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 #include <linux/memory.h>
  47
  48 #include <asm-generic/sections.h>
  49 #include <asm/cacheflush.h>
  50 #include <asm/errno.h>
  51 #include <asm/uaccess.h>
  52
  53 #define KPROBE_HASH_BITS 6
  54 #define KPROBE_TABLE_SIZE (1 << KPROBE_HASH_BITS)
  55
  56
  57 /*
  58  * Some oddball architectures like 64bit powerpc have function descriptors
  59  * so this must be overridable.
  60  */
  61 #ifndef kprobe_lookup_name
  62 #define kprobe_lookup_name(name, addr) \
  63         addr = ((kprobe_opcode_t *)(kallsyms_lookup_name(name)))
  64 #endif
  65
  66 static int kprobes_initialized;
  67 static struct hlist_head kprobe_table[KPROBE_TABLE_SIZE];
  68 static struct hlist_head kretprobe_inst_table[KPROBE_TABLE_SIZE];
  69
  70 /* NOTE: change this value only with kprobe_mutex held */
  71 static bool kprobes_all_disarmed;
  72
  73 static DEFINE_MUTEX(kprobe_mutex);      /* Protects kprobe_table */
  74 static DEFINE_PER_CPU(struct kprobe *, kprobe_instance) = NULL;
  75 static struct {
  76         spinlock_t lock ____cacheline_aligned_in_smp;
  77 } kretprobe_table_locks[KPROBE_TABLE_SIZE];
  78
  79 static spinlock_t *kretprobe_table_lock_ptr(unsigned long hash)
  80 {
  81         return &(kretprobe_table_locks[hash].lock);
  82 }
  83
  84 /*
  85  * Normally, functions that we'd want to prohibit kprobes in, are marked
  86  * __kprobes. But, there are cases where such functions already belong to
  87  * a different section (__sched for preempt_schedule)
  88  *
  89  * For such cases, we now have a blacklist
  90  */
  91 static struct kprobe_blackpoint kprobe_blacklist[] = {
  92         {"preempt_schedule",},
  93         {NULL}    /* Terminator */
  94 };
  95
  96 #ifdef __ARCH_WANT_KPROBES_INSN_SLOT
  97 /*
  98  * kprobe->ainsn.insn points to the copy of the instruction to be
  99  * single-stepped. x86_64, POWER4 and above have no-exec support and
 100  * stepping on the instruction on a vmalloced/kmalloced/data page
 101  * is a recipe for disaster
 102  */
 103 #define INSNS_PER_PAGE  (PAGE_SIZE/(MAX_INSN_SIZE * sizeof(kprobe_opcode_t)))
 104
 105 struct kprobe_insn_page {
 106         struct hlist_node hlist;
 107         kprobe_opcode_t *insns;         /* Page of instruction slots */
 108         char slot_used[INSNS_PER_PAGE];
 109         int nused;
 110         int ngarbage;
 111 };
 112
 113 enum kprobe_slot_state {
 114         SLOT_CLEAN = 0,
 115         SLOT_DIRTY = 1,
 116         SLOT_USED = 2,
 117 };
 118
 119 static DEFINE_MUTEX(kprobe_insn_mutex); /* Protects kprobe_insn_pages */
 120 static struct hlist_head kprobe_insn_pages;
 121 static int kprobe_garbage_slots;
 122 static int collect_garbage_slots(void);
 123
 124 static int __kprobes check_safety(void)
 125 {
 126         int ret = 0;
 127 #if defined(CONFIG_PREEMPT) && defined(CONFIG_FREEZER)
 128         ret = freeze_processes();
 129         if (ret == 0) {
 130                 struct task_struct *p, *q;
 131                 do_each_thread(p, q) {
 132                         if (p != current && p->state == TASK_RUNNING &&
 133                             p->pid != 0) {
 134                                 printk("Check failed: %s is running\n",p->comm);
 135                                 ret = -1;
 136                                 goto loop_end;
 137                         }
 138                 } while_each_thread(p, q);
 139         }
 140 loop_end:
 141         thaw_processes();
 142 #else
 143         synchronize_sched();
 144 #endif
 145         return ret;
 146 }
 147
 148 /**
 149  * __get_insn_slot() - Find a slot on an executable page for an instruction.
 150  * We allocate an executable page if there's no room on existing ones.
 151  */
 152 static kprobe_opcode_t __kprobes *__get_insn_slot(void)
 153 {
 154         struct kprobe_insn_page *kip;
 155         struct hlist_node *pos;
 156
 157  retry:
 158         hlist_for_each_entry(kip, pos, &kprobe_insn_pages, hlist) {
 159                 if (kip->nused < INSNS_PER_PAGE) {
 160                         int i;
 161                         for (i = 0; i < INSNS_PER_PAGE; i++) {
 162                                 if (kip->slot_used[i] == SLOT_CLEAN) {
 163                                         kip->slot_used[i] = SLOT_USED;
 164                                         kip->nused++;
 165                                         return kip->insns + (i * MAX_INSN_SIZE);
 166                                 }
 167                         }
 168                         /* Surprise!  No unused slots.  Fix kip->nused. */
 169                         kip->nused = INSNS_PER_PAGE;
 170                 }
 171         }
 172
 173         /* If there are any garbage slots, collect it and try again. */
 174         if (kprobe_garbage_slots && collect_garbage_slots() == 0) {
 175                 goto retry;
 176         }
 177         /* All out of space.  Need to allocate a new page. Use slot 0. */
 178         kip = kmalloc(sizeof(struct kprobe_insn_page), GFP_KERNEL);
 179         if (!kip)
 180                 return NULL;
 181
 182         /*
 183          * Use module_alloc so this page is within +/- 2GB of where the
 184          * kernel image and loaded module images reside. This is required
 185          * so x86_64 can correctly handle the %rip-relative fixups.
 186          */
 187         kip->insns = module_alloc(PAGE_SIZE);
 188         if (!kip->insns) {
 189                 kfree(kip);
 190                 return NULL;
 191         }
 192         INIT_HLIST_NODE(&kip->hlist);
 193         hlist_add_head(&kip->hlist, &kprobe_insn_pages);
 194         memset(kip->slot_used, SLOT_CLEAN, INSNS_PER_PAGE);
 195         kip->slot_used[0] = SLOT_USED;
 196         kip->nused = 1;
 197         kip->ngarbage = 0;
 198         return kip->insns;
 199 }
 200
 201 kprobe_opcode_t __kprobes *get_insn_slot(void)
 202 {
 203         kprobe_opcode_t *ret;
 204         mutex_lock(&kprobe_insn_mutex);
 205         ret = __get_insn_slot();
 206         mutex_unlock(&kprobe_insn_mutex);
 207         return ret;
 208 }
 209
 210 /* Return 1 if all garbages are collected, otherwise 0. */
 211 static int __kprobes collect_one_slot(struct kprobe_insn_page *kip, int idx)
 212 {
 213         kip->slot_used[idx] = SLOT_CLEAN;
 214         kip->nused--;
 215         if (kip->nused == 0) {
 216                 /*
 217                  * Page is no longer in use.  Free it unless
 218                  * it's the last one.  We keep the last one
 219                  * so as not to have to set it up again the
 220                  * next time somebody inserts a probe.
 221                  */
 222                 hlist_del(&kip->hlist);
 223                 if (hlist_empty(&kprobe_insn_pages)) {
 224                         INIT_HLIST_NODE(&kip->hlist);
 225                         hlist_add_head(&kip->hlist,
 226                                        &kprobe_insn_pages);
 227                 } else {
 228                         module_free(NULL, kip->insns);
 229                         kfree(kip);
 230                 }
 231                 return 1;
 232         }
 233         return 0;
 234 }
 235
 236 static int __kprobes collect_garbage_slots(void)
 237 {
 238         struct kprobe_insn_page *kip;
 239         struct hlist_node *pos, *next;
 240         int safety;
 241
 242         /* Ensure no-one is preepmted on the garbages */
 243         mutex_unlock(&kprobe_insn_mutex);
 244         safety = check_safety();
 245         mutex_lock(&kprobe_insn_mutex);
 246         if (safety != 0)
 247                 return -EAGAIN;
 248
 249         hlist_for_each_entry_safe(kip, pos, next, &kprobe_insn_pages, hlist) {
 250                 int i;
 251                 if (kip->ngarbage == 0)
 252                         continue;
 253                 kip->ngarbage = 0;      /* we will collect all garbages */
 254                 for (i = 0; i < INSNS_PER_PAGE; i++) {
 255                         if (kip->slot_used[i] == SLOT_DIRTY &&
 256                             collect_one_slot(kip, i))
 257                                 break;
 258                 }
 259         }
 260         kprobe_garbage_slots = 0;
 261         return 0;
 262 }
 263
 264 void __kprobes free_insn_slot(kprobe_opcode_t * slot, int dirty)
 265 {
 266         struct kprobe_insn_page *kip;
 267         struct hlist_node *pos;
 268
 269         mutex_lock(&kprobe_insn_mutex);
 270         hlist_for_each_entry(kip, pos, &kprobe_insn_pages, hlist) {
 271                 if (kip->insns <= slot &&
 272                     slot < kip->insns + (INSNS_PER_PAGE * MAX_INSN_SIZE)) {
 273                         int i = (slot - kip->insns) / MAX_INSN_SIZE;
 274                         if (dirty) {
 275                                 kip->slot_used[i] = SLOT_DIRTY;
 276                                 kip->ngarbage++;
 277                         } else {
 278                                 collect_one_slot(kip, i);
 279                         }
 280                         break;
 281                 }
 282         }
 283
 284         if (dirty && ++kprobe_garbage_slots > INSNS_PER_PAGE)
 285                 collect_garbage_slots();
 286
 287         mutex_unlock(&kprobe_insn_mutex);
 288 }
 289 #endif
 290
 291 /* We have preemption disabled.. so it is safe to use __ versions */
 292 static inline void set_kprobe_instance(struct kprobe *kp)
 293 {
 294         __get_cpu_var(kprobe_instance) = kp;
 295 }
 296
 297 static inline void reset_kprobe_instance(void)
 298 {
 299         __get_cpu_var(kprobe_instance) = NULL;
 300 }
 301
 302 /*
 303  * This routine is called either:
 304  *      - under the kprobe_mutex - during kprobe_[un]register()
 305  *                              OR
 306  *      - with preemption disabled - from arch/xxx/kernel/kprobes.c
 307  */
 308 struct kprobe __kprobes *get_kprobe(void *addr)
 309 {
 310         struct hlist_head *head;
 311         struct hlist_node *node;
 312         struct kprobe *p;
 313
 314         head = &kprobe_table[hash_ptr(addr, KPROBE_HASH_BITS)];
 315         hlist_for_each_entry_rcu(p, node, head, hlist) {
 316                 if (p->addr == addr)
 317                         return p;
 318         }
 319         return NULL;
 320 }
 321
 322 /*
 323  * Aggregate handlers for multiple kprobes support - these handlers
 324  * take care of invoking the individual kprobe handlers on p->list
 325  */
 326 static int __kprobes aggr_pre_handler(struct kprobe *p, struct pt_regs *regs)
 327 {
 328         struct kprobe *kp;
 329
 330         list_for_each_entry_rcu(kp, &p->list, list) {
 331                 if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
 332                         set_kprobe_instance(kp);
 333                         if (kp->pre_handler(kp, regs))
 334                                 return 1;
 335                 }
 336                 reset_kprobe_instance();
 337         }
 338         return 0;
 339 }
 340
 341 static void __kprobes aggr_post_handler(struct kprobe *p, struct pt_regs *regs,
 342                                         unsigned long flags)
 343 {
 344         struct kprobe *kp;
 345
 346         list_for_each_entry_rcu(kp, &p->list, list) {
 347                 if (kp->post_handler && likely(!kprobe_disabled(kp))) {
 348                         set_kprobe_instance(kp);
 349                         kp->post_handler(kp, regs, flags);
 350                         reset_kprobe_instance();
 351                 }
 352         }
 353 }
 354
 355 static int __kprobes aggr_fault_handler(struct kprobe *p, struct pt_regs *regs,
 356                                         int trapnr)
 357 {
 358         struct kprobe *cur = __get_cpu_var(kprobe_instance);
 359
 360         /*
 361          * if we faulted "during" the execution of a user specified
 362          * probe handler, invoke just that probe's fault handler
 363          */
 364         if (cur && cur->fault_handler) {
 365                 if (cur->fault_handler(cur, regs, trapnr))
 366                         return 1;
 367         }
 368         return 0;
 369 }
 370
 371 static int __kprobes aggr_break_handler(struct kprobe *p, struct pt_regs *regs)
 372 {
 373         struct kprobe *cur = __get_cpu_var(kprobe_instance);
 374         int ret = 0;
 375
 376         if (cur && cur->break_handler) {
 377                 if (cur->break_handler(cur, regs))
 378                         ret = 1;
 379         }
 380         reset_kprobe_instance();
 381         return ret;
 382 }
 383
 384 /* Walks the list and increments nmissed count for multiprobe case */
 385 void __kprobes kprobes_inc_nmissed_count(struct kprobe *p)
 386 {
 387         struct kprobe *kp;
 388         if (p->pre_handler != aggr_pre_handler) {
 389                 p->nmissed++;
 390         } else {
 391                 list_for_each_entry_rcu(kp, &p->list, list)
 392                         kp->nmissed++;
 393         }
 394         return;
 395 }
 396
 397 void __kprobes recycle_rp_inst(struct kretprobe_instance *ri,
 398                                 struct hlist_head *head)
 399 {
 400         struct kretprobe *rp = ri->rp;
 401
 402         /* remove rp inst off the rprobe_inst_table */
 403         hlist_del(&ri->hlist);
 404         INIT_HLIST_NODE(&ri->hlist);
 405         if (likely(rp)) {
 406                 spin_lock(&rp->lock);
 407                 hlist_add_head(&ri->hlist, &rp->free_instances);
 408                 spin_unlock(&rp->lock);
 409         } else
 410                 /* Unregistering */
 411                 hlist_add_head(&ri->hlist, head);
 412 }
 413
 414 void __kprobes kretprobe_hash_lock(struct task_struct *tsk,
 415                          struct hlist_head **head, unsigned long *flags)
 416 {
 417         unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
 418         spinlock_t *hlist_lock;
 419
 420         *head = &kretprobe_inst_table[hash];
 421         hlist_lock = kretprobe_table_lock_ptr(hash);
 422         spin_lock_irqsave(hlist_lock, *flags);
 423 }
 424
 425 static void __kprobes kretprobe_table_lock(unsigned long hash,
 426         unsigned long *flags)
 427 {
 428         spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
 429         spin_lock_irqsave(hlist_lock, *flags);
 430 }
 431
 432 void __kprobes kretprobe_hash_unlock(struct task_struct *tsk,
 433         unsigned long *flags)
 434 {
 435         unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
 436         spinlock_t *hlist_lock;
 437
 438         hlist_lock = kretprobe_table_lock_ptr(hash);
 439         spin_unlock_irqrestore(hlist_lock, *flags);
 440 }
 441
 442 void __kprobes kretprobe_table_unlock(unsigned long hash, unsigned long *flags)
 443 {
 444         spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
 445         spin_unlock_irqrestore(hlist_lock, *flags);
 446 }
 447
 448 /*
 449  * This function is called from finish_task_switch when task tk becomes dead,
 450  * so that we can recycle any function-return probe instances associated
 451  * with this task. These left over instances represent probed functions
 452  * that have been called but will never return.
 453  */
 454 void __kprobes kprobe_flush_task(struct task_struct *tk)
 455 {
 456         struct kretprobe_instance *ri;
 457         struct hlist_head *head, empty_rp;
 458         struct hlist_node *node, *tmp;
 459         unsigned long hash, flags = 0;
 460
 461         if (unlikely(!kprobes_initialized))
 462                 /* Early boot.  kretprobe_table_locks not yet initialized. */
 463                 return;
 464
 465         hash = hash_ptr(tk, KPROBE_HASH_BITS);
 466         head = &kretprobe_inst_table[hash];
 467         kretprobe_table_lock(hash, &flags);
 468         hlist_for_each_entry_safe(ri, node, tmp, head, hlist) {
 469                 if (ri->task == tk)
 470                         recycle_rp_inst(ri, &empty_rp);
 471         }
 472         kretprobe_table_unlock(hash, &flags);
 473         INIT_HLIST_HEAD(&empty_rp);
 474         hlist_for_each_entry_safe(ri, node, tmp, &empty_rp, hlist) {
 475                 hlist_del(&ri->hlist);
 476                 kfree(ri);
 477         }
 478 }
 479
 480 static inline void free_rp_inst(struct kretprobe *rp)
 481 {
 482         struct kretprobe_instance *ri;
 483         struct hlist_node *pos, *next;
 484
 485         hlist_for_each_entry_safe(ri, pos, next, &rp->free_instances, hlist) {
 486                 hlist_del(&ri->hlist);
 487                 kfree(ri);
 488         }
 489 }
 490
 491 static void __kprobes cleanup_rp_inst(struct kretprobe *rp)
 492 {
 493         unsigned long flags, hash;
 494         struct kretprobe_instance *ri;
 495         struct hlist_node *pos, *next;
 496         struct hlist_head *head;
 497
 498         /* No race here */
 499         for (hash = 0; hash < KPROBE_TABLE_SIZE; hash++) {
 500                 kretprobe_table_lock(hash, &flags);
 501                 head = &kretprobe_inst_table[hash];
 502                 hlist_for_each_entry_safe(ri, pos, next, head, hlist) {
 503                         if (ri->rp == rp)
 504                                 ri->rp = NULL;
 505                 }
 506                 kretprobe_table_unlock(hash, &flags);
 507         }
 508         free_rp_inst(rp);
 509 }
 510
 511 /*
 512  * Keep all fields in the kprobe consistent
 513  */
 514 static inline void copy_kprobe(struct kprobe *old_p, struct kprobe *p)
 515 {
 516         memcpy(&p->opcode, &old_p->opcode, sizeof(kprobe_opcode_t));
 517         memcpy(&p->ainsn, &old_p->ainsn, sizeof(struct arch_specific_insn));
 518 }
 519
 520 /*
 521 * Add the new probe to ap->list. Fail if this is the
 522 * second jprobe at the address - two jprobes can't coexist
 523 */
 524 static int __kprobes add_new_kprobe(struct kprobe *ap, struct kprobe *p)
 525 {
 526         BUG_ON(kprobe_gone(ap) || kprobe_gone(p));
 527         if (p->break_handler) {
 528                 if (ap->break_handler)
 529                         return -EEXIST;
 530                 list_add_tail_rcu(&p->list, &ap->list);
 531                 ap->break_handler = aggr_break_handler;
 532         } else
 533                 list_add_rcu(&p->list, &ap->list);
 534         if (p->post_handler && !ap->post_handler)
 535                 ap->post_handler = aggr_post_handler;
 536
 537         if (kprobe_disabled(ap) && !kprobe_disabled(p)) {
 538                 ap->flags &= ~KPROBE_FLAG_DISABLED;
 539                 if (!kprobes_all_disarmed)
 540                         /* Arm the breakpoint again. */
 541                         arch_arm_kprobe(ap);
 542         }
 543         return 0;
 544 }
 545
 546 /*
 547  * Fill in the required fields of the "manager kprobe". Replace the
 548  * earlier kprobe in the hlist with the manager kprobe
 549  */
 550 static inline void add_aggr_kprobe(struct kprobe *ap, struct kprobe *p)
 551 {
 552         copy_kprobe(p, ap);
 553         flush_insn_slot(ap);
 554         ap->addr = p->addr;
 555         ap->flags = p->flags;
 556         ap->pre_handler = aggr_pre_handler;
 557         ap->fault_handler = aggr_fault_handler;
 558         /* We don't care the kprobe which has gone. */
 559         if (p->post_handler && !kprobe_gone(p))
 560                 ap->post_handler = aggr_post_handler;
 561         if (p->break_handler && !kprobe_gone(p))
 562                 ap->break_handler = aggr_break_handler;
 563
 564         INIT_LIST_HEAD(&ap->list);
 565         list_add_rcu(&p->list, &ap->list);
 566
 567         hlist_replace_rcu(&p->hlist, &ap->hlist);
 568 }
 569
 570 /*
 571  * This is the second or subsequent kprobe at the address - handle
 572  * the intricacies
 573  */
 574 static int __kprobes register_aggr_kprobe(struct kprobe *old_p,
 575                                           struct kprobe *p)
 576 {
 577         int ret = 0;
 578         struct kprobe *ap = old_p;
 579
 580         if (old_p->pre_handler != aggr_pre_handler) {
 581                 /* If old_p is not an aggr_probe, create new aggr_kprobe. */
 582                 ap = kzalloc(sizeof(struct kprobe), GFP_KERNEL);
 583                 if (!ap)
 584                         return -ENOMEM;
 585                 add_aggr_kprobe(ap, old_p);
 586         }
 587
 588         if (kprobe_gone(ap)) {
 589                 /*
 590                  * Attempting to insert new probe at the same location that
 591                  * had a probe in the module vaddr area which already
 592                  * freed. So, the instruction slot has already been
 593                  * released. We need a new slot for the new probe.
 594                  */
 595                 ret = arch_prepare_kprobe(ap);
 596                 if (ret)
 597                         /*
 598                          * Even if fail to allocate new slot, don't need to
 599                          * free aggr_probe. It will be used next time, or
 600                          * freed by unregister_kprobe.
 601                          */
 602                         return ret;
 603
 604                 /*
 605                  * Clear gone flag to prevent allocating new slot again, and
 606                  * set disabled flag because it is not armed yet.
 607                  */
 608                 ap->flags = (ap->flags & ~KPROBE_FLAG_GONE)
 609                             | KPROBE_FLAG_DISABLED;
 610         }
 611
 612         copy_kprobe(ap, p);
 613         return add_new_kprobe(ap, p);
 614 }
 615
 616 /* Try to disable aggr_kprobe, and return 1 if succeeded.*/
 617 static int __kprobes try_to_disable_aggr_kprobe(struct kprobe *p)
 618 {
 619         struct kprobe *kp;
 620
 621         list_for_each_entry_rcu(kp, &p->list, list) {
 622                 if (!kprobe_disabled(kp))
 623                         /*
 624                          * There is an active probe on the list.
 625                          * We can't disable aggr_kprobe.
 626                          */
 627                         return 0;
 628         }
 629         p->flags |= KPROBE_FLAG_DISABLED;
 630         return 1;
 631 }
 632
 633 static int __kprobes in_kprobes_functions(unsigned long addr)
 634 {
 635         struct kprobe_blackpoint *kb;
 636
 637         if (addr >= (unsigned long)__kprobes_text_start &&
 638             addr < (unsigned long)__kprobes_text_end)
 639                 return -EINVAL;
 640         /*
 641          * If there exists a kprobe_blacklist, verify and
 642          * fail any probe registration in the prohibited area
 643          */
 644         for (kb = kprobe_blacklist; kb->name != NULL; kb++) {
 645                 if (kb->start_addr) {
 646                         if (addr >= kb->start_addr &&
 647                             addr < (kb->start_addr + kb->range))
 648                                 return -EINVAL;
 649                 }
 650         }
 651         return 0;
 652 }
 653
 654 /*
 655  * If we have a symbol_name argument, look it up and add the offset field
 656  * to it. This way, we can specify a relative address to a symbol.
 657  */
 658 static kprobe_opcode_t __kprobes *kprobe_addr(struct kprobe *p)
 659 {
 660         kprobe_opcode_t *addr = p->addr;
 661         if (p->symbol_name) {
 662                 if (addr)
 663                         return NULL;
 664                 kprobe_lookup_name(p->symbol_name, addr);
 665         }
 666
 667         if (!addr)
 668                 return NULL;
 669         return (kprobe_opcode_t *)(((char *)addr) + p->offset);
 670 }
 671
 672 int __kprobes register_kprobe(struct kprobe *p)
 673 {
 674         int ret = 0;
 675         struct kprobe *old_p;
 676         struct module *probed_mod;
 677         kprobe_opcode_t *addr;
 678
 679         addr = kprobe_addr(p);
 680         if (!addr)
 681                 return -EINVAL;
 682         p->addr = addr;
 683
 684         preempt_disable();
 685         if (!__kernel_text_address((unsigned long) p->addr) ||
 686             in_kprobes_functions((unsigned long) p->addr)) {
 687                 preempt_enable();
 688                 return -EINVAL;
 689         }
 690
 691         /* User can pass only KPROBE_FLAG_DISABLED to register_kprobe */
 692         p->flags &= KPROBE_FLAG_DISABLED;
 693
 694         /*
 695          * Check if are we probing a module.
 696          */
 697         probed_mod = __module_text_address((unsigned long) p->addr);
 698         if (probed_mod) {
 699                 /*
 700                  * We must hold a refcount of the probed module while updating
 701                  * its code to prohibit unexpected unloading.
 702                  */
 703                 if (unlikely(!try_module_get(probed_mod))) {
 704                         preempt_enable();
 705                         return -EINVAL;
 706                 }
 707                 /*
 708                  * If the module freed .init.text, we couldn't insert
 709                  * kprobes in there.
 710                  */
 711                 if (within_module_init((unsigned long)p->addr, probed_mod) &&
 712                     probed_mod->state != MODULE_STATE_COMING) {
 713                         module_put(probed_mod);
 714                         preempt_enable();
 715                         return -EINVAL;
 716                 }
 717         }
 718         preempt_enable();
 719
 720         p->nmissed = 0;
 721         INIT_LIST_HEAD(&p->list);
 722         mutex_lock(&kprobe_mutex);
 723         old_p = get_kprobe(p->addr);
 724         if (old_p) {
 725                 ret = register_aggr_kprobe(old_p, p);
 726                 goto out;
 727         }
 728
 729         mutex_lock(&text_mutex);
 730         ret = arch_prepare_kprobe(p);
 731         if (ret)
 732                 goto out_unlock_text;
 733
 734         INIT_HLIST_NODE(&p->hlist);
 735         hlist_add_head_rcu(&p->hlist,
 736                        &kprobe_table[hash_ptr(p->addr, KPROBE_HASH_BITS)]);
 737
 738         if (!kprobes_all_disarmed && !kprobe_disabled(p))
 739                 arch_arm_kprobe(p);
 740
 741 out_unlock_text:
 742         mutex_unlock(&text_mutex);
 743 out:
 744         mutex_unlock(&kprobe_mutex);
 745
 746         if (probed_mod)
 747                 module_put(probed_mod);
 748
 749         return ret;
 750 }
 751 EXPORT_SYMBOL_GPL(register_kprobe);
 752
 753 /* Check passed kprobe is valid and return kprobe in kprobe_table. */
 754 static struct kprobe * __kprobes __get_valid_kprobe(struct kprobe *p)
 755 {
 756         struct kprobe *old_p, *list_p;
 757
 758         old_p = get_kprobe(p->addr);
 759         if (unlikely(!old_p))
 760                 return NULL;
 761
 762         if (p != old_p) {
 763                 list_for_each_entry_rcu(list_p, &old_p->list, list)
 764                         if (list_p == p)
 765                         /* kprobe p is a valid probe */
 766                                 goto valid;
 767                 return NULL;
 768         }
 769 valid:
 770         return old_p;
 771 }
 772
 773 /*
 774  * Unregister a kprobe without a scheduler synchronization.
 775  */
 776 static int __kprobes __unregister_kprobe_top(struct kprobe *p)
 777 {
 778         struct kprobe *old_p, *list_p;
 779
 780         old_p = __get_valid_kprobe(p);
 781         if (old_p == NULL)
 782                 return -EINVAL;
 783
 784         if (old_p == p ||
 785             (old_p->pre_handler == aggr_pre_handler &&
 786              list_is_singular(&old_p->list))) {
 787                 /*
 788                  * Only probe on the hash list. Disarm only if kprobes are
 789                  * enabled and not gone - otherwise, the breakpoint would
 790                  * already have been removed. We save on flushing icache.
 791                  */
 792                 if (!kprobes_all_disarmed && !kprobe_disabled(old_p)) {
 793                         mutex_lock(&text_mutex);
 794                         arch_disarm_kprobe(p);
 795                         mutex_unlock(&text_mutex);
 796                 }
 797                 hlist_del_rcu(&old_p->hlist);
 798         } else {
 799                 if (p->break_handler && !kprobe_gone(p))
 800                         old_p->break_handler = NULL;
 801                 if (p->post_handler && !kprobe_gone(p)) {
 802                         list_for_each_entry_rcu(list_p, &old_p->list, list) {
 803                                 if ((list_p != p) && (list_p->post_handler))
 804                                         goto noclean;
 805                         }
 806                         old_p->post_handler = NULL;
 807                 }
 808 noclean:
 809                 list_del_rcu(&p->list);
 810                 if (!kprobe_disabled(old_p)) {
 811                         try_to_disable_aggr_kprobe(old_p);
 812                         if (!kprobes_all_disarmed && kprobe_disabled(old_p))
 813                                 arch_disarm_kprobe(old_p);
 814                 }
 815         }
 816         return 0;
 817 }
 818
 819 static void __kprobes __unregister_kprobe_bottom(struct kprobe *p)
 820 {
 821         struct kprobe *old_p;
 822
 823         if (list_empty(&p->list))
 824                 arch_remove_kprobe(p);
 825         else if (list_is_singular(&p->list)) {
 826                 /* "p" is the last child of an aggr_kprobe */
 827                 old_p = list_entry(p->list.next, struct kprobe, list);
 828                 list_del(&p->list);
 829                 arch_remove_kprobe(old_p);
 830                 kfree(old_p);
 831         }
 832 }
 833
 834 int __kprobes register_kprobes(struct kprobe **kps, int num)
 835 {
 836         int i, ret = 0;
 837
 838         if (num <= 0)
 839                 return -EINVAL;
 840         for (i = 0; i < num; i++) {
 841                 ret = register_kprobe(kps[i]);
 842                 if (ret < 0) {
 843                         if (i > 0)
 844                                 unregister_kprobes(kps, i);
 845                         break;
 846                 }
 847         }
 848         return ret;
 849 }
 850 EXPORT_SYMBOL_GPL(register_kprobes);
 851
 852 void __kprobes unregister_kprobe(struct kprobe *p)
 853 {
 854         unregister_kprobes(&p, 1);
 855 }
 856 EXPORT_SYMBOL_GPL(unregister_kprobe);
 857
 858 void __kprobes unregister_kprobes(struct kprobe **kps, int num)
 859 {
 860         int i;
 861
 862         if (num <= 0)
 863                 return;
 864         mutex_lock(&kprobe_mutex);
 865         for (i = 0; i < num; i++)
 866                 if (__unregister_kprobe_top(kps[i]) < 0)
 867                         kps[i]->addr = NULL;
 868         mutex_unlock(&kprobe_mutex);
 869
 870         synchronize_sched();
 871         for (i = 0; i < num; i++)
 872                 if (kps[i]->addr)
 873                         __unregister_kprobe_bottom(kps[i]);
 874 }
 875 EXPORT_SYMBOL_GPL(unregister_kprobes);
 876
 877 static struct notifier_block kprobe_exceptions_nb = {
 878         .notifier_call = kprobe_exceptions_notify,
 879         .priority = 0x7fffffff /* we need to be notified first */
 880 };
 881
 882 unsigned long __weak arch_deref_entry_point(void *entry)
 883 {
 884         return (unsigned long)entry;
 885 }
 886
 887 int __kprobes register_jprobes(struct jprobe **jps, int num)
 888 {
 889         struct jprobe *jp;
 890         int ret = 0, i;
 891
 892         if (num <= 0)
 893                 return -EINVAL;
 894         for (i = 0; i < num; i++) {
 895                 unsigned long addr;
 896                 jp = jps[i];
 897                 addr = arch_deref_entry_point(jp->entry);
 898
 899                 if (!kernel_text_address(addr))
 900                         ret = -EINVAL;
 901                 else {
 902                         /* Todo: Verify probepoint is a function entry point */
 903                         jp->kp.pre_handler = setjmp_pre_handler;
 904                         jp->kp.break_handler = longjmp_break_handler;
 905                         ret = register_kprobe(&jp->kp);
 906                 }
 907                 if (ret < 0) {
 908                         if (i > 0)
 909                                 unregister_jprobes(jps, i);
 910                         break;
 911                 }
 912         }
 913         return ret;
 914 }
 915 EXPORT_SYMBOL_GPL(register_jprobes);
 916
 917 int __kprobes register_jprobe(struct jprobe *jp)
 918 {
 919         return register_jprobes(&jp, 1);
 920 }
 921 EXPORT_SYMBOL_GPL(register_jprobe);
 922
 923 void __kprobes unregister_jprobe(struct jprobe *jp)
 924 {
 925         unregister_jprobes(&jp, 1);
 926 }
 927 EXPORT_SYMBOL_GPL(unregister_jprobe);
 928
 929 void __kprobes unregister_jprobes(struct jprobe **jps, int num)
 930 {
 931         int i;
 932
 933         if (num <= 0)
 934                 return;
 935         mutex_lock(&kprobe_mutex);
 936         for (i = 0; i < num; i++)
 937                 if (__unregister_kprobe_top(&jps[i]->kp) < 0)
 938                         jps[i]->kp.addr = NULL;
 939         mutex_unlock(&kprobe_mutex);
 940
 941         synchronize_sched();
 942         for (i = 0; i < num; i++) {
 943                 if (jps[i]->kp.addr)
 944                         __unregister_kprobe_bottom(&jps[i]->kp);
 945         }
 946 }
 947 EXPORT_SYMBOL_GPL(unregister_jprobes);
 948
 949 #ifdef CONFIG_KRETPROBES
 950 /*
 951  * This kprobe pre_handler is registered with every kretprobe. When probe
 952  * hits it will set up the return probe.
 953  */
 954 static int __kprobes pre_handler_kretprobe(struct kprobe *p,
 955                                            struct pt_regs *regs)
 956 {
 957         struct kretprobe *rp = container_of(p, struct kretprobe, kp);
 958         unsigned long hash, flags = 0;
 959         struct kretprobe_instance *ri;
 960
 961         /*TODO: consider to only swap the RA after the last pre_handler fired */
 962         hash = hash_ptr(current, KPROBE_HASH_BITS);
 963         spin_lock_irqsave(&rp->lock, flags);
 964         if (!hlist_empty(&rp->free_instances)) {
 965                 ri = hlist_entry(rp->free_instances.first,
 966                                 struct kretprobe_instance, hlist);
 967                 hlist_del(&ri->hlist);
 968                 spin_unlock_irqrestore(&rp->lock, flags);
 969
 970                 ri->rp = rp;
 971                 ri->task = current;
 972
 973                 if (rp->entry_handler && rp->entry_handler(ri, regs))
 974                         return 0;
 975
 976                 arch_prepare_kretprobe(ri, regs);
 977
 978                 /* XXX(hch): why is there no hlist_move_head? */
 979                 INIT_HLIST_NODE(&ri->hlist);
 980                 kretprobe_table_lock(hash, &flags);
 981                 hlist_add_head(&ri->hlist, &kretprobe_inst_table[hash]);
 982                 kretprobe_table_unlock(hash, &flags);
 983         } else {
 984                 rp->nmissed++;
 985                 spin_unlock_irqrestore(&rp->lock, flags);
 986         }
 987         return 0;
 988 }
 989
 990 int __kprobes register_kretprobe(struct kretprobe *rp)
 991 {
 992         int ret = 0;
 993         struct kretprobe_instance *inst;
 994         int i;
 995         void *addr;
 996
 997         if (kretprobe_blacklist_size) {
 998                 addr = kprobe_addr(&rp->kp);
 999                 if (!addr)
1000                         return -EINVAL;
1001
1002                 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
1003                         if (kretprobe_blacklist[i].addr == addr)
1004                                 return -EINVAL;
1005                 }
1006         }
1007
1008         rp->kp.pre_handler = pre_handler_kretprobe;
1009         rp->kp.post_handler = NULL;
1010         rp->kp.fault_handler = NULL;
1011         rp->kp.break_handler = NULL;
1012
1013         /* Pre-allocate memory for max kretprobe instances */
1014         if (rp->maxactive <= 0) {
1015 #ifdef CONFIG_PREEMPT
1016                 rp->maxactive = max(10, 2 * NR_CPUS);
1017 #else
1018                 rp->maxactive = NR_CPUS;
1019 #endif
1020         }
1021         spin_lock_init(&rp->lock);
1022         INIT_HLIST_HEAD(&rp->free_instances);
1023         for (i = 0; i < rp->maxactive; i++) {
1024                 inst = kmalloc(sizeof(struct kretprobe_instance) +
1025                                rp->data_size, GFP_KERNEL);
1026                 if (inst == NULL) {
1027                         free_rp_inst(rp);
1028                         return -ENOMEM;
1029                 }
1030                 INIT_HLIST_NODE(&inst->hlist);
1031                 hlist_add_head(&inst->hlist, &rp->free_instances);
1032         }
1033
1034         rp->nmissed = 0;
1035         /* Establish function entry probe point */
1036         ret = register_kprobe(&rp->kp);
1037         if (ret != 0)
1038                 free_rp_inst(rp);
1039         return ret;
1040 }
1041 EXPORT_SYMBOL_GPL(register_kretprobe);
1042
1043 int __kprobes register_kretprobes(struct kretprobe **rps, int num)
1044 {
1045         int ret = 0, i;
1046
1047         if (num <= 0)
1048                 return -EINVAL;
1049         for (i = 0; i < num; i++) {
1050                 ret = register_kretprobe(rps[i]);
1051                 if (ret < 0) {
1052                         if (i > 0)
1053                                 unregister_kretprobes(rps, i);
1054                         break;
1055                 }
1056         }
1057         return ret;
1058 }
1059 EXPORT_SYMBOL_GPL(register_kretprobes);
1060
1061 void __kprobes unregister_kretprobe(struct kretprobe *rp)
1062 {
1063         unregister_kretprobes(&rp, 1);
1064 }
1065 EXPORT_SYMBOL_GPL(unregister_kretprobe);
1066
1067 void __kprobes unregister_kretprobes(struct kretprobe **rps, int num)
1068 {
1069         int i;
1070
1071         if (num <= 0)
1072                 return;
1073         mutex_lock(&kprobe_mutex);
1074         for (i = 0; i < num; i++)
1075                 if (__unregister_kprobe_top(&rps[i]->kp) < 0)
1076                         rps[i]->kp.addr = NULL;
1077         mutex_unlock(&kprobe_mutex);
1078
1079         synchronize_sched();
1080         for (i = 0; i < num; i++) {
1081                 if (rps[i]->kp.addr) {
1082                         __unregister_kprobe_bottom(&rps[i]->kp);
1083                         cleanup_rp_inst(rps[i]);
1084                 }
1085         }
1086 }
1087 EXPORT_SYMBOL_GPL(unregister_kretprobes);
1088
1089 #else /* CONFIG_KRETPROBES */
1090 int __kprobes register_kretprobe(struct kretprobe *rp)
1091 {
1092         return -ENOSYS;
1093 }
1094 EXPORT_SYMBOL_GPL(register_kretprobe);
1095
1096 int __kprobes register_kretprobes(struct kretprobe **rps, int num)
1097 {
1098         return -ENOSYS;
1099 }
1100 EXPORT_SYMBOL_GPL(register_kretprobes);
1101
1102 void __kprobes unregister_kretprobe(struct kretprobe *rp)
1103 {
1104 }
1105 EXPORT_SYMBOL_GPL(unregister_kretprobe);
1106
1107 void __kprobes unregister_kretprobes(struct kretprobe **rps, int num)
1108 {
1109 }
1110 EXPORT_SYMBOL_GPL(unregister_kretprobes);
1111
1112 static int __kprobes pre_handler_kretprobe(struct kprobe *p,
1113                                            struct pt_regs *regs)
1114 {
1115         return 0;
1116 }
1117
1118 #endif /* CONFIG_KRETPROBES */
1119
1120 /* Set the kprobe gone and remove its instruction buffer. */
1121 static void __kprobes kill_kprobe(struct kprobe *p)
1122 {
1123         struct kprobe *kp;
1124
1125         p->flags |= KPROBE_FLAG_GONE;
1126         if (p->pre_handler == aggr_pre_handler) {
1127                 /*
1128                  * If this is an aggr_kprobe, we have to list all the
1129                  * chained probes and mark them GONE.
1130                  */
1131                 list_for_each_entry_rcu(kp, &p->list, list)
1132                         kp->flags |= KPROBE_FLAG_GONE;
1133                 p->post_handler = NULL;
1134                 p->break_handler = NULL;
1135         }
1136         /*
1137          * Here, we can remove insn_slot safely, because no thread calls
1138          * the original probed function (which will be freed soon) any more.
1139          */
1140         arch_remove_kprobe(p);
1141 }
1142
1143 /* Module notifier call back, checking kprobes on the module */
1144 static int __kprobes kprobes_module_callback(struct notifier_block *nb,
1145                                              unsigned long val, void *data)
1146 {
1147         struct module *mod = data;
1148         struct hlist_head *head;
1149         struct hlist_node *node;
1150         struct kprobe *p;
1151         unsigned int i;
1152         int checkcore = (val == MODULE_STATE_GOING);
1153
1154         if (val != MODULE_STATE_GOING && val != MODULE_STATE_LIVE)
1155                 return NOTIFY_DONE;
1156
1157         /*
1158          * When MODULE_STATE_GOING was notified, both of module .text and
1159          * .init.text sections would be freed. When MODULE_STATE_LIVE was
1160          * notified, only .init.text section would be freed. We need to
1161          * disable kprobes which have been inserted in the sections.
1162          */
1163         mutex_lock(&kprobe_mutex);
1164         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
1165                 head = &kprobe_table[i];
1166                 hlist_for_each_entry_rcu(p, node, head, hlist)
1167                         if (within_module_init((unsigned long)p->addr, mod) ||
1168                             (checkcore &&
1169                              within_module_core((unsigned long)p->addr, mod))) {
1170                                 /*
1171                                  * The vaddr this probe is installed will soon
1172                                  * be vfreed buy not synced to disk. Hence,
1173                                  * disarming the breakpoint isn't needed.
1174                                  */
1175                                 kill_kprobe(p);
1176                         }
1177         }
1178         mutex_unlock(&kprobe_mutex);
1179         return NOTIFY_DONE;
1180 }
1181
1182 static struct notifier_block kprobe_module_nb = {
1183         .notifier_call = kprobes_module_callback,
1184         .priority = 0
1185 };
1186
1187 static int __init init_kprobes(void)
1188 {
1189         int i, err = 0;
1190         unsigned long offset = 0, size = 0;
1191         char *modname, namebuf[128];
1192         const char *symbol_name;
1193         void *addr;
1194         struct kprobe_blackpoint *kb;
1195
1196         /* FIXME allocate the probe table, currently defined statically */
1197         /* initialize all list heads */
1198         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
1199                 INIT_HLIST_HEAD(&kprobe_table[i]);
1200                 INIT_HLIST_HEAD(&kretprobe_inst_table[i]);
1201                 spin_lock_init(&(kretprobe_table_locks[i].lock));
1202         }
1203
1204         /*
1205          * Lookup and populate the kprobe_blacklist.
1206          *
1207          * Unlike the kretprobe blacklist, we'll need to determine
1208          * the range of addresses that belong to the said functions,
1209          * since a kprobe need not necessarily be at the beginning
1210          * of a function.
1211          */
1212         for (kb = kprobe_blacklist; kb->name != NULL; kb++) {
1213                 kprobe_lookup_name(kb->name, addr);
1214                 if (!addr)
1215                         continue;
1216
1217                 kb->start_addr = (unsigned long)addr;
1218                 symbol_name = kallsyms_lookup(kb->start_addr,
1219                                 &size, &offset, &modname, namebuf);
1220                 if (!symbol_name)
1221                         kb->range = 0;
1222                 else
1223                         kb->range = size;
1224         }
1225
1226         if (kretprobe_blacklist_size) {
1227                 /* lookup the function address from its name */
1228                 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
1229                         kprobe_lookup_name(kretprobe_blacklist[i].name,
1230                                            kretprobe_blacklist[i].addr);
1231                         if (!kretprobe_blacklist[i].addr)
1232                                 printk("kretprobe: lookup failed: %s\n",
1233                                        kretprobe_blacklist[i].name);
1234                 }
1235         }
1236
1237         /* By default, kprobes are armed */
1238         kprobes_all_disarmed = false;
1239
1240         err = arch_init_kprobes();
1241         if (!err)
1242                 err = register_die_notifier(&kprobe_exceptions_nb);
1243         if (!err)
1244                 err = register_module_notifier(&kprobe_module_nb);
1245
1246         kprobes_initialized = (err == 0);
1247
1248         if (!err)
1249                 init_test_probes();
1250         return err;
1251 }
1252
1253 #ifdef CONFIG_DEBUG_FS
1254 static void __kprobes report_probe(struct seq_file *pi, struct kprobe *p,
1255                 const char *sym, int offset,char *modname)
1256 {
1257         char *kprobe_type;
1258
1259         if (p->pre_handler == pre_handler_kretprobe)
1260                 kprobe_type = "r";
1261         else if (p->pre_handler == setjmp_pre_handler)
1262                 kprobe_type = "j";
1263         else
1264                 kprobe_type = "k";
1265         if (sym)
1266                 seq_printf(pi, "%p  %s  %s+0x%x  %s %s%s\n",
1267                         p->addr, kprobe_type, sym, offset,
1268                         (modname ? modname : " "),
1269                         (kprobe_gone(p) ? "[GONE]" : ""),
1270                         ((kprobe_disabled(p) && !kprobe_gone(p)) ?
1271                          "[DISABLED]" : ""));
1272         else
1273                 seq_printf(pi, "%p  %s  %p %s%s\n",
1274                         p->addr, kprobe_type, p->addr,
1275                         (kprobe_gone(p) ? "[GONE]" : ""),
1276                         ((kprobe_disabled(p) && !kprobe_gone(p)) ?
1277                          "[DISABLED]" : ""));
1278 }
1279
1280 static void __kprobes *kprobe_seq_start(struct seq_file *f, loff_t *pos)
1281 {
1282         return (*pos < KPROBE_TABLE_SIZE) ? pos : NULL;
1283 }
1284
1285 static void __kprobes *kprobe_seq_next(struct seq_file *f, void *v, loff_t *pos)
1286 {
1287         (*pos)++;
1288         if (*pos >= KPROBE_TABLE_SIZE)
1289                 return NULL;
1290         return pos;
1291 }
1292
1293 static void __kprobes kprobe_seq_stop(struct seq_file *f, void *v)
1294 {
1295         /* Nothing to do */
1296 }
1297
1298 static int __kprobes show_kprobe_addr(struct seq_file *pi, void *v)
1299 {
1300         struct hlist_head *head;
1301         struct hlist_node *node;
1302         struct kprobe *p, *kp;
1303         const char *sym = NULL;
1304         unsigned int i = *(loff_t *) v;
1305         unsigned long offset = 0;
1306         char *modname, namebuf[128];
1307
1308         head = &kprobe_table[i];
1309         preempt_disable();
1310         hlist_for_each_entry_rcu(p, node, head, hlist) {
1311                 sym = kallsyms_lookup((unsigned long)p->addr, NULL,
1312                                         &offset, &modname, namebuf);
1313                 if (p->pre_handler == aggr_pre_handler) {
1314                         list_for_each_entry_rcu(kp, &p->list, list)
1315                                 report_probe(pi, kp, sym, offset, modname);
1316                 } else
1317                         report_probe(pi, p, sym, offset, modname);
1318         }
1319         preempt_enable();
1320         return 0;
1321 }
1322
1323 static struct seq_operations kprobes_seq_ops = {
1324         .start = kprobe_seq_start,
1325         .next  = kprobe_seq_next,
1326         .stop  = kprobe_seq_stop,
1327         .show  = show_kprobe_addr
1328 };
1329
1330 static int __kprobes kprobes_open(struct inode *inode, struct file *filp)
1331 {
1332         return seq_open(filp, &kprobes_seq_ops);
1333 }
1334
1335 static struct file_operations debugfs_kprobes_operations = {
1336         .open           = kprobes_open,
1337         .read           = seq_read,
1338         .llseek         = seq_lseek,
1339         .release        = seq_release,
1340 };
1341
1342 /* Disable one kprobe */
1343 int __kprobes disable_kprobe(struct kprobe *kp)
1344 {
1345         int ret = 0;
1346         struct kprobe *p;
1347
1348         mutex_lock(&kprobe_mutex);
1349
1350         /* Check whether specified probe is valid. */
1351         p = __get_valid_kprobe(kp);
1352         if (unlikely(p == NULL)) {
1353                 ret = -EINVAL;
1354                 goto out;
1355         }
1356
1357         /* If the probe is already disabled (or gone), just return */
1358         if (kprobe_disabled(kp))
1359                 goto out;
1360
1361         kp->flags |= KPROBE_FLAG_DISABLED;
1362         if (p != kp)
1363                 /* When kp != p, p is always enabled. */
1364                 try_to_disable_aggr_kprobe(p);
1365
1366         if (!kprobes_all_disarmed && kprobe_disabled(p))
1367                 arch_disarm_kprobe(p);
1368 out:
1369         mutex_unlock(&kprobe_mutex);
1370         return ret;
1371 }
1372 EXPORT_SYMBOL_GPL(disable_kprobe);
1373
1374 /* Enable one kprobe */
1375 int __kprobes enable_kprobe(struct kprobe *kp)
1376 {
1377         int ret = 0;
1378         struct kprobe *p;
1379
1380         mutex_lock(&kprobe_mutex);
1381
1382         /* Check whether specified probe is valid. */
1383         p = __get_valid_kprobe(kp);
1384         if (unlikely(p == NULL)) {
1385                 ret = -EINVAL;
1386                 goto out;
1387         }
1388
1389         if (kprobe_gone(kp)) {
1390                 /* This kprobe has gone, we couldn't enable it. */
1391                 ret = -EINVAL;
1392                 goto out;
1393         }
1394
1395         if (!kprobes_all_disarmed && kprobe_disabled(p))
1396                 arch_arm_kprobe(p);
1397
1398         p->flags &= ~KPROBE_FLAG_DISABLED;
1399         if (p != kp)
1400                 kp->flags &= ~KPROBE_FLAG_DISABLED;
1401 out:
1402         mutex_unlock(&kprobe_mutex);
1403         return ret;
1404 }
1405 EXPORT_SYMBOL_GPL(enable_kprobe);
1406
1407 static void __kprobes arm_all_kprobes(void)
1408 {
1409         struct hlist_head *head;
1410         struct hlist_node *node;
1411         struct kprobe *p;
1412         unsigned int i;
1413
1414         mutex_lock(&kprobe_mutex);
1415
1416         /* If kprobes are armed, just return */
1417         if (!kprobes_all_disarmed)
1418                 goto already_enabled;
1419
1420         mutex_lock(&text_mutex);
1421         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
1422                 head = &kprobe_table[i];
1423                 hlist_for_each_entry_rcu(p, node, head, hlist)
1424                         if (!kprobe_disabled(p))
1425                                 arch_arm_kprobe(p);
1426         }
1427         mutex_unlock(&text_mutex);
1428
1429         kprobes_all_disarmed = false;
1430         printk(KERN_INFO "Kprobes globally enabled\n");
1431
1432 already_enabled:
1433         mutex_unlock(&kprobe_mutex);
1434         return;
1435 }
1436
1437 static void __kprobes disarm_all_kprobes(void)
1438 {
1439         struct hlist_head *head;
1440         struct hlist_node *node;
1441         struct kprobe *p;
1442         unsigned int i;
1443
1444         mutex_lock(&kprobe_mutex);
1445
1446         /* If kprobes are already disarmed, just return */
1447         if (kprobes_all_disarmed)
1448                 goto already_disabled;
1449
1450         kprobes_all_disarmed = true;
1451         printk(KERN_INFO "Kprobes globally disabled\n");
1452         mutex_lock(&text_mutex);
1453         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
1454                 head = &kprobe_table[i];
1455                 hlist_for_each_entry_rcu(p, node, head, hlist) {
1456                         if (!arch_trampoline_kprobe(p) && !kprobe_disabled(p))
1457                                 arch_disarm_kprobe(p);
1458                 }
1459         }
1460
1461         mutex_unlock(&text_mutex);
1462         mutex_unlock(&kprobe_mutex);
1463         /* Allow all currently running kprobes to complete */
1464         synchronize_sched();
1465         return;
1466
1467 already_disabled:
1468         mutex_unlock(&kprobe_mutex);
1469         return;
1470 }
1471
1472 /*
1473  * XXX: The debugfs bool file interface doesn't allow for callbacks
1474  * when the bool state is switched. We can reuse that facility when
1475  * available
1476  */
1477 static ssize_t read_enabled_file_bool(struct file *file,
1478                char __user *user_buf, size_t count, loff_t *ppos)
1479 {
1480         char buf[3];
1481
1482         if (!kprobes_all_disarmed)
1483                 buf[0] = '1';
1484         else
1485                 buf[0] = '0';
1486         buf[1] = '\n';
1487         buf[2] = 0x00;
1488         return simple_read_from_buffer(user_buf, count, ppos, buf, 2);
1489 }
1490
1491 static ssize_t write_enabled_file_bool(struct file *file,
1492                const char __user *user_buf, size_t count, loff_t *ppos)
1493 {
1494         char buf[32];
1495         int buf_size;
1496
1497         buf_size = min(count, (sizeof(buf)-1));
1498         if (copy_from_user(buf, user_buf, buf_size))
1499                 return -EFAULT;
1500
1501         switch (buf[0]) {
1502         case 'y':
1503         case 'Y':
1504         case '1':
1505                 arm_all_kprobes();
1506                 break;
1507         case 'n':
1508         case 'N':
1509         case '0':
1510                 disarm_all_kprobes();
1511                 break;
1512         }
1513
1514         return count;
1515 }
1516
1517 static struct file_operations fops_kp = {
1518         .read =         read_enabled_file_bool,
1519         .write =        write_enabled_file_bool,
1520 };
1521
1522 static int __kprobes debugfs_kprobe_init(void)
1523 {
1524         struct dentry *dir, *file;
1525         unsigned int value = 1;
1526
1527         dir = debugfs_create_dir("kprobes", NULL);
1528         if (!dir)
1529                 return -ENOMEM;
1530
1531         file = debugfs_create_file("list", 0444, dir, NULL,
1532                                 &debugfs_kprobes_operations);
1533         if (!file) {
1534                 debugfs_remove(dir);
1535                 return -ENOMEM;
1536         }
1537
1538         file = debugfs_create_file("enabled", 0600, dir,
1539                                         &value, &fops_kp);
1540         if (!file) {
1541                 debugfs_remove(dir);
1542                 return -ENOMEM;
1543         }
1544
1545         return 0;
1546 }
1547
1548 late_initcall(debugfs_kprobe_init);
1549 #endif /* CONFIG_DEBUG_FS */
1550
1551 module_init(init_kprobes);
1552
1553 /* defined in arch/.../kernel/kprobes.c */
1554 EXPORT_SYMBOL_GPL(jprobe_return);