linux/kernel/kprobes.c
<<
>>
Prefs
   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/export.h>
  40#include <linux/moduleloader.h>
  41#include <linux/kallsyms.h>
  42#include <linux/freezer.h>
  43#include <linux/seq_file.h>
  44#include <linux/debugfs.h>
  45#include <linux/sysctl.h>
  46#include <linux/kdebug.h>
  47#include <linux/memory.h>
  48#include <linux/ftrace.h>
  49#include <linux/cpu.h>
  50#include <linux/jump_label.h>
  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        raw_spinlock_t lock ____cacheline_aligned_in_smp;
  82} kretprobe_table_locks[KPROBE_TABLE_SIZE];
  83
  84static raw_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 kprobe *p;
 338
 339        head = &kprobe_table[hash_ptr(addr, KPROBE_HASH_BITS)];
 340        hlist_for_each_entry_rcu(p, head, hlist) {
 341                if (p->addr == addr)
 342                        return p;
 343        }
 344
 345        return NULL;
 346}
 347
 348static int __kprobes aggr_pre_handler(struct kprobe *p, struct pt_regs *regs);
 349
 350/* Return true if the kprobe is an aggregator */
 351static inline int kprobe_aggrprobe(struct kprobe *p)
 352{
 353        return p->pre_handler == aggr_pre_handler;
 354}
 355
 356/* Return true(!0) if the kprobe is unused */
 357static inline int kprobe_unused(struct kprobe *p)
 358{
 359        return kprobe_aggrprobe(p) && kprobe_disabled(p) &&
 360               list_empty(&p->list);
 361}
 362
 363/*
 364 * Keep all fields in the kprobe consistent
 365 */
 366static inline void copy_kprobe(struct kprobe *ap, struct kprobe *p)
 367{
 368        memcpy(&p->opcode, &ap->opcode, sizeof(kprobe_opcode_t));
 369        memcpy(&p->ainsn, &ap->ainsn, sizeof(struct arch_specific_insn));
 370}
 371
 372#ifdef CONFIG_OPTPROBES
 373/* NOTE: change this value only with kprobe_mutex held */
 374static bool kprobes_allow_optimization;
 375
 376/*
 377 * Call all pre_handler on the list, but ignores its return value.
 378 * This must be called from arch-dep optimized caller.
 379 */
 380void __kprobes opt_pre_handler(struct kprobe *p, struct pt_regs *regs)
 381{
 382        struct kprobe *kp;
 383
 384        list_for_each_entry_rcu(kp, &p->list, list) {
 385                if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
 386                        set_kprobe_instance(kp);
 387                        kp->pre_handler(kp, regs);
 388                }
 389                reset_kprobe_instance();
 390        }
 391}
 392
 393/* Free optimized instructions and optimized_kprobe */
 394static __kprobes void free_aggr_kprobe(struct kprobe *p)
 395{
 396        struct optimized_kprobe *op;
 397
 398        op = container_of(p, struct optimized_kprobe, kp);
 399        arch_remove_optimized_kprobe(op);
 400        arch_remove_kprobe(p);
 401        kfree(op);
 402}
 403
 404/* Return true(!0) if the kprobe is ready for optimization. */
 405static inline int kprobe_optready(struct kprobe *p)
 406{
 407        struct optimized_kprobe *op;
 408
 409        if (kprobe_aggrprobe(p)) {
 410                op = container_of(p, struct optimized_kprobe, kp);
 411                return arch_prepared_optinsn(&op->optinsn);
 412        }
 413
 414        return 0;
 415}
 416
 417/* Return true(!0) if the kprobe is disarmed. Note: p must be on hash list */
 418static inline int kprobe_disarmed(struct kprobe *p)
 419{
 420        struct optimized_kprobe *op;
 421
 422        /* If kprobe is not aggr/opt probe, just return kprobe is disabled */
 423        if (!kprobe_aggrprobe(p))
 424                return kprobe_disabled(p);
 425
 426        op = container_of(p, struct optimized_kprobe, kp);
 427
 428        return kprobe_disabled(p) && list_empty(&op->list);
 429}
 430
 431/* Return true(!0) if the probe is queued on (un)optimizing lists */
 432static int __kprobes kprobe_queued(struct kprobe *p)
 433{
 434        struct optimized_kprobe *op;
 435
 436        if (kprobe_aggrprobe(p)) {
 437                op = container_of(p, struct optimized_kprobe, kp);
 438                if (!list_empty(&op->list))
 439                        return 1;
 440        }
 441        return 0;
 442}
 443
 444/*
 445 * Return an optimized kprobe whose optimizing code replaces
 446 * instructions including addr (exclude breakpoint).
 447 */
 448static struct kprobe *__kprobes get_optimized_kprobe(unsigned long addr)
 449{
 450        int i;
 451        struct kprobe *p = NULL;
 452        struct optimized_kprobe *op;
 453
 454        /* Don't check i == 0, since that is a breakpoint case. */
 455        for (i = 1; !p && i < MAX_OPTIMIZED_LENGTH; i++)
 456                p = get_kprobe((void *)(addr - i));
 457
 458        if (p && kprobe_optready(p)) {
 459                op = container_of(p, struct optimized_kprobe, kp);
 460                if (arch_within_optimized_kprobe(op, addr))
 461                        return p;
 462        }
 463
 464        return NULL;
 465}
 466
 467/* Optimization staging list, protected by kprobe_mutex */
 468static LIST_HEAD(optimizing_list);
 469static LIST_HEAD(unoptimizing_list);
 470
 471static void kprobe_optimizer(struct work_struct *work);
 472static DECLARE_DELAYED_WORK(optimizing_work, kprobe_optimizer);
 473#define OPTIMIZE_DELAY 5
 474
 475/*
 476 * Optimize (replace a breakpoint with a jump) kprobes listed on
 477 * optimizing_list.
 478 */
 479static __kprobes void do_optimize_kprobes(void)
 480{
 481        /* Optimization never be done when disarmed */
 482        if (kprobes_all_disarmed || !kprobes_allow_optimization ||
 483            list_empty(&optimizing_list))
 484                return;
 485
 486        /*
 487         * The optimization/unoptimization refers online_cpus via
 488         * stop_machine() and cpu-hotplug modifies online_cpus.
 489         * And same time, text_mutex will be held in cpu-hotplug and here.
 490         * This combination can cause a deadlock (cpu-hotplug try to lock
 491         * text_mutex but stop_machine can not be done because online_cpus
 492         * has been changed)
 493         * To avoid this deadlock, we need to call get_online_cpus()
 494         * for preventing cpu-hotplug outside of text_mutex locking.
 495         */
 496        get_online_cpus();
 497        mutex_lock(&text_mutex);
 498        arch_optimize_kprobes(&optimizing_list);
 499        mutex_unlock(&text_mutex);
 500        put_online_cpus();
 501}
 502
 503/*
 504 * Unoptimize (replace a jump with a breakpoint and remove the breakpoint
 505 * if need) kprobes listed on unoptimizing_list.
 506 */
 507static __kprobes void do_unoptimize_kprobes(struct list_head *free_list)
 508{
 509        struct optimized_kprobe *op, *tmp;
 510
 511        /* Unoptimization must be done anytime */
 512        if (list_empty(&unoptimizing_list))
 513                return;
 514
 515        /* Ditto to do_optimize_kprobes */
 516        get_online_cpus();
 517        mutex_lock(&text_mutex);
 518        arch_unoptimize_kprobes(&unoptimizing_list, free_list);
 519        /* Loop free_list for disarming */
 520        list_for_each_entry_safe(op, tmp, free_list, list) {
 521                /* Disarm probes if marked disabled */
 522                if (kprobe_disabled(&op->kp))
 523                        arch_disarm_kprobe(&op->kp);
 524                if (kprobe_unused(&op->kp)) {
 525                        /*
 526                         * Remove unused probes from hash list. After waiting
 527                         * for synchronization, these probes are reclaimed.
 528                         * (reclaiming is done by do_free_cleaned_kprobes.)
 529                         */
 530                        hlist_del_rcu(&op->kp.hlist);
 531                } else
 532                        list_del_init(&op->list);
 533        }
 534        mutex_unlock(&text_mutex);
 535        put_online_cpus();
 536}
 537
 538/* Reclaim all kprobes on the free_list */
 539static __kprobes void do_free_cleaned_kprobes(struct list_head *free_list)
 540{
 541        struct optimized_kprobe *op, *tmp;
 542
 543        list_for_each_entry_safe(op, tmp, free_list, list) {
 544                BUG_ON(!kprobe_unused(&op->kp));
 545                list_del_init(&op->list);
 546                free_aggr_kprobe(&op->kp);
 547        }
 548}
 549
 550/* Start optimizer after OPTIMIZE_DELAY passed */
 551static __kprobes void kick_kprobe_optimizer(void)
 552{
 553        schedule_delayed_work(&optimizing_work, OPTIMIZE_DELAY);
 554}
 555
 556/* Kprobe jump optimizer */
 557static __kprobes void kprobe_optimizer(struct work_struct *work)
 558{
 559        LIST_HEAD(free_list);
 560
 561        mutex_lock(&kprobe_mutex);
 562        /* Lock modules while optimizing kprobes */
 563        mutex_lock(&module_mutex);
 564
 565        /*
 566         * Step 1: Unoptimize kprobes and collect cleaned (unused and disarmed)
 567         * kprobes before waiting for quiesence period.
 568         */
 569        do_unoptimize_kprobes(&free_list);
 570
 571        /*
 572         * Step 2: Wait for quiesence period to ensure all running interrupts
 573         * are done. Because optprobe may modify multiple instructions
 574         * there is a chance that Nth instruction is interrupted. In that
 575         * case, running interrupt can return to 2nd-Nth byte of jump
 576         * instruction. This wait is for avoiding it.
 577         */
 578        synchronize_sched();
 579
 580        /* Step 3: Optimize kprobes after quiesence period */
 581        do_optimize_kprobes();
 582
 583        /* Step 4: Free cleaned kprobes after quiesence period */
 584        do_free_cleaned_kprobes(&free_list);
 585
 586        mutex_unlock(&module_mutex);
 587        mutex_unlock(&kprobe_mutex);
 588
 589        /* Step 5: Kick optimizer again if needed */
 590        if (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list))
 591                kick_kprobe_optimizer();
 592}
 593
 594/* Wait for completing optimization and unoptimization */
 595static __kprobes void wait_for_kprobe_optimizer(void)
 596{
 597        mutex_lock(&kprobe_mutex);
 598
 599        while (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list)) {
 600                mutex_unlock(&kprobe_mutex);
 601
 602                /* this will also make optimizing_work execute immmediately */
 603                flush_delayed_work(&optimizing_work);
 604                /* @optimizing_work might not have been queued yet, relax */
 605                cpu_relax();
 606
 607                mutex_lock(&kprobe_mutex);
 608        }
 609
 610        mutex_unlock(&kprobe_mutex);
 611}
 612
 613/* Optimize kprobe if p is ready to be optimized */
 614static __kprobes void optimize_kprobe(struct kprobe *p)
 615{
 616        struct optimized_kprobe *op;
 617
 618        /* Check if the kprobe is disabled or not ready for optimization. */
 619        if (!kprobe_optready(p) || !kprobes_allow_optimization ||
 620            (kprobe_disabled(p) || kprobes_all_disarmed))
 621                return;
 622
 623        /* Both of break_handler and post_handler are not supported. */
 624        if (p->break_handler || p->post_handler)
 625                return;
 626
 627        op = container_of(p, struct optimized_kprobe, kp);
 628
 629        /* Check there is no other kprobes at the optimized instructions */
 630        if (arch_check_optimized_kprobe(op) < 0)
 631                return;
 632
 633        /* Check if it is already optimized. */
 634        if (op->kp.flags & KPROBE_FLAG_OPTIMIZED)
 635                return;
 636        op->kp.flags |= KPROBE_FLAG_OPTIMIZED;
 637
 638        if (!list_empty(&op->list))
 639                /* This is under unoptimizing. Just dequeue the probe */
 640                list_del_init(&op->list);
 641        else {
 642                list_add(&op->list, &optimizing_list);
 643                kick_kprobe_optimizer();
 644        }
 645}
 646
 647/* Short cut to direct unoptimizing */
 648static __kprobes void force_unoptimize_kprobe(struct optimized_kprobe *op)
 649{
 650        get_online_cpus();
 651        arch_unoptimize_kprobe(op);
 652        put_online_cpus();
 653        if (kprobe_disabled(&op->kp))
 654                arch_disarm_kprobe(&op->kp);
 655}
 656
 657/* Unoptimize a kprobe if p is optimized */
 658static __kprobes void unoptimize_kprobe(struct kprobe *p, bool force)
 659{
 660        struct optimized_kprobe *op;
 661
 662        if (!kprobe_aggrprobe(p) || kprobe_disarmed(p))
 663                return; /* This is not an optprobe nor optimized */
 664
 665        op = container_of(p, struct optimized_kprobe, kp);
 666        if (!kprobe_optimized(p)) {
 667                /* Unoptimized or unoptimizing case */
 668                if (force && !list_empty(&op->list)) {
 669                        /*
 670                         * Only if this is unoptimizing kprobe and forced,
 671                         * forcibly unoptimize it. (No need to unoptimize
 672                         * unoptimized kprobe again :)
 673                         */
 674                        list_del_init(&op->list);
 675                        force_unoptimize_kprobe(op);
 676                }
 677                return;
 678        }
 679
 680        op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
 681        if (!list_empty(&op->list)) {
 682                /* Dequeue from the optimization queue */
 683                list_del_init(&op->list);
 684                return;
 685        }
 686        /* Optimized kprobe case */
 687        if (force)
 688                /* Forcibly update the code: this is a special case */
 689                force_unoptimize_kprobe(op);
 690        else {
 691                list_add(&op->list, &unoptimizing_list);
 692                kick_kprobe_optimizer();
 693        }
 694}
 695
 696/* Cancel unoptimizing for reusing */
 697static void reuse_unused_kprobe(struct kprobe *ap)
 698{
 699        struct optimized_kprobe *op;
 700
 701        BUG_ON(!kprobe_unused(ap));
 702        /*
 703         * Unused kprobe MUST be on the way of delayed unoptimizing (means
 704         * there is still a relative jump) and disabled.
 705         */
 706        op = container_of(ap, struct optimized_kprobe, kp);
 707        if (unlikely(list_empty(&op->list)))
 708                printk(KERN_WARNING "Warning: found a stray unused "
 709                        "aggrprobe@%p\n", ap->addr);
 710        /* Enable the probe again */
 711        ap->flags &= ~KPROBE_FLAG_DISABLED;
 712        /* Optimize it again (remove from op->list) */
 713        BUG_ON(!kprobe_optready(ap));
 714        optimize_kprobe(ap);
 715}
 716
 717/* Remove optimized instructions */
 718static void __kprobes kill_optimized_kprobe(struct kprobe *p)
 719{
 720        struct optimized_kprobe *op;
 721
 722        op = container_of(p, struct optimized_kprobe, kp);
 723        if (!list_empty(&op->list))
 724                /* Dequeue from the (un)optimization queue */
 725                list_del_init(&op->list);
 726
 727        op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
 728        /* Don't touch the code, because it is already freed. */
 729        arch_remove_optimized_kprobe(op);
 730}
 731
 732/* Try to prepare optimized instructions */
 733static __kprobes void prepare_optimized_kprobe(struct kprobe *p)
 734{
 735        struct optimized_kprobe *op;
 736
 737        op = container_of(p, struct optimized_kprobe, kp);
 738        arch_prepare_optimized_kprobe(op);
 739}
 740
 741/* Allocate new optimized_kprobe and try to prepare optimized instructions */
 742static __kprobes struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
 743{
 744        struct optimized_kprobe *op;
 745
 746        op = kzalloc(sizeof(struct optimized_kprobe), GFP_KERNEL);
 747        if (!op)
 748                return NULL;
 749
 750        INIT_LIST_HEAD(&op->list);
 751        op->kp.addr = p->addr;
 752        arch_prepare_optimized_kprobe(op);
 753
 754        return &op->kp;
 755}
 756
 757static void __kprobes init_aggr_kprobe(struct kprobe *ap, struct kprobe *p);
 758
 759/*
 760 * Prepare an optimized_kprobe and optimize it
 761 * NOTE: p must be a normal registered kprobe
 762 */
 763static __kprobes void try_to_optimize_kprobe(struct kprobe *p)
 764{
 765        struct kprobe *ap;
 766        struct optimized_kprobe *op;
 767
 768        /* Impossible to optimize ftrace-based kprobe */
 769        if (kprobe_ftrace(p))
 770                return;
 771
 772        /* For preparing optimization, jump_label_text_reserved() is called */
 773        jump_label_lock();
 774        mutex_lock(&text_mutex);
 775
 776        ap = alloc_aggr_kprobe(p);
 777        if (!ap)
 778                goto out;
 779
 780        op = container_of(ap, struct optimized_kprobe, kp);
 781        if (!arch_prepared_optinsn(&op->optinsn)) {
 782                /* If failed to setup optimizing, fallback to kprobe */
 783                arch_remove_optimized_kprobe(op);
 784                kfree(op);
 785                goto out;
 786        }
 787
 788        init_aggr_kprobe(ap, p);
 789        optimize_kprobe(ap);    /* This just kicks optimizer thread */
 790
 791out:
 792        mutex_unlock(&text_mutex);
 793        jump_label_unlock();
 794}
 795
 796#ifdef CONFIG_SYSCTL
 797static void __kprobes optimize_all_kprobes(void)
 798{
 799        struct hlist_head *head;
 800        struct kprobe *p;
 801        unsigned int i;
 802
 803        mutex_lock(&kprobe_mutex);
 804        /* If optimization is already allowed, just return */
 805        if (kprobes_allow_optimization)
 806                goto out;
 807
 808        kprobes_allow_optimization = true;
 809        for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
 810                head = &kprobe_table[i];
 811                hlist_for_each_entry_rcu(p, head, hlist)
 812                        if (!kprobe_disabled(p))
 813                                optimize_kprobe(p);
 814        }
 815        printk(KERN_INFO "Kprobes globally optimized\n");
 816out:
 817        mutex_unlock(&kprobe_mutex);
 818}
 819
 820static void __kprobes unoptimize_all_kprobes(void)
 821{
 822        struct hlist_head *head;
 823        struct kprobe *p;
 824        unsigned int i;
 825
 826        mutex_lock(&kprobe_mutex);
 827        /* If optimization is already prohibited, just return */
 828        if (!kprobes_allow_optimization) {
 829                mutex_unlock(&kprobe_mutex);
 830                return;
 831        }
 832
 833        kprobes_allow_optimization = false;
 834        for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
 835                head = &kprobe_table[i];
 836                hlist_for_each_entry_rcu(p, head, hlist) {
 837                        if (!kprobe_disabled(p))
 838                                unoptimize_kprobe(p, false);
 839                }
 840        }
 841        mutex_unlock(&kprobe_mutex);
 842
 843        /* Wait for unoptimizing completion */
 844        wait_for_kprobe_optimizer();
 845        printk(KERN_INFO "Kprobes globally unoptimized\n");
 846}
 847
 848static DEFINE_MUTEX(kprobe_sysctl_mutex);
 849int sysctl_kprobes_optimization;
 850int proc_kprobes_optimization_handler(struct ctl_table *table, int write,
 851                                      void __user *buffer, size_t *length,
 852                                      loff_t *ppos)
 853{
 854        int ret;
 855
 856        mutex_lock(&kprobe_sysctl_mutex);
 857        sysctl_kprobes_optimization = kprobes_allow_optimization ? 1 : 0;
 858        ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
 859
 860        if (sysctl_kprobes_optimization)
 861                optimize_all_kprobes();
 862        else
 863                unoptimize_all_kprobes();
 864        mutex_unlock(&kprobe_sysctl_mutex);
 865
 866        return ret;
 867}
 868#endif /* CONFIG_SYSCTL */
 869
 870/* Put a breakpoint for a probe. Must be called with text_mutex locked */
 871static void __kprobes __arm_kprobe(struct kprobe *p)
 872{
 873        struct kprobe *_p;
 874
 875        /* Check collision with other optimized kprobes */
 876        _p = get_optimized_kprobe((unsigned long)p->addr);
 877        if (unlikely(_p))
 878                /* Fallback to unoptimized kprobe */
 879                unoptimize_kprobe(_p, true);
 880
 881        arch_arm_kprobe(p);
 882        optimize_kprobe(p);     /* Try to optimize (add kprobe to a list) */
 883}
 884
 885/* Remove the breakpoint of a probe. Must be called with text_mutex locked */
 886static void __kprobes __disarm_kprobe(struct kprobe *p, bool reopt)
 887{
 888        struct kprobe *_p;
 889
 890        unoptimize_kprobe(p, false);    /* Try to unoptimize */
 891
 892        if (!kprobe_queued(p)) {
 893                arch_disarm_kprobe(p);
 894                /* If another kprobe was blocked, optimize it. */
 895                _p = get_optimized_kprobe((unsigned long)p->addr);
 896                if (unlikely(_p) && reopt)
 897                        optimize_kprobe(_p);
 898        }
 899        /* TODO: reoptimize others after unoptimized this probe */
 900}
 901
 902#else /* !CONFIG_OPTPROBES */
 903
 904#define optimize_kprobe(p)                      do {} while (0)
 905#define unoptimize_kprobe(p, f)                 do {} while (0)
 906#define kill_optimized_kprobe(p)                do {} while (0)
 907#define prepare_optimized_kprobe(p)             do {} while (0)
 908#define try_to_optimize_kprobe(p)               do {} while (0)
 909#define __arm_kprobe(p)                         arch_arm_kprobe(p)
 910#define __disarm_kprobe(p, o)                   arch_disarm_kprobe(p)
 911#define kprobe_disarmed(p)                      kprobe_disabled(p)
 912#define wait_for_kprobe_optimizer()             do {} while (0)
 913
 914/* There should be no unused kprobes can be reused without optimization */
 915static void reuse_unused_kprobe(struct kprobe *ap)
 916{
 917        printk(KERN_ERR "Error: There should be no unused kprobe here.\n");
 918        BUG_ON(kprobe_unused(ap));
 919}
 920
 921static __kprobes void free_aggr_kprobe(struct kprobe *p)
 922{
 923        arch_remove_kprobe(p);
 924        kfree(p);
 925}
 926
 927static __kprobes struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
 928{
 929        return kzalloc(sizeof(struct kprobe), GFP_KERNEL);
 930}
 931#endif /* CONFIG_OPTPROBES */
 932
 933#ifdef CONFIG_KPROBES_ON_FTRACE
 934static struct ftrace_ops kprobe_ftrace_ops __read_mostly = {
 935        .func = kprobe_ftrace_handler,
 936        .flags = FTRACE_OPS_FL_SAVE_REGS,
 937};
 938static int kprobe_ftrace_enabled;
 939
 940/* Must ensure p->addr is really on ftrace */
 941static int __kprobes prepare_kprobe(struct kprobe *p)
 942{
 943        if (!kprobe_ftrace(p))
 944                return arch_prepare_kprobe(p);
 945
 946        return arch_prepare_kprobe_ftrace(p);
 947}
 948
 949/* Caller must lock kprobe_mutex */
 950static void __kprobes arm_kprobe_ftrace(struct kprobe *p)
 951{
 952        int ret;
 953
 954        ret = ftrace_set_filter_ip(&kprobe_ftrace_ops,
 955                                   (unsigned long)p->addr, 0, 0);
 956        WARN(ret < 0, "Failed to arm kprobe-ftrace at %p (%d)\n", p->addr, ret);
 957        kprobe_ftrace_enabled++;
 958        if (kprobe_ftrace_enabled == 1) {
 959                ret = register_ftrace_function(&kprobe_ftrace_ops);
 960                WARN(ret < 0, "Failed to init kprobe-ftrace (%d)\n", ret);
 961        }
 962}
 963
 964/* Caller must lock kprobe_mutex */
 965static void __kprobes disarm_kprobe_ftrace(struct kprobe *p)
 966{
 967        int ret;
 968
 969        kprobe_ftrace_enabled--;
 970        if (kprobe_ftrace_enabled == 0) {
 971                ret = unregister_ftrace_function(&kprobe_ftrace_ops);
 972                WARN(ret < 0, "Failed to init kprobe-ftrace (%d)\n", ret);
 973        }
 974        ret = ftrace_set_filter_ip(&kprobe_ftrace_ops,
 975                           (unsigned long)p->addr, 1, 0);
 976        WARN(ret < 0, "Failed to disarm kprobe-ftrace at %p (%d)\n", p->addr, ret);
 977}
 978#else   /* !CONFIG_KPROBES_ON_FTRACE */
 979#define prepare_kprobe(p)       arch_prepare_kprobe(p)
 980#define arm_kprobe_ftrace(p)    do {} while (0)
 981#define disarm_kprobe_ftrace(p) do {} while (0)
 982#endif
 983
 984/* Arm a kprobe with text_mutex */
 985static void __kprobes arm_kprobe(struct kprobe *kp)
 986{
 987        if (unlikely(kprobe_ftrace(kp))) {
 988                arm_kprobe_ftrace(kp);
 989                return;
 990        }
 991        /*
 992         * Here, since __arm_kprobe() doesn't use stop_machine(),
 993         * this doesn't cause deadlock on text_mutex. So, we don't
 994         * need get_online_cpus().
 995         */
 996        mutex_lock(&text_mutex);
 997        __arm_kprobe(kp);
 998        mutex_unlock(&text_mutex);
 999}
1000
1001/* Disarm a kprobe with text_mutex */
1002static void __kprobes disarm_kprobe(struct kprobe *kp, bool reopt)
1003{
1004        if (unlikely(kprobe_ftrace(kp))) {
1005                disarm_kprobe_ftrace(kp);
1006                return;
1007        }
1008        /* Ditto */
1009        mutex_lock(&text_mutex);
1010        __disarm_kprobe(kp, reopt);
1011        mutex_unlock(&text_mutex);
1012}
1013
1014/*
1015 * Aggregate handlers for multiple kprobes support - these handlers
1016 * take care of invoking the individual kprobe handlers on p->list
1017 */
1018static int __kprobes aggr_pre_handler(struct kprobe *p, struct pt_regs *regs)
1019{
1020        struct kprobe *kp;
1021
1022        list_for_each_entry_rcu(kp, &p->list, list) {
1023                if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
1024                        set_kprobe_instance(kp);
1025                        if (kp->pre_handler(kp, regs))
1026                                return 1;
1027                }
1028                reset_kprobe_instance();
1029        }
1030        return 0;
1031}
1032
1033static void __kprobes aggr_post_handler(struct kprobe *p, struct pt_regs *regs,
1034                                        unsigned long flags)
1035{
1036        struct kprobe *kp;
1037
1038        list_for_each_entry_rcu(kp, &p->list, list) {
1039                if (kp->post_handler && likely(!kprobe_disabled(kp))) {
1040                        set_kprobe_instance(kp);
1041                        kp->post_handler(kp, regs, flags);
1042                        reset_kprobe_instance();
1043                }
1044        }
1045}
1046
1047static int __kprobes aggr_fault_handler(struct kprobe *p, struct pt_regs *regs,
1048                                        int trapnr)
1049{
1050        struct kprobe *cur = __this_cpu_read(kprobe_instance);
1051
1052        /*
1053         * if we faulted "during" the execution of a user specified
1054         * probe handler, invoke just that probe's fault handler
1055         */
1056        if (cur && cur->fault_handler) {
1057                if (cur->fault_handler(cur, regs, trapnr))
1058                        return 1;
1059        }
1060        return 0;
1061}
1062
1063static int __kprobes aggr_break_handler(struct kprobe *p, struct pt_regs *regs)
1064{
1065        struct kprobe *cur = __this_cpu_read(kprobe_instance);
1066        int ret = 0;
1067
1068        if (cur && cur->break_handler) {
1069                if (cur->break_handler(cur, regs))
1070                        ret = 1;
1071        }
1072        reset_kprobe_instance();
1073        return ret;
1074}
1075
1076/* Walks the list and increments nmissed count for multiprobe case */
1077void __kprobes kprobes_inc_nmissed_count(struct kprobe *p)
1078{
1079        struct kprobe *kp;
1080        if (!kprobe_aggrprobe(p)) {
1081                p->nmissed++;
1082        } else {
1083                list_for_each_entry_rcu(kp, &p->list, list)
1084                        kp->nmissed++;
1085        }
1086        return;
1087}
1088
1089void __kprobes recycle_rp_inst(struct kretprobe_instance *ri,
1090                                struct hlist_head *head)
1091{
1092        struct kretprobe *rp = ri->rp;
1093
1094        /* remove rp inst off the rprobe_inst_table */
1095        hlist_del(&ri->hlist);
1096        INIT_HLIST_NODE(&ri->hlist);
1097        if (likely(rp)) {
1098                raw_spin_lock(&rp->lock);
1099                hlist_add_head(&ri->hlist, &rp->free_instances);
1100                raw_spin_unlock(&rp->lock);
1101        } else
1102                /* Unregistering */
1103                hlist_add_head(&ri->hlist, head);
1104}
1105
1106void __kprobes kretprobe_hash_lock(struct task_struct *tsk,
1107                         struct hlist_head **head, unsigned long *flags)
1108__acquires(hlist_lock)
1109{
1110        unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
1111        raw_spinlock_t *hlist_lock;
1112
1113        *head = &kretprobe_inst_table[hash];
1114        hlist_lock = kretprobe_table_lock_ptr(hash);
1115        raw_spin_lock_irqsave(hlist_lock, *flags);
1116}
1117
1118static void __kprobes kretprobe_table_lock(unsigned long hash,
1119        unsigned long *flags)
1120__acquires(hlist_lock)
1121{
1122        raw_spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
1123        raw_spin_lock_irqsave(hlist_lock, *flags);
1124}
1125
1126void __kprobes kretprobe_hash_unlock(struct task_struct *tsk,
1127        unsigned long *flags)
1128__releases(hlist_lock)
1129{
1130        unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
1131        raw_spinlock_t *hlist_lock;
1132
1133        hlist_lock = kretprobe_table_lock_ptr(hash);
1134        raw_spin_unlock_irqrestore(hlist_lock, *flags);
1135}
1136
1137static void __kprobes kretprobe_table_unlock(unsigned long hash,
1138       unsigned long *flags)
1139__releases(hlist_lock)
1140{
1141        raw_spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
1142        raw_spin_unlock_irqrestore(hlist_lock, *flags);
1143}
1144
1145/*
1146 * This function is called from finish_task_switch when task tk becomes dead,
1147 * so that we can recycle any function-return probe instances associated
1148 * with this task. These left over instances represent probed functions
1149 * that have been called but will never return.
1150 */
1151void __kprobes kprobe_flush_task(struct task_struct *tk)
1152{
1153        struct kretprobe_instance *ri;
1154        struct hlist_head *head, empty_rp;
1155        struct hlist_node *tmp;
1156        unsigned long hash, flags = 0;
1157
1158        if (unlikely(!kprobes_initialized))
1159                /* Early boot.  kretprobe_table_locks not yet initialized. */
1160                return;
1161
1162        INIT_HLIST_HEAD(&empty_rp);
1163        hash = hash_ptr(tk, KPROBE_HASH_BITS);
1164        head = &kretprobe_inst_table[hash];
1165        kretprobe_table_lock(hash, &flags);
1166        hlist_for_each_entry_safe(ri, tmp, head, hlist) {
1167                if (ri->task == tk)
1168                        recycle_rp_inst(ri, &empty_rp);
1169        }
1170        kretprobe_table_unlock(hash, &flags);
1171        hlist_for_each_entry_safe(ri, tmp, &empty_rp, hlist) {
1172                hlist_del(&ri->hlist);
1173                kfree(ri);
1174        }
1175}
1176
1177static inline void free_rp_inst(struct kretprobe *rp)
1178{
1179        struct kretprobe_instance *ri;
1180        struct hlist_node *next;
1181
1182        hlist_for_each_entry_safe(ri, next, &rp->free_instances, hlist) {
1183                hlist_del(&ri->hlist);
1184                kfree(ri);
1185        }
1186}
1187
1188static void __kprobes cleanup_rp_inst(struct kretprobe *rp)
1189{
1190        unsigned long flags, hash;
1191        struct kretprobe_instance *ri;
1192        struct hlist_node *next;
1193        struct hlist_head *head;
1194
1195        /* No race here */
1196        for (hash = 0; hash < KPROBE_TABLE_SIZE; hash++) {
1197                kretprobe_table_lock(hash, &flags);
1198                head = &kretprobe_inst_table[hash];
1199                hlist_for_each_entry_safe(ri, next, head, hlist) {
1200                        if (ri->rp == rp)
1201                                ri->rp = NULL;
1202                }
1203                kretprobe_table_unlock(hash, &flags);
1204        }
1205        free_rp_inst(rp);
1206}
1207
1208/*
1209* Add the new probe to ap->list. Fail if this is the
1210* second jprobe at the address - two jprobes can't coexist
1211*/
1212static int __kprobes add_new_kprobe(struct kprobe *ap, struct kprobe *p)
1213{
1214        BUG_ON(kprobe_gone(ap) || kprobe_gone(p));
1215
1216        if (p->break_handler || p->post_handler)
1217                unoptimize_kprobe(ap, true);    /* Fall back to normal kprobe */
1218
1219        if (p->break_handler) {
1220                if (ap->break_handler)
1221                        return -EEXIST;
1222                list_add_tail_rcu(&p->list, &ap->list);
1223                ap->break_handler = aggr_break_handler;
1224        } else
1225                list_add_rcu(&p->list, &ap->list);
1226        if (p->post_handler && !ap->post_handler)
1227                ap->post_handler = aggr_post_handler;
1228
1229        return 0;
1230}
1231
1232/*
1233 * Fill in the required fields of the "manager kprobe". Replace the
1234 * earlier kprobe in the hlist with the manager kprobe
1235 */
1236static void __kprobes init_aggr_kprobe(struct kprobe *ap, struct kprobe *p)
1237{
1238        /* Copy p's insn slot to ap */
1239        copy_kprobe(p, ap);
1240        flush_insn_slot(ap);
1241        ap->addr = p->addr;
1242        ap->flags = p->flags & ~KPROBE_FLAG_OPTIMIZED;
1243        ap->pre_handler = aggr_pre_handler;
1244        ap->fault_handler = aggr_fault_handler;
1245        /* We don't care the kprobe which has gone. */
1246        if (p->post_handler && !kprobe_gone(p))
1247                ap->post_handler = aggr_post_handler;
1248        if (p->break_handler && !kprobe_gone(p))
1249                ap->break_handler = aggr_break_handler;
1250
1251        INIT_LIST_HEAD(&ap->list);
1252        INIT_HLIST_NODE(&ap->hlist);
1253
1254        list_add_rcu(&p->list, &ap->list);
1255        hlist_replace_rcu(&p->hlist, &ap->hlist);
1256}
1257
1258/*
1259 * This is the second or subsequent kprobe at the address - handle
1260 * the intricacies
1261 */
1262static int __kprobes register_aggr_kprobe(struct kprobe *orig_p,
1263                                          struct kprobe *p)
1264{
1265        int ret = 0;
1266        struct kprobe *ap = orig_p;
1267
1268        /* For preparing optimization, jump_label_text_reserved() is called */
1269        jump_label_lock();
1270        /*
1271         * Get online CPUs to avoid text_mutex deadlock.with stop machine,
1272         * which is invoked by unoptimize_kprobe() in add_new_kprobe()
1273         */
1274        get_online_cpus();
1275        mutex_lock(&text_mutex);
1276
1277        if (!kprobe_aggrprobe(orig_p)) {
1278                /* If orig_p is not an aggr_kprobe, create new aggr_kprobe. */
1279                ap = alloc_aggr_kprobe(orig_p);
1280                if (!ap) {
1281                        ret = -ENOMEM;
1282                        goto out;
1283                }
1284                init_aggr_kprobe(ap, orig_p);
1285        } else if (kprobe_unused(ap))
1286                /* This probe is going to die. Rescue it */
1287                reuse_unused_kprobe(ap);
1288
1289        if (kprobe_gone(ap)) {
1290                /*
1291                 * Attempting to insert new probe at the same location that
1292                 * had a probe in the module vaddr area which already
1293                 * freed. So, the instruction slot has already been
1294                 * released. We need a new slot for the new probe.
1295                 */
1296                ret = arch_prepare_kprobe(ap);
1297                if (ret)
1298                        /*
1299                         * Even if fail to allocate new slot, don't need to
1300                         * free aggr_probe. It will be used next time, or
1301                         * freed by unregister_kprobe.
1302                         */
1303                        goto out;
1304
1305                /* Prepare optimized instructions if possible. */
1306                prepare_optimized_kprobe(ap);
1307
1308                /*
1309                 * Clear gone flag to prevent allocating new slot again, and
1310                 * set disabled flag because it is not armed yet.
1311                 */
1312                ap->flags = (ap->flags & ~KPROBE_FLAG_GONE)
1313                            | KPROBE_FLAG_DISABLED;
1314        }
1315
1316        /* Copy ap's insn slot to p */
1317        copy_kprobe(ap, p);
1318        ret = add_new_kprobe(ap, p);
1319
1320out:
1321        mutex_unlock(&text_mutex);
1322        put_online_cpus();
1323        jump_label_unlock();
1324
1325        if (ret == 0 && kprobe_disabled(ap) && !kprobe_disabled(p)) {
1326                ap->flags &= ~KPROBE_FLAG_DISABLED;
1327                if (!kprobes_all_disarmed)
1328                        /* Arm the breakpoint again. */
1329                        arm_kprobe(ap);
1330        }
1331        return ret;
1332}
1333
1334static int __kprobes in_kprobes_functions(unsigned long addr)
1335{
1336        struct kprobe_blackpoint *kb;
1337
1338        if (addr >= (unsigned long)__kprobes_text_start &&
1339            addr < (unsigned long)__kprobes_text_end)
1340                return -EINVAL;
1341        /*
1342         * If there exists a kprobe_blacklist, verify and
1343         * fail any probe registration in the prohibited area
1344         */
1345        for (kb = kprobe_blacklist; kb->name != NULL; kb++) {
1346                if (kb->start_addr) {
1347                        if (addr >= kb->start_addr &&
1348                            addr < (kb->start_addr + kb->range))
1349                                return -EINVAL;
1350                }
1351        }
1352        return 0;
1353}
1354
1355/*
1356 * If we have a symbol_name argument, look it up and add the offset field
1357 * to it. This way, we can specify a relative address to a symbol.
1358 * This returns encoded errors if it fails to look up symbol or invalid
1359 * combination of parameters.
1360 */
1361static kprobe_opcode_t __kprobes *kprobe_addr(struct kprobe *p)
1362{
1363        kprobe_opcode_t *addr = p->addr;
1364
1365        if ((p->symbol_name && p->addr) ||
1366            (!p->symbol_name && !p->addr))
1367                goto invalid;
1368
1369        if (p->symbol_name) {
1370                kprobe_lookup_name(p->symbol_name, addr);
1371                if (!addr)
1372                        return ERR_PTR(-ENOENT);
1373        }
1374
1375        addr = (kprobe_opcode_t *)(((char *)addr) + p->offset);
1376        if (addr)
1377                return addr;
1378
1379invalid:
1380        return ERR_PTR(-EINVAL);
1381}
1382
1383/* Check passed kprobe is valid and return kprobe in kprobe_table. */
1384static struct kprobe * __kprobes __get_valid_kprobe(struct kprobe *p)
1385{
1386        struct kprobe *ap, *list_p;
1387
1388        ap = get_kprobe(p->addr);
1389        if (unlikely(!ap))
1390                return NULL;
1391
1392        if (p != ap) {
1393                list_for_each_entry_rcu(list_p, &ap->list, list)
1394                        if (list_p == p)
1395                        /* kprobe p is a valid probe */
1396                                goto valid;
1397                return NULL;
1398        }
1399valid:
1400        return ap;
1401}
1402
1403/* Return error if the kprobe is being re-registered */
1404static inline int check_kprobe_rereg(struct kprobe *p)
1405{
1406        int ret = 0;
1407
1408        mutex_lock(&kprobe_mutex);
1409        if (__get_valid_kprobe(p))
1410                ret = -EINVAL;
1411        mutex_unlock(&kprobe_mutex);
1412
1413        return ret;
1414}
1415
1416static __kprobes int check_kprobe_address_safe(struct kprobe *p,
1417                                               struct module **probed_mod)
1418{
1419        int ret = 0;
1420        unsigned long ftrace_addr;
1421
1422        /*
1423         * If the address is located on a ftrace nop, set the
1424         * breakpoint to the following instruction.
1425         */
1426        ftrace_addr = ftrace_location((unsigned long)p->addr);
1427        if (ftrace_addr) {
1428#ifdef CONFIG_KPROBES_ON_FTRACE
1429                /* Given address is not on the instruction boundary */
1430                if ((unsigned long)p->addr != ftrace_addr)
1431                        return -EILSEQ;
1432                p->flags |= KPROBE_FLAG_FTRACE;
1433#else   /* !CONFIG_KPROBES_ON_FTRACE */
1434                return -EINVAL;
1435#endif
1436        }
1437
1438        jump_label_lock();
1439        preempt_disable();
1440
1441        /* Ensure it is not in reserved area nor out of text */
1442        if (!kernel_text_address((unsigned long) p->addr) ||
1443            in_kprobes_functions((unsigned long) p->addr) ||
1444            jump_label_text_reserved(p->addr, p->addr)) {
1445                ret = -EINVAL;
1446                goto out;
1447        }
1448
1449        /* Check if are we probing a module */
1450        *probed_mod = __module_text_address((unsigned long) p->addr);
1451        if (*probed_mod) {
1452                /*
1453                 * We must hold a refcount of the probed module while updating
1454                 * its code to prohibit unexpected unloading.
1455                 */
1456                if (unlikely(!try_module_get(*probed_mod))) {
1457                        ret = -ENOENT;
1458                        goto out;
1459                }
1460
1461                /*
1462                 * If the module freed .init.text, we couldn't insert
1463                 * kprobes in there.
1464                 */
1465                if (within_module_init((unsigned long)p->addr, *probed_mod) &&
1466                    (*probed_mod)->state != MODULE_STATE_COMING) {
1467                        module_put(*probed_mod);
1468                        *probed_mod = NULL;
1469                        ret = -ENOENT;
1470                }
1471        }
1472out:
1473        preempt_enable();
1474        jump_label_unlock();
1475
1476        return ret;
1477}
1478
1479int __kprobes register_kprobe(struct kprobe *p)
1480{
1481        int ret;
1482        struct kprobe *old_p;
1483        struct module *probed_mod;
1484        kprobe_opcode_t *addr;
1485
1486        /* Adjust probe address from symbol */
1487        addr = kprobe_addr(p);
1488        if (IS_ERR(addr))
1489                return PTR_ERR(addr);
1490        p->addr = addr;
1491
1492        ret = check_kprobe_rereg(p);
1493        if (ret)
1494                return ret;
1495
1496        /* User can pass only KPROBE_FLAG_DISABLED to register_kprobe */
1497        p->flags &= KPROBE_FLAG_DISABLED;
1498        p->nmissed = 0;
1499        INIT_LIST_HEAD(&p->list);
1500
1501        ret = check_kprobe_address_safe(p, &probed_mod);
1502        if (ret)
1503                return ret;
1504
1505        mutex_lock(&kprobe_mutex);
1506
1507        old_p = get_kprobe(p->addr);
1508        if (old_p) {
1509                /* Since this may unoptimize old_p, locking text_mutex. */
1510                ret = register_aggr_kprobe(old_p, p);
1511                goto out;
1512        }
1513
1514        mutex_lock(&text_mutex);        /* Avoiding text modification */
1515        ret = prepare_kprobe(p);
1516        mutex_unlock(&text_mutex);
1517        if (ret)
1518                goto out;
1519
1520        INIT_HLIST_NODE(&p->hlist);
1521        hlist_add_head_rcu(&p->hlist,
1522                       &kprobe_table[hash_ptr(p->addr, KPROBE_HASH_BITS)]);
1523
1524        if (!kprobes_all_disarmed && !kprobe_disabled(p))
1525                arm_kprobe(p);
1526
1527        /* Try to optimize kprobe */
1528        try_to_optimize_kprobe(p);
1529
1530out:
1531        mutex_unlock(&kprobe_mutex);
1532
1533        if (probed_mod)
1534                module_put(probed_mod);
1535
1536        return ret;
1537}
1538EXPORT_SYMBOL_GPL(register_kprobe);
1539
1540/* Check if all probes on the aggrprobe are disabled */
1541static int __kprobes aggr_kprobe_disabled(struct kprobe *ap)
1542{
1543        struct kprobe *kp;
1544
1545        list_for_each_entry_rcu(kp, &ap->list, list)
1546                if (!kprobe_disabled(kp))
1547                        /*
1548                         * There is an active probe on the list.
1549                         * We can't disable this ap.
1550                         */
1551                        return 0;
1552
1553        return 1;
1554}
1555
1556/* Disable one kprobe: Make sure called under kprobe_mutex is locked */
1557static struct kprobe *__kprobes __disable_kprobe(struct kprobe *p)
1558{
1559        struct kprobe *orig_p;
1560
1561        /* Get an original kprobe for return */
1562        orig_p = __get_valid_kprobe(p);
1563        if (unlikely(orig_p == NULL))
1564                return NULL;
1565
1566        if (!kprobe_disabled(p)) {
1567                /* Disable probe if it is a child probe */
1568                if (p != orig_p)
1569                        p->flags |= KPROBE_FLAG_DISABLED;
1570
1571                /* Try to disarm and disable this/parent probe */
1572                if (p == orig_p || aggr_kprobe_disabled(orig_p)) {
1573                        disarm_kprobe(orig_p, true);
1574                        orig_p->flags |= KPROBE_FLAG_DISABLED;
1575                }
1576        }
1577
1578        return orig_p;
1579}
1580
1581/*
1582 * Unregister a kprobe without a scheduler synchronization.
1583 */
1584static int __kprobes __unregister_kprobe_top(struct kprobe *p)
1585{
1586        struct kprobe *ap, *list_p;
1587
1588        /* Disable kprobe. This will disarm it if needed. */
1589        ap = __disable_kprobe(p);
1590        if (ap == NULL)
1591                return -EINVAL;
1592
1593        if (ap == p)
1594                /*
1595                 * This probe is an independent(and non-optimized) kprobe
1596                 * (not an aggrprobe). Remove from the hash list.
1597                 */
1598                goto disarmed;
1599
1600        /* Following process expects this probe is an aggrprobe */
1601        WARN_ON(!kprobe_aggrprobe(ap));
1602
1603        if (list_is_singular(&ap->list) && kprobe_disarmed(ap))
1604                /*
1605                 * !disarmed could be happen if the probe is under delayed
1606                 * unoptimizing.
1607                 */
1608                goto disarmed;
1609        else {
1610                /* If disabling probe has special handlers, update aggrprobe */
1611                if (p->break_handler && !kprobe_gone(p))
1612                        ap->break_handler = NULL;
1613                if (p->post_handler && !kprobe_gone(p)) {
1614                        list_for_each_entry_rcu(list_p, &ap->list, list) {
1615                                if ((list_p != p) && (list_p->post_handler))
1616                                        goto noclean;
1617                        }
1618                        ap->post_handler = NULL;
1619                }
1620noclean:
1621                /*
1622                 * Remove from the aggrprobe: this path will do nothing in
1623                 * __unregister_kprobe_bottom().
1624                 */
1625                list_del_rcu(&p->list);
1626                if (!kprobe_disabled(ap) && !kprobes_all_disarmed)
1627                        /*
1628                         * Try to optimize this probe again, because post
1629                         * handler may have been changed.
1630                         */
1631                        optimize_kprobe(ap);
1632        }
1633        return 0;
1634
1635disarmed:
1636        BUG_ON(!kprobe_disarmed(ap));
1637        hlist_del_rcu(&ap->hlist);
1638        return 0;
1639}
1640
1641static void __kprobes __unregister_kprobe_bottom(struct kprobe *p)
1642{
1643        struct kprobe *ap;
1644
1645        if (list_empty(&p->list))
1646                /* This is an independent kprobe */
1647                arch_remove_kprobe(p);
1648        else if (list_is_singular(&p->list)) {
1649                /* This is the last child of an aggrprobe */
1650                ap = list_entry(p->list.next, struct kprobe, list);
1651                list_del(&p->list);
1652                free_aggr_kprobe(ap);
1653        }
1654        /* Otherwise, do nothing. */
1655}
1656
1657int __kprobes register_kprobes(struct kprobe **kps, int num)
1658{
1659        int i, ret = 0;
1660
1661        if (num <= 0)
1662                return -EINVAL;
1663        for (i = 0; i < num; i++) {
1664                ret = register_kprobe(kps[i]);
1665                if (ret < 0) {
1666                        if (i > 0)
1667                                unregister_kprobes(kps, i);
1668                        break;
1669                }
1670        }
1671        return ret;
1672}
1673EXPORT_SYMBOL_GPL(register_kprobes);
1674
1675void __kprobes unregister_kprobe(struct kprobe *p)
1676{
1677        unregister_kprobes(&p, 1);
1678}
1679EXPORT_SYMBOL_GPL(unregister_kprobe);
1680
1681void __kprobes unregister_kprobes(struct kprobe **kps, int num)
1682{
1683        int i;
1684
1685        if (num <= 0)
1686                return;
1687        mutex_lock(&kprobe_mutex);
1688        for (i = 0; i < num; i++)
1689                if (__unregister_kprobe_top(kps[i]) < 0)
1690                        kps[i]->addr = NULL;
1691        mutex_unlock(&kprobe_mutex);
1692
1693        synchronize_sched();
1694        for (i = 0; i < num; i++)
1695                if (kps[i]->addr)
1696                        __unregister_kprobe_bottom(kps[i]);
1697}
1698EXPORT_SYMBOL_GPL(unregister_kprobes);
1699
1700static struct notifier_block kprobe_exceptions_nb = {
1701        .notifier_call = kprobe_exceptions_notify,
1702        .priority = 0x7fffffff /* we need to be notified first */
1703};
1704
1705unsigned long __weak arch_deref_entry_point(void *entry)
1706{
1707        return (unsigned long)entry;
1708}
1709
1710int __kprobes register_jprobes(struct jprobe **jps, int num)
1711{
1712        struct jprobe *jp;
1713        int ret = 0, i;
1714
1715        if (num <= 0)
1716                return -EINVAL;
1717        for (i = 0; i < num; i++) {
1718                unsigned long addr, offset;
1719                jp = jps[i];
1720                addr = arch_deref_entry_point(jp->entry);
1721
1722                /* Verify probepoint is a function entry point */
1723                if (kallsyms_lookup_size_offset(addr, NULL, &offset) &&
1724                    offset == 0) {
1725                        jp->kp.pre_handler = setjmp_pre_handler;
1726                        jp->kp.break_handler = longjmp_break_handler;
1727                        ret = register_kprobe(&jp->kp);
1728                } else
1729                        ret = -EINVAL;
1730
1731                if (ret < 0) {
1732                        if (i > 0)
1733                                unregister_jprobes(jps, i);
1734                        break;
1735                }
1736        }
1737        return ret;
1738}
1739EXPORT_SYMBOL_GPL(register_jprobes);
1740
1741int __kprobes register_jprobe(struct jprobe *jp)
1742{
1743        return register_jprobes(&jp, 1);
1744}
1745EXPORT_SYMBOL_GPL(register_jprobe);
1746
1747void __kprobes unregister_jprobe(struct jprobe *jp)
1748{
1749        unregister_jprobes(&jp, 1);
1750}
1751EXPORT_SYMBOL_GPL(unregister_jprobe);
1752
1753void __kprobes unregister_jprobes(struct jprobe **jps, int num)
1754{
1755        int i;
1756
1757        if (num <= 0)
1758                return;
1759        mutex_lock(&kprobe_mutex);
1760        for (i = 0; i < num; i++)
1761                if (__unregister_kprobe_top(&jps[i]->kp) < 0)
1762                        jps[i]->kp.addr = NULL;
1763        mutex_unlock(&kprobe_mutex);
1764
1765        synchronize_sched();
1766        for (i = 0; i < num; i++) {
1767                if (jps[i]->kp.addr)
1768                        __unregister_kprobe_bottom(&jps[i]->kp);
1769        }
1770}
1771EXPORT_SYMBOL_GPL(unregister_jprobes);
1772
1773#ifdef CONFIG_KRETPROBES
1774/*
1775 * This kprobe pre_handler is registered with every kretprobe. When probe
1776 * hits it will set up the return probe.
1777 */
1778static int __kprobes pre_handler_kretprobe(struct kprobe *p,
1779                                           struct pt_regs *regs)
1780{
1781        struct kretprobe *rp = container_of(p, struct kretprobe, kp);
1782        unsigned long hash, flags = 0;
1783        struct kretprobe_instance *ri;
1784
1785        /*TODO: consider to only swap the RA after the last pre_handler fired */
1786        hash = hash_ptr(current, KPROBE_HASH_BITS);
1787        raw_spin_lock_irqsave(&rp->lock, flags);
1788        if (!hlist_empty(&rp->free_instances)) {
1789                ri = hlist_entry(rp->free_instances.first,
1790                                struct kretprobe_instance, hlist);
1791                hlist_del(&ri->hlist);
1792                raw_spin_unlock_irqrestore(&rp->lock, flags);
1793
1794                ri->rp = rp;
1795                ri->task = current;
1796
1797                if (rp->entry_handler && rp->entry_handler(ri, regs)) {
1798                        raw_spin_lock_irqsave(&rp->lock, flags);
1799                        hlist_add_head(&ri->hlist, &rp->free_instances);
1800                        raw_spin_unlock_irqrestore(&rp->lock, flags);
1801                        return 0;
1802                }
1803
1804                arch_prepare_kretprobe(ri, regs);
1805
1806                /* XXX(hch): why is there no hlist_move_head? */
1807                INIT_HLIST_NODE(&ri->hlist);
1808                kretprobe_table_lock(hash, &flags);
1809                hlist_add_head(&ri->hlist, &kretprobe_inst_table[hash]);
1810                kretprobe_table_unlock(hash, &flags);
1811        } else {
1812                rp->nmissed++;
1813                raw_spin_unlock_irqrestore(&rp->lock, flags);
1814        }
1815        return 0;
1816}
1817
1818int __kprobes register_kretprobe(struct kretprobe *rp)
1819{
1820        int ret = 0;
1821        struct kretprobe_instance *inst;
1822        int i;
1823        void *addr;
1824
1825        if (kretprobe_blacklist_size) {
1826                addr = kprobe_addr(&rp->kp);
1827                if (IS_ERR(addr))
1828                        return PTR_ERR(addr);
1829
1830                for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
1831                        if (kretprobe_blacklist[i].addr == addr)
1832                                return -EINVAL;
1833                }
1834        }
1835
1836        rp->kp.pre_handler = pre_handler_kretprobe;
1837        rp->kp.post_handler = NULL;
1838        rp->kp.fault_handler = NULL;
1839        rp->kp.break_handler = NULL;
1840
1841        /* Pre-allocate memory for max kretprobe instances */
1842        if (rp->maxactive <= 0) {
1843#ifdef CONFIG_PREEMPT
1844                rp->maxactive = max_t(unsigned int, 10, 2*num_possible_cpus());
1845#else
1846                rp->maxactive = num_possible_cpus();
1847#endif
1848        }
1849        raw_spin_lock_init(&rp->lock);
1850        INIT_HLIST_HEAD(&rp->free_instances);
1851        for (i = 0; i < rp->maxactive; i++) {
1852                inst = kmalloc(sizeof(struct kretprobe_instance) +
1853                               rp->data_size, GFP_KERNEL);
1854                if (inst == NULL) {
1855                        free_rp_inst(rp);
1856                        return -ENOMEM;
1857                }
1858                INIT_HLIST_NODE(&inst->hlist);
1859                hlist_add_head(&inst->hlist, &rp->free_instances);
1860        }
1861
1862        rp->nmissed = 0;
1863        /* Establish function entry probe point */
1864        ret = register_kprobe(&rp->kp);
1865        if (ret != 0)
1866                free_rp_inst(rp);
1867        return ret;
1868}
1869EXPORT_SYMBOL_GPL(register_kretprobe);
1870
1871int __kprobes register_kretprobes(struct kretprobe **rps, int num)
1872{
1873        int ret = 0, i;
1874
1875        if (num <= 0)
1876                return -EINVAL;
1877        for (i = 0; i < num; i++) {
1878                ret = register_kretprobe(rps[i]);
1879                if (ret < 0) {
1880                        if (i > 0)
1881                                unregister_kretprobes(rps, i);
1882                        break;
1883                }
1884        }
1885        return ret;
1886}
1887EXPORT_SYMBOL_GPL(register_kretprobes);
1888
1889void __kprobes unregister_kretprobe(struct kretprobe *rp)
1890{
1891        unregister_kretprobes(&rp, 1);
1892}
1893EXPORT_SYMBOL_GPL(unregister_kretprobe);
1894
1895void __kprobes unregister_kretprobes(struct kretprobe **rps, int num)
1896{
1897        int i;
1898
1899        if (num <= 0)
1900                return;
1901        mutex_lock(&kprobe_mutex);
1902        for (i = 0; i < num; i++)
1903                if (__unregister_kprobe_top(&rps[i]->kp) < 0)
1904                        rps[i]->kp.addr = NULL;
1905        mutex_unlock(&kprobe_mutex);
1906
1907        synchronize_sched();
1908        for (i = 0; i < num; i++) {
1909                if (rps[i]->kp.addr) {
1910                        __unregister_kprobe_bottom(&rps[i]->kp);
1911                        cleanup_rp_inst(rps[i]);
1912                }
1913        }
1914}
1915EXPORT_SYMBOL_GPL(unregister_kretprobes);
1916
1917#else /* CONFIG_KRETPROBES */
1918int __kprobes register_kretprobe(struct kretprobe *rp)
1919{
1920        return -ENOSYS;
1921}
1922EXPORT_SYMBOL_GPL(register_kretprobe);
1923
1924int __kprobes register_kretprobes(struct kretprobe **rps, int num)
1925{
1926        return -ENOSYS;
1927}
1928EXPORT_SYMBOL_GPL(register_kretprobes);
1929
1930void __kprobes unregister_kretprobe(struct kretprobe *rp)
1931{
1932}
1933EXPORT_SYMBOL_GPL(unregister_kretprobe);
1934
1935void __kprobes unregister_kretprobes(struct kretprobe **rps, int num)
1936{
1937}
1938EXPORT_SYMBOL_GPL(unregister_kretprobes);
1939
1940static int __kprobes pre_handler_kretprobe(struct kprobe *p,
1941                                           struct pt_regs *regs)
1942{
1943        return 0;
1944}
1945
1946#endif /* CONFIG_KRETPROBES */
1947
1948/* Set the kprobe gone and remove its instruction buffer. */
1949static void __kprobes kill_kprobe(struct kprobe *p)
1950{
1951        struct kprobe *kp;
1952
1953        p->flags |= KPROBE_FLAG_GONE;
1954        if (kprobe_aggrprobe(p)) {
1955                /*
1956                 * If this is an aggr_kprobe, we have to list all the
1957                 * chained probes and mark them GONE.
1958                 */
1959                list_for_each_entry_rcu(kp, &p->list, list)
1960                        kp->flags |= KPROBE_FLAG_GONE;
1961                p->post_handler = NULL;
1962                p->break_handler = NULL;
1963                kill_optimized_kprobe(p);
1964        }
1965        /*
1966         * Here, we can remove insn_slot safely, because no thread calls
1967         * the original probed function (which will be freed soon) any more.
1968         */
1969        arch_remove_kprobe(p);
1970}
1971
1972/* Disable one kprobe */
1973int __kprobes disable_kprobe(struct kprobe *kp)
1974{
1975        int ret = 0;
1976
1977        mutex_lock(&kprobe_mutex);
1978
1979        /* Disable this kprobe */
1980        if (__disable_kprobe(kp) == NULL)
1981                ret = -EINVAL;
1982
1983        mutex_unlock(&kprobe_mutex);
1984        return ret;
1985}
1986EXPORT_SYMBOL_GPL(disable_kprobe);
1987
1988/* Enable one kprobe */
1989int __kprobes enable_kprobe(struct kprobe *kp)
1990{
1991        int ret = 0;
1992        struct kprobe *p;
1993
1994        mutex_lock(&kprobe_mutex);
1995
1996        /* Check whether specified probe is valid. */
1997        p = __get_valid_kprobe(kp);
1998        if (unlikely(p == NULL)) {
1999                ret = -EINVAL;
2000                goto out;
2001        }
2002
2003        if (kprobe_gone(kp)) {
2004                /* This kprobe has gone, we couldn't enable it. */
2005                ret = -EINVAL;
2006                goto out;
2007        }
2008
2009        if (p != kp)
2010                kp->flags &= ~KPROBE_FLAG_DISABLED;
2011
2012        if (!kprobes_all_disarmed && kprobe_disabled(p)) {
2013                p->flags &= ~KPROBE_FLAG_DISABLED;
2014                arm_kprobe(p);
2015        }
2016out:
2017        mutex_unlock(&kprobe_mutex);
2018        return ret;
2019}
2020EXPORT_SYMBOL_GPL(enable_kprobe);
2021
2022void __kprobes dump_kprobe(struct kprobe *kp)
2023{
2024        printk(KERN_WARNING "Dumping kprobe:\n");
2025        printk(KERN_WARNING "Name: %s\nAddress: %p\nOffset: %x\n",
2026               kp->symbol_name, kp->addr, kp->offset);
2027}
2028
2029/* Module notifier call back, checking kprobes on the module */
2030static int __kprobes kprobes_module_callback(struct notifier_block *nb,
2031                                             unsigned long val, void *data)
2032{
2033        struct module *mod = data;
2034        struct hlist_head *head;
2035        struct kprobe *p;
2036        unsigned int i;
2037        int checkcore = (val == MODULE_STATE_GOING);
2038
2039        if (val != MODULE_STATE_GOING && val != MODULE_STATE_LIVE)
2040                return NOTIFY_DONE;
2041
2042        /*
2043         * When MODULE_STATE_GOING was notified, both of module .text and
2044         * .init.text sections would be freed. When MODULE_STATE_LIVE was
2045         * notified, only .init.text section would be freed. We need to
2046         * disable kprobes which have been inserted in the sections.
2047         */
2048        mutex_lock(&kprobe_mutex);
2049        for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2050                head = &kprobe_table[i];
2051                hlist_for_each_entry_rcu(p, head, hlist)
2052                        if (within_module_init((unsigned long)p->addr, mod) ||
2053                            (checkcore &&
2054                             within_module_core((unsigned long)p->addr, mod))) {
2055                                /*
2056                                 * The vaddr this probe is installed will soon
2057                                 * be vfreed buy not synced to disk. Hence,
2058                                 * disarming the breakpoint isn't needed.
2059                                 */
2060                                kill_kprobe(p);
2061                        }
2062        }
2063        mutex_unlock(&kprobe_mutex);
2064        return NOTIFY_DONE;
2065}
2066
2067static struct notifier_block kprobe_module_nb = {
2068        .notifier_call = kprobes_module_callback,
2069        .priority = 0
2070};
2071
2072static int __init init_kprobes(void)
2073{
2074        int i, err = 0;
2075        unsigned long offset = 0, size = 0;
2076        char *modname, namebuf[128];
2077        const char *symbol_name;
2078        void *addr;
2079        struct kprobe_blackpoint *kb;
2080
2081        /* FIXME allocate the probe table, currently defined statically */
2082        /* initialize all list heads */
2083        for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2084                INIT_HLIST_HEAD(&kprobe_table[i]);
2085                INIT_HLIST_HEAD(&kretprobe_inst_table[i]);
2086                raw_spin_lock_init(&(kretprobe_table_locks[i].lock));
2087        }
2088
2089        /*
2090         * Lookup and populate the kprobe_blacklist.
2091         *
2092         * Unlike the kretprobe blacklist, we'll need to determine
2093         * the range of addresses that belong to the said functions,
2094         * since a kprobe need not necessarily be at the beginning
2095         * of a function.
2096         */
2097        for (kb = kprobe_blacklist; kb->name != NULL; kb++) {
2098                kprobe_lookup_name(kb->name, addr);
2099                if (!addr)
2100                        continue;
2101
2102                kb->start_addr = (unsigned long)addr;
2103                symbol_name = kallsyms_lookup(kb->start_addr,
2104                                &size, &offset, &modname, namebuf);
2105                if (!symbol_name)
2106                        kb->range = 0;
2107                else
2108                        kb->range = size;
2109        }
2110
2111        if (kretprobe_blacklist_size) {
2112                /* lookup the function address from its name */
2113                for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
2114                        kprobe_lookup_name(kretprobe_blacklist[i].name,
2115                                           kretprobe_blacklist[i].addr);
2116                        if (!kretprobe_blacklist[i].addr)
2117                                printk("kretprobe: lookup failed: %s\n",
2118                                       kretprobe_blacklist[i].name);
2119                }
2120        }
2121
2122#if defined(CONFIG_OPTPROBES)
2123#if defined(__ARCH_WANT_KPROBES_INSN_SLOT)
2124        /* Init kprobe_optinsn_slots */
2125        kprobe_optinsn_slots.insn_size = MAX_OPTINSN_SIZE;
2126#endif
2127        /* By default, kprobes can be optimized */
2128        kprobes_allow_optimization = true;
2129#endif
2130
2131        /* By default, kprobes are armed */
2132        kprobes_all_disarmed = false;
2133
2134        err = arch_init_kprobes();
2135        if (!err)
2136                err = register_die_notifier(&kprobe_exceptions_nb);
2137        if (!err)
2138                err = register_module_notifier(&kprobe_module_nb);
2139
2140        kprobes_initialized = (err == 0);
2141
2142        if (!err)
2143                init_test_probes();
2144        return err;
2145}
2146
2147#ifdef CONFIG_DEBUG_FS
2148static void __kprobes report_probe(struct seq_file *pi, struct kprobe *p,
2149                const char *sym, int offset, char *modname, struct kprobe *pp)
2150{
2151        char *kprobe_type;
2152
2153        if (p->pre_handler == pre_handler_kretprobe)
2154                kprobe_type = "r";
2155        else if (p->pre_handler == setjmp_pre_handler)
2156                kprobe_type = "j";
2157        else
2158                kprobe_type = "k";
2159
2160        if (sym)
2161                seq_printf(pi, "%p  %s  %s+0x%x  %s ",
2162                        p->addr, kprobe_type, sym, offset,
2163                        (modname ? modname : " "));
2164        else
2165                seq_printf(pi, "%p  %s  %p ",
2166                        p->addr, kprobe_type, p->addr);
2167
2168        if (!pp)
2169                pp = p;
2170        seq_printf(pi, "%s%s%s%s\n",
2171                (kprobe_gone(p) ? "[GONE]" : ""),
2172                ((kprobe_disabled(p) && !kprobe_gone(p)) ?  "[DISABLED]" : ""),
2173                (kprobe_optimized(pp) ? "[OPTIMIZED]" : ""),
2174                (kprobe_ftrace(pp) ? "[FTRACE]" : ""));
2175}
2176
2177static void __kprobes *kprobe_seq_start(struct seq_file *f, loff_t *pos)
2178{
2179        return (*pos < KPROBE_TABLE_SIZE) ? pos : NULL;
2180}
2181
2182static void __kprobes *kprobe_seq_next(struct seq_file *f, void *v, loff_t *pos)
2183{
2184        (*pos)++;
2185        if (*pos >= KPROBE_TABLE_SIZE)
2186                return NULL;
2187        return pos;
2188}
2189
2190static void __kprobes kprobe_seq_stop(struct seq_file *f, void *v)
2191{
2192        /* Nothing to do */
2193}
2194
2195static int __kprobes show_kprobe_addr(struct seq_file *pi, void *v)
2196{
2197        struct hlist_head *head;
2198        struct kprobe *p, *kp;
2199        const char *sym = NULL;
2200        unsigned int i = *(loff_t *) v;
2201        unsigned long offset = 0;
2202        char *modname, namebuf[128];
2203
2204        head = &kprobe_table[i];
2205        preempt_disable();
2206        hlist_for_each_entry_rcu(p, head, hlist) {
2207                sym = kallsyms_lookup((unsigned long)p->addr, NULL,
2208                                        &offset, &modname, namebuf);
2209                if (kprobe_aggrprobe(p)) {
2210                        list_for_each_entry_rcu(kp, &p->list, list)
2211                                report_probe(pi, kp, sym, offset, modname, p);
2212                } else
2213                        report_probe(pi, p, sym, offset, modname, NULL);
2214        }
2215        preempt_enable();
2216        return 0;
2217}
2218
2219static const struct seq_operations kprobes_seq_ops = {
2220        .start = kprobe_seq_start,
2221        .next  = kprobe_seq_next,
2222        .stop  = kprobe_seq_stop,
2223        .show  = show_kprobe_addr
2224};
2225
2226static int __kprobes kprobes_open(struct inode *inode, struct file *filp)
2227{
2228        return seq_open(filp, &kprobes_seq_ops);
2229}
2230
2231static const struct file_operations debugfs_kprobes_operations = {
2232        .open           = kprobes_open,
2233        .read           = seq_read,
2234        .llseek         = seq_lseek,
2235        .release        = seq_release,
2236};
2237
2238static void __kprobes arm_all_kprobes(void)
2239{
2240        struct hlist_head *head;
2241        struct kprobe *p;
2242        unsigned int i;
2243
2244        mutex_lock(&kprobe_mutex);
2245
2246        /* If kprobes are armed, just return */
2247        if (!kprobes_all_disarmed)
2248                goto already_enabled;
2249
2250        /* Arming kprobes doesn't optimize kprobe itself */
2251        for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2252                head = &kprobe_table[i];
2253                hlist_for_each_entry_rcu(p, head, hlist)
2254                        if (!kprobe_disabled(p))
2255                                arm_kprobe(p);
2256        }
2257
2258        kprobes_all_disarmed = false;
2259        printk(KERN_INFO "Kprobes globally enabled\n");
2260
2261already_enabled:
2262        mutex_unlock(&kprobe_mutex);
2263        return;
2264}
2265
2266static void __kprobes disarm_all_kprobes(void)
2267{
2268        struct hlist_head *head;
2269        struct kprobe *p;
2270        unsigned int i;
2271
2272        mutex_lock(&kprobe_mutex);
2273
2274        /* If kprobes are already disarmed, just return */
2275        if (kprobes_all_disarmed) {
2276                mutex_unlock(&kprobe_mutex);
2277                return;
2278        }
2279
2280        kprobes_all_disarmed = true;
2281        printk(KERN_INFO "Kprobes globally disabled\n");
2282
2283        for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2284                head = &kprobe_table[i];
2285                hlist_for_each_entry_rcu(p, head, hlist) {
2286                        if (!arch_trampoline_kprobe(p) && !kprobe_disabled(p))
2287                                disarm_kprobe(p, false);
2288                }
2289        }
2290        mutex_unlock(&kprobe_mutex);
2291
2292        /* Wait for disarming all kprobes by optimizer */
2293        wait_for_kprobe_optimizer();
2294}
2295
2296/*
2297 * XXX: The debugfs bool file interface doesn't allow for callbacks
2298 * when the bool state is switched. We can reuse that facility when
2299 * available
2300 */
2301static ssize_t read_enabled_file_bool(struct file *file,
2302               char __user *user_buf, size_t count, loff_t *ppos)
2303{
2304        char buf[3];
2305
2306        if (!kprobes_all_disarmed)
2307                buf[0] = '1';
2308        else
2309                buf[0] = '0';
2310        buf[1] = '\n';
2311        buf[2] = 0x00;
2312        return simple_read_from_buffer(user_buf, count, ppos, buf, 2);
2313}
2314
2315static ssize_t write_enabled_file_bool(struct file *file,
2316               const char __user *user_buf, size_t count, loff_t *ppos)
2317{
2318        char buf[32];
2319        size_t buf_size;
2320
2321        buf_size = min(count, (sizeof(buf)-1));
2322        if (copy_from_user(buf, user_buf, buf_size))
2323                return -EFAULT;
2324
2325        switch (buf[0]) {
2326        case 'y':
2327        case 'Y':
2328        case '1':
2329                arm_all_kprobes();
2330                break;
2331        case 'n':
2332        case 'N':
2333        case '0':
2334                disarm_all_kprobes();
2335                break;
2336        }
2337
2338        return count;
2339}
2340
2341static const struct file_operations fops_kp = {
2342        .read =         read_enabled_file_bool,
2343        .write =        write_enabled_file_bool,
2344        .llseek =       default_llseek,
2345};
2346
2347static int __kprobes debugfs_kprobe_init(void)
2348{
2349        struct dentry *dir, *file;
2350        unsigned int value = 1;
2351
2352        dir = debugfs_create_dir("kprobes", NULL);
2353        if (!dir)
2354                return -ENOMEM;
2355
2356        file = debugfs_create_file("list", 0444, dir, NULL,
2357                                &debugfs_kprobes_operations);
2358        if (!file) {
2359                debugfs_remove(dir);
2360                return -ENOMEM;
2361        }
2362
2363        file = debugfs_create_file("enabled", 0600, dir,
2364                                        &value, &fops_kp);
2365        if (!file) {
2366                debugfs_remove(dir);
2367                return -ENOMEM;
2368        }
2369
2370        return 0;
2371}
2372
2373late_initcall(debugfs_kprobe_init);
2374#endif /* CONFIG_DEBUG_FS */
2375
2376module_init(init_kprobes);
2377
2378/* defined in arch/.../kernel/kprobes.c */
2379EXPORT_SYMBOL_GPL(jprobe_return);
2380
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.