git.oblomov.eu Git - linux-2.6/blob - kernel/kprobes.c

   1 /*
   2  *  Kernel Probes (KProbes)
   3  *  kernel/kprobes.c
   4  *
   5  * This program is free software; you can redistribute it and/or modify
   6  * it under the terms of the GNU General Public License as published by
   7  * the Free Software Foundation; either version 2 of the License, or
   8  * (at your option) any later version.
   9  *
  10  * This program is distributed in the hope that it will be useful,
  11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  * GNU General Public License for more details.
  14  *
  15  * You should have received a copy of the GNU General Public License
  16  * along with this program; if not, write to the Free Software
  17  * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
  18  *
  19  * Copyright (C) IBM Corporation, 2002, 2004
  20  *
  21  * 2002-Oct     Created by Vamsi Krishna S <vamsi_krishna@in.ibm.com> Kernel
  22  *              Probes initial implementation (includes suggestions from
  23  *              Rusty Russell).
  24  * 2004-Aug     Updated by Prasanna S Panchamukhi <prasanna@in.ibm.com> with
  25  *              hlists and exceptions notifier as suggested by Andi Kleen.
  26  * 2004-July    Suparna Bhattacharya <suparna@in.ibm.com> added jumper probes
  27  *              interface to access function arguments.
  28  * 2004-Sep     Prasanna S Panchamukhi <prasanna@in.ibm.com> Changed Kprobes
  29  *              exceptions notifier to be first on the priority list.
  30  * 2005-May     Hien Nguyen <hien@us.ibm.com>, Jim Keniston
  31  *              <jkenisto@us.ibm.com> and Prasanna S Panchamukhi
  32  *              <prasanna@in.ibm.com> added function-return probes.
  33  */
  34 #include <linux/kprobes.h>
  35 #include <linux/hash.h>
  36 #include <linux/init.h>
  37 #include <linux/slab.h>
  38 #include <linux/stddef.h>
  39 #include <linux/module.h>
  40 #include <linux/moduleloader.h>
  41 #include <linux/kallsyms.h>
  42 #include <linux/freezer.h>
  43 #include <linux/seq_file.h>
  44 #include <linux/debugfs.h>
  45 #include <linux/kdebug.h>
  46
  47 #include <asm-generic/sections.h>
  48 #include <asm/cacheflush.h>
  49 #include <asm/errno.h>
  50 #include <asm/uaccess.h>
  51
  52 #define KPROBE_HASH_BITS 6
  53 #define KPROBE_TABLE_SIZE (1 << KPROBE_HASH_BITS)
  54
  55
  56 /*
  57  * Some oddball architectures like 64bit powerpc have function descriptors
  58  * so this must be overridable.
  59  */
  60 #ifndef kprobe_lookup_name
  61 #define kprobe_lookup_name(name, addr) \
  62         addr = ((kprobe_opcode_t *)(kallsyms_lookup_name(name)))
  63 #endif
  64
  65 static struct hlist_head kprobe_table[KPROBE_TABLE_SIZE];
  66 static struct hlist_head kretprobe_inst_table[KPROBE_TABLE_SIZE];
  67
  68 /* NOTE: change this value only with kprobe_mutex held */
  69 static bool kprobe_enabled;
  70
  71 DEFINE_MUTEX(kprobe_mutex);             /* Protects kprobe_table */
  72 DEFINE_SPINLOCK(kretprobe_lock);        /* Protects kretprobe_inst_table */
  73 static DEFINE_PER_CPU(struct kprobe *, kprobe_instance) = NULL;
  74
  75 /*
  76  * Normally, functions that we'd want to prohibit kprobes in, are marked
  77  * __kprobes. But, there are cases where such functions already belong to
  78  * a different section (__sched for preempt_schedule)
  79  *
  80  * For such cases, we now have a blacklist
  81  */
  82 struct kprobe_blackpoint kprobe_blacklist[] = {
  83         {"preempt_schedule",},
  84         {NULL}    /* Terminator */
  85 };
  86
  87 #ifdef __ARCH_WANT_KPROBES_INSN_SLOT
  88 /*
  89  * kprobe->ainsn.insn points to the copy of the instruction to be
  90  * single-stepped. x86_64, POWER4 and above have no-exec support and
  91  * stepping on the instruction on a vmalloced/kmalloced/data page
  92  * is a recipe for disaster
  93  */
  94 #define INSNS_PER_PAGE  (PAGE_SIZE/(MAX_INSN_SIZE * sizeof(kprobe_opcode_t)))
  95
  96 struct kprobe_insn_page {
  97         struct hlist_node hlist;
  98         kprobe_opcode_t *insns;         /* Page of instruction slots */
  99         char slot_used[INSNS_PER_PAGE];
 100         int nused;
 101         int ngarbage;
 102 };
 103
 104 enum kprobe_slot_state {
 105         SLOT_CLEAN = 0,
 106         SLOT_DIRTY = 1,
 107         SLOT_USED = 2,
 108 };
 109
 110 static struct hlist_head kprobe_insn_pages;
 111 static int kprobe_garbage_slots;
 112 static int collect_garbage_slots(void);
 113
 114 static int __kprobes check_safety(void)
 115 {
 116         int ret = 0;
 117 #if defined(CONFIG_PREEMPT) && defined(CONFIG_PM)
 118         ret = freeze_processes();
 119         if (ret == 0) {
 120                 struct task_struct *p, *q;
 121                 do_each_thread(p, q) {
 122                         if (p != current && p->state == TASK_RUNNING &&
 123                             p->pid != 0) {
 124                                 printk("Check failed: %s is running\n",p->comm);
 125                                 ret = -1;
 126                                 goto loop_end;
 127                         }
 128                 } while_each_thread(p, q);
 129         }
 130 loop_end:
 131         thaw_processes();
 132 #else
 133         synchronize_sched();
 134 #endif
 135         return ret;
 136 }
 137
 138 /**
 139  * get_insn_slot() - Find a slot on an executable page for an instruction.
 140  * We allocate an executable page if there's no room on existing ones.
 141  */
 142 kprobe_opcode_t __kprobes *get_insn_slot(void)
 143 {
 144         struct kprobe_insn_page *kip;
 145         struct hlist_node *pos;
 146
 147  retry:
 148         hlist_for_each_entry(kip, pos, &kprobe_insn_pages, hlist) {
 149                 if (kip->nused < INSNS_PER_PAGE) {
 150                         int i;
 151                         for (i = 0; i < INSNS_PER_PAGE; i++) {
 152                                 if (kip->slot_used[i] == SLOT_CLEAN) {
 153                                         kip->slot_used[i] = SLOT_USED;
 154                                         kip->nused++;
 155                                         return kip->insns + (i * MAX_INSN_SIZE);
 156                                 }
 157                         }
 158                         /* Surprise!  No unused slots.  Fix kip->nused. */
 159                         kip->nused = INSNS_PER_PAGE;
 160                 }
 161         }
 162
 163         /* If there are any garbage slots, collect it and try again. */
 164         if (kprobe_garbage_slots && collect_garbage_slots() == 0) {
 165                 goto retry;
 166         }
 167         /* All out of space.  Need to allocate a new page. Use slot 0. */
 168         kip = kmalloc(sizeof(struct kprobe_insn_page), GFP_KERNEL);
 169         if (!kip)
 170                 return NULL;
 171
 172         /*
 173          * Use module_alloc so this page is within +/- 2GB of where the
 174          * kernel image and loaded module images reside. This is required
 175          * so x86_64 can correctly handle the %rip-relative fixups.
 176          */
 177         kip->insns = module_alloc(PAGE_SIZE);
 178         if (!kip->insns) {
 179                 kfree(kip);
 180                 return NULL;
 181         }
 182         INIT_HLIST_NODE(&kip->hlist);
 183         hlist_add_head(&kip->hlist, &kprobe_insn_pages);
 184         memset(kip->slot_used, SLOT_CLEAN, INSNS_PER_PAGE);
 185         kip->slot_used[0] = SLOT_USED;
 186         kip->nused = 1;
 187         kip->ngarbage = 0;
 188         return kip->insns;
 189 }
 190
 191 /* Return 1 if all garbages are collected, otherwise 0. */
 192 static int __kprobes collect_one_slot(struct kprobe_insn_page *kip, int idx)
 193 {
 194         kip->slot_used[idx] = SLOT_CLEAN;
 195         kip->nused--;
 196         if (kip->nused == 0) {
 197                 /*
 198                  * Page is no longer in use.  Free it unless
 199                  * it's the last one.  We keep the last one
 200                  * so as not to have to set it up again the
 201                  * next time somebody inserts a probe.
 202                  */
 203                 hlist_del(&kip->hlist);
 204                 if (hlist_empty(&kprobe_insn_pages)) {
 205                         INIT_HLIST_NODE(&kip->hlist);
 206                         hlist_add_head(&kip->hlist,
 207                                        &kprobe_insn_pages);
 208                 } else {
 209                         module_free(NULL, kip->insns);
 210                         kfree(kip);
 211                 }
 212                 return 1;
 213         }
 214         return 0;
 215 }
 216
 217 static int __kprobes collect_garbage_slots(void)
 218 {
 219         struct kprobe_insn_page *kip;
 220         struct hlist_node *pos, *next;
 221
 222         /* Ensure no-one is preepmted on the garbages */
 223         if (check_safety() != 0)
 224                 return -EAGAIN;
 225
 226         hlist_for_each_entry_safe(kip, pos, next, &kprobe_insn_pages, hlist) {
 227                 int i;
 228                 if (kip->ngarbage == 0)
 229                         continue;
 230                 kip->ngarbage = 0;      /* we will collect all garbages */
 231                 for (i = 0; i < INSNS_PER_PAGE; i++) {
 232                         if (kip->slot_used[i] == SLOT_DIRTY &&
 233                             collect_one_slot(kip, i))
 234                                 break;
 235                 }
 236         }
 237         kprobe_garbage_slots = 0;
 238         return 0;
 239 }
 240
 241 void __kprobes free_insn_slot(kprobe_opcode_t * slot, int dirty)
 242 {
 243         struct kprobe_insn_page *kip;
 244         struct hlist_node *pos;
 245
 246         hlist_for_each_entry(kip, pos, &kprobe_insn_pages, hlist) {
 247                 if (kip->insns <= slot &&
 248                     slot < kip->insns + (INSNS_PER_PAGE * MAX_INSN_SIZE)) {
 249                         int i = (slot - kip->insns) / MAX_INSN_SIZE;
 250                         if (dirty) {
 251                                 kip->slot_used[i] = SLOT_DIRTY;
 252                                 kip->ngarbage++;
 253                         } else {
 254                                 collect_one_slot(kip, i);
 255                         }
 256                         break;
 257                 }
 258         }
 259
 260         if (dirty && ++kprobe_garbage_slots > INSNS_PER_PAGE)
 261                 collect_garbage_slots();
 262 }
 263 #endif
 264
 265 /* We have preemption disabled.. so it is safe to use __ versions */
 266 static inline void set_kprobe_instance(struct kprobe *kp)
 267 {
 268         __get_cpu_var(kprobe_instance) = kp;
 269 }
 270
 271 static inline void reset_kprobe_instance(void)
 272 {
 273         __get_cpu_var(kprobe_instance) = NULL;
 274 }
 275
 276 /*
 277  * This routine is called either:
 278  *      - under the kprobe_mutex - during kprobe_[un]register()
 279  *                              OR
 280  *      - with preemption disabled - from arch/xxx/kernel/kprobes.c
 281  */
 282 struct kprobe __kprobes *get_kprobe(void *addr)
 283 {
 284         struct hlist_head *head;
 285         struct hlist_node *node;
 286         struct kprobe *p;
 287
 288         head = &kprobe_table[hash_ptr(addr, KPROBE_HASH_BITS)];
 289         hlist_for_each_entry_rcu(p, node, head, hlist) {
 290                 if (p->addr == addr)
 291                         return p;
 292         }
 293         return NULL;
 294 }
 295
 296 /*
 297  * Aggregate handlers for multiple kprobes support - these handlers
 298  * take care of invoking the individual kprobe handlers on p->list
 299  */
 300 static int __kprobes aggr_pre_handler(struct kprobe *p, struct pt_regs *regs)
 301 {
 302         struct kprobe *kp;
 303
 304         list_for_each_entry_rcu(kp, &p->list, list) {
 305                 if (kp->pre_handler) {
 306                         set_kprobe_instance(kp);
 307                         if (kp->pre_handler(kp, regs))
 308                                 return 1;
 309                 }
 310                 reset_kprobe_instance();
 311         }
 312         return 0;
 313 }
 314
 315 static void __kprobes aggr_post_handler(struct kprobe *p, struct pt_regs *regs,
 316                                         unsigned long flags)
 317 {
 318         struct kprobe *kp;
 319
 320         list_for_each_entry_rcu(kp, &p->list, list) {
 321                 if (kp->post_handler) {
 322                         set_kprobe_instance(kp);
 323                         kp->post_handler(kp, regs, flags);
 324                         reset_kprobe_instance();
 325                 }
 326         }
 327 }
 328
 329 static int __kprobes aggr_fault_handler(struct kprobe *p, struct pt_regs *regs,
 330                                         int trapnr)
 331 {
 332         struct kprobe *cur = __get_cpu_var(kprobe_instance);
 333
 334         /*
 335          * if we faulted "during" the execution of a user specified
 336          * probe handler, invoke just that probe's fault handler
 337          */
 338         if (cur && cur->fault_handler) {
 339                 if (cur->fault_handler(cur, regs, trapnr))
 340                         return 1;
 341         }
 342         return 0;
 343 }
 344
 345 static int __kprobes aggr_break_handler(struct kprobe *p, struct pt_regs *regs)
 346 {
 347         struct kprobe *cur = __get_cpu_var(kprobe_instance);
 348         int ret = 0;
 349
 350         if (cur && cur->break_handler) {
 351                 if (cur->break_handler(cur, regs))
 352                         ret = 1;
 353         }
 354         reset_kprobe_instance();
 355         return ret;
 356 }
 357
 358 /* Walks the list and increments nmissed count for multiprobe case */
 359 void __kprobes kprobes_inc_nmissed_count(struct kprobe *p)
 360 {
 361         struct kprobe *kp;
 362         if (p->pre_handler != aggr_pre_handler) {
 363                 p->nmissed++;
 364         } else {
 365                 list_for_each_entry_rcu(kp, &p->list, list)
 366                         kp->nmissed++;
 367         }
 368         return;
 369 }
 370
 371 /* Called with kretprobe_lock held */
 372 void __kprobes recycle_rp_inst(struct kretprobe_instance *ri,
 373                                 struct hlist_head *head)
 374 {
 375         /* remove rp inst off the rprobe_inst_table */
 376         hlist_del(&ri->hlist);
 377         if (ri->rp) {
 378                 /* remove rp inst off the used list */
 379                 hlist_del(&ri->uflist);
 380                 /* put rp inst back onto the free list */
 381                 INIT_HLIST_NODE(&ri->uflist);
 382                 hlist_add_head(&ri->uflist, &ri->rp->free_instances);
 383         } else
 384                 /* Unregistering */
 385                 hlist_add_head(&ri->hlist, head);
 386 }
 387
 388 struct hlist_head __kprobes *kretprobe_inst_table_head(struct task_struct *tsk)
 389 {
 390         return &kretprobe_inst_table[hash_ptr(tsk, KPROBE_HASH_BITS)];
 391 }
 392
 393 /*
 394  * This function is called from finish_task_switch when task tk becomes dead,
 395  * so that we can recycle any function-return probe instances associated
 396  * with this task. These left over instances represent probed functions
 397  * that have been called but will never return.
 398  */
 399 void __kprobes kprobe_flush_task(struct task_struct *tk)
 400 {
 401         struct kretprobe_instance *ri;
 402         struct hlist_head *head, empty_rp;
 403         struct hlist_node *node, *tmp;
 404         unsigned long flags = 0;
 405
 406         INIT_HLIST_HEAD(&empty_rp);
 407         spin_lock_irqsave(&kretprobe_lock, flags);
 408         head = kretprobe_inst_table_head(tk);
 409         hlist_for_each_entry_safe(ri, node, tmp, head, hlist) {
 410                 if (ri->task == tk)
 411                         recycle_rp_inst(ri, &empty_rp);
 412         }
 413         spin_unlock_irqrestore(&kretprobe_lock, flags);
 414
 415         hlist_for_each_entry_safe(ri, node, tmp, &empty_rp, hlist) {
 416                 hlist_del(&ri->hlist);
 417                 kfree(ri);
 418         }
 419 }
 420
 421 static inline void free_rp_inst(struct kretprobe *rp)
 422 {
 423         struct kretprobe_instance *ri;
 424         struct hlist_node *pos, *next;
 425
 426         hlist_for_each_entry_safe(ri, pos, next, &rp->free_instances, uflist) {
 427                 hlist_del(&ri->uflist);
 428                 kfree(ri);
 429         }
 430 }
 431
 432 static void __kprobes cleanup_rp_inst(struct kretprobe *rp)
 433 {
 434         unsigned long flags;
 435         struct kretprobe_instance *ri;
 436         struct hlist_node *pos, *next;
 437         /* No race here */
 438         spin_lock_irqsave(&kretprobe_lock, flags);
 439         hlist_for_each_entry_safe(ri, pos, next, &rp->used_instances, uflist) {
 440                 ri->rp = NULL;
 441                 hlist_del(&ri->uflist);
 442         }
 443         spin_unlock_irqrestore(&kretprobe_lock, flags);
 444         free_rp_inst(rp);
 445 }
 446
 447 /*
 448  * Keep all fields in the kprobe consistent
 449  */
 450 static inline void copy_kprobe(struct kprobe *old_p, struct kprobe *p)
 451 {
 452         memcpy(&p->opcode, &old_p->opcode, sizeof(kprobe_opcode_t));
 453         memcpy(&p->ainsn, &old_p->ainsn, sizeof(struct arch_specific_insn));
 454 }
 455
 456 /*
 457 * Add the new probe to old_p->list. Fail if this is the
 458 * second jprobe at the address - two jprobes can't coexist
 459 */
 460 static int __kprobes add_new_kprobe(struct kprobe *old_p, struct kprobe *p)
 461 {
 462         if (p->break_handler) {
 463                 if (old_p->break_handler)
 464                         return -EEXIST;
 465                 list_add_tail_rcu(&p->list, &old_p->list);
 466                 old_p->break_handler = aggr_break_handler;
 467         } else
 468                 list_add_rcu(&p->list, &old_p->list);
 469         if (p->post_handler && !old_p->post_handler)
 470                 old_p->post_handler = aggr_post_handler;
 471         return 0;
 472 }
 473
 474 /*
 475  * Fill in the required fields of the "manager kprobe". Replace the
 476  * earlier kprobe in the hlist with the manager kprobe
 477  */
 478 static inline void add_aggr_kprobe(struct kprobe *ap, struct kprobe *p)
 479 {
 480         copy_kprobe(p, ap);
 481         flush_insn_slot(ap);
 482         ap->addr = p->addr;
 483         ap->pre_handler = aggr_pre_handler;
 484         ap->fault_handler = aggr_fault_handler;
 485         if (p->post_handler)
 486                 ap->post_handler = aggr_post_handler;
 487         if (p->break_handler)
 488                 ap->break_handler = aggr_break_handler;
 489
 490         INIT_LIST_HEAD(&ap->list);
 491         list_add_rcu(&p->list, &ap->list);
 492
 493         hlist_replace_rcu(&p->hlist, &ap->hlist);
 494 }
 495
 496 /*
 497  * This is the second or subsequent kprobe at the address - handle
 498  * the intricacies
 499  */
 500 static int __kprobes register_aggr_kprobe(struct kprobe *old_p,
 501                                           struct kprobe *p)
 502 {
 503         int ret = 0;
 504         struct kprobe *ap;
 505
 506         if (old_p->pre_handler == aggr_pre_handler) {
 507                 copy_kprobe(old_p, p);
 508                 ret = add_new_kprobe(old_p, p);
 509         } else {
 510                 ap = kzalloc(sizeof(struct kprobe), GFP_KERNEL);
 511                 if (!ap)
 512                         return -ENOMEM;
 513                 add_aggr_kprobe(ap, old_p);
 514                 copy_kprobe(ap, p);
 515                 ret = add_new_kprobe(ap, p);
 516         }
 517         return ret;
 518 }
 519
 520 static int __kprobes in_kprobes_functions(unsigned long addr)
 521 {
 522         struct kprobe_blackpoint *kb;
 523
 524         if (addr >= (unsigned long)__kprobes_text_start &&
 525             addr < (unsigned long)__kprobes_text_end)
 526                 return -EINVAL;
 527         /*
 528          * If there exists a kprobe_blacklist, verify and
 529          * fail any probe registration in the prohibited area
 530          */
 531         for (kb = kprobe_blacklist; kb->name != NULL; kb++) {
 532                 if (kb->start_addr) {
 533                         if (addr >= kb->start_addr &&
 534                             addr < (kb->start_addr + kb->range))
 535                                 return -EINVAL;
 536                 }
 537         }
 538         return 0;
 539 }
 540
 541 /*
 542  * If we have a symbol_name argument, look it up and add the offset field
 543  * to it. This way, we can specify a relative address to a symbol.
 544  */
 545 static kprobe_opcode_t __kprobes *kprobe_addr(struct kprobe *p)
 546 {
 547         kprobe_opcode_t *addr = p->addr;
 548         if (p->symbol_name) {
 549                 if (addr)
 550                         return NULL;
 551                 kprobe_lookup_name(p->symbol_name, addr);
 552         }
 553
 554         if (!addr)
 555                 return NULL;
 556         return (kprobe_opcode_t *)(((char *)addr) + p->offset);
 557 }
 558
 559 static int __kprobes __register_kprobe(struct kprobe *p,
 560         unsigned long called_from)
 561 {
 562         int ret = 0;
 563         struct kprobe *old_p;
 564         struct module *probed_mod;
 565         kprobe_opcode_t *addr;
 566
 567         addr = kprobe_addr(p);
 568         if (!addr)
 569                 return -EINVAL;
 570         p->addr = addr;
 571
 572         if (!kernel_text_address((unsigned long) p->addr) ||
 573             in_kprobes_functions((unsigned long) p->addr))
 574                 return -EINVAL;
 575
 576         p->mod_refcounted = 0;
 577
 578         /*
 579          * Check if are we probing a module.
 580          */
 581         probed_mod = module_text_address((unsigned long) p->addr);
 582         if (probed_mod) {
 583                 struct module *calling_mod = module_text_address(called_from);
 584                 /*
 585                  * We must allow modules to probe themself and in this case
 586                  * avoid incrementing the module refcount, so as to allow
 587                  * unloading of self probing modules.
 588                  */
 589                 if (calling_mod && calling_mod != probed_mod) {
 590                         if (unlikely(!try_module_get(probed_mod)))
 591                                 return -EINVAL;
 592                         p->mod_refcounted = 1;
 593                 } else
 594                         probed_mod = NULL;
 595         }
 596
 597         p->nmissed = 0;
 598         INIT_LIST_HEAD(&p->list);
 599         mutex_lock(&kprobe_mutex);
 600         old_p = get_kprobe(p->addr);
 601         if (old_p) {
 602                 ret = register_aggr_kprobe(old_p, p);
 603                 goto out;
 604         }
 605
 606         ret = arch_prepare_kprobe(p);
 607         if (ret)
 608                 goto out;
 609
 610         INIT_HLIST_NODE(&p->hlist);
 611         hlist_add_head_rcu(&p->hlist,
 612                        &kprobe_table[hash_ptr(p->addr, KPROBE_HASH_BITS)]);
 613
 614         if (kprobe_enabled)
 615                 arch_arm_kprobe(p);
 616
 617 out:
 618         mutex_unlock(&kprobe_mutex);
 619
 620         if (ret && probed_mod)
 621                 module_put(probed_mod);
 622         return ret;
 623 }
 624
 625 /*
 626  * Unregister a kprobe without a scheduler synchronization.
 627  */
 628 static int __kprobes __unregister_kprobe_top(struct kprobe *p)
 629 {
 630         struct kprobe *old_p, *list_p;
 631
 632         old_p = get_kprobe(p->addr);
 633         if (unlikely(!old_p))
 634                 return -EINVAL;
 635
 636         if (p != old_p) {
 637                 list_for_each_entry_rcu(list_p, &old_p->list, list)
 638                         if (list_p == p)
 639                         /* kprobe p is a valid probe */
 640                                 goto valid_p;
 641                 return -EINVAL;
 642         }
 643 valid_p:
 644         if (old_p == p ||
 645             (old_p->pre_handler == aggr_pre_handler &&
 646              list_is_singular(&old_p->list))) {
 647                 /*
 648                  * Only probe on the hash list. Disarm only if kprobes are
 649                  * enabled - otherwise, the breakpoint would already have
 650                  * been removed. We save on flushing icache.
 651                  */
 652                 if (kprobe_enabled)
 653                         arch_disarm_kprobe(p);
 654                 hlist_del_rcu(&old_p->hlist);
 655         } else {
 656                 if (p->break_handler)
 657                         old_p->break_handler = NULL;
 658                 if (p->post_handler) {
 659                         list_for_each_entry_rcu(list_p, &old_p->list, list) {
 660                                 if ((list_p != p) && (list_p->post_handler))
 661                                         goto noclean;
 662                         }
 663                         old_p->post_handler = NULL;
 664                 }
 665 noclean:
 666                 list_del_rcu(&p->list);
 667         }
 668         return 0;
 669 }
 670
 671 static void __kprobes __unregister_kprobe_bottom(struct kprobe *p)
 672 {
 673         struct module *mod;
 674         struct kprobe *old_p;
 675
 676         if (p->mod_refcounted) {
 677                 mod = module_text_address((unsigned long)p->addr);
 678                 if (mod)
 679                         module_put(mod);
 680         }
 681
 682         if (list_empty(&p->list) || list_is_singular(&p->list)) {
 683                 if (!list_empty(&p->list)) {
 684                         /* "p" is the last child of an aggr_kprobe */
 685                         old_p = list_entry(p->list.next, struct kprobe, list);
 686                         list_del(&p->list);
 687                         kfree(old_p);
 688                 }
 689                 arch_remove_kprobe(p);
 690         }
 691 }
 692
 693 static int __register_kprobes(struct kprobe **kps, int num,
 694         unsigned long called_from)
 695 {
 696         int i, ret = 0;
 697
 698         if (num <= 0)
 699                 return -EINVAL;
 700         for (i = 0; i < num; i++) {
 701                 ret = __register_kprobe(kps[i], called_from);
 702                 if (ret < 0) {
 703                         if (i > 0)
 704                                 unregister_kprobes(kps, i);
 705                         break;
 706                 }
 707         }
 708         return ret;
 709 }
 710
 711 /*
 712  * Registration and unregistration functions for kprobe.
 713  */
 714 int __kprobes register_kprobe(struct kprobe *p)
 715 {
 716         return __register_kprobes(&p, 1,
 717                                   (unsigned long)__builtin_return_address(0));
 718 }
 719
 720 void __kprobes unregister_kprobe(struct kprobe *p)
 721 {
 722         unregister_kprobes(&p, 1);
 723 }
 724
 725 int __kprobes register_kprobes(struct kprobe **kps, int num)
 726 {
 727         return __register_kprobes(kps, num,
 728                                   (unsigned long)__builtin_return_address(0));
 729 }
 730
 731 void __kprobes unregister_kprobes(struct kprobe **kps, int num)
 732 {
 733         int i;
 734
 735         if (num <= 0)
 736                 return;
 737         mutex_lock(&kprobe_mutex);
 738         for (i = 0; i < num; i++)
 739                 if (__unregister_kprobe_top(kps[i]) < 0)
 740                         kps[i]->addr = NULL;
 741         mutex_unlock(&kprobe_mutex);
 742
 743         synchronize_sched();
 744         for (i = 0; i < num; i++)
 745                 if (kps[i]->addr)
 746                         __unregister_kprobe_bottom(kps[i]);
 747 }
 748
 749 static struct notifier_block kprobe_exceptions_nb = {
 750         .notifier_call = kprobe_exceptions_notify,
 751         .priority = 0x7fffffff /* we need to be notified first */
 752 };
 753
 754 unsigned long __weak arch_deref_entry_point(void *entry)
 755 {
 756         return (unsigned long)entry;
 757 }
 758
 759 static int __register_jprobes(struct jprobe **jps, int num,
 760         unsigned long called_from)
 761 {
 762         struct jprobe *jp;
 763         int ret = 0, i;
 764
 765         if (num <= 0)
 766                 return -EINVAL;
 767         for (i = 0; i < num; i++) {
 768                 unsigned long addr;
 769                 jp = jps[i];
 770                 addr = arch_deref_entry_point(jp->entry);
 771
 772                 if (!kernel_text_address(addr))
 773                         ret = -EINVAL;
 774                 else {
 775                         /* Todo: Verify probepoint is a function entry point */
 776                         jp->kp.pre_handler = setjmp_pre_handler;
 777                         jp->kp.break_handler = longjmp_break_handler;
 778                         ret = __register_kprobe(&jp->kp, called_from);
 779                 }
 780                 if (ret < 0) {
 781                         if (i > 0)
 782                                 unregister_jprobes(jps, i);
 783                         break;
 784                 }
 785         }
 786         return ret;
 787 }
 788
 789 int __kprobes register_jprobe(struct jprobe *jp)
 790 {
 791         return __register_jprobes(&jp, 1,
 792                 (unsigned long)__builtin_return_address(0));
 793 }
 794
 795 void __kprobes unregister_jprobe(struct jprobe *jp)
 796 {
 797         unregister_jprobes(&jp, 1);
 798 }
 799
 800 int __kprobes register_jprobes(struct jprobe **jps, int num)
 801 {
 802         return __register_jprobes(jps, num,
 803                 (unsigned long)__builtin_return_address(0));
 804 }
 805
 806 void __kprobes unregister_jprobes(struct jprobe **jps, int num)
 807 {
 808         int i;
 809
 810         if (num <= 0)
 811                 return;
 812         mutex_lock(&kprobe_mutex);
 813         for (i = 0; i < num; i++)
 814                 if (__unregister_kprobe_top(&jps[i]->kp) < 0)
 815                         jps[i]->kp.addr = NULL;
 816         mutex_unlock(&kprobe_mutex);
 817
 818         synchronize_sched();
 819         for (i = 0; i < num; i++) {
 820                 if (jps[i]->kp.addr)
 821                         __unregister_kprobe_bottom(&jps[i]->kp);
 822         }
 823 }
 824
 825 #ifdef CONFIG_KRETPROBES
 826 /*
 827  * This kprobe pre_handler is registered with every kretprobe. When probe
 828  * hits it will set up the return probe.
 829  */
 830 static int __kprobes pre_handler_kretprobe(struct kprobe *p,
 831                                            struct pt_regs *regs)
 832 {
 833         struct kretprobe *rp = container_of(p, struct kretprobe, kp);
 834         unsigned long flags = 0;
 835
 836         /*TODO: consider to only swap the RA after the last pre_handler fired */
 837         spin_lock_irqsave(&kretprobe_lock, flags);
 838         if (!hlist_empty(&rp->free_instances)) {
 839                 struct kretprobe_instance *ri;
 840
 841                 ri = hlist_entry(rp->free_instances.first,
 842                                  struct kretprobe_instance, uflist);
 843                 ri->rp = rp;
 844                 ri->task = current;
 845
 846                 if (rp->entry_handler && rp->entry_handler(ri, regs)) {
 847                         spin_unlock_irqrestore(&kretprobe_lock, flags);
 848                         return 0;
 849                 }
 850
 851                 arch_prepare_kretprobe(ri, regs);
 852
 853                 /* XXX(hch): why is there no hlist_move_head? */
 854                 hlist_del(&ri->uflist);
 855                 hlist_add_head(&ri->uflist, &ri->rp->used_instances);
 856                 hlist_add_head(&ri->hlist, kretprobe_inst_table_head(ri->task));
 857         } else
 858                 rp->nmissed++;
 859         spin_unlock_irqrestore(&kretprobe_lock, flags);
 860         return 0;
 861 }
 862
 863 static int __kprobes __register_kretprobe(struct kretprobe *rp,
 864                                           unsigned long called_from)
 865 {
 866         int ret = 0;
 867         struct kretprobe_instance *inst;
 868         int i;
 869         void *addr;
 870
 871         if (kretprobe_blacklist_size) {
 872                 addr = kprobe_addr(&rp->kp);
 873                 if (!addr)
 874                         return -EINVAL;
 875
 876                 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
 877                         if (kretprobe_blacklist[i].addr == addr)
 878                                 return -EINVAL;
 879                 }
 880         }
 881
 882         rp->kp.pre_handler = pre_handler_kretprobe;
 883         rp->kp.post_handler = NULL;
 884         rp->kp.fault_handler = NULL;
 885         rp->kp.break_handler = NULL;
 886
 887         /* Pre-allocate memory for max kretprobe instances */
 888         if (rp->maxactive <= 0) {
 889 #ifdef CONFIG_PREEMPT
 890                 rp->maxactive = max(10, 2 * NR_CPUS);
 891 #else
 892                 rp->maxactive = NR_CPUS;
 893 #endif
 894         }
 895         INIT_HLIST_HEAD(&rp->used_instances);
 896         INIT_HLIST_HEAD(&rp->free_instances);
 897         for (i = 0; i < rp->maxactive; i++) {
 898                 inst = kmalloc(sizeof(struct kretprobe_instance) +
 899                                rp->data_size, GFP_KERNEL);
 900                 if (inst == NULL) {
 901                         free_rp_inst(rp);
 902                         return -ENOMEM;
 903                 }
 904                 INIT_HLIST_NODE(&inst->uflist);
 905                 hlist_add_head(&inst->uflist, &rp->free_instances);
 906         }
 907
 908         rp->nmissed = 0;
 909         /* Establish function entry probe point */
 910         ret = __register_kprobe(&rp->kp, called_from);
 911         if (ret != 0)
 912                 free_rp_inst(rp);
 913         return ret;
 914 }
 915
 916 static int __register_kretprobes(struct kretprobe **rps, int num,
 917         unsigned long called_from)
 918 {
 919         int ret = 0, i;
 920
 921         if (num <= 0)
 922                 return -EINVAL;
 923         for (i = 0; i < num; i++) {
 924                 ret = __register_kretprobe(rps[i], called_from);
 925                 if (ret < 0) {
 926                         if (i > 0)
 927                                 unregister_kretprobes(rps, i);
 928                         break;
 929                 }
 930         }
 931         return ret;
 932 }
 933
 934 int __kprobes register_kretprobe(struct kretprobe *rp)
 935 {
 936         return __register_kretprobes(&rp, 1,
 937                         (unsigned long)__builtin_return_address(0));
 938 }
 939
 940 void __kprobes unregister_kretprobe(struct kretprobe *rp)
 941 {
 942         unregister_kretprobes(&rp, 1);
 943 }
 944
 945 int __kprobes register_kretprobes(struct kretprobe **rps, int num)
 946 {
 947         return __register_kretprobes(rps, num,
 948                         (unsigned long)__builtin_return_address(0));
 949 }
 950
 951 void __kprobes unregister_kretprobes(struct kretprobe **rps, 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(&rps[i]->kp) < 0)
 960                         rps[i]->kp.addr = NULL;
 961         mutex_unlock(&kprobe_mutex);
 962
 963         synchronize_sched();
 964         for (i = 0; i < num; i++) {
 965                 if (rps[i]->kp.addr) {
 966                         __unregister_kprobe_bottom(&rps[i]->kp);
 967                         cleanup_rp_inst(rps[i]);
 968                 }
 969         }
 970 }
 971
 972 #else /* CONFIG_KRETPROBES */
 973 int __kprobes register_kretprobe(struct kretprobe *rp)
 974 {
 975         return -ENOSYS;
 976 }
 977
 978 int __kprobes register_kretprobes(struct kretprobe **rps, int num)
 979 {
 980         return -ENOSYS;
 981 }
 982 void __kprobes unregister_kretprobe(struct kretprobe *rp)
 983 {
 984 }
 985
 986 void __kprobes unregister_kretprobes(struct kretprobe **rps, int num)
 987 {
 988 }
 989
 990 static int __kprobes pre_handler_kretprobe(struct kprobe *p,
 991                                            struct pt_regs *regs)
 992 {
 993         return 0;
 994 }
 995
 996 #endif /* CONFIG_KRETPROBES */
 997
 998 static int __init init_kprobes(void)
 999 {
1000         int i, err = 0;
1001         unsigned long offset = 0, size = 0;
1002         char *modname, namebuf[128];
1003         const char *symbol_name;
1004         void *addr;
1005         struct kprobe_blackpoint *kb;
1006
1007         /* FIXME allocate the probe table, currently defined statically */
1008         /* initialize all list heads */
1009         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
1010                 INIT_HLIST_HEAD(&kprobe_table[i]);
1011                 INIT_HLIST_HEAD(&kretprobe_inst_table[i]);
1012         }
1013
1014         /*
1015          * Lookup and populate the kprobe_blacklist.
1016          *
1017          * Unlike the kretprobe blacklist, we'll need to determine
1018          * the range of addresses that belong to the said functions,
1019          * since a kprobe need not necessarily be at the beginning
1020          * of a function.
1021          */
1022         for (kb = kprobe_blacklist; kb->name != NULL; kb++) {
1023                 kprobe_lookup_name(kb->name, addr);
1024                 if (!addr)
1025                         continue;
1026
1027                 kb->start_addr = (unsigned long)addr;
1028                 symbol_name = kallsyms_lookup(kb->start_addr,
1029                                 &size, &offset, &modname, namebuf);
1030                 if (!symbol_name)
1031                         kb->range = 0;
1032                 else
1033                         kb->range = size;
1034         }
1035
1036         if (kretprobe_blacklist_size) {
1037                 /* lookup the function address from its name */
1038                 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
1039                         kprobe_lookup_name(kretprobe_blacklist[i].name,
1040                                            kretprobe_blacklist[i].addr);
1041                         if (!kretprobe_blacklist[i].addr)
1042                                 printk("kretprobe: lookup failed: %s\n",
1043                                        kretprobe_blacklist[i].name);
1044                 }
1045         }
1046
1047         /* By default, kprobes are enabled */
1048         kprobe_enabled = true;
1049
1050         err = arch_init_kprobes();
1051         if (!err)
1052                 err = register_die_notifier(&kprobe_exceptions_nb);
1053
1054         if (!err)
1055                 init_test_probes();
1056         return err;
1057 }
1058
1059 #ifdef CONFIG_DEBUG_FS
1060 static void __kprobes report_probe(struct seq_file *pi, struct kprobe *p,
1061                 const char *sym, int offset,char *modname)
1062 {
1063         char *kprobe_type;
1064
1065         if (p->pre_handler == pre_handler_kretprobe)
1066                 kprobe_type = "r";
1067         else if (p->pre_handler == setjmp_pre_handler)
1068                 kprobe_type = "j";
1069         else
1070                 kprobe_type = "k";
1071         if (sym)
1072                 seq_printf(pi, "%p  %s  %s+0x%x  %s\n", p->addr, kprobe_type,
1073                         sym, offset, (modname ? modname : " "));
1074         else
1075                 seq_printf(pi, "%p  %s  %p\n", p->addr, kprobe_type, p->addr);
1076 }
1077
1078 static void __kprobes *kprobe_seq_start(struct seq_file *f, loff_t *pos)
1079 {
1080         return (*pos < KPROBE_TABLE_SIZE) ? pos : NULL;
1081 }
1082
1083 static void __kprobes *kprobe_seq_next(struct seq_file *f, void *v, loff_t *pos)
1084 {
1085         (*pos)++;
1086         if (*pos >= KPROBE_TABLE_SIZE)
1087                 return NULL;
1088         return pos;
1089 }
1090
1091 static void __kprobes kprobe_seq_stop(struct seq_file *f, void *v)
1092 {
1093         /* Nothing to do */
1094 }
1095
1096 static int __kprobes show_kprobe_addr(struct seq_file *pi, void *v)
1097 {
1098         struct hlist_head *head;
1099         struct hlist_node *node;
1100         struct kprobe *p, *kp;
1101         const char *sym = NULL;
1102         unsigned int i = *(loff_t *) v;
1103         unsigned long offset = 0;
1104         char *modname, namebuf[128];
1105
1106         head = &kprobe_table[i];
1107         preempt_disable();
1108         hlist_for_each_entry_rcu(p, node, head, hlist) {
1109                 sym = kallsyms_lookup((unsigned long)p->addr, NULL,
1110                                         &offset, &modname, namebuf);
1111                 if (p->pre_handler == aggr_pre_handler) {
1112                         list_for_each_entry_rcu(kp, &p->list, list)
1113                                 report_probe(pi, kp, sym, offset, modname);
1114                 } else
1115                         report_probe(pi, p, sym, offset, modname);
1116         }
1117         preempt_enable();
1118         return 0;
1119 }
1120
1121 static struct seq_operations kprobes_seq_ops = {
1122         .start = kprobe_seq_start,
1123         .next  = kprobe_seq_next,
1124         .stop  = kprobe_seq_stop,
1125         .show  = show_kprobe_addr
1126 };
1127
1128 static int __kprobes kprobes_open(struct inode *inode, struct file *filp)
1129 {
1130         return seq_open(filp, &kprobes_seq_ops);
1131 }
1132
1133 static struct file_operations debugfs_kprobes_operations = {
1134         .open           = kprobes_open,
1135         .read           = seq_read,
1136         .llseek         = seq_lseek,
1137         .release        = seq_release,
1138 };
1139
1140 static void __kprobes enable_all_kprobes(void)
1141 {
1142         struct hlist_head *head;
1143         struct hlist_node *node;
1144         struct kprobe *p;
1145         unsigned int i;
1146
1147         mutex_lock(&kprobe_mutex);
1148
1149         /* If kprobes are already enabled, just return */
1150         if (kprobe_enabled)
1151                 goto already_enabled;
1152
1153         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
1154                 head = &kprobe_table[i];
1155                 hlist_for_each_entry_rcu(p, node, head, hlist)
1156                         arch_arm_kprobe(p);
1157         }
1158
1159         kprobe_enabled = true;
1160         printk(KERN_INFO "Kprobes globally enabled\n");
1161
1162 already_enabled:
1163         mutex_unlock(&kprobe_mutex);
1164         return;
1165 }
1166
1167 static void __kprobes disable_all_kprobes(void)
1168 {
1169         struct hlist_head *head;
1170         struct hlist_node *node;
1171         struct kprobe *p;
1172         unsigned int i;
1173
1174         mutex_lock(&kprobe_mutex);
1175
1176         /* If kprobes are already disabled, just return */
1177         if (!kprobe_enabled)
1178                 goto already_disabled;
1179
1180         kprobe_enabled = false;
1181         printk(KERN_INFO "Kprobes globally disabled\n");
1182         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
1183                 head = &kprobe_table[i];
1184                 hlist_for_each_entry_rcu(p, node, head, hlist) {
1185                         if (!arch_trampoline_kprobe(p))
1186                                 arch_disarm_kprobe(p);
1187                 }
1188         }
1189
1190         mutex_unlock(&kprobe_mutex);
1191         /* Allow all currently running kprobes to complete */
1192         synchronize_sched();
1193         return;
1194
1195 already_disabled:
1196         mutex_unlock(&kprobe_mutex);
1197         return;
1198 }
1199
1200 /*
1201  * XXX: The debugfs bool file interface doesn't allow for callbacks
1202  * when the bool state is switched. We can reuse that facility when
1203  * available
1204  */
1205 static ssize_t read_enabled_file_bool(struct file *file,
1206                char __user *user_buf, size_t count, loff_t *ppos)
1207 {
1208         char buf[3];
1209
1210         if (kprobe_enabled)
1211                 buf[0] = '1';
1212         else
1213                 buf[0] = '0';
1214         buf[1] = '\n';
1215         buf[2] = 0x00;
1216         return simple_read_from_buffer(user_buf, count, ppos, buf, 2);
1217 }
1218
1219 static ssize_t write_enabled_file_bool(struct file *file,
1220                const char __user *user_buf, size_t count, loff_t *ppos)
1221 {
1222         char buf[32];
1223         int buf_size;
1224
1225         buf_size = min(count, (sizeof(buf)-1));
1226         if (copy_from_user(buf, user_buf, buf_size))
1227                 return -EFAULT;
1228
1229         switch (buf[0]) {
1230         case 'y':
1231         case 'Y':
1232         case '1':
1233                 enable_all_kprobes();
1234                 break;
1235         case 'n':
1236         case 'N':
1237         case '0':
1238                 disable_all_kprobes();
1239                 break;
1240         }
1241
1242         return count;
1243 }
1244
1245 static struct file_operations fops_kp = {
1246         .read =         read_enabled_file_bool,
1247         .write =        write_enabled_file_bool,
1248 };
1249
1250 static int __kprobes debugfs_kprobe_init(void)
1251 {
1252         struct dentry *dir, *file;
1253         unsigned int value = 1;
1254
1255         dir = debugfs_create_dir("kprobes", NULL);
1256         if (!dir)
1257                 return -ENOMEM;
1258
1259         file = debugfs_create_file("list", 0444, dir, NULL,
1260                                 &debugfs_kprobes_operations);
1261         if (!file) {
1262                 debugfs_remove(dir);
1263                 return -ENOMEM;
1264         }
1265
1266         file = debugfs_create_file("enabled", 0600, dir,
1267                                         &value, &fops_kp);
1268         if (!file) {
1269                 debugfs_remove(dir);
1270                 return -ENOMEM;
1271         }
1272
1273         return 0;
1274 }
1275
1276 late_initcall(debugfs_kprobe_init);
1277 #endif /* CONFIG_DEBUG_FS */
1278
1279 module_init(init_kprobes);
1280
1281 EXPORT_SYMBOL_GPL(register_kprobe);
1282 EXPORT_SYMBOL_GPL(unregister_kprobe);
1283 EXPORT_SYMBOL_GPL(register_kprobes);
1284 EXPORT_SYMBOL_GPL(unregister_kprobes);
1285 EXPORT_SYMBOL_GPL(register_jprobe);
1286 EXPORT_SYMBOL_GPL(unregister_jprobe);
1287 EXPORT_SYMBOL_GPL(register_jprobes);
1288 EXPORT_SYMBOL_GPL(unregister_jprobes);
1289 #ifdef CONFIG_KPROBES
1290 EXPORT_SYMBOL_GPL(jprobe_return);
1291 #endif
1292
1293 #ifdef CONFIG_KPROBES
1294 EXPORT_SYMBOL_GPL(register_kretprobe);
1295 EXPORT_SYMBOL_GPL(unregister_kretprobe);
1296 EXPORT_SYMBOL_GPL(register_kretprobes);
1297 EXPORT_SYMBOL_GPL(unregister_kretprobes);
1298 #endif