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/export.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         raw_spinlock_t lock ____cacheline_aligned_in_smp;
  82 } kretprobe_table_locks[KPROBE_TABLE_SIZE];
  83
  84 static raw_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         __this_cpu_write(kprobe_instance, kp);
 321 }
 322
 323 static inline void reset_kprobe_instance(void)
 324 {
 325         __this_cpu_write(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 /* Return true(!0) if the kprobe is unused */
 358 static inline int kprobe_unused(struct kprobe *p)
 359 {
 360         return kprobe_aggrprobe(p) && kprobe_disabled(p) &&
 361                list_empty(&p->list);
 362 }
 363
 364 /*
 365  * Keep all fields in the kprobe consistent
 366  */
 367 static inline void copy_kprobe(struct kprobe *ap, struct kprobe *p)
 368 {
 369         memcpy(&p->opcode, &ap->opcode, sizeof(kprobe_opcode_t));
 370         memcpy(&p->ainsn, &ap->ainsn, sizeof(struct arch_specific_insn));
 371 }
 372
 373 #ifdef CONFIG_OPTPROBES
 374 /* NOTE: change this value only with kprobe_mutex held */
 375 static bool kprobes_allow_optimization;
 376
 377 /*
 378  * Call all pre_handler on the list, but ignores its return value.
 379  * This must be called from arch-dep optimized caller.
 380  */
 381 void __kprobes opt_pre_handler(struct kprobe *p, struct pt_regs *regs)
 382 {
 383         struct kprobe *kp;
 384
 385         list_for_each_entry_rcu(kp, &p->list, list) {
 386                 if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
 387                         set_kprobe_instance(kp);
 388                         kp->pre_handler(kp, regs);
 389                 }
 390                 reset_kprobe_instance();
 391         }
 392 }
 393
 394 /* Free optimized instructions and optimized_kprobe */
 395 static __kprobes void free_aggr_kprobe(struct kprobe *p)
 396 {
 397         struct optimized_kprobe *op;
 398
 399         op = container_of(p, struct optimized_kprobe, kp);
 400         arch_remove_optimized_kprobe(op);
 401         arch_remove_kprobe(p);
 402         kfree(op);
 403 }
 404
 405 /* Return true(!0) if the kprobe is ready for optimization. */
 406 static inline int kprobe_optready(struct kprobe *p)
 407 {
 408         struct optimized_kprobe *op;
 409
 410         if (kprobe_aggrprobe(p)) {
 411                 op = container_of(p, struct optimized_kprobe, kp);
 412                 return arch_prepared_optinsn(&op->optinsn);
 413         }
 414
 415         return 0;
 416 }
 417
 418 /* Return true(!0) if the kprobe is disarmed. Note: p must be on hash list */
 419 static inline int kprobe_disarmed(struct kprobe *p)
 420 {
 421         struct optimized_kprobe *op;
 422
 423         /* If kprobe is not aggr/opt probe, just return kprobe is disabled */
 424         if (!kprobe_aggrprobe(p))
 425                 return kprobe_disabled(p);
 426
 427         op = container_of(p, struct optimized_kprobe, kp);
 428
 429         return kprobe_disabled(p) && list_empty(&op->list);
 430 }
 431
 432 /* Return true(!0) if the probe is queued on (un)optimizing lists */
 433 static int __kprobes kprobe_queued(struct kprobe *p)
 434 {
 435         struct optimized_kprobe *op;
 436
 437         if (kprobe_aggrprobe(p)) {
 438                 op = container_of(p, struct optimized_kprobe, kp);
 439                 if (!list_empty(&op->list))
 440                         return 1;
 441         }
 442         return 0;
 443 }
 444
 445 /*
 446  * Return an optimized kprobe whose optimizing code replaces
 447  * instructions including addr (exclude breakpoint).
 448  */
 449 static struct kprobe *__kprobes get_optimized_kprobe(unsigned long addr)
 450 {
 451         int i;
 452         struct kprobe *p = NULL;
 453         struct optimized_kprobe *op;
 454
 455         /* Don't check i == 0, since that is a breakpoint case. */
 456         for (i = 1; !p && i < MAX_OPTIMIZED_LENGTH; i++)
 457                 p = get_kprobe((void *)(addr - i));
 458
 459         if (p && kprobe_optready(p)) {
 460                 op = container_of(p, struct optimized_kprobe, kp);
 461                 if (arch_within_optimized_kprobe(op, addr))
 462                         return p;
 463         }
 464
 465         return NULL;
 466 }
 467
 468 /* Optimization staging list, protected by kprobe_mutex */
 469 static LIST_HEAD(optimizing_list);
 470 static LIST_HEAD(unoptimizing_list);
 471
 472 static void kprobe_optimizer(struct work_struct *work);
 473 static DECLARE_DELAYED_WORK(optimizing_work, kprobe_optimizer);
 474 static DECLARE_COMPLETION(optimizer_comp);
 475 #define OPTIMIZE_DELAY 5
 476
 477 /*
 478  * Optimize (replace a breakpoint with a jump) kprobes listed on
 479  * optimizing_list.
 480  */
 481 static __kprobes void do_optimize_kprobes(void)
 482 {
 483         /* Optimization never be done when disarmed */
 484         if (kprobes_all_disarmed || !kprobes_allow_optimization ||
 485             list_empty(&optimizing_list))
 486                 return;
 487
 488         /*
 489          * The optimization/unoptimization refers online_cpus via
 490          * stop_machine() and cpu-hotplug modifies online_cpus.
 491          * And same time, text_mutex will be held in cpu-hotplug and here.
 492          * This combination can cause a deadlock (cpu-hotplug try to lock
 493          * text_mutex but stop_machine can not be done because online_cpus
 494          * has been changed)
 495          * To avoid this deadlock, we need to call get_online_cpus()
 496          * for preventing cpu-hotplug outside of text_mutex locking.
 497          */
 498         get_online_cpus();
 499         mutex_lock(&text_mutex);
 500         arch_optimize_kprobes(&optimizing_list);
 501         mutex_unlock(&text_mutex);
 502         put_online_cpus();
 503 }
 504
 505 /*
 506  * Unoptimize (replace a jump with a breakpoint and remove the breakpoint
 507  * if need) kprobes listed on unoptimizing_list.
 508  */
 509 static __kprobes void do_unoptimize_kprobes(struct list_head *free_list)
 510 {
 511         struct optimized_kprobe *op, *tmp;
 512
 513         /* Unoptimization must be done anytime */
 514         if (list_empty(&unoptimizing_list))
 515                 return;
 516
 517         /* Ditto to do_optimize_kprobes */
 518         get_online_cpus();
 519         mutex_lock(&text_mutex);
 520         arch_unoptimize_kprobes(&unoptimizing_list, free_list);
 521         /* Loop free_list for disarming */
 522         list_for_each_entry_safe(op, tmp, free_list, list) {
 523                 /* Disarm probes if marked disabled */
 524                 if (kprobe_disabled(&op->kp))
 525                         arch_disarm_kprobe(&op->kp);
 526                 if (kprobe_unused(&op->kp)) {
 527                         /*
 528                          * Remove unused probes from hash list. After waiting
 529                          * for synchronization, these probes are reclaimed.
 530                          * (reclaiming is done by do_free_cleaned_kprobes.)
 531                          */
 532                         hlist_del_rcu(&op->kp.hlist);
 533                 } else
 534                         list_del_init(&op->list);
 535         }
 536         mutex_unlock(&text_mutex);
 537         put_online_cpus();
 538 }
 539
 540 /* Reclaim all kprobes on the free_list */
 541 static __kprobes void do_free_cleaned_kprobes(struct list_head *free_list)
 542 {
 543         struct optimized_kprobe *op, *tmp;
 544
 545         list_for_each_entry_safe(op, tmp, free_list, list) {
 546                 BUG_ON(!kprobe_unused(&op->kp));
 547                 list_del_init(&op->list);
 548                 free_aggr_kprobe(&op->kp);
 549         }
 550 }
 551
 552 /* Start optimizer after OPTIMIZE_DELAY passed */
 553 static __kprobes void kick_kprobe_optimizer(void)
 554 {
 555         if (!delayed_work_pending(&optimizing_work))
 556                 schedule_delayed_work(&optimizing_work, OPTIMIZE_DELAY);
 557 }
 558
 559 /* Kprobe jump optimizer */
 560 static __kprobes void kprobe_optimizer(struct work_struct *work)
 561 {
 562         LIST_HEAD(free_list);
 563
 564         mutex_lock(&kprobe_mutex);
 565         /* Lock modules while optimizing kprobes */
 566         mutex_lock(&module_mutex);
 567
 568         /*
 569          * Step 1: Unoptimize kprobes and collect cleaned (unused and disarmed)
 570          * kprobes before waiting for quiesence period.
 571          */
 572         do_unoptimize_kprobes(&free_list);
 573
 574         /*
 575          * Step 2: Wait for quiesence period to ensure all running interrupts
 576          * are done. Because optprobe may modify multiple instructions
 577          * there is a chance that Nth instruction is interrupted. In that
 578          * case, running interrupt can return to 2nd-Nth byte of jump
 579          * instruction. This wait is for avoiding it.
 580          */
 581         synchronize_sched();
 582
 583         /* Step 3: Optimize kprobes after quiesence period */
 584         do_optimize_kprobes();
 585
 586         /* Step 4: Free cleaned kprobes after quiesence period */
 587         do_free_cleaned_kprobes(&free_list);
 588
 589         mutex_unlock(&module_mutex);
 590         mutex_unlock(&kprobe_mutex);
 591
 592         /* Step 5: Kick optimizer again if needed */
 593         if (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list))
 594                 kick_kprobe_optimizer();
 595         else
 596                 /* Wake up all waiters */
 597                 complete_all(&optimizer_comp);
 598 }
 599
 600 /* Wait for completing optimization and unoptimization */
 601 static __kprobes void wait_for_kprobe_optimizer(void)
 602 {
 603         if (delayed_work_pending(&optimizing_work))
 604                 wait_for_completion(&optimizer_comp);
 605 }
 606
 607 /* Optimize kprobe if p is ready to be optimized */
 608 static __kprobes void optimize_kprobe(struct kprobe *p)
 609 {
 610         struct optimized_kprobe *op;
 611
 612         /* Check if the kprobe is disabled or not ready for optimization. */
 613         if (!kprobe_optready(p) || !kprobes_allow_optimization ||
 614             (kprobe_disabled(p) || kprobes_all_disarmed))
 615                 return;
 616
 617         /* Both of break_handler and post_handler are not supported. */
 618         if (p->break_handler || p->post_handler)
 619                 return;
 620
 621         op = container_of(p, struct optimized_kprobe, kp);
 622
 623         /* Check there is no other kprobes at the optimized instructions */
 624         if (arch_check_optimized_kprobe(op) < 0)
 625                 return;
 626
 627         /* Check if it is already optimized. */
 628         if (op->kp.flags & KPROBE_FLAG_OPTIMIZED)
 629                 return;
 630         op->kp.flags |= KPROBE_FLAG_OPTIMIZED;
 631
 632         if (!list_empty(&op->list))
 633                 /* This is under unoptimizing. Just dequeue the probe */
 634                 list_del_init(&op->list);
 635         else {
 636                 list_add(&op->list, &optimizing_list);
 637                 kick_kprobe_optimizer();
 638         }
 639 }
 640
 641 /* Short cut to direct unoptimizing */
 642 static __kprobes void force_unoptimize_kprobe(struct optimized_kprobe *op)
 643 {
 644         get_online_cpus();
 645         arch_unoptimize_kprobe(op);
 646         put_online_cpus();
 647         if (kprobe_disabled(&op->kp))
 648                 arch_disarm_kprobe(&op->kp);
 649 }
 650
 651 /* Unoptimize a kprobe if p is optimized */
 652 static __kprobes void unoptimize_kprobe(struct kprobe *p, bool force)
 653 {
 654         struct optimized_kprobe *op;
 655
 656         if (!kprobe_aggrprobe(p) || kprobe_disarmed(p))
 657                 return; /* This is not an optprobe nor optimized */
 658
 659         op = container_of(p, struct optimized_kprobe, kp);
 660         if (!kprobe_optimized(p)) {
 661                 /* Unoptimized or unoptimizing case */
 662                 if (force && !list_empty(&op->list)) {
 663                         /*
 664                          * Only if this is unoptimizing kprobe and forced,
 665                          * forcibly unoptimize it. (No need to unoptimize
 666                          * unoptimized kprobe again :)
 667                          */
 668                         list_del_init(&op->list);
 669                         force_unoptimize_kprobe(op);
 670                 }
 671                 return;
 672         }
 673
 674         op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
 675         if (!list_empty(&op->list)) {
 676                 /* Dequeue from the optimization queue */
 677                 list_del_init(&op->list);
 678                 return;
 679         }
 680         /* Optimized kprobe case */
 681         if (force)
 682                 /* Forcibly update the code: this is a special case */
 683                 force_unoptimize_kprobe(op);
 684         else {
 685                 list_add(&op->list, &unoptimizing_list);
 686                 kick_kprobe_optimizer();
 687         }
 688 }
 689
 690 /* Cancel unoptimizing for reusing */
 691 static void reuse_unused_kprobe(struct kprobe *ap)
 692 {
 693         struct optimized_kprobe *op;
 694
 695         BUG_ON(!kprobe_unused(ap));
 696         /*
 697          * Unused kprobe MUST be on the way of delayed unoptimizing (means
 698          * there is still a relative jump) and disabled.
 699          */
 700         op = container_of(ap, struct optimized_kprobe, kp);
 701         if (unlikely(list_empty(&op->list)))
 702                 printk(KERN_WARNING "Warning: found a stray unused "
 703                         "aggrprobe@%p\n", ap->addr);
 704         /* Enable the probe again */
 705         ap->flags &= ~KPROBE_FLAG_DISABLED;
 706         /* Optimize it again (remove from op->list) */
 707         BUG_ON(!kprobe_optready(ap));
 708         optimize_kprobe(ap);
 709 }
 710
 711 /* Remove optimized instructions */
 712 static void __kprobes kill_optimized_kprobe(struct kprobe *p)
 713 {
 714         struct optimized_kprobe *op;
 715
 716         op = container_of(p, struct optimized_kprobe, kp);
 717         if (!list_empty(&op->list))
 718                 /* Dequeue from the (un)optimization queue */
 719                 list_del_init(&op->list);
 720
 721         op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
 722         /* Don't touch the code, because it is already freed. */
 723         arch_remove_optimized_kprobe(op);
 724 }
 725
 726 /* Try to prepare optimized instructions */
 727 static __kprobes void prepare_optimized_kprobe(struct kprobe *p)
 728 {
 729         struct optimized_kprobe *op;
 730
 731         op = container_of(p, struct optimized_kprobe, kp);
 732         arch_prepare_optimized_kprobe(op);
 733 }
 734
 735 /* Allocate new optimized_kprobe and try to prepare optimized instructions */
 736 static __kprobes struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
 737 {
 738         struct optimized_kprobe *op;
 739
 740         op = kzalloc(sizeof(struct optimized_kprobe), GFP_KERNEL);
 741         if (!op)
 742                 return NULL;
 743
 744         INIT_LIST_HEAD(&op->list);
 745         op->kp.addr = p->addr;
 746         arch_prepare_optimized_kprobe(op);
 747
 748         return &op->kp;
 749 }
 750
 751 static void __kprobes init_aggr_kprobe(struct kprobe *ap, struct kprobe *p);
 752
 753 /*
 754  * Prepare an optimized_kprobe and optimize it
 755  * NOTE: p must be a normal registered kprobe
 756  */
 757 static __kprobes void try_to_optimize_kprobe(struct kprobe *p)
 758 {
 759         struct kprobe *ap;
 760         struct optimized_kprobe *op;
 761
 762         /* For preparing optimization, jump_label_text_reserved() is called */
 763         jump_label_lock();
 764         mutex_lock(&text_mutex);
 765
 766         ap = alloc_aggr_kprobe(p);
 767         if (!ap)
 768                 goto out;
 769
 770         op = container_of(ap, struct optimized_kprobe, kp);
 771         if (!arch_prepared_optinsn(&op->optinsn)) {
 772                 /* If failed to setup optimizing, fallback to kprobe */
 773                 arch_remove_optimized_kprobe(op);
 774                 kfree(op);
 775                 goto out;
 776         }
 777
 778         init_aggr_kprobe(ap, p);
 779         optimize_kprobe(ap);    /* This just kicks optimizer thread */
 780
 781 out:
 782         mutex_unlock(&text_mutex);
 783         jump_label_unlock();
 784 }
 785
 786 #ifdef CONFIG_SYSCTL
 787 /* This should be called with kprobe_mutex locked */
 788 static void __kprobes optimize_all_kprobes(void)
 789 {
 790         struct hlist_head *head;
 791         struct hlist_node *node;
 792         struct kprobe *p;
 793         unsigned int i;
 794
 795         /* If optimization is already allowed, just return */
 796         if (kprobes_allow_optimization)
 797                 return;
 798
 799         kprobes_allow_optimization = true;
 800         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
 801                 head = &kprobe_table[i];
 802                 hlist_for_each_entry_rcu(p, node, head, hlist)
 803                         if (!kprobe_disabled(p))
 804                                 optimize_kprobe(p);
 805         }
 806         printk(KERN_INFO "Kprobes globally optimized\n");
 807 }
 808
 809 /* This should be called with kprobe_mutex locked */
 810 static void __kprobes unoptimize_all_kprobes(void)
 811 {
 812         struct hlist_head *head;
 813         struct hlist_node *node;
 814         struct kprobe *p;
 815         unsigned int i;
 816
 817         /* If optimization is already prohibited, just return */
 818         if (!kprobes_allow_optimization)
 819                 return;
 820
 821         kprobes_allow_optimization = false;
 822         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
 823                 head = &kprobe_table[i];
 824                 hlist_for_each_entry_rcu(p, node, head, hlist) {
 825                         if (!kprobe_disabled(p))
 826                                 unoptimize_kprobe(p, false);
 827                 }
 828         }
 829         /* Wait for unoptimizing completion */
 830         wait_for_kprobe_optimizer();
 831         printk(KERN_INFO "Kprobes globally unoptimized\n");
 832 }
 833
 834 int sysctl_kprobes_optimization;
 835 int proc_kprobes_optimization_handler(struct ctl_table *table, int write,
 836                                       void __user *buffer, size_t *length,
 837                                       loff_t *ppos)
 838 {
 839         int ret;
 840
 841         mutex_lock(&kprobe_mutex);
 842         sysctl_kprobes_optimization = kprobes_allow_optimization ? 1 : 0;
 843         ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
 844
 845         if (sysctl_kprobes_optimization)
 846                 optimize_all_kprobes();
 847         else
 848                 unoptimize_all_kprobes();
 849         mutex_unlock(&kprobe_mutex);
 850
 851         return ret;
 852 }
 853 #endif /* CONFIG_SYSCTL */
 854
 855 /* Put a breakpoint for a probe. Must be called with text_mutex locked */
 856 static void __kprobes __arm_kprobe(struct kprobe *p)
 857 {
 858         struct kprobe *_p;
 859
 860         /* Check collision with other optimized kprobes */
 861         _p = get_optimized_kprobe((unsigned long)p->addr);
 862         if (unlikely(_p))
 863                 /* Fallback to unoptimized kprobe */
 864                 unoptimize_kprobe(_p, true);
 865
 866         arch_arm_kprobe(p);
 867         optimize_kprobe(p);     /* Try to optimize (add kprobe to a list) */
 868 }
 869
 870 /* Remove the breakpoint of a probe. Must be called with text_mutex locked */
 871 static void __kprobes __disarm_kprobe(struct kprobe *p, bool reopt)
 872 {
 873         struct kprobe *_p;
 874
 875         unoptimize_kprobe(p, false);    /* Try to unoptimize */
 876
 877         if (!kprobe_queued(p)) {
 878                 arch_disarm_kprobe(p);
 879                 /* If another kprobe was blocked, optimize it. */
 880                 _p = get_optimized_kprobe((unsigned long)p->addr);
 881                 if (unlikely(_p) && reopt)
 882                         optimize_kprobe(_p);
 883         }
 884         /* TODO: reoptimize others after unoptimized this probe */
 885 }
 886
 887 #else /* !CONFIG_OPTPROBES */
 888
 889 #define optimize_kprobe(p)                      do {} while (0)
 890 #define unoptimize_kprobe(p, f)                 do {} while (0)
 891 #define kill_optimized_kprobe(p)                do {} while (0)
 892 #define prepare_optimized_kprobe(p)             do {} while (0)
 893 #define try_to_optimize_kprobe(p)               do {} while (0)
 894 #define __arm_kprobe(p)                         arch_arm_kprobe(p)
 895 #define __disarm_kprobe(p, o)                   arch_disarm_kprobe(p)
 896 #define kprobe_disarmed(p)                      kprobe_disabled(p)
 897 #define wait_for_kprobe_optimizer()             do {} while (0)
 898
 899 /* There should be no unused kprobes can be reused without optimization */
 900 static void reuse_unused_kprobe(struct kprobe *ap)
 901 {
 902         printk(KERN_ERR "Error: There should be no unused kprobe here.\n");
 903         BUG_ON(kprobe_unused(ap));
 904 }
 905
 906 static __kprobes void free_aggr_kprobe(struct kprobe *p)
 907 {
 908         arch_remove_kprobe(p);
 909         kfree(p);
 910 }
 911
 912 static __kprobes struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
 913 {
 914         return kzalloc(sizeof(struct kprobe), GFP_KERNEL);
 915 }
 916 #endif /* CONFIG_OPTPROBES */
 917
 918 /* Arm a kprobe with text_mutex */
 919 static void __kprobes arm_kprobe(struct kprobe *kp)
 920 {
 921         /*
 922          * Here, since __arm_kprobe() doesn't use stop_machine(),
 923          * this doesn't cause deadlock on text_mutex. So, we don't
 924          * need get_online_cpus().
 925          */
 926         mutex_lock(&text_mutex);
 927         __arm_kprobe(kp);
 928         mutex_unlock(&text_mutex);
 929 }
 930
 931 /* Disarm a kprobe with text_mutex */
 932 static void __kprobes disarm_kprobe(struct kprobe *kp)
 933 {
 934         /* Ditto */
 935         mutex_lock(&text_mutex);
 936         __disarm_kprobe(kp, true);
 937         mutex_unlock(&text_mutex);
 938 }
 939
 940 /*
 941  * Aggregate handlers for multiple kprobes support - these handlers
 942  * take care of invoking the individual kprobe handlers on p->list
 943  */
 944 static int __kprobes aggr_pre_handler(struct kprobe *p, struct pt_regs *regs)
 945 {
 946         struct kprobe *kp;
 947
 948         list_for_each_entry_rcu(kp, &p->list, list) {
 949                 if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
 950                         set_kprobe_instance(kp);
 951                         if (kp->pre_handler(kp, regs))
 952                                 return 1;
 953                 }
 954                 reset_kprobe_instance();
 955         }
 956         return 0;
 957 }
 958
 959 static void __kprobes aggr_post_handler(struct kprobe *p, struct pt_regs *regs,
 960                                         unsigned long flags)
 961 {
 962         struct kprobe *kp;
 963
 964         list_for_each_entry_rcu(kp, &p->list, list) {
 965                 if (kp->post_handler && likely(!kprobe_disabled(kp))) {
 966                         set_kprobe_instance(kp);
 967                         kp->post_handler(kp, regs, flags);
 968                         reset_kprobe_instance();
 969                 }
 970         }
 971 }
 972
 973 static int __kprobes aggr_fault_handler(struct kprobe *p, struct pt_regs *regs,
 974                                         int trapnr)
 975 {
 976         struct kprobe *cur = __this_cpu_read(kprobe_instance);
 977
 978         /*
 979          * if we faulted "during" the execution of a user specified
 980          * probe handler, invoke just that probe's fault handler
 981          */
 982         if (cur && cur->fault_handler) {
 983                 if (cur->fault_handler(cur, regs, trapnr))
 984                         return 1;
 985         }
 986         return 0;
 987 }
 988
 989 static int __kprobes aggr_break_handler(struct kprobe *p, struct pt_regs *regs)
 990 {
 991         struct kprobe *cur = __this_cpu_read(kprobe_instance);
 992         int ret = 0;
 993
 994         if (cur && cur->break_handler) {
 995                 if (cur->break_handler(cur, regs))
 996                         ret = 1;
 997         }
 998         reset_kprobe_instance();
 999         return ret;
1000 }
1001
1002 /* Walks the list and increments nmissed count for multiprobe case */
1003 void __kprobes kprobes_inc_nmissed_count(struct kprobe *p)
1004 {
1005         struct kprobe *kp;
1006         if (!kprobe_aggrprobe(p)) {
1007                 p->nmissed++;
1008         } else {
1009                 list_for_each_entry_rcu(kp, &p->list, list)
1010                         kp->nmissed++;
1011         }
1012         return;
1013 }
1014
1015 void __kprobes recycle_rp_inst(struct kretprobe_instance *ri,
1016                                 struct hlist_head *head)
1017 {
1018         struct kretprobe *rp = ri->rp;
1019
1020         /* remove rp inst off the rprobe_inst_table */
1021         hlist_del(&ri->hlist);
1022         INIT_HLIST_NODE(&ri->hlist);
1023         if (likely(rp)) {
1024                 raw_spin_lock(&rp->lock);
1025                 hlist_add_head(&ri->hlist, &rp->free_instances);
1026                 raw_spin_unlock(&rp->lock);
1027         } else
1028                 /* Unregistering */
1029                 hlist_add_head(&ri->hlist, head);
1030 }
1031
1032 void __kprobes kretprobe_hash_lock(struct task_struct *tsk,
1033                          struct hlist_head **head, unsigned long *flags)
1034 __acquires(hlist_lock)
1035 {
1036         unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
1037         raw_spinlock_t *hlist_lock;
1038
1039         *head = &kretprobe_inst_table[hash];
1040         hlist_lock = kretprobe_table_lock_ptr(hash);
1041         raw_spin_lock_irqsave(hlist_lock, *flags);
1042 }
1043
1044 static void __kprobes kretprobe_table_lock(unsigned long hash,
1045         unsigned long *flags)
1046 __acquires(hlist_lock)
1047 {
1048         raw_spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
1049         raw_spin_lock_irqsave(hlist_lock, *flags);
1050 }
1051
1052 void __kprobes kretprobe_hash_unlock(struct task_struct *tsk,
1053         unsigned long *flags)
1054 __releases(hlist_lock)
1055 {
1056         unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
1057         raw_spinlock_t *hlist_lock;
1058
1059         hlist_lock = kretprobe_table_lock_ptr(hash);
1060         raw_spin_unlock_irqrestore(hlist_lock, *flags);
1061 }
1062
1063 static void __kprobes kretprobe_table_unlock(unsigned long hash,
1064        unsigned long *flags)
1065 __releases(hlist_lock)
1066 {
1067         raw_spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
1068         raw_spin_unlock_irqrestore(hlist_lock, *flags);
1069 }
1070
1071 /*
1072  * This function is called from finish_task_switch when task tk becomes dead,
1073  * so that we can recycle any function-return probe instances associated
1074  * with this task. These left over instances represent probed functions
1075  * that have been called but will never return.
1076  */
1077 void __kprobes kprobe_flush_task(struct task_struct *tk)
1078 {
1079         struct kretprobe_instance *ri;
1080         struct hlist_head *head, empty_rp;
1081         struct hlist_node *node, *tmp;
1082         unsigned long hash, flags = 0;
1083
1084         if (unlikely(!kprobes_initialized))
1085                 /* Early boot.  kretprobe_table_locks not yet initialized. */
1086                 return;
1087
1088         INIT_HLIST_HEAD(&empty_rp);
1089         hash = hash_ptr(tk, KPROBE_HASH_BITS);
1090         head = &kretprobe_inst_table[hash];
1091         kretprobe_table_lock(hash, &flags);
1092         hlist_for_each_entry_safe(ri, node, tmp, head, hlist) {
1093                 if (ri->task == tk)
1094                         recycle_rp_inst(ri, &empty_rp);
1095         }
1096         kretprobe_table_unlock(hash, &flags);
1097         hlist_for_each_entry_safe(ri, node, tmp, &empty_rp, hlist) {
1098                 hlist_del(&ri->hlist);
1099                 kfree(ri);
1100         }
1101 }
1102
1103 static inline void free_rp_inst(struct kretprobe *rp)
1104 {
1105         struct kretprobe_instance *ri;
1106         struct hlist_node *pos, *next;
1107
1108         hlist_for_each_entry_safe(ri, pos, next, &rp->free_instances, hlist) {
1109                 hlist_del(&ri->hlist);
1110                 kfree(ri);
1111         }
1112 }
1113
1114 static void __kprobes cleanup_rp_inst(struct kretprobe *rp)
1115 {
1116         unsigned long flags, hash;
1117         struct kretprobe_instance *ri;
1118         struct hlist_node *pos, *next;
1119         struct hlist_head *head;
1120
1121         /* No race here */
1122         for (hash = 0; hash < KPROBE_TABLE_SIZE; hash++) {
1123                 kretprobe_table_lock(hash, &flags);
1124                 head = &kretprobe_inst_table[hash];
1125                 hlist_for_each_entry_safe(ri, pos, next, head, hlist) {
1126                         if (ri->rp == rp)
1127                                 ri->rp = NULL;
1128                 }
1129                 kretprobe_table_unlock(hash, &flags);
1130         }
1131         free_rp_inst(rp);
1132 }
1133
1134 /*
1135 * Add the new probe to ap->list. Fail if this is the
1136 * second jprobe at the address - two jprobes can't coexist
1137 */
1138 static int __kprobes add_new_kprobe(struct kprobe *ap, struct kprobe *p)
1139 {
1140         BUG_ON(kprobe_gone(ap) || kprobe_gone(p));
1141
1142         if (p->break_handler || p->post_handler)
1143                 unoptimize_kprobe(ap, true);    /* Fall back to normal kprobe */
1144
1145         if (p->break_handler) {
1146                 if (ap->break_handler)
1147                         return -EEXIST;
1148                 list_add_tail_rcu(&p->list, &ap->list);
1149                 ap->break_handler = aggr_break_handler;
1150         } else
1151                 list_add_rcu(&p->list, &ap->list);
1152         if (p->post_handler && !ap->post_handler)
1153                 ap->post_handler = aggr_post_handler;
1154
1155         return 0;
1156 }
1157
1158 /*
1159  * Fill in the required fields of the "manager kprobe". Replace the
1160  * earlier kprobe in the hlist with the manager kprobe
1161  */
1162 static void __kprobes init_aggr_kprobe(struct kprobe *ap, struct kprobe *p)
1163 {
1164         /* Copy p's insn slot to ap */
1165         copy_kprobe(p, ap);
1166         flush_insn_slot(ap);
1167         ap->addr = p->addr;
1168         ap->flags = p->flags & ~KPROBE_FLAG_OPTIMIZED;
1169         ap->pre_handler = aggr_pre_handler;
1170         ap->fault_handler = aggr_fault_handler;
1171         /* We don't care the kprobe which has gone. */
1172         if (p->post_handler && !kprobe_gone(p))
1173                 ap->post_handler = aggr_post_handler;
1174         if (p->break_handler && !kprobe_gone(p))
1175                 ap->break_handler = aggr_break_handler;
1176
1177         INIT_LIST_HEAD(&ap->list);
1178         INIT_HLIST_NODE(&ap->hlist);
1179
1180         list_add_rcu(&p->list, &ap->list);
1181         hlist_replace_rcu(&p->hlist, &ap->hlist);
1182 }
1183
1184 /*
1185  * This is the second or subsequent kprobe at the address - handle
1186  * the intricacies
1187  */
1188 static int __kprobes register_aggr_kprobe(struct kprobe *orig_p,
1189                                           struct kprobe *p)
1190 {
1191         int ret = 0;
1192         struct kprobe *ap = orig_p;
1193
1194         /* For preparing optimization, jump_label_text_reserved() is called */
1195         jump_label_lock();
1196         /*
1197          * Get online CPUs to avoid text_mutex deadlock.with stop machine,
1198          * which is invoked by unoptimize_kprobe() in add_new_kprobe()
1199          */
1200         get_online_cpus();
1201         mutex_lock(&text_mutex);
1202
1203         if (!kprobe_aggrprobe(orig_p)) {
1204                 /* If orig_p is not an aggr_kprobe, create new aggr_kprobe. */
1205                 ap = alloc_aggr_kprobe(orig_p);
1206                 if (!ap) {
1207                         ret = -ENOMEM;
1208                         goto out;
1209                 }
1210                 init_aggr_kprobe(ap, orig_p);
1211         } else if (kprobe_unused(ap))
1212                 /* This probe is going to die. Rescue it */
1213                 reuse_unused_kprobe(ap);
1214
1215         if (kprobe_gone(ap)) {
1216                 /*
1217                  * Attempting to insert new probe at the same location that
1218                  * had a probe in the module vaddr area which already
1219                  * freed. So, the instruction slot has already been
1220                  * released. We need a new slot for the new probe.
1221                  */
1222                 ret = arch_prepare_kprobe(ap);
1223                 if (ret)
1224                         /*
1225                          * Even if fail to allocate new slot, don't need to
1226                          * free aggr_probe. It will be used next time, or
1227                          * freed by unregister_kprobe.
1228                          */
1229                         goto out;
1230
1231                 /* Prepare optimized instructions if possible. */
1232                 prepare_optimized_kprobe(ap);
1233
1234                 /*
1235                  * Clear gone flag to prevent allocating new slot again, and
1236                  * set disabled flag because it is not armed yet.
1237                  */
1238                 ap->flags = (ap->flags & ~KPROBE_FLAG_GONE)
1239                             | KPROBE_FLAG_DISABLED;
1240         }
1241
1242         /* Copy ap's insn slot to p */
1243         copy_kprobe(ap, p);
1244         ret = add_new_kprobe(ap, p);
1245
1246 out:
1247         mutex_unlock(&text_mutex);
1248         put_online_cpus();
1249         jump_label_unlock();
1250
1251         if (ret == 0 && kprobe_disabled(ap) && !kprobe_disabled(p)) {
1252                 ap->flags &= ~KPROBE_FLAG_DISABLED;
1253                 if (!kprobes_all_disarmed)
1254                         /* Arm the breakpoint again. */
1255                         arm_kprobe(ap);
1256         }
1257         return ret;
1258 }
1259
1260 static int __kprobes in_kprobes_functions(unsigned long addr)
1261 {
1262         struct kprobe_blackpoint *kb;
1263
1264         if (addr >= (unsigned long)__kprobes_text_start &&
1265             addr < (unsigned long)__kprobes_text_end)
1266                 return -EINVAL;
1267         /*
1268          * If there exists a kprobe_blacklist, verify and
1269          * fail any probe registration in the prohibited area
1270          */
1271         for (kb = kprobe_blacklist; kb->name != NULL; kb++) {
1272                 if (kb->start_addr) {
1273                         if (addr >= kb->start_addr &&
1274                             addr < (kb->start_addr + kb->range))
1275                                 return -EINVAL;
1276                 }
1277         }
1278         return 0;
1279 }
1280
1281 /*
1282  * If we have a symbol_name argument, look it up and add the offset field
1283  * to it. This way, we can specify a relative address to a symbol.
1284  * This returns encoded errors if it fails to look up symbol or invalid
1285  * combination of parameters.
1286  */
1287 static kprobe_opcode_t __kprobes *kprobe_addr(struct kprobe *p)
1288 {
1289         kprobe_opcode_t *addr = p->addr;
1290
1291         if ((p->symbol_name && p->addr) ||
1292             (!p->symbol_name && !p->addr))
1293                 goto invalid;
1294
1295         if (p->symbol_name) {
1296                 kprobe_lookup_name(p->symbol_name, addr);
1297                 if (!addr)
1298                         return ERR_PTR(-ENOENT);
1299         }
1300
1301         addr = (kprobe_opcode_t *)(((char *)addr) + p->offset);
1302         if (addr)
1303                 return addr;
1304
1305 invalid:
1306         return ERR_PTR(-EINVAL);
1307 }
1308
1309 /* Check passed kprobe is valid and return kprobe in kprobe_table. */
1310 static struct kprobe * __kprobes __get_valid_kprobe(struct kprobe *p)
1311 {
1312         struct kprobe *ap, *list_p;
1313
1314         ap = get_kprobe(p->addr);
1315         if (unlikely(!ap))
1316                 return NULL;
1317
1318         if (p != ap) {
1319                 list_for_each_entry_rcu(list_p, &ap->list, list)
1320                         if (list_p == p)
1321                         /* kprobe p is a valid probe */
1322                                 goto valid;
1323                 return NULL;
1324         }
1325 valid:
1326         return ap;
1327 }
1328
1329 /* Return error if the kprobe is being re-registered */
1330 static inline int check_kprobe_rereg(struct kprobe *p)
1331 {
1332         int ret = 0;
1333
1334         mutex_lock(&kprobe_mutex);
1335         if (__get_valid_kprobe(p))
1336                 ret = -EINVAL;
1337         mutex_unlock(&kprobe_mutex);
1338
1339         return ret;
1340 }
1341
1342 static __kprobes int check_kprobe_address_safe(struct kprobe *p,
1343                                                struct module **probed_mod)
1344 {
1345         int ret = 0;
1346
1347         jump_label_lock();
1348         preempt_disable();
1349
1350         /* Ensure it is not in reserved area nor out of text */
1351         if (!kernel_text_address((unsigned long) p->addr) ||
1352             in_kprobes_functions((unsigned long) p->addr) ||
1353             ftrace_text_reserved(p->addr, p->addr) ||
1354             jump_label_text_reserved(p->addr, p->addr)) {
1355                 ret = -EINVAL;
1356                 goto out;
1357         }
1358
1359         /* Check if are we probing a module */
1360         *probed_mod = __module_text_address((unsigned long) p->addr);
1361         if (*probed_mod) {
1362                 /*
1363                  * We must hold a refcount of the probed module while updating
1364                  * its code to prohibit unexpected unloading.
1365                  */
1366                 if (unlikely(!try_module_get(*probed_mod))) {
1367                         ret = -ENOENT;
1368                         goto out;
1369                 }
1370
1371                 /*
1372                  * If the module freed .init.text, we couldn't insert
1373                  * kprobes in there.
1374                  */
1375                 if (within_module_init((unsigned long)p->addr, *probed_mod) &&
1376                     (*probed_mod)->state != MODULE_STATE_COMING) {
1377                         module_put(*probed_mod);
1378                         *probed_mod = NULL;
1379                         ret = -ENOENT;
1380                 }
1381         }
1382 out:
1383         preempt_enable();
1384         jump_label_unlock();
1385
1386         return ret;
1387 }
1388
1389 int __kprobes register_kprobe(struct kprobe *p)
1390 {
1391         int ret;
1392         struct kprobe *old_p;
1393         struct module *probed_mod;
1394         kprobe_opcode_t *addr;
1395
1396         /* Adjust probe address from symbol */
1397         addr = kprobe_addr(p);
1398         if (IS_ERR(addr))
1399                 return PTR_ERR(addr);
1400         p->addr = addr;
1401
1402         ret = check_kprobe_rereg(p);
1403         if (ret)
1404                 return ret;
1405
1406         /* User can pass only KPROBE_FLAG_DISABLED to register_kprobe */
1407         p->flags &= KPROBE_FLAG_DISABLED;
1408         p->nmissed = 0;
1409         INIT_LIST_HEAD(&p->list);
1410
1411         ret = check_kprobe_address_safe(p, &probed_mod);
1412         if (ret)
1413                 return ret;
1414
1415         mutex_lock(&kprobe_mutex);
1416
1417         old_p = get_kprobe(p->addr);
1418         if (old_p) {
1419                 /* Since this may unoptimize old_p, locking text_mutex. */
1420                 ret = register_aggr_kprobe(old_p, p);
1421                 goto out;
1422         }
1423
1424         mutex_lock(&text_mutex);        /* Avoiding text modification */
1425         ret = arch_prepare_kprobe(p);
1426         mutex_unlock(&text_mutex);
1427         if (ret)
1428                 goto out;
1429
1430         INIT_HLIST_NODE(&p->hlist);
1431         hlist_add_head_rcu(&p->hlist,
1432                        &kprobe_table[hash_ptr(p->addr, KPROBE_HASH_BITS)]);
1433
1434         if (!kprobes_all_disarmed && !kprobe_disabled(p))
1435                 arm_kprobe(p);
1436
1437         /* Try to optimize kprobe */
1438         try_to_optimize_kprobe(p);
1439
1440 out:
1441         mutex_unlock(&kprobe_mutex);
1442
1443         if (probed_mod)
1444                 module_put(probed_mod);
1445
1446         return ret;
1447 }
1448 EXPORT_SYMBOL_GPL(register_kprobe);
1449
1450 /* Check if all probes on the aggrprobe are disabled */
1451 static int __kprobes aggr_kprobe_disabled(struct kprobe *ap)
1452 {
1453         struct kprobe *kp;
1454
1455         list_for_each_entry_rcu(kp, &ap->list, list)
1456                 if (!kprobe_disabled(kp))
1457                         /*
1458                          * There is an active probe on the list.
1459                          * We can't disable this ap.
1460                          */
1461                         return 0;
1462
1463         return 1;
1464 }
1465
1466 /* Disable one kprobe: Make sure called under kprobe_mutex is locked */
1467 static struct kprobe *__kprobes __disable_kprobe(struct kprobe *p)
1468 {
1469         struct kprobe *orig_p;
1470
1471         /* Get an original kprobe for return */
1472         orig_p = __get_valid_kprobe(p);
1473         if (unlikely(orig_p == NULL))
1474                 return NULL;
1475
1476         if (!kprobe_disabled(p)) {
1477                 /* Disable probe if it is a child probe */
1478                 if (p != orig_p)
1479                         p->flags |= KPROBE_FLAG_DISABLED;
1480
1481                 /* Try to disarm and disable this/parent probe */
1482                 if (p == orig_p || aggr_kprobe_disabled(orig_p)) {
1483                         disarm_kprobe(orig_p);
1484                         orig_p->flags |= KPROBE_FLAG_DISABLED;
1485                 }
1486         }
1487
1488         return orig_p;
1489 }
1490
1491 /*
1492  * Unregister a kprobe without a scheduler synchronization.
1493  */
1494 static int __kprobes __unregister_kprobe_top(struct kprobe *p)
1495 {
1496         struct kprobe *ap, *list_p;
1497
1498         /* Disable kprobe. This will disarm it if needed. */
1499         ap = __disable_kprobe(p);
1500         if (ap == NULL)
1501                 return -EINVAL;
1502
1503         if (ap == p)
1504                 /*
1505                  * This probe is an independent(and non-optimized) kprobe
1506                  * (not an aggrprobe). Remove from the hash list.
1507                  */
1508                 goto disarmed;
1509
1510         /* Following process expects this probe is an aggrprobe */
1511         WARN_ON(!kprobe_aggrprobe(ap));
1512
1513         if (list_is_singular(&ap->list) && kprobe_disarmed(ap))
1514                 /*
1515                  * !disarmed could be happen if the probe is under delayed
1516                  * unoptimizing.
1517                  */
1518                 goto disarmed;
1519         else {
1520                 /* If disabling probe has special handlers, update aggrprobe */
1521                 if (p->break_handler && !kprobe_gone(p))
1522                         ap->break_handler = NULL;
1523                 if (p->post_handler && !kprobe_gone(p)) {
1524                         list_for_each_entry_rcu(list_p, &ap->list, list) {
1525                                 if ((list_p != p) && (list_p->post_handler))
1526                                         goto noclean;
1527                         }
1528                         ap->post_handler = NULL;
1529                 }
1530 noclean:
1531                 /*
1532                  * Remove from the aggrprobe: this path will do nothing in
1533                  * __unregister_kprobe_bottom().
1534                  */
1535                 list_del_rcu(&p->list);
1536                 if (!kprobe_disabled(ap) && !kprobes_all_disarmed)
1537                         /*
1538                          * Try to optimize this probe again, because post
1539                          * handler may have been changed.
1540                          */
1541                         optimize_kprobe(ap);
1542         }
1543         return 0;
1544
1545 disarmed:
1546         BUG_ON(!kprobe_disarmed(ap));
1547         hlist_del_rcu(&ap->hlist);
1548         return 0;
1549 }
1550
1551 static void __kprobes __unregister_kprobe_bottom(struct kprobe *p)
1552 {
1553         struct kprobe *ap;
1554
1555         if (list_empty(&p->list))
1556                 /* This is an independent kprobe */
1557                 arch_remove_kprobe(p);
1558         else if (list_is_singular(&p->list)) {
1559                 /* This is the last child of an aggrprobe */
1560                 ap = list_entry(p->list.next, struct kprobe, list);
1561                 list_del(&p->list);
1562                 free_aggr_kprobe(ap);
1563         }
1564         /* Otherwise, do nothing. */
1565 }
1566
1567 int __kprobes register_kprobes(struct kprobe **kps, int num)
1568 {
1569         int i, ret = 0;
1570
1571         if (num <= 0)
1572                 return -EINVAL;
1573         for (i = 0; i < num; i++) {
1574                 ret = register_kprobe(kps[i]);
1575                 if (ret < 0) {
1576                         if (i > 0)
1577                                 unregister_kprobes(kps, i);
1578                         break;
1579                 }
1580         }
1581         return ret;
1582 }
1583 EXPORT_SYMBOL_GPL(register_kprobes);
1584
1585 void __kprobes unregister_kprobe(struct kprobe *p)
1586 {
1587         unregister_kprobes(&p, 1);
1588 }
1589 EXPORT_SYMBOL_GPL(unregister_kprobe);
1590
1591 void __kprobes unregister_kprobes(struct kprobe **kps, int num)
1592 {
1593         int i;
1594
1595         if (num <= 0)
1596                 return;
1597         mutex_lock(&kprobe_mutex);
1598         for (i = 0; i < num; i++)
1599                 if (__unregister_kprobe_top(kps[i]) < 0)
1600                         kps[i]->addr = NULL;
1601         mutex_unlock(&kprobe_mutex);
1602
1603         synchronize_sched();
1604         for (i = 0; i < num; i++)
1605                 if (kps[i]->addr)
1606                         __unregister_kprobe_bottom(kps[i]);
1607 }
1608 EXPORT_SYMBOL_GPL(unregister_kprobes);
1609
1610 static struct notifier_block kprobe_exceptions_nb = {
1611         .notifier_call = kprobe_exceptions_notify,
1612         .priority = 0x7fffffff /* we need to be notified first */
1613 };
1614
1615 unsigned long __weak arch_deref_entry_point(void *entry)
1616 {
1617         return (unsigned long)entry;
1618 }
1619
1620 int __kprobes register_jprobes(struct jprobe **jps, int num)
1621 {
1622         struct jprobe *jp;
1623         int ret = 0, i;
1624
1625         if (num <= 0)
1626                 return -EINVAL;
1627         for (i = 0; i < num; i++) {
1628                 unsigned long addr, offset;
1629                 jp = jps[i];
1630                 addr = arch_deref_entry_point(jp->entry);
1631
1632                 /* Verify probepoint is a function entry point */
1633                 if (kallsyms_lookup_size_offset(addr, NULL, &offset) &&
1634                     offset == 0) {
1635                         jp->kp.pre_handler = setjmp_pre_handler;
1636                         jp->kp.break_handler = longjmp_break_handler;
1637                         ret = register_kprobe(&jp->kp);
1638                 } else
1639                         ret = -EINVAL;
1640
1641                 if (ret < 0) {
1642                         if (i > 0)
1643                                 unregister_jprobes(jps, i);
1644                         break;
1645                 }
1646         }
1647         return ret;
1648 }
1649 EXPORT_SYMBOL_GPL(register_jprobes);
1650
1651 int __kprobes register_jprobe(struct jprobe *jp)
1652 {
1653         return register_jprobes(&jp, 1);
1654 }
1655 EXPORT_SYMBOL_GPL(register_jprobe);
1656
1657 void __kprobes unregister_jprobe(struct jprobe *jp)
1658 {
1659         unregister_jprobes(&jp, 1);
1660 }
1661 EXPORT_SYMBOL_GPL(unregister_jprobe);
1662
1663 void __kprobes unregister_jprobes(struct jprobe **jps, int num)
1664 {
1665         int i;
1666
1667         if (num <= 0)
1668                 return;
1669         mutex_lock(&kprobe_mutex);
1670         for (i = 0; i < num; i++)
1671                 if (__unregister_kprobe_top(&jps[i]->kp) < 0)
1672                         jps[i]->kp.addr = NULL;
1673         mutex_unlock(&kprobe_mutex);
1674
1675         synchronize_sched();
1676         for (i = 0; i < num; i++) {
1677                 if (jps[i]->kp.addr)
1678                         __unregister_kprobe_bottom(&jps[i]->kp);
1679         }
1680 }
1681 EXPORT_SYMBOL_GPL(unregister_jprobes);
1682
1683 #ifdef CONFIG_KRETPROBES
1684 /*
1685  * This kprobe pre_handler is registered with every kretprobe. When probe
1686  * hits it will set up the return probe.
1687  */
1688 static int __kprobes pre_handler_kretprobe(struct kprobe *p,
1689                                            struct pt_regs *regs)
1690 {
1691         struct kretprobe *rp = container_of(p, struct kretprobe, kp);
1692         unsigned long hash, flags = 0;
1693         struct kretprobe_instance *ri;
1694
1695         /*TODO: consider to only swap the RA after the last pre_handler fired */
1696         hash = hash_ptr(current, KPROBE_HASH_BITS);
1697         raw_spin_lock_irqsave(&rp->lock, flags);
1698         if (!hlist_empty(&rp->free_instances)) {
1699                 ri = hlist_entry(rp->free_instances.first,
1700                                 struct kretprobe_instance, hlist);
1701                 hlist_del(&ri->hlist);
1702                 raw_spin_unlock_irqrestore(&rp->lock, flags);
1703
1704                 ri->rp = rp;
1705                 ri->task = current;
1706
1707                 if (rp->entry_handler && rp->entry_handler(ri, regs)) {
1708                         raw_spin_lock_irqsave(&rp->lock, flags);
1709                         hlist_add_head(&ri->hlist, &rp->free_instances);
1710                         raw_spin_unlock_irqrestore(&rp->lock, flags);
1711                         return 0;
1712                 }
1713
1714                 arch_prepare_kretprobe(ri, regs);
1715
1716                 /* XXX(hch): why is there no hlist_move_head? */
1717                 INIT_HLIST_NODE(&ri->hlist);
1718                 kretprobe_table_lock(hash, &flags);
1719                 hlist_add_head(&ri->hlist, &kretprobe_inst_table[hash]);
1720                 kretprobe_table_unlock(hash, &flags);
1721         } else {
1722                 rp->nmissed++;
1723                 raw_spin_unlock_irqrestore(&rp->lock, flags);
1724         }
1725         return 0;
1726 }
1727
1728 int __kprobes register_kretprobe(struct kretprobe *rp)
1729 {
1730         int ret = 0;
1731         struct kretprobe_instance *inst;
1732         int i;
1733         void *addr;
1734
1735         if (kretprobe_blacklist_size) {
1736                 addr = kprobe_addr(&rp->kp);
1737                 if (IS_ERR(addr))
1738                         return PTR_ERR(addr);
1739
1740                 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
1741                         if (kretprobe_blacklist[i].addr == addr)
1742                                 return -EINVAL;
1743                 }
1744         }
1745
1746         rp->kp.pre_handler = pre_handler_kretprobe;
1747         rp->kp.post_handler = NULL;
1748         rp->kp.fault_handler = NULL;
1749         rp->kp.break_handler = NULL;
1750
1751         /* Pre-allocate memory for max kretprobe instances */
1752         if (rp->maxactive <= 0) {
1753 #ifdef CONFIG_PREEMPT
1754                 rp->maxactive = max_t(unsigned int, 10, 2*num_possible_cpus());
1755 #else
1756                 rp->maxactive = num_possible_cpus();
1757 #endif
1758         }
1759         raw_spin_lock_init(&rp->lock);
1760         INIT_HLIST_HEAD(&rp->free_instances);
1761         for (i = 0; i < rp->maxactive; i++) {
1762                 inst = kmalloc(sizeof(struct kretprobe_instance) +
1763                                rp->data_size, GFP_KERNEL);
1764                 if (inst == NULL) {
1765                         free_rp_inst(rp);
1766                         return -ENOMEM;
1767                 }
1768                 INIT_HLIST_NODE(&inst->hlist);
1769                 hlist_add_head(&inst->hlist, &rp->free_instances);
1770         }
1771
1772         rp->nmissed = 0;
1773         /* Establish function entry probe point */
1774         ret = register_kprobe(&rp->kp);
1775         if (ret != 0)
1776                 free_rp_inst(rp);
1777         return ret;
1778 }
1779 EXPORT_SYMBOL_GPL(register_kretprobe);
1780
1781 int __kprobes register_kretprobes(struct kretprobe **rps, int num)
1782 {
1783         int ret = 0, i;
1784
1785         if (num <= 0)
1786                 return -EINVAL;
1787         for (i = 0; i < num; i++) {
1788                 ret = register_kretprobe(rps[i]);
1789                 if (ret < 0) {
1790                         if (i > 0)
1791                                 unregister_kretprobes(rps, i);
1792                         break;
1793                 }
1794         }
1795         return ret;
1796 }
1797 EXPORT_SYMBOL_GPL(register_kretprobes);
1798
1799 void __kprobes unregister_kretprobe(struct kretprobe *rp)
1800 {
1801         unregister_kretprobes(&rp, 1);
1802 }
1803 EXPORT_SYMBOL_GPL(unregister_kretprobe);
1804
1805 void __kprobes unregister_kretprobes(struct kretprobe **rps, int num)
1806 {
1807         int i;
1808
1809         if (num <= 0)
1810                 return;
1811         mutex_lock(&kprobe_mutex);
1812         for (i = 0; i < num; i++)
1813                 if (__unregister_kprobe_top(&rps[i]->kp) < 0)
1814                         rps[i]->kp.addr = NULL;
1815         mutex_unlock(&kprobe_mutex);
1816
1817         synchronize_sched();
1818         for (i = 0; i < num; i++) {
1819                 if (rps[i]->kp.addr) {
1820                         __unregister_kprobe_bottom(&rps[i]->kp);
1821                         cleanup_rp_inst(rps[i]);
1822                 }
1823         }
1824 }
1825 EXPORT_SYMBOL_GPL(unregister_kretprobes);
1826
1827 #else /* CONFIG_KRETPROBES */
1828 int __kprobes register_kretprobe(struct kretprobe *rp)
1829 {
1830         return -ENOSYS;
1831 }
1832 EXPORT_SYMBOL_GPL(register_kretprobe);
1833
1834 int __kprobes register_kretprobes(struct kretprobe **rps, int num)
1835 {
1836         return -ENOSYS;
1837 }
1838 EXPORT_SYMBOL_GPL(register_kretprobes);
1839
1840 void __kprobes unregister_kretprobe(struct kretprobe *rp)
1841 {
1842 }
1843 EXPORT_SYMBOL_GPL(unregister_kretprobe);
1844
1845 void __kprobes unregister_kretprobes(struct kretprobe **rps, int num)
1846 {
1847 }
1848 EXPORT_SYMBOL_GPL(unregister_kretprobes);
1849
1850 static int __kprobes pre_handler_kretprobe(struct kprobe *p,
1851                                            struct pt_regs *regs)
1852 {
1853         return 0;
1854 }
1855
1856 #endif /* CONFIG_KRETPROBES */
1857
1858 /* Set the kprobe gone and remove its instruction buffer. */
1859 static void __kprobes kill_kprobe(struct kprobe *p)
1860 {
1861         struct kprobe *kp;
1862
1863         p->flags |= KPROBE_FLAG_GONE;
1864         if (kprobe_aggrprobe(p)) {
1865                 /*
1866                  * If this is an aggr_kprobe, we have to list all the
1867                  * chained probes and mark them GONE.
1868                  */
1869                 list_for_each_entry_rcu(kp, &p->list, list)
1870                         kp->flags |= KPROBE_FLAG_GONE;
1871                 p->post_handler = NULL;
1872                 p->break_handler = NULL;
1873                 kill_optimized_kprobe(p);
1874         }
1875         /*
1876          * Here, we can remove insn_slot safely, because no thread calls
1877          * the original probed function (which will be freed soon) any more.
1878          */
1879         arch_remove_kprobe(p);
1880 }
1881
1882 /* Disable one kprobe */
1883 int __kprobes disable_kprobe(struct kprobe *kp)
1884 {
1885         int ret = 0;
1886
1887         mutex_lock(&kprobe_mutex);
1888
1889         /* Disable this kprobe */
1890         if (__disable_kprobe(kp) == NULL)
1891                 ret = -EINVAL;
1892
1893         mutex_unlock(&kprobe_mutex);
1894         return ret;
1895 }
1896 EXPORT_SYMBOL_GPL(disable_kprobe);
1897
1898 /* Enable one kprobe */
1899 int __kprobes enable_kprobe(struct kprobe *kp)
1900 {
1901         int ret = 0;
1902         struct kprobe *p;
1903
1904         mutex_lock(&kprobe_mutex);
1905
1906         /* Check whether specified probe is valid. */
1907         p = __get_valid_kprobe(kp);
1908         if (unlikely(p == NULL)) {
1909                 ret = -EINVAL;
1910                 goto out;
1911         }
1912
1913         if (kprobe_gone(kp)) {
1914                 /* This kprobe has gone, we couldn't enable it. */
1915                 ret = -EINVAL;
1916                 goto out;
1917         }
1918
1919         if (p != kp)
1920                 kp->flags &= ~KPROBE_FLAG_DISABLED;
1921
1922         if (!kprobes_all_disarmed && kprobe_disabled(p)) {
1923                 p->flags &= ~KPROBE_FLAG_DISABLED;
1924                 arm_kprobe(p);
1925         }
1926 out:
1927         mutex_unlock(&kprobe_mutex);
1928         return ret;
1929 }
1930 EXPORT_SYMBOL_GPL(enable_kprobe);
1931
1932 void __kprobes dump_kprobe(struct kprobe *kp)
1933 {
1934         printk(KERN_WARNING "Dumping kprobe:\n");
1935         printk(KERN_WARNING "Name: %s\nAddress: %p\nOffset: %x\n",
1936                kp->symbol_name, kp->addr, kp->offset);
1937 }
1938
1939 /* Module notifier call back, checking kprobes on the module */
1940 static int __kprobes kprobes_module_callback(struct notifier_block *nb,
1941                                              unsigned long val, void *data)
1942 {
1943         struct module *mod = data;
1944         struct hlist_head *head;
1945         struct hlist_node *node;
1946         struct kprobe *p;
1947         unsigned int i;
1948         int checkcore = (val == MODULE_STATE_GOING);
1949
1950         if (val != MODULE_STATE_GOING && val != MODULE_STATE_LIVE)
1951                 return NOTIFY_DONE;
1952
1953         /*
1954          * When MODULE_STATE_GOING was notified, both of module .text and
1955          * .init.text sections would be freed. When MODULE_STATE_LIVE was
1956          * notified, only .init.text section would be freed. We need to
1957          * disable kprobes which have been inserted in the sections.
1958          */
1959         mutex_lock(&kprobe_mutex);
1960         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
1961                 head = &kprobe_table[i];
1962                 hlist_for_each_entry_rcu(p, node, head, hlist)
1963                         if (within_module_init((unsigned long)p->addr, mod) ||
1964                             (checkcore &&
1965                              within_module_core((unsigned long)p->addr, mod))) {
1966                                 /*
1967                                  * The vaddr this probe is installed will soon
1968                                  * be vfreed buy not synced to disk. Hence,
1969                                  * disarming the breakpoint isn't needed.
1970                                  */
1971                                 kill_kprobe(p);
1972                         }
1973         }
1974         mutex_unlock(&kprobe_mutex);
1975         return NOTIFY_DONE;
1976 }
1977
1978 static struct notifier_block kprobe_module_nb = {
1979         .notifier_call = kprobes_module_callback,
1980         .priority = 0
1981 };
1982
1983 static int __init init_kprobes(void)
1984 {
1985         int i, err = 0;
1986         unsigned long offset = 0, size = 0;
1987         char *modname, namebuf[128];
1988         const char *symbol_name;
1989         void *addr;
1990         struct kprobe_blackpoint *kb;
1991
1992         /* FIXME allocate the probe table, currently defined statically */
1993         /* initialize all list heads */
1994         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
1995                 INIT_HLIST_HEAD(&kprobe_table[i]);
1996                 INIT_HLIST_HEAD(&kretprobe_inst_table[i]);
1997                 raw_spin_lock_init(&(kretprobe_table_locks[i].lock));
1998         }
1999
2000         /*
2001          * Lookup and populate the kprobe_blacklist.
2002          *
2003          * Unlike the kretprobe blacklist, we'll need to determine
2004          * the range of addresses that belong to the said functions,
2005          * since a kprobe need not necessarily be at the beginning
2006          * of a function.
2007          */
2008         for (kb = kprobe_blacklist; kb->name != NULL; kb++) {
2009                 kprobe_lookup_name(kb->name, addr);
2010                 if (!addr)
2011                         continue;
2012
2013                 kb->start_addr = (unsigned long)addr;
2014                 symbol_name = kallsyms_lookup(kb->start_addr,
2015                                 &size, &offset, &modname, namebuf);
2016                 if (!symbol_name)
2017                         kb->range = 0;
2018                 else
2019                         kb->range = size;
2020         }
2021
2022         if (kretprobe_blacklist_size) {
2023                 /* lookup the function address from its name */
2024                 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
2025                         kprobe_lookup_name(kretprobe_blacklist[i].name,
2026                                            kretprobe_blacklist[i].addr);
2027                         if (!kretprobe_blacklist[i].addr)
2028                                 printk("kretprobe: lookup failed: %s\n",
2029                                        kretprobe_blacklist[i].name);
2030                 }
2031         }
2032
2033 #if defined(CONFIG_OPTPROBES)
2034 #if defined(__ARCH_WANT_KPROBES_INSN_SLOT)
2035         /* Init kprobe_optinsn_slots */
2036         kprobe_optinsn_slots.insn_size = MAX_OPTINSN_SIZE;
2037 #endif
2038         /* By default, kprobes can be optimized */
2039         kprobes_allow_optimization = true;
2040 #endif
2041
2042         /* By default, kprobes are armed */
2043         kprobes_all_disarmed = false;
2044
2045         err = arch_init_kprobes();
2046         if (!err)
2047                 err = register_die_notifier(&kprobe_exceptions_nb);
2048         if (!err)
2049                 err = register_module_notifier(&kprobe_module_nb);
2050
2051         kprobes_initialized = (err == 0);
2052
2053         if (!err)
2054                 init_test_probes();
2055         return err;
2056 }
2057
2058 #ifdef CONFIG_DEBUG_FS
2059 static void __kprobes report_probe(struct seq_file *pi, struct kprobe *p,
2060                 const char *sym, int offset, char *modname, struct kprobe *pp)
2061 {
2062         char *kprobe_type;
2063
2064         if (p->pre_handler == pre_handler_kretprobe)
2065                 kprobe_type = "r";
2066         else if (p->pre_handler == setjmp_pre_handler)
2067                 kprobe_type = "j";
2068         else
2069                 kprobe_type = "k";
2070
2071         if (sym)
2072                 seq_printf(pi, "%p  %s  %s+0x%x  %s ",
2073                         p->addr, kprobe_type, sym, offset,
2074                         (modname ? modname : " "));
2075         else
2076                 seq_printf(pi, "%p  %s  %p ",
2077                         p->addr, kprobe_type, p->addr);
2078
2079         if (!pp)
2080                 pp = p;
2081         seq_printf(pi, "%s%s%s\n",
2082                 (kprobe_gone(p) ? "[GONE]" : ""),
2083                 ((kprobe_disabled(p) && !kprobe_gone(p)) ?  "[DISABLED]" : ""),
2084                 (kprobe_optimized(pp) ? "[OPTIMIZED]" : ""));
2085 }
2086
2087 static void __kprobes *kprobe_seq_start(struct seq_file *f, loff_t *pos)
2088 {
2089         return (*pos < KPROBE_TABLE_SIZE) ? pos : NULL;
2090 }
2091
2092 static void __kprobes *kprobe_seq_next(struct seq_file *f, void *v, loff_t *pos)
2093 {
2094         (*pos)++;
2095         if (*pos >= KPROBE_TABLE_SIZE)
2096                 return NULL;
2097         return pos;
2098 }
2099
2100 static void __kprobes kprobe_seq_stop(struct seq_file *f, void *v)
2101 {
2102         /* Nothing to do */
2103 }
2104
2105 static int __kprobes show_kprobe_addr(struct seq_file *pi, void *v)
2106 {
2107         struct hlist_head *head;
2108         struct hlist_node *node;
2109         struct kprobe *p, *kp;
2110         const char *sym = NULL;
2111         unsigned int i = *(loff_t *) v;
2112         unsigned long offset = 0;
2113         char *modname, namebuf[128];
2114
2115         head = &kprobe_table[i];
2116         preempt_disable();
2117         hlist_for_each_entry_rcu(p, node, head, hlist) {
2118                 sym = kallsyms_lookup((unsigned long)p->addr, NULL,
2119                                         &offset, &modname, namebuf);
2120                 if (kprobe_aggrprobe(p)) {
2121                         list_for_each_entry_rcu(kp, &p->list, list)
2122                                 report_probe(pi, kp, sym, offset, modname, p);
2123                 } else
2124                         report_probe(pi, p, sym, offset, modname, NULL);
2125         }
2126         preempt_enable();
2127         return 0;
2128 }
2129
2130 static const struct seq_operations kprobes_seq_ops = {
2131         .start = kprobe_seq_start,
2132         .next  = kprobe_seq_next,
2133         .stop  = kprobe_seq_stop,
2134         .show  = show_kprobe_addr
2135 };
2136
2137 static int __kprobes kprobes_open(struct inode *inode, struct file *filp)
2138 {
2139         return seq_open(filp, &kprobes_seq_ops);
2140 }
2141
2142 static const struct file_operations debugfs_kprobes_operations = {
2143         .open           = kprobes_open,
2144         .read           = seq_read,
2145         .llseek         = seq_lseek,
2146         .release        = seq_release,
2147 };
2148
2149 static void __kprobes arm_all_kprobes(void)
2150 {
2151         struct hlist_head *head;
2152         struct hlist_node *node;
2153         struct kprobe *p;
2154         unsigned int i;
2155
2156         mutex_lock(&kprobe_mutex);
2157
2158         /* If kprobes are armed, just return */
2159         if (!kprobes_all_disarmed)
2160                 goto already_enabled;
2161
2162         /* Arming kprobes doesn't optimize kprobe itself */
2163         mutex_lock(&text_mutex);
2164         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2165                 head = &kprobe_table[i];
2166                 hlist_for_each_entry_rcu(p, node, head, hlist)
2167                         if (!kprobe_disabled(p))
2168                                 __arm_kprobe(p);
2169         }
2170         mutex_unlock(&text_mutex);
2171
2172         kprobes_all_disarmed = false;
2173         printk(KERN_INFO "Kprobes globally enabled\n");
2174
2175 already_enabled:
2176         mutex_unlock(&kprobe_mutex);
2177         return;
2178 }
2179
2180 static void __kprobes disarm_all_kprobes(void)
2181 {
2182         struct hlist_head *head;
2183         struct hlist_node *node;
2184         struct kprobe *p;
2185         unsigned int i;
2186
2187         mutex_lock(&kprobe_mutex);
2188
2189         /* If kprobes are already disarmed, just return */
2190         if (kprobes_all_disarmed) {
2191                 mutex_unlock(&kprobe_mutex);
2192                 return;
2193         }
2194
2195         kprobes_all_disarmed = true;
2196         printk(KERN_INFO "Kprobes globally disabled\n");
2197
2198         mutex_lock(&text_mutex);
2199         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2200                 head = &kprobe_table[i];
2201                 hlist_for_each_entry_rcu(p, node, head, hlist) {
2202                         if (!arch_trampoline_kprobe(p) && !kprobe_disabled(p))
2203                                 __disarm_kprobe(p, false);
2204                 }
2205         }
2206         mutex_unlock(&text_mutex);
2207         mutex_unlock(&kprobe_mutex);
2208
2209         /* Wait for disarming all kprobes by optimizer */
2210         wait_for_kprobe_optimizer();
2211 }
2212
2213 /*
2214  * XXX: The debugfs bool file interface doesn't allow for callbacks
2215  * when the bool state is switched. We can reuse that facility when
2216  * available
2217  */
2218 static ssize_t read_enabled_file_bool(struct file *file,
2219                char __user *user_buf, size_t count, loff_t *ppos)
2220 {
2221         char buf[3];
2222
2223         if (!kprobes_all_disarmed)
2224                 buf[0] = '1';
2225         else
2226                 buf[0] = '0';
2227         buf[1] = '\n';
2228         buf[2] = 0x00;
2229         return simple_read_from_buffer(user_buf, count, ppos, buf, 2);
2230 }
2231
2232 static ssize_t write_enabled_file_bool(struct file *file,
2233                const char __user *user_buf, size_t count, loff_t *ppos)
2234 {
2235         char buf[32];
2236         size_t buf_size;
2237
2238         buf_size = min(count, (sizeof(buf)-1));
2239         if (copy_from_user(buf, user_buf, buf_size))
2240                 return -EFAULT;
2241
2242         switch (buf[0]) {
2243         case 'y':
2244         case 'Y':
2245         case '1':
2246                 arm_all_kprobes();
2247                 break;
2248         case 'n':
2249         case 'N':
2250         case '0':
2251                 disarm_all_kprobes();
2252                 break;
2253         }
2254
2255         return count;
2256 }
2257
2258 static const struct file_operations fops_kp = {
2259         .read =         read_enabled_file_bool,
2260         .write =        write_enabled_file_bool,
2261         .llseek =       default_llseek,
2262 };
2263
2264 static int __kprobes debugfs_kprobe_init(void)
2265 {
2266         struct dentry *dir, *file;
2267         unsigned int value = 1;
2268
2269         dir = debugfs_create_dir("kprobes", NULL);
2270         if (!dir)
2271                 return -ENOMEM;
2272
2273         file = debugfs_create_file("list", 0444, dir, NULL,
2274                                 &debugfs_kprobes_operations);
2275         if (!file) {
2276                 debugfs_remove(dir);
2277                 return -ENOMEM;
2278         }
2279
2280         file = debugfs_create_file("enabled", 0600, dir,
2281                                         &value, &fops_kp);
2282         if (!file) {
2283                 debugfs_remove(dir);
2284                 return -ENOMEM;
2285         }
2286
2287         return 0;
2288 }
2289
2290 late_initcall(debugfs_kprobe_init);
2291 #endif /* CONFIG_DEBUG_FS */
2292
2293 module_init(init_kprobes);
2294
2295 /* defined in arch/.../kernel/kprobes.c */
2296 EXPORT_SYMBOL_GPL(jprobe_return);