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 /* Arm a kprobe with text_mutex */
 323 static void __kprobes arm_kprobe(struct kprobe *kp)
 324 {
 325         mutex_lock(&text_mutex);
 326         arch_arm_kprobe(kp);
 327         mutex_unlock(&text_mutex);
 328 }
 329
 330 /* Disarm a kprobe with text_mutex */
 331 static void __kprobes disarm_kprobe(struct kprobe *kp)
 332 {
 333         mutex_lock(&text_mutex);
 334         arch_disarm_kprobe(kp);
 335         mutex_unlock(&text_mutex);
 336 }
 337
 338 /*
 339  * Aggregate handlers for multiple kprobes support - these handlers
 340  * take care of invoking the individual kprobe handlers on p->list
 341  */
 342 static int __kprobes aggr_pre_handler(struct kprobe *p, struct pt_regs *regs)
 343 {
 344         struct kprobe *kp;
 345
 346         list_for_each_entry_rcu(kp, &p->list, list) {
 347                 if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
 348                         set_kprobe_instance(kp);
 349                         if (kp->pre_handler(kp, regs))
 350                                 return 1;
 351                 }
 352                 reset_kprobe_instance();
 353         }
 354         return 0;
 355 }
 356
 357 static void __kprobes aggr_post_handler(struct kprobe *p, struct pt_regs *regs,
 358                                         unsigned long flags)
 359 {
 360         struct kprobe *kp;
 361
 362         list_for_each_entry_rcu(kp, &p->list, list) {
 363                 if (kp->post_handler && likely(!kprobe_disabled(kp))) {
 364                         set_kprobe_instance(kp);
 365                         kp->post_handler(kp, regs, flags);
 366                         reset_kprobe_instance();
 367                 }
 368         }
 369 }
 370
 371 static int __kprobes aggr_fault_handler(struct kprobe *p, struct pt_regs *regs,
 372                                         int trapnr)
 373 {
 374         struct kprobe *cur = __get_cpu_var(kprobe_instance);
 375
 376         /*
 377          * if we faulted "during" the execution of a user specified
 378          * probe handler, invoke just that probe's fault handler
 379          */
 380         if (cur && cur->fault_handler) {
 381                 if (cur->fault_handler(cur, regs, trapnr))
 382                         return 1;
 383         }
 384         return 0;
 385 }
 386
 387 static int __kprobes aggr_break_handler(struct kprobe *p, struct pt_regs *regs)
 388 {
 389         struct kprobe *cur = __get_cpu_var(kprobe_instance);
 390         int ret = 0;
 391
 392         if (cur && cur->break_handler) {
 393                 if (cur->break_handler(cur, regs))
 394                         ret = 1;
 395         }
 396         reset_kprobe_instance();
 397         return ret;
 398 }
 399
 400 /* Walks the list and increments nmissed count for multiprobe case */
 401 void __kprobes kprobes_inc_nmissed_count(struct kprobe *p)
 402 {
 403         struct kprobe *kp;
 404         if (p->pre_handler != aggr_pre_handler) {
 405                 p->nmissed++;
 406         } else {
 407                 list_for_each_entry_rcu(kp, &p->list, list)
 408                         kp->nmissed++;
 409         }
 410         return;
 411 }
 412
 413 void __kprobes recycle_rp_inst(struct kretprobe_instance *ri,
 414                                 struct hlist_head *head)
 415 {
 416         struct kretprobe *rp = ri->rp;
 417
 418         /* remove rp inst off the rprobe_inst_table */
 419         hlist_del(&ri->hlist);
 420         INIT_HLIST_NODE(&ri->hlist);
 421         if (likely(rp)) {
 422                 spin_lock(&rp->lock);
 423                 hlist_add_head(&ri->hlist, &rp->free_instances);
 424                 spin_unlock(&rp->lock);
 425         } else
 426                 /* Unregistering */
 427                 hlist_add_head(&ri->hlist, head);
 428 }
 429
 430 void __kprobes kretprobe_hash_lock(struct task_struct *tsk,
 431                          struct hlist_head **head, unsigned long *flags)
 432 {
 433         unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
 434         spinlock_t *hlist_lock;
 435
 436         *head = &kretprobe_inst_table[hash];
 437         hlist_lock = kretprobe_table_lock_ptr(hash);
 438         spin_lock_irqsave(hlist_lock, *flags);
 439 }
 440
 441 static void __kprobes kretprobe_table_lock(unsigned long hash,
 442         unsigned long *flags)
 443 {
 444         spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
 445         spin_lock_irqsave(hlist_lock, *flags);
 446 }
 447
 448 void __kprobes kretprobe_hash_unlock(struct task_struct *tsk,
 449         unsigned long *flags)
 450 {
 451         unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
 452         spinlock_t *hlist_lock;
 453
 454         hlist_lock = kretprobe_table_lock_ptr(hash);
 455         spin_unlock_irqrestore(hlist_lock, *flags);
 456 }
 457
 458 void __kprobes kretprobe_table_unlock(unsigned long hash, unsigned long *flags)
 459 {
 460         spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
 461         spin_unlock_irqrestore(hlist_lock, *flags);
 462 }
 463
 464 /*
 465  * This function is called from finish_task_switch when task tk becomes dead,
 466  * so that we can recycle any function-return probe instances associated
 467  * with this task. These left over instances represent probed functions
 468  * that have been called but will never return.
 469  */
 470 void __kprobes kprobe_flush_task(struct task_struct *tk)
 471 {
 472         struct kretprobe_instance *ri;
 473         struct hlist_head *head, empty_rp;
 474         struct hlist_node *node, *tmp;
 475         unsigned long hash, flags = 0;
 476
 477         if (unlikely(!kprobes_initialized))
 478                 /* Early boot.  kretprobe_table_locks not yet initialized. */
 479                 return;
 480
 481         hash = hash_ptr(tk, KPROBE_HASH_BITS);
 482         head = &kretprobe_inst_table[hash];
 483         kretprobe_table_lock(hash, &flags);
 484         hlist_for_each_entry_safe(ri, node, tmp, head, hlist) {
 485                 if (ri->task == tk)
 486                         recycle_rp_inst(ri, &empty_rp);
 487         }
 488         kretprobe_table_unlock(hash, &flags);
 489         INIT_HLIST_HEAD(&empty_rp);
 490         hlist_for_each_entry_safe(ri, node, tmp, &empty_rp, hlist) {
 491                 hlist_del(&ri->hlist);
 492                 kfree(ri);
 493         }
 494 }
 495
 496 static inline void free_rp_inst(struct kretprobe *rp)
 497 {
 498         struct kretprobe_instance *ri;
 499         struct hlist_node *pos, *next;
 500
 501         hlist_for_each_entry_safe(ri, pos, next, &rp->free_instances, hlist) {
 502                 hlist_del(&ri->hlist);
 503                 kfree(ri);
 504         }
 505 }
 506
 507 static void __kprobes cleanup_rp_inst(struct kretprobe *rp)
 508 {
 509         unsigned long flags, hash;
 510         struct kretprobe_instance *ri;
 511         struct hlist_node *pos, *next;
 512         struct hlist_head *head;
 513
 514         /* No race here */
 515         for (hash = 0; hash < KPROBE_TABLE_SIZE; hash++) {
 516                 kretprobe_table_lock(hash, &flags);
 517                 head = &kretprobe_inst_table[hash];
 518                 hlist_for_each_entry_safe(ri, pos, next, head, hlist) {
 519                         if (ri->rp == rp)
 520                                 ri->rp = NULL;
 521                 }
 522                 kretprobe_table_unlock(hash, &flags);
 523         }
 524         free_rp_inst(rp);
 525 }
 526
 527 /*
 528  * Keep all fields in the kprobe consistent
 529  */
 530 static inline void copy_kprobe(struct kprobe *old_p, struct kprobe *p)
 531 {
 532         memcpy(&p->opcode, &old_p->opcode, sizeof(kprobe_opcode_t));
 533         memcpy(&p->ainsn, &old_p->ainsn, sizeof(struct arch_specific_insn));
 534 }
 535
 536 /*
 537 * Add the new probe to ap->list. Fail if this is the
 538 * second jprobe at the address - two jprobes can't coexist
 539 */
 540 static int __kprobes add_new_kprobe(struct kprobe *ap, struct kprobe *p)
 541 {
 542         BUG_ON(kprobe_gone(ap) || kprobe_gone(p));
 543         if (p->break_handler) {
 544                 if (ap->break_handler)
 545                         return -EEXIST;
 546                 list_add_tail_rcu(&p->list, &ap->list);
 547                 ap->break_handler = aggr_break_handler;
 548         } else
 549                 list_add_rcu(&p->list, &ap->list);
 550         if (p->post_handler && !ap->post_handler)
 551                 ap->post_handler = aggr_post_handler;
 552
 553         if (kprobe_disabled(ap) && !kprobe_disabled(p)) {
 554                 ap->flags &= ~KPROBE_FLAG_DISABLED;
 555                 if (!kprobes_all_disarmed)
 556                         /* Arm the breakpoint again. */
 557                         arm_kprobe(ap);
 558         }
 559         return 0;
 560 }
 561
 562 /*
 563  * Fill in the required fields of the "manager kprobe". Replace the
 564  * earlier kprobe in the hlist with the manager kprobe
 565  */
 566 static inline void add_aggr_kprobe(struct kprobe *ap, struct kprobe *p)
 567 {
 568         copy_kprobe(p, ap);
 569         flush_insn_slot(ap);
 570         ap->addr = p->addr;
 571         ap->flags = p->flags;
 572         ap->pre_handler = aggr_pre_handler;
 573         ap->fault_handler = aggr_fault_handler;
 574         /* We don't care the kprobe which has gone. */
 575         if (p->post_handler && !kprobe_gone(p))
 576                 ap->post_handler = aggr_post_handler;
 577         if (p->break_handler && !kprobe_gone(p))
 578                 ap->break_handler = aggr_break_handler;
 579
 580         INIT_LIST_HEAD(&ap->list);
 581         list_add_rcu(&p->list, &ap->list);
 582
 583         hlist_replace_rcu(&p->hlist, &ap->hlist);
 584 }
 585
 586 /*
 587  * This is the second or subsequent kprobe at the address - handle
 588  * the intricacies
 589  */
 590 static int __kprobes register_aggr_kprobe(struct kprobe *old_p,
 591                                           struct kprobe *p)
 592 {
 593         int ret = 0;
 594         struct kprobe *ap = old_p;
 595
 596         if (old_p->pre_handler != aggr_pre_handler) {
 597                 /* If old_p is not an aggr_probe, create new aggr_kprobe. */
 598                 ap = kzalloc(sizeof(struct kprobe), GFP_KERNEL);
 599                 if (!ap)
 600                         return -ENOMEM;
 601                 add_aggr_kprobe(ap, old_p);
 602         }
 603
 604         if (kprobe_gone(ap)) {
 605                 /*
 606                  * Attempting to insert new probe at the same location that
 607                  * had a probe in the module vaddr area which already
 608                  * freed. So, the instruction slot has already been
 609                  * released. We need a new slot for the new probe.
 610                  */
 611                 ret = arch_prepare_kprobe(ap);
 612                 if (ret)
 613                         /*
 614                          * Even if fail to allocate new slot, don't need to
 615                          * free aggr_probe. It will be used next time, or
 616                          * freed by unregister_kprobe.
 617                          */
 618                         return ret;
 619
 620                 /*
 621                  * Clear gone flag to prevent allocating new slot again, and
 622                  * set disabled flag because it is not armed yet.
 623                  */
 624                 ap->flags = (ap->flags & ~KPROBE_FLAG_GONE)
 625                             | KPROBE_FLAG_DISABLED;
 626         }
 627
 628         copy_kprobe(ap, p);
 629         return add_new_kprobe(ap, p);
 630 }
 631
 632 /* Try to disable aggr_kprobe, and return 1 if succeeded.*/
 633 static int __kprobes try_to_disable_aggr_kprobe(struct kprobe *p)
 634 {
 635         struct kprobe *kp;
 636
 637         list_for_each_entry_rcu(kp, &p->list, list) {
 638                 if (!kprobe_disabled(kp))
 639                         /*
 640                          * There is an active probe on the list.
 641                          * We can't disable aggr_kprobe.
 642                          */
 643                         return 0;
 644         }
 645         p->flags |= KPROBE_FLAG_DISABLED;
 646         return 1;
 647 }
 648
 649 static int __kprobes in_kprobes_functions(unsigned long addr)
 650 {
 651         struct kprobe_blackpoint *kb;
 652
 653         if (addr >= (unsigned long)__kprobes_text_start &&
 654             addr < (unsigned long)__kprobes_text_end)
 655                 return -EINVAL;
 656         /*
 657          * If there exists a kprobe_blacklist, verify and
 658          * fail any probe registration in the prohibited area
 659          */
 660         for (kb = kprobe_blacklist; kb->name != NULL; kb++) {
 661                 if (kb->start_addr) {
 662                         if (addr >= kb->start_addr &&
 663                             addr < (kb->start_addr + kb->range))
 664                                 return -EINVAL;
 665                 }
 666         }
 667         return 0;
 668 }
 669
 670 /*
 671  * If we have a symbol_name argument, look it up and add the offset field
 672  * to it. This way, we can specify a relative address to a symbol.
 673  */
 674 static kprobe_opcode_t __kprobes *kprobe_addr(struct kprobe *p)
 675 {
 676         kprobe_opcode_t *addr = p->addr;
 677         if (p->symbol_name) {
 678                 if (addr)
 679                         return NULL;
 680                 kprobe_lookup_name(p->symbol_name, addr);
 681         }
 682
 683         if (!addr)
 684                 return NULL;
 685         return (kprobe_opcode_t *)(((char *)addr) + p->offset);
 686 }
 687
 688 int __kprobes register_kprobe(struct kprobe *p)
 689 {
 690         int ret = 0;
 691         struct kprobe *old_p;
 692         struct module *probed_mod;
 693         kprobe_opcode_t *addr;
 694
 695         addr = kprobe_addr(p);
 696         if (!addr)
 697                 return -EINVAL;
 698         p->addr = addr;
 699
 700         preempt_disable();
 701         if (!__kernel_text_address((unsigned long) p->addr) ||
 702             in_kprobes_functions((unsigned long) p->addr)) {
 703                 preempt_enable();
 704                 return -EINVAL;
 705         }
 706
 707         /* User can pass only KPROBE_FLAG_DISABLED to register_kprobe */
 708         p->flags &= KPROBE_FLAG_DISABLED;
 709
 710         /*
 711          * Check if are we probing a module.
 712          */
 713         probed_mod = __module_text_address((unsigned long) p->addr);
 714         if (probed_mod) {
 715                 /*
 716                  * We must hold a refcount of the probed module while updating
 717                  * its code to prohibit unexpected unloading.
 718                  */
 719                 if (unlikely(!try_module_get(probed_mod))) {
 720                         preempt_enable();
 721                         return -EINVAL;
 722                 }
 723                 /*
 724                  * If the module freed .init.text, we couldn't insert
 725                  * kprobes in there.
 726                  */
 727                 if (within_module_init((unsigned long)p->addr, probed_mod) &&
 728                     probed_mod->state != MODULE_STATE_COMING) {
 729                         module_put(probed_mod);
 730                         preempt_enable();
 731                         return -EINVAL;
 732                 }
 733         }
 734         preempt_enable();
 735
 736         p->nmissed = 0;
 737         INIT_LIST_HEAD(&p->list);
 738         mutex_lock(&kprobe_mutex);
 739         old_p = get_kprobe(p->addr);
 740         if (old_p) {
 741                 ret = register_aggr_kprobe(old_p, p);
 742                 goto out;
 743         }
 744
 745         mutex_lock(&text_mutex);
 746         ret = arch_prepare_kprobe(p);
 747         if (ret)
 748                 goto out_unlock_text;
 749
 750         INIT_HLIST_NODE(&p->hlist);
 751         hlist_add_head_rcu(&p->hlist,
 752                        &kprobe_table[hash_ptr(p->addr, KPROBE_HASH_BITS)]);
 753
 754         if (!kprobes_all_disarmed && !kprobe_disabled(p))
 755                 arch_arm_kprobe(p);
 756
 757 out_unlock_text:
 758         mutex_unlock(&text_mutex);
 759 out:
 760         mutex_unlock(&kprobe_mutex);
 761
 762         if (probed_mod)
 763                 module_put(probed_mod);
 764
 765         return ret;
 766 }
 767 EXPORT_SYMBOL_GPL(register_kprobe);
 768
 769 /* Check passed kprobe is valid and return kprobe in kprobe_table. */
 770 static struct kprobe * __kprobes __get_valid_kprobe(struct kprobe *p)
 771 {
 772         struct kprobe *old_p, *list_p;
 773
 774         old_p = get_kprobe(p->addr);
 775         if (unlikely(!old_p))
 776                 return NULL;
 777
 778         if (p != old_p) {
 779                 list_for_each_entry_rcu(list_p, &old_p->list, list)
 780                         if (list_p == p)
 781                         /* kprobe p is a valid probe */
 782                                 goto valid;
 783                 return NULL;
 784         }
 785 valid:
 786         return old_p;
 787 }
 788
 789 /*
 790  * Unregister a kprobe without a scheduler synchronization.
 791  */
 792 static int __kprobes __unregister_kprobe_top(struct kprobe *p)
 793 {
 794         struct kprobe *old_p, *list_p;
 795
 796         old_p = __get_valid_kprobe(p);
 797         if (old_p == NULL)
 798                 return -EINVAL;
 799
 800         if (old_p == p ||
 801             (old_p->pre_handler == aggr_pre_handler &&
 802              list_is_singular(&old_p->list))) {
 803                 /*
 804                  * Only probe on the hash list. Disarm only if kprobes are
 805                  * enabled and not gone - otherwise, the breakpoint would
 806                  * already have been removed. We save on flushing icache.
 807                  */
 808                 if (!kprobes_all_disarmed && !kprobe_disabled(old_p))
 809                         disarm_kprobe(p);
 810                 hlist_del_rcu(&old_p->hlist);
 811         } else {
 812                 if (p->break_handler && !kprobe_gone(p))
 813                         old_p->break_handler = NULL;
 814                 if (p->post_handler && !kprobe_gone(p)) {
 815                         list_for_each_entry_rcu(list_p, &old_p->list, list) {
 816                                 if ((list_p != p) && (list_p->post_handler))
 817                                         goto noclean;
 818                         }
 819                         old_p->post_handler = NULL;
 820                 }
 821 noclean:
 822                 list_del_rcu(&p->list);
 823                 if (!kprobe_disabled(old_p)) {
 824                         try_to_disable_aggr_kprobe(old_p);
 825                         if (!kprobes_all_disarmed && kprobe_disabled(old_p))
 826                                 disarm_kprobe(old_p);
 827                 }
 828         }
 829         return 0;
 830 }
 831
 832 static void __kprobes __unregister_kprobe_bottom(struct kprobe *p)
 833 {
 834         struct kprobe *old_p;
 835
 836         if (list_empty(&p->list))
 837                 arch_remove_kprobe(p);
 838         else if (list_is_singular(&p->list)) {
 839                 /* "p" is the last child of an aggr_kprobe */
 840                 old_p = list_entry(p->list.next, struct kprobe, list);
 841                 list_del(&p->list);
 842                 arch_remove_kprobe(old_p);
 843                 kfree(old_p);
 844         }
 845 }
 846
 847 int __kprobes register_kprobes(struct kprobe **kps, int num)
 848 {
 849         int i, ret = 0;
 850
 851         if (num <= 0)
 852                 return -EINVAL;
 853         for (i = 0; i < num; i++) {
 854                 ret = register_kprobe(kps[i]);
 855                 if (ret < 0) {
 856                         if (i > 0)
 857                                 unregister_kprobes(kps, i);
 858                         break;
 859                 }
 860         }
 861         return ret;
 862 }
 863 EXPORT_SYMBOL_GPL(register_kprobes);
 864
 865 void __kprobes unregister_kprobe(struct kprobe *p)
 866 {
 867         unregister_kprobes(&p, 1);
 868 }
 869 EXPORT_SYMBOL_GPL(unregister_kprobe);
 870
 871 void __kprobes unregister_kprobes(struct kprobe **kps, int num)
 872 {
 873         int i;
 874
 875         if (num <= 0)
 876                 return;
 877         mutex_lock(&kprobe_mutex);
 878         for (i = 0; i < num; i++)
 879                 if (__unregister_kprobe_top(kps[i]) < 0)
 880                         kps[i]->addr = NULL;
 881         mutex_unlock(&kprobe_mutex);
 882
 883         synchronize_sched();
 884         for (i = 0; i < num; i++)
 885                 if (kps[i]->addr)
 886                         __unregister_kprobe_bottom(kps[i]);
 887 }
 888 EXPORT_SYMBOL_GPL(unregister_kprobes);
 889
 890 static struct notifier_block kprobe_exceptions_nb = {
 891         .notifier_call = kprobe_exceptions_notify,
 892         .priority = 0x7fffffff /* we need to be notified first */
 893 };
 894
 895 unsigned long __weak arch_deref_entry_point(void *entry)
 896 {
 897         return (unsigned long)entry;
 898 }
 899
 900 int __kprobes register_jprobes(struct jprobe **jps, int num)
 901 {
 902         struct jprobe *jp;
 903         int ret = 0, i;
 904
 905         if (num <= 0)
 906                 return -EINVAL;
 907         for (i = 0; i < num; i++) {
 908                 unsigned long addr;
 909                 jp = jps[i];
 910                 addr = arch_deref_entry_point(jp->entry);
 911
 912                 if (!kernel_text_address(addr))
 913                         ret = -EINVAL;
 914                 else {
 915                         /* Todo: Verify probepoint is a function entry point */
 916                         jp->kp.pre_handler = setjmp_pre_handler;
 917                         jp->kp.break_handler = longjmp_break_handler;
 918                         ret = register_kprobe(&jp->kp);
 919                 }
 920                 if (ret < 0) {
 921                         if (i > 0)
 922                                 unregister_jprobes(jps, i);
 923                         break;
 924                 }
 925         }
 926         return ret;
 927 }
 928 EXPORT_SYMBOL_GPL(register_jprobes);
 929
 930 int __kprobes register_jprobe(struct jprobe *jp)
 931 {
 932         return register_jprobes(&jp, 1);
 933 }
 934 EXPORT_SYMBOL_GPL(register_jprobe);
 935
 936 void __kprobes unregister_jprobe(struct jprobe *jp)
 937 {
 938         unregister_jprobes(&jp, 1);
 939 }
 940 EXPORT_SYMBOL_GPL(unregister_jprobe);
 941
 942 void __kprobes unregister_jprobes(struct jprobe **jps, int num)
 943 {
 944         int i;
 945
 946         if (num <= 0)
 947                 return;
 948         mutex_lock(&kprobe_mutex);
 949         for (i = 0; i < num; i++)
 950                 if (__unregister_kprobe_top(&jps[i]->kp) < 0)
 951                         jps[i]->kp.addr = NULL;
 952         mutex_unlock(&kprobe_mutex);
 953
 954         synchronize_sched();
 955         for (i = 0; i < num; i++) {
 956                 if (jps[i]->kp.addr)
 957                         __unregister_kprobe_bottom(&jps[i]->kp);
 958         }
 959 }
 960 EXPORT_SYMBOL_GPL(unregister_jprobes);
 961
 962 #ifdef CONFIG_KRETPROBES
 963 /*
 964  * This kprobe pre_handler is registered with every kretprobe. When probe
 965  * hits it will set up the return probe.
 966  */
 967 static int __kprobes pre_handler_kretprobe(struct kprobe *p,
 968                                            struct pt_regs *regs)
 969 {
 970         struct kretprobe *rp = container_of(p, struct kretprobe, kp);
 971         unsigned long hash, flags = 0;
 972         struct kretprobe_instance *ri;
 973
 974         /*TODO: consider to only swap the RA after the last pre_handler fired */
 975         hash = hash_ptr(current, KPROBE_HASH_BITS);
 976         spin_lock_irqsave(&rp->lock, flags);
 977         if (!hlist_empty(&rp->free_instances)) {
 978                 ri = hlist_entry(rp->free_instances.first,
 979                                 struct kretprobe_instance, hlist);
 980                 hlist_del(&ri->hlist);
 981                 spin_unlock_irqrestore(&rp->lock, flags);
 982
 983                 ri->rp = rp;
 984                 ri->task = current;
 985
 986                 if (rp->entry_handler && rp->entry_handler(ri, regs))
 987                         return 0;
 988
 989                 arch_prepare_kretprobe(ri, regs);
 990
 991                 /* XXX(hch): why is there no hlist_move_head? */
 992                 INIT_HLIST_NODE(&ri->hlist);
 993                 kretprobe_table_lock(hash, &flags);
 994                 hlist_add_head(&ri->hlist, &kretprobe_inst_table[hash]);
 995                 kretprobe_table_unlock(hash, &flags);
 996         } else {
 997                 rp->nmissed++;
 998                 spin_unlock_irqrestore(&rp->lock, flags);
 999         }
1000         return 0;
1001 }
1002
1003 int __kprobes register_kretprobe(struct kretprobe *rp)
1004 {
1005         int ret = 0;
1006         struct kretprobe_instance *inst;
1007         int i;
1008         void *addr;
1009
1010         if (kretprobe_blacklist_size) {
1011                 addr = kprobe_addr(&rp->kp);
1012                 if (!addr)
1013                         return -EINVAL;
1014
1015                 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
1016                         if (kretprobe_blacklist[i].addr == addr)
1017                                 return -EINVAL;
1018                 }
1019         }
1020
1021         rp->kp.pre_handler = pre_handler_kretprobe;
1022         rp->kp.post_handler = NULL;
1023         rp->kp.fault_handler = NULL;
1024         rp->kp.break_handler = NULL;
1025
1026         /* Pre-allocate memory for max kretprobe instances */
1027         if (rp->maxactive <= 0) {
1028 #ifdef CONFIG_PREEMPT
1029                 rp->maxactive = max(10, 2 * NR_CPUS);
1030 #else
1031                 rp->maxactive = NR_CPUS;
1032 #endif
1033         }
1034         spin_lock_init(&rp->lock);
1035         INIT_HLIST_HEAD(&rp->free_instances);
1036         for (i = 0; i < rp->maxactive; i++) {
1037                 inst = kmalloc(sizeof(struct kretprobe_instance) +
1038                                rp->data_size, GFP_KERNEL);
1039                 if (inst == NULL) {
1040                         free_rp_inst(rp);
1041                         return -ENOMEM;
1042                 }
1043                 INIT_HLIST_NODE(&inst->hlist);
1044                 hlist_add_head(&inst->hlist, &rp->free_instances);
1045         }
1046
1047         rp->nmissed = 0;
1048         /* Establish function entry probe point */
1049         ret = register_kprobe(&rp->kp);
1050         if (ret != 0)
1051                 free_rp_inst(rp);
1052         return ret;
1053 }
1054 EXPORT_SYMBOL_GPL(register_kretprobe);
1055
1056 int __kprobes register_kretprobes(struct kretprobe **rps, int num)
1057 {
1058         int ret = 0, i;
1059
1060         if (num <= 0)
1061                 return -EINVAL;
1062         for (i = 0; i < num; i++) {
1063                 ret = register_kretprobe(rps[i]);
1064                 if (ret < 0) {
1065                         if (i > 0)
1066                                 unregister_kretprobes(rps, i);
1067                         break;
1068                 }
1069         }
1070         return ret;
1071 }
1072 EXPORT_SYMBOL_GPL(register_kretprobes);
1073
1074 void __kprobes unregister_kretprobe(struct kretprobe *rp)
1075 {
1076         unregister_kretprobes(&rp, 1);
1077 }
1078 EXPORT_SYMBOL_GPL(unregister_kretprobe);
1079
1080 void __kprobes unregister_kretprobes(struct kretprobe **rps, int num)
1081 {
1082         int i;
1083
1084         if (num <= 0)
1085                 return;
1086         mutex_lock(&kprobe_mutex);
1087         for (i = 0; i < num; i++)
1088                 if (__unregister_kprobe_top(&rps[i]->kp) < 0)
1089                         rps[i]->kp.addr = NULL;
1090         mutex_unlock(&kprobe_mutex);
1091
1092         synchronize_sched();
1093         for (i = 0; i < num; i++) {
1094                 if (rps[i]->kp.addr) {
1095                         __unregister_kprobe_bottom(&rps[i]->kp);
1096                         cleanup_rp_inst(rps[i]);
1097                 }
1098         }
1099 }
1100 EXPORT_SYMBOL_GPL(unregister_kretprobes);
1101
1102 #else /* CONFIG_KRETPROBES */
1103 int __kprobes register_kretprobe(struct kretprobe *rp)
1104 {
1105         return -ENOSYS;
1106 }
1107 EXPORT_SYMBOL_GPL(register_kretprobe);
1108
1109 int __kprobes register_kretprobes(struct kretprobe **rps, int num)
1110 {
1111         return -ENOSYS;
1112 }
1113 EXPORT_SYMBOL_GPL(register_kretprobes);
1114
1115 void __kprobes unregister_kretprobe(struct kretprobe *rp)
1116 {
1117 }
1118 EXPORT_SYMBOL_GPL(unregister_kretprobe);
1119
1120 void __kprobes unregister_kretprobes(struct kretprobe **rps, int num)
1121 {
1122 }
1123 EXPORT_SYMBOL_GPL(unregister_kretprobes);
1124
1125 static int __kprobes pre_handler_kretprobe(struct kprobe *p,
1126                                            struct pt_regs *regs)
1127 {
1128         return 0;
1129 }
1130
1131 #endif /* CONFIG_KRETPROBES */
1132
1133 /* Set the kprobe gone and remove its instruction buffer. */
1134 static void __kprobes kill_kprobe(struct kprobe *p)
1135 {
1136         struct kprobe *kp;
1137
1138         p->flags |= KPROBE_FLAG_GONE;
1139         if (p->pre_handler == aggr_pre_handler) {
1140                 /*
1141                  * If this is an aggr_kprobe, we have to list all the
1142                  * chained probes and mark them GONE.
1143                  */
1144                 list_for_each_entry_rcu(kp, &p->list, list)
1145                         kp->flags |= KPROBE_FLAG_GONE;
1146                 p->post_handler = NULL;
1147                 p->break_handler = NULL;
1148         }
1149         /*
1150          * Here, we can remove insn_slot safely, because no thread calls
1151          * the original probed function (which will be freed soon) any more.
1152          */
1153         arch_remove_kprobe(p);
1154 }
1155
1156 /* Module notifier call back, checking kprobes on the module */
1157 static int __kprobes kprobes_module_callback(struct notifier_block *nb,
1158                                              unsigned long val, void *data)
1159 {
1160         struct module *mod = data;
1161         struct hlist_head *head;
1162         struct hlist_node *node;
1163         struct kprobe *p;
1164         unsigned int i;
1165         int checkcore = (val == MODULE_STATE_GOING);
1166
1167         if (val != MODULE_STATE_GOING && val != MODULE_STATE_LIVE)
1168                 return NOTIFY_DONE;
1169
1170         /*
1171          * When MODULE_STATE_GOING was notified, both of module .text and
1172          * .init.text sections would be freed. When MODULE_STATE_LIVE was
1173          * notified, only .init.text section would be freed. We need to
1174          * disable kprobes which have been inserted in the sections.
1175          */
1176         mutex_lock(&kprobe_mutex);
1177         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
1178                 head = &kprobe_table[i];
1179                 hlist_for_each_entry_rcu(p, node, head, hlist)
1180                         if (within_module_init((unsigned long)p->addr, mod) ||
1181                             (checkcore &&
1182                              within_module_core((unsigned long)p->addr, mod))) {
1183                                 /*
1184                                  * The vaddr this probe is installed will soon
1185                                  * be vfreed buy not synced to disk. Hence,
1186                                  * disarming the breakpoint isn't needed.
1187                                  */
1188                                 kill_kprobe(p);
1189                         }
1190         }
1191         mutex_unlock(&kprobe_mutex);
1192         return NOTIFY_DONE;
1193 }
1194
1195 static struct notifier_block kprobe_module_nb = {
1196         .notifier_call = kprobes_module_callback,
1197         .priority = 0
1198 };
1199
1200 static int __init init_kprobes(void)
1201 {
1202         int i, err = 0;
1203         unsigned long offset = 0, size = 0;
1204         char *modname, namebuf[128];
1205         const char *symbol_name;
1206         void *addr;
1207         struct kprobe_blackpoint *kb;
1208
1209         /* FIXME allocate the probe table, currently defined statically */
1210         /* initialize all list heads */
1211         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
1212                 INIT_HLIST_HEAD(&kprobe_table[i]);
1213                 INIT_HLIST_HEAD(&kretprobe_inst_table[i]);
1214                 spin_lock_init(&(kretprobe_table_locks[i].lock));
1215         }
1216
1217         /*
1218          * Lookup and populate the kprobe_blacklist.
1219          *
1220          * Unlike the kretprobe blacklist, we'll need to determine
1221          * the range of addresses that belong to the said functions,
1222          * since a kprobe need not necessarily be at the beginning
1223          * of a function.
1224          */
1225         for (kb = kprobe_blacklist; kb->name != NULL; kb++) {
1226                 kprobe_lookup_name(kb->name, addr);
1227                 if (!addr)
1228                         continue;
1229
1230                 kb->start_addr = (unsigned long)addr;
1231                 symbol_name = kallsyms_lookup(kb->start_addr,
1232                                 &size, &offset, &modname, namebuf);
1233                 if (!symbol_name)
1234                         kb->range = 0;
1235                 else
1236                         kb->range = size;
1237         }
1238
1239         if (kretprobe_blacklist_size) {
1240                 /* lookup the function address from its name */
1241                 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
1242                         kprobe_lookup_name(kretprobe_blacklist[i].name,
1243                                            kretprobe_blacklist[i].addr);
1244                         if (!kretprobe_blacklist[i].addr)
1245                                 printk("kretprobe: lookup failed: %s\n",
1246                                        kretprobe_blacklist[i].name);
1247                 }
1248         }
1249
1250         /* By default, kprobes are armed */
1251         kprobes_all_disarmed = false;
1252
1253         err = arch_init_kprobes();
1254         if (!err)
1255                 err = register_die_notifier(&kprobe_exceptions_nb);
1256         if (!err)
1257                 err = register_module_notifier(&kprobe_module_nb);
1258
1259         kprobes_initialized = (err == 0);
1260
1261         if (!err)
1262                 init_test_probes();
1263         return err;
1264 }
1265
1266 #ifdef CONFIG_DEBUG_FS
1267 static void __kprobes report_probe(struct seq_file *pi, struct kprobe *p,
1268                 const char *sym, int offset,char *modname)
1269 {
1270         char *kprobe_type;
1271
1272         if (p->pre_handler == pre_handler_kretprobe)
1273                 kprobe_type = "r";
1274         else if (p->pre_handler == setjmp_pre_handler)
1275                 kprobe_type = "j";
1276         else
1277                 kprobe_type = "k";
1278         if (sym)
1279                 seq_printf(pi, "%p  %s  %s+0x%x  %s %s%s\n",
1280                         p->addr, kprobe_type, sym, offset,
1281                         (modname ? modname : " "),
1282                         (kprobe_gone(p) ? "[GONE]" : ""),
1283                         ((kprobe_disabled(p) && !kprobe_gone(p)) ?
1284                          "[DISABLED]" : ""));
1285         else
1286                 seq_printf(pi, "%p  %s  %p %s%s\n",
1287                         p->addr, kprobe_type, p->addr,
1288                         (kprobe_gone(p) ? "[GONE]" : ""),
1289                         ((kprobe_disabled(p) && !kprobe_gone(p)) ?
1290                          "[DISABLED]" : ""));
1291 }
1292
1293 static void __kprobes *kprobe_seq_start(struct seq_file *f, loff_t *pos)
1294 {
1295         return (*pos < KPROBE_TABLE_SIZE) ? pos : NULL;
1296 }
1297
1298 static void __kprobes *kprobe_seq_next(struct seq_file *f, void *v, loff_t *pos)
1299 {
1300         (*pos)++;
1301         if (*pos >= KPROBE_TABLE_SIZE)
1302                 return NULL;
1303         return pos;
1304 }
1305
1306 static void __kprobes kprobe_seq_stop(struct seq_file *f, void *v)
1307 {
1308         /* Nothing to do */
1309 }
1310
1311 static int __kprobes show_kprobe_addr(struct seq_file *pi, void *v)
1312 {
1313         struct hlist_head *head;
1314         struct hlist_node *node;
1315         struct kprobe *p, *kp;
1316         const char *sym = NULL;
1317         unsigned int i = *(loff_t *) v;
1318         unsigned long offset = 0;
1319         char *modname, namebuf[128];
1320
1321         head = &kprobe_table[i];
1322         preempt_disable();
1323         hlist_for_each_entry_rcu(p, node, head, hlist) {
1324                 sym = kallsyms_lookup((unsigned long)p->addr, NULL,
1325                                         &offset, &modname, namebuf);
1326                 if (p->pre_handler == aggr_pre_handler) {
1327                         list_for_each_entry_rcu(kp, &p->list, list)
1328                                 report_probe(pi, kp, sym, offset, modname);
1329                 } else
1330                         report_probe(pi, p, sym, offset, modname);
1331         }
1332         preempt_enable();
1333         return 0;
1334 }
1335
1336 static struct seq_operations kprobes_seq_ops = {
1337         .start = kprobe_seq_start,
1338         .next  = kprobe_seq_next,
1339         .stop  = kprobe_seq_stop,
1340         .show  = show_kprobe_addr
1341 };
1342
1343 static int __kprobes kprobes_open(struct inode *inode, struct file *filp)
1344 {
1345         return seq_open(filp, &kprobes_seq_ops);
1346 }
1347
1348 static struct file_operations debugfs_kprobes_operations = {
1349         .open           = kprobes_open,
1350         .read           = seq_read,
1351         .llseek         = seq_lseek,
1352         .release        = seq_release,
1353 };
1354
1355 /* Disable one kprobe */
1356 int __kprobes disable_kprobe(struct kprobe *kp)
1357 {
1358         int ret = 0;
1359         struct kprobe *p;
1360
1361         mutex_lock(&kprobe_mutex);
1362
1363         /* Check whether specified probe is valid. */
1364         p = __get_valid_kprobe(kp);
1365         if (unlikely(p == NULL)) {
1366                 ret = -EINVAL;
1367                 goto out;
1368         }
1369
1370         /* If the probe is already disabled (or gone), just return */
1371         if (kprobe_disabled(kp))
1372                 goto out;
1373
1374         kp->flags |= KPROBE_FLAG_DISABLED;
1375         if (p != kp)
1376                 /* When kp != p, p is always enabled. */
1377                 try_to_disable_aggr_kprobe(p);
1378
1379         if (!kprobes_all_disarmed && kprobe_disabled(p))
1380                 disarm_kprobe(p);
1381 out:
1382         mutex_unlock(&kprobe_mutex);
1383         return ret;
1384 }
1385 EXPORT_SYMBOL_GPL(disable_kprobe);
1386
1387 /* Enable one kprobe */
1388 int __kprobes enable_kprobe(struct kprobe *kp)
1389 {
1390         int ret = 0;
1391         struct kprobe *p;
1392
1393         mutex_lock(&kprobe_mutex);
1394
1395         /* Check whether specified probe is valid. */
1396         p = __get_valid_kprobe(kp);
1397         if (unlikely(p == NULL)) {
1398                 ret = -EINVAL;
1399                 goto out;
1400         }
1401
1402         if (kprobe_gone(kp)) {
1403                 /* This kprobe has gone, we couldn't enable it. */
1404                 ret = -EINVAL;
1405                 goto out;
1406         }
1407
1408         if (!kprobes_all_disarmed && kprobe_disabled(p))
1409                 arm_kprobe(p);
1410
1411         p->flags &= ~KPROBE_FLAG_DISABLED;
1412         if (p != kp)
1413                 kp->flags &= ~KPROBE_FLAG_DISABLED;
1414 out:
1415         mutex_unlock(&kprobe_mutex);
1416         return ret;
1417 }
1418 EXPORT_SYMBOL_GPL(enable_kprobe);
1419
1420 static void __kprobes arm_all_kprobes(void)
1421 {
1422         struct hlist_head *head;
1423         struct hlist_node *node;
1424         struct kprobe *p;
1425         unsigned int i;
1426
1427         mutex_lock(&kprobe_mutex);
1428
1429         /* If kprobes are armed, just return */
1430         if (!kprobes_all_disarmed)
1431                 goto already_enabled;
1432
1433         mutex_lock(&text_mutex);
1434         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
1435                 head = &kprobe_table[i];
1436                 hlist_for_each_entry_rcu(p, node, head, hlist)
1437                         if (!kprobe_disabled(p))
1438                                 arch_arm_kprobe(p);
1439         }
1440         mutex_unlock(&text_mutex);
1441
1442         kprobes_all_disarmed = false;
1443         printk(KERN_INFO "Kprobes globally enabled\n");
1444
1445 already_enabled:
1446         mutex_unlock(&kprobe_mutex);
1447         return;
1448 }
1449
1450 static void __kprobes disarm_all_kprobes(void)
1451 {
1452         struct hlist_head *head;
1453         struct hlist_node *node;
1454         struct kprobe *p;
1455         unsigned int i;
1456
1457         mutex_lock(&kprobe_mutex);
1458
1459         /* If kprobes are already disarmed, just return */
1460         if (kprobes_all_disarmed)
1461                 goto already_disabled;
1462
1463         kprobes_all_disarmed = true;
1464         printk(KERN_INFO "Kprobes globally disabled\n");
1465         mutex_lock(&text_mutex);
1466         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
1467                 head = &kprobe_table[i];
1468                 hlist_for_each_entry_rcu(p, node, head, hlist) {
1469                         if (!arch_trampoline_kprobe(p) && !kprobe_disabled(p))
1470                                 arch_disarm_kprobe(p);
1471                 }
1472         }
1473
1474         mutex_unlock(&text_mutex);
1475         mutex_unlock(&kprobe_mutex);
1476         /* Allow all currently running kprobes to complete */
1477         synchronize_sched();
1478         return;
1479
1480 already_disabled:
1481         mutex_unlock(&kprobe_mutex);
1482         return;
1483 }
1484
1485 /*
1486  * XXX: The debugfs bool file interface doesn't allow for callbacks
1487  * when the bool state is switched. We can reuse that facility when
1488  * available
1489  */
1490 static ssize_t read_enabled_file_bool(struct file *file,
1491                char __user *user_buf, size_t count, loff_t *ppos)
1492 {
1493         char buf[3];
1494
1495         if (!kprobes_all_disarmed)
1496                 buf[0] = '1';
1497         else
1498                 buf[0] = '0';
1499         buf[1] = '\n';
1500         buf[2] = 0x00;
1501         return simple_read_from_buffer(user_buf, count, ppos, buf, 2);
1502 }
1503
1504 static ssize_t write_enabled_file_bool(struct file *file,
1505                const char __user *user_buf, size_t count, loff_t *ppos)
1506 {
1507         char buf[32];
1508         int buf_size;
1509
1510         buf_size = min(count, (sizeof(buf)-1));
1511         if (copy_from_user(buf, user_buf, buf_size))
1512                 return -EFAULT;
1513
1514         switch (buf[0]) {
1515         case 'y':
1516         case 'Y':
1517         case '1':
1518                 arm_all_kprobes();
1519                 break;
1520         case 'n':
1521         case 'N':
1522         case '0':
1523                 disarm_all_kprobes();
1524                 break;
1525         }
1526
1527         return count;
1528 }
1529
1530 static struct file_operations fops_kp = {
1531         .read =         read_enabled_file_bool,
1532         .write =        write_enabled_file_bool,
1533 };
1534
1535 static int __kprobes debugfs_kprobe_init(void)
1536 {
1537         struct dentry *dir, *file;
1538         unsigned int value = 1;
1539
1540         dir = debugfs_create_dir("kprobes", NULL);
1541         if (!dir)
1542                 return -ENOMEM;
1543
1544         file = debugfs_create_file("list", 0444, dir, NULL,
1545                                 &debugfs_kprobes_operations);
1546         if (!file) {
1547                 debugfs_remove(dir);
1548                 return -ENOMEM;
1549         }
1550
1551         file = debugfs_create_file("enabled", 0600, dir,
1552                                         &value, &fops_kp);
1553         if (!file) {
1554                 debugfs_remove(dir);
1555                 return -ENOMEM;
1556         }
1557
1558         return 0;
1559 }
1560
1561 late_initcall(debugfs_kprobe_init);
1562 #endif /* CONFIG_DEBUG_FS */
1563
1564 module_init(init_kprobes);
1565
1566 /* defined in arch/.../kernel/kprobes.c */
1567 EXPORT_SYMBOL_GPL(jprobe_return);