uflist);
-       hlist_add_head(&ri->uflist, &ri->rp->used_instances);
+       /*
+        * if we faulted "during" the execution of a user specified
+        * probe handler, invoke just that probe's fault handler
+        */
+       if (cur && cur->fault_handler) {
+               if (cur->fault_handler(cur, regs, trapnr))
+                       return 1;
+       }
+       return 0;
+}
+
+static int __kprobes aggr_break_handler(struct kprobe *p, struct pt_regs *regs)
+{
+       struct kprobe *cur = __get_cpu_var(kprobe_instance);
+       int ret = 0;
+
+       if (cur && cur->break_handler) {
+               if (cur->break_handler(cur, regs))
+                       ret = 1;
+       }
+       reset_kprobe_instance();
+       return ret;
+}
+
+/* Walks the list and increments nmissed count for multiprobe case */
+void __kprobes kprobes_inc_nmissed_count(struct kprobe *p)
+{
+       struct kprobe *kp;
+       if (!kprobe_aggrprobe(p)) {
+               p->nmissed++;
+       } else {
+               list_for_each_entry_rcu(kp, &p->list, list)
+                       kp->nmissed++;
+       }
+       return;
 }
 
-void recycle_rp_inst(struct kretprobe_instance *ri)
+void __kprobes recycle_rp_inst(struct kretprobe_instance *ri,
+                               struct hlist_head *head)
 {
+       struct kretprobe *rp = ri->rp;
+
        /* remove rp inst off the rprobe_inst_table */
        hlist_del(&ri->hlist);
-       if (ri->rp) {
-               /* remove rp inst off the used list */
-               hlist_del(&ri->uflist);
-               /* put rp inst back onto the free list */
-               INIT_HLIST_NODE(&ri->uflist);
-               hlist_add_head(&ri->uflist, &ri->rp->free_instances);
+       INIT_HLIST_NODE(&ri->hlist);
+       if (likely(rp)) {
+               spin_lock(&rp->lock);
+               hlist_add_head(&ri->hlist, &rp->free_instances);
+               spin_unlock(&rp->lock);
        } else
                /* Unregistering */
-               kfree(ri);
+               hlist_add_head(&ri->hlist, head);
 }
 
-struct hlist_head * kretprobe_inst_table_head(struct task_struct *tsk)
+void __kprobes kretprobe_hash_lock(struct task_struct *tsk,
+                        struct hlist_head **head, unsigned long *flags)
 {
-       return &kretprobe_inst_table[hash_ptr(tsk, KPROBE_HASH_BITS)];
+       unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
+       spinlock_t *hlist_lock;
+
+       *head = &kretprobe_inst_table[hash];
+       hlist_lock = kretprobe_table_lock_ptr(hash);
+       spin_lock_irqsave(hlist_lock, *flags);
 }
 
-struct kretprobe_instance *get_rp_inst_tsk(struct task_struct *tk)
+static void __kprobes kretprobe_table_lock(unsigned long hash,
+       unsigned long *flags)
 {
-       struct task_struct *tsk;
-       struct hlist_head *head;
-       struct hlist_node *node;
-       struct kretprobe_instance *ri;
+       spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
+       spin_lock_irqsave(hlist_lock, *flags);
+}
 
-       head = &kretprobe_inst_table[hash_ptr(tk, KPROBE_HASH_BITS)];
+void __kprobes kretprobe_hash_unlock(struct task_struct *tsk,
+       unsigned long *flags)
+{
+       unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
+       spinlock_t *hlist_lock;
 
-       hlist_for_each_entry(ri, node, head, hlist) {
-               tsk = arch_get_kprobe_task(ri->stack_addr);
-               if (tsk == tk)
-                       return ri;
-       }
-       return NULL;
+       hlist_lock = kretprobe_table_lock_ptr(hash);
+       spin_unlock_irqrestore(hlist_lock, *flags);
 }
 
-/*
- * This function is called from do_exit or do_execv when task tk's stack is
- * about to be recycled. Recycle any function-return probe instances
- * associated with this task. These represent probed functions that have
- * been called but may never return.
- */
-void kprobe_flush_task(struct task_struct *tk)
+void __kprobes kretprobe_table_unlock(unsigned long hash, unsigned long *flags)
 {
-       unsigned long flags = 0;
-       spin_lock_irqsave(&kprobe_lock, flags);
-       arch_kprobe_flush_task(tk);
-       spin_unlock_irqrestore(&kprobe_lock, flags);
+       spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
+       spin_unlock_irqrestore(hlist_lock, *flags);
 }
 
 /*
- * This kprobe pre_handler is registered with every kretprobe. When probe
- * hits it will set up the return probe.
+ * This function is called from finish_task_switch when task tk becomes dead,
+ * so that we can recycle any function-return probe instances associated
+ * with this task. These left over instances represent probed functions
+ * that have been called but will never return.
  */
-static int pre_handler_kretprobe(struct kprobe *p, struct pt_regs *regs)
+void __kprobes kprobe_flush_task(struct task_struct *tk)
 {
-       struct kretprobe *rp = container_of(p, struct kretprobe, kp);
-
-       /*TODO: consider to only swap the RA after the last pre_handler fired */
-       arch_prepare_kretprobe(rp, regs);
-       return 0;
+       struct kretprobe_instance *ri;
+       struct hlist_head *head, empty_rp;
+       struct hlist_node *node, *tmp;
+       unsigned long hash, flags = 0;
+
+       if (unlikely(!kprobes_initialized))
+               /* Early boot.  kretprobe_table_locks not yet initialized. */
+               return;
+
+       hash = hash_ptr(tk, KPROBE_HASH_BITS);
+       head = &kretprobe_inst_table[hash];
+       kretprobe_table_lock(hash, &flags);
+       hlist_for_each_entry_safe(ri, node, tmp, head, hlist) {
+               if (ri->task == tk)
+                       recycle_rp_inst(ri, &empty_rp);
+       }
+       kretprobe_table_unlock(hash, &flags);
+       INIT_HLIST_HEAD(&empty_rp);
+       hlist_for_each_entry_safe(ri, node, tmp, &empty_rp, hlist) {
+               hlist_del(&ri->hlist);
+               kfree(ri);
+       }
 }
 
 static inline void free_rp_inst(struct kretprobe *rp)
 {
        struct kretprobe_instance *ri;
-       while ((ri = get_free_rp_inst(rp)) != NULL) {
-               hlist_del(&ri->uflist);
+       struct hlist_node *pos, *next;
+
+       hlist_for_each_entry_safe(ri, pos, next, &rp->free_instances, hlist) {
+               hlist_del(&ri->hlist);
                kfree(ri);
        }
 }
 
-/*
- * Keep all fields in the kprobe consistent
- */
-static inline void copy_kprobe(struct kprobe *old_p, struct kprobe *p)
+static void __kprobes cleanup_rp_inst(struct kretprobe *rp)
 {
-       memcpy(&p->opcode, &old_p->opcode, sizeof(kprobe_opcode_t));
-       memcpy(&p->ainsn, &old_p->ainsn, sizeof(struct arch_specific_insn));
+       unsigned long flags, hash;
+       struct kretprobe_instance *ri;
+       struct hlist_node *pos, *next;
+       struct hlist_head *head;
+
+       /* No race here */
+       for (hash = 0; hash < KPROBE_TABLE_SIZE; hash++) {
+               kretprobe_table_lock(hash, &flags);
+               head = &kretprobe_inst_table[hash];
+               hlist_for_each_entry_safe(ri, pos, next, head, hlist) {
+                       if (ri->rp == rp)
+                               ri->rp = NULL;
+               }
+               kretprobe_table_unlock(hash, &flags);
+       }
+       free_rp_inst(rp);
 }
 
 /*
-* Add the new probe to old_p->list. Fail if this is the
+* Add the new probe to ap->list. Fail if this is the
 * second jprobe at the address - two jprobes can't coexist
 */
-static int add_new_kprobe(struct kprobe *old_p, struct kprobe *p)
+static int __kprobes add_new_kprobe(struct kprobe *ap, struct kprobe *p)
 {
-        struct kprobe *kp;
+       BUG_ON(kprobe_gone(ap) || kprobe_gone(p));
+
+       if (p->break_handler || p->post_handler)
+               unoptimize_kprobe(ap);  /* Fall back to normal kprobe */
 
        if (p->break_handler) {
-               list_for_each_entry(kp, &old_p->list, list) {
-                       if (kp->break_handler)
-                               return -EEXIST;
-               }
-               list_add_tail(&p->list, &old_p->list);
+               if (ap->break_handler)
+                       return -EEXIST;
+               list_add_tail_rcu(&p->list, &ap->list);
+               ap->break_handler = aggr_break_handler;
        } else
-               list_add(&p->list, &old_p->list);
+               list_add_rcu(&p->list, &ap->list);
+       if (p->post_handler && !ap->post_handler)
+               ap->post_handler = aggr_post_handler;
+
+       if (kprobe_disabled(ap) && !kprobe_disabled(p)) {
+               ap->flags &= ~KPROBE_FLAG_DISABLED;
+               if (!kprobes_all_disarmed)
+                       /* Arm the breakpoint again. */
+                       __arm_kprobe(ap);
+       }
        return 0;
 }
 
@@ -404,221 +960,1070 @@ static int add_new_kprobe(struct kprobe *old_p, struct kprobe *p)
  * Fill in the required fields of the "manager kprobe". Replace the
  * earlier kprobe in the hlist with the manager kprobe
  */
-static inline void add_aggr_kprobe(struct kprobe *ap, struct kprobe *p)
+static void __kprobes init_aggr_kprobe(struct kprobe *ap, struct kprobe *p)
 {
+       /* Copy p's insn slot to ap */
        copy_kprobe(p, ap);
+       flush_insn_slot(ap);
        ap->addr = p->addr;
+       ap->flags = p->flags & ~KPROBE_FLAG_OPTIMIZED;
        ap->pre_handler = aggr_pre_handler;
-       ap->post_handler = aggr_post_handler;
        ap->fault_handler = aggr_fault_handler;
-       ap->break_handler = aggr_break_handler;
+       /* We don't care the kprobe which has gone. */
+       if (p->post_handler && !kprobe_gone(p))
+               ap->post_handler = aggr_post_handler;
+       if (p->break_handler && !kprobe_gone(p))
+               ap->break_handler = aggr_break_handler;
 
        INIT_LIST_HEAD(&ap->list);
-       list_add(&p->list, &ap->list);
-
        INIT_HLIST_NODE(&ap->hlist);
-       hlist_del(&p->hlist);
-       hlist_add_head(&ap->hlist,
-               &kprobe_table[hash_ptr(ap->addr, KPROBE_HASH_BITS)]);
+
+       list_add_rcu(&p->list, &ap->list);
+       hlist_replace_rcu(&p->hlist, &ap->hlist);
 }
 
 /*
  * This is the second or subsequent kprobe at the address - handle
  * the intricacies
- * TODO: Move kcalloc outside the spinlock
  */
-static int register_aggr_kprobe(struct kprobe *old_p, struct kprobe *p)
+static int __kprobes register_aggr_kprobe(struct kprobe *old_p,
+                                         struct kprobe *p)
 {
        int ret = 0;
-       struct kprobe *ap;
+       struct kprobe *ap = old_p;
 
-       if (old_p->pre_handler == aggr_pre_handler) {
-               copy_kprobe(old_p, p);
-               ret = add_new_kprobe(old_p, p);
-       } else {
-               ap = kcalloc(1, sizeof(struct kprobe), GFP_ATOMIC);
+       if (!kprobe_aggrprobe(old_p)) {
+               /* If old_p is not an aggr_kprobe, create new aggr_kprobe. */
+               ap = alloc_aggr_kprobe(old_p);
                if (!ap)
                        return -ENOMEM;
-               add_aggr_kprobe(ap, old_p);
-               copy_kprobe(ap, p);
-               ret = add_new_kprobe(ap, p);
+               init_aggr_kprobe(ap, old_p);
        }
-       return ret;
+
+       if (kprobe_gone(ap)) {
+               /*
+                * Attempting to insert new probe at the same location that
+                * had a probe in the module vaddr area which already
+                * freed. So, the instruction slot has already been
+                * released. We need a new slot for the new probe.
+                */
+               ret = arch_prepare_kprobe(ap);
+               if (ret)
+                       /*
+                        * Even if fail to allocate new slot, don't need to
+                        * free aggr_probe. It will be used next time, or
+                        * freed by unregister_kprobe.
+                        */
+                       return ret;
+
+               /* Prepare optimized instructions if possible. */
+               prepare_optimized_kprobe(ap);
+
+               /*
+                * Clear gone flag to prevent allocating new slot again, and
+                * set disabled flag because it is not armed yet.
+                */
+               ap->flags = (ap->flags & ~KPROBE_FLAG_GONE)
+                           | KPROBE_FLAG_DISABLED;
+       }
+
+       /* Copy ap's insn slot to p */
+       copy_kprobe(ap, p);
+       return add_new_kprobe(ap, p);
 }
 
-/* kprobe removal house-keeping routines */
-static inline void cleanup_kprobe(struct kprobe *p, unsigned long flags)
+/* Try to disable aggr_kprobe, and return 1 if succeeded.*/
+static int __kprobes try_to_disable_aggr_kprobe(struct kprobe *p)
 {
-       arch_disarm_kprobe(p);
-       hlist_del(&p->hlist);
-       spin_unlock_irqrestore(&kprobe_lock, flags);
-       arch_remove_kprobe(p);
+       struct kprobe *kp;
+
+       list_for_each_entry_rcu(kp, &p->list, list) {
+               if (!kprobe_disabled(kp))
+                       /*
+                        * There is an active probe on the list.
+                        * We can't disable aggr_kprobe.
+                        */
+                       return 0;
+       }
+       p->flags |= KPROBE_FLAG_DISABLED;
+       return 1;
 }
 
-static inline void cleanup_aggr_kprobe(struct kprobe *old_p,
-               struct kprobe *p, unsigned long flags)
+static int __kprobes in_kprobes_functions(unsigned long addr)
 {
-       list_del(&p->list);
-       if (list_empty(&old_p->list)) {
-               cleanup_kprobe(old_p, flags);
-               kfree(old_p);
-       } else
-               spin_unlock_irqrestore(&kprobe_lock, flags);
+       struct kprobe_blackpoint *kb;
+
+       if (addr >= (unsigned long)__kprobes_text_start &&
+           addr < (unsigned long)__kprobes_text_end)
+               return -EINVAL;
+       /*
+        * If there exists a kprobe_blacklist, verify and
+        * fail any probe registration in the prohibited area
+        */
+       for (kb = kprobe_blacklist; kb->name != NULL; kb++) {
+               if (kb->start_addr) {
+                       if (addr >= kb->start_addr &&
+                           addr < (kb->start_addr + kb->range))
+                               return -EINVAL;
+               }
+       }
+       return 0;
+}
+
+/*
+ * If we have a symbol_name argument, look it up and add the offset field
+ * to it. This way, we can specify a relative address to a symbol.
+ */
+static kprobe_opcode_t __kprobes *kprobe_addr(struct kprobe *p)
+{
+       kprobe_opcode_t *addr = p->addr;
+       if (p->symbol_name) {
+               if (addr)
+                       return NULL;
+               kprobe_lookup_name(p->symbol_name, addr);
+       }
+
+       if (!addr)
+               return NULL;
+       return (kprobe_opcode_t *)(((char *)addr) + p->offset);
+}
+
+/* Check passed kprobe is valid and return kprobe in kprobe_table. */
+static struct kprobe * __kprobes __get_valid_kprobe(struct kprobe *p)
+{
+       struct kprobe *old_p, *list_p;
+
+       old_p = get_kprobe(p->addr);
+       if (unlikely(!old_p))
+               return NULL;
+
+       if (p != old_p) {
+               list_for_each_entry_rcu(list_p, &old_p->list, list)
+                       if (list_p == p)
+                       /* kprobe p is a valid probe */
+                               goto valid;
+               return NULL;
+       }
+valid:
+       return old_p;
 }
 
-int register_kprobe(struct kprobe *p)
+/* Return error if the kprobe is being re-registered */
+static inline int check_kprobe_rereg(struct kprobe *p)
 {
        int ret = 0;
-       unsigned long flags = 0;
        struct kprobe *old_p;
 
-       if ((ret = arch_prepare_kprobe(p)) != 0) {
-               goto rm_kprobe;
+       mutex_lock(&kprobe_mutex);
+       old_p = __get_valid_kprobe(p);
+       if (old_p)
+               ret = -EINVAL;
+       mutex_unlock(&kprobe_mutex);
+       return ret;
+}
+
+int __kprobes register_kprobe(struct kprobe *p)
+{
+       int ret = 0;
+       struct kprobe *old_p;
+       struct module *probed_mod;
+       kprobe_opcode_t *addr;
+
+       addr = kprobe_addr(p);
+       if (!addr)
+               return -EINVAL;
+       p->addr = addr;
+
+       ret = check_kprobe_rereg(p);
+       if (ret)
+               return ret;
+
+       preempt_disable();
+       if (!kernel_text_address((unsigned long) p->addr) ||
+           in_kprobes_functions((unsigned long) p->addr) ||
+           ftrace_text_reserved(p->addr, p->addr)) {
+               preempt_enable();
+               return -EINVAL;
        }
-       spin_lock_irqsave(&kprobe_lock, flags);
-       old_p = get_kprobe(p->addr);
+
+       /* User can pass only KPROBE_FLAG_DISABLED to register_kprobe */
+       p->flags &= KPROBE_FLAG_DISABLED;
+
+       /*
+        * Check if are we probing a module.
+        */
+       probed_mod = __module_text_address((unsigned long) p->addr);
+       if (probed_mod) {
+               /*
+                * We must hold a refcount of the probed module while updating
+                * its code to prohibit unexpected unloading.
+                */
+               if (unlikely(!try_module_get(probed_mod))) {
+                       preempt_enable();
+                       return -EINVAL;
+               }
+               /*
+                * If the module freed .init.text, we couldn't insert
+                * kprobes in there.
+                */
+               if (within_module_init((unsigned long)p->addr, probed_mod) &&
+                   probed_mod->state != MODULE_STATE_COMING) {
+                       module_put(probed_mod);
+                       preempt_enable();
+                       return -EINVAL;
+               }
+       }
+       preempt_enable();
+
        p->nmissed = 0;
+       INIT_LIST_HEAD(&p->list);
+       mutex_lock(&kprobe_mutex);
+
+       get_online_cpus();      /* For avoiding text_mutex deadlock. */
+       mutex_lock(&text_mutex);
+
+       old_p = get_kprobe(p->addr);
        if (old_p) {
+               /* Since this may unoptimize old_p, locking text_mutex. */
                ret = register_aggr_kprobe(old_p, p);
                goto out;
        }
 
-       arch_copy_kprobe(p);
+       ret = arch_prepare_kprobe(p);
+       if (ret)
+               goto out;
+
        INIT_HLIST_NODE(&p->hlist);
-       hlist_add_head(&p->hlist,
+       hlist_add_head_rcu(&p->hlist,
                       &kprobe_table[hash_ptr(p->addr, KPROBE_HASH_BITS)]);
 
-       arch_arm_kprobe(p);
+       if (!kprobes_all_disarmed && !kprobe_disabled(p))
+               __arm_kprobe(p);
+
+       /* Try to optimize kprobe */
+       try_to_optimize_kprobe(p);
 
 out:
-       spin_unlock_irqrestore(&kprobe_lock, flags);
-rm_kprobe:
-       if (ret == -EEXIST)
-               arch_remove_kprobe(p);
+       mutex_unlock(&text_mutex);
+       put_online_cpus();
+       mutex_unlock(&kprobe_mutex);
+
+       if (probed_mod)
+               module_put(probed_mod);
+
        return ret;
 }
+EXPORT_SYMBOL_GPL(register_kprobe);
+
+/*
+ * Unregister a kprobe without a scheduler synchronization.
+ */
+static int __kprobes __unregister_kprobe_top(struct kprobe *p)
+{
+       struct kprobe *old_p, *list_p;
+
+       old_p = __get_valid_kprobe(p);
+       if (old_p == NULL)
+               return -EINVAL;
+
+       if (old_p == p ||
+           (kprobe_aggrprobe(old_p) &&
+            list_is_singular(&old_p->list))) {
+               /*
+                * Only probe on the hash list. Disarm only if kprobes are
+                * enabled and not gone - otherwise, the breakpoint would
+                * already have been removed. We save on flushing icache.
+                */
+               if (!kprobes_all_disarmed && !kprobe_disabled(old_p))
+                       disarm_kprobe(old_p);
+               hlist_del_rcu(&old_p->hlist);
+       } else {
+               if (p->break_handler && !kprobe_gone(p))
+                       old_p->break_handler = NULL;
+               if (p->post_handler && !kprobe_gone(p)) {
+                       list_for_each_entry_rcu(list_p, &old_p->list, list) {
+                               if ((list_p != p) && (list_p->post_handler))
+                                       goto noclean;
+                       }
+                       old_p->post_handler = NULL;
+               }
+noclean:
+               list_del_rcu(&p->list);
+               if (!kprobe_disabled(old_p)) {
+                       try_to_disable_aggr_kprobe(old_p);
+                       if (!kprobes_all_disarmed) {
+                               if (kprobe_disabled(old_p))
+                                       disarm_kprobe(old_p);
+                               else
+                                       /* Try to optimize this probe again */
+                                       optimize_kprobe(old_p);
+                       }
+               }
+       }
+       return 0;
+}
 
-void unregister_kprobe(struct kprobe *p)
+static void __kprobes __unregister_kprobe_bottom(struct kprobe *p)
 {
-       unsigned long flags;
        struct kprobe *old_p;
 
-       spin_lock_irqsave(&kprobe_lock, flags);
-       old_p = get_kprobe(p->addr);
-       if (old_p) {
-               if (old_p->pre_handler == aggr_pre_handler)
-                       cleanup_aggr_kprobe(old_p, p, flags);
-               else
-                       cleanup_kprobe(p, flags);
-       } else
-               spin_unlock_irqrestore(&kprobe_lock, flags);
+       if (list_empty(&p->list))
+               arch_remove_kprobe(p);
+       else if (list_is_singular(&p->list)) {
+               /* "p" is the last child of an aggr_kprobe */
+               old_p = list_entry(p->list.next, struct kprobe, list);
+               list_del(&p->list);
+               arch_remove_kprobe(old_p);
+               free_aggr_kprobe(old_p);
+       }
+}
+
+int __kprobes register_kprobes(struct kprobe **kps, int num)
+{
+       int i, ret = 0;
+
+       if (num <= 0)
+               return -EINVAL;
+       for (i = 0; i < num; i++) {
+               ret = register_kprobe(kps[i]);
+               if (ret < 0) {
+                       if (i > 0)
+                               unregister_kprobes(kps, i);
+                       break;
+               }
+       }
+       return ret;
+}
+EXPORT_SYMBOL_GPL(register_kprobes);
+
+void __kprobes unregister_kprobe(struct kprobe *p)
+{
+       unregister_kprobes(&p, 1);
+}
+EXPORT_SYMBOL_GPL(unregister_kprobe);
+
+void __kprobes unregister_kprobes(struct kprobe **kps, int num)
+{
+       int i;
+
+       if (num <= 0)
+               return;
+       mutex_lock(&kprobe_mutex);
+       for (i = 0; i < num; i++)
+               if (__unregister_kprobe_top(kps[i]) < 0)
+                       kps[i]->addr = NULL;
+       mutex_unlock(&kprobe_mutex);
+
+       synchronize_sched();
+       for (i = 0; i < num; i++)
+               if (kps[i]->addr)
+                       __unregister_kprobe_bottom(kps[i]);
 }
+EXPORT_SYMBOL_GPL(unregister_kprobes);
 
 static struct notifier_block kprobe_exceptions_nb = {
        .notifier_call = kprobe_exceptions_notify,
-       .priority = 0x7fffffff /* we need to notified first */
+       .priority = 0x7fffffff /* we need to be notified first */
 };
 
-int register_jprobe(struct jprobe *jp)
+unsigned long __weak arch_deref_entry_point(void *entry)
 {
-       /* Todo: Verify probepoint is a function entry point */
-       jp->kp.pre_handler = setjmp_pre_handler;
-       jp->kp.break_handler = longjmp_break_handler;
+       return (unsigned long)entry;
+}
 
-       return register_kprobe(&jp->kp);
+int __kprobes register_jprobes(struct jprobe **jps, int num)
+{
+       struct jprobe *jp;
+       int ret = 0, i;
+
+       if (num <= 0)
+               return -EINVAL;
+       for (i = 0; i < num; i++) {
+               unsigned long addr;
+               jp = jps[i];
+               addr = arch_deref_entry_point(jp->entry);
+
+               if (!kernel_text_address(addr))
+                       ret = -EINVAL;
+               else {
+                       /* Todo: Verify probepoint is a function entry point */
+                       jp->kp.pre_handler = setjmp_pre_handler;
+                       jp->kp.break_handler = longjmp_break_handler;
+                       ret = register_kprobe(&jp->kp);
+               }
+               if (ret < 0) {
+                       if (i > 0)
+                               unregister_jprobes(jps, i);
+                       break;
+               }
+       }
+       return ret;
 }
+EXPORT_SYMBOL_GPL(register_jprobes);
 
-void unregister_jprobe(struct jprobe *jp)
+int __kprobes register_jprobe(struct jprobe *jp)
 {
-       unregister_kprobe(&jp->kp);
+       return register_jprobes(&jp, 1);
+}
+EXPORT_SYMBOL_GPL(register_jprobe);
+
+void __kprobes unregister_jprobe(struct jprobe *jp)
+{
+       unregister_jprobes(&jp, 1);
+}
+EXPORT_SYMBOL_GPL(unregister_jprobe);
+
+void __kprobes unregister_jprobes(struct jprobe **jps, int num)
+{
+       int i;
+
+       if (num <= 0)
+               return;
+       mutex_lock(&kprobe_mutex);
+       for (i = 0; i < num; i++)
+               if (__unregister_kprobe_top(&jps[i]->kp) < 0)
+                       jps[i]->kp.addr = NULL;
+       mutex_unlock(&kprobe_mutex);
+
+       synchronize_sched();
+       for (i = 0; i < num; i++) {
+               if (jps[i]->kp.addr)
+                       __unregister_kprobe_bottom(&jps[i]->kp);
+       }
 }
+EXPORT_SYMBOL_GPL(unregister_jprobes);
 
-#ifdef ARCH_SUPPORTS_KRETPROBES
+#ifdef CONFIG_KRETPROBES
+/*
+ * This kprobe pre_handler is registered with every kretprobe. When probe
+ * hits it will set up the return probe.
+ */
+static int __kprobes pre_handler_kretprobe(struct kprobe *p,
+                                          struct pt_regs *regs)
+{
+       struct kretprobe *rp = container_of(p, struct kretprobe, kp);
+       unsigned long hash, flags = 0;
+       struct kretprobe_instance *ri;
 
-int register_kretprobe(struct kretprobe *rp)
+       /*TODO: consider to only swap the RA after the last pre_handler fired */
+       hash = hash_ptr(current, KPROBE_HASH_BITS);
+       spin_lock_irqsave(&rp->lock, flags);
+       if (!hlist_empty(&rp->free_instances)) {
+               ri = hlist_entry(rp->free_instances.first,
+                               struct kretprobe_instance, hlist);
+               hlist_del(&ri->hlist);
+               spin_unlock_irqrestore(&rp->lock, flags);
+
+               ri->rp = rp;
+               ri->task = current;
+
+               if (rp->entry_handler && rp->entry_handler(ri, regs))
+                       return 0;
+
+               arch_prepare_kretprobe(ri, regs);
+
+               /* XXX(hch): why is there no hlist_move_head? */
+               INIT_HLIST_NODE(&ri->hlist);
+               kretprobe_table_lock(hash, &flags);
+               hlist_add_head(&ri->hlist, &kretprobe_inst_table[hash]);
+               kretprobe_table_unlock(hash, &flags);
+       } else {
+               rp->nmissed++;
+               spin_unlock_irqrestore(&rp->lock, flags);
+       }
+       return 0;
+}
+
+int __kprobes register_kretprobe(struct kretprobe *rp)
 {
        int ret = 0;
        struct kretprobe_instance *inst;
        int i;
+       void *addr;
+
+       if (kretprobe_blacklist_size) {
+               addr = kprobe_addr(&rp->kp);
+               if (!addr)
+                       return -EINVAL;
+
+               for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
+                       if (kretprobe_blacklist[i].addr == addr)
+                               return -EINVAL;
+               }
+       }
 
        rp->kp.pre_handler = pre_handler_kretprobe;
+       rp->kp.post_handler = NULL;
+       rp->kp.fault_handler = NULL;
+       rp->kp.break_handler = NULL;
 
        /* Pre-allocate memory for max kretprobe instances */
        if (rp->maxactive <= 0) {
 #ifdef CONFIG_PREEMPT
-               rp->maxactive = max(10, 2 * NR_CPUS);
+               rp->maxactive = max_t(unsigned int, 10, 2*num_possible_cpus());
 #else
-               rp->maxactive = NR_CPUS;
+               rp->maxactive = num_possible_cpus();
 #endif
        }
-       INIT_HLIST_HEAD(&rp->used_instances);
+       spin_lock_init(&rp->lock);
        INIT_HLIST_HEAD(&rp->free_instances);
        for (i = 0; i < rp->maxactive; i++) {
-               inst = kmalloc(sizeof(struct kretprobe_instance), GFP_KERNEL);
+               inst = kmalloc(sizeof(struct kretprobe_instance) +
+                              rp->data_size, GFP_KERNEL);
                if (inst == NULL) {
                        free_rp_inst(rp);
                        return -ENOMEM;
                }
-               INIT_HLIST_NODE(&inst->uflist);
-               hlist_add_head(&inst->uflist, &rp->free_instances);
+               INIT_HLIST_NODE(&inst->hlist);
+               hlist_add_head(&inst->hlist, &rp->free_instances);
        }
 
        rp->nmissed = 0;
        /* Establish function entry probe point */
-       if ((ret = register_kprobe(&rp->kp)) != 0)
+       ret = register_kprobe(&rp->kp);
+       if (ret != 0)
                free_rp_inst(rp);
        return ret;
 }
+EXPORT_SYMBOL_GPL(register_kretprobe);
+
+int __kprobes register_kretprobes(struct kretprobe **rps, int num)
+{
+       int ret = 0, i;
+
+       if (num <= 0)
+               return -EINVAL;
+       for (i = 0; i < num; i++) {
+               ret = register_kretprobe(rps[i]);
+               if (ret < 0) {
+                       if (i > 0)
+                               unregister_kretprobes(rps, i);
+                       break;
+               }
+       }
+       return ret;
+}
+EXPORT_SYMBOL_GPL(register_kretprobes);
 
-#else /* ARCH_SUPPORTS_KRETPROBES */
+void __kprobes unregister_kretprobe(struct kretprobe *rp)
+{
+       unregister_kretprobes(&rp, 1);
+}
+EXPORT_SYMBOL_GPL(unregister_kretprobe);
+
+void __kprobes unregister_kretprobes(struct kretprobe **rps, int num)
+{
+       int i;
+
+       if (num <= 0)
+               return;
+       mutex_lock(&kprobe_mutex);
+       for (i = 0; i < num; i++)
+               if (__unregister_kprobe_top(&rps[i]->kp) < 0)
+                       rps[i]->kp.addr = NULL;
+       mutex_unlock(&kprobe_mutex);
+
+       synchronize_sched();
+       for (i = 0; i < num; i++) {
+               if (rps[i]->kp.addr) {
+                       __unregister_kprobe_bottom(&rps[i]->kp);
+                       cleanup_rp_inst(rps[i]);
+               }
+       }
+}
+EXPORT_SYMBOL_GPL(unregister_kretprobes);
+
+#else /* CONFIG_KRETPROBES */
+int __kprobes register_kretprobe(struct kretprobe *rp)
+{
+       return -ENOSYS;
+}
+EXPORT_SYMBOL_GPL(register_kretprobe);
 
-int register_kretprobe(struct kretprobe *rp)
+int __kprobes register_kretprobes(struct kretprobe **rps, int num)
 {
        return -ENOSYS;
 }
+EXPORT_SYMBOL_GPL(register_kretprobes);
 
-#endif /* ARCH_SUPPORTS_KRETPROBES */
+void __kprobes unregister_kretprobe(struct kretprobe *rp)
+{
+}
+EXPORT_SYMBOL_GPL(unregister_kretprobe);
 
-void unregister_kretprobe(struct kretprobe *rp)
+void __kprobes unregister_kretprobes(struct kretprobe **rps, int num)
 {
-       unsigned long flags;
-       struct kretprobe_instance *ri;
+}
+EXPORT_SYMBOL_GPL(unregister_kretprobes);
 
-       unregister_kprobe(&rp->kp);
-       /* No race here */
-       spin_lock_irqsave(&kprobe_lock, flags);
-       free_rp_inst(rp);
-       while ((ri = get_used_rp_inst(rp)) != NULL) {
-               ri->rp = NULL;
-               hlist_del(&ri->uflist);
+static int __kprobes pre_handler_kretprobe(struct kprobe *p,
+                                          struct pt_regs *regs)
+{
+       return 0;
+}
+
+#endif /* CONFIG_KRETPROBES */
+
+/* Set the kprobe gone and remove its instruction buffer. */
+static void __kprobes kill_kprobe(struct kprobe *p)
+{
+       struct kprobe *kp;
+
+       p->flags |= KPROBE_FLAG_GONE;
+       if (kprobe_aggrprobe(p)) {
+               /*
+                * If this is an aggr_kprobe, we have to list all the
+                * chained probes and mark them GONE.
+                */
+               list_for_each_entry_rcu(kp, &p->list, list)
+                       kp->flags |= KPROBE_FLAG_GONE;
+               p->post_handler = NULL;
+               p->break_handler = NULL;
+               kill_optimized_kprobe(p);
        }
-       spin_unlock_irqrestore(&kprobe_lock, flags);
+       /*
+        * Here, we can remove insn_slot safely, because no thread calls
+        * the original probed function (which will be freed soon) any more.
+        */
+       arch_remove_kprobe(p);
+}
+
+/* Disable one kprobe */
+int __kprobes disable_kprobe(struct kprobe *kp)
+{
+       int ret = 0;
+       struct kprobe *p;
+
+       mutex_lock(&kprobe_mutex);
+
+       /* Check whether specified probe is valid. */
+       p = __get_valid_kprobe(kp);
+       if (unlikely(p == NULL)) {
+               ret = -EINVAL;
+               goto out;
+       }
+
+       /* If the probe is already disabled (or gone), just return */
+       if (kprobe_disabled(kp))
+               goto out;
+
+       kp->flags |= KPROBE_FLAG_DISABLED;
+       if (p != kp)
+               /* When kp != p, p is always enabled. */
+               try_to_disable_aggr_kprobe(p);
+
+       if (!kprobes_all_disarmed && kprobe_disabled(p))
+               disarm_kprobe(p);
+out:
+       mutex_unlock(&kprobe_mutex);
+       return ret;
+}
+EXPORT_SYMBOL_GPL(disable_kprobe);
+
+/* Enable one kprobe */
+int __kprobes enable_kprobe(struct kprobe *kp)
+{
+       int ret = 0;
+       struct kprobe *p;
+
+       mutex_lock(&kprobe_mutex);
+
+       /* Check whether specified probe is valid. */
+       p = __get_valid_kprobe(kp);
+       if (unlikely(p == NULL)) {
+               ret = -EINVAL;
+               goto out;
+       }
+
+       if (kprobe_gone(kp)) {
+               /* This kprobe has gone, we couldn't enable it. */
+               ret = -EINVAL;
+               goto out;
+       }
+
+       if (p != kp)
+               kp->flags &= ~KPROBE_FLAG_DISABLED;
+
+       if (!kprobes_all_disarmed && kprobe_disabled(p)) {
+               p->flags &= ~KPROBE_FLAG_DISABLED;
+               arm_kprobe(p);
+       }
+out:
+       mutex_unlock(&kprobe_mutex);
+       return ret;
+}
+EXPORT_SYMBOL_GPL(enable_kprobe);
+
+void __kprobes dump_kprobe(struct kprobe *kp)
+{
+       printk(KERN_WARNING "Dumping kprobe:\n");
+       printk(KERN_WARNING "Name: %s\nAddress: %p\nOffset: %x\n",
+              kp->symbol_name, kp->addr, kp->offset);
+}
+
+/* Module notifier call back, checking kprobes on the module */
+static int __kprobes kprobes_module_callback(struct notifier_block *nb,
+                                            unsigned long val, void *data)
+{
+       struct module *mod = data;
+       struct hlist_head *head;
+       struct hlist_node *node;
+       struct kprobe *p;
+       unsigned int i;
+       int checkcore = (val == MODULE_STATE_GOING);
+
+       if (val != MODULE_STATE_GOING && val != MODULE_STATE_LIVE)
+               return NOTIFY_DONE;
+
+       /*
+        * When MODULE_STATE_GOING was notified, both of module .text and
+        * .init.text sections would be freed. When MODULE_STATE_LIVE was
+        * notified, only .init.text section would be freed. We need to
+        * disable kprobes which have been inserted in the sections.
+        */
+       mutex_lock(&kprobe_mutex);
+       for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
+               head = &kprobe_table[i];
+               hlist_for_each_entry_rcu(p, node, head, hlist)
+                       if (within_module_init((unsigned long)p->addr, mod) ||
+                           (checkcore &&
+                            within_module_core((unsigned long)p->addr, mod))) {
+                               /*
+                                * The vaddr this probe is installed will soon
+                                * be vfreed buy not synced to disk. Hence,
+                                * disarming the breakpoint isn't needed.
+                                */
+                               kill_kprobe(p);
+                       }
+       }
+       mutex_unlock(&kprobe_mutex);
+       return NOTIFY_DONE;
 }
 
+static struct notifier_block kprobe_module_nb = {
+       .notifier_call = kprobes_module_callback,
+       .priority = 0
+};
+
 static int __init init_kprobes(void)
 {
        int i, err = 0;
+       unsigned long offset = 0, size = 0;
+       char *modname, namebuf[128];
+       const char *symbol_name;
+       void *addr;
+       struct kprobe_blackpoint *kb;
 
        /* FIXME allocate the probe table, currently defined statically */
        /* initialize all list heads */
        for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
                INIT_HLIST_HEAD(&kprobe_table[i]);
                INIT_HLIST_HEAD(&kretprobe_inst_table[i]);
+               spin_lock_init(&(kretprobe_table_locks[i].lock));
+       }
+
+       /*
+        * Lookup and populate the kprobe_blacklist.
+        *
+        * Unlike the kretprobe blacklist, we'll need to determine
+        * the range of addresses that belong to the said functions,
+        * since a kprobe need not necessarily be at the beginning
+        * of a function.
+        */
+       for (kb = kprobe_blacklist; kb->name != NULL; kb++) {
+               kprobe_lookup_name(kb->name, addr);
+               if (!addr)
+                       continue;
+
+               kb->start_addr = (unsigned long)addr;
+               symbol_name = kallsyms_lookup(kb->start_addr,
+                               &size, &offset, &modname, namebuf);
+               if (!symbol_name)
+                       kb->range = 0;
+               else
+                       kb->range = size;
+       }
+
+       if (kretprobe_blacklist_size) {
+               /* lookup the function address from its name */
+               for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
+                       kprobe_lookup_name(kretprobe_blacklist[i].name,
+                                          kretprobe_blacklist[i].addr);
+                       if (!kretprobe_blacklist[i].addr)
+                               printk("kretprobe: lookup failed: %s\n",
+                                      kretprobe_blacklist[i].name);
+               }
        }
 
-       err = register_die_notifier(&kprobe_exceptions_nb);
-       /* Register the trampoline probe for return probe */
-       register_kprobe(&trampoline_p);
+#if defined(CONFIG_OPTPROBES)
+#if defined(__ARCH_WANT_KPROBES_INSN_SLOT)
+       /* Init kprobe_optinsn_slots */
+       kprobe_optinsn_slots.insn_size = MAX_OPTINSN_SIZE;
+#endif
+       /* By default, kprobes can be optimized */
+       kprobes_allow_optimization = true;
+#endif
+
+       /* By default, kprobes are armed */
+       kprobes_all_disarmed = false;
+
+       err = arch_init_kprobes();
+       if (!err)
+               err = register_die_notifier(&kprobe_exceptions_nb);
+       if (!err)
+               err = register_module_notifier(&kprobe_module_nb);
+
+       kprobes_initialized = (err == 0);
+
+       if (!err)
+               init_test_probes();
        return err;
 }
 
-__initcall(init_kprobes);
+#ifdef CONFIG_DEBUG_FS
+static void __kprobes report_probe(struct seq_file *pi, struct kprobe *p,
+               const char *sym, int offset, char *modname, struct kprobe *pp)
+{
+       char *kprobe_type;
+
+       if (p->pre_handler == pre_handler_kretprobe)
+               kprobe_type = "r";
+       else if (p->pre_handler == setjmp_pre_handler)
+               kprobe_type = "j";
+       else
+               kprobe_type = "k";
+
+       if (sym)
+               seq_printf(pi, "%p  %s  %s+0x%x  %s ",
+                       p->addr, kprobe_type, sym, offset,
+                       (modname ? modname : " "));
+       else
+               seq_printf(pi, "%p  %s  %p ",
+                       p->addr, kprobe_type, p->addr);
+
+       if (!pp)
+               pp = p;
+       seq_printf(pi, "%s%s%s\n",
+               (kprobe_gone(p) ? "[GONE]" : ""),
+               ((kprobe_disabled(p) && !kprobe_gone(p)) ?  "[DISABLED]" : ""),
+               (kprobe_optimized(pp) ? "[OPTIMIZED]" : ""));
+}
 
-EXPORT_SYMBOL_GPL(register_kprobe);
-EXPORT_SYMBOL_GPL(unregister_kprobe);
-EXPORT_SYMBOL_GPL(register_jprobe);
-EXPORT_SYMBOL_GPL(unregister_jprobe);
-EXPORT_SYMBOL_GPL(jprobe_return);
-EXPORT_SYMBOL_GPL(register_kretprobe);
-EXPORT_SYMBOL_GPL(unregister_kretprobe);
+static void __kprobes *kprobe_seq_start(struct seq_file *f, loff_t *pos)
+{
+       return (*pos < KPROBE_TABLE_SIZE) ? pos : NULL;
+}
 
+static void __kprobes *kprobe_seq_next(struct seq_file *f, void *v, loff_t *pos)
+{
+       (*pos)++;
+       if (*pos >= KPROBE_TABLE_SIZE)
+               return NULL;
+       return pos;
+}
+
+static void __kprobes kprobe_seq_stop(struct seq_file *f, void *v)
+{
+       /* Nothing to do */
+}
+
+static int __kprobes show_kprobe_addr(struct seq_file *pi, void *v)
+{
+       struct hlist_head *head;
+       struct hlist_node *node;
+       struct kprobe *p, *kp;
+       const char *sym = NULL;
+       unsigned int i = *(loff_t *) v;
+       unsigned long offset = 0;
+       char *modname, namebuf[128];
+
+       head = &kprobe_table[i];
+       preempt_disable();
+       hlist_for_each_entry_rcu(p, node, head, hlist) {
+               sym = kallsyms_lookup((unsigned long)p->addr, NULL,
+                                       &offset, &modname, namebuf);
+               if (kprobe_aggrprobe(p)) {
+                       list_for_each_entry_rcu(kp, &p->list, list)
+                               report_probe(pi, kp, sym, offset, modname, p);
+               } else
+                       report_probe(pi, p, sym, offset, modname, NULL);
+       }
+       preempt_enable();
+       return 0;
+}
+
+static const struct seq_operations kprobes_seq_ops = {
+       .start = kprobe_seq_start,
+       .next  = kprobe_seq_next,
+       .stop  = kprobe_seq_stop,
+       .show  = show_kprobe_addr
+};
+
+static int __kprobes kprobes_open(struct inode *inode, struct file *filp)
+{
+       return seq_open(filp, &kprobes_seq_ops);
+}
+
+static const struct file_operations debugfs_kprobes_operations = {
+       .open           = kprobes_open,
+       .read           = seq_read,
+       .llseek         = seq_lseek,
+       .release        = seq_release,
+};
+
+static void __kprobes arm_all_kprobes(void)
+{
+       struct hlist_head *head;
+       struct hlist_node *node;
+       struct kprobe *p;
+       unsigned int i;
+
+       mutex_lock(&kprobe_mutex);
+
+       /* If kprobes are armed, just return */
+       if (!kprobes_all_disarmed)
+               goto already_enabled;
+
+       /* Arming kprobes doesn't optimize kprobe itself */
+       mutex_lock(&text_mutex);
+       for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
+               head = &kprobe_table[i];
+               hlist_for_each_entry_rcu(p, node, head, hlist)
+                       if (!kprobe_disabled(p))
+                               __arm_kprobe(p);
+       }
+       mutex_unlock(&text_mutex);
+
+       kprobes_all_disarmed = false;
+       printk(KERN_INFO "Kprobes globally enabled\n");
+
+already_enabled:
+       mutex_unlock(&kprobe_mutex);
+       return;
+}
+
+static void __kprobes disarm_all_kprobes(void)
+{
+       struct hlist_head *head;
+       struct hlist_node *node;
+       struct kprobe *p;
+       unsigned int i;
+
+       mutex_lock(&kprobe_mutex);
+
+       /* If kprobes are already disarmed, just return */
+       if (kprobes_all_disarmed)
+               goto already_disabled;
+
+       kprobes_all_disarmed = true;
+       printk(KERN_INFO "Kprobes globally disabled\n");
+
+       /*
+        * Here we call get_online_cpus() for avoiding text_mutex deadlock,
+        * because disarming may also unoptimize kprobes.
+        */
+       get_online_cpus();
+       mutex_lock(&text_mutex);
+       for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
+               head = &kprobe_table[i];
+               hlist_for_each_entry_rcu(p, node, head, hlist) {
+                       if (!arch_trampoline_kprobe(p) && !kprobe_disabled(p))
+                               __disarm_kprobe(p);
+               }
+       }
+
+       mutex_unlock(&text_mutex);
+       put_online_cpus();
+       mutex_unlock(&kprobe_mutex);
+       /* Allow all currently running kprobes to complete */
+       synchronize_sched();
+       return;
+
+already_disabled:
+       mutex_unlock(&kprobe_mutex);
+       return;
+}
+
+/*
+ * XXX: The debugfs bool file interface doesn't allow for callbacks
+ * when the bool state is switched. We can reuse that facility when
+ * available
+ */
+static ssize_t read_enabled_file_bool(struct file *file,
+              char __user *user_buf, size_t count, loff_t *ppos)
+{
+       char buf[3];
+
+       if (!kprobes_all_disarmed)
+               buf[0] = '1';
+       else
+               buf[0] = '0';
+       buf[1] = '\n';
+       buf[2] = 0x00;
+       return simple_read_from_buffer(user_buf, count, ppos, buf, 2);
+}
+
+static ssize_t write_enabled_file_bool(struct file *file,
+              const char __user *user_buf, size_t count, loff_t *ppos)
+{
+       char buf[32];
+       int buf_size;
+
+       buf_size = min(count, (sizeof(buf)-1));
+       if (copy_from_user(buf, user_buf, buf_size))
+               return -EFAULT;
+
+       switch (buf[0]) {
+       case 'y':
+       case 'Y':
+       case '1':
+               arm_all_kprobes();
+               break;
+       case 'n':
+       case 'N':
+       case '0':
+               disarm_all_kprobes();
+               break;
+       }
+
+       return count;
+}
+
+static const struct file_operations fops_kp = {
+       .read =         read_enabled_file_bool,
+       .write =        write_enabled_file_bool,
+};
+
+static int __kprobes debugfs_kprobe_init(void)
+{
+       struct dentry *dir, *file;
+       unsigned int value = 1;
+
+       dir = debugfs_create_dir("kprobes", NULL);
+       if (!dir)
+               return -ENOMEM;
+
+       file = debugfs_create_file("list", 0444, dir, NULL,
+                               &debugfs_kprobes_operations);
+       if (!file) {
+               debugfs_remove(dir);
+               return -ENOMEM;
+       }
+
+       file = debugfs_create_file("enabled", 0600, dir,
+                                       &value, &fops_kp);
+       if (!file) {
+               debugfs_remove(dir);
+               return -ENOMEM;
+       }
+
+       return 0;
+}
+
+late_initcall(debugfs_kprobe_init);
+#endif /* CONFIG_DEBUG_FS */
+
+module_init(init_kprobes);
+
+/* defined in arch/.../kernel/kprobes.c */
+EXPORT_SYMBOL_GPL(jprobe_return);