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         {"native_get_debugreg",},
  94         {"irq_entries_start",},
  95         {"common_interrupt",},
  96         {NULL}    /* Terminator */
  97 };
  98
  99 #ifdef __ARCH_WANT_KPROBES_INSN_SLOT
 100 /*
 101  * kprobe->ainsn.insn points to the copy of the instruction to be
 102  * single-stepped. x86_64, POWER4 and above have no-exec support and
 103  * stepping on the instruction on a vmalloced/kmalloced/data page
 104  * is a recipe for disaster
 105  */
 106 #define INSNS_PER_PAGE  (PAGE_SIZE/(MAX_INSN_SIZE * sizeof(kprobe_opcode_t)))
 107
 108 struct kprobe_insn_page {
 109         struct list_head list;
 110         kprobe_opcode_t *insns;         /* Page of instruction slots */
 111         char slot_used[INSNS_PER_PAGE];
 112         int nused;
 113         int ngarbage;
 114 };
 115
 116 enum kprobe_slot_state {
 117         SLOT_CLEAN = 0,
 118         SLOT_DIRTY = 1,
 119         SLOT_USED = 2,
 120 };
 121
 122 static DEFINE_MUTEX(kprobe_insn_mutex); /* Protects kprobe_insn_pages */
 123 static LIST_HEAD(kprobe_insn_pages);
 124 static int kprobe_garbage_slots;
 125 static int collect_garbage_slots(void);
 126
 127 static int __kprobes check_safety(void)
 128 {
 129         int ret = 0;
 130 #if defined(CONFIG_PREEMPT) && defined(CONFIG_FREEZER)
 131         ret = freeze_processes();
 132         if (ret == 0) {
 133                 struct task_struct *p, *q;
 134                 do_each_thread(p, q) {
 135                         if (p != current && p->state == TASK_RUNNING &&
 136                             p->pid != 0) {
 137                                 printk("Check failed: %s is running\n",p->comm);
 138                                 ret = -1;
 139                                 goto loop_end;
 140                         }
 141                 } while_each_thread(p, q);
 142         }
 143 loop_end:
 144         thaw_processes();
 145 #else
 146         synchronize_sched();
 147 #endif
 148         return ret;
 149 }
 150
 151 /**
 152  * __get_insn_slot() - Find a slot on an executable page for an instruction.
 153  * We allocate an executable page if there's no room on existing ones.
 154  */
 155 static kprobe_opcode_t __kprobes *__get_insn_slot(void)
 156 {
 157         struct kprobe_insn_page *kip;
 158
 159  retry:
 160         list_for_each_entry(kip, &kprobe_insn_pages, list) {
 161                 if (kip->nused < INSNS_PER_PAGE) {
 162                         int i;
 163                         for (i = 0; i < INSNS_PER_PAGE; i++) {
 164                                 if (kip->slot_used[i] == SLOT_CLEAN) {
 165                                         kip->slot_used[i] = SLOT_USED;
 166                                         kip->nused++;
 167                                         return kip->insns + (i * MAX_INSN_SIZE);
 168                                 }
 169                         }
 170                         /* Surprise!  No unused slots.  Fix kip->nused. */
 171                         kip->nused = INSNS_PER_PAGE;
 172                 }
 173         }
 174
 175         /* If there are any garbage slots, collect it and try again. */
 176         if (kprobe_garbage_slots && collect_garbage_slots() == 0) {
 177                 goto retry;
 178         }
 179         /* All out of space.  Need to allocate a new page. Use slot 0. */
 180         kip = kmalloc(sizeof(struct kprobe_insn_page), GFP_KERNEL);
 181         if (!kip)
 182                 return NULL;
 183
 184         /*
 185          * Use module_alloc so this page is within +/- 2GB of where the
 186          * kernel image and loaded module images reside. This is required
 187          * so x86_64 can correctly handle the %rip-relative fixups.
 188          */
 189         kip->insns = module_alloc(PAGE_SIZE);
 190         if (!kip->insns) {
 191                 kfree(kip);
 192                 return NULL;
 193         }
 194         INIT_LIST_HEAD(&kip->list);
 195         list_add(&kip->list, &kprobe_insn_pages);
 196         memset(kip->slot_used, SLOT_CLEAN, INSNS_PER_PAGE);
 197         kip->slot_used[0] = SLOT_USED;
 198         kip->nused = 1;
 199         kip->ngarbage = 0;
 200         return kip->insns;
 201 }
 202
 203 kprobe_opcode_t __kprobes *get_insn_slot(void)
 204 {
 205         kprobe_opcode_t *ret;
 206         mutex_lock(&kprobe_insn_mutex);
 207         ret = __get_insn_slot();
 208         mutex_unlock(&kprobe_insn_mutex);
 209         return ret;
 210 }
 211
 212 /* Return 1 if all garbages are collected, otherwise 0. */
 213 static int __kprobes collect_one_slot(struct kprobe_insn_page *kip, int idx)
 214 {
 215         kip->slot_used[idx] = SLOT_CLEAN;
 216         kip->nused--;
 217         if (kip->nused == 0) {
 218                 /*
 219                  * Page is no longer in use.  Free it unless
 220                  * it's the last one.  We keep the last one
 221                  * so as not to have to set it up again the
 222                  * next time somebody inserts a probe.
 223                  */
 224                 if (!list_is_singular(&kprobe_insn_pages)) {
 225                         list_del(&kip->list);
 226                         module_free(NULL, kip->insns);
 227                         kfree(kip);
 228                 }
 229                 return 1;
 230         }
 231         return 0;
 232 }
 233
 234 static int __kprobes collect_garbage_slots(void)
 235 {
 236         struct kprobe_insn_page *kip, *next;
 237
 238         /* Ensure no-one is preepmted on the garbages */
 239         if (check_safety())
 240                 return -EAGAIN;
 241
 242         list_for_each_entry_safe(kip, next, &kprobe_insn_pages, list) {
 243                 int i;
 244                 if (kip->ngarbage == 0)
 245                         continue;
 246                 kip->ngarbage = 0;      /* we will collect all garbages */
 247                 for (i = 0; i < INSNS_PER_PAGE; i++) {
 248                         if (kip->slot_used[i] == SLOT_DIRTY &&
 249                             collect_one_slot(kip, i))
 250                                 break;
 251                 }
 252         }
 253         kprobe_garbage_slots = 0;
 254         return 0;
 255 }
 256
 257 void __kprobes free_insn_slot(kprobe_opcode_t * slot, int dirty)
 258 {
 259         struct kprobe_insn_page *kip;
 260
 261         mutex_lock(&kprobe_insn_mutex);
 262         list_for_each_entry(kip, &kprobe_insn_pages, list) {
 263                 if (kip->insns <= slot &&
 264                     slot < kip->insns + (INSNS_PER_PAGE * MAX_INSN_SIZE)) {
 265                         int i = (slot - kip->insns) / MAX_INSN_SIZE;
 266                         if (dirty) {
 267                                 kip->slot_used[i] = SLOT_DIRTY;
 268                                 kip->ngarbage++;
 269                         } else
 270                                 collect_one_slot(kip, i);
 271                         break;
 272                 }
 273         }
 274
 275         if (dirty && ++kprobe_garbage_slots > INSNS_PER_PAGE)
 276                 collect_garbage_slots();
 277
 278         mutex_unlock(&kprobe_insn_mutex);
 279 }
 280 #endif
 281
 282 /* We have preemption disabled.. so it is safe to use __ versions */
 283 static inline void set_kprobe_instance(struct kprobe *kp)
 284 {
 285         __get_cpu_var(kprobe_instance) = kp;
 286 }
 287
 288 static inline void reset_kprobe_instance(void)
 289 {
 290         __get_cpu_var(kprobe_instance) = NULL;
 291 }
 292
 293 /*
 294  * This routine is called either:
 295  *      - under the kprobe_mutex - during kprobe_[un]register()
 296  *                              OR
 297  *      - with preemption disabled - from arch/xxx/kernel/kprobes.c
 298  */
 299 struct kprobe __kprobes *get_kprobe(void *addr)
 300 {
 301         struct hlist_head *head;
 302         struct hlist_node *node;
 303         struct kprobe *p;
 304
 305         head = &kprobe_table[hash_ptr(addr, KPROBE_HASH_BITS)];
 306         hlist_for_each_entry_rcu(p, node, head, hlist) {
 307                 if (p->addr == addr)
 308                         return p;
 309         }
 310         return NULL;
 311 }
 312
 313 /* Arm a kprobe with text_mutex */
 314 static void __kprobes arm_kprobe(struct kprobe *kp)
 315 {
 316         mutex_lock(&text_mutex);
 317         arch_arm_kprobe(kp);
 318         mutex_unlock(&text_mutex);
 319 }
 320
 321 /* Disarm a kprobe with text_mutex */
 322 static void __kprobes disarm_kprobe(struct kprobe *kp)
 323 {
 324         mutex_lock(&text_mutex);
 325         arch_disarm_kprobe(kp);
 326         mutex_unlock(&text_mutex);
 327 }
 328
 329 /*
 330  * Aggregate handlers for multiple kprobes support - these handlers
 331  * take care of invoking the individual kprobe handlers on p->list
 332  */
 333 static int __kprobes aggr_pre_handler(struct kprobe *p, struct pt_regs *regs)
 334 {
 335         struct kprobe *kp;
 336
 337         list_for_each_entry_rcu(kp, &p->list, list) {
 338                 if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
 339                         set_kprobe_instance(kp);
 340                         if (kp->pre_handler(kp, regs))
 341                                 return 1;
 342                 }
 343                 reset_kprobe_instance();
 344         }
 345         return 0;
 346 }
 347
 348 static void __kprobes aggr_post_handler(struct kprobe *p, struct pt_regs *regs,
 349                                         unsigned long flags)
 350 {
 351         struct kprobe *kp;
 352
 353         list_for_each_entry_rcu(kp, &p->list, list) {
 354                 if (kp->post_handler && likely(!kprobe_disabled(kp))) {
 355                         set_kprobe_instance(kp);
 356                         kp->post_handler(kp, regs, flags);
 357                         reset_kprobe_instance();
 358                 }
 359         }
 360 }
 361
 362 static int __kprobes aggr_fault_handler(struct kprobe *p, struct pt_regs *regs,
 363                                         int trapnr)
 364 {
 365         struct kprobe *cur = __get_cpu_var(kprobe_instance);
 366
 367         /*
 368          * if we faulted "during" the execution of a user specified
 369          * probe handler, invoke just that probe's fault handler
 370          */
 371         if (cur && cur->fault_handler) {
 372                 if (cur->fault_handler(cur, regs, trapnr))
 373                         return 1;
 374         }
 375         return 0;
 376 }
 377
 378 static int __kprobes aggr_break_handler(struct kprobe *p, struct pt_regs *regs)
 379 {
 380         struct kprobe *cur = __get_cpu_var(kprobe_instance);
 381         int ret = 0;
 382
 383         if (cur && cur->break_handler) {
 384                 if (cur->break_handler(cur, regs))
 385                         ret = 1;
 386         }
 387         reset_kprobe_instance();
 388         return ret;
 389 }
 390
 391 /* Walks the list and increments nmissed count for multiprobe case */
 392 void __kprobes kprobes_inc_nmissed_count(struct kprobe *p)
 393 {
 394         struct kprobe *kp;
 395         if (p->pre_handler != aggr_pre_handler) {
 396                 p->nmissed++;
 397         } else {
 398                 list_for_each_entry_rcu(kp, &p->list, list)
 399                         kp->nmissed++;
 400         }
 401         return;
 402 }
 403
 404 void __kprobes recycle_rp_inst(struct kretprobe_instance *ri,
 405                                 struct hlist_head *head)
 406 {
 407         struct kretprobe *rp = ri->rp;
 408
 409         /* remove rp inst off the rprobe_inst_table */
 410         hlist_del(&ri->hlist);
 411         INIT_HLIST_NODE(&ri->hlist);
 412         if (likely(rp)) {
 413                 spin_lock(&rp->lock);
 414                 hlist_add_head(&ri->hlist, &rp->free_instances);
 415                 spin_unlock(&rp->lock);
 416         } else
 417                 /* Unregistering */
 418                 hlist_add_head(&ri->hlist, head);
 419 }
 420
 421 void __kprobes kretprobe_hash_lock(struct task_struct *tsk,
 422                          struct hlist_head **head, unsigned long *flags)
 423 {
 424         unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
 425         spinlock_t *hlist_lock;
 426
 427         *head = &kretprobe_inst_table[hash];
 428         hlist_lock = kretprobe_table_lock_ptr(hash);
 429         spin_lock_irqsave(hlist_lock, *flags);
 430 }
 431
 432 static void __kprobes kretprobe_table_lock(unsigned long hash,
 433         unsigned long *flags)
 434 {
 435         spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
 436         spin_lock_irqsave(hlist_lock, *flags);
 437 }
 438
 439 void __kprobes kretprobe_hash_unlock(struct task_struct *tsk,
 440         unsigned long *flags)
 441 {
 442         unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
 443         spinlock_t *hlist_lock;
 444
 445         hlist_lock = kretprobe_table_lock_ptr(hash);
 446         spin_unlock_irqrestore(hlist_lock, *flags);
 447 }
 448
 449 void __kprobes kretprobe_table_unlock(unsigned long hash, unsigned long *flags)
 450 {
 451         spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
 452         spin_unlock_irqrestore(hlist_lock, *flags);
 453 }
 454
 455 /*
 456  * This function is called from finish_task_switch when task tk becomes dead,
 457  * so that we can recycle any function-return probe instances associated
 458  * with this task. These left over instances represent probed functions
 459  * that have been called but will never return.
 460  */
 461 void __kprobes kprobe_flush_task(struct task_struct *tk)
 462 {
 463         struct kretprobe_instance *ri;
 464         struct hlist_head *head, empty_rp;
 465         struct hlist_node *node, *tmp;
 466         unsigned long hash, flags = 0;
 467
 468         if (unlikely(!kprobes_initialized))
 469                 /* Early boot.  kretprobe_table_locks not yet initialized. */
 470                 return;
 471
 472         hash = hash_ptr(tk, KPROBE_HASH_BITS);
 473         head = &kretprobe_inst_table[hash];
 474         kretprobe_table_lock(hash, &flags);
 475         hlist_for_each_entry_safe(ri, node, tmp, head, hlist) {
 476                 if (ri->task == tk)
 477                         recycle_rp_inst(ri, &empty_rp);
 478         }
 479         kretprobe_table_unlock(hash, &flags);
 480         INIT_HLIST_HEAD(&empty_rp);
 481         hlist_for_each_entry_safe(ri, node, tmp, &empty_rp, hlist) {
 482                 hlist_del(&ri->hlist);
 483                 kfree(ri);
 484         }
 485 }
 486
 487 static inline void free_rp_inst(struct kretprobe *rp)
 488 {
 489         struct kretprobe_instance *ri;
 490         struct hlist_node *pos, *next;
 491
 492         hlist_for_each_entry_safe(ri, pos, next, &rp->free_instances, hlist) {
 493                 hlist_del(&ri->hlist);
 494                 kfree(ri);
 495         }
 496 }
 497
 498 static void __kprobes cleanup_rp_inst(struct kretprobe *rp)
 499 {
 500         unsigned long flags, hash;
 501         struct kretprobe_instance *ri;
 502         struct hlist_node *pos, *next;
 503         struct hlist_head *head;
 504
 505         /* No race here */
 506         for (hash = 0; hash < KPROBE_TABLE_SIZE; hash++) {
 507                 kretprobe_table_lock(hash, &flags);
 508                 head = &kretprobe_inst_table[hash];
 509                 hlist_for_each_entry_safe(ri, pos, next, head, hlist) {
 510                         if (ri->rp == rp)
 511                                 ri->rp = NULL;
 512                 }
 513                 kretprobe_table_unlock(hash, &flags);
 514         }
 515         free_rp_inst(rp);
 516 }
 517
 518 /*
 519  * Keep all fields in the kprobe consistent
 520  */
 521 static inline void copy_kprobe(struct kprobe *old_p, struct kprobe *p)
 522 {
 523         memcpy(&p->opcode, &old_p->opcode, sizeof(kprobe_opcode_t));
 524         memcpy(&p->ainsn, &old_p->ainsn, sizeof(struct arch_specific_insn));
 525 }
 526
 527 /*
 528 * Add the new probe to ap->list. Fail if this is the
 529 * second jprobe at the address - two jprobes can't coexist
 530 */
 531 static int __kprobes add_new_kprobe(struct kprobe *ap, struct kprobe *p)
 532 {
 533         BUG_ON(kprobe_gone(ap) || kprobe_gone(p));
 534         if (p->break_handler) {
 535                 if (ap->break_handler)
 536                         return -EEXIST;
 537                 list_add_tail_rcu(&p->list, &ap->list);
 538                 ap->break_handler = aggr_break_handler;
 539         } else
 540                 list_add_rcu(&p->list, &ap->list);
 541         if (p->post_handler && !ap->post_handler)
 542                 ap->post_handler = aggr_post_handler;
 543
 544         if (kprobe_disabled(ap) && !kprobe_disabled(p)) {
 545                 ap->flags &= ~KPROBE_FLAG_DISABLED;
 546                 if (!kprobes_all_disarmed)
 547                         /* Arm the breakpoint again. */
 548                         arm_kprobe(ap);
 549         }
 550         return 0;
 551 }
 552
 553 /*
 554  * Fill in the required fields of the "manager kprobe". Replace the
 555  * earlier kprobe in the hlist with the manager kprobe
 556  */
 557 static inline void add_aggr_kprobe(struct kprobe *ap, struct kprobe *p)
 558 {
 559         copy_kprobe(p, ap);
 560         flush_insn_slot(ap);
 561         ap->addr = p->addr;
 562         ap->flags = p->flags;
 563         ap->pre_handler = aggr_pre_handler;
 564         ap->fault_handler = aggr_fault_handler;
 565         /* We don't care the kprobe which has gone. */
 566         if (p->post_handler && !kprobe_gone(p))
 567                 ap->post_handler = aggr_post_handler;
 568         if (p->break_handler && !kprobe_gone(p))
 569                 ap->break_handler = aggr_break_handler;
 570
 571         INIT_LIST_HEAD(&ap->list);
 572         list_add_rcu(&p->list, &ap->list);
 573
 574         hlist_replace_rcu(&p->hlist, &ap->hlist);
 575 }
 576
 577 /*
 578  * This is the second or subsequent kprobe at the address - handle
 579  * the intricacies
 580  */
 581 static int __kprobes register_aggr_kprobe(struct kprobe *old_p,
 582                                           struct kprobe *p)
 583 {
 584         int ret = 0;
 585         struct kprobe *ap = old_p;
 586
 587         if (old_p->pre_handler != aggr_pre_handler) {
 588                 /* If old_p is not an aggr_probe, create new aggr_kprobe. */
 589                 ap = kzalloc(sizeof(struct kprobe), GFP_KERNEL);
 590                 if (!ap)
 591                         return -ENOMEM;
 592                 add_aggr_kprobe(ap, old_p);
 593         }
 594
 595         if (kprobe_gone(ap)) {
 596                 /*
 597                  * Attempting to insert new probe at the same location that
 598                  * had a probe in the module vaddr area which already
 599                  * freed. So, the instruction slot has already been
 600                  * released. We need a new slot for the new probe.
 601                  */
 602                 ret = arch_prepare_kprobe(ap);
 603                 if (ret)
 604                         /*
 605                          * Even if fail to allocate new slot, don't need to
 606                          * free aggr_probe. It will be used next time, or
 607                          * freed by unregister_kprobe.
 608                          */
 609                         return ret;
 610
 611                 /*
 612                  * Clear gone flag to prevent allocating new slot again, and
 613                  * set disabled flag because it is not armed yet.
 614                  */
 615                 ap->flags = (ap->flags & ~KPROBE_FLAG_GONE)
 616                             | KPROBE_FLAG_DISABLED;
 617         }
 618
 619         copy_kprobe(ap, p);
 620         return add_new_kprobe(ap, p);
 621 }
 622
 623 /* Try to disable aggr_kprobe, and return 1 if succeeded.*/
 624 static int __kprobes try_to_disable_aggr_kprobe(struct kprobe *p)
 625 {
 626         struct kprobe *kp;
 627
 628         list_for_each_entry_rcu(kp, &p->list, list) {
 629                 if (!kprobe_disabled(kp))
 630                         /*
 631                          * There is an active probe on the list.
 632                          * We can't disable aggr_kprobe.
 633                          */
 634                         return 0;
 635         }
 636         p->flags |= KPROBE_FLAG_DISABLED;
 637         return 1;
 638 }
 639
 640 static int __kprobes in_kprobes_functions(unsigned long addr)
 641 {
 642         struct kprobe_blackpoint *kb;
 643
 644         if (addr >= (unsigned long)__kprobes_text_start &&
 645             addr < (unsigned long)__kprobes_text_end)
 646                 return -EINVAL;
 647         /*
 648          * If there exists a kprobe_blacklist, verify and
 649          * fail any probe registration in the prohibited area
 650          */
 651         for (kb = kprobe_blacklist; kb->name != NULL; kb++) {
 652                 if (kb->start_addr) {
 653                         if (addr >= kb->start_addr &&
 654                             addr < (kb->start_addr + kb->range))
 655                                 return -EINVAL;
 656                 }
 657         }
 658         return 0;
 659 }
 660
 661 /*
 662  * If we have a symbol_name argument, look it up and add the offset field
 663  * to it. This way, we can specify a relative address to a symbol.
 664  */
 665 static kprobe_opcode_t __kprobes *kprobe_addr(struct kprobe *p)
 666 {
 667         kprobe_opcode_t *addr = p->addr;
 668         if (p->symbol_name) {
 669                 if (addr)
 670                         return NULL;
 671                 kprobe_lookup_name(p->symbol_name, addr);
 672         }
 673
 674         if (!addr)
 675                 return NULL;
 676         return (kprobe_opcode_t *)(((char *)addr) + p->offset);
 677 }
 678
 679 /* Check passed kprobe is valid and return kprobe in kprobe_table. */
 680 static struct kprobe * __kprobes __get_valid_kprobe(struct kprobe *p)
 681 {
 682         struct kprobe *old_p, *list_p;
 683
 684         old_p = get_kprobe(p->addr);
 685         if (unlikely(!old_p))
 686                 return NULL;
 687
 688         if (p != old_p) {
 689                 list_for_each_entry_rcu(list_p, &old_p->list, list)
 690                         if (list_p == p)
 691                         /* kprobe p is a valid probe */
 692                                 goto valid;
 693                 return NULL;
 694         }
 695 valid:
 696         return old_p;
 697 }
 698
 699 /* Return error if the kprobe is being re-registered */
 700 static inline int check_kprobe_rereg(struct kprobe *p)
 701 {
 702         int ret = 0;
 703         struct kprobe *old_p;
 704
 705         mutex_lock(&kprobe_mutex);
 706         old_p = __get_valid_kprobe(p);
 707         if (old_p)
 708                 ret = -EINVAL;
 709         mutex_unlock(&kprobe_mutex);
 710         return ret;
 711 }
 712
 713 int __kprobes register_kprobe(struct kprobe *p)
 714 {
 715         int ret = 0;
 716         struct kprobe *old_p;
 717         struct module *probed_mod;
 718         kprobe_opcode_t *addr;
 719
 720         addr = kprobe_addr(p);
 721         if (!addr)
 722                 return -EINVAL;
 723         p->addr = addr;
 724
 725         ret = check_kprobe_rereg(p);
 726         if (ret)
 727                 return ret;
 728
 729         preempt_disable();
 730         if (!kernel_text_address((unsigned long) p->addr) ||
 731             in_kprobes_functions((unsigned long) p->addr)) {
 732                 preempt_enable();
 733                 return -EINVAL;
 734         }
 735
 736         /* User can pass only KPROBE_FLAG_DISABLED to register_kprobe */
 737         p->flags &= KPROBE_FLAG_DISABLED;
 738
 739         /*
 740          * Check if are we probing a module.
 741          */
 742         probed_mod = __module_text_address((unsigned long) p->addr);
 743         if (probed_mod) {
 744                 /*
 745                  * We must hold a refcount of the probed module while updating
 746                  * its code to prohibit unexpected unloading.
 747                  */
 748                 if (unlikely(!try_module_get(probed_mod))) {
 749                         preempt_enable();
 750                         return -EINVAL;
 751                 }
 752                 /*
 753                  * If the module freed .init.text, we couldn't insert
 754                  * kprobes in there.
 755                  */
 756                 if (within_module_init((unsigned long)p->addr, probed_mod) &&
 757                     probed_mod->state != MODULE_STATE_COMING) {
 758                         module_put(probed_mod);
 759                         preempt_enable();
 760                         return -EINVAL;
 761                 }
 762         }
 763         preempt_enable();
 764
 765         p->nmissed = 0;
 766         INIT_LIST_HEAD(&p->list);
 767         mutex_lock(&kprobe_mutex);
 768         old_p = get_kprobe(p->addr);
 769         if (old_p) {
 770                 ret = register_aggr_kprobe(old_p, p);
 771                 goto out;
 772         }
 773
 774         mutex_lock(&text_mutex);
 775         ret = arch_prepare_kprobe(p);
 776         if (ret)
 777                 goto out_unlock_text;
 778
 779         INIT_HLIST_NODE(&p->hlist);
 780         hlist_add_head_rcu(&p->hlist,
 781                        &kprobe_table[hash_ptr(p->addr, KPROBE_HASH_BITS)]);
 782
 783         if (!kprobes_all_disarmed && !kprobe_disabled(p))
 784                 arch_arm_kprobe(p);
 785
 786 out_unlock_text:
 787         mutex_unlock(&text_mutex);
 788 out:
 789         mutex_unlock(&kprobe_mutex);
 790
 791         if (probed_mod)
 792                 module_put(probed_mod);
 793
 794         return ret;
 795 }
 796 EXPORT_SYMBOL_GPL(register_kprobe);
 797
 798 /*
 799  * Unregister a kprobe without a scheduler synchronization.
 800  */
 801 static int __kprobes __unregister_kprobe_top(struct kprobe *p)
 802 {
 803         struct kprobe *old_p, *list_p;
 804
 805         old_p = __get_valid_kprobe(p);
 806         if (old_p == NULL)
 807                 return -EINVAL;
 808
 809         if (old_p == p ||
 810             (old_p->pre_handler == aggr_pre_handler &&
 811              list_is_singular(&old_p->list))) {
 812                 /*
 813                  * Only probe on the hash list. Disarm only if kprobes are
 814                  * enabled and not gone - otherwise, the breakpoint would
 815                  * already have been removed. We save on flushing icache.
 816                  */
 817                 if (!kprobes_all_disarmed && !kprobe_disabled(old_p))
 818                         disarm_kprobe(p);
 819                 hlist_del_rcu(&old_p->hlist);
 820         } else {
 821                 if (p->break_handler && !kprobe_gone(p))
 822                         old_p->break_handler = NULL;
 823                 if (p->post_handler && !kprobe_gone(p)) {
 824                         list_for_each_entry_rcu(list_p, &old_p->list, list) {
 825                                 if ((list_p != p) && (list_p->post_handler))
 826                                         goto noclean;
 827                         }
 828                         old_p->post_handler = NULL;
 829                 }
 830 noclean:
 831                 list_del_rcu(&p->list);
 832                 if (!kprobe_disabled(old_p)) {
 833                         try_to_disable_aggr_kprobe(old_p);
 834                         if (!kprobes_all_disarmed && kprobe_disabled(old_p))
 835                                 disarm_kprobe(old_p);
 836                 }
 837         }
 838         return 0;
 839 }
 840
 841 static void __kprobes __unregister_kprobe_bottom(struct kprobe *p)
 842 {
 843         struct kprobe *old_p;
 844
 845         if (list_empty(&p->list))
 846                 arch_remove_kprobe(p);
 847         else if (list_is_singular(&p->list)) {
 848                 /* "p" is the last child of an aggr_kprobe */
 849                 old_p = list_entry(p->list.next, struct kprobe, list);
 850                 list_del(&p->list);
 851                 arch_remove_kprobe(old_p);
 852                 kfree(old_p);
 853         }
 854 }
 855
 856 int __kprobes register_kprobes(struct kprobe **kps, int num)
 857 {
 858         int i, ret = 0;
 859
 860         if (num <= 0)
 861                 return -EINVAL;
 862         for (i = 0; i < num; i++) {
 863                 ret = register_kprobe(kps[i]);
 864                 if (ret < 0) {
 865                         if (i > 0)
 866                                 unregister_kprobes(kps, i);
 867                         break;
 868                 }
 869         }
 870         return ret;
 871 }
 872 EXPORT_SYMBOL_GPL(register_kprobes);
 873
 874 void __kprobes unregister_kprobe(struct kprobe *p)
 875 {
 876         unregister_kprobes(&p, 1);
 877 }
 878 EXPORT_SYMBOL_GPL(unregister_kprobe);
 879
 880 void __kprobes unregister_kprobes(struct kprobe **kps, int num)
 881 {
 882         int i;
 883
 884         if (num <= 0)
 885                 return;
 886         mutex_lock(&kprobe_mutex);
 887         for (i = 0; i < num; i++)
 888                 if (__unregister_kprobe_top(kps[i]) < 0)
 889                         kps[i]->addr = NULL;
 890         mutex_unlock(&kprobe_mutex);
 891
 892         synchronize_sched();
 893         for (i = 0; i < num; i++)
 894                 if (kps[i]->addr)
 895                         __unregister_kprobe_bottom(kps[i]);
 896 }
 897 EXPORT_SYMBOL_GPL(unregister_kprobes);
 898
 899 static struct notifier_block kprobe_exceptions_nb = {
 900         .notifier_call = kprobe_exceptions_notify,
 901         .priority = 0x7fffffff /* we need to be notified first */
 902 };
 903
 904 unsigned long __weak arch_deref_entry_point(void *entry)
 905 {
 906         return (unsigned long)entry;
 907 }
 908
 909 int __kprobes register_jprobes(struct jprobe **jps, int num)
 910 {
 911         struct jprobe *jp;
 912         int ret = 0, i;
 913
 914         if (num <= 0)
 915                 return -EINVAL;
 916         for (i = 0; i < num; i++) {
 917                 unsigned long addr;
 918                 jp = jps[i];
 919                 addr = arch_deref_entry_point(jp->entry);
 920
 921                 if (!kernel_text_address(addr))
 922                         ret = -EINVAL;
 923                 else {
 924                         /* Todo: Verify probepoint is a function entry point */
 925                         jp->kp.pre_handler = setjmp_pre_handler;
 926                         jp->kp.break_handler = longjmp_break_handler;
 927                         ret = register_kprobe(&jp->kp);
 928                 }
 929                 if (ret < 0) {
 930                         if (i > 0)
 931                                 unregister_jprobes(jps, i);
 932                         break;
 933                 }
 934         }
 935         return ret;
 936 }
 937 EXPORT_SYMBOL_GPL(register_jprobes);
 938
 939 int __kprobes register_jprobe(struct jprobe *jp)
 940 {
 941         return register_jprobes(&jp, 1);
 942 }
 943 EXPORT_SYMBOL_GPL(register_jprobe);
 944
 945 void __kprobes unregister_jprobe(struct jprobe *jp)
 946 {
 947         unregister_jprobes(&jp, 1);
 948 }
 949 EXPORT_SYMBOL_GPL(unregister_jprobe);
 950
 951 void __kprobes unregister_jprobes(struct jprobe **jps, int num)
 952 {
 953         int i;
 954
 955         if (num <= 0)
 956                 return;
 957         mutex_lock(&kprobe_mutex);
 958         for (i = 0; i < num; i++)
 959                 if (__unregister_kprobe_top(&jps[i]->kp) < 0)
 960                         jps[i]->kp.addr = NULL;
 961         mutex_unlock(&kprobe_mutex);
 962
 963         synchronize_sched();
 964         for (i = 0; i < num; i++) {
 965                 if (jps[i]->kp.addr)
 966                         __unregister_kprobe_bottom(&jps[i]->kp);
 967         }
 968 }
 969 EXPORT_SYMBOL_GPL(unregister_jprobes);
 970
 971 #ifdef CONFIG_KRETPROBES
 972 /*
 973  * This kprobe pre_handler is registered with every kretprobe. When probe
 974  * hits it will set up the return probe.
 975  */
 976 static int __kprobes pre_handler_kretprobe(struct kprobe *p,
 977                                            struct pt_regs *regs)
 978 {
 979         struct kretprobe *rp = container_of(p, struct kretprobe, kp);
 980         unsigned long hash, flags = 0;
 981         struct kretprobe_instance *ri;
 982
 983         /*TODO: consider to only swap the RA after the last pre_handler fired */
 984         hash = hash_ptr(current, KPROBE_HASH_BITS);
 985         spin_lock_irqsave(&rp->lock, flags);
 986         if (!hlist_empty(&rp->free_instances)) {
 987                 ri = hlist_entry(rp->free_instances.first,
 988                                 struct kretprobe_instance, hlist);
 989                 hlist_del(&ri->hlist);
 990                 spin_unlock_irqrestore(&rp->lock, flags);
 991
 992                 ri->rp = rp;
 993                 ri->task = current;
 994
 995                 if (rp->entry_handler && rp->entry_handler(ri, regs))
 996                         return 0;
 997
 998                 arch_prepare_kretprobe(ri, regs);
 999
1000                 /* XXX(hch): why is there no hlist_move_head? */
1001                 INIT_HLIST_NODE(&ri->hlist);
1002                 kretprobe_table_lock(hash, &flags);
1003                 hlist_add_head(&ri->hlist, &kretprobe_inst_table[hash]);
1004                 kretprobe_table_unlock(hash, &flags);
1005         } else {
1006                 rp->nmissed++;
1007                 spin_unlock_irqrestore(&rp->lock, flags);
1008         }
1009         return 0;
1010 }
1011
1012 int __kprobes register_kretprobe(struct kretprobe *rp)
1013 {
1014         int ret = 0;
1015         struct kretprobe_instance *inst;
1016         int i;
1017         void *addr;
1018
1019         if (kretprobe_blacklist_size) {
1020                 addr = kprobe_addr(&rp->kp);
1021                 if (!addr)
1022                         return -EINVAL;
1023
1024                 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
1025                         if (kretprobe_blacklist[i].addr == addr)
1026                                 return -EINVAL;
1027                 }
1028         }
1029
1030         rp->kp.pre_handler = pre_handler_kretprobe;
1031         rp->kp.post_handler = NULL;
1032         rp->kp.fault_handler = NULL;
1033         rp->kp.break_handler = NULL;
1034
1035         /* Pre-allocate memory for max kretprobe instances */
1036         if (rp->maxactive <= 0) {
1037 #ifdef CONFIG_PREEMPT
1038                 rp->maxactive = max(10, 2 * num_possible_cpus());
1039 #else
1040                 rp->maxactive = num_possible_cpus();
1041 #endif
1042         }
1043         spin_lock_init(&rp->lock);
1044         INIT_HLIST_HEAD(&rp->free_instances);
1045         for (i = 0; i < rp->maxactive; i++) {
1046                 inst = kmalloc(sizeof(struct kretprobe_instance) +
1047                                rp->data_size, GFP_KERNEL);
1048                 if (inst == NULL) {
1049                         free_rp_inst(rp);
1050                         return -ENOMEM;
1051                 }
1052                 INIT_HLIST_NODE(&inst->hlist);
1053                 hlist_add_head(&inst->hlist, &rp->free_instances);
1054         }
1055
1056         rp->nmissed = 0;
1057         /* Establish function entry probe point */
1058         ret = register_kprobe(&rp->kp);
1059         if (ret != 0)
1060                 free_rp_inst(rp);
1061         return ret;
1062 }
1063 EXPORT_SYMBOL_GPL(register_kretprobe);
1064
1065 int __kprobes register_kretprobes(struct kretprobe **rps, int num)
1066 {
1067         int ret = 0, i;
1068
1069         if (num <= 0)
1070                 return -EINVAL;
1071         for (i = 0; i < num; i++) {
1072                 ret = register_kretprobe(rps[i]);
1073                 if (ret < 0) {
1074                         if (i > 0)
1075                                 unregister_kretprobes(rps, i);
1076                         break;
1077                 }
1078         }
1079         return ret;
1080 }
1081 EXPORT_SYMBOL_GPL(register_kretprobes);
1082
1083 void __kprobes unregister_kretprobe(struct kretprobe *rp)
1084 {
1085         unregister_kretprobes(&rp, 1);
1086 }
1087 EXPORT_SYMBOL_GPL(unregister_kretprobe);
1088
1089 void __kprobes unregister_kretprobes(struct kretprobe **rps, int num)
1090 {
1091         int i;
1092
1093         if (num <= 0)
1094                 return;
1095         mutex_lock(&kprobe_mutex);
1096         for (i = 0; i < num; i++)
1097                 if (__unregister_kprobe_top(&rps[i]->kp) < 0)
1098                         rps[i]->kp.addr = NULL;
1099         mutex_unlock(&kprobe_mutex);
1100
1101         synchronize_sched();
1102         for (i = 0; i < num; i++) {
1103                 if (rps[i]->kp.addr) {
1104                         __unregister_kprobe_bottom(&rps[i]->kp);
1105                         cleanup_rp_inst(rps[i]);
1106                 }
1107         }
1108 }
1109 EXPORT_SYMBOL_GPL(unregister_kretprobes);
1110
1111 #else /* CONFIG_KRETPROBES */
1112 int __kprobes register_kretprobe(struct kretprobe *rp)
1113 {
1114         return -ENOSYS;
1115 }
1116 EXPORT_SYMBOL_GPL(register_kretprobe);
1117
1118 int __kprobes register_kretprobes(struct kretprobe **rps, int num)
1119 {
1120         return -ENOSYS;
1121 }
1122 EXPORT_SYMBOL_GPL(register_kretprobes);
1123
1124 void __kprobes unregister_kretprobe(struct kretprobe *rp)
1125 {
1126 }
1127 EXPORT_SYMBOL_GPL(unregister_kretprobe);
1128
1129 void __kprobes unregister_kretprobes(struct kretprobe **rps, int num)
1130 {
1131 }
1132 EXPORT_SYMBOL_GPL(unregister_kretprobes);
1133
1134 static int __kprobes pre_handler_kretprobe(struct kprobe *p,
1135                                            struct pt_regs *regs)
1136 {
1137         return 0;
1138 }
1139
1140 #endif /* CONFIG_KRETPROBES */
1141
1142 /* Set the kprobe gone and remove its instruction buffer. */
1143 static void __kprobes kill_kprobe(struct kprobe *p)
1144 {
1145         struct kprobe *kp;
1146
1147         p->flags |= KPROBE_FLAG_GONE;
1148         if (p->pre_handler == aggr_pre_handler) {
1149                 /*
1150                  * If this is an aggr_kprobe, we have to list all the
1151                  * chained probes and mark them GONE.
1152                  */
1153                 list_for_each_entry_rcu(kp, &p->list, list)
1154                         kp->flags |= KPROBE_FLAG_GONE;
1155                 p->post_handler = NULL;
1156                 p->break_handler = NULL;
1157         }
1158         /*
1159          * Here, we can remove insn_slot safely, because no thread calls
1160          * the original probed function (which will be freed soon) any more.
1161          */
1162         arch_remove_kprobe(p);
1163 }
1164
1165 void __kprobes dump_kprobe(struct kprobe *kp)
1166 {
1167         printk(KERN_WARNING "Dumping kprobe:\n");
1168         printk(KERN_WARNING "Name: %s\nAddress: %p\nOffset: %x\n",
1169                kp->symbol_name, kp->addr, kp->offset);
1170 }
1171
1172 /* Module notifier call back, checking kprobes on the module */
1173 static int __kprobes kprobes_module_callback(struct notifier_block *nb,
1174                                              unsigned long val, void *data)
1175 {
1176         struct module *mod = data;
1177         struct hlist_head *head;
1178         struct hlist_node *node;
1179         struct kprobe *p;
1180         unsigned int i;
1181         int checkcore = (val == MODULE_STATE_GOING);
1182
1183         if (val != MODULE_STATE_GOING && val != MODULE_STATE_LIVE)
1184                 return NOTIFY_DONE;
1185
1186         /*
1187          * When MODULE_STATE_GOING was notified, both of module .text and
1188          * .init.text sections would be freed. When MODULE_STATE_LIVE was
1189          * notified, only .init.text section would be freed. We need to
1190          * disable kprobes which have been inserted in the sections.
1191          */
1192         mutex_lock(&kprobe_mutex);
1193         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
1194                 head = &kprobe_table[i];
1195                 hlist_for_each_entry_rcu(p, node, head, hlist)
1196                         if (within_module_init((unsigned long)p->addr, mod) ||
1197                             (checkcore &&
1198                              within_module_core((unsigned long)p->addr, mod))) {
1199                                 /*
1200                                  * The vaddr this probe is installed will soon
1201                                  * be vfreed buy not synced to disk. Hence,
1202                                  * disarming the breakpoint isn't needed.
1203                                  */
1204                                 kill_kprobe(p);
1205                         }
1206         }
1207         mutex_unlock(&kprobe_mutex);
1208         return NOTIFY_DONE;
1209 }
1210
1211 static struct notifier_block kprobe_module_nb = {
1212         .notifier_call = kprobes_module_callback,
1213         .priority = 0
1214 };
1215
1216 static int __init init_kprobes(void)
1217 {
1218         int i, err = 0;
1219         unsigned long offset = 0, size = 0;
1220         char *modname, namebuf[128];
1221         const char *symbol_name;
1222         void *addr;
1223         struct kprobe_blackpoint *kb;
1224
1225         /* FIXME allocate the probe table, currently defined statically */
1226         /* initialize all list heads */
1227         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
1228                 INIT_HLIST_HEAD(&kprobe_table[i]);
1229                 INIT_HLIST_HEAD(&kretprobe_inst_table[i]);
1230                 spin_lock_init(&(kretprobe_table_locks[i].lock));
1231         }
1232
1233         /*
1234          * Lookup and populate the kprobe_blacklist.
1235          *
1236          * Unlike the kretprobe blacklist, we'll need to determine
1237          * the range of addresses that belong to the said functions,
1238          * since a kprobe need not necessarily be at the beginning
1239          * of a function.
1240          */
1241         for (kb = kprobe_blacklist; kb->name != NULL; kb++) {
1242                 kprobe_lookup_name(kb->name, addr);
1243                 if (!addr)
1244                         continue;
1245
1246                 kb->start_addr = (unsigned long)addr;
1247                 symbol_name = kallsyms_lookup(kb->start_addr,
1248                                 &size, &offset, &modname, namebuf);
1249                 if (!symbol_name)
1250                         kb->range = 0;
1251                 else
1252                         kb->range = size;
1253         }
1254
1255         if (kretprobe_blacklist_size) {
1256                 /* lookup the function address from its name */
1257                 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
1258                         kprobe_lookup_name(kretprobe_blacklist[i].name,
1259                                            kretprobe_blacklist[i].addr);
1260                         if (!kretprobe_blacklist[i].addr)
1261                                 printk("kretprobe: lookup failed: %s\n",
1262                                        kretprobe_blacklist[i].name);
1263                 }
1264         }
1265
1266         /* By default, kprobes are armed */
1267         kprobes_all_disarmed = false;
1268
1269         err = arch_init_kprobes();
1270         if (!err)
1271                 err = register_die_notifier(&kprobe_exceptions_nb);
1272         if (!err)
1273                 err = register_module_notifier(&kprobe_module_nb);
1274
1275         kprobes_initialized = (err == 0);
1276
1277         if (!err)
1278                 init_test_probes();
1279         return err;
1280 }
1281
1282 #ifdef CONFIG_DEBUG_FS
1283 static void __kprobes report_probe(struct seq_file *pi, struct kprobe *p,
1284                 const char *sym, int offset,char *modname)
1285 {
1286         char *kprobe_type;
1287
1288         if (p->pre_handler == pre_handler_kretprobe)
1289                 kprobe_type = "r";
1290         else if (p->pre_handler == setjmp_pre_handler)
1291                 kprobe_type = "j";
1292         else
1293                 kprobe_type = "k";
1294         if (sym)
1295                 seq_printf(pi, "%p  %s  %s+0x%x  %s %s%s\n",
1296                         p->addr, kprobe_type, sym, offset,
1297                         (modname ? modname : " "),
1298                         (kprobe_gone(p) ? "[GONE]" : ""),
1299                         ((kprobe_disabled(p) && !kprobe_gone(p)) ?
1300                          "[DISABLED]" : ""));
1301         else
1302                 seq_printf(pi, "%p  %s  %p %s%s\n",
1303                         p->addr, kprobe_type, p->addr,
1304                         (kprobe_gone(p) ? "[GONE]" : ""),
1305                         ((kprobe_disabled(p) && !kprobe_gone(p)) ?
1306                          "[DISABLED]" : ""));
1307 }
1308
1309 static void __kprobes *kprobe_seq_start(struct seq_file *f, loff_t *pos)
1310 {
1311         return (*pos < KPROBE_TABLE_SIZE) ? pos : NULL;
1312 }
1313
1314 static void __kprobes *kprobe_seq_next(struct seq_file *f, void *v, loff_t *pos)
1315 {
1316         (*pos)++;
1317         if (*pos >= KPROBE_TABLE_SIZE)
1318                 return NULL;
1319         return pos;
1320 }
1321
1322 static void __kprobes kprobe_seq_stop(struct seq_file *f, void *v)
1323 {
1324         /* Nothing to do */
1325 }
1326
1327 static int __kprobes show_kprobe_addr(struct seq_file *pi, void *v)
1328 {
1329         struct hlist_head *head;
1330         struct hlist_node *node;
1331         struct kprobe *p, *kp;
1332         const char *sym = NULL;
1333         unsigned int i = *(loff_t *) v;
1334         unsigned long offset = 0;
1335         char *modname, namebuf[128];
1336
1337         head = &kprobe_table[i];
1338         preempt_disable();
1339         hlist_for_each_entry_rcu(p, node, head, hlist) {
1340                 sym = kallsyms_lookup((unsigned long)p->addr, NULL,
1341                                         &offset, &modname, namebuf);
1342                 if (p->pre_handler == aggr_pre_handler) {
1343                         list_for_each_entry_rcu(kp, &p->list, list)
1344                                 report_probe(pi, kp, sym, offset, modname);
1345                 } else
1346                         report_probe(pi, p, sym, offset, modname);
1347         }
1348         preempt_enable();
1349         return 0;
1350 }
1351
1352 static const struct seq_operations kprobes_seq_ops = {
1353         .start = kprobe_seq_start,
1354         .next  = kprobe_seq_next,
1355         .stop  = kprobe_seq_stop,
1356         .show  = show_kprobe_addr
1357 };
1358
1359 static int __kprobes kprobes_open(struct inode *inode, struct file *filp)
1360 {
1361         return seq_open(filp, &kprobes_seq_ops);
1362 }
1363
1364 static const struct file_operations debugfs_kprobes_operations = {
1365         .open           = kprobes_open,
1366         .read           = seq_read,
1367         .llseek         = seq_lseek,
1368         .release        = seq_release,
1369 };
1370
1371 /* Disable one kprobe */
1372 int __kprobes disable_kprobe(struct kprobe *kp)
1373 {
1374         int ret = 0;
1375         struct kprobe *p;
1376
1377         mutex_lock(&kprobe_mutex);
1378
1379         /* Check whether specified probe is valid. */
1380         p = __get_valid_kprobe(kp);
1381         if (unlikely(p == NULL)) {
1382                 ret = -EINVAL;
1383                 goto out;
1384         }
1385
1386         /* If the probe is already disabled (or gone), just return */
1387         if (kprobe_disabled(kp))
1388                 goto out;
1389
1390         kp->flags |= KPROBE_FLAG_DISABLED;
1391         if (p != kp)
1392                 /* When kp != p, p is always enabled. */
1393                 try_to_disable_aggr_kprobe(p);
1394
1395         if (!kprobes_all_disarmed && kprobe_disabled(p))
1396                 disarm_kprobe(p);
1397 out:
1398         mutex_unlock(&kprobe_mutex);
1399         return ret;
1400 }
1401 EXPORT_SYMBOL_GPL(disable_kprobe);
1402
1403 /* Enable one kprobe */
1404 int __kprobes enable_kprobe(struct kprobe *kp)
1405 {
1406         int ret = 0;
1407         struct kprobe *p;
1408
1409         mutex_lock(&kprobe_mutex);
1410
1411         /* Check whether specified probe is valid. */
1412         p = __get_valid_kprobe(kp);
1413         if (unlikely(p == NULL)) {
1414                 ret = -EINVAL;
1415                 goto out;
1416         }
1417
1418         if (kprobe_gone(kp)) {
1419                 /* This kprobe has gone, we couldn't enable it. */
1420                 ret = -EINVAL;
1421                 goto out;
1422         }
1423
1424         if (!kprobes_all_disarmed && kprobe_disabled(p))
1425                 arm_kprobe(p);
1426
1427         p->flags &= ~KPROBE_FLAG_DISABLED;
1428         if (p != kp)
1429                 kp->flags &= ~KPROBE_FLAG_DISABLED;
1430 out:
1431         mutex_unlock(&kprobe_mutex);
1432         return ret;
1433 }
1434 EXPORT_SYMBOL_GPL(enable_kprobe);
1435
1436 static void __kprobes arm_all_kprobes(void)
1437 {
1438         struct hlist_head *head;
1439         struct hlist_node *node;
1440         struct kprobe *p;
1441         unsigned int i;
1442
1443         mutex_lock(&kprobe_mutex);
1444
1445         /* If kprobes are armed, just return */
1446         if (!kprobes_all_disarmed)
1447                 goto already_enabled;
1448
1449         mutex_lock(&text_mutex);
1450         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
1451                 head = &kprobe_table[i];
1452                 hlist_for_each_entry_rcu(p, node, head, hlist)
1453                         if (!kprobe_disabled(p))
1454                                 arch_arm_kprobe(p);
1455         }
1456         mutex_unlock(&text_mutex);
1457
1458         kprobes_all_disarmed = false;
1459         printk(KERN_INFO "Kprobes globally enabled\n");
1460
1461 already_enabled:
1462         mutex_unlock(&kprobe_mutex);
1463         return;
1464 }
1465
1466 static void __kprobes disarm_all_kprobes(void)
1467 {
1468         struct hlist_head *head;
1469         struct hlist_node *node;
1470         struct kprobe *p;
1471         unsigned int i;
1472
1473         mutex_lock(&kprobe_mutex);
1474
1475         /* If kprobes are already disarmed, just return */
1476         if (kprobes_all_disarmed)
1477                 goto already_disabled;
1478
1479         kprobes_all_disarmed = true;
1480         printk(KERN_INFO "Kprobes globally disabled\n");
1481         mutex_lock(&text_mutex);
1482         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
1483                 head = &kprobe_table[i];
1484                 hlist_for_each_entry_rcu(p, node, head, hlist) {
1485                         if (!arch_trampoline_kprobe(p) && !kprobe_disabled(p))
1486                                 arch_disarm_kprobe(p);
1487                 }
1488         }
1489
1490         mutex_unlock(&text_mutex);
1491         mutex_unlock(&kprobe_mutex);
1492         /* Allow all currently running kprobes to complete */
1493         synchronize_sched();
1494         return;
1495
1496 already_disabled:
1497         mutex_unlock(&kprobe_mutex);
1498         return;
1499 }
1500
1501 /*
1502  * XXX: The debugfs bool file interface doesn't allow for callbacks
1503  * when the bool state is switched. We can reuse that facility when
1504  * available
1505  */
1506 static ssize_t read_enabled_file_bool(struct file *file,
1507                char __user *user_buf, size_t count, loff_t *ppos)
1508 {
1509         char buf[3];
1510
1511         if (!kprobes_all_disarmed)
1512                 buf[0] = '1';
1513         else
1514                 buf[0] = '0';
1515         buf[1] = '\n';
1516         buf[2] = 0x00;
1517         return simple_read_from_buffer(user_buf, count, ppos, buf, 2);
1518 }
1519
1520 static ssize_t write_enabled_file_bool(struct file *file,
1521                const char __user *user_buf, size_t count, loff_t *ppos)
1522 {
1523         char buf[32];
1524         int buf_size;
1525
1526         buf_size = min(count, (sizeof(buf)-1));
1527         if (copy_from_user(buf, user_buf, buf_size))
1528                 return -EFAULT;
1529
1530         switch (buf[0]) {
1531         case 'y':
1532         case 'Y':
1533         case '1':
1534                 arm_all_kprobes();
1535                 break;
1536         case 'n':
1537         case 'N':
1538         case '0':
1539                 disarm_all_kprobes();
1540                 break;
1541         }
1542
1543         return count;
1544 }
1545
1546 static const struct file_operations fops_kp = {
1547         .read =         read_enabled_file_bool,
1548         .write =        write_enabled_file_bool,
1549 };
1550
1551 static int __kprobes debugfs_kprobe_init(void)
1552 {
1553         struct dentry *dir, *file;
1554         unsigned int value = 1;
1555
1556         dir = debugfs_create_dir("kprobes", NULL);
1557         if (!dir)
1558                 return -ENOMEM;
1559
1560         file = debugfs_create_file("list", 0444, dir, NULL,
1561                                 &debugfs_kprobes_operations);
1562         if (!file) {
1563                 debugfs_remove(dir);
1564                 return -ENOMEM;
1565         }
1566
1567         file = debugfs_create_file("enabled", 0600, dir,
1568                                         &value, &fops_kp);
1569         if (!file) {
1570                 debugfs_remove(dir);
1571                 return -ENOMEM;
1572         }
1573
1574         return 0;
1575 }
1576
1577 late_initcall(debugfs_kprobe_init);
1578 #endif /* CONFIG_DEBUG_FS */
1579
1580 module_init(init_kprobes);
1581
1582 /* defined in arch/.../kernel/kprobes.c */
1583 EXPORT_SYMBOL_GPL(jprobe_return);