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/kdebug.h>
  46
  47 #include <asm-generic/sections.h>
  48 #include <asm/cacheflush.h>
  49 #include <asm/errno.h>
  50 #include <asm/uaccess.h>
  51
  52 #define KPROBE_HASH_BITS 6
  53 #define KPROBE_TABLE_SIZE (1 << KPROBE_HASH_BITS)
  54
  55
  56 /*
  57  * Some oddball architectures like 64bit powerpc have function descriptors
  58  * so this must be overridable.
  59  */
  60 #ifndef kprobe_lookup_name
  61 #define kprobe_lookup_name(name, addr) \
  62         addr = ((kprobe_opcode_t *)(kallsyms_lookup_name(name)))
  63 #endif
  64
  65 static int kprobes_initialized;
  66 static struct hlist_head kprobe_table[KPROBE_TABLE_SIZE];
  67 static struct hlist_head kretprobe_inst_table[KPROBE_TABLE_SIZE];
  68
  69 /* NOTE: change this value only with kprobe_mutex held */
  70 static bool kprobe_enabled;
  71
  72 static DEFINE_MUTEX(kprobe_mutex);      /* Protects kprobe_table */
  73 static DEFINE_PER_CPU(struct kprobe *, kprobe_instance) = NULL;
  74 static struct {
  75         spinlock_t lock ____cacheline_aligned_in_smp;
  76 } kretprobe_table_locks[KPROBE_TABLE_SIZE];
  77
  78 static spinlock_t *kretprobe_table_lock_ptr(unsigned long hash)
  79 {
  80         return &(kretprobe_table_locks[hash].lock);
  81 }
  82
  83 /*
  84  * Normally, functions that we'd want to prohibit kprobes in, are marked
  85  * __kprobes. But, there are cases where such functions already belong to
  86  * a different section (__sched for preempt_schedule)
  87  *
  88  * For such cases, we now have a blacklist
  89  */
  90 static struct kprobe_blackpoint kprobe_blacklist[] = {
  91         {"preempt_schedule",},
  92         {NULL}    /* Terminator */
  93 };
  94
  95 #ifdef __ARCH_WANT_KPROBES_INSN_SLOT
  96 /*
  97  * kprobe->ainsn.insn points to the copy of the instruction to be
  98  * single-stepped. x86_64, POWER4 and above have no-exec support and
  99  * stepping on the instruction on a vmalloced/kmalloced/data page
 100  * is a recipe for disaster
 101  */
 102 #define INSNS_PER_PAGE  (PAGE_SIZE/(MAX_INSN_SIZE * sizeof(kprobe_opcode_t)))
 103
 104 struct kprobe_insn_page {
 105         struct hlist_node hlist;
 106         kprobe_opcode_t *insns;         /* Page of instruction slots */
 107         char slot_used[INSNS_PER_PAGE];
 108         int nused;
 109         int ngarbage;
 110 };
 111
 112 enum kprobe_slot_state {
 113         SLOT_CLEAN = 0,
 114         SLOT_DIRTY = 1,
 115         SLOT_USED = 2,
 116 };
 117
 118 static DEFINE_MUTEX(kprobe_insn_mutex); /* Protects kprobe_insn_pages */
 119 static struct hlist_head kprobe_insn_pages;
 120 static int kprobe_garbage_slots;
 121 static int collect_garbage_slots(void);
 122
 123 static int __kprobes check_safety(void)
 124 {
 125         int ret = 0;
 126 #if defined(CONFIG_PREEMPT) && defined(CONFIG_PM)
 127         ret = freeze_processes();
 128         if (ret == 0) {
 129                 struct task_struct *p, *q;
 130                 do_each_thread(p, q) {
 131                         if (p != current && p->state == TASK_RUNNING &&
 132                             p->pid != 0) {
 133                                 printk("Check failed: %s is running\n",p->comm);
 134                                 ret = -1;
 135                                 goto loop_end;
 136                         }
 137                 } while_each_thread(p, q);
 138         }
 139 loop_end:
 140         thaw_processes();
 141 #else
 142         synchronize_sched();
 143 #endif
 144         return ret;
 145 }
 146
 147 /**
 148  * __get_insn_slot() - Find a slot on an executable page for an instruction.
 149  * We allocate an executable page if there's no room on existing ones.
 150  */
 151 static kprobe_opcode_t __kprobes *__get_insn_slot(void)
 152 {
 153         struct kprobe_insn_page *kip;
 154         struct hlist_node *pos;
 155
 156  retry:
 157         hlist_for_each_entry(kip, pos, &kprobe_insn_pages, hlist) {
 158                 if (kip->nused < INSNS_PER_PAGE) {
 159                         int i;
 160                         for (i = 0; i < INSNS_PER_PAGE; i++) {
 161                                 if (kip->slot_used[i] == SLOT_CLEAN) {
 162                                         kip->slot_used[i] = SLOT_USED;
 163                                         kip->nused++;
 164                                         return kip->insns + (i * MAX_INSN_SIZE);
 165                                 }
 166                         }
 167                         /* Surprise!  No unused slots.  Fix kip->nused. */
 168                         kip->nused = INSNS_PER_PAGE;
 169                 }
 170         }
 171
 172         /* If there are any garbage slots, collect it and try again. */
 173         if (kprobe_garbage_slots && collect_garbage_slots() == 0) {
 174                 goto retry;
 175         }
 176         /* All out of space.  Need to allocate a new page. Use slot 0. */
 177         kip = kmalloc(sizeof(struct kprobe_insn_page), GFP_KERNEL);
 178         if (!kip)
 179                 return NULL;
 180
 181         /*
 182          * Use module_alloc so this page is within +/- 2GB of where the
 183          * kernel image and loaded module images reside. This is required
 184          * so x86_64 can correctly handle the %rip-relative fixups.
 185          */
 186         kip->insns = module_alloc(PAGE_SIZE);
 187         if (!kip->insns) {
 188                 kfree(kip);
 189                 return NULL;
 190         }
 191         INIT_HLIST_NODE(&kip->hlist);
 192         hlist_add_head(&kip->hlist, &kprobe_insn_pages);
 193         memset(kip->slot_used, SLOT_CLEAN, INSNS_PER_PAGE);
 194         kip->slot_used[0] = SLOT_USED;
 195         kip->nused = 1;
 196         kip->ngarbage = 0;
 197         return kip->insns;
 198 }
 199
 200 kprobe_opcode_t __kprobes *get_insn_slot(void)
 201 {
 202         kprobe_opcode_t *ret;
 203         mutex_lock(&kprobe_insn_mutex);
 204         ret = __get_insn_slot();
 205         mutex_unlock(&kprobe_insn_mutex);
 206         return ret;
 207 }
 208
 209 /* Return 1 if all garbages are collected, otherwise 0. */
 210 static int __kprobes collect_one_slot(struct kprobe_insn_page *kip, int idx)
 211 {
 212         kip->slot_used[idx] = SLOT_CLEAN;
 213         kip->nused--;
 214         if (kip->nused == 0) {
 215                 /*
 216                  * Page is no longer in use.  Free it unless
 217                  * it's the last one.  We keep the last one
 218                  * so as not to have to set it up again the
 219                  * next time somebody inserts a probe.
 220                  */
 221                 hlist_del(&kip->hlist);
 222                 if (hlist_empty(&kprobe_insn_pages)) {
 223                         INIT_HLIST_NODE(&kip->hlist);
 224                         hlist_add_head(&kip->hlist,
 225                                        &kprobe_insn_pages);
 226                 } else {
 227                         module_free(NULL, kip->insns);
 228                         kfree(kip);
 229                 }
 230                 return 1;
 231         }
 232         return 0;
 233 }
 234
 235 static int __kprobes collect_garbage_slots(void)
 236 {
 237         struct kprobe_insn_page *kip;
 238         struct hlist_node *pos, *next;
 239         int safety;
 240
 241         /* Ensure no-one is preepmted on the garbages */
 242         mutex_unlock(&kprobe_insn_mutex);
 243         safety = check_safety();
 244         mutex_lock(&kprobe_insn_mutex);
 245         if (safety != 0)
 246                 return -EAGAIN;
 247
 248         hlist_for_each_entry_safe(kip, pos, next, &kprobe_insn_pages, hlist) {
 249                 int i;
 250                 if (kip->ngarbage == 0)
 251                         continue;
 252                 kip->ngarbage = 0;      /* we will collect all garbages */
 253                 for (i = 0; i < INSNS_PER_PAGE; i++) {
 254                         if (kip->slot_used[i] == SLOT_DIRTY &&
 255                             collect_one_slot(kip, i))
 256                                 break;
 257                 }
 258         }
 259         kprobe_garbage_slots = 0;
 260         return 0;
 261 }
 262
 263 void __kprobes free_insn_slot(kprobe_opcode_t * slot, int dirty)
 264 {
 265         struct kprobe_insn_page *kip;
 266         struct hlist_node *pos;
 267
 268         mutex_lock(&kprobe_insn_mutex);
 269         hlist_for_each_entry(kip, pos, &kprobe_insn_pages, hlist) {
 270                 if (kip->insns <= slot &&
 271                     slot < kip->insns + (INSNS_PER_PAGE * MAX_INSN_SIZE)) {
 272                         int i = (slot - kip->insns) / MAX_INSN_SIZE;
 273                         if (dirty) {
 274                                 kip->slot_used[i] = SLOT_DIRTY;
 275                                 kip->ngarbage++;
 276                         } else {
 277                                 collect_one_slot(kip, i);
 278                         }
 279                         break;
 280                 }
 281         }
 282
 283         if (dirty && ++kprobe_garbage_slots > INSNS_PER_PAGE)
 284                 collect_garbage_slots();
 285
 286         mutex_unlock(&kprobe_insn_mutex);
 287 }
 288 #endif
 289
 290 /* We have preemption disabled.. so it is safe to use __ versions */
 291 static inline void set_kprobe_instance(struct kprobe *kp)
 292 {
 293         __get_cpu_var(kprobe_instance) = kp;
 294 }
 295
 296 static inline void reset_kprobe_instance(void)
 297 {
 298         __get_cpu_var(kprobe_instance) = NULL;
 299 }
 300
 301 /*
 302  * This routine is called either:
 303  *      - under the kprobe_mutex - during kprobe_[un]register()
 304  *                              OR
 305  *      - with preemption disabled - from arch/xxx/kernel/kprobes.c
 306  */
 307 struct kprobe __kprobes *get_kprobe(void *addr)
 308 {
 309         struct hlist_head *head;
 310         struct hlist_node *node;
 311         struct kprobe *p;
 312
 313         head = &kprobe_table[hash_ptr(addr, KPROBE_HASH_BITS)];
 314         hlist_for_each_entry_rcu(p, node, head, hlist) {
 315                 if (p->addr == addr)
 316                         return p;
 317         }
 318         return NULL;
 319 }
 320
 321 /*
 322  * Aggregate handlers for multiple kprobes support - these handlers
 323  * take care of invoking the individual kprobe handlers on p->list
 324  */
 325 static int __kprobes aggr_pre_handler(struct kprobe *p, struct pt_regs *regs)
 326 {
 327         struct kprobe *kp;
 328
 329         list_for_each_entry_rcu(kp, &p->list, list) {
 330                 if (kp->pre_handler && !kprobe_gone(kp)) {
 331                         set_kprobe_instance(kp);
 332                         if (kp->pre_handler(kp, regs))
 333                                 return 1;
 334                 }
 335                 reset_kprobe_instance();
 336         }
 337         return 0;
 338 }
 339
 340 static void __kprobes aggr_post_handler(struct kprobe *p, struct pt_regs *regs,
 341                                         unsigned long flags)
 342 {
 343         struct kprobe *kp;
 344
 345         list_for_each_entry_rcu(kp, &p->list, list) {
 346                 if (kp->post_handler && !kprobe_gone(kp)) {
 347                         set_kprobe_instance(kp);
 348                         kp->post_handler(kp, regs, flags);
 349                         reset_kprobe_instance();
 350                 }
 351         }
 352 }
 353
 354 static int __kprobes aggr_fault_handler(struct kprobe *p, struct pt_regs *regs,
 355                                         int trapnr)
 356 {
 357         struct kprobe *cur = __get_cpu_var(kprobe_instance);
 358
 359         /*
 360          * if we faulted "during" the execution of a user specified
 361          * probe handler, invoke just that probe's fault handler
 362          */
 363         if (cur && cur->fault_handler) {
 364                 if (cur->fault_handler(cur, regs, trapnr))
 365                         return 1;
 366         }
 367         return 0;
 368 }
 369
 370 static int __kprobes aggr_break_handler(struct kprobe *p, struct pt_regs *regs)
 371 {
 372         struct kprobe *cur = __get_cpu_var(kprobe_instance);
 373         int ret = 0;
 374
 375         if (cur && cur->break_handler) {
 376                 if (cur->break_handler(cur, regs))
 377                         ret = 1;
 378         }
 379         reset_kprobe_instance();
 380         return ret;
 381 }
 382
 383 /* Walks the list and increments nmissed count for multiprobe case */
 384 void __kprobes kprobes_inc_nmissed_count(struct kprobe *p)
 385 {
 386         struct kprobe *kp;
 387         if (p->pre_handler != aggr_pre_handler) {
 388                 p->nmissed++;
 389         } else {
 390                 list_for_each_entry_rcu(kp, &p->list, list)
 391                         kp->nmissed++;
 392         }
 393         return;
 394 }
 395
 396 void __kprobes recycle_rp_inst(struct kretprobe_instance *ri,
 397                                 struct hlist_head *head)
 398 {
 399         struct kretprobe *rp = ri->rp;
 400
 401         /* remove rp inst off the rprobe_inst_table */
 402         hlist_del(&ri->hlist);
 403         INIT_HLIST_NODE(&ri->hlist);
 404         if (likely(rp)) {
 405                 spin_lock(&rp->lock);
 406                 hlist_add_head(&ri->hlist, &rp->free_instances);
 407                 spin_unlock(&rp->lock);
 408         } else
 409                 /* Unregistering */
 410                 hlist_add_head(&ri->hlist, head);
 411 }
 412
 413 void __kprobes kretprobe_hash_lock(struct task_struct *tsk,
 414                          struct hlist_head **head, unsigned long *flags)
 415 {
 416         unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
 417         spinlock_t *hlist_lock;
 418
 419         *head = &kretprobe_inst_table[hash];
 420         hlist_lock = kretprobe_table_lock_ptr(hash);
 421         spin_lock_irqsave(hlist_lock, *flags);
 422 }
 423
 424 static void __kprobes kretprobe_table_lock(unsigned long hash,
 425         unsigned long *flags)
 426 {
 427         spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
 428         spin_lock_irqsave(hlist_lock, *flags);
 429 }
 430
 431 void __kprobes kretprobe_hash_unlock(struct task_struct *tsk,
 432         unsigned long *flags)
 433 {
 434         unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
 435         spinlock_t *hlist_lock;
 436
 437         hlist_lock = kretprobe_table_lock_ptr(hash);
 438         spin_unlock_irqrestore(hlist_lock, *flags);
 439 }
 440
 441 void __kprobes kretprobe_table_unlock(unsigned long hash, unsigned long *flags)
 442 {
 443         spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
 444         spin_unlock_irqrestore(hlist_lock, *flags);
 445 }
 446
 447 /*
 448  * This function is called from finish_task_switch when task tk becomes dead,
 449  * so that we can recycle any function-return probe instances associated
 450  * with this task. These left over instances represent probed functions
 451  * that have been called but will never return.
 452  */
 453 void __kprobes kprobe_flush_task(struct task_struct *tk)
 454 {
 455         struct kretprobe_instance *ri;
 456         struct hlist_head *head, empty_rp;
 457         struct hlist_node *node, *tmp;
 458         unsigned long hash, flags = 0;
 459
 460         if (unlikely(!kprobes_initialized))
 461                 /* Early boot.  kretprobe_table_locks not yet initialized. */
 462                 return;
 463
 464         hash = hash_ptr(tk, KPROBE_HASH_BITS);
 465         head = &kretprobe_inst_table[hash];
 466         kretprobe_table_lock(hash, &flags);
 467         hlist_for_each_entry_safe(ri, node, tmp, head, hlist) {
 468                 if (ri->task == tk)
 469                         recycle_rp_inst(ri, &empty_rp);
 470         }
 471         kretprobe_table_unlock(hash, &flags);
 472         INIT_HLIST_HEAD(&empty_rp);
 473         hlist_for_each_entry_safe(ri, node, tmp, &empty_rp, hlist) {
 474                 hlist_del(&ri->hlist);
 475                 kfree(ri);
 476         }
 477 }
 478
 479 static inline void free_rp_inst(struct kretprobe *rp)
 480 {
 481         struct kretprobe_instance *ri;
 482         struct hlist_node *pos, *next;
 483
 484         hlist_for_each_entry_safe(ri, pos, next, &rp->free_instances, hlist) {
 485                 hlist_del(&ri->hlist);
 486                 kfree(ri);
 487         }
 488 }
 489
 490 static void __kprobes cleanup_rp_inst(struct kretprobe *rp)
 491 {
 492         unsigned long flags, hash;
 493         struct kretprobe_instance *ri;
 494         struct hlist_node *pos, *next;
 495         struct hlist_head *head;
 496
 497         /* No race here */
 498         for (hash = 0; hash < KPROBE_TABLE_SIZE; hash++) {
 499                 kretprobe_table_lock(hash, &flags);
 500                 head = &kretprobe_inst_table[hash];
 501                 hlist_for_each_entry_safe(ri, pos, next, head, hlist) {
 502                         if (ri->rp == rp)
 503                                 ri->rp = NULL;
 504                 }
 505                 kretprobe_table_unlock(hash, &flags);
 506         }
 507         free_rp_inst(rp);
 508 }
 509
 510 /*
 511  * Keep all fields in the kprobe consistent
 512  */
 513 static inline void copy_kprobe(struct kprobe *old_p, struct kprobe *p)
 514 {
 515         memcpy(&p->opcode, &old_p->opcode, sizeof(kprobe_opcode_t));
 516         memcpy(&p->ainsn, &old_p->ainsn, sizeof(struct arch_specific_insn));
 517 }
 518
 519 /*
 520 * Add the new probe to old_p->list. Fail if this is the
 521 * second jprobe at the address - two jprobes can't coexist
 522 */
 523 static int __kprobes add_new_kprobe(struct kprobe *old_p, struct kprobe *p)
 524 {
 525         if (p->break_handler) {
 526                 if (old_p->break_handler)
 527                         return -EEXIST;
 528                 list_add_tail_rcu(&p->list, &old_p->list);
 529                 old_p->break_handler = aggr_break_handler;
 530         } else
 531                 list_add_rcu(&p->list, &old_p->list);
 532         if (p->post_handler && !old_p->post_handler)
 533                 old_p->post_handler = aggr_post_handler;
 534         return 0;
 535 }
 536
 537 /*
 538  * Fill in the required fields of the "manager kprobe". Replace the
 539  * earlier kprobe in the hlist with the manager kprobe
 540  */
 541 static inline void add_aggr_kprobe(struct kprobe *ap, struct kprobe *p)
 542 {
 543         copy_kprobe(p, ap);
 544         flush_insn_slot(ap);
 545         ap->addr = p->addr;
 546         ap->pre_handler = aggr_pre_handler;
 547         ap->fault_handler = aggr_fault_handler;
 548         /* We don't care the kprobe which has gone. */
 549         if (p->post_handler && !kprobe_gone(p))
 550                 ap->post_handler = aggr_post_handler;
 551         if (p->break_handler && !kprobe_gone(p))
 552                 ap->break_handler = aggr_break_handler;
 553
 554         INIT_LIST_HEAD(&ap->list);
 555         list_add_rcu(&p->list, &ap->list);
 556
 557         hlist_replace_rcu(&p->hlist, &ap->hlist);
 558 }
 559
 560 /*
 561  * This is the second or subsequent kprobe at the address - handle
 562  * the intricacies
 563  */
 564 static int __kprobes register_aggr_kprobe(struct kprobe *old_p,
 565                                           struct kprobe *p)
 566 {
 567         int ret = 0;
 568         struct kprobe *ap;
 569
 570         if (kprobe_gone(old_p)) {
 571                 /*
 572                  * Attempting to insert new probe at the same location that
 573                  * had a probe in the module vaddr area which already
 574                  * freed. So, the instruction slot has already been
 575                  * released. We need a new slot for the new probe.
 576                  */
 577                 ret = arch_prepare_kprobe(old_p);
 578                 if (ret)
 579                         return ret;
 580         }
 581         if (old_p->pre_handler == aggr_pre_handler) {
 582                 copy_kprobe(old_p, p);
 583                 ret = add_new_kprobe(old_p, p);
 584                 ap = old_p;
 585         } else {
 586                 ap = kzalloc(sizeof(struct kprobe), GFP_KERNEL);
 587                 if (!ap) {
 588                         if (kprobe_gone(old_p))
 589                                 arch_remove_kprobe(old_p);
 590                         return -ENOMEM;
 591                 }
 592                 add_aggr_kprobe(ap, old_p);
 593                 copy_kprobe(ap, p);
 594                 ret = add_new_kprobe(ap, p);
 595         }
 596         if (kprobe_gone(old_p)) {
 597                 /*
 598                  * If the old_p has gone, its breakpoint has been disarmed.
 599                  * We have to arm it again after preparing real kprobes.
 600                  */
 601                 ap->flags &= ~KPROBE_FLAG_GONE;
 602                 if (kprobe_enabled)
 603                         arch_arm_kprobe(ap);
 604         }
 605         return ret;
 606 }
 607
 608 static int __kprobes in_kprobes_functions(unsigned long addr)
 609 {
 610         struct kprobe_blackpoint *kb;
 611
 612         if (addr >= (unsigned long)__kprobes_text_start &&
 613             addr < (unsigned long)__kprobes_text_end)
 614                 return -EINVAL;
 615         /*
 616          * If there exists a kprobe_blacklist, verify and
 617          * fail any probe registration in the prohibited area
 618          */
 619         for (kb = kprobe_blacklist; kb->name != NULL; kb++) {
 620                 if (kb->start_addr) {
 621                         if (addr >= kb->start_addr &&
 622                             addr < (kb->start_addr + kb->range))
 623                                 return -EINVAL;
 624                 }
 625         }
 626         return 0;
 627 }
 628
 629 /*
 630  * If we have a symbol_name argument, look it up and add the offset field
 631  * to it. This way, we can specify a relative address to a symbol.
 632  */
 633 static kprobe_opcode_t __kprobes *kprobe_addr(struct kprobe *p)
 634 {
 635         kprobe_opcode_t *addr = p->addr;
 636         if (p->symbol_name) {
 637                 if (addr)
 638                         return NULL;
 639                 kprobe_lookup_name(p->symbol_name, addr);
 640         }
 641
 642         if (!addr)
 643                 return NULL;
 644         return (kprobe_opcode_t *)(((char *)addr) + p->offset);
 645 }
 646
 647 static int __kprobes __register_kprobe(struct kprobe *p,
 648         unsigned long called_from)
 649 {
 650         int ret = 0;
 651         struct kprobe *old_p;
 652         struct module *probed_mod;
 653         kprobe_opcode_t *addr;
 654
 655         addr = kprobe_addr(p);
 656         if (!addr)
 657                 return -EINVAL;
 658         p->addr = addr;
 659
 660         preempt_disable();
 661         if (!__kernel_text_address((unsigned long) p->addr) ||
 662             in_kprobes_functions((unsigned long) p->addr)) {
 663                 preempt_enable();
 664                 return -EINVAL;
 665         }
 666
 667         p->flags = 0;
 668         /*
 669          * Check if are we probing a module.
 670          */
 671         probed_mod = __module_text_address((unsigned long) p->addr);
 672         if (probed_mod) {
 673                 struct module *calling_mod;
 674                 calling_mod = __module_text_address(called_from);
 675                 /*
 676                  * We must hold a refcount of the probed module while updating
 677                  * its code to prohibit unexpected unloading.
 678                  */
 679                 if (calling_mod != probed_mod) {
 680                         if (unlikely(!try_module_get(probed_mod))) {
 681                                 preempt_enable();
 682                                 return -EINVAL;
 683                         }
 684                 } else
 685                         probed_mod = NULL;
 686         }
 687         preempt_enable();
 688
 689         p->nmissed = 0;
 690         INIT_LIST_HEAD(&p->list);
 691         mutex_lock(&kprobe_mutex);
 692         old_p = get_kprobe(p->addr);
 693         if (old_p) {
 694                 ret = register_aggr_kprobe(old_p, p);
 695                 goto out;
 696         }
 697
 698         ret = arch_prepare_kprobe(p);
 699         if (ret)
 700                 goto out;
 701
 702         INIT_HLIST_NODE(&p->hlist);
 703         hlist_add_head_rcu(&p->hlist,
 704                        &kprobe_table[hash_ptr(p->addr, KPROBE_HASH_BITS)]);
 705
 706         if (kprobe_enabled)
 707                 arch_arm_kprobe(p);
 708
 709 out:
 710         mutex_unlock(&kprobe_mutex);
 711
 712         if (probed_mod)
 713                 module_put(probed_mod);
 714
 715         return ret;
 716 }
 717
 718 /*
 719  * Unregister a kprobe without a scheduler synchronization.
 720  */
 721 static int __kprobes __unregister_kprobe_top(struct kprobe *p)
 722 {
 723         struct kprobe *old_p, *list_p;
 724
 725         old_p = get_kprobe(p->addr);
 726         if (unlikely(!old_p))
 727                 return -EINVAL;
 728
 729         if (p != old_p) {
 730                 list_for_each_entry_rcu(list_p, &old_p->list, list)
 731                         if (list_p == p)
 732                         /* kprobe p is a valid probe */
 733                                 goto valid_p;
 734                 return -EINVAL;
 735         }
 736 valid_p:
 737         if (old_p == p ||
 738             (old_p->pre_handler == aggr_pre_handler &&
 739              list_is_singular(&old_p->list))) {
 740                 /*
 741                  * Only probe on the hash list. Disarm only if kprobes are
 742                  * enabled and not gone - otherwise, the breakpoint would
 743                  * already have been removed. We save on flushing icache.
 744                  */
 745                 if (kprobe_enabled && !kprobe_gone(old_p))
 746                         arch_disarm_kprobe(p);
 747                 hlist_del_rcu(&old_p->hlist);
 748         } else {
 749                 if (p->break_handler && !kprobe_gone(p))
 750                         old_p->break_handler = NULL;
 751                 if (p->post_handler && !kprobe_gone(p)) {
 752                         list_for_each_entry_rcu(list_p, &old_p->list, list) {
 753                                 if ((list_p != p) && (list_p->post_handler))
 754                                         goto noclean;
 755                         }
 756                         old_p->post_handler = NULL;
 757                 }
 758 noclean:
 759                 list_del_rcu(&p->list);
 760         }
 761         return 0;
 762 }
 763
 764 static void __kprobes __unregister_kprobe_bottom(struct kprobe *p)
 765 {
 766         struct kprobe *old_p;
 767
 768         if (list_empty(&p->list))
 769                 arch_remove_kprobe(p);
 770         else if (list_is_singular(&p->list)) {
 771                 /* "p" is the last child of an aggr_kprobe */
 772                 old_p = list_entry(p->list.next, struct kprobe, list);
 773                 list_del(&p->list);
 774                 arch_remove_kprobe(old_p);
 775                 kfree(old_p);
 776         }
 777 }
 778
 779 static int __kprobes __register_kprobes(struct kprobe **kps, int num,
 780         unsigned long called_from)
 781 {
 782         int i, ret = 0;
 783
 784         if (num <= 0)
 785                 return -EINVAL;
 786         for (i = 0; i < num; i++) {
 787                 ret = __register_kprobe(kps[i], called_from);
 788                 if (ret < 0) {
 789                         if (i > 0)
 790                                 unregister_kprobes(kps, i);
 791                         break;
 792                 }
 793         }
 794         return ret;
 795 }
 796
 797 /*
 798  * Registration and unregistration functions for kprobe.
 799  */
 800 int __kprobes register_kprobe(struct kprobe *p)
 801 {
 802         return __register_kprobes(&p, 1,
 803                                   (unsigned long)__builtin_return_address(0));
 804 }
 805
 806 void __kprobes unregister_kprobe(struct kprobe *p)
 807 {
 808         unregister_kprobes(&p, 1);
 809 }
 810
 811 int __kprobes register_kprobes(struct kprobe **kps, int num)
 812 {
 813         return __register_kprobes(kps, num,
 814                                   (unsigned long)__builtin_return_address(0));
 815 }
 816
 817 void __kprobes unregister_kprobes(struct kprobe **kps, int num)
 818 {
 819         int i;
 820
 821         if (num <= 0)
 822                 return;
 823         mutex_lock(&kprobe_mutex);
 824         for (i = 0; i < num; i++)
 825                 if (__unregister_kprobe_top(kps[i]) < 0)
 826                         kps[i]->addr = NULL;
 827         mutex_unlock(&kprobe_mutex);
 828
 829         synchronize_sched();
 830         for (i = 0; i < num; i++)
 831                 if (kps[i]->addr)
 832                         __unregister_kprobe_bottom(kps[i]);
 833 }
 834
 835 static struct notifier_block kprobe_exceptions_nb = {
 836         .notifier_call = kprobe_exceptions_notify,
 837         .priority = 0x7fffffff /* we need to be notified first */
 838 };
 839
 840 unsigned long __weak arch_deref_entry_point(void *entry)
 841 {
 842         return (unsigned long)entry;
 843 }
 844
 845 static int __kprobes __register_jprobes(struct jprobe **jps, int num,
 846         unsigned long called_from)
 847 {
 848         struct jprobe *jp;
 849         int ret = 0, i;
 850
 851         if (num <= 0)
 852                 return -EINVAL;
 853         for (i = 0; i < num; i++) {
 854                 unsigned long addr;
 855                 jp = jps[i];
 856                 addr = arch_deref_entry_point(jp->entry);
 857
 858                 if (!kernel_text_address(addr))
 859                         ret = -EINVAL;
 860                 else {
 861                         /* Todo: Verify probepoint is a function entry point */
 862                         jp->kp.pre_handler = setjmp_pre_handler;
 863                         jp->kp.break_handler = longjmp_break_handler;
 864                         ret = __register_kprobe(&jp->kp, called_from);
 865                 }
 866                 if (ret < 0) {
 867                         if (i > 0)
 868                                 unregister_jprobes(jps, i);
 869                         break;
 870                 }
 871         }
 872         return ret;
 873 }
 874
 875 int __kprobes register_jprobe(struct jprobe *jp)
 876 {
 877         return __register_jprobes(&jp, 1,
 878                 (unsigned long)__builtin_return_address(0));
 879 }
 880
 881 void __kprobes unregister_jprobe(struct jprobe *jp)
 882 {
 883         unregister_jprobes(&jp, 1);
 884 }
 885
 886 int __kprobes register_jprobes(struct jprobe **jps, int num)
 887 {
 888         return __register_jprobes(jps, num,
 889                 (unsigned long)__builtin_return_address(0));
 890 }
 891
 892 void __kprobes unregister_jprobes(struct jprobe **jps, int num)
 893 {
 894         int i;
 895
 896         if (num <= 0)
 897                 return;
 898         mutex_lock(&kprobe_mutex);
 899         for (i = 0; i < num; i++)
 900                 if (__unregister_kprobe_top(&jps[i]->kp) < 0)
 901                         jps[i]->kp.addr = NULL;
 902         mutex_unlock(&kprobe_mutex);
 903
 904         synchronize_sched();
 905         for (i = 0; i < num; i++) {
 906                 if (jps[i]->kp.addr)
 907                         __unregister_kprobe_bottom(&jps[i]->kp);
 908         }
 909 }
 910
 911 #ifdef CONFIG_KRETPROBES
 912 /*
 913  * This kprobe pre_handler is registered with every kretprobe. When probe
 914  * hits it will set up the return probe.
 915  */
 916 static int __kprobes pre_handler_kretprobe(struct kprobe *p,
 917                                            struct pt_regs *regs)
 918 {
 919         struct kretprobe *rp = container_of(p, struct kretprobe, kp);
 920         unsigned long hash, flags = 0;
 921         struct kretprobe_instance *ri;
 922
 923         /*TODO: consider to only swap the RA after the last pre_handler fired */
 924         hash = hash_ptr(current, KPROBE_HASH_BITS);
 925         spin_lock_irqsave(&rp->lock, flags);
 926         if (!hlist_empty(&rp->free_instances)) {
 927                 ri = hlist_entry(rp->free_instances.first,
 928                                 struct kretprobe_instance, hlist);
 929                 hlist_del(&ri->hlist);
 930                 spin_unlock_irqrestore(&rp->lock, flags);
 931
 932                 ri->rp = rp;
 933                 ri->task = current;
 934
 935                 if (rp->entry_handler && rp->entry_handler(ri, regs)) {
 936                         spin_unlock_irqrestore(&rp->lock, flags);
 937                         return 0;
 938                 }
 939
 940                 arch_prepare_kretprobe(ri, regs);
 941
 942                 /* XXX(hch): why is there no hlist_move_head? */
 943                 INIT_HLIST_NODE(&ri->hlist);
 944                 kretprobe_table_lock(hash, &flags);
 945                 hlist_add_head(&ri->hlist, &kretprobe_inst_table[hash]);
 946                 kretprobe_table_unlock(hash, &flags);
 947         } else {
 948                 rp->nmissed++;
 949                 spin_unlock_irqrestore(&rp->lock, flags);
 950         }
 951         return 0;
 952 }
 953
 954 static int __kprobes __register_kretprobe(struct kretprobe *rp,
 955                                           unsigned long called_from)
 956 {
 957         int ret = 0;
 958         struct kretprobe_instance *inst;
 959         int i;
 960         void *addr;
 961
 962         if (kretprobe_blacklist_size) {
 963                 addr = kprobe_addr(&rp->kp);
 964                 if (!addr)
 965                         return -EINVAL;
 966
 967                 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
 968                         if (kretprobe_blacklist[i].addr == addr)
 969                                 return -EINVAL;
 970                 }
 971         }
 972
 973         rp->kp.pre_handler = pre_handler_kretprobe;
 974         rp->kp.post_handler = NULL;
 975         rp->kp.fault_handler = NULL;
 976         rp->kp.break_handler = NULL;
 977
 978         /* Pre-allocate memory for max kretprobe instances */
 979         if (rp->maxactive <= 0) {
 980 #ifdef CONFIG_PREEMPT
 981                 rp->maxactive = max(10, 2 * NR_CPUS);
 982 #else
 983                 rp->maxactive = NR_CPUS;
 984 #endif
 985         }
 986         spin_lock_init(&rp->lock);
 987         INIT_HLIST_HEAD(&rp->free_instances);
 988         for (i = 0; i < rp->maxactive; i++) {
 989                 inst = kmalloc(sizeof(struct kretprobe_instance) +
 990                                rp->data_size, GFP_KERNEL);
 991                 if (inst == NULL) {
 992                         free_rp_inst(rp);
 993                         return -ENOMEM;
 994                 }
 995                 INIT_HLIST_NODE(&inst->hlist);
 996                 hlist_add_head(&inst->hlist, &rp->free_instances);
 997         }
 998
 999         rp->nmissed = 0;
1000         /* Establish function entry probe point */
1001         ret = __register_kprobe(&rp->kp, called_from);
1002         if (ret != 0)
1003                 free_rp_inst(rp);
1004         return ret;
1005 }
1006
1007 static int __kprobes __register_kretprobes(struct kretprobe **rps, int num,
1008         unsigned long called_from)
1009 {
1010         int ret = 0, i;
1011
1012         if (num <= 0)
1013                 return -EINVAL;
1014         for (i = 0; i < num; i++) {
1015                 ret = __register_kretprobe(rps[i], called_from);
1016                 if (ret < 0) {
1017                         if (i > 0)
1018                                 unregister_kretprobes(rps, i);
1019                         break;
1020                 }
1021         }
1022         return ret;
1023 }
1024
1025 int __kprobes register_kretprobe(struct kretprobe *rp)
1026 {
1027         return __register_kretprobes(&rp, 1,
1028                         (unsigned long)__builtin_return_address(0));
1029 }
1030
1031 void __kprobes unregister_kretprobe(struct kretprobe *rp)
1032 {
1033         unregister_kretprobes(&rp, 1);
1034 }
1035
1036 int __kprobes register_kretprobes(struct kretprobe **rps, int num)
1037 {
1038         return __register_kretprobes(rps, num,
1039                         (unsigned long)__builtin_return_address(0));
1040 }
1041
1042 void __kprobes unregister_kretprobes(struct kretprobe **rps, int num)
1043 {
1044         int i;
1045
1046         if (num <= 0)
1047                 return;
1048         mutex_lock(&kprobe_mutex);
1049         for (i = 0; i < num; i++)
1050                 if (__unregister_kprobe_top(&rps[i]->kp) < 0)
1051                         rps[i]->kp.addr = NULL;
1052         mutex_unlock(&kprobe_mutex);
1053
1054         synchronize_sched();
1055         for (i = 0; i < num; i++) {
1056                 if (rps[i]->kp.addr) {
1057                         __unregister_kprobe_bottom(&rps[i]->kp);
1058                         cleanup_rp_inst(rps[i]);
1059                 }
1060         }
1061 }
1062
1063 #else /* CONFIG_KRETPROBES */
1064 int __kprobes register_kretprobe(struct kretprobe *rp)
1065 {
1066         return -ENOSYS;
1067 }
1068
1069 int __kprobes register_kretprobes(struct kretprobe **rps, int num)
1070 {
1071         return -ENOSYS;
1072 }
1073 void __kprobes unregister_kretprobe(struct kretprobe *rp)
1074 {
1075 }
1076
1077 void __kprobes unregister_kretprobes(struct kretprobe **rps, int num)
1078 {
1079 }
1080
1081 static int __kprobes pre_handler_kretprobe(struct kprobe *p,
1082                                            struct pt_regs *regs)
1083 {
1084         return 0;
1085 }
1086
1087 #endif /* CONFIG_KRETPROBES */
1088
1089 /* Set the kprobe gone and remove its instruction buffer. */
1090 static void __kprobes kill_kprobe(struct kprobe *p)
1091 {
1092         struct kprobe *kp;
1093         p->flags |= KPROBE_FLAG_GONE;
1094         if (p->pre_handler == aggr_pre_handler) {
1095                 /*
1096                  * If this is an aggr_kprobe, we have to list all the
1097                  * chained probes and mark them GONE.
1098                  */
1099                 list_for_each_entry_rcu(kp, &p->list, list)
1100                         kp->flags |= KPROBE_FLAG_GONE;
1101                 p->post_handler = NULL;
1102                 p->break_handler = NULL;
1103         }
1104         /*
1105          * Here, we can remove insn_slot safely, because no thread calls
1106          * the original probed function (which will be freed soon) any more.
1107          */
1108         arch_remove_kprobe(p);
1109 }
1110
1111 /* Module notifier call back, checking kprobes on the module */
1112 static int __kprobes kprobes_module_callback(struct notifier_block *nb,
1113                                              unsigned long val, void *data)
1114 {
1115         struct module *mod = data;
1116         struct hlist_head *head;
1117         struct hlist_node *node;
1118         struct kprobe *p;
1119         unsigned int i;
1120
1121         if (val != MODULE_STATE_GOING)
1122                 return NOTIFY_DONE;
1123
1124         /*
1125          * module .text section will be freed. We need to
1126          * disable kprobes which have been inserted in the section.
1127          */
1128         mutex_lock(&kprobe_mutex);
1129         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
1130                 head = &kprobe_table[i];
1131                 hlist_for_each_entry_rcu(p, node, head, hlist)
1132                         if (within_module_core((unsigned long)p->addr, mod)) {
1133                                 /*
1134                                  * The vaddr this probe is installed will soon
1135                                  * be vfreed buy not synced to disk. Hence,
1136                                  * disarming the breakpoint isn't needed.
1137                                  */
1138                                 kill_kprobe(p);
1139                         }
1140         }
1141         mutex_unlock(&kprobe_mutex);
1142         return NOTIFY_DONE;
1143 }
1144
1145 static struct notifier_block kprobe_module_nb = {
1146         .notifier_call = kprobes_module_callback,
1147         .priority = 0
1148 };
1149
1150 static int __init init_kprobes(void)
1151 {
1152         int i, err = 0;
1153         unsigned long offset = 0, size = 0;
1154         char *modname, namebuf[128];
1155         const char *symbol_name;
1156         void *addr;
1157         struct kprobe_blackpoint *kb;
1158
1159         /* FIXME allocate the probe table, currently defined statically */
1160         /* initialize all list heads */
1161         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
1162                 INIT_HLIST_HEAD(&kprobe_table[i]);
1163                 INIT_HLIST_HEAD(&kretprobe_inst_table[i]);
1164                 spin_lock_init(&(kretprobe_table_locks[i].lock));
1165         }
1166
1167         /*
1168          * Lookup and populate the kprobe_blacklist.
1169          *
1170          * Unlike the kretprobe blacklist, we'll need to determine
1171          * the range of addresses that belong to the said functions,
1172          * since a kprobe need not necessarily be at the beginning
1173          * of a function.
1174          */
1175         for (kb = kprobe_blacklist; kb->name != NULL; kb++) {
1176                 kprobe_lookup_name(kb->name, addr);
1177                 if (!addr)
1178                         continue;
1179
1180                 kb->start_addr = (unsigned long)addr;
1181                 symbol_name = kallsyms_lookup(kb->start_addr,
1182                                 &size, &offset, &modname, namebuf);
1183                 if (!symbol_name)
1184                         kb->range = 0;
1185                 else
1186                         kb->range = size;
1187         }
1188
1189         if (kretprobe_blacklist_size) {
1190                 /* lookup the function address from its name */
1191                 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
1192                         kprobe_lookup_name(kretprobe_blacklist[i].name,
1193                                            kretprobe_blacklist[i].addr);
1194                         if (!kretprobe_blacklist[i].addr)
1195                                 printk("kretprobe: lookup failed: %s\n",
1196                                        kretprobe_blacklist[i].name);
1197                 }
1198         }
1199
1200         /* By default, kprobes are enabled */
1201         kprobe_enabled = true;
1202
1203         err = arch_init_kprobes();
1204         if (!err)
1205                 err = register_die_notifier(&kprobe_exceptions_nb);
1206         if (!err)
1207                 err = register_module_notifier(&kprobe_module_nb);
1208
1209         kprobes_initialized = (err == 0);
1210
1211         if (!err)
1212                 init_test_probes();
1213         return err;
1214 }
1215
1216 #ifdef CONFIG_DEBUG_FS
1217 static void __kprobes report_probe(struct seq_file *pi, struct kprobe *p,
1218                 const char *sym, int offset,char *modname)
1219 {
1220         char *kprobe_type;
1221
1222         if (p->pre_handler == pre_handler_kretprobe)
1223                 kprobe_type = "r";
1224         else if (p->pre_handler == setjmp_pre_handler)
1225                 kprobe_type = "j";
1226         else
1227                 kprobe_type = "k";
1228         if (sym)
1229                 seq_printf(pi, "%p  %s  %s+0x%x  %s %s\n", p->addr, kprobe_type,
1230                         sym, offset, (modname ? modname : " "),
1231                         (kprobe_gone(p) ? "[GONE]" : ""));
1232         else
1233                 seq_printf(pi, "%p  %s  %p %s\n", p->addr, kprobe_type, p->addr,
1234                         (kprobe_gone(p) ? "[GONE]" : ""));
1235 }
1236
1237 static void __kprobes *kprobe_seq_start(struct seq_file *f, loff_t *pos)
1238 {
1239         return (*pos < KPROBE_TABLE_SIZE) ? pos : NULL;
1240 }
1241
1242 static void __kprobes *kprobe_seq_next(struct seq_file *f, void *v, loff_t *pos)
1243 {
1244         (*pos)++;
1245         if (*pos >= KPROBE_TABLE_SIZE)
1246                 return NULL;
1247         return pos;
1248 }
1249
1250 static void __kprobes kprobe_seq_stop(struct seq_file *f, void *v)
1251 {
1252         /* Nothing to do */
1253 }
1254
1255 static int __kprobes show_kprobe_addr(struct seq_file *pi, void *v)
1256 {
1257         struct hlist_head *head;
1258         struct hlist_node *node;
1259         struct kprobe *p, *kp;
1260         const char *sym = NULL;
1261         unsigned int i = *(loff_t *) v;
1262         unsigned long offset = 0;
1263         char *modname, namebuf[128];
1264
1265         head = &kprobe_table[i];
1266         preempt_disable();
1267         hlist_for_each_entry_rcu(p, node, head, hlist) {
1268                 sym = kallsyms_lookup((unsigned long)p->addr, NULL,
1269                                         &offset, &modname, namebuf);
1270                 if (p->pre_handler == aggr_pre_handler) {
1271                         list_for_each_entry_rcu(kp, &p->list, list)
1272                                 report_probe(pi, kp, sym, offset, modname);
1273                 } else
1274                         report_probe(pi, p, sym, offset, modname);
1275         }
1276         preempt_enable();
1277         return 0;
1278 }
1279
1280 static struct seq_operations kprobes_seq_ops = {
1281         .start = kprobe_seq_start,
1282         .next  = kprobe_seq_next,
1283         .stop  = kprobe_seq_stop,
1284         .show  = show_kprobe_addr
1285 };
1286
1287 static int __kprobes kprobes_open(struct inode *inode, struct file *filp)
1288 {
1289         return seq_open(filp, &kprobes_seq_ops);
1290 }
1291
1292 static struct file_operations debugfs_kprobes_operations = {
1293         .open           = kprobes_open,
1294         .read           = seq_read,
1295         .llseek         = seq_lseek,
1296         .release        = seq_release,
1297 };
1298
1299 static void __kprobes enable_all_kprobes(void)
1300 {
1301         struct hlist_head *head;
1302         struct hlist_node *node;
1303         struct kprobe *p;
1304         unsigned int i;
1305
1306         mutex_lock(&kprobe_mutex);
1307
1308         /* If kprobes are already enabled, just return */
1309         if (kprobe_enabled)
1310                 goto already_enabled;
1311
1312         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
1313                 head = &kprobe_table[i];
1314                 hlist_for_each_entry_rcu(p, node, head, hlist)
1315                         if (!kprobe_gone(p))
1316                                 arch_arm_kprobe(p);
1317         }
1318
1319         kprobe_enabled = true;
1320         printk(KERN_INFO "Kprobes globally enabled\n");
1321
1322 already_enabled:
1323         mutex_unlock(&kprobe_mutex);
1324         return;
1325 }
1326
1327 static void __kprobes disable_all_kprobes(void)
1328 {
1329         struct hlist_head *head;
1330         struct hlist_node *node;
1331         struct kprobe *p;
1332         unsigned int i;
1333
1334         mutex_lock(&kprobe_mutex);
1335
1336         /* If kprobes are already disabled, just return */
1337         if (!kprobe_enabled)
1338                 goto already_disabled;
1339
1340         kprobe_enabled = false;
1341         printk(KERN_INFO "Kprobes globally disabled\n");
1342         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
1343                 head = &kprobe_table[i];
1344                 hlist_for_each_entry_rcu(p, node, head, hlist) {
1345                         if (!arch_trampoline_kprobe(p) && !kprobe_gone(p))
1346                                 arch_disarm_kprobe(p);
1347                 }
1348         }
1349
1350         mutex_unlock(&kprobe_mutex);
1351         /* Allow all currently running kprobes to complete */
1352         synchronize_sched();
1353         return;
1354
1355 already_disabled:
1356         mutex_unlock(&kprobe_mutex);
1357         return;
1358 }
1359
1360 /*
1361  * XXX: The debugfs bool file interface doesn't allow for callbacks
1362  * when the bool state is switched. We can reuse that facility when
1363  * available
1364  */
1365 static ssize_t read_enabled_file_bool(struct file *file,
1366                char __user *user_buf, size_t count, loff_t *ppos)
1367 {
1368         char buf[3];
1369
1370         if (kprobe_enabled)
1371                 buf[0] = '1';
1372         else
1373                 buf[0] = '0';
1374         buf[1] = '\n';
1375         buf[2] = 0x00;
1376         return simple_read_from_buffer(user_buf, count, ppos, buf, 2);
1377 }
1378
1379 static ssize_t write_enabled_file_bool(struct file *file,
1380                const char __user *user_buf, size_t count, loff_t *ppos)
1381 {
1382         char buf[32];
1383         int buf_size;
1384
1385         buf_size = min(count, (sizeof(buf)-1));
1386         if (copy_from_user(buf, user_buf, buf_size))
1387                 return -EFAULT;
1388
1389         switch (buf[0]) {
1390         case 'y':
1391         case 'Y':
1392         case '1':
1393                 enable_all_kprobes();
1394                 break;
1395         case 'n':
1396         case 'N':
1397         case '0':
1398                 disable_all_kprobes();
1399                 break;
1400         }
1401
1402         return count;
1403 }
1404
1405 static struct file_operations fops_kp = {
1406         .read =         read_enabled_file_bool,
1407         .write =        write_enabled_file_bool,
1408 };
1409
1410 static int __kprobes debugfs_kprobe_init(void)
1411 {
1412         struct dentry *dir, *file;
1413         unsigned int value = 1;
1414
1415         dir = debugfs_create_dir("kprobes", NULL);
1416         if (!dir)
1417                 return -ENOMEM;
1418
1419         file = debugfs_create_file("list", 0444, dir, NULL,
1420                                 &debugfs_kprobes_operations);
1421         if (!file) {
1422                 debugfs_remove(dir);
1423                 return -ENOMEM;
1424         }
1425
1426         file = debugfs_create_file("enabled", 0600, dir,
1427                                         &value, &fops_kp);
1428         if (!file) {
1429                 debugfs_remove(dir);
1430                 return -ENOMEM;
1431         }
1432
1433         return 0;
1434 }
1435
1436 late_initcall(debugfs_kprobe_init);
1437 #endif /* CONFIG_DEBUG_FS */
1438
1439 module_init(init_kprobes);
1440
1441 EXPORT_SYMBOL_GPL(register_kprobe);
1442 EXPORT_SYMBOL_GPL(unregister_kprobe);
1443 EXPORT_SYMBOL_GPL(register_kprobes);
1444 EXPORT_SYMBOL_GPL(unregister_kprobes);
1445 EXPORT_SYMBOL_GPL(register_jprobe);
1446 EXPORT_SYMBOL_GPL(unregister_jprobe);
1447 EXPORT_SYMBOL_GPL(register_jprobes);
1448 EXPORT_SYMBOL_GPL(unregister_jprobes);
1449 EXPORT_SYMBOL_GPL(jprobe_return);
1450 EXPORT_SYMBOL_GPL(register_kretprobe);
1451 EXPORT_SYMBOL_GPL(unregister_kretprobe);
1452 EXPORT_SYMBOL_GPL(register_kretprobes);
1453 EXPORT_SYMBOL_GPL(unregister_kretprobes);