Root/kernel/kprobes.c

1/*
2 * Kernel Probes (KProbes)
3 * kernel/kprobes.c
4 *
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.
9 *
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.
14 *
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.
18 *
19 * Copyright (C) IBM Corporation, 2002, 2004
20 *
21 * 2002-Oct Created by Vamsi Krishna S <vamsi_krishna@in.ibm.com> Kernel
22 * Probes initial implementation (includes suggestions from
23 * Rusty Russell).
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.
33 */
34#include <linux/kprobes.h>
35#include <linux/hash.h>
36#include <linux/init.h>
37#include <linux/slab.h>
38#include <linux/stddef.h>
39#include <linux/module.h>
40#include <linux/moduleloader.h>
41#include <linux/kallsyms.h>
42#include <linux/freezer.h>
43#include <linux/seq_file.h>
44#include <linux/debugfs.h>
45#include <linux/sysctl.h>
46#include <linux/kdebug.h>
47#include <linux/memory.h>
48#include <linux/ftrace.h>
49#include <linux/cpu.h>
50#include <linux/jump_label.h>
51
52#include <asm-generic/sections.h>
53#include <asm/cacheflush.h>
54#include <asm/errno.h>
55#include <asm/uaccess.h>
56
57#define KPROBE_HASH_BITS 6
58#define KPROBE_TABLE_SIZE (1 << KPROBE_HASH_BITS)
59
60
61/*
62 * Some oddball architectures like 64bit powerpc have function descriptors
63 * so this must be overridable.
64 */
65#ifndef kprobe_lookup_name
66#define kprobe_lookup_name(name, addr) \
67    addr = ((kprobe_opcode_t *)(kallsyms_lookup_name(name)))
68#endif
69
70static int kprobes_initialized;
71static struct hlist_head kprobe_table[KPROBE_TABLE_SIZE];
72static struct hlist_head kretprobe_inst_table[KPROBE_TABLE_SIZE];
73
74/* NOTE: change this value only with kprobe_mutex held */
75static bool kprobes_all_disarmed;
76
77/* This protects kprobe_table and optimizing_list */
78static DEFINE_MUTEX(kprobe_mutex);
79static DEFINE_PER_CPU(struct kprobe *, kprobe_instance) = NULL;
80static struct {
81    spinlock_t lock ____cacheline_aligned_in_smp;
82} kretprobe_table_locks[KPROBE_TABLE_SIZE];
83
84static spinlock_t *kretprobe_table_lock_ptr(unsigned long hash)
85{
86    return &(kretprobe_table_locks[hash].lock);
87}
88
89/*
90 * Normally, functions that we'd want to prohibit kprobes in, are marked
91 * __kprobes. But, there are cases where such functions already belong to
92 * a different section (__sched for preempt_schedule)
93 *
94 * For such cases, we now have a blacklist
95 */
96static struct kprobe_blackpoint kprobe_blacklist[] = {
97    {"preempt_schedule",},
98    {"native_get_debugreg",},
99    {"irq_entries_start",},
100    {"common_interrupt",},
101    {"mcount",}, /* mcount can be called from everywhere */
102    {NULL} /* Terminator */
103};
104
105#ifdef __ARCH_WANT_KPROBES_INSN_SLOT
106/*
107 * kprobe->ainsn.insn points to the copy of the instruction to be
108 * single-stepped. x86_64, POWER4 and above have no-exec support and
109 * stepping on the instruction on a vmalloced/kmalloced/data page
110 * is a recipe for disaster
111 */
112struct kprobe_insn_page {
113    struct list_head list;
114    kprobe_opcode_t *insns; /* Page of instruction slots */
115    int nused;
116    int ngarbage;
117    char slot_used[];
118};
119
120#define KPROBE_INSN_PAGE_SIZE(slots) \
121    (offsetof(struct kprobe_insn_page, slot_used) + \
122     (sizeof(char) * (slots)))
123
124struct kprobe_insn_cache {
125    struct list_head pages; /* list of kprobe_insn_page */
126    size_t insn_size; /* size of instruction slot */
127    int nr_garbage;
128};
129
130static int slots_per_page(struct kprobe_insn_cache *c)
131{
132    return PAGE_SIZE/(c->insn_size * sizeof(kprobe_opcode_t));
133}
134
135enum kprobe_slot_state {
136    SLOT_CLEAN = 0,
137    SLOT_DIRTY = 1,
138    SLOT_USED = 2,
139};
140
141static DEFINE_MUTEX(kprobe_insn_mutex); /* Protects kprobe_insn_slots */
142static struct kprobe_insn_cache kprobe_insn_slots = {
143    .pages = LIST_HEAD_INIT(kprobe_insn_slots.pages),
144    .insn_size = MAX_INSN_SIZE,
145    .nr_garbage = 0,
146};
147static int __kprobes collect_garbage_slots(struct kprobe_insn_cache *c);
148
149/**
150 * __get_insn_slot() - Find a slot on an executable page for an instruction.
151 * We allocate an executable page if there's no room on existing ones.
152 */
153static kprobe_opcode_t __kprobes *__get_insn_slot(struct kprobe_insn_cache *c)
154{
155    struct kprobe_insn_page *kip;
156
157 retry:
158    list_for_each_entry(kip, &c->pages, list) {
159        if (kip->nused < slots_per_page(c)) {
160            int i;
161            for (i = 0; i < slots_per_page(c); i++) {
162                if (kip->slot_used[i] == SLOT_CLEAN) {
163                    kip->slot_used[i] = SLOT_USED;
164                    kip->nused++;
165                    return kip->insns + (i * c->insn_size);
166                }
167            }
168            /* kip->nused is broken. Fix it. */
169            kip->nused = slots_per_page(c);
170            WARN_ON(1);
171        }
172    }
173
174    /* If there are any garbage slots, collect it and try again. */
175    if (c->nr_garbage && collect_garbage_slots(c) == 0)
176        goto retry;
177
178    /* All out of space. Need to allocate a new page. */
179    kip = kmalloc(KPROBE_INSN_PAGE_SIZE(slots_per_page(c)), GFP_KERNEL);
180    if (!kip)
181        return NULL;
182
183    /*
184     * Use module_alloc so this page is within +/- 2GB of where the
185     * kernel image and loaded module images reside. This is required
186     * so x86_64 can correctly handle the %rip-relative fixups.
187     */
188    kip->insns = module_alloc(PAGE_SIZE);
189    if (!kip->insns) {
190        kfree(kip);
191        return NULL;
192    }
193    INIT_LIST_HEAD(&kip->list);
194    memset(kip->slot_used, SLOT_CLEAN, slots_per_page(c));
195    kip->slot_used[0] = SLOT_USED;
196    kip->nused = 1;
197    kip->ngarbage = 0;
198    list_add(&kip->list, &c->pages);
199    return kip->insns;
200}
201
202
203kprobe_opcode_t __kprobes *get_insn_slot(void)
204{
205    kprobe_opcode_t *ret = NULL;
206
207    mutex_lock(&kprobe_insn_mutex);
208    ret = __get_insn_slot(&kprobe_insn_slots);
209    mutex_unlock(&kprobe_insn_mutex);
210
211    return ret;
212}
213
214/* Return 1 if all garbages are collected, otherwise 0. */
215static int __kprobes collect_one_slot(struct kprobe_insn_page *kip, int idx)
216{
217    kip->slot_used[idx] = SLOT_CLEAN;
218    kip->nused--;
219    if (kip->nused == 0) {
220        /*
221         * Page is no longer in use. Free it unless
222         * it's the last one. We keep the last one
223         * so as not to have to set it up again the
224         * next time somebody inserts a probe.
225         */
226        if (!list_is_singular(&kip->list)) {
227            list_del(&kip->list);
228            module_free(NULL, kip->insns);
229            kfree(kip);
230        }
231        return 1;
232    }
233    return 0;
234}
235
236static int __kprobes collect_garbage_slots(struct kprobe_insn_cache *c)
237{
238    struct kprobe_insn_page *kip, *next;
239
240    /* Ensure no-one is interrupted on the garbages */
241    synchronize_sched();
242
243    list_for_each_entry_safe(kip, next, &c->pages, list) {
244        int i;
245        if (kip->ngarbage == 0)
246            continue;
247        kip->ngarbage = 0; /* we will collect all garbages */
248        for (i = 0; i < slots_per_page(c); i++) {
249            if (kip->slot_used[i] == SLOT_DIRTY &&
250                collect_one_slot(kip, i))
251                break;
252        }
253    }
254    c->nr_garbage = 0;
255    return 0;
256}
257
258static void __kprobes __free_insn_slot(struct kprobe_insn_cache *c,
259                       kprobe_opcode_t *slot, int dirty)
260{
261    struct kprobe_insn_page *kip;
262
263    list_for_each_entry(kip, &c->pages, list) {
264        long idx = ((long)slot - (long)kip->insns) /
265                (c->insn_size * sizeof(kprobe_opcode_t));
266        if (idx >= 0 && idx < slots_per_page(c)) {
267            WARN_ON(kip->slot_used[idx] != SLOT_USED);
268            if (dirty) {
269                kip->slot_used[idx] = SLOT_DIRTY;
270                kip->ngarbage++;
271                if (++c->nr_garbage > slots_per_page(c))
272                    collect_garbage_slots(c);
273            } else
274                collect_one_slot(kip, idx);
275            return;
276        }
277    }
278    /* Could not free this slot. */
279    WARN_ON(1);
280}
281
282void __kprobes free_insn_slot(kprobe_opcode_t * slot, int dirty)
283{
284    mutex_lock(&kprobe_insn_mutex);
285    __free_insn_slot(&kprobe_insn_slots, slot, dirty);
286    mutex_unlock(&kprobe_insn_mutex);
287}
288#ifdef CONFIG_OPTPROBES
289/* For optimized_kprobe buffer */
290static DEFINE_MUTEX(kprobe_optinsn_mutex); /* Protects kprobe_optinsn_slots */
291static struct kprobe_insn_cache kprobe_optinsn_slots = {
292    .pages = LIST_HEAD_INIT(kprobe_optinsn_slots.pages),
293    /* .insn_size is initialized later */
294    .nr_garbage = 0,
295};
296/* Get a slot for optimized_kprobe buffer */
297kprobe_opcode_t __kprobes *get_optinsn_slot(void)
298{
299    kprobe_opcode_t *ret = NULL;
300
301    mutex_lock(&kprobe_optinsn_mutex);
302    ret = __get_insn_slot(&kprobe_optinsn_slots);
303    mutex_unlock(&kprobe_optinsn_mutex);
304
305    return ret;
306}
307
308void __kprobes free_optinsn_slot(kprobe_opcode_t * slot, int dirty)
309{
310    mutex_lock(&kprobe_optinsn_mutex);
311    __free_insn_slot(&kprobe_optinsn_slots, slot, dirty);
312    mutex_unlock(&kprobe_optinsn_mutex);
313}
314#endif
315#endif
316
317/* We have preemption disabled.. so it is safe to use __ versions */
318static inline void set_kprobe_instance(struct kprobe *kp)
319{
320    __this_cpu_write(kprobe_instance, kp);
321}
322
323static inline void reset_kprobe_instance(void)
324{
325    __this_cpu_write(kprobe_instance, NULL);
326}
327
328/*
329 * This routine is called either:
330 * - under the kprobe_mutex - during kprobe_[un]register()
331 * OR
332 * - with preemption disabled - from arch/xxx/kernel/kprobes.c
333 */
334struct kprobe __kprobes *get_kprobe(void *addr)
335{
336    struct hlist_head *head;
337    struct hlist_node *node;
338    struct kprobe *p;
339
340    head = &kprobe_table[hash_ptr(addr, KPROBE_HASH_BITS)];
341    hlist_for_each_entry_rcu(p, node, head, hlist) {
342        if (p->addr == addr)
343            return p;
344    }
345
346    return NULL;
347}
348
349static int __kprobes aggr_pre_handler(struct kprobe *p, struct pt_regs *regs);
350
351/* Return true if the kprobe is an aggregator */
352static inline int kprobe_aggrprobe(struct kprobe *p)
353{
354    return p->pre_handler == aggr_pre_handler;
355}
356
357/* Return true(!0) if the kprobe is unused */
358static inline int kprobe_unused(struct kprobe *p)
359{
360    return kprobe_aggrprobe(p) && kprobe_disabled(p) &&
361           list_empty(&p->list);
362}
363
364/*
365 * Keep all fields in the kprobe consistent
366 */
367static inline void copy_kprobe(struct kprobe *ap, struct kprobe *p)
368{
369    memcpy(&p->opcode, &ap->opcode, sizeof(kprobe_opcode_t));
370    memcpy(&p->ainsn, &ap->ainsn, sizeof(struct arch_specific_insn));
371}
372
373#ifdef CONFIG_OPTPROBES
374/* NOTE: change this value only with kprobe_mutex held */
375static bool kprobes_allow_optimization;
376
377/*
378 * Call all pre_handler on the list, but ignores its return value.
379 * This must be called from arch-dep optimized caller.
380 */
381void __kprobes opt_pre_handler(struct kprobe *p, struct pt_regs *regs)
382{
383    struct kprobe *kp;
384
385    list_for_each_entry_rcu(kp, &p->list, list) {
386        if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
387            set_kprobe_instance(kp);
388            kp->pre_handler(kp, regs);
389        }
390        reset_kprobe_instance();
391    }
392}
393
394/* Free optimized instructions and optimized_kprobe */
395static __kprobes void free_aggr_kprobe(struct kprobe *p)
396{
397    struct optimized_kprobe *op;
398
399    op = container_of(p, struct optimized_kprobe, kp);
400    arch_remove_optimized_kprobe(op);
401    arch_remove_kprobe(p);
402    kfree(op);
403}
404
405/* Return true(!0) if the kprobe is ready for optimization. */
406static inline int kprobe_optready(struct kprobe *p)
407{
408    struct optimized_kprobe *op;
409
410    if (kprobe_aggrprobe(p)) {
411        op = container_of(p, struct optimized_kprobe, kp);
412        return arch_prepared_optinsn(&op->optinsn);
413    }
414
415    return 0;
416}
417
418/* Return true(!0) if the kprobe is disarmed. Note: p must be on hash list */
419static inline int kprobe_disarmed(struct kprobe *p)
420{
421    struct optimized_kprobe *op;
422
423    /* If kprobe is not aggr/opt probe, just return kprobe is disabled */
424    if (!kprobe_aggrprobe(p))
425        return kprobe_disabled(p);
426
427    op = container_of(p, struct optimized_kprobe, kp);
428
429    return kprobe_disabled(p) && list_empty(&op->list);
430}
431
432/* Return true(!0) if the probe is queued on (un)optimizing lists */
433static int __kprobes kprobe_queued(struct kprobe *p)
434{
435    struct optimized_kprobe *op;
436
437    if (kprobe_aggrprobe(p)) {
438        op = container_of(p, struct optimized_kprobe, kp);
439        if (!list_empty(&op->list))
440            return 1;
441    }
442    return 0;
443}
444
445/*
446 * Return an optimized kprobe whose optimizing code replaces
447 * instructions including addr (exclude breakpoint).
448 */
449static struct kprobe *__kprobes get_optimized_kprobe(unsigned long addr)
450{
451    int i;
452    struct kprobe *p = NULL;
453    struct optimized_kprobe *op;
454
455    /* Don't check i == 0, since that is a breakpoint case. */
456    for (i = 1; !p && i < MAX_OPTIMIZED_LENGTH; i++)
457        p = get_kprobe((void *)(addr - i));
458
459    if (p && kprobe_optready(p)) {
460        op = container_of(p, struct optimized_kprobe, kp);
461        if (arch_within_optimized_kprobe(op, addr))
462            return p;
463    }
464
465    return NULL;
466}
467
468/* Optimization staging list, protected by kprobe_mutex */
469static LIST_HEAD(optimizing_list);
470static LIST_HEAD(unoptimizing_list);
471
472static void kprobe_optimizer(struct work_struct *work);
473static DECLARE_DELAYED_WORK(optimizing_work, kprobe_optimizer);
474static DECLARE_COMPLETION(optimizer_comp);
475#define OPTIMIZE_DELAY 5
476
477/*
478 * Optimize (replace a breakpoint with a jump) kprobes listed on
479 * optimizing_list.
480 */
481static __kprobes void do_optimize_kprobes(void)
482{
483    /* Optimization never be done when disarmed */
484    if (kprobes_all_disarmed || !kprobes_allow_optimization ||
485        list_empty(&optimizing_list))
486        return;
487
488    /*
489     * The optimization/unoptimization refers online_cpus via
490     * stop_machine() and cpu-hotplug modifies online_cpus.
491     * And same time, text_mutex will be held in cpu-hotplug and here.
492     * This combination can cause a deadlock (cpu-hotplug try to lock
493     * text_mutex but stop_machine can not be done because online_cpus
494     * has been changed)
495     * To avoid this deadlock, we need to call get_online_cpus()
496     * for preventing cpu-hotplug outside of text_mutex locking.
497     */
498    get_online_cpus();
499    mutex_lock(&text_mutex);
500    arch_optimize_kprobes(&optimizing_list);
501    mutex_unlock(&text_mutex);
502    put_online_cpus();
503}
504
505/*
506 * Unoptimize (replace a jump with a breakpoint and remove the breakpoint
507 * if need) kprobes listed on unoptimizing_list.
508 */
509static __kprobes void do_unoptimize_kprobes(struct list_head *free_list)
510{
511    struct optimized_kprobe *op, *tmp;
512
513    /* Unoptimization must be done anytime */
514    if (list_empty(&unoptimizing_list))
515        return;
516
517    /* Ditto to do_optimize_kprobes */
518    get_online_cpus();
519    mutex_lock(&text_mutex);
520    arch_unoptimize_kprobes(&unoptimizing_list, free_list);
521    /* Loop free_list for disarming */
522    list_for_each_entry_safe(op, tmp, free_list, list) {
523        /* Disarm probes if marked disabled */
524        if (kprobe_disabled(&op->kp))
525            arch_disarm_kprobe(&op->kp);
526        if (kprobe_unused(&op->kp)) {
527            /*
528             * Remove unused probes from hash list. After waiting
529             * for synchronization, these probes are reclaimed.
530             * (reclaiming is done by do_free_cleaned_kprobes.)
531             */
532            hlist_del_rcu(&op->kp.hlist);
533        } else
534            list_del_init(&op->list);
535    }
536    mutex_unlock(&text_mutex);
537    put_online_cpus();
538}
539
540/* Reclaim all kprobes on the free_list */
541static __kprobes void do_free_cleaned_kprobes(struct list_head *free_list)
542{
543    struct optimized_kprobe *op, *tmp;
544
545    list_for_each_entry_safe(op, tmp, free_list, list) {
546        BUG_ON(!kprobe_unused(&op->kp));
547        list_del_init(&op->list);
548        free_aggr_kprobe(&op->kp);
549    }
550}
551
552/* Start optimizer after OPTIMIZE_DELAY passed */
553static __kprobes void kick_kprobe_optimizer(void)
554{
555    if (!delayed_work_pending(&optimizing_work))
556        schedule_delayed_work(&optimizing_work, OPTIMIZE_DELAY);
557}
558
559/* Kprobe jump optimizer */
560static __kprobes void kprobe_optimizer(struct work_struct *work)
561{
562    LIST_HEAD(free_list);
563
564    /* Lock modules while optimizing kprobes */
565    mutex_lock(&module_mutex);
566    mutex_lock(&kprobe_mutex);
567
568    /*
569     * Step 1: Unoptimize kprobes and collect cleaned (unused and disarmed)
570     * kprobes before waiting for quiesence period.
571     */
572    do_unoptimize_kprobes(&free_list);
573
574    /*
575     * Step 2: Wait for quiesence period to ensure all running interrupts
576     * are done. Because optprobe may modify multiple instructions
577     * there is a chance that Nth instruction is interrupted. In that
578     * case, running interrupt can return to 2nd-Nth byte of jump
579     * instruction. This wait is for avoiding it.
580     */
581    synchronize_sched();
582
583    /* Step 3: Optimize kprobes after quiesence period */
584    do_optimize_kprobes();
585
586    /* Step 4: Free cleaned kprobes after quiesence period */
587    do_free_cleaned_kprobes(&free_list);
588
589    mutex_unlock(&kprobe_mutex);
590    mutex_unlock(&module_mutex);
591
592    /* Step 5: Kick optimizer again if needed */
593    if (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list))
594        kick_kprobe_optimizer();
595    else
596        /* Wake up all waiters */
597        complete_all(&optimizer_comp);
598}
599
600/* Wait for completing optimization and unoptimization */
601static __kprobes void wait_for_kprobe_optimizer(void)
602{
603    if (delayed_work_pending(&optimizing_work))
604        wait_for_completion(&optimizer_comp);
605}
606
607/* Optimize kprobe if p is ready to be optimized */
608static __kprobes void optimize_kprobe(struct kprobe *p)
609{
610    struct optimized_kprobe *op;
611
612    /* Check if the kprobe is disabled or not ready for optimization. */
613    if (!kprobe_optready(p) || !kprobes_allow_optimization ||
614        (kprobe_disabled(p) || kprobes_all_disarmed))
615        return;
616
617    /* Both of break_handler and post_handler are not supported. */
618    if (p->break_handler || p->post_handler)
619        return;
620
621    op = container_of(p, struct optimized_kprobe, kp);
622
623    /* Check there is no other kprobes at the optimized instructions */
624    if (arch_check_optimized_kprobe(op) < 0)
625        return;
626
627    /* Check if it is already optimized. */
628    if (op->kp.flags & KPROBE_FLAG_OPTIMIZED)
629        return;
630    op->kp.flags |= KPROBE_FLAG_OPTIMIZED;
631
632    if (!list_empty(&op->list))
633        /* This is under unoptimizing. Just dequeue the probe */
634        list_del_init(&op->list);
635    else {
636        list_add(&op->list, &optimizing_list);
637        kick_kprobe_optimizer();
638    }
639}
640
641/* Short cut to direct unoptimizing */
642static __kprobes void force_unoptimize_kprobe(struct optimized_kprobe *op)
643{
644    get_online_cpus();
645    arch_unoptimize_kprobe(op);
646    put_online_cpus();
647    if (kprobe_disabled(&op->kp))
648        arch_disarm_kprobe(&op->kp);
649}
650
651/* Unoptimize a kprobe if p is optimized */
652static __kprobes void unoptimize_kprobe(struct kprobe *p, bool force)
653{
654    struct optimized_kprobe *op;
655
656    if (!kprobe_aggrprobe(p) || kprobe_disarmed(p))
657        return; /* This is not an optprobe nor optimized */
658
659    op = container_of(p, struct optimized_kprobe, kp);
660    if (!kprobe_optimized(p)) {
661        /* Unoptimized or unoptimizing case */
662        if (force && !list_empty(&op->list)) {
663            /*
664             * Only if this is unoptimizing kprobe and forced,
665             * forcibly unoptimize it. (No need to unoptimize
666             * unoptimized kprobe again :)
667             */
668            list_del_init(&op->list);
669            force_unoptimize_kprobe(op);
670        }
671        return;
672    }
673
674    op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
675    if (!list_empty(&op->list)) {
676        /* Dequeue from the optimization queue */
677        list_del_init(&op->list);
678        return;
679    }
680    /* Optimized kprobe case */
681    if (force)
682        /* Forcibly update the code: this is a special case */
683        force_unoptimize_kprobe(op);
684    else {
685        list_add(&op->list, &unoptimizing_list);
686        kick_kprobe_optimizer();
687    }
688}
689
690/* Cancel unoptimizing for reusing */
691static void reuse_unused_kprobe(struct kprobe *ap)
692{
693    struct optimized_kprobe *op;
694
695    BUG_ON(!kprobe_unused(ap));
696    /*
697     * Unused kprobe MUST be on the way of delayed unoptimizing (means
698     * there is still a relative jump) and disabled.
699     */
700    op = container_of(ap, struct optimized_kprobe, kp);
701    if (unlikely(list_empty(&op->list)))
702        printk(KERN_WARNING "Warning: found a stray unused "
703            "aggrprobe@%p\n", ap->addr);
704    /* Enable the probe again */
705    ap->flags &= ~KPROBE_FLAG_DISABLED;
706    /* Optimize it again (remove from op->list) */
707    BUG_ON(!kprobe_optready(ap));
708    optimize_kprobe(ap);
709}
710
711/* Remove optimized instructions */
712static void __kprobes kill_optimized_kprobe(struct kprobe *p)
713{
714    struct optimized_kprobe *op;
715
716    op = container_of(p, struct optimized_kprobe, kp);
717    if (!list_empty(&op->list))
718        /* Dequeue from the (un)optimization queue */
719        list_del_init(&op->list);
720
721    op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
722    /* Don't touch the code, because it is already freed. */
723    arch_remove_optimized_kprobe(op);
724}
725
726/* Try to prepare optimized instructions */
727static __kprobes void prepare_optimized_kprobe(struct kprobe *p)
728{
729    struct optimized_kprobe *op;
730
731    op = container_of(p, struct optimized_kprobe, kp);
732    arch_prepare_optimized_kprobe(op);
733}
734
735/* Allocate new optimized_kprobe and try to prepare optimized instructions */
736static __kprobes struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
737{
738    struct optimized_kprobe *op;
739
740    op = kzalloc(sizeof(struct optimized_kprobe), GFP_KERNEL);
741    if (!op)
742        return NULL;
743
744    INIT_LIST_HEAD(&op->list);
745    op->kp.addr = p->addr;
746    arch_prepare_optimized_kprobe(op);
747
748    return &op->kp;
749}
750
751static void __kprobes init_aggr_kprobe(struct kprobe *ap, struct kprobe *p);
752
753/*
754 * Prepare an optimized_kprobe and optimize it
755 * NOTE: p must be a normal registered kprobe
756 */
757static __kprobes void try_to_optimize_kprobe(struct kprobe *p)
758{
759    struct kprobe *ap;
760    struct optimized_kprobe *op;
761
762    ap = alloc_aggr_kprobe(p);
763    if (!ap)
764        return;
765
766    op = container_of(ap, struct optimized_kprobe, kp);
767    if (!arch_prepared_optinsn(&op->optinsn)) {
768        /* If failed to setup optimizing, fallback to kprobe */
769        arch_remove_optimized_kprobe(op);
770        kfree(op);
771        return;
772    }
773
774    init_aggr_kprobe(ap, p);
775    optimize_kprobe(ap);
776}
777
778#ifdef CONFIG_SYSCTL
779/* This should be called with kprobe_mutex locked */
780static void __kprobes optimize_all_kprobes(void)
781{
782    struct hlist_head *head;
783    struct hlist_node *node;
784    struct kprobe *p;
785    unsigned int i;
786
787    /* If optimization is already allowed, just return */
788    if (kprobes_allow_optimization)
789        return;
790
791    kprobes_allow_optimization = true;
792    for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
793        head = &kprobe_table[i];
794        hlist_for_each_entry_rcu(p, node, head, hlist)
795            if (!kprobe_disabled(p))
796                optimize_kprobe(p);
797    }
798    printk(KERN_INFO "Kprobes globally optimized\n");
799}
800
801/* This should be called with kprobe_mutex locked */
802static void __kprobes unoptimize_all_kprobes(void)
803{
804    struct hlist_head *head;
805    struct hlist_node *node;
806    struct kprobe *p;
807    unsigned int i;
808
809    /* If optimization is already prohibited, just return */
810    if (!kprobes_allow_optimization)
811        return;
812
813    kprobes_allow_optimization = false;
814    for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
815        head = &kprobe_table[i];
816        hlist_for_each_entry_rcu(p, node, head, hlist) {
817            if (!kprobe_disabled(p))
818                unoptimize_kprobe(p, false);
819        }
820    }
821    /* Wait for unoptimizing completion */
822    wait_for_kprobe_optimizer();
823    printk(KERN_INFO "Kprobes globally unoptimized\n");
824}
825
826int sysctl_kprobes_optimization;
827int proc_kprobes_optimization_handler(struct ctl_table *table, int write,
828                      void __user *buffer, size_t *length,
829                      loff_t *ppos)
830{
831    int ret;
832
833    mutex_lock(&kprobe_mutex);
834    sysctl_kprobes_optimization = kprobes_allow_optimization ? 1 : 0;
835    ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
836
837    if (sysctl_kprobes_optimization)
838        optimize_all_kprobes();
839    else
840        unoptimize_all_kprobes();
841    mutex_unlock(&kprobe_mutex);
842
843    return ret;
844}
845#endif /* CONFIG_SYSCTL */
846
847/* Put a breakpoint for a probe. Must be called with text_mutex locked */
848static void __kprobes __arm_kprobe(struct kprobe *p)
849{
850    struct kprobe *_p;
851
852    /* Check collision with other optimized kprobes */
853    _p = get_optimized_kprobe((unsigned long)p->addr);
854    if (unlikely(_p))
855        /* Fallback to unoptimized kprobe */
856        unoptimize_kprobe(_p, true);
857
858    arch_arm_kprobe(p);
859    optimize_kprobe(p); /* Try to optimize (add kprobe to a list) */
860}
861
862/* Remove the breakpoint of a probe. Must be called with text_mutex locked */
863static void __kprobes __disarm_kprobe(struct kprobe *p, bool reopt)
864{
865    struct kprobe *_p;
866
867    unoptimize_kprobe(p, false); /* Try to unoptimize */
868
869    if (!kprobe_queued(p)) {
870        arch_disarm_kprobe(p);
871        /* If another kprobe was blocked, optimize it. */
872        _p = get_optimized_kprobe((unsigned long)p->addr);
873        if (unlikely(_p) && reopt)
874            optimize_kprobe(_p);
875    }
876    /* TODO: reoptimize others after unoptimized this probe */
877}
878
879#else /* !CONFIG_OPTPROBES */
880
881#define optimize_kprobe(p) do {} while (0)
882#define unoptimize_kprobe(p, f) do {} while (0)
883#define kill_optimized_kprobe(p) do {} while (0)
884#define prepare_optimized_kprobe(p) do {} while (0)
885#define try_to_optimize_kprobe(p) do {} while (0)
886#define __arm_kprobe(p) arch_arm_kprobe(p)
887#define __disarm_kprobe(p, o) arch_disarm_kprobe(p)
888#define kprobe_disarmed(p) kprobe_disabled(p)
889#define wait_for_kprobe_optimizer() do {} while (0)
890
891/* There should be no unused kprobes can be reused without optimization */
892static void reuse_unused_kprobe(struct kprobe *ap)
893{
894    printk(KERN_ERR "Error: There should be no unused kprobe here.\n");
895    BUG_ON(kprobe_unused(ap));
896}
897
898static __kprobes void free_aggr_kprobe(struct kprobe *p)
899{
900    arch_remove_kprobe(p);
901    kfree(p);
902}
903
904static __kprobes struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
905{
906    return kzalloc(sizeof(struct kprobe), GFP_KERNEL);
907}
908#endif /* CONFIG_OPTPROBES */
909
910/* Arm a kprobe with text_mutex */
911static void __kprobes arm_kprobe(struct kprobe *kp)
912{
913    /*
914     * Here, since __arm_kprobe() doesn't use stop_machine(),
915     * this doesn't cause deadlock on text_mutex. So, we don't
916     * need get_online_cpus().
917     */
918    mutex_lock(&text_mutex);
919    __arm_kprobe(kp);
920    mutex_unlock(&text_mutex);
921}
922
923/* Disarm a kprobe with text_mutex */
924static void __kprobes disarm_kprobe(struct kprobe *kp)
925{
926    /* Ditto */
927    mutex_lock(&text_mutex);
928    __disarm_kprobe(kp, true);
929    mutex_unlock(&text_mutex);
930}
931
932/*
933 * Aggregate handlers for multiple kprobes support - these handlers
934 * take care of invoking the individual kprobe handlers on p->list
935 */
936static int __kprobes aggr_pre_handler(struct kprobe *p, struct pt_regs *regs)
937{
938    struct kprobe *kp;
939
940    list_for_each_entry_rcu(kp, &p->list, list) {
941        if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
942            set_kprobe_instance(kp);
943            if (kp->pre_handler(kp, regs))
944                return 1;
945        }
946        reset_kprobe_instance();
947    }
948    return 0;
949}
950
951static void __kprobes aggr_post_handler(struct kprobe *p, struct pt_regs *regs,
952                    unsigned long flags)
953{
954    struct kprobe *kp;
955
956    list_for_each_entry_rcu(kp, &p->list, list) {
957        if (kp->post_handler && likely(!kprobe_disabled(kp))) {
958            set_kprobe_instance(kp);
959            kp->post_handler(kp, regs, flags);
960            reset_kprobe_instance();
961        }
962    }
963}
964
965static int __kprobes aggr_fault_handler(struct kprobe *p, struct pt_regs *regs,
966                    int trapnr)
967{
968    struct kprobe *cur = __this_cpu_read(kprobe_instance);
969
970    /*
971     * if we faulted "during" the execution of a user specified
972     * probe handler, invoke just that probe's fault handler
973     */
974    if (cur && cur->fault_handler) {
975        if (cur->fault_handler(cur, regs, trapnr))
976            return 1;
977    }
978    return 0;
979}
980
981static int __kprobes aggr_break_handler(struct kprobe *p, struct pt_regs *regs)
982{
983    struct kprobe *cur = __this_cpu_read(kprobe_instance);
984    int ret = 0;
985
986    if (cur && cur->break_handler) {
987        if (cur->break_handler(cur, regs))
988            ret = 1;
989    }
990    reset_kprobe_instance();
991    return ret;
992}
993
994/* Walks the list and increments nmissed count for multiprobe case */
995void __kprobes kprobes_inc_nmissed_count(struct kprobe *p)
996{
997    struct kprobe *kp;
998    if (!kprobe_aggrprobe(p)) {
999        p->nmissed++;
1000    } else {
1001        list_for_each_entry_rcu(kp, &p->list, list)
1002            kp->nmissed++;
1003    }
1004    return;
1005}
1006
1007void __kprobes recycle_rp_inst(struct kretprobe_instance *ri,
1008                struct hlist_head *head)
1009{
1010    struct kretprobe *rp = ri->rp;
1011
1012    /* remove rp inst off the rprobe_inst_table */
1013    hlist_del(&ri->hlist);
1014    INIT_HLIST_NODE(&ri->hlist);
1015    if (likely(rp)) {
1016        spin_lock(&rp->lock);
1017        hlist_add_head(&ri->hlist, &rp->free_instances);
1018        spin_unlock(&rp->lock);
1019    } else
1020        /* Unregistering */
1021        hlist_add_head(&ri->hlist, head);
1022}
1023
1024void __kprobes kretprobe_hash_lock(struct task_struct *tsk,
1025             struct hlist_head **head, unsigned long *flags)
1026__acquires(hlist_lock)
1027{
1028    unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
1029    spinlock_t *hlist_lock;
1030
1031    *head = &kretprobe_inst_table[hash];
1032    hlist_lock = kretprobe_table_lock_ptr(hash);
1033    spin_lock_irqsave(hlist_lock, *flags);
1034}
1035
1036static void __kprobes kretprobe_table_lock(unsigned long hash,
1037    unsigned long *flags)
1038__acquires(hlist_lock)
1039{
1040    spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
1041    spin_lock_irqsave(hlist_lock, *flags);
1042}
1043
1044void __kprobes kretprobe_hash_unlock(struct task_struct *tsk,
1045    unsigned long *flags)
1046__releases(hlist_lock)
1047{
1048    unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
1049    spinlock_t *hlist_lock;
1050
1051    hlist_lock = kretprobe_table_lock_ptr(hash);
1052    spin_unlock_irqrestore(hlist_lock, *flags);
1053}
1054
1055static void __kprobes kretprobe_table_unlock(unsigned long hash,
1056       unsigned long *flags)
1057__releases(hlist_lock)
1058{
1059    spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
1060    spin_unlock_irqrestore(hlist_lock, *flags);
1061}
1062
1063/*
1064 * This function is called from finish_task_switch when task tk becomes dead,
1065 * so that we can recycle any function-return probe instances associated
1066 * with this task. These left over instances represent probed functions
1067 * that have been called but will never return.
1068 */
1069void __kprobes kprobe_flush_task(struct task_struct *tk)
1070{
1071    struct kretprobe_instance *ri;
1072    struct hlist_head *head, empty_rp;
1073    struct hlist_node *node, *tmp;
1074    unsigned long hash, flags = 0;
1075
1076    if (unlikely(!kprobes_initialized))
1077        /* Early boot. kretprobe_table_locks not yet initialized. */
1078        return;
1079
1080    hash = hash_ptr(tk, KPROBE_HASH_BITS);
1081    head = &kretprobe_inst_table[hash];
1082    kretprobe_table_lock(hash, &flags);
1083    hlist_for_each_entry_safe(ri, node, tmp, head, hlist) {
1084        if (ri->task == tk)
1085            recycle_rp_inst(ri, &empty_rp);
1086    }
1087    kretprobe_table_unlock(hash, &flags);
1088    INIT_HLIST_HEAD(&empty_rp);
1089    hlist_for_each_entry_safe(ri, node, tmp, &empty_rp, hlist) {
1090        hlist_del(&ri->hlist);
1091        kfree(ri);
1092    }
1093}
1094
1095static inline void free_rp_inst(struct kretprobe *rp)
1096{
1097    struct kretprobe_instance *ri;
1098    struct hlist_node *pos, *next;
1099
1100    hlist_for_each_entry_safe(ri, pos, next, &rp->free_instances, hlist) {
1101        hlist_del(&ri->hlist);
1102        kfree(ri);
1103    }
1104}
1105
1106static void __kprobes cleanup_rp_inst(struct kretprobe *rp)
1107{
1108    unsigned long flags, hash;
1109    struct kretprobe_instance *ri;
1110    struct hlist_node *pos, *next;
1111    struct hlist_head *head;
1112
1113    /* No race here */
1114    for (hash = 0; hash < KPROBE_TABLE_SIZE; hash++) {
1115        kretprobe_table_lock(hash, &flags);
1116        head = &kretprobe_inst_table[hash];
1117        hlist_for_each_entry_safe(ri, pos, next, head, hlist) {
1118            if (ri->rp == rp)
1119                ri->rp = NULL;
1120        }
1121        kretprobe_table_unlock(hash, &flags);
1122    }
1123    free_rp_inst(rp);
1124}
1125
1126/*
1127* Add the new probe to ap->list. Fail if this is the
1128* second jprobe at the address - two jprobes can't coexist
1129*/
1130static int __kprobes add_new_kprobe(struct kprobe *ap, struct kprobe *p)
1131{
1132    BUG_ON(kprobe_gone(ap) || kprobe_gone(p));
1133
1134    if (p->break_handler || p->post_handler)
1135        unoptimize_kprobe(ap, true); /* Fall back to normal kprobe */
1136
1137    if (p->break_handler) {
1138        if (ap->break_handler)
1139            return -EEXIST;
1140        list_add_tail_rcu(&p->list, &ap->list);
1141        ap->break_handler = aggr_break_handler;
1142    } else
1143        list_add_rcu(&p->list, &ap->list);
1144    if (p->post_handler && !ap->post_handler)
1145        ap->post_handler = aggr_post_handler;
1146
1147    if (kprobe_disabled(ap) && !kprobe_disabled(p)) {
1148        ap->flags &= ~KPROBE_FLAG_DISABLED;
1149        if (!kprobes_all_disarmed)
1150            /* Arm the breakpoint again. */
1151            __arm_kprobe(ap);
1152    }
1153    return 0;
1154}
1155
1156/*
1157 * Fill in the required fields of the "manager kprobe". Replace the
1158 * earlier kprobe in the hlist with the manager kprobe
1159 */
1160static void __kprobes init_aggr_kprobe(struct kprobe *ap, struct kprobe *p)
1161{
1162    /* Copy p's insn slot to ap */
1163    copy_kprobe(p, ap);
1164    flush_insn_slot(ap);
1165    ap->addr = p->addr;
1166    ap->flags = p->flags & ~KPROBE_FLAG_OPTIMIZED;
1167    ap->pre_handler = aggr_pre_handler;
1168    ap->fault_handler = aggr_fault_handler;
1169    /* We don't care the kprobe which has gone. */
1170    if (p->post_handler && !kprobe_gone(p))
1171        ap->post_handler = aggr_post_handler;
1172    if (p->break_handler && !kprobe_gone(p))
1173        ap->break_handler = aggr_break_handler;
1174
1175    INIT_LIST_HEAD(&ap->list);
1176    INIT_HLIST_NODE(&ap->hlist);
1177
1178    list_add_rcu(&p->list, &ap->list);
1179    hlist_replace_rcu(&p->hlist, &ap->hlist);
1180}
1181
1182/*
1183 * This is the second or subsequent kprobe at the address - handle
1184 * the intricacies
1185 */
1186static int __kprobes register_aggr_kprobe(struct kprobe *orig_p,
1187                      struct kprobe *p)
1188{
1189    int ret = 0;
1190    struct kprobe *ap = orig_p;
1191
1192    if (!kprobe_aggrprobe(orig_p)) {
1193        /* If orig_p is not an aggr_kprobe, create new aggr_kprobe. */
1194        ap = alloc_aggr_kprobe(orig_p);
1195        if (!ap)
1196            return -ENOMEM;
1197        init_aggr_kprobe(ap, orig_p);
1198    } else if (kprobe_unused(ap))
1199        /* This probe is going to die. Rescue it */
1200        reuse_unused_kprobe(ap);
1201
1202    if (kprobe_gone(ap)) {
1203        /*
1204         * Attempting to insert new probe at the same location that
1205         * had a probe in the module vaddr area which already
1206         * freed. So, the instruction slot has already been
1207         * released. We need a new slot for the new probe.
1208         */
1209        ret = arch_prepare_kprobe(ap);
1210        if (ret)
1211            /*
1212             * Even if fail to allocate new slot, don't need to
1213             * free aggr_probe. It will be used next time, or
1214             * freed by unregister_kprobe.
1215             */
1216            return ret;
1217
1218        /* Prepare optimized instructions if possible. */
1219        prepare_optimized_kprobe(ap);
1220
1221        /*
1222         * Clear gone flag to prevent allocating new slot again, and
1223         * set disabled flag because it is not armed yet.
1224         */
1225        ap->flags = (ap->flags & ~KPROBE_FLAG_GONE)
1226                | KPROBE_FLAG_DISABLED;
1227    }
1228
1229    /* Copy ap's insn slot to p */
1230    copy_kprobe(ap, p);
1231    return add_new_kprobe(ap, p);
1232}
1233
1234static int __kprobes in_kprobes_functions(unsigned long addr)
1235{
1236    struct kprobe_blackpoint *kb;
1237
1238    if (addr >= (unsigned long)__kprobes_text_start &&
1239        addr < (unsigned long)__kprobes_text_end)
1240        return -EINVAL;
1241    /*
1242     * If there exists a kprobe_blacklist, verify and
1243     * fail any probe registration in the prohibited area
1244     */
1245    for (kb = kprobe_blacklist; kb->name != NULL; kb++) {
1246        if (kb->start_addr) {
1247            if (addr >= kb->start_addr &&
1248                addr < (kb->start_addr + kb->range))
1249                return -EINVAL;
1250        }
1251    }
1252    return 0;
1253}
1254
1255/*
1256 * If we have a symbol_name argument, look it up and add the offset field
1257 * to it. This way, we can specify a relative address to a symbol.
1258 */
1259static kprobe_opcode_t __kprobes *kprobe_addr(struct kprobe *p)
1260{
1261    kprobe_opcode_t *addr = p->addr;
1262    if (p->symbol_name) {
1263        if (addr)
1264            return NULL;
1265        kprobe_lookup_name(p->symbol_name, addr);
1266    }
1267
1268    if (!addr)
1269        return NULL;
1270    return (kprobe_opcode_t *)(((char *)addr) + p->offset);
1271}
1272
1273/* Check passed kprobe is valid and return kprobe in kprobe_table. */
1274static struct kprobe * __kprobes __get_valid_kprobe(struct kprobe *p)
1275{
1276    struct kprobe *ap, *list_p;
1277
1278    ap = get_kprobe(p->addr);
1279    if (unlikely(!ap))
1280        return NULL;
1281
1282    if (p != ap) {
1283        list_for_each_entry_rcu(list_p, &ap->list, list)
1284            if (list_p == p)
1285            /* kprobe p is a valid probe */
1286                goto valid;
1287        return NULL;
1288    }
1289valid:
1290    return ap;
1291}
1292
1293/* Return error if the kprobe is being re-registered */
1294static inline int check_kprobe_rereg(struct kprobe *p)
1295{
1296    int ret = 0;
1297
1298    mutex_lock(&kprobe_mutex);
1299    if (__get_valid_kprobe(p))
1300        ret = -EINVAL;
1301    mutex_unlock(&kprobe_mutex);
1302
1303    return ret;
1304}
1305
1306int __kprobes register_kprobe(struct kprobe *p)
1307{
1308    int ret = 0;
1309    struct kprobe *old_p;
1310    struct module *probed_mod;
1311    kprobe_opcode_t *addr;
1312
1313    addr = kprobe_addr(p);
1314    if (!addr)
1315        return -EINVAL;
1316    p->addr = addr;
1317
1318    ret = check_kprobe_rereg(p);
1319    if (ret)
1320        return ret;
1321
1322    jump_label_lock();
1323    preempt_disable();
1324    if (!kernel_text_address((unsigned long) p->addr) ||
1325        in_kprobes_functions((unsigned long) p->addr) ||
1326        ftrace_text_reserved(p->addr, p->addr) ||
1327        jump_label_text_reserved(p->addr, p->addr))
1328        goto fail_with_jump_label;
1329
1330    /* User can pass only KPROBE_FLAG_DISABLED to register_kprobe */
1331    p->flags &= KPROBE_FLAG_DISABLED;
1332
1333    /*
1334     * Check if are we probing a module.
1335     */
1336    probed_mod = __module_text_address((unsigned long) p->addr);
1337    if (probed_mod) {
1338        /*
1339         * We must hold a refcount of the probed module while updating
1340         * its code to prohibit unexpected unloading.
1341         */
1342        if (unlikely(!try_module_get(probed_mod)))
1343            goto fail_with_jump_label;
1344
1345        /*
1346         * If the module freed .init.text, we couldn't insert
1347         * kprobes in there.
1348         */
1349        if (within_module_init((unsigned long)p->addr, probed_mod) &&
1350            probed_mod->state != MODULE_STATE_COMING) {
1351            module_put(probed_mod);
1352            goto fail_with_jump_label;
1353        }
1354    }
1355    preempt_enable();
1356    jump_label_unlock();
1357
1358    p->nmissed = 0;
1359    INIT_LIST_HEAD(&p->list);
1360    mutex_lock(&kprobe_mutex);
1361
1362    jump_label_lock(); /* needed to call jump_label_text_reserved() */
1363
1364    get_online_cpus(); /* For avoiding text_mutex deadlock. */
1365    mutex_lock(&text_mutex);
1366
1367    old_p = get_kprobe(p->addr);
1368    if (old_p) {
1369        /* Since this may unoptimize old_p, locking text_mutex. */
1370        ret = register_aggr_kprobe(old_p, p);
1371        goto out;
1372    }
1373
1374    ret = arch_prepare_kprobe(p);
1375    if (ret)
1376        goto out;
1377
1378    INIT_HLIST_NODE(&p->hlist);
1379    hlist_add_head_rcu(&p->hlist,
1380               &kprobe_table[hash_ptr(p->addr, KPROBE_HASH_BITS)]);
1381
1382    if (!kprobes_all_disarmed && !kprobe_disabled(p))
1383        __arm_kprobe(p);
1384
1385    /* Try to optimize kprobe */
1386    try_to_optimize_kprobe(p);
1387
1388out:
1389    mutex_unlock(&text_mutex);
1390    put_online_cpus();
1391    jump_label_unlock();
1392    mutex_unlock(&kprobe_mutex);
1393
1394    if (probed_mod)
1395        module_put(probed_mod);
1396
1397    return ret;
1398
1399fail_with_jump_label:
1400    preempt_enable();
1401    jump_label_unlock();
1402    return -EINVAL;
1403}
1404EXPORT_SYMBOL_GPL(register_kprobe);
1405
1406/* Check if all probes on the aggrprobe are disabled */
1407static int __kprobes aggr_kprobe_disabled(struct kprobe *ap)
1408{
1409    struct kprobe *kp;
1410
1411    list_for_each_entry_rcu(kp, &ap->list, list)
1412        if (!kprobe_disabled(kp))
1413            /*
1414             * There is an active probe on the list.
1415             * We can't disable this ap.
1416             */
1417            return 0;
1418
1419    return 1;
1420}
1421
1422/* Disable one kprobe: Make sure called under kprobe_mutex is locked */
1423static struct kprobe *__kprobes __disable_kprobe(struct kprobe *p)
1424{
1425    struct kprobe *orig_p;
1426
1427    /* Get an original kprobe for return */
1428    orig_p = __get_valid_kprobe(p);
1429    if (unlikely(orig_p == NULL))
1430        return NULL;
1431
1432    if (!kprobe_disabled(p)) {
1433        /* Disable probe if it is a child probe */
1434        if (p != orig_p)
1435            p->flags |= KPROBE_FLAG_DISABLED;
1436
1437        /* Try to disarm and disable this/parent probe */
1438        if (p == orig_p || aggr_kprobe_disabled(orig_p)) {
1439            disarm_kprobe(orig_p);
1440            orig_p->flags |= KPROBE_FLAG_DISABLED;
1441        }
1442    }
1443
1444    return orig_p;
1445}
1446
1447/*
1448 * Unregister a kprobe without a scheduler synchronization.
1449 */
1450static int __kprobes __unregister_kprobe_top(struct kprobe *p)
1451{
1452    struct kprobe *ap, *list_p;
1453
1454    /* Disable kprobe. This will disarm it if needed. */
1455    ap = __disable_kprobe(p);
1456    if (ap == NULL)
1457        return -EINVAL;
1458
1459    if (ap == p)
1460        /*
1461         * This probe is an independent(and non-optimized) kprobe
1462         * (not an aggrprobe). Remove from the hash list.
1463         */
1464        goto disarmed;
1465
1466    /* Following process expects this probe is an aggrprobe */
1467    WARN_ON(!kprobe_aggrprobe(ap));
1468
1469    if (list_is_singular(&ap->list) && kprobe_disarmed(ap))
1470        /*
1471         * !disarmed could be happen if the probe is under delayed
1472         * unoptimizing.
1473         */
1474        goto disarmed;
1475    else {
1476        /* If disabling probe has special handlers, update aggrprobe */
1477        if (p->break_handler && !kprobe_gone(p))
1478            ap->break_handler = NULL;
1479        if (p->post_handler && !kprobe_gone(p)) {
1480            list_for_each_entry_rcu(list_p, &ap->list, list) {
1481                if ((list_p != p) && (list_p->post_handler))
1482                    goto noclean;
1483            }
1484            ap->post_handler = NULL;
1485        }
1486noclean:
1487        /*
1488         * Remove from the aggrprobe: this path will do nothing in
1489         * __unregister_kprobe_bottom().
1490         */
1491        list_del_rcu(&p->list);
1492        if (!kprobe_disabled(ap) && !kprobes_all_disarmed)
1493            /*
1494             * Try to optimize this probe again, because post
1495             * handler may have been changed.
1496             */
1497            optimize_kprobe(ap);
1498    }
1499    return 0;
1500
1501disarmed:
1502    BUG_ON(!kprobe_disarmed(ap));
1503    hlist_del_rcu(&ap->hlist);
1504    return 0;
1505}
1506
1507static void __kprobes __unregister_kprobe_bottom(struct kprobe *p)
1508{
1509    struct kprobe *ap;
1510
1511    if (list_empty(&p->list))
1512        /* This is an independent kprobe */
1513        arch_remove_kprobe(p);
1514    else if (list_is_singular(&p->list)) {
1515        /* This is the last child of an aggrprobe */
1516        ap = list_entry(p->list.next, struct kprobe, list);
1517        list_del(&p->list);
1518        free_aggr_kprobe(ap);
1519    }
1520    /* Otherwise, do nothing. */
1521}
1522
1523int __kprobes register_kprobes(struct kprobe **kps, int num)
1524{
1525    int i, ret = 0;
1526
1527    if (num <= 0)
1528        return -EINVAL;
1529    for (i = 0; i < num; i++) {
1530        ret = register_kprobe(kps[i]);
1531        if (ret < 0) {
1532            if (i > 0)
1533                unregister_kprobes(kps, i);
1534            break;
1535        }
1536    }
1537    return ret;
1538}
1539EXPORT_SYMBOL_GPL(register_kprobes);
1540
1541void __kprobes unregister_kprobe(struct kprobe *p)
1542{
1543    unregister_kprobes(&p, 1);
1544}
1545EXPORT_SYMBOL_GPL(unregister_kprobe);
1546
1547void __kprobes unregister_kprobes(struct kprobe **kps, int num)
1548{
1549    int i;
1550
1551    if (num <= 0)
1552        return;
1553    mutex_lock(&kprobe_mutex);
1554    for (i = 0; i < num; i++)
1555        if (__unregister_kprobe_top(kps[i]) < 0)
1556            kps[i]->addr = NULL;
1557    mutex_unlock(&kprobe_mutex);
1558
1559    synchronize_sched();
1560    for (i = 0; i < num; i++)
1561        if (kps[i]->addr)
1562            __unregister_kprobe_bottom(kps[i]);
1563}
1564EXPORT_SYMBOL_GPL(unregister_kprobes);
1565
1566static struct notifier_block kprobe_exceptions_nb = {
1567    .notifier_call = kprobe_exceptions_notify,
1568    .priority = 0x7fffffff /* we need to be notified first */
1569};
1570
1571unsigned long __weak arch_deref_entry_point(void *entry)
1572{
1573    return (unsigned long)entry;
1574}
1575
1576int __kprobes register_jprobes(struct jprobe **jps, int num)
1577{
1578    struct jprobe *jp;
1579    int ret = 0, i;
1580
1581    if (num <= 0)
1582        return -EINVAL;
1583    for (i = 0; i < num; i++) {
1584        unsigned long addr, offset;
1585        jp = jps[i];
1586        addr = arch_deref_entry_point(jp->entry);
1587
1588        /* Verify probepoint is a function entry point */
1589        if (kallsyms_lookup_size_offset(addr, NULL, &offset) &&
1590            offset == 0) {
1591            jp->kp.pre_handler = setjmp_pre_handler;
1592            jp->kp.break_handler = longjmp_break_handler;
1593            ret = register_kprobe(&jp->kp);
1594        } else
1595            ret = -EINVAL;
1596
1597        if (ret < 0) {
1598            if (i > 0)
1599                unregister_jprobes(jps, i);
1600            break;
1601        }
1602    }
1603    return ret;
1604}
1605EXPORT_SYMBOL_GPL(register_jprobes);
1606
1607int __kprobes register_jprobe(struct jprobe *jp)
1608{
1609    return register_jprobes(&jp, 1);
1610}
1611EXPORT_SYMBOL_GPL(register_jprobe);
1612
1613void __kprobes unregister_jprobe(struct jprobe *jp)
1614{
1615    unregister_jprobes(&jp, 1);
1616}
1617EXPORT_SYMBOL_GPL(unregister_jprobe);
1618
1619void __kprobes unregister_jprobes(struct jprobe **jps, int num)
1620{
1621    int i;
1622
1623    if (num <= 0)
1624        return;
1625    mutex_lock(&kprobe_mutex);
1626    for (i = 0; i < num; i++)
1627        if (__unregister_kprobe_top(&jps[i]->kp) < 0)
1628            jps[i]->kp.addr = NULL;
1629    mutex_unlock(&kprobe_mutex);
1630
1631    synchronize_sched();
1632    for (i = 0; i < num; i++) {
1633        if (jps[i]->kp.addr)
1634            __unregister_kprobe_bottom(&jps[i]->kp);
1635    }
1636}
1637EXPORT_SYMBOL_GPL(unregister_jprobes);
1638
1639#ifdef CONFIG_KRETPROBES
1640/*
1641 * This kprobe pre_handler is registered with every kretprobe. When probe
1642 * hits it will set up the return probe.
1643 */
1644static int __kprobes pre_handler_kretprobe(struct kprobe *p,
1645                       struct pt_regs *regs)
1646{
1647    struct kretprobe *rp = container_of(p, struct kretprobe, kp);
1648    unsigned long hash, flags = 0;
1649    struct kretprobe_instance *ri;
1650
1651    /*TODO: consider to only swap the RA after the last pre_handler fired */
1652    hash = hash_ptr(current, KPROBE_HASH_BITS);
1653    spin_lock_irqsave(&rp->lock, flags);
1654    if (!hlist_empty(&rp->free_instances)) {
1655        ri = hlist_entry(rp->free_instances.first,
1656                struct kretprobe_instance, hlist);
1657        hlist_del(&ri->hlist);
1658        spin_unlock_irqrestore(&rp->lock, flags);
1659
1660        ri->rp = rp;
1661        ri->task = current;
1662
1663        if (rp->entry_handler && rp->entry_handler(ri, regs))
1664            return 0;
1665
1666        arch_prepare_kretprobe(ri, regs);
1667
1668        /* XXX(hch): why is there no hlist_move_head? */
1669        INIT_HLIST_NODE(&ri->hlist);
1670        kretprobe_table_lock(hash, &flags);
1671        hlist_add_head(&ri->hlist, &kretprobe_inst_table[hash]);
1672        kretprobe_table_unlock(hash, &flags);
1673    } else {
1674        rp->nmissed++;
1675        spin_unlock_irqrestore(&rp->lock, flags);
1676    }
1677    return 0;
1678}
1679
1680int __kprobes register_kretprobe(struct kretprobe *rp)
1681{
1682    int ret = 0;
1683    struct kretprobe_instance *inst;
1684    int i;
1685    void *addr;
1686
1687    if (kretprobe_blacklist_size) {
1688        addr = kprobe_addr(&rp->kp);
1689        if (!addr)
1690            return -EINVAL;
1691
1692        for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
1693            if (kretprobe_blacklist[i].addr == addr)
1694                return -EINVAL;
1695        }
1696    }
1697
1698    rp->kp.pre_handler = pre_handler_kretprobe;
1699    rp->kp.post_handler = NULL;
1700    rp->kp.fault_handler = NULL;
1701    rp->kp.break_handler = NULL;
1702
1703    /* Pre-allocate memory for max kretprobe instances */
1704    if (rp->maxactive <= 0) {
1705#ifdef CONFIG_PREEMPT
1706        rp->maxactive = max_t(unsigned int, 10, 2*num_possible_cpus());
1707#else
1708        rp->maxactive = num_possible_cpus();
1709#endif
1710    }
1711    spin_lock_init(&rp->lock);
1712    INIT_HLIST_HEAD(&rp->free_instances);
1713    for (i = 0; i < rp->maxactive; i++) {
1714        inst = kmalloc(sizeof(struct kretprobe_instance) +
1715                   rp->data_size, GFP_KERNEL);
1716        if (inst == NULL) {
1717            free_rp_inst(rp);
1718            return -ENOMEM;
1719        }
1720        INIT_HLIST_NODE(&inst->hlist);
1721        hlist_add_head(&inst->hlist, &rp->free_instances);
1722    }
1723
1724    rp->nmissed = 0;
1725    /* Establish function entry probe point */
1726    ret = register_kprobe(&rp->kp);
1727    if (ret != 0)
1728        free_rp_inst(rp);
1729    return ret;
1730}
1731EXPORT_SYMBOL_GPL(register_kretprobe);
1732
1733int __kprobes register_kretprobes(struct kretprobe **rps, int num)
1734{
1735    int ret = 0, i;
1736
1737    if (num <= 0)
1738        return -EINVAL;
1739    for (i = 0; i < num; i++) {
1740        ret = register_kretprobe(rps[i]);
1741        if (ret < 0) {
1742            if (i > 0)
1743                unregister_kretprobes(rps, i);
1744            break;
1745        }
1746    }
1747    return ret;
1748}
1749EXPORT_SYMBOL_GPL(register_kretprobes);
1750
1751void __kprobes unregister_kretprobe(struct kretprobe *rp)
1752{
1753    unregister_kretprobes(&rp, 1);
1754}
1755EXPORT_SYMBOL_GPL(unregister_kretprobe);
1756
1757void __kprobes unregister_kretprobes(struct kretprobe **rps, int num)
1758{
1759    int i;
1760
1761    if (num <= 0)
1762        return;
1763    mutex_lock(&kprobe_mutex);
1764    for (i = 0; i < num; i++)
1765        if (__unregister_kprobe_top(&rps[i]->kp) < 0)
1766            rps[i]->kp.addr = NULL;
1767    mutex_unlock(&kprobe_mutex);
1768
1769    synchronize_sched();
1770    for (i = 0; i < num; i++) {
1771        if (rps[i]->kp.addr) {
1772            __unregister_kprobe_bottom(&rps[i]->kp);
1773            cleanup_rp_inst(rps[i]);
1774        }
1775    }
1776}
1777EXPORT_SYMBOL_GPL(unregister_kretprobes);
1778
1779#else /* CONFIG_KRETPROBES */
1780int __kprobes register_kretprobe(struct kretprobe *rp)
1781{
1782    return -ENOSYS;
1783}
1784EXPORT_SYMBOL_GPL(register_kretprobe);
1785
1786int __kprobes register_kretprobes(struct kretprobe **rps, int num)
1787{
1788    return -ENOSYS;
1789}
1790EXPORT_SYMBOL_GPL(register_kretprobes);
1791
1792void __kprobes unregister_kretprobe(struct kretprobe *rp)
1793{
1794}
1795EXPORT_SYMBOL_GPL(unregister_kretprobe);
1796
1797void __kprobes unregister_kretprobes(struct kretprobe **rps, int num)
1798{
1799}
1800EXPORT_SYMBOL_GPL(unregister_kretprobes);
1801
1802static int __kprobes pre_handler_kretprobe(struct kprobe *p,
1803                       struct pt_regs *regs)
1804{
1805    return 0;
1806}
1807
1808#endif /* CONFIG_KRETPROBES */
1809
1810/* Set the kprobe gone and remove its instruction buffer. */
1811static void __kprobes kill_kprobe(struct kprobe *p)
1812{
1813    struct kprobe *kp;
1814
1815    p->flags |= KPROBE_FLAG_GONE;
1816    if (kprobe_aggrprobe(p)) {
1817        /*
1818         * If this is an aggr_kprobe, we have to list all the
1819         * chained probes and mark them GONE.
1820         */
1821        list_for_each_entry_rcu(kp, &p->list, list)
1822            kp->flags |= KPROBE_FLAG_GONE;
1823        p->post_handler = NULL;
1824        p->break_handler = NULL;
1825        kill_optimized_kprobe(p);
1826    }
1827    /*
1828     * Here, we can remove insn_slot safely, because no thread calls
1829     * the original probed function (which will be freed soon) any more.
1830     */
1831    arch_remove_kprobe(p);
1832}
1833
1834/* Disable one kprobe */
1835int __kprobes disable_kprobe(struct kprobe *kp)
1836{
1837    int ret = 0;
1838
1839    mutex_lock(&kprobe_mutex);
1840
1841    /* Disable this kprobe */
1842    if (__disable_kprobe(kp) == NULL)
1843        ret = -EINVAL;
1844
1845    mutex_unlock(&kprobe_mutex);
1846    return ret;
1847}
1848EXPORT_SYMBOL_GPL(disable_kprobe);
1849
1850/* Enable one kprobe */
1851int __kprobes enable_kprobe(struct kprobe *kp)
1852{
1853    int ret = 0;
1854    struct kprobe *p;
1855
1856    mutex_lock(&kprobe_mutex);
1857
1858    /* Check whether specified probe is valid. */
1859    p = __get_valid_kprobe(kp);
1860    if (unlikely(p == NULL)) {
1861        ret = -EINVAL;
1862        goto out;
1863    }
1864
1865    if (kprobe_gone(kp)) {
1866        /* This kprobe has gone, we couldn't enable it. */
1867        ret = -EINVAL;
1868        goto out;
1869    }
1870
1871    if (p != kp)
1872        kp->flags &= ~KPROBE_FLAG_DISABLED;
1873
1874    if (!kprobes_all_disarmed && kprobe_disabled(p)) {
1875        p->flags &= ~KPROBE_FLAG_DISABLED;
1876        arm_kprobe(p);
1877    }
1878out:
1879    mutex_unlock(&kprobe_mutex);
1880    return ret;
1881}
1882EXPORT_SYMBOL_GPL(enable_kprobe);
1883
1884void __kprobes dump_kprobe(struct kprobe *kp)
1885{
1886    printk(KERN_WARNING "Dumping kprobe:\n");
1887    printk(KERN_WARNING "Name: %s\nAddress: %p\nOffset: %x\n",
1888           kp->symbol_name, kp->addr, kp->offset);
1889}
1890
1891/* Module notifier call back, checking kprobes on the module */
1892static int __kprobes kprobes_module_callback(struct notifier_block *nb,
1893                         unsigned long val, void *data)
1894{
1895    struct module *mod = data;
1896    struct hlist_head *head;
1897    struct hlist_node *node;
1898    struct kprobe *p;
1899    unsigned int i;
1900    int checkcore = (val == MODULE_STATE_GOING);
1901
1902    if (val != MODULE_STATE_GOING && val != MODULE_STATE_LIVE)
1903        return NOTIFY_DONE;
1904
1905    /*
1906     * When MODULE_STATE_GOING was notified, both of module .text and
1907     * .init.text sections would be freed. When MODULE_STATE_LIVE was
1908     * notified, only .init.text section would be freed. We need to
1909     * disable kprobes which have been inserted in the sections.
1910     */
1911    mutex_lock(&kprobe_mutex);
1912    for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
1913        head = &kprobe_table[i];
1914        hlist_for_each_entry_rcu(p, node, head, hlist)
1915            if (within_module_init((unsigned long)p->addr, mod) ||
1916                (checkcore &&
1917                 within_module_core((unsigned long)p->addr, mod))) {
1918                /*
1919                 * The vaddr this probe is installed will soon
1920                 * be vfreed buy not synced to disk. Hence,
1921                 * disarming the breakpoint isn't needed.
1922                 */
1923                kill_kprobe(p);
1924            }
1925    }
1926    mutex_unlock(&kprobe_mutex);
1927    return NOTIFY_DONE;
1928}
1929
1930static struct notifier_block kprobe_module_nb = {
1931    .notifier_call = kprobes_module_callback,
1932    .priority = 0
1933};
1934
1935static int __init init_kprobes(void)
1936{
1937    int i, err = 0;
1938    unsigned long offset = 0, size = 0;
1939    char *modname, namebuf[128];
1940    const char *symbol_name;
1941    void *addr;
1942    struct kprobe_blackpoint *kb;
1943
1944    /* FIXME allocate the probe table, currently defined statically */
1945    /* initialize all list heads */
1946    for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
1947        INIT_HLIST_HEAD(&kprobe_table[i]);
1948        INIT_HLIST_HEAD(&kretprobe_inst_table[i]);
1949        spin_lock_init(&(kretprobe_table_locks[i].lock));
1950    }
1951
1952    /*
1953     * Lookup and populate the kprobe_blacklist.
1954     *
1955     * Unlike the kretprobe blacklist, we'll need to determine
1956     * the range of addresses that belong to the said functions,
1957     * since a kprobe need not necessarily be at the beginning
1958     * of a function.
1959     */
1960    for (kb = kprobe_blacklist; kb->name != NULL; kb++) {
1961        kprobe_lookup_name(kb->name, addr);
1962        if (!addr)
1963            continue;
1964
1965        kb->start_addr = (unsigned long)addr;
1966        symbol_name = kallsyms_lookup(kb->start_addr,
1967                &size, &offset, &modname, namebuf);
1968        if (!symbol_name)
1969            kb->range = 0;
1970        else
1971            kb->range = size;
1972    }
1973
1974    if (kretprobe_blacklist_size) {
1975        /* lookup the function address from its name */
1976        for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
1977            kprobe_lookup_name(kretprobe_blacklist[i].name,
1978                       kretprobe_blacklist[i].addr);
1979            if (!kretprobe_blacklist[i].addr)
1980                printk("kretprobe: lookup failed: %s\n",
1981                       kretprobe_blacklist[i].name);
1982        }
1983    }
1984
1985#if defined(CONFIG_OPTPROBES)
1986#if defined(__ARCH_WANT_KPROBES_INSN_SLOT)
1987    /* Init kprobe_optinsn_slots */
1988    kprobe_optinsn_slots.insn_size = MAX_OPTINSN_SIZE;
1989#endif
1990    /* By default, kprobes can be optimized */
1991    kprobes_allow_optimization = true;
1992#endif
1993
1994    /* By default, kprobes are armed */
1995    kprobes_all_disarmed = false;
1996
1997    err = arch_init_kprobes();
1998    if (!err)
1999        err = register_die_notifier(&kprobe_exceptions_nb);
2000    if (!err)
2001        err = register_module_notifier(&kprobe_module_nb);
2002
2003    kprobes_initialized = (err == 0);
2004
2005    if (!err)
2006        init_test_probes();
2007    return err;
2008}
2009
2010#ifdef CONFIG_DEBUG_FS
2011static void __kprobes report_probe(struct seq_file *pi, struct kprobe *p,
2012        const char *sym, int offset, char *modname, struct kprobe *pp)
2013{
2014    char *kprobe_type;
2015
2016    if (p->pre_handler == pre_handler_kretprobe)
2017        kprobe_type = "r";
2018    else if (p->pre_handler == setjmp_pre_handler)
2019        kprobe_type = "j";
2020    else
2021        kprobe_type = "k";
2022
2023    if (sym)
2024        seq_printf(pi, "%p %s %s+0x%x %s ",
2025            p->addr, kprobe_type, sym, offset,
2026            (modname ? modname : " "));
2027    else
2028        seq_printf(pi, "%p %s %p ",
2029            p->addr, kprobe_type, p->addr);
2030
2031    if (!pp)
2032        pp = p;
2033    seq_printf(pi, "%s%s%s\n",
2034        (kprobe_gone(p) ? "[GONE]" : ""),
2035        ((kprobe_disabled(p) && !kprobe_gone(p)) ? "[DISABLED]" : ""),
2036        (kprobe_optimized(pp) ? "[OPTIMIZED]" : ""));
2037}
2038
2039static void __kprobes *kprobe_seq_start(struct seq_file *f, loff_t *pos)
2040{
2041    return (*pos < KPROBE_TABLE_SIZE) ? pos : NULL;
2042}
2043
2044static void __kprobes *kprobe_seq_next(struct seq_file *f, void *v, loff_t *pos)
2045{
2046    (*pos)++;
2047    if (*pos >= KPROBE_TABLE_SIZE)
2048        return NULL;
2049    return pos;
2050}
2051
2052static void __kprobes kprobe_seq_stop(struct seq_file *f, void *v)
2053{
2054    /* Nothing to do */
2055}
2056
2057static int __kprobes show_kprobe_addr(struct seq_file *pi, void *v)
2058{
2059    struct hlist_head *head;
2060    struct hlist_node *node;
2061    struct kprobe *p, *kp;
2062    const char *sym = NULL;
2063    unsigned int i = *(loff_t *) v;
2064    unsigned long offset = 0;
2065    char *modname, namebuf[128];
2066
2067    head = &kprobe_table[i];
2068    preempt_disable();
2069    hlist_for_each_entry_rcu(p, node, head, hlist) {
2070        sym = kallsyms_lookup((unsigned long)p->addr, NULL,
2071                    &offset, &modname, namebuf);
2072        if (kprobe_aggrprobe(p)) {
2073            list_for_each_entry_rcu(kp, &p->list, list)
2074                report_probe(pi, kp, sym, offset, modname, p);
2075        } else
2076            report_probe(pi, p, sym, offset, modname, NULL);
2077    }
2078    preempt_enable();
2079    return 0;
2080}
2081
2082static const struct seq_operations kprobes_seq_ops = {
2083    .start = kprobe_seq_start,
2084    .next = kprobe_seq_next,
2085    .stop = kprobe_seq_stop,
2086    .show = show_kprobe_addr
2087};
2088
2089static int __kprobes kprobes_open(struct inode *inode, struct file *filp)
2090{
2091    return seq_open(filp, &kprobes_seq_ops);
2092}
2093
2094static const struct file_operations debugfs_kprobes_operations = {
2095    .open = kprobes_open,
2096    .read = seq_read,
2097    .llseek = seq_lseek,
2098    .release = seq_release,
2099};
2100
2101static void __kprobes arm_all_kprobes(void)
2102{
2103    struct hlist_head *head;
2104    struct hlist_node *node;
2105    struct kprobe *p;
2106    unsigned int i;
2107
2108    mutex_lock(&kprobe_mutex);
2109
2110    /* If kprobes are armed, just return */
2111    if (!kprobes_all_disarmed)
2112        goto already_enabled;
2113
2114    /* Arming kprobes doesn't optimize kprobe itself */
2115    mutex_lock(&text_mutex);
2116    for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2117        head = &kprobe_table[i];
2118        hlist_for_each_entry_rcu(p, node, head, hlist)
2119            if (!kprobe_disabled(p))
2120                __arm_kprobe(p);
2121    }
2122    mutex_unlock(&text_mutex);
2123
2124    kprobes_all_disarmed = false;
2125    printk(KERN_INFO "Kprobes globally enabled\n");
2126
2127already_enabled:
2128    mutex_unlock(&kprobe_mutex);
2129    return;
2130}
2131
2132static void __kprobes disarm_all_kprobes(void)
2133{
2134    struct hlist_head *head;
2135    struct hlist_node *node;
2136    struct kprobe *p;
2137    unsigned int i;
2138
2139    mutex_lock(&kprobe_mutex);
2140
2141    /* If kprobes are already disarmed, just return */
2142    if (kprobes_all_disarmed) {
2143        mutex_unlock(&kprobe_mutex);
2144        return;
2145    }
2146
2147    kprobes_all_disarmed = true;
2148    printk(KERN_INFO "Kprobes globally disabled\n");
2149
2150    mutex_lock(&text_mutex);
2151    for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2152        head = &kprobe_table[i];
2153        hlist_for_each_entry_rcu(p, node, head, hlist) {
2154            if (!arch_trampoline_kprobe(p) && !kprobe_disabled(p))
2155                __disarm_kprobe(p, false);
2156        }
2157    }
2158    mutex_unlock(&text_mutex);
2159    mutex_unlock(&kprobe_mutex);
2160
2161    /* Wait for disarming all kprobes by optimizer */
2162    wait_for_kprobe_optimizer();
2163}
2164
2165/*
2166 * XXX: The debugfs bool file interface doesn't allow for callbacks
2167 * when the bool state is switched. We can reuse that facility when
2168 * available
2169 */
2170static ssize_t read_enabled_file_bool(struct file *file,
2171           char __user *user_buf, size_t count, loff_t *ppos)
2172{
2173    char buf[3];
2174
2175    if (!kprobes_all_disarmed)
2176        buf[0] = '1';
2177    else
2178        buf[0] = '0';
2179    buf[1] = '\n';
2180    buf[2] = 0x00;
2181    return simple_read_from_buffer(user_buf, count, ppos, buf, 2);
2182}
2183
2184static ssize_t write_enabled_file_bool(struct file *file,
2185           const char __user *user_buf, size_t count, loff_t *ppos)
2186{
2187    char buf[32];
2188    int buf_size;
2189
2190    buf_size = min(count, (sizeof(buf)-1));
2191    if (copy_from_user(buf, user_buf, buf_size))
2192        return -EFAULT;
2193
2194    switch (buf[0]) {
2195    case 'y':
2196    case 'Y':
2197    case '1':
2198        arm_all_kprobes();
2199        break;
2200    case 'n':
2201    case 'N':
2202    case '0':
2203        disarm_all_kprobes();
2204        break;
2205    }
2206
2207    return count;
2208}
2209
2210static const struct file_operations fops_kp = {
2211    .read = read_enabled_file_bool,
2212    .write = write_enabled_file_bool,
2213    .llseek = default_llseek,
2214};
2215
2216static int __kprobes debugfs_kprobe_init(void)
2217{
2218    struct dentry *dir, *file;
2219    unsigned int value = 1;
2220
2221    dir = debugfs_create_dir("kprobes", NULL);
2222    if (!dir)
2223        return -ENOMEM;
2224
2225    file = debugfs_create_file("list", 0444, dir, NULL,
2226                &debugfs_kprobes_operations);
2227    if (!file) {
2228        debugfs_remove(dir);
2229        return -ENOMEM;
2230    }
2231
2232    file = debugfs_create_file("enabled", 0600, dir,
2233                    &value, &fops_kp);
2234    if (!file) {
2235        debugfs_remove(dir);
2236        return -ENOMEM;
2237    }
2238
2239    return 0;
2240}
2241
2242late_initcall(debugfs_kprobe_init);
2243#endif /* CONFIG_DEBUG_FS */
2244
2245module_init(init_kprobes);
2246
2247/* defined in arch/.../kernel/kprobes.c */
2248EXPORT_SYMBOL_GPL(jprobe_return);
2249

Archive Download this file



interactive