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