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/sysctl.h>
  46 #include <linux/kdebug.h>
  47 #include <linux/memory.h>
  48 #include <linux/ftrace.h>
  49 #include <linux/cpu.h>
  50 #include <linux/jump_label.h>
  51
  52 #include <asm-generic/sections.h>
  53 #include <asm/cacheflush.h>
  54 #include <asm/errno.h>
  55 #include <asm/uaccess.h>
  56
  57 #define KPROBE_HASH_BITS 6
  58 #define KPROBE_TABLE_SIZE (1 << KPROBE_HASH_BITS)
  59
  60
  61 /*
  62  * Some oddball architectures like 64bit powerpc have function descriptors
  63  * so this must be overridable.
  64  */
  65 #ifndef kprobe_lookup_name
  66 #define kprobe_lookup_name(name, addr) \
  67         addr = ((kprobe_opcode_t *)(kallsyms_lookup_name(name)))
  68 #endif
  69
  70 static int kprobes_initialized;
  71 static struct hlist_head kprobe_table[KPROBE_TABLE_SIZE];
  72 static struct hlist_head kretprobe_inst_table[KPROBE_TABLE_SIZE];
  73
  74 /* NOTE: change this value only with kprobe_mutex held */
  75 static bool kprobes_all_disarmed;
  76
  77 /* This protects kprobe_table and optimizing_list */
  78 static DEFINE_MUTEX(kprobe_mutex);
  79 static DEFINE_PER_CPU(struct kprobe *, kprobe_instance) = NULL;
  80 static struct {
  81         spinlock_t lock ____cacheline_aligned_in_smp;
  82 } kretprobe_table_locks[KPROBE_TABLE_SIZE];
  83
  84 static spinlock_t *kretprobe_table_lock_ptr(unsigned long hash)
  85 {
  86         return &(kretprobe_table_locks[hash].lock);
  87 }
  88
  89 /*
  90  * Normally, functions that we'd want to prohibit kprobes in, are marked
  91  * __kprobes. But, there are cases where such functions already belong to
  92  * a different section (__sched for preempt_schedule)
  93  *
  94  * For such cases, we now have a blacklist
  95  */
  96 static struct kprobe_blackpoint kprobe_blacklist[] = {
  97         {"preempt_schedule",},
  98         {"native_get_debugreg",},
  99         {"irq_entries_start",},
 100         {"common_interrupt",},
 101         {"mcount",},    /* mcount can be called from everywhere */
 102         {NULL}    /* Terminator */
 103 };
 104
 105 #ifdef __ARCH_WANT_KPROBES_INSN_SLOT
 106 /*
 107  * kprobe->ainsn.insn points to the copy of the instruction to be
 108  * single-stepped. x86_64, POWER4 and above have no-exec support and
 109  * stepping on the instruction on a vmalloced/kmalloced/data page
 110  * is a recipe for disaster
 111  */
 112 struct kprobe_insn_page {
 113         struct list_head list;
 114         kprobe_opcode_t *insns;         /* Page of instruction slots */
 115         int nused;
 116         int ngarbage;
 117         char slot_used[];
 118 };
 119
 120 #define KPROBE_INSN_PAGE_SIZE(slots)                    \
 121         (offsetof(struct kprobe_insn_page, slot_used) + \
 122          (sizeof(char) * (slots)))
 123
 124 struct kprobe_insn_cache {
 125         struct list_head pages; /* list of kprobe_insn_page */
 126         size_t insn_size;       /* size of instruction slot */
 127         int nr_garbage;
 128 };
 129
 130 static int slots_per_page(struct kprobe_insn_cache *c)
 131 {
 132         return PAGE_SIZE/(c->insn_size * sizeof(kprobe_opcode_t));
 133 }
 134
 135 enum kprobe_slot_state {
 136         SLOT_CLEAN = 0,
 137         SLOT_DIRTY = 1,
 138         SLOT_USED = 2,
 139 };
 140
 141 static DEFINE_MUTEX(kprobe_insn_mutex); /* Protects kprobe_insn_slots */
 142 static struct kprobe_insn_cache kprobe_insn_slots = {
 143         .pages = LIST_HEAD_INIT(kprobe_insn_slots.pages),
 144         .insn_size = MAX_INSN_SIZE,
 145         .nr_garbage = 0,
 146 };
 147 static int __kprobes collect_garbage_slots(struct kprobe_insn_cache *c);
 148
 149 /**
 150  * __get_insn_slot() - Find a slot on an executable page for an instruction.
 151  * We allocate an executable page if there's no room on existing ones.
 152  */
 153 static kprobe_opcode_t __kprobes *__get_insn_slot(struct kprobe_insn_cache *c)
 154 {
 155         struct kprobe_insn_page *kip;
 156
 157  retry:
 158         list_for_each_entry(kip, &c->pages, list) {
 159                 if (kip->nused < slots_per_page(c)) {
 160                         int i;
 161                         for (i = 0; i < slots_per_page(c); i++) {
 162                                 if (kip->slot_used[i] == SLOT_CLEAN) {
 163                                         kip->slot_used[i] = SLOT_USED;
 164                                         kip->nused++;
 165                                         return kip->insns + (i * c->insn_size);
 166                                 }
 167                         }
 168                         /* kip->nused is broken. Fix it. */
 169                         kip->nused = slots_per_page(c);
 170                         WARN_ON(1);
 171                 }
 172         }
 173
 174         /* If there are any garbage slots, collect it and try again. */
 175         if (c->nr_garbage && collect_garbage_slots(c) == 0)
 176                 goto retry;
 177
 178         /* All out of space.  Need to allocate a new page. */
 179         kip = kmalloc(KPROBE_INSN_PAGE_SIZE(slots_per_page(c)), GFP_KERNEL);
 180         if (!kip)
 181                 return NULL;
 182
 183         /*
 184          * Use module_alloc so this page is within +/- 2GB of where the
 185          * kernel image and loaded module images reside. This is required
 186          * so x86_64 can correctly handle the %rip-relative fixups.
 187          */
 188         kip->insns = module_alloc(PAGE_SIZE);
 189         if (!kip->insns) {
 190                 kfree(kip);
 191                 return NULL;
 192         }
 193         INIT_LIST_HEAD(&kip->list);
 194         memset(kip->slot_used, SLOT_CLEAN, slots_per_page(c));
 195         kip->slot_used[0] = SLOT_USED;
 196         kip->nused = 1;
 197         kip->ngarbage = 0;
 198         list_add(&kip->list, &c->pages);
 199         return kip->insns;
 200 }
 201
 202
 203 kprobe_opcode_t __kprobes *get_insn_slot(void)
 204 {
 205         kprobe_opcode_t *ret = NULL;
 206
 207         mutex_lock(&kprobe_insn_mutex);
 208         ret = __get_insn_slot(&kprobe_insn_slots);
 209         mutex_unlock(&kprobe_insn_mutex);
 210
 211         return ret;
 212 }
 213
 214 /* Return 1 if all garbages are collected, otherwise 0. */
 215 static int __kprobes collect_one_slot(struct kprobe_insn_page *kip, int idx)
 216 {
 217         kip->slot_used[idx] = SLOT_CLEAN;
 218         kip->nused--;
 219         if (kip->nused == 0) {
 220                 /*
 221                  * Page is no longer in use.  Free it unless
 222                  * it's the last one.  We keep the last one
 223                  * so as not to have to set it up again the
 224                  * next time somebody inserts a probe.
 225                  */
 226                 if (!list_is_singular(&kip->list)) {
 227                         list_del(&kip->list);
 228                         module_free(NULL, kip->insns);
 229                         kfree(kip);
 230                 }
 231                 return 1;
 232         }
 233         return 0;
 234 }
 235
 236 static int __kprobes collect_garbage_slots(struct kprobe_insn_cache *c)
 237 {
 238         struct kprobe_insn_page *kip, *next;
 239
 240         /* Ensure no-one is interrupted on the garbages */
 241         synchronize_sched();
 242
 243         list_for_each_entry_safe(kip, next, &c->pages, list) {
 244                 int i;
 245                 if (kip->ngarbage == 0)
 246                         continue;
 247                 kip->ngarbage = 0;      /* we will collect all garbages */
 248                 for (i = 0; i < slots_per_page(c); i++) {
 249                         if (kip->slot_used[i] == SLOT_DIRTY &&
 250                             collect_one_slot(kip, i))
 251                                 break;
 252                 }
 253         }
 254         c->nr_garbage = 0;
 255         return 0;
 256 }
 257
 258 static void __kprobes __free_insn_slot(struct kprobe_insn_cache *c,
 259                                        kprobe_opcode_t *slot, int dirty)
 260 {
 261         struct kprobe_insn_page *kip;
 262
 263         list_for_each_entry(kip, &c->pages, list) {
 264                 long idx = ((long)slot - (long)kip->insns) /
 265                                 (c->insn_size * sizeof(kprobe_opcode_t));
 266                 if (idx >= 0 && idx < slots_per_page(c)) {
 267                         WARN_ON(kip->slot_used[idx] != SLOT_USED);
 268                         if (dirty) {
 269                                 kip->slot_used[idx] = SLOT_DIRTY;
 270                                 kip->ngarbage++;
 271                                 if (++c->nr_garbage > slots_per_page(c))
 272                                         collect_garbage_slots(c);
 273                         } else
 274                                 collect_one_slot(kip, idx);
 275                         return;
 276                 }
 277         }
 278         /* Could not free this slot. */
 279         WARN_ON(1);
 280 }
 281
 282 void __kprobes free_insn_slot(kprobe_opcode_t * slot, int dirty)
 283 {
 284         mutex_lock(&kprobe_insn_mutex);
 285         __free_insn_slot(&kprobe_insn_slots, slot, dirty);
 286         mutex_unlock(&kprobe_insn_mutex);
 287 }
 288 #ifdef CONFIG_OPTPROBES
 289 /* For optimized_kprobe buffer */
 290 static DEFINE_MUTEX(kprobe_optinsn_mutex); /* Protects kprobe_optinsn_slots */
 291 static struct kprobe_insn_cache kprobe_optinsn_slots = {
 292         .pages = LIST_HEAD_INIT(kprobe_optinsn_slots.pages),
 293         /* .insn_size is initialized later */
 294         .nr_garbage = 0,
 295 };
 296 /* Get a slot for optimized_kprobe buffer */
 297 kprobe_opcode_t __kprobes *get_optinsn_slot(void)
 298 {
 299         kprobe_opcode_t *ret = NULL;
 300
 301         mutex_lock(&kprobe_optinsn_mutex);
 302         ret = __get_insn_slot(&kprobe_optinsn_slots);
 303         mutex_unlock(&kprobe_optinsn_mutex);
 304
 305         return ret;
 306 }
 307
 308 void __kprobes free_optinsn_slot(kprobe_opcode_t * slot, int dirty)
 309 {
 310         mutex_lock(&kprobe_optinsn_mutex);
 311         __free_insn_slot(&kprobe_optinsn_slots, slot, dirty);
 312         mutex_unlock(&kprobe_optinsn_mutex);
 313 }
 314 #endif
 315 #endif
 316
 317 /* We have preemption disabled.. so it is safe to use __ versions */
 318 static inline void set_kprobe_instance(struct kprobe *kp)
 319 {
 320         __get_cpu_var(kprobe_instance) = kp;
 321 }
 322
 323 static inline void reset_kprobe_instance(void)
 324 {
 325         __get_cpu_var(kprobe_instance) = NULL;
 326 }
 327
 328 /*
 329  * This routine is called either:
 330  *      - under the kprobe_mutex - during kprobe_[un]register()
 331  *                              OR
 332  *      - with preemption disabled - from arch/xxx/kernel/kprobes.c
 333  */
 334 struct kprobe __kprobes *get_kprobe(void *addr)
 335 {
 336         struct hlist_head *head;
 337         struct hlist_node *node;
 338         struct kprobe *p;
 339
 340         head = &kprobe_table[hash_ptr(addr, KPROBE_HASH_BITS)];
 341         hlist_for_each_entry_rcu(p, node, head, hlist) {
 342                 if (p->addr == addr)
 343                         return p;
 344         }
 345
 346         return NULL;
 347 }
 348
 349 static int __kprobes aggr_pre_handler(struct kprobe *p, struct pt_regs *regs);
 350
 351 /* Return true if the kprobe is an aggregator */
 352 static inline int kprobe_aggrprobe(struct kprobe *p)
 353 {
 354         return p->pre_handler == aggr_pre_handler;
 355 }
 356
 357 /*
 358  * Keep all fields in the kprobe consistent
 359  */
 360 static inline void copy_kprobe(struct kprobe *old_p, struct kprobe *p)
 361 {
 362         memcpy(&p->opcode, &old_p->opcode, sizeof(kprobe_opcode_t));
 363         memcpy(&p->ainsn, &old_p->ainsn, sizeof(struct arch_specific_insn));
 364 }
 365
 366 #ifdef CONFIG_OPTPROBES
 367 /* NOTE: change this value only with kprobe_mutex held */
 368 static bool kprobes_allow_optimization;
 369
 370 /*
 371  * Call all pre_handler on the list, but ignores its return value.
 372  * This must be called from arch-dep optimized caller.
 373  */
 374 void __kprobes opt_pre_handler(struct kprobe *p, struct pt_regs *regs)
 375 {
 376         struct kprobe *kp;
 377
 378         list_for_each_entry_rcu(kp, &p->list, list) {
 379                 if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
 380                         set_kprobe_instance(kp);
 381                         kp->pre_handler(kp, regs);
 382                 }
 383                 reset_kprobe_instance();
 384         }
 385 }
 386
 387 /* Return true(!0) if the kprobe is ready for optimization. */
 388 static inline int kprobe_optready(struct kprobe *p)
 389 {
 390         struct optimized_kprobe *op;
 391
 392         if (kprobe_aggrprobe(p)) {
 393                 op = container_of(p, struct optimized_kprobe, kp);
 394                 return arch_prepared_optinsn(&op->optinsn);
 395         }
 396
 397         return 0;
 398 }
 399
 400 /*
 401  * Return an optimized kprobe whose optimizing code replaces
 402  * instructions including addr (exclude breakpoint).
 403  */
 404 static struct kprobe *__kprobes get_optimized_kprobe(unsigned long addr)
 405 {
 406         int i;
 407         struct kprobe *p = NULL;
 408         struct optimized_kprobe *op;
 409
 410         /* Don't check i == 0, since that is a breakpoint case. */
 411         for (i = 1; !p && i < MAX_OPTIMIZED_LENGTH; i++)
 412                 p = get_kprobe((void *)(addr - i));
 413
 414         if (p && kprobe_optready(p)) {
 415                 op = container_of(p, struct optimized_kprobe, kp);
 416                 if (arch_within_optimized_kprobe(op, addr))
 417                         return p;
 418         }
 419
 420         return NULL;
 421 }
 422
 423 /* Optimization staging list, protected by kprobe_mutex */
 424 static LIST_HEAD(optimizing_list);
 425
 426 static void kprobe_optimizer(struct work_struct *work);
 427 static DECLARE_DELAYED_WORK(optimizing_work, kprobe_optimizer);
 428 #define OPTIMIZE_DELAY 5
 429
 430 /* Kprobe jump optimizer */
 431 static __kprobes void kprobe_optimizer(struct work_struct *work)
 432 {
 433         struct optimized_kprobe *op, *tmp;
 434
 435         /* Lock modules while optimizing kprobes */
 436         mutex_lock(&module_mutex);
 437         mutex_lock(&kprobe_mutex);
 438         if (kprobes_all_disarmed || !kprobes_allow_optimization)
 439                 goto end;
 440
 441         /*
 442          * Wait for quiesence period to ensure all running interrupts
 443          * are done. Because optprobe may modify multiple instructions
 444          * there is a chance that Nth instruction is interrupted. In that
 445          * case, running interrupt can return to 2nd-Nth byte of jump
 446          * instruction. This wait is for avoiding it.
 447          */
 448         synchronize_sched();
 449
 450         /*
 451          * The optimization/unoptimization refers online_cpus via
 452          * stop_machine() and cpu-hotplug modifies online_cpus.
 453          * And same time, text_mutex will be held in cpu-hotplug and here.
 454          * This combination can cause a deadlock (cpu-hotplug try to lock
 455          * text_mutex but stop_machine can not be done because online_cpus
 456          * has been changed)
 457          * To avoid this deadlock, we need to call get_online_cpus()
 458          * for preventing cpu-hotplug outside of text_mutex locking.
 459          */
 460         get_online_cpus();
 461         mutex_lock(&text_mutex);
 462         list_for_each_entry_safe(op, tmp, &optimizing_list, list) {
 463                 WARN_ON(kprobe_disabled(&op->kp));
 464                 if (arch_optimize_kprobe(op) < 0)
 465                         op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
 466                 list_del_init(&op->list);
 467         }
 468         mutex_unlock(&text_mutex);
 469         put_online_cpus();
 470 end:
 471         mutex_unlock(&kprobe_mutex);
 472         mutex_unlock(&module_mutex);
 473 }
 474
 475 /* Optimize kprobe if p is ready to be optimized */
 476 static __kprobes void optimize_kprobe(struct kprobe *p)
 477 {
 478         struct optimized_kprobe *op;
 479
 480         /* Check if the kprobe is disabled or not ready for optimization. */
 481         if (!kprobe_optready(p) || !kprobes_allow_optimization ||
 482             (kprobe_disabled(p) || kprobes_all_disarmed))
 483                 return;
 484
 485         /* Both of break_handler and post_handler are not supported. */
 486         if (p->break_handler || p->post_handler)
 487                 return;
 488
 489         op = container_of(p, struct optimized_kprobe, kp);
 490
 491         /* Check there is no other kprobes at the optimized instructions */
 492         if (arch_check_optimized_kprobe(op) < 0)
 493                 return;
 494
 495         /* Check if it is already optimized. */
 496         if (op->kp.flags & KPROBE_FLAG_OPTIMIZED)
 497                 return;
 498
 499         op->kp.flags |= KPROBE_FLAG_OPTIMIZED;
 500         list_add(&op->list, &optimizing_list);
 501         if (!delayed_work_pending(&optimizing_work))
 502                 schedule_delayed_work(&optimizing_work, OPTIMIZE_DELAY);
 503 }
 504
 505 /* Unoptimize a kprobe if p is optimized */
 506 static __kprobes void unoptimize_kprobe(struct kprobe *p)
 507 {
 508         struct optimized_kprobe *op;
 509
 510         if ((p->flags & KPROBE_FLAG_OPTIMIZED) && kprobe_aggrprobe(p)) {
 511                 op = container_of(p, struct optimized_kprobe, kp);
 512                 if (!list_empty(&op->list))
 513                         /* Dequeue from the optimization queue */
 514                         list_del_init(&op->list);
 515                 else
 516                         /* Replace jump with break */
 517                         arch_unoptimize_kprobe(op);
 518                 op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
 519         }
 520 }
 521
 522 /* Remove optimized instructions */
 523 static void __kprobes kill_optimized_kprobe(struct kprobe *p)
 524 {
 525         struct optimized_kprobe *op;
 526
 527         op = container_of(p, struct optimized_kprobe, kp);
 528         if (!list_empty(&op->list)) {
 529                 /* Dequeue from the optimization queue */
 530                 list_del_init(&op->list);
 531                 op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
 532         }
 533         /* Don't unoptimize, because the target code will be freed. */
 534         arch_remove_optimized_kprobe(op);
 535 }
 536
 537 /* Try to prepare optimized instructions */
 538 static __kprobes void prepare_optimized_kprobe(struct kprobe *p)
 539 {
 540         struct optimized_kprobe *op;
 541
 542         op = container_of(p, struct optimized_kprobe, kp);
 543         arch_prepare_optimized_kprobe(op);
 544 }
 545
 546 /* Free optimized instructions and optimized_kprobe */
 547 static __kprobes void free_aggr_kprobe(struct kprobe *p)
 548 {
 549         struct optimized_kprobe *op;
 550
 551         op = container_of(p, struct optimized_kprobe, kp);
 552         arch_remove_optimized_kprobe(op);
 553         kfree(op);
 554 }
 555
 556 /* Allocate new optimized_kprobe and try to prepare optimized instructions */
 557 static __kprobes struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
 558 {
 559         struct optimized_kprobe *op;
 560
 561         op = kzalloc(sizeof(struct optimized_kprobe), GFP_KERNEL);
 562         if (!op)
 563                 return NULL;
 564
 565         INIT_LIST_HEAD(&op->list);
 566         op->kp.addr = p->addr;
 567         arch_prepare_optimized_kprobe(op);
 568
 569         return &op->kp;
 570 }
 571
 572 static void __kprobes init_aggr_kprobe(struct kprobe *ap, struct kprobe *p);
 573
 574 /*
 575  * Prepare an optimized_kprobe and optimize it
 576  * NOTE: p must be a normal registered kprobe
 577  */
 578 static __kprobes void try_to_optimize_kprobe(struct kprobe *p)
 579 {
 580         struct kprobe *ap;
 581         struct optimized_kprobe *op;
 582
 583         ap = alloc_aggr_kprobe(p);
 584         if (!ap)
 585                 return;
 586
 587         op = container_of(ap, struct optimized_kprobe, kp);
 588         if (!arch_prepared_optinsn(&op->optinsn)) {
 589                 /* If failed to setup optimizing, fallback to kprobe */
 590                 free_aggr_kprobe(ap);
 591                 return;
 592         }
 593
 594         init_aggr_kprobe(ap, p);
 595         optimize_kprobe(ap);
 596 }
 597
 598 #ifdef CONFIG_SYSCTL
 599 /* This should be called with kprobe_mutex locked */
 600 static void __kprobes optimize_all_kprobes(void)
 601 {
 602         struct hlist_head *head;
 603         struct hlist_node *node;
 604         struct kprobe *p;
 605         unsigned int i;
 606
 607         /* If optimization is already allowed, just return */
 608         if (kprobes_allow_optimization)
 609                 return;
 610
 611         kprobes_allow_optimization = true;
 612         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
 613                 head = &kprobe_table[i];
 614                 hlist_for_each_entry_rcu(p, node, head, hlist)
 615                         if (!kprobe_disabled(p))
 616                                 optimize_kprobe(p);
 617         }
 618         printk(KERN_INFO "Kprobes globally optimized\n");
 619 }
 620
 621 /* This should be called with kprobe_mutex locked */
 622 static void __kprobes unoptimize_all_kprobes(void)
 623 {
 624         struct hlist_head *head;
 625         struct hlist_node *node;
 626         struct kprobe *p;
 627         unsigned int i;
 628
 629         /* If optimization is already prohibited, just return */
 630         if (!kprobes_allow_optimization)
 631                 return;
 632
 633         kprobes_allow_optimization = false;
 634         printk(KERN_INFO "Kprobes globally unoptimized\n");
 635         get_online_cpus();      /* For avoiding text_mutex deadlock */
 636         mutex_lock(&text_mutex);
 637         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
 638                 head = &kprobe_table[i];
 639                 hlist_for_each_entry_rcu(p, node, head, hlist) {
 640                         if (!kprobe_disabled(p))
 641                                 unoptimize_kprobe(p);
 642                 }
 643         }
 644
 645         mutex_unlock(&text_mutex);
 646         put_online_cpus();
 647         /* Allow all currently running kprobes to complete */
 648         synchronize_sched();
 649 }
 650
 651 int sysctl_kprobes_optimization;
 652 int proc_kprobes_optimization_handler(struct ctl_table *table, int write,
 653                                       void __user *buffer, size_t *length,
 654                                       loff_t *ppos)
 655 {
 656         int ret;
 657
 658         mutex_lock(&kprobe_mutex);
 659         sysctl_kprobes_optimization = kprobes_allow_optimization ? 1 : 0;
 660         ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
 661
 662         if (sysctl_kprobes_optimization)
 663                 optimize_all_kprobes();
 664         else
 665                 unoptimize_all_kprobes();
 666         mutex_unlock(&kprobe_mutex);
 667
 668         return ret;
 669 }
 670 #endif /* CONFIG_SYSCTL */
 671
 672 static void __kprobes __arm_kprobe(struct kprobe *p)
 673 {
 674         struct kprobe *old_p;
 675
 676         /* Check collision with other optimized kprobes */
 677         old_p = get_optimized_kprobe((unsigned long)p->addr);
 678         if (unlikely(old_p))
 679                 unoptimize_kprobe(old_p); /* Fallback to unoptimized kprobe */
 680
 681         arch_arm_kprobe(p);
 682         optimize_kprobe(p);     /* Try to optimize (add kprobe to a list) */
 683 }
 684
 685 static void __kprobes __disarm_kprobe(struct kprobe *p)
 686 {
 687         struct kprobe *old_p;
 688
 689         unoptimize_kprobe(p);   /* Try to unoptimize */
 690         arch_disarm_kprobe(p);
 691
 692         /* If another kprobe was blocked, optimize it. */
 693         old_p = get_optimized_kprobe((unsigned long)p->addr);
 694         if (unlikely(old_p))
 695                 optimize_kprobe(old_p);
 696 }
 697
 698 #else /* !CONFIG_OPTPROBES */
 699
 700 #define optimize_kprobe(p)                      do {} while (0)
 701 #define unoptimize_kprobe(p)                    do {} while (0)
 702 #define kill_optimized_kprobe(p)                do {} while (0)
 703 #define prepare_optimized_kprobe(p)             do {} while (0)
 704 #define try_to_optimize_kprobe(p)               do {} while (0)
 705 #define __arm_kprobe(p)                         arch_arm_kprobe(p)
 706 #define __disarm_kprobe(p)                      arch_disarm_kprobe(p)
 707
 708 static __kprobes void free_aggr_kprobe(struct kprobe *p)
 709 {
 710         kfree(p);
 711 }
 712
 713 static __kprobes struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
 714 {
 715         return kzalloc(sizeof(struct kprobe), GFP_KERNEL);
 716 }
 717 #endif /* CONFIG_OPTPROBES */
 718
 719 /* Arm a kprobe with text_mutex */
 720 static void __kprobes arm_kprobe(struct kprobe *kp)
 721 {
 722         /*
 723          * Here, since __arm_kprobe() doesn't use stop_machine(),
 724          * this doesn't cause deadlock on text_mutex. So, we don't
 725          * need get_online_cpus().
 726          */
 727         mutex_lock(&text_mutex);
 728         __arm_kprobe(kp);
 729         mutex_unlock(&text_mutex);
 730 }
 731
 732 /* Disarm a kprobe with text_mutex */
 733 static void __kprobes disarm_kprobe(struct kprobe *kp)
 734 {
 735         get_online_cpus();      /* For avoiding text_mutex deadlock */
 736         mutex_lock(&text_mutex);
 737         __disarm_kprobe(kp);
 738         mutex_unlock(&text_mutex);
 739         put_online_cpus();
 740 }
 741
 742 /*
 743  * Aggregate handlers for multiple kprobes support - these handlers
 744  * take care of invoking the individual kprobe handlers on p->list
 745  */
 746 static int __kprobes aggr_pre_handler(struct kprobe *p, struct pt_regs *regs)
 747 {
 748         struct kprobe *kp;
 749
 750         list_for_each_entry_rcu(kp, &p->list, list) {
 751                 if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
 752                         set_kprobe_instance(kp);
 753                         if (kp->pre_handler(kp, regs))
 754                                 return 1;
 755                 }
 756                 reset_kprobe_instance();
 757         }
 758         return 0;
 759 }
 760
 761 static void __kprobes aggr_post_handler(struct kprobe *p, struct pt_regs *regs,
 762                                         unsigned long flags)
 763 {
 764         struct kprobe *kp;
 765
 766         list_for_each_entry_rcu(kp, &p->list, list) {
 767                 if (kp->post_handler && likely(!kprobe_disabled(kp))) {
 768                         set_kprobe_instance(kp);
 769                         kp->post_handler(kp, regs, flags);
 770                         reset_kprobe_instance();
 771                 }
 772         }
 773 }
 774
 775 static int __kprobes aggr_fault_handler(struct kprobe *p, struct pt_regs *regs,
 776                                         int trapnr)
 777 {
 778         struct kprobe *cur = __get_cpu_var(kprobe_instance);
 779
 780         /*
 781          * if we faulted "during" the execution of a user specified
 782          * probe handler, invoke just that probe's fault handler
 783          */
 784         if (cur && cur->fault_handler) {
 785                 if (cur->fault_handler(cur, regs, trapnr))
 786                         return 1;
 787         }
 788         return 0;
 789 }
 790
 791 static int __kprobes aggr_break_handler(struct kprobe *p, struct pt_regs *regs)
 792 {
 793         struct kprobe *cur = __get_cpu_var(kprobe_instance);
 794         int ret = 0;
 795
 796         if (cur && cur->break_handler) {
 797                 if (cur->break_handler(cur, regs))
 798                         ret = 1;
 799         }
 800         reset_kprobe_instance();
 801         return ret;
 802 }
 803
 804 /* Walks the list and increments nmissed count for multiprobe case */
 805 void __kprobes kprobes_inc_nmissed_count(struct kprobe *p)
 806 {
 807         struct kprobe *kp;
 808         if (!kprobe_aggrprobe(p)) {
 809                 p->nmissed++;
 810         } else {
 811                 list_for_each_entry_rcu(kp, &p->list, list)
 812                         kp->nmissed++;
 813         }
 814         return;
 815 }
 816
 817 void __kprobes recycle_rp_inst(struct kretprobe_instance *ri,
 818                                 struct hlist_head *head)
 819 {
 820         struct kretprobe *rp = ri->rp;
 821
 822         /* remove rp inst off the rprobe_inst_table */
 823         hlist_del(&ri->hlist);
 824         INIT_HLIST_NODE(&ri->hlist);
 825         if (likely(rp)) {
 826                 spin_lock(&rp->lock);
 827                 hlist_add_head(&ri->hlist, &rp->free_instances);
 828                 spin_unlock(&rp->lock);
 829         } else
 830                 /* Unregistering */
 831                 hlist_add_head(&ri->hlist, head);
 832 }
 833
 834 void __kprobes kretprobe_hash_lock(struct task_struct *tsk,
 835                          struct hlist_head **head, unsigned long *flags)
 836 __acquires(hlist_lock)
 837 {
 838         unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
 839         spinlock_t *hlist_lock;
 840
 841         *head = &kretprobe_inst_table[hash];
 842         hlist_lock = kretprobe_table_lock_ptr(hash);
 843         spin_lock_irqsave(hlist_lock, *flags);
 844 }
 845
 846 static void __kprobes kretprobe_table_lock(unsigned long hash,
 847         unsigned long *flags)
 848 __acquires(hlist_lock)
 849 {
 850         spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
 851         spin_lock_irqsave(hlist_lock, *flags);
 852 }
 853
 854 void __kprobes kretprobe_hash_unlock(struct task_struct *tsk,
 855         unsigned long *flags)
 856 __releases(hlist_lock)
 857 {
 858         unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
 859         spinlock_t *hlist_lock;
 860
 861         hlist_lock = kretprobe_table_lock_ptr(hash);
 862         spin_unlock_irqrestore(hlist_lock, *flags);
 863 }
 864
 865 static void __kprobes kretprobe_table_unlock(unsigned long hash,
 866        unsigned long *flags)
 867 __releases(hlist_lock)
 868 {
 869         spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
 870         spin_unlock_irqrestore(hlist_lock, *flags);
 871 }
 872
 873 /*
 874  * This function is called from finish_task_switch when task tk becomes dead,
 875  * so that we can recycle any function-return probe instances associated
 876  * with this task. These left over instances represent probed functions
 877  * that have been called but will never return.
 878  */
 879 void __kprobes kprobe_flush_task(struct task_struct *tk)
 880 {
 881         struct kretprobe_instance *ri;
 882         struct hlist_head *head, empty_rp;
 883         struct hlist_node *node, *tmp;
 884         unsigned long hash, flags = 0;
 885
 886         if (unlikely(!kprobes_initialized))
 887                 /* Early boot.  kretprobe_table_locks not yet initialized. */
 888                 return;
 889
 890         hash = hash_ptr(tk, KPROBE_HASH_BITS);
 891         head = &kretprobe_inst_table[hash];
 892         kretprobe_table_lock(hash, &flags);
 893         hlist_for_each_entry_safe(ri, node, tmp, head, hlist) {
 894                 if (ri->task == tk)
 895                         recycle_rp_inst(ri, &empty_rp);
 896         }
 897         kretprobe_table_unlock(hash, &flags);
 898         INIT_HLIST_HEAD(&empty_rp);
 899         hlist_for_each_entry_safe(ri, node, tmp, &empty_rp, hlist) {
 900                 hlist_del(&ri->hlist);
 901                 kfree(ri);
 902         }
 903 }
 904
 905 static inline void free_rp_inst(struct kretprobe *rp)
 906 {
 907         struct kretprobe_instance *ri;
 908         struct hlist_node *pos, *next;
 909
 910         hlist_for_each_entry_safe(ri, pos, next, &rp->free_instances, hlist) {
 911                 hlist_del(&ri->hlist);
 912                 kfree(ri);
 913         }
 914 }
 915
 916 static void __kprobes cleanup_rp_inst(struct kretprobe *rp)
 917 {
 918         unsigned long flags, hash;
 919         struct kretprobe_instance *ri;
 920         struct hlist_node *pos, *next;
 921         struct hlist_head *head;
 922
 923         /* No race here */
 924         for (hash = 0; hash < KPROBE_TABLE_SIZE; hash++) {
 925                 kretprobe_table_lock(hash, &flags);
 926                 head = &kretprobe_inst_table[hash];
 927                 hlist_for_each_entry_safe(ri, pos, next, head, hlist) {
 928                         if (ri->rp == rp)
 929                                 ri->rp = NULL;
 930                 }
 931                 kretprobe_table_unlock(hash, &flags);
 932         }
 933         free_rp_inst(rp);
 934 }
 935
 936 /*
 937 * Add the new probe to ap->list. Fail if this is the
 938 * second jprobe at the address - two jprobes can't coexist
 939 */
 940 static int __kprobes add_new_kprobe(struct kprobe *ap, struct kprobe *p)
 941 {
 942         BUG_ON(kprobe_gone(ap) || kprobe_gone(p));
 943
 944         if (p->break_handler || p->post_handler)
 945                 unoptimize_kprobe(ap);  /* Fall back to normal kprobe */
 946
 947         if (p->break_handler) {
 948                 if (ap->break_handler)
 949                         return -EEXIST;
 950                 list_add_tail_rcu(&p->list, &ap->list);
 951                 ap->break_handler = aggr_break_handler;
 952         } else
 953                 list_add_rcu(&p->list, &ap->list);
 954         if (p->post_handler && !ap->post_handler)
 955                 ap->post_handler = aggr_post_handler;
 956
 957         if (kprobe_disabled(ap) && !kprobe_disabled(p)) {
 958                 ap->flags &= ~KPROBE_FLAG_DISABLED;
 959                 if (!kprobes_all_disarmed)
 960                         /* Arm the breakpoint again. */
 961                         __arm_kprobe(ap);
 962         }
 963         return 0;
 964 }
 965
 966 /*
 967  * Fill in the required fields of the "manager kprobe". Replace the
 968  * earlier kprobe in the hlist with the manager kprobe
 969  */
 970 static void __kprobes init_aggr_kprobe(struct kprobe *ap, struct kprobe *p)
 971 {
 972         /* Copy p's insn slot to ap */
 973         copy_kprobe(p, ap);
 974         flush_insn_slot(ap);
 975         ap->addr = p->addr;
 976         ap->flags = p->flags & ~KPROBE_FLAG_OPTIMIZED;
 977         ap->pre_handler = aggr_pre_handler;
 978         ap->fault_handler = aggr_fault_handler;
 979         /* We don't care the kprobe which has gone. */
 980         if (p->post_handler && !kprobe_gone(p))
 981                 ap->post_handler = aggr_post_handler;
 982         if (p->break_handler && !kprobe_gone(p))
 983                 ap->break_handler = aggr_break_handler;
 984
 985         INIT_LIST_HEAD(&ap->list);
 986         INIT_HLIST_NODE(&ap->hlist);
 987
 988         list_add_rcu(&p->list, &ap->list);
 989         hlist_replace_rcu(&p->hlist, &ap->hlist);
 990 }
 991
 992 /*
 993  * This is the second or subsequent kprobe at the address - handle
 994  * the intricacies
 995  */
 996 static int __kprobes register_aggr_kprobe(struct kprobe *old_p,
 997                                           struct kprobe *p)
 998 {
 999         int ret = 0;
1000         struct kprobe *ap = old_p;
1001
1002         if (!kprobe_aggrprobe(old_p)) {
1003                 /* If old_p is not an aggr_kprobe, create new aggr_kprobe. */
1004                 ap = alloc_aggr_kprobe(old_p);
1005                 if (!ap)
1006                         return -ENOMEM;
1007                 init_aggr_kprobe(ap, old_p);
1008         }
1009
1010         if (kprobe_gone(ap)) {
1011                 /*
1012                  * Attempting to insert new probe at the same location that
1013                  * had a probe in the module vaddr area which already
1014                  * freed. So, the instruction slot has already been
1015                  * released. We need a new slot for the new probe.
1016                  */
1017                 ret = arch_prepare_kprobe(ap);
1018                 if (ret)
1019                         /*
1020                          * Even if fail to allocate new slot, don't need to
1021                          * free aggr_probe. It will be used next time, or
1022                          * freed by unregister_kprobe.
1023                          */
1024                         return ret;
1025
1026                 /* Prepare optimized instructions if possible. */
1027                 prepare_optimized_kprobe(ap);
1028
1029                 /*
1030                  * Clear gone flag to prevent allocating new slot again, and
1031                  * set disabled flag because it is not armed yet.
1032                  */
1033                 ap->flags = (ap->flags & ~KPROBE_FLAG_GONE)
1034                             | KPROBE_FLAG_DISABLED;
1035         }
1036
1037         /* Copy ap's insn slot to p */
1038         copy_kprobe(ap, p);
1039         return add_new_kprobe(ap, p);
1040 }
1041
1042 /* Try to disable aggr_kprobe, and return 1 if succeeded.*/
1043 static int __kprobes try_to_disable_aggr_kprobe(struct kprobe *p)
1044 {
1045         struct kprobe *kp;
1046
1047         list_for_each_entry_rcu(kp, &p->list, list) {
1048                 if (!kprobe_disabled(kp))
1049                         /*
1050                          * There is an active probe on the list.
1051                          * We can't disable aggr_kprobe.
1052                          */
1053                         return 0;
1054         }
1055         p->flags |= KPROBE_FLAG_DISABLED;
1056         return 1;
1057 }
1058
1059 static int __kprobes in_kprobes_functions(unsigned long addr)
1060 {
1061         struct kprobe_blackpoint *kb;
1062
1063         if (addr >= (unsigned long)__kprobes_text_start &&
1064             addr < (unsigned long)__kprobes_text_end)
1065                 return -EINVAL;
1066         /*
1067          * If there exists a kprobe_blacklist, verify and
1068          * fail any probe registration in the prohibited area
1069          */
1070         for (kb = kprobe_blacklist; kb->name != NULL; kb++) {
1071                 if (kb->start_addr) {
1072                         if (addr >= kb->start_addr &&
1073                             addr < (kb->start_addr + kb->range))
1074                                 return -EINVAL;
1075                 }
1076         }
1077         return 0;
1078 }
1079
1080 /*
1081  * If we have a symbol_name argument, look it up and add the offset field
1082  * to it. This way, we can specify a relative address to a symbol.
1083  */
1084 static kprobe_opcode_t __kprobes *kprobe_addr(struct kprobe *p)
1085 {
1086         kprobe_opcode_t *addr = p->addr;
1087         if (p->symbol_name) {
1088                 if (addr)
1089                         return NULL;
1090                 kprobe_lookup_name(p->symbol_name, addr);
1091         }
1092
1093         if (!addr)
1094                 return NULL;
1095         return (kprobe_opcode_t *)(((char *)addr) + p->offset);
1096 }
1097
1098 /* Check passed kprobe is valid and return kprobe in kprobe_table. */
1099 static struct kprobe * __kprobes __get_valid_kprobe(struct kprobe *p)
1100 {
1101         struct kprobe *old_p, *list_p;
1102
1103         old_p = get_kprobe(p->addr);
1104         if (unlikely(!old_p))
1105                 return NULL;
1106
1107         if (p != old_p) {
1108                 list_for_each_entry_rcu(list_p, &old_p->list, list)
1109                         if (list_p == p)
1110                         /* kprobe p is a valid probe */
1111                                 goto valid;
1112                 return NULL;
1113         }
1114 valid:
1115         return old_p;
1116 }
1117
1118 /* Return error if the kprobe is being re-registered */
1119 static inline int check_kprobe_rereg(struct kprobe *p)
1120 {
1121         int ret = 0;
1122         struct kprobe *old_p;
1123
1124         mutex_lock(&kprobe_mutex);
1125         old_p = __get_valid_kprobe(p);
1126         if (old_p)
1127                 ret = -EINVAL;
1128         mutex_unlock(&kprobe_mutex);
1129         return ret;
1130 }
1131
1132 int __kprobes register_kprobe(struct kprobe *p)
1133 {
1134         int ret = 0;
1135         struct kprobe *old_p;
1136         struct module *probed_mod;
1137         kprobe_opcode_t *addr;
1138
1139         addr = kprobe_addr(p);
1140         if (!addr)
1141                 return -EINVAL;
1142         p->addr = addr;
1143
1144         ret = check_kprobe_rereg(p);
1145         if (ret)
1146                 return ret;
1147
1148         jump_label_lock();
1149         preempt_disable();
1150         if (!kernel_text_address((unsigned long) p->addr) ||
1151             in_kprobes_functions((unsigned long) p->addr) ||
1152             ftrace_text_reserved(p->addr, p->addr) ||
1153             jump_label_text_reserved(p->addr, p->addr))
1154                 goto fail_with_jump_label;
1155
1156         /* User can pass only KPROBE_FLAG_DISABLED to register_kprobe */
1157         p->flags &= KPROBE_FLAG_DISABLED;
1158
1159         /*
1160          * Check if are we probing a module.
1161          */
1162         probed_mod = __module_text_address((unsigned long) p->addr);
1163         if (probed_mod) {
1164                 /*
1165                  * We must hold a refcount of the probed module while updating
1166                  * its code to prohibit unexpected unloading.
1167                  */
1168                 if (unlikely(!try_module_get(probed_mod)))
1169                         goto fail_with_jump_label;
1170
1171                 /*
1172                  * If the module freed .init.text, we couldn't insert
1173                  * kprobes in there.
1174                  */
1175                 if (within_module_init((unsigned long)p->addr, probed_mod) &&
1176                     probed_mod->state != MODULE_STATE_COMING) {
1177                         module_put(probed_mod);
1178                         goto fail_with_jump_label;
1179                 }
1180         }
1181         preempt_enable();
1182         jump_label_unlock();
1183
1184         p->nmissed = 0;
1185         INIT_LIST_HEAD(&p->list);
1186         mutex_lock(&kprobe_mutex);
1187
1188         jump_label_lock(); /* needed to call jump_label_text_reserved() */
1189
1190         get_online_cpus();      /* For avoiding text_mutex deadlock. */
1191         mutex_lock(&text_mutex);
1192
1193         old_p = get_kprobe(p->addr);
1194         if (old_p) {
1195                 /* Since this may unoptimize old_p, locking text_mutex. */
1196                 ret = register_aggr_kprobe(old_p, p);
1197                 goto out;
1198         }
1199
1200         ret = arch_prepare_kprobe(p);
1201         if (ret)
1202                 goto out;
1203
1204         INIT_HLIST_NODE(&p->hlist);
1205         hlist_add_head_rcu(&p->hlist,
1206                        &kprobe_table[hash_ptr(p->addr, KPROBE_HASH_BITS)]);
1207
1208         if (!kprobes_all_disarmed && !kprobe_disabled(p))
1209                 __arm_kprobe(p);
1210
1211         /* Try to optimize kprobe */
1212         try_to_optimize_kprobe(p);
1213
1214 out:
1215         mutex_unlock(&text_mutex);
1216         put_online_cpus();
1217         jump_label_unlock();
1218         mutex_unlock(&kprobe_mutex);
1219
1220         if (probed_mod)
1221                 module_put(probed_mod);
1222
1223         return ret;
1224
1225 fail_with_jump_label:
1226         preempt_enable();
1227         jump_label_unlock();
1228         return -EINVAL;
1229 }
1230 EXPORT_SYMBOL_GPL(register_kprobe);
1231
1232 /*
1233  * Unregister a kprobe without a scheduler synchronization.
1234  */
1235 static int __kprobes __unregister_kprobe_top(struct kprobe *p)
1236 {
1237         struct kprobe *old_p, *list_p;
1238
1239         old_p = __get_valid_kprobe(p);
1240         if (old_p == NULL)
1241                 return -EINVAL;
1242
1243         if (old_p == p ||
1244             (kprobe_aggrprobe(old_p) &&
1245              list_is_singular(&old_p->list))) {
1246                 /*
1247                  * Only probe on the hash list. Disarm only if kprobes are
1248                  * enabled and not gone - otherwise, the breakpoint would
1249                  * already have been removed. We save on flushing icache.
1250                  */
1251                 if (!kprobes_all_disarmed && !kprobe_disabled(old_p))
1252                         disarm_kprobe(old_p);
1253                 hlist_del_rcu(&old_p->hlist);
1254         } else {
1255                 if (p->break_handler && !kprobe_gone(p))
1256                         old_p->break_handler = NULL;
1257                 if (p->post_handler && !kprobe_gone(p)) {
1258                         list_for_each_entry_rcu(list_p, &old_p->list, list) {
1259                                 if ((list_p != p) && (list_p->post_handler))
1260                                         goto noclean;
1261                         }
1262                         old_p->post_handler = NULL;
1263                 }
1264 noclean:
1265                 list_del_rcu(&p->list);
1266                 if (!kprobe_disabled(old_p)) {
1267                         try_to_disable_aggr_kprobe(old_p);
1268                         if (!kprobes_all_disarmed) {
1269                                 if (kprobe_disabled(old_p))
1270                                         disarm_kprobe(old_p);
1271                                 else
1272                                         /* Try to optimize this probe again */
1273                                         optimize_kprobe(old_p);
1274                         }
1275                 }
1276         }
1277         return 0;
1278 }
1279
1280 static void __kprobes __unregister_kprobe_bottom(struct kprobe *p)
1281 {
1282         struct kprobe *old_p;
1283
1284         if (list_empty(&p->list))
1285                 arch_remove_kprobe(p);
1286         else if (list_is_singular(&p->list)) {
1287                 /* "p" is the last child of an aggr_kprobe */
1288                 old_p = list_entry(p->list.next, struct kprobe, list);
1289                 list_del(&p->list);
1290                 arch_remove_kprobe(old_p);
1291                 free_aggr_kprobe(old_p);
1292         }
1293 }
1294
1295 int __kprobes register_kprobes(struct kprobe **kps, int num)
1296 {
1297         int i, ret = 0;
1298
1299         if (num <= 0)
1300                 return -EINVAL;
1301         for (i = 0; i < num; i++) {
1302                 ret = register_kprobe(kps[i]);
1303                 if (ret < 0) {
1304                         if (i > 0)
1305                                 unregister_kprobes(kps, i);
1306                         break;
1307                 }
1308         }
1309         return ret;
1310 }
1311 EXPORT_SYMBOL_GPL(register_kprobes);
1312
1313 void __kprobes unregister_kprobe(struct kprobe *p)
1314 {
1315         unregister_kprobes(&p, 1);
1316 }
1317 EXPORT_SYMBOL_GPL(unregister_kprobe);
1318
1319 void __kprobes unregister_kprobes(struct kprobe **kps, int num)
1320 {
1321         int i;
1322
1323         if (num <= 0)
1324                 return;
1325         mutex_lock(&kprobe_mutex);
1326         for (i = 0; i < num; i++)
1327                 if (__unregister_kprobe_top(kps[i]) < 0)
1328                         kps[i]->addr = NULL;
1329         mutex_unlock(&kprobe_mutex);
1330
1331         synchronize_sched();
1332         for (i = 0; i < num; i++)
1333                 if (kps[i]->addr)
1334                         __unregister_kprobe_bottom(kps[i]);
1335 }
1336 EXPORT_SYMBOL_GPL(unregister_kprobes);
1337
1338 static struct notifier_block kprobe_exceptions_nb = {
1339         .notifier_call = kprobe_exceptions_notify,
1340         .priority = 0x7fffffff /* we need to be notified first */
1341 };
1342
1343 unsigned long __weak arch_deref_entry_point(void *entry)
1344 {
1345         return (unsigned long)entry;
1346 }
1347
1348 int __kprobes register_jprobes(struct jprobe **jps, int num)
1349 {
1350         struct jprobe *jp;
1351         int ret = 0, i;
1352
1353         if (num <= 0)
1354                 return -EINVAL;
1355         for (i = 0; i < num; i++) {
1356                 unsigned long addr, offset;
1357                 jp = jps[i];
1358                 addr = arch_deref_entry_point(jp->entry);
1359
1360                 /* Verify probepoint is a function entry point */
1361                 if (kallsyms_lookup_size_offset(addr, NULL, &offset) &&
1362                     offset == 0) {
1363                         jp->kp.pre_handler = setjmp_pre_handler;
1364                         jp->kp.break_handler = longjmp_break_handler;
1365                         ret = register_kprobe(&jp->kp);
1366                 } else
1367                         ret = -EINVAL;
1368
1369                 if (ret < 0) {
1370                         if (i > 0)
1371                                 unregister_jprobes(jps, i);
1372                         break;
1373                 }
1374         }
1375         return ret;
1376 }
1377 EXPORT_SYMBOL_GPL(register_jprobes);
1378
1379 int __kprobes register_jprobe(struct jprobe *jp)
1380 {
1381         return register_jprobes(&jp, 1);
1382 }
1383 EXPORT_SYMBOL_GPL(register_jprobe);
1384
1385 void __kprobes unregister_jprobe(struct jprobe *jp)
1386 {
1387         unregister_jprobes(&jp, 1);
1388 }
1389 EXPORT_SYMBOL_GPL(unregister_jprobe);
1390
1391 void __kprobes unregister_jprobes(struct jprobe **jps, int num)
1392 {
1393         int i;
1394
1395         if (num <= 0)
1396                 return;
1397         mutex_lock(&kprobe_mutex);
1398         for (i = 0; i < num; i++)
1399                 if (__unregister_kprobe_top(&jps[i]->kp) < 0)
1400                         jps[i]->kp.addr = NULL;
1401         mutex_unlock(&kprobe_mutex);
1402
1403         synchronize_sched();
1404         for (i = 0; i < num; i++) {
1405                 if (jps[i]->kp.addr)
1406                         __unregister_kprobe_bottom(&jps[i]->kp);
1407         }
1408 }
1409 EXPORT_SYMBOL_GPL(unregister_jprobes);
1410
1411 #ifdef CONFIG_KRETPROBES
1412 /*
1413  * This kprobe pre_handler is registered with every kretprobe. When probe
1414  * hits it will set up the return probe.
1415  */
1416 static int __kprobes pre_handler_kretprobe(struct kprobe *p,
1417                                            struct pt_regs *regs)
1418 {
1419         struct kretprobe *rp = container_of(p, struct kretprobe, kp);
1420         unsigned long hash, flags = 0;
1421         struct kretprobe_instance *ri;
1422
1423         /*TODO: consider to only swap the RA after the last pre_handler fired */
1424         hash = hash_ptr(current, KPROBE_HASH_BITS);
1425         spin_lock_irqsave(&rp->lock, flags);
1426         if (!hlist_empty(&rp->free_instances)) {
1427                 ri = hlist_entry(rp->free_instances.first,
1428                                 struct kretprobe_instance, hlist);
1429                 hlist_del(&ri->hlist);
1430                 spin_unlock_irqrestore(&rp->lock, flags);
1431
1432                 ri->rp = rp;
1433                 ri->task = current;
1434
1435                 if (rp->entry_handler && rp->entry_handler(ri, regs))
1436                         return 0;
1437
1438                 arch_prepare_kretprobe(ri, regs);
1439
1440                 /* XXX(hch): why is there no hlist_move_head? */
1441                 INIT_HLIST_NODE(&ri->hlist);
1442                 kretprobe_table_lock(hash, &flags);
1443                 hlist_add_head(&ri->hlist, &kretprobe_inst_table[hash]);
1444                 kretprobe_table_unlock(hash, &flags);
1445         } else {
1446                 rp->nmissed++;
1447                 spin_unlock_irqrestore(&rp->lock, flags);
1448         }
1449         return 0;
1450 }
1451
1452 int __kprobes register_kretprobe(struct kretprobe *rp)
1453 {
1454         int ret = 0;
1455         struct kretprobe_instance *inst;
1456         int i;
1457         void *addr;
1458
1459         if (kretprobe_blacklist_size) {
1460                 addr = kprobe_addr(&rp->kp);
1461                 if (!addr)
1462                         return -EINVAL;
1463
1464                 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
1465                         if (kretprobe_blacklist[i].addr == addr)
1466                                 return -EINVAL;
1467                 }
1468         }
1469
1470         rp->kp.pre_handler = pre_handler_kretprobe;
1471         rp->kp.post_handler = NULL;
1472         rp->kp.fault_handler = NULL;
1473         rp->kp.break_handler = NULL;
1474
1475         /* Pre-allocate memory for max kretprobe instances */
1476         if (rp->maxactive <= 0) {
1477 #ifdef CONFIG_PREEMPT
1478                 rp->maxactive = max_t(unsigned int, 10, 2*num_possible_cpus());
1479 #else
1480                 rp->maxactive = num_possible_cpus();
1481 #endif
1482         }
1483         spin_lock_init(&rp->lock);
1484         INIT_HLIST_HEAD(&rp->free_instances);
1485         for (i = 0; i < rp->maxactive; i++) {
1486                 inst = kmalloc(sizeof(struct kretprobe_instance) +
1487                                rp->data_size, GFP_KERNEL);
1488                 if (inst == NULL) {
1489                         free_rp_inst(rp);
1490                         return -ENOMEM;
1491                 }
1492                 INIT_HLIST_NODE(&inst->hlist);
1493                 hlist_add_head(&inst->hlist, &rp->free_instances);
1494         }
1495
1496         rp->nmissed = 0;
1497         /* Establish function entry probe point */
1498         ret = register_kprobe(&rp->kp);
1499         if (ret != 0)
1500                 free_rp_inst(rp);
1501         return ret;
1502 }
1503 EXPORT_SYMBOL_GPL(register_kretprobe);
1504
1505 int __kprobes register_kretprobes(struct kretprobe **rps, int num)
1506 {
1507         int ret = 0, i;
1508
1509         if (num <= 0)
1510                 return -EINVAL;
1511         for (i = 0; i < num; i++) {
1512                 ret = register_kretprobe(rps[i]);
1513                 if (ret < 0) {
1514                         if (i > 0)
1515                                 unregister_kretprobes(rps, i);
1516                         break;
1517                 }
1518         }
1519         return ret;
1520 }
1521 EXPORT_SYMBOL_GPL(register_kretprobes);
1522
1523 void __kprobes unregister_kretprobe(struct kretprobe *rp)
1524 {
1525         unregister_kretprobes(&rp, 1);
1526 }
1527 EXPORT_SYMBOL_GPL(unregister_kretprobe);
1528
1529 void __kprobes unregister_kretprobes(struct kretprobe **rps, int num)
1530 {
1531         int i;
1532
1533         if (num <= 0)
1534                 return;
1535         mutex_lock(&kprobe_mutex);
1536         for (i = 0; i < num; i++)
1537                 if (__unregister_kprobe_top(&rps[i]->kp) < 0)
1538                         rps[i]->kp.addr = NULL;
1539         mutex_unlock(&kprobe_mutex);
1540
1541         synchronize_sched();
1542         for (i = 0; i < num; i++) {
1543                 if (rps[i]->kp.addr) {
1544                         __unregister_kprobe_bottom(&rps[i]->kp);
1545                         cleanup_rp_inst(rps[i]);
1546                 }
1547         }
1548 }
1549 EXPORT_SYMBOL_GPL(unregister_kretprobes);
1550
1551 #else /* CONFIG_KRETPROBES */
1552 int __kprobes register_kretprobe(struct kretprobe *rp)
1553 {
1554         return -ENOSYS;
1555 }
1556 EXPORT_SYMBOL_GPL(register_kretprobe);
1557
1558 int __kprobes register_kretprobes(struct kretprobe **rps, int num)
1559 {
1560         return -ENOSYS;
1561 }
1562 EXPORT_SYMBOL_GPL(register_kretprobes);
1563
1564 void __kprobes unregister_kretprobe(struct kretprobe *rp)
1565 {
1566 }
1567 EXPORT_SYMBOL_GPL(unregister_kretprobe);
1568
1569 void __kprobes unregister_kretprobes(struct kretprobe **rps, int num)
1570 {
1571 }
1572 EXPORT_SYMBOL_GPL(unregister_kretprobes);
1573
1574 static int __kprobes pre_handler_kretprobe(struct kprobe *p,
1575                                            struct pt_regs *regs)
1576 {
1577         return 0;
1578 }
1579
1580 #endif /* CONFIG_KRETPROBES */
1581
1582 /* Set the kprobe gone and remove its instruction buffer. */
1583 static void __kprobes kill_kprobe(struct kprobe *p)
1584 {
1585         struct kprobe *kp;
1586
1587         p->flags |= KPROBE_FLAG_GONE;
1588         if (kprobe_aggrprobe(p)) {
1589                 /*
1590                  * If this is an aggr_kprobe, we have to list all the
1591                  * chained probes and mark them GONE.
1592                  */
1593                 list_for_each_entry_rcu(kp, &p->list, list)
1594                         kp->flags |= KPROBE_FLAG_GONE;
1595                 p->post_handler = NULL;
1596                 p->break_handler = NULL;
1597                 kill_optimized_kprobe(p);
1598         }
1599         /*
1600          * Here, we can remove insn_slot safely, because no thread calls
1601          * the original probed function (which will be freed soon) any more.
1602          */
1603         arch_remove_kprobe(p);
1604 }
1605
1606 /* Disable one kprobe */
1607 int __kprobes disable_kprobe(struct kprobe *kp)
1608 {
1609         int ret = 0;
1610         struct kprobe *p;
1611
1612         mutex_lock(&kprobe_mutex);
1613
1614         /* Check whether specified probe is valid. */
1615         p = __get_valid_kprobe(kp);
1616         if (unlikely(p == NULL)) {
1617                 ret = -EINVAL;
1618                 goto out;
1619         }
1620
1621         /* If the probe is already disabled (or gone), just return */
1622         if (kprobe_disabled(kp))
1623                 goto out;
1624
1625         kp->flags |= KPROBE_FLAG_DISABLED;
1626         if (p != kp)
1627                 /* When kp != p, p is always enabled. */
1628                 try_to_disable_aggr_kprobe(p);
1629
1630         if (!kprobes_all_disarmed && kprobe_disabled(p))
1631                 disarm_kprobe(p);
1632 out:
1633         mutex_unlock(&kprobe_mutex);
1634         return ret;
1635 }
1636 EXPORT_SYMBOL_GPL(disable_kprobe);
1637
1638 /* Enable one kprobe */
1639 int __kprobes enable_kprobe(struct kprobe *kp)
1640 {
1641         int ret = 0;
1642         struct kprobe *p;
1643
1644         mutex_lock(&kprobe_mutex);
1645
1646         /* Check whether specified probe is valid. */
1647         p = __get_valid_kprobe(kp);
1648         if (unlikely(p == NULL)) {
1649                 ret = -EINVAL;
1650                 goto out;
1651         }
1652
1653         if (kprobe_gone(kp)) {
1654                 /* This kprobe has gone, we couldn't enable it. */
1655                 ret = -EINVAL;
1656                 goto out;
1657         }
1658
1659         if (p != kp)
1660                 kp->flags &= ~KPROBE_FLAG_DISABLED;
1661
1662         if (!kprobes_all_disarmed && kprobe_disabled(p)) {
1663                 p->flags &= ~KPROBE_FLAG_DISABLED;
1664                 arm_kprobe(p);
1665         }
1666 out:
1667         mutex_unlock(&kprobe_mutex);
1668         return ret;
1669 }
1670 EXPORT_SYMBOL_GPL(enable_kprobe);
1671
1672 void __kprobes dump_kprobe(struct kprobe *kp)
1673 {
1674         printk(KERN_WARNING "Dumping kprobe:\n");
1675         printk(KERN_WARNING "Name: %s\nAddress: %p\nOffset: %x\n",
1676                kp->symbol_name, kp->addr, kp->offset);
1677 }
1678
1679 /* Module notifier call back, checking kprobes on the module */
1680 static int __kprobes kprobes_module_callback(struct notifier_block *nb,
1681                                              unsigned long val, void *data)
1682 {
1683         struct module *mod = data;
1684         struct hlist_head *head;
1685         struct hlist_node *node;
1686         struct kprobe *p;
1687         unsigned int i;
1688         int checkcore = (val == MODULE_STATE_GOING);
1689
1690         if (val != MODULE_STATE_GOING && val != MODULE_STATE_LIVE)
1691                 return NOTIFY_DONE;
1692
1693         /*
1694          * When MODULE_STATE_GOING was notified, both of module .text and
1695          * .init.text sections would be freed. When MODULE_STATE_LIVE was
1696          * notified, only .init.text section would be freed. We need to
1697          * disable kprobes which have been inserted in the sections.
1698          */
1699         mutex_lock(&kprobe_mutex);
1700         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
1701                 head = &kprobe_table[i];
1702                 hlist_for_each_entry_rcu(p, node, head, hlist)
1703                         if (within_module_init((unsigned long)p->addr, mod) ||
1704                             (checkcore &&
1705                              within_module_core((unsigned long)p->addr, mod))) {
1706                                 /*
1707                                  * The vaddr this probe is installed will soon
1708                                  * be vfreed buy not synced to disk. Hence,
1709                                  * disarming the breakpoint isn't needed.
1710                                  */
1711                                 kill_kprobe(p);
1712                         }
1713         }
1714         mutex_unlock(&kprobe_mutex);
1715         return NOTIFY_DONE;
1716 }
1717
1718 static struct notifier_block kprobe_module_nb = {
1719         .notifier_call = kprobes_module_callback,
1720         .priority = 0
1721 };
1722
1723 static int __init init_kprobes(void)
1724 {
1725         int i, err = 0;
1726         unsigned long offset = 0, size = 0;
1727         char *modname, namebuf[128];
1728         const char *symbol_name;
1729         void *addr;
1730         struct kprobe_blackpoint *kb;
1731
1732         /* FIXME allocate the probe table, currently defined statically */
1733         /* initialize all list heads */
1734         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
1735                 INIT_HLIST_HEAD(&kprobe_table[i]);
1736                 INIT_HLIST_HEAD(&kretprobe_inst_table[i]);
1737                 spin_lock_init(&(kretprobe_table_locks[i].lock));
1738         }
1739
1740         /*
1741          * Lookup and populate the kprobe_blacklist.
1742          *
1743          * Unlike the kretprobe blacklist, we'll need to determine
1744          * the range of addresses that belong to the said functions,
1745          * since a kprobe need not necessarily be at the beginning
1746          * of a function.
1747          */
1748         for (kb = kprobe_blacklist; kb->name != NULL; kb++) {
1749                 kprobe_lookup_name(kb->name, addr);
1750                 if (!addr)
1751                         continue;
1752
1753                 kb->start_addr = (unsigned long)addr;
1754                 symbol_name = kallsyms_lookup(kb->start_addr,
1755                                 &size, &offset, &modname, namebuf);
1756                 if (!symbol_name)
1757                         kb->range = 0;
1758                 else
1759                         kb->range = size;
1760         }
1761
1762         if (kretprobe_blacklist_size) {
1763                 /* lookup the function address from its name */
1764                 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
1765                         kprobe_lookup_name(kretprobe_blacklist[i].name,
1766                                            kretprobe_blacklist[i].addr);
1767                         if (!kretprobe_blacklist[i].addr)
1768                                 printk("kretprobe: lookup failed: %s\n",
1769                                        kretprobe_blacklist[i].name);
1770                 }
1771         }
1772
1773 #if defined(CONFIG_OPTPROBES)
1774 #if defined(__ARCH_WANT_KPROBES_INSN_SLOT)
1775         /* Init kprobe_optinsn_slots */
1776         kprobe_optinsn_slots.insn_size = MAX_OPTINSN_SIZE;
1777 #endif
1778         /* By default, kprobes can be optimized */
1779         kprobes_allow_optimization = true;
1780 #endif
1781
1782         /* By default, kprobes are armed */
1783         kprobes_all_disarmed = false;
1784
1785         err = arch_init_kprobes();
1786         if (!err)
1787                 err = register_die_notifier(&kprobe_exceptions_nb);
1788         if (!err)
1789                 err = register_module_notifier(&kprobe_module_nb);
1790
1791         kprobes_initialized = (err == 0);
1792
1793         if (!err)
1794                 init_test_probes();
1795         return err;
1796 }
1797
1798 #ifdef CONFIG_DEBUG_FS
1799 static void __kprobes report_probe(struct seq_file *pi, struct kprobe *p,
1800                 const char *sym, int offset, char *modname, struct kprobe *pp)
1801 {
1802         char *kprobe_type;
1803
1804         if (p->pre_handler == pre_handler_kretprobe)
1805                 kprobe_type = "r";
1806         else if (p->pre_handler == setjmp_pre_handler)
1807                 kprobe_type = "j";
1808         else
1809                 kprobe_type = "k";
1810
1811         if (sym)
1812                 seq_printf(pi, "%p  %s  %s+0x%x  %s ",
1813                         p->addr, kprobe_type, sym, offset,
1814                         (modname ? modname : " "));
1815         else
1816                 seq_printf(pi, "%p  %s  %p ",
1817                         p->addr, kprobe_type, p->addr);
1818
1819         if (!pp)
1820                 pp = p;
1821         seq_printf(pi, "%s%s%s\n",
1822                 (kprobe_gone(p) ? "[GONE]" : ""),
1823                 ((kprobe_disabled(p) && !kprobe_gone(p)) ?  "[DISABLED]" : ""),
1824                 (kprobe_optimized(pp) ? "[OPTIMIZED]" : ""));
1825 }
1826
1827 static void __kprobes *kprobe_seq_start(struct seq_file *f, loff_t *pos)
1828 {
1829         return (*pos < KPROBE_TABLE_SIZE) ? pos : NULL;
1830 }
1831
1832 static void __kprobes *kprobe_seq_next(struct seq_file *f, void *v, loff_t *pos)
1833 {
1834         (*pos)++;
1835         if (*pos >= KPROBE_TABLE_SIZE)
1836                 return NULL;
1837         return pos;
1838 }
1839
1840 static void __kprobes kprobe_seq_stop(struct seq_file *f, void *v)
1841 {
1842         /* Nothing to do */
1843 }
1844
1845 static int __kprobes show_kprobe_addr(struct seq_file *pi, void *v)
1846 {
1847         struct hlist_head *head;
1848         struct hlist_node *node;
1849         struct kprobe *p, *kp;
1850         const char *sym = NULL;
1851         unsigned int i = *(loff_t *) v;
1852         unsigned long offset = 0;
1853         char *modname, namebuf[128];
1854
1855         head = &kprobe_table[i];
1856         preempt_disable();
1857         hlist_for_each_entry_rcu(p, node, head, hlist) {
1858                 sym = kallsyms_lookup((unsigned long)p->addr, NULL,
1859                                         &offset, &modname, namebuf);
1860                 if (kprobe_aggrprobe(p)) {
1861                         list_for_each_entry_rcu(kp, &p->list, list)
1862                                 report_probe(pi, kp, sym, offset, modname, p);
1863                 } else
1864                         report_probe(pi, p, sym, offset, modname, NULL);
1865         }
1866         preempt_enable();
1867         return 0;
1868 }
1869
1870 static const struct seq_operations kprobes_seq_ops = {
1871         .start = kprobe_seq_start,
1872         .next  = kprobe_seq_next,
1873         .stop  = kprobe_seq_stop,
1874         .show  = show_kprobe_addr
1875 };
1876
1877 static int __kprobes kprobes_open(struct inode *inode, struct file *filp)
1878 {
1879         return seq_open(filp, &kprobes_seq_ops);
1880 }
1881
1882 static const struct file_operations debugfs_kprobes_operations = {
1883         .open           = kprobes_open,
1884         .read           = seq_read,
1885         .llseek         = seq_lseek,
1886         .release        = seq_release,
1887 };
1888
1889 static void __kprobes arm_all_kprobes(void)
1890 {
1891         struct hlist_head *head;
1892         struct hlist_node *node;
1893         struct kprobe *p;
1894         unsigned int i;
1895
1896         mutex_lock(&kprobe_mutex);
1897
1898         /* If kprobes are armed, just return */
1899         if (!kprobes_all_disarmed)
1900                 goto already_enabled;
1901
1902         /* Arming kprobes doesn't optimize kprobe itself */
1903         mutex_lock(&text_mutex);
1904         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
1905                 head = &kprobe_table[i];
1906                 hlist_for_each_entry_rcu(p, node, head, hlist)
1907                         if (!kprobe_disabled(p))
1908                                 __arm_kprobe(p);
1909         }
1910         mutex_unlock(&text_mutex);
1911
1912         kprobes_all_disarmed = false;
1913         printk(KERN_INFO "Kprobes globally enabled\n");
1914
1915 already_enabled:
1916         mutex_unlock(&kprobe_mutex);
1917         return;
1918 }
1919
1920 static void __kprobes disarm_all_kprobes(void)
1921 {
1922         struct hlist_head *head;
1923         struct hlist_node *node;
1924         struct kprobe *p;
1925         unsigned int i;
1926
1927         mutex_lock(&kprobe_mutex);
1928
1929         /* If kprobes are already disarmed, just return */
1930         if (kprobes_all_disarmed)
1931                 goto already_disabled;
1932
1933         kprobes_all_disarmed = true;
1934         printk(KERN_INFO "Kprobes globally disabled\n");
1935
1936         /*
1937          * Here we call get_online_cpus() for avoiding text_mutex deadlock,
1938          * because disarming may also unoptimize kprobes.
1939          */
1940         get_online_cpus();
1941         mutex_lock(&text_mutex);
1942         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
1943                 head = &kprobe_table[i];
1944                 hlist_for_each_entry_rcu(p, node, head, hlist) {
1945                         if (!arch_trampoline_kprobe(p) && !kprobe_disabled(p))
1946                                 __disarm_kprobe(p);
1947                 }
1948         }
1949
1950         mutex_unlock(&text_mutex);
1951         put_online_cpus();
1952         mutex_unlock(&kprobe_mutex);
1953         /* Allow all currently running kprobes to complete */
1954         synchronize_sched();
1955         return;
1956
1957 already_disabled:
1958         mutex_unlock(&kprobe_mutex);
1959         return;
1960 }
1961
1962 /*
1963  * XXX: The debugfs bool file interface doesn't allow for callbacks
1964  * when the bool state is switched. We can reuse that facility when
1965  * available
1966  */
1967 static ssize_t read_enabled_file_bool(struct file *file,
1968                char __user *user_buf, size_t count, loff_t *ppos)
1969 {
1970         char buf[3];
1971
1972         if (!kprobes_all_disarmed)
1973                 buf[0] = '1';
1974         else
1975                 buf[0] = '0';
1976         buf[1] = '\n';
1977         buf[2] = 0x00;
1978         return simple_read_from_buffer(user_buf, count, ppos, buf, 2);
1979 }
1980
1981 static ssize_t write_enabled_file_bool(struct file *file,
1982                const char __user *user_buf, size_t count, loff_t *ppos)
1983 {
1984         char buf[32];
1985         int buf_size;
1986
1987         buf_size = min(count, (sizeof(buf)-1));
1988         if (copy_from_user(buf, user_buf, buf_size))
1989                 return -EFAULT;
1990
1991         switch (buf[0]) {
1992         case 'y':
1993         case 'Y':
1994         case '1':
1995                 arm_all_kprobes();
1996                 break;
1997         case 'n':
1998         case 'N':
1999         case '0':
2000                 disarm_all_kprobes();
2001                 break;
2002         }
2003
2004         return count;
2005 }
2006
2007 static const struct file_operations fops_kp = {
2008         .read =         read_enabled_file_bool,
2009         .write =        write_enabled_file_bool,
2010         .llseek =       default_llseek,
2011 };
2012
2013 static int __kprobes debugfs_kprobe_init(void)
2014 {
2015         struct dentry *dir, *file;
2016         unsigned int value = 1;
2017
2018         dir = debugfs_create_dir("kprobes", NULL);
2019         if (!dir)
2020                 return -ENOMEM;
2021
2022         file = debugfs_create_file("list", 0444, dir, NULL,
2023                                 &debugfs_kprobes_operations);
2024         if (!file) {
2025                 debugfs_remove(dir);
2026                 return -ENOMEM;
2027         }
2028
2029         file = debugfs_create_file("enabled", 0600, dir,
2030                                         &value, &fops_kp);
2031         if (!file) {
2032                 debugfs_remove(dir);
2033                 return -ENOMEM;
2034         }
2035
2036         return 0;
2037 }
2038
2039 late_initcall(debugfs_kprobe_init);
2040 #endif /* CONFIG_DEBUG_FS */
2041
2042 module_init(init_kprobes);
2043
2044 /* defined in arch/.../kernel/kprobes.c */
2045 EXPORT_SYMBOL_GPL(jprobe_return);