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 && i > 0) {
 703                         unregister_kprobes(kps, i);
 704                         break;
 705                 }
 706         }
 707         return ret;
 708 }
 709
 710 /*
 711  * Registration and unregistration functions for kprobe.
 712  */
 713 int __kprobes register_kprobe(struct kprobe *p)
 714 {
 715         return __register_kprobes(&p, 1,
 716                                   (unsigned long)__builtin_return_address(0));
 717 }
 718
 719 void __kprobes unregister_kprobe(struct kprobe *p)
 720 {
 721         unregister_kprobes(&p, 1);
 722 }
 723
 724 int __kprobes register_kprobes(struct kprobe **kps, int num)
 725 {
 726         return __register_kprobes(kps, num,
 727                                   (unsigned long)__builtin_return_address(0));
 728 }
 729
 730 void __kprobes unregister_kprobes(struct kprobe **kps, int num)
 731 {
 732         int i;
 733
 734         if (num <= 0)
 735                 return;
 736         mutex_lock(&kprobe_mutex);
 737         for (i = 0; i < num; i++)
 738                 if (__unregister_kprobe_top(kps[i]) < 0)
 739                         kps[i]->addr = NULL;
 740         mutex_unlock(&kprobe_mutex);
 741
 742         synchronize_sched();
 743         for (i = 0; i < num; i++)
 744                 if (kps[i]->addr)
 745                         __unregister_kprobe_bottom(kps[i]);
 746 }
 747
 748 static struct notifier_block kprobe_exceptions_nb = {
 749         .notifier_call = kprobe_exceptions_notify,
 750         .priority = 0x7fffffff /* we need to be notified first */
 751 };
 752
 753 unsigned long __weak arch_deref_entry_point(void *entry)
 754 {
 755         return (unsigned long)entry;
 756 }
 757
 758 static int __register_jprobes(struct jprobe **jps, int num,
 759         unsigned long called_from)
 760 {
 761         struct jprobe *jp;
 762         int ret = 0, i;
 763
 764         if (num <= 0)
 765                 return -EINVAL;
 766         for (i = 0; i < num; i++) {
 767                 unsigned long addr;
 768                 jp = jps[i];
 769                 addr = arch_deref_entry_point(jp->entry);
 770
 771                 if (!kernel_text_address(addr))
 772                         ret = -EINVAL;
 773                 else {
 774                         /* Todo: Verify probepoint is a function entry point */
 775                         jp->kp.pre_handler = setjmp_pre_handler;
 776                         jp->kp.break_handler = longjmp_break_handler;
 777                         ret = __register_kprobe(&jp->kp, called_from);
 778                 }
 779                 if (ret < 0 && i > 0) {
 780                         unregister_jprobes(jps, i);
 781                         break;
 782                 }
 783         }
 784         return ret;
 785 }
 786
 787 int __kprobes register_jprobe(struct jprobe *jp)
 788 {
 789         return __register_jprobes(&jp, 1,
 790                 (unsigned long)__builtin_return_address(0));
 791 }
 792
 793 void __kprobes unregister_jprobe(struct jprobe *jp)
 794 {
 795         unregister_jprobes(&jp, 1);
 796 }
 797
 798 int __kprobes register_jprobes(struct jprobe **jps, int num)
 799 {
 800         return __register_jprobes(jps, num,
 801                 (unsigned long)__builtin_return_address(0));
 802 }
 803
 804 void __kprobes unregister_jprobes(struct jprobe **jps, int num)
 805 {
 806         int i;
 807
 808         if (num <= 0)
 809                 return;
 810         mutex_lock(&kprobe_mutex);
 811         for (i = 0; i < num; i++)
 812                 if (__unregister_kprobe_top(&jps[i]->kp) < 0)
 813                         jps[i]->kp.addr = NULL;
 814         mutex_unlock(&kprobe_mutex);
 815
 816         synchronize_sched();
 817         for (i = 0; i < num; i++) {
 818                 if (jps[i]->kp.addr)
 819                         __unregister_kprobe_bottom(&jps[i]->kp);
 820         }
 821 }
 822
 823 #ifdef CONFIG_KRETPROBES
 824 /*
 825  * This kprobe pre_handler is registered with every kretprobe. When probe
 826  * hits it will set up the return probe.
 827  */
 828 static int __kprobes pre_handler_kretprobe(struct kprobe *p,
 829                                            struct pt_regs *regs)
 830 {
 831         struct kretprobe *rp = container_of(p, struct kretprobe, kp);
 832         unsigned long flags = 0;
 833
 834         /*TODO: consider to only swap the RA after the last pre_handler fired */
 835         spin_lock_irqsave(&kretprobe_lock, flags);
 836         if (!hlist_empty(&rp->free_instances)) {
 837                 struct kretprobe_instance *ri;
 838
 839                 ri = hlist_entry(rp->free_instances.first,
 840                                  struct kretprobe_instance, uflist);
 841                 ri->rp = rp;
 842                 ri->task = current;
 843
 844                 if (rp->entry_handler && rp->entry_handler(ri, regs)) {
 845                         spin_unlock_irqrestore(&kretprobe_lock, flags);
 846                         return 0;
 847                 }
 848
 849                 arch_prepare_kretprobe(ri, regs);
 850
 851                 /* XXX(hch): why is there no hlist_move_head? */
 852                 hlist_del(&ri->uflist);
 853                 hlist_add_head(&ri->uflist, &ri->rp->used_instances);
 854                 hlist_add_head(&ri->hlist, kretprobe_inst_table_head(ri->task));
 855         } else
 856                 rp->nmissed++;
 857         spin_unlock_irqrestore(&kretprobe_lock, flags);
 858         return 0;
 859 }
 860
 861 static int __kprobes __register_kretprobe(struct kretprobe *rp,
 862                                           unsigned long called_from)
 863 {
 864         int ret = 0;
 865         struct kretprobe_instance *inst;
 866         int i;
 867         void *addr;
 868
 869         if (kretprobe_blacklist_size) {
 870                 addr = kprobe_addr(&rp->kp);
 871                 if (!addr)
 872                         return -EINVAL;
 873
 874                 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
 875                         if (kretprobe_blacklist[i].addr == addr)
 876                                 return -EINVAL;
 877                 }
 878         }
 879
 880         rp->kp.pre_handler = pre_handler_kretprobe;
 881         rp->kp.post_handler = NULL;
 882         rp->kp.fault_handler = NULL;
 883         rp->kp.break_handler = NULL;
 884
 885         /* Pre-allocate memory for max kretprobe instances */
 886         if (rp->maxactive <= 0) {
 887 #ifdef CONFIG_PREEMPT
 888                 rp->maxactive = max(10, 2 * NR_CPUS);
 889 #else
 890                 rp->maxactive = NR_CPUS;
 891 #endif
 892         }
 893         INIT_HLIST_HEAD(&rp->used_instances);
 894         INIT_HLIST_HEAD(&rp->free_instances);
 895         for (i = 0; i < rp->maxactive; i++) {
 896                 inst = kmalloc(sizeof(struct kretprobe_instance) +
 897                                rp->data_size, GFP_KERNEL);
 898                 if (inst == NULL) {
 899                         free_rp_inst(rp);
 900                         return -ENOMEM;
 901                 }
 902                 INIT_HLIST_NODE(&inst->uflist);
 903                 hlist_add_head(&inst->uflist, &rp->free_instances);
 904         }
 905
 906         rp->nmissed = 0;
 907         /* Establish function entry probe point */
 908         ret = __register_kprobe(&rp->kp, called_from);
 909         if (ret != 0)
 910                 free_rp_inst(rp);
 911         return ret;
 912 }
 913
 914 static int __register_kretprobes(struct kretprobe **rps, int num,
 915         unsigned long called_from)
 916 {
 917         int ret = 0, i;
 918
 919         if (num <= 0)
 920                 return -EINVAL;
 921         for (i = 0; i < num; i++) {
 922                 ret = __register_kretprobe(rps[i], called_from);
 923                 if (ret < 0 && i > 0) {
 924                         unregister_kretprobes(rps, i);
 925                         break;
 926                 }
 927         }
 928         return ret;
 929 }
 930
 931 int __kprobes register_kretprobe(struct kretprobe *rp)
 932 {
 933         return __register_kretprobes(&rp, 1,
 934                         (unsigned long)__builtin_return_address(0));
 935 }
 936
 937 void __kprobes unregister_kretprobe(struct kretprobe *rp)
 938 {
 939         unregister_kretprobes(&rp, 1);
 940 }
 941
 942 int __kprobes register_kretprobes(struct kretprobe **rps, int num)
 943 {
 944         return __register_kretprobes(rps, num,
 945                         (unsigned long)__builtin_return_address(0));
 946 }
 947
 948 void __kprobes unregister_kretprobes(struct kretprobe **rps, int num)
 949 {
 950         int i;
 951
 952         if (num <= 0)
 953                 return;
 954         mutex_lock(&kprobe_mutex);
 955         for (i = 0; i < num; i++)
 956                 if (__unregister_kprobe_top(&rps[i]->kp) < 0)
 957                         rps[i]->kp.addr = NULL;
 958         mutex_unlock(&kprobe_mutex);
 959
 960         synchronize_sched();
 961         for (i = 0; i < num; i++) {
 962                 if (rps[i]->kp.addr) {
 963                         __unregister_kprobe_bottom(&rps[i]->kp);
 964                         cleanup_rp_inst(rps[i]);
 965                 }
 966         }
 967 }
 968
 969 #else /* CONFIG_KRETPROBES */
 970 int __kprobes register_kretprobe(struct kretprobe *rp)
 971 {
 972         return -ENOSYS;
 973 }
 974
 975 int __kprobes register_kretprobes(struct kretprobe **rps, int num)
 976 {
 977         return -ENOSYS;
 978 }
 979 void __kprobes unregister_kretprobe(struct kretprobe *rp)
 980 {
 981 }
 982
 983 void __kprobes unregister_kretprobes(struct kretprobe **rps, int num)
 984 {
 985 }
 986
 987 static int __kprobes pre_handler_kretprobe(struct kprobe *p,
 988                                            struct pt_regs *regs)
 989 {
 990         return 0;
 991 }
 992
 993 #endif /* CONFIG_KRETPROBES */
 994
 995 static int __init init_kprobes(void)
 996 {
 997         int i, err = 0;
 998         unsigned long offset = 0, size = 0;
 999         char *modname, namebuf[128];
1000         const char *symbol_name;
1001         void *addr;
1002         struct kprobe_blackpoint *kb;
1003
1004         /* FIXME allocate the probe table, currently defined statically */
1005         /* initialize all list heads */
1006         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
1007                 INIT_HLIST_HEAD(&kprobe_table[i]);
1008                 INIT_HLIST_HEAD(&kretprobe_inst_table[i]);
1009         }
1010
1011         /*
1012          * Lookup and populate the kprobe_blacklist.
1013          *
1014          * Unlike the kretprobe blacklist, we'll need to determine
1015          * the range of addresses that belong to the said functions,
1016          * since a kprobe need not necessarily be at the beginning
1017          * of a function.
1018          */
1019         for (kb = kprobe_blacklist; kb->name != NULL; kb++) {
1020                 kprobe_lookup_name(kb->name, addr);
1021                 if (!addr)
1022                         continue;
1023
1024                 kb->start_addr = (unsigned long)addr;
1025                 symbol_name = kallsyms_lookup(kb->start_addr,
1026                                 &size, &offset, &modname, namebuf);
1027                 if (!symbol_name)
1028                         kb->range = 0;
1029                 else
1030                         kb->range = size;
1031         }
1032
1033         if (kretprobe_blacklist_size) {
1034                 /* lookup the function address from its name */
1035                 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
1036                         kprobe_lookup_name(kretprobe_blacklist[i].name,
1037                                            kretprobe_blacklist[i].addr);
1038                         if (!kretprobe_blacklist[i].addr)
1039                                 printk("kretprobe: lookup failed: %s\n",
1040                                        kretprobe_blacklist[i].name);
1041                 }
1042         }
1043
1044         /* By default, kprobes are enabled */
1045         kprobe_enabled = true;
1046
1047         err = arch_init_kprobes();
1048         if (!err)
1049                 err = register_die_notifier(&kprobe_exceptions_nb);
1050
1051         if (!err)
1052                 init_test_probes();
1053         return err;
1054 }
1055
1056 #ifdef CONFIG_DEBUG_FS
1057 static void __kprobes report_probe(struct seq_file *pi, struct kprobe *p,
1058                 const char *sym, int offset,char *modname)
1059 {
1060         char *kprobe_type;
1061
1062         if (p->pre_handler == pre_handler_kretprobe)
1063                 kprobe_type = "r";
1064         else if (p->pre_handler == setjmp_pre_handler)
1065                 kprobe_type = "j";
1066         else
1067                 kprobe_type = "k";
1068         if (sym)
1069                 seq_printf(pi, "%p  %s  %s+0x%x  %s\n", p->addr, kprobe_type,
1070                         sym, offset, (modname ? modname : " "));
1071         else
1072                 seq_printf(pi, "%p  %s  %p\n", p->addr, kprobe_type, p->addr);
1073 }
1074
1075 static void __kprobes *kprobe_seq_start(struct seq_file *f, loff_t *pos)
1076 {
1077         return (*pos < KPROBE_TABLE_SIZE) ? pos : NULL;
1078 }
1079
1080 static void __kprobes *kprobe_seq_next(struct seq_file *f, void *v, loff_t *pos)
1081 {
1082         (*pos)++;
1083         if (*pos >= KPROBE_TABLE_SIZE)
1084                 return NULL;
1085         return pos;
1086 }
1087
1088 static void __kprobes kprobe_seq_stop(struct seq_file *f, void *v)
1089 {
1090         /* Nothing to do */
1091 }
1092
1093 static int __kprobes show_kprobe_addr(struct seq_file *pi, void *v)
1094 {
1095         struct hlist_head *head;
1096         struct hlist_node *node;
1097         struct kprobe *p, *kp;
1098         const char *sym = NULL;
1099         unsigned int i = *(loff_t *) v;
1100         unsigned long offset = 0;
1101         char *modname, namebuf[128];
1102
1103         head = &kprobe_table[i];
1104         preempt_disable();
1105         hlist_for_each_entry_rcu(p, node, head, hlist) {
1106                 sym = kallsyms_lookup((unsigned long)p->addr, NULL,
1107                                         &offset, &modname, namebuf);
1108                 if (p->pre_handler == aggr_pre_handler) {
1109                         list_for_each_entry_rcu(kp, &p->list, list)
1110                                 report_probe(pi, kp, sym, offset, modname);
1111                 } else
1112                         report_probe(pi, p, sym, offset, modname);
1113         }
1114         preempt_enable();
1115         return 0;
1116 }
1117
1118 static struct seq_operations kprobes_seq_ops = {
1119         .start = kprobe_seq_start,
1120         .next  = kprobe_seq_next,
1121         .stop  = kprobe_seq_stop,
1122         .show  = show_kprobe_addr
1123 };
1124
1125 static int __kprobes kprobes_open(struct inode *inode, struct file *filp)
1126 {
1127         return seq_open(filp, &kprobes_seq_ops);
1128 }
1129
1130 static struct file_operations debugfs_kprobes_operations = {
1131         .open           = kprobes_open,
1132         .read           = seq_read,
1133         .llseek         = seq_lseek,
1134         .release        = seq_release,
1135 };
1136
1137 static void __kprobes enable_all_kprobes(void)
1138 {
1139         struct hlist_head *head;
1140         struct hlist_node *node;
1141         struct kprobe *p;
1142         unsigned int i;
1143
1144         mutex_lock(&kprobe_mutex);
1145
1146         /* If kprobes are already enabled, just return */
1147         if (kprobe_enabled)
1148                 goto already_enabled;
1149
1150         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
1151                 head = &kprobe_table[i];
1152                 hlist_for_each_entry_rcu(p, node, head, hlist)
1153                         arch_arm_kprobe(p);
1154         }
1155
1156         kprobe_enabled = true;
1157         printk(KERN_INFO "Kprobes globally enabled\n");
1158
1159 already_enabled:
1160         mutex_unlock(&kprobe_mutex);
1161         return;
1162 }
1163
1164 static void __kprobes disable_all_kprobes(void)
1165 {
1166         struct hlist_head *head;
1167         struct hlist_node *node;
1168         struct kprobe *p;
1169         unsigned int i;
1170
1171         mutex_lock(&kprobe_mutex);
1172
1173         /* If kprobes are already disabled, just return */
1174         if (!kprobe_enabled)
1175                 goto already_disabled;
1176
1177         kprobe_enabled = false;
1178         printk(KERN_INFO "Kprobes globally disabled\n");
1179         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
1180                 head = &kprobe_table[i];
1181                 hlist_for_each_entry_rcu(p, node, head, hlist) {
1182                         if (!arch_trampoline_kprobe(p))
1183                                 arch_disarm_kprobe(p);
1184                 }
1185         }
1186
1187         mutex_unlock(&kprobe_mutex);
1188         /* Allow all currently running kprobes to complete */
1189         synchronize_sched();
1190         return;
1191
1192 already_disabled:
1193         mutex_unlock(&kprobe_mutex);
1194         return;
1195 }
1196
1197 /*
1198  * XXX: The debugfs bool file interface doesn't allow for callbacks
1199  * when the bool state is switched. We can reuse that facility when
1200  * available
1201  */
1202 static ssize_t read_enabled_file_bool(struct file *file,
1203                char __user *user_buf, size_t count, loff_t *ppos)
1204 {
1205         char buf[3];
1206
1207         if (kprobe_enabled)
1208                 buf[0] = '1';
1209         else
1210                 buf[0] = '0';
1211         buf[1] = '\n';
1212         buf[2] = 0x00;
1213         return simple_read_from_buffer(user_buf, count, ppos, buf, 2);
1214 }
1215
1216 static ssize_t write_enabled_file_bool(struct file *file,
1217                const char __user *user_buf, size_t count, loff_t *ppos)
1218 {
1219         char buf[32];
1220         int buf_size;
1221
1222         buf_size = min(count, (sizeof(buf)-1));
1223         if (copy_from_user(buf, user_buf, buf_size))
1224                 return -EFAULT;
1225
1226         switch (buf[0]) {
1227         case 'y':
1228         case 'Y':
1229         case '1':
1230                 enable_all_kprobes();
1231                 break;
1232         case 'n':
1233         case 'N':
1234         case '0':
1235                 disable_all_kprobes();
1236                 break;
1237         }
1238
1239         return count;
1240 }
1241
1242 static struct file_operations fops_kp = {
1243         .read =         read_enabled_file_bool,
1244         .write =        write_enabled_file_bool,
1245 };
1246
1247 static int __kprobes debugfs_kprobe_init(void)
1248 {
1249         struct dentry *dir, *file;
1250         unsigned int value = 1;
1251
1252         dir = debugfs_create_dir("kprobes", NULL);
1253         if (!dir)
1254                 return -ENOMEM;
1255
1256         file = debugfs_create_file("list", 0444, dir, NULL,
1257                                 &debugfs_kprobes_operations);
1258         if (!file) {
1259                 debugfs_remove(dir);
1260                 return -ENOMEM;
1261         }
1262
1263         file = debugfs_create_file("enabled", 0600, dir,
1264                                         &value, &fops_kp);
1265         if (!file) {
1266                 debugfs_remove(dir);
1267                 return -ENOMEM;
1268         }
1269
1270         return 0;
1271 }
1272
1273 late_initcall(debugfs_kprobe_init);
1274 #endif /* CONFIG_DEBUG_FS */
1275
1276 module_init(init_kprobes);
1277
1278 EXPORT_SYMBOL_GPL(register_kprobe);
1279 EXPORT_SYMBOL_GPL(unregister_kprobe);
1280 EXPORT_SYMBOL_GPL(register_kprobes);
1281 EXPORT_SYMBOL_GPL(unregister_kprobes);
1282 EXPORT_SYMBOL_GPL(register_jprobe);
1283 EXPORT_SYMBOL_GPL(unregister_jprobe);
1284 EXPORT_SYMBOL_GPL(register_jprobes);
1285 EXPORT_SYMBOL_GPL(unregister_jprobes);
1286 #ifdef CONFIG_KPROBES
1287 EXPORT_SYMBOL_GPL(jprobe_return);
1288 #endif
1289
1290 #ifdef CONFIG_KPROBES
1291 EXPORT_SYMBOL_GPL(register_kretprobe);
1292 EXPORT_SYMBOL_GPL(unregister_kretprobe);
1293 EXPORT_SYMBOL_GPL(register_kretprobes);
1294 EXPORT_SYMBOL_GPL(unregister_kretprobes);
1295 #endif