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/sections.h>
  53 #include <asm/cacheflush.h>
  54 #include <asm/errno.h>
  55 #include <linux/uaccess.h>
  56
  57 #define KPROBE_HASH_BITS 6
  58 #define KPROBE_TABLE_SIZE (1 << KPROBE_HASH_BITS)
  59
  60
  61 static int kprobes_initialized;
  62 static struct hlist_head kprobe_table[KPROBE_TABLE_SIZE];
  63 static struct hlist_head kretprobe_inst_table[KPROBE_TABLE_SIZE];
  64
  65 /* NOTE: change this value only with kprobe_mutex held */
  66 static bool kprobes_all_disarmed;
  67
  68 /* This protects kprobe_table and optimizing_list */
  69 static DEFINE_MUTEX(kprobe_mutex);
  70 static DEFINE_PER_CPU(struct kprobe *, kprobe_instance) = NULL;
  71 static struct {
  72         raw_spinlock_t lock ____cacheline_aligned_in_smp;
  73 } kretprobe_table_locks[KPROBE_TABLE_SIZE];
  74
  75 kprobe_opcode_t * __weak kprobe_lookup_name(const char *name,
  76                                         unsigned int __unused)
  77 {
  78         return ((kprobe_opcode_t *)(kallsyms_lookup_name(name)));
  79 }
  80
  81 static raw_spinlock_t *kretprobe_table_lock_ptr(unsigned long hash)
  82 {
  83         return &(kretprobe_table_locks[hash].lock);
  84 }
  85
  86 /* Blacklist -- list of struct kprobe_blacklist_entry */
  87 static LIST_HEAD(kprobe_blacklist);
  88
  89 #ifdef __ARCH_WANT_KPROBES_INSN_SLOT
  90 /*
  91  * kprobe->ainsn.insn points to the copy of the instruction to be
  92  * single-stepped. x86_64, POWER4 and above have no-exec support and
  93  * stepping on the instruction on a vmalloced/kmalloced/data page
  94  * is a recipe for disaster
  95  */
  96 struct kprobe_insn_page {
  97         struct list_head list;
  98         kprobe_opcode_t *insns;         /* Page of instruction slots */
  99         struct kprobe_insn_cache *cache;
 100         int nused;
 101         int ngarbage;
 102         char slot_used[];
 103 };
 104
 105 #define KPROBE_INSN_PAGE_SIZE(slots)                    \
 106         (offsetof(struct kprobe_insn_page, slot_used) + \
 107          (sizeof(char) * (slots)))
 108
 109 static int slots_per_page(struct kprobe_insn_cache *c)
 110 {
 111         return PAGE_SIZE/(c->insn_size * sizeof(kprobe_opcode_t));
 112 }
 113
 114 enum kprobe_slot_state {
 115         SLOT_CLEAN = 0,
 116         SLOT_DIRTY = 1,
 117         SLOT_USED = 2,
 118 };
 119
 120 static void *alloc_insn_page(void)
 121 {
 122         return module_alloc(PAGE_SIZE);
 123 }
 124
 125 void __weak free_insn_page(void *page)
 126 {
 127         module_memfree(page);
 128 }
 129
 130 struct kprobe_insn_cache kprobe_insn_slots = {
 131         .mutex = __MUTEX_INITIALIZER(kprobe_insn_slots.mutex),
 132         .alloc = alloc_insn_page,
 133         .free = free_insn_page,
 134         .pages = LIST_HEAD_INIT(kprobe_insn_slots.pages),
 135         .insn_size = MAX_INSN_SIZE,
 136         .nr_garbage = 0,
 137 };
 138 static int collect_garbage_slots(struct kprobe_insn_cache *c);
 139
 140 /**
 141  * __get_insn_slot() - Find a slot on an executable page for an instruction.
 142  * We allocate an executable page if there's no room on existing ones.
 143  */
 144 kprobe_opcode_t *__get_insn_slot(struct kprobe_insn_cache *c)
 145 {
 146         struct kprobe_insn_page *kip;
 147         kprobe_opcode_t *slot = NULL;
 148
 149         /* Since the slot array is not protected by rcu, we need a mutex */
 150         mutex_lock(&c->mutex);
 151  retry:
 152         rcu_read_lock();
 153         list_for_each_entry_rcu(kip, &c->pages, list) {
 154                 if (kip->nused < slots_per_page(c)) {
 155                         int i;
 156                         for (i = 0; i < slots_per_page(c); i++) {
 157                                 if (kip->slot_used[i] == SLOT_CLEAN) {
 158                                         kip->slot_used[i] = SLOT_USED;
 159                                         kip->nused++;
 160                                         slot = kip->insns + (i * c->insn_size);
 161                                         rcu_read_unlock();
 162                                         goto out;
 163                                 }
 164                         }
 165                         /* kip->nused is broken. Fix it. */
 166                         kip->nused = slots_per_page(c);
 167                         WARN_ON(1);
 168                 }
 169         }
 170         rcu_read_unlock();
 171
 172         /* If there are any garbage slots, collect it and try again. */
 173         if (c->nr_garbage && collect_garbage_slots(c) == 0)
 174                 goto retry;
 175
 176         /* All out of space.  Need to allocate a new page. */
 177         kip = kmalloc(KPROBE_INSN_PAGE_SIZE(slots_per_page(c)), GFP_KERNEL);
 178         if (!kip)
 179                 goto out;
 180
 181         /*
 182          * Use module_alloc so this page is within +/- 2GB of where the
 183          * kernel image and loaded module images reside. This is required
 184          * so x86_64 can correctly handle the %rip-relative fixups.
 185          */
 186         kip->insns = c->alloc();
 187         if (!kip->insns) {
 188                 kfree(kip);
 189                 goto out;
 190         }
 191         INIT_LIST_HEAD(&kip->list);
 192         memset(kip->slot_used, SLOT_CLEAN, slots_per_page(c));
 193         kip->slot_used[0] = SLOT_USED;
 194         kip->nused = 1;
 195         kip->ngarbage = 0;
 196         kip->cache = c;
 197         list_add_rcu(&kip->list, &c->pages);
 198         slot = kip->insns;
 199 out:
 200         mutex_unlock(&c->mutex);
 201         return slot;
 202 }
 203
 204 /* Return 1 if all garbages are collected, otherwise 0. */
 205 static int collect_one_slot(struct kprobe_insn_page *kip, int idx)
 206 {
 207         kip->slot_used[idx] = SLOT_CLEAN;
 208         kip->nused--;
 209         if (kip->nused == 0) {
 210                 /*
 211                  * Page is no longer in use.  Free it unless
 212                  * it's the last one.  We keep the last one
 213                  * so as not to have to set it up again the
 214                  * next time somebody inserts a probe.
 215                  */
 216                 if (!list_is_singular(&kip->list)) {
 217                         list_del_rcu(&kip->list);
 218                         synchronize_rcu();
 219                         kip->cache->free(kip->insns);
 220                         kfree(kip);
 221                 }
 222                 return 1;
 223         }
 224         return 0;
 225 }
 226
 227 static int collect_garbage_slots(struct kprobe_insn_cache *c)
 228 {
 229         struct kprobe_insn_page *kip, *next;
 230
 231         /* Ensure no-one is interrupted on the garbages */
 232         synchronize_sched();
 233
 234         list_for_each_entry_safe(kip, next, &c->pages, list) {
 235                 int i;
 236                 if (kip->ngarbage == 0)
 237                         continue;
 238                 kip->ngarbage = 0;      /* we will collect all garbages */
 239                 for (i = 0; i < slots_per_page(c); i++) {
 240                         if (kip->slot_used[i] == SLOT_DIRTY && collect_one_slot(kip, i))
 241                                 break;
 242                 }
 243         }
 244         c->nr_garbage = 0;
 245         return 0;
 246 }
 247
 248 void __free_insn_slot(struct kprobe_insn_cache *c,
 249                       kprobe_opcode_t *slot, int dirty)
 250 {
 251         struct kprobe_insn_page *kip;
 252         long idx;
 253
 254         mutex_lock(&c->mutex);
 255         rcu_read_lock();
 256         list_for_each_entry_rcu(kip, &c->pages, list) {
 257                 idx = ((long)slot - (long)kip->insns) /
 258                         (c->insn_size * sizeof(kprobe_opcode_t));
 259                 if (idx >= 0 && idx < slots_per_page(c))
 260                         goto out;
 261         }
 262         /* Could not find this slot. */
 263         WARN_ON(1);
 264         kip = NULL;
 265 out:
 266         rcu_read_unlock();
 267         /* Mark and sweep: this may sleep */
 268         if (kip) {
 269                 /* Check double free */
 270                 WARN_ON(kip->slot_used[idx] != SLOT_USED);
 271                 if (dirty) {
 272                         kip->slot_used[idx] = SLOT_DIRTY;
 273                         kip->ngarbage++;
 274                         if (++c->nr_garbage > slots_per_page(c))
 275                                 collect_garbage_slots(c);
 276                 } else {
 277                         collect_one_slot(kip, idx);
 278                 }
 279         }
 280         mutex_unlock(&c->mutex);
 281 }
 282
 283 /*
 284  * Check given address is on the page of kprobe instruction slots.
 285  * This will be used for checking whether the address on a stack
 286  * is on a text area or not.
 287  */
 288 bool __is_insn_slot_addr(struct kprobe_insn_cache *c, unsigned long addr)
 289 {
 290         struct kprobe_insn_page *kip;
 291         bool ret = false;
 292
 293         rcu_read_lock();
 294         list_for_each_entry_rcu(kip, &c->pages, list) {
 295                 if (addr >= (unsigned long)kip->insns &&
 296                     addr < (unsigned long)kip->insns + PAGE_SIZE) {
 297                         ret = true;
 298                         break;
 299                 }
 300         }
 301         rcu_read_unlock();
 302
 303         return ret;
 304 }
 305
 306 #ifdef CONFIG_OPTPROBES
 307 /* For optimized_kprobe buffer */
 308 struct kprobe_insn_cache kprobe_optinsn_slots = {
 309         .mutex = __MUTEX_INITIALIZER(kprobe_optinsn_slots.mutex),
 310         .alloc = alloc_insn_page,
 311         .free = free_insn_page,
 312         .pages = LIST_HEAD_INIT(kprobe_optinsn_slots.pages),
 313         /* .insn_size is initialized later */
 314         .nr_garbage = 0,
 315 };
 316 #endif
 317 #endif
 318
 319 /* We have preemption disabled.. so it is safe to use __ versions */
 320 static inline void set_kprobe_instance(struct kprobe *kp)
 321 {
 322         __this_cpu_write(kprobe_instance, kp);
 323 }
 324
 325 static inline void reset_kprobe_instance(void)
 326 {
 327         __this_cpu_write(kprobe_instance, NULL);
 328 }
 329
 330 /*
 331  * This routine is called either:
 332  *      - under the kprobe_mutex - during kprobe_[un]register()
 333  *                              OR
 334  *      - with preemption disabled - from arch/xxx/kernel/kprobes.c
 335  */
 336 struct kprobe *get_kprobe(void *addr)
 337 {
 338         struct hlist_head *head;
 339         struct kprobe *p;
 340
 341         head = &kprobe_table[hash_ptr(addr, KPROBE_HASH_BITS)];
 342         hlist_for_each_entry_rcu(p, head, hlist) {
 343                 if (p->addr == addr)
 344                         return p;
 345         }
 346
 347         return NULL;
 348 }
 349 NOKPROBE_SYMBOL(get_kprobe);
 350
 351 static int aggr_pre_handler(struct kprobe *p, struct pt_regs *regs);
 352
 353 /* Return true if the kprobe is an aggregator */
 354 static inline int kprobe_aggrprobe(struct kprobe *p)
 355 {
 356         return p->pre_handler == aggr_pre_handler;
 357 }
 358
 359 /* Return true(!0) if the kprobe is unused */
 360 static inline int kprobe_unused(struct kprobe *p)
 361 {
 362         return kprobe_aggrprobe(p) && kprobe_disabled(p) &&
 363                list_empty(&p->list);
 364 }
 365
 366 /*
 367  * Keep all fields in the kprobe consistent
 368  */
 369 static inline void copy_kprobe(struct kprobe *ap, struct kprobe *p)
 370 {
 371         memcpy(&p->opcode, &ap->opcode, sizeof(kprobe_opcode_t));
 372         memcpy(&p->ainsn, &ap->ainsn, sizeof(struct arch_specific_insn));
 373 }
 374
 375 #ifdef CONFIG_OPTPROBES
 376 /* NOTE: change this value only with kprobe_mutex held */
 377 static bool kprobes_allow_optimization;
 378
 379 /*
 380  * Call all pre_handler on the list, but ignores its return value.
 381  * This must be called from arch-dep optimized caller.
 382  */
 383 void opt_pre_handler(struct kprobe *p, struct pt_regs *regs)
 384 {
 385         struct kprobe *kp;
 386
 387         list_for_each_entry_rcu(kp, &p->list, list) {
 388                 if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
 389                         set_kprobe_instance(kp);
 390                         kp->pre_handler(kp, regs);
 391                 }
 392                 reset_kprobe_instance();
 393         }
 394 }
 395 NOKPROBE_SYMBOL(opt_pre_handler);
 396
 397 /* Free optimized instructions and optimized_kprobe */
 398 static void free_aggr_kprobe(struct kprobe *p)
 399 {
 400         struct optimized_kprobe *op;
 401
 402         op = container_of(p, struct optimized_kprobe, kp);
 403         arch_remove_optimized_kprobe(op);
 404         arch_remove_kprobe(p);
 405         kfree(op);
 406 }
 407
 408 /* Return true(!0) if the kprobe is ready for optimization. */
 409 static inline int kprobe_optready(struct kprobe *p)
 410 {
 411         struct optimized_kprobe *op;
 412
 413         if (kprobe_aggrprobe(p)) {
 414                 op = container_of(p, struct optimized_kprobe, kp);
 415                 return arch_prepared_optinsn(&op->optinsn);
 416         }
 417
 418         return 0;
 419 }
 420
 421 /* Return true(!0) if the kprobe is disarmed. Note: p must be on hash list */
 422 static inline int kprobe_disarmed(struct kprobe *p)
 423 {
 424         struct optimized_kprobe *op;
 425
 426         /* If kprobe is not aggr/opt probe, just return kprobe is disabled */
 427         if (!kprobe_aggrprobe(p))
 428                 return kprobe_disabled(p);
 429
 430         op = container_of(p, struct optimized_kprobe, kp);
 431
 432         return kprobe_disabled(p) && list_empty(&op->list);
 433 }
 434
 435 /* Return true(!0) if the probe is queued on (un)optimizing lists */
 436 static int kprobe_queued(struct kprobe *p)
 437 {
 438         struct optimized_kprobe *op;
 439
 440         if (kprobe_aggrprobe(p)) {
 441                 op = container_of(p, struct optimized_kprobe, kp);
 442                 if (!list_empty(&op->list))
 443                         return 1;
 444         }
 445         return 0;
 446 }
 447
 448 /*
 449  * Return an optimized kprobe whose optimizing code replaces
 450  * instructions including addr (exclude breakpoint).
 451  */
 452 static struct kprobe *get_optimized_kprobe(unsigned long addr)
 453 {
 454         int i;
 455         struct kprobe *p = NULL;
 456         struct optimized_kprobe *op;
 457
 458         /* Don't check i == 0, since that is a breakpoint case. */
 459         for (i = 1; !p && i < MAX_OPTIMIZED_LENGTH; i++)
 460                 p = get_kprobe((void *)(addr - i));
 461
 462         if (p && kprobe_optready(p)) {
 463                 op = container_of(p, struct optimized_kprobe, kp);
 464                 if (arch_within_optimized_kprobe(op, addr))
 465                         return p;
 466         }
 467
 468         return NULL;
 469 }
 470
 471 /* Optimization staging list, protected by kprobe_mutex */
 472 static LIST_HEAD(optimizing_list);
 473 static LIST_HEAD(unoptimizing_list);
 474 static LIST_HEAD(freeing_list);
 475
 476 static void kprobe_optimizer(struct work_struct *work);
 477 static DECLARE_DELAYED_WORK(optimizing_work, kprobe_optimizer);
 478 #define OPTIMIZE_DELAY 5
 479
 480 /*
 481  * Optimize (replace a breakpoint with a jump) kprobes listed on
 482  * optimizing_list.
 483  */
 484 static void do_optimize_kprobes(void)
 485 {
 486         /*
 487          * The optimization/unoptimization refers online_cpus via
 488          * stop_machine() and cpu-hotplug modifies online_cpus.
 489          * And same time, text_mutex will be held in cpu-hotplug and here.
 490          * This combination can cause a deadlock (cpu-hotplug try to lock
 491          * text_mutex but stop_machine can not be done because online_cpus
 492          * has been changed)
 493          * To avoid this deadlock, caller must have locked cpu hotplug
 494          * for preventing cpu-hotplug outside of text_mutex locking.
 495          */
 496         lockdep_assert_cpus_held();
 497
 498         /* Optimization never be done when disarmed */
 499         if (kprobes_all_disarmed || !kprobes_allow_optimization ||
 500             list_empty(&optimizing_list))
 501                 return;
 502
 503         mutex_lock(&text_mutex);
 504         arch_optimize_kprobes(&optimizing_list);
 505         mutex_unlock(&text_mutex);
 506 }
 507
 508 /*
 509  * Unoptimize (replace a jump with a breakpoint and remove the breakpoint
 510  * if need) kprobes listed on unoptimizing_list.
 511  */
 512 static void do_unoptimize_kprobes(void)
 513 {
 514         struct optimized_kprobe *op, *tmp;
 515
 516         /* See comment in do_optimize_kprobes() */
 517         lockdep_assert_cpus_held();
 518
 519         /* Unoptimization must be done anytime */
 520         if (list_empty(&unoptimizing_list))
 521                 return;
 522
 523         mutex_lock(&text_mutex);
 524         arch_unoptimize_kprobes(&unoptimizing_list, &freeing_list);
 525         /* Loop free_list for disarming */
 526         list_for_each_entry_safe(op, tmp, &freeing_list, list) {
 527                 /* Disarm probes if marked disabled */
 528                 if (kprobe_disabled(&op->kp))
 529                         arch_disarm_kprobe(&op->kp);
 530                 if (kprobe_unused(&op->kp)) {
 531                         /*
 532                          * Remove unused probes from hash list. After waiting
 533                          * for synchronization, these probes are reclaimed.
 534                          * (reclaiming is done by do_free_cleaned_kprobes.)
 535                          */
 536                         hlist_del_rcu(&op->kp.hlist);
 537                 } else
 538                         list_del_init(&op->list);
 539         }
 540         mutex_unlock(&text_mutex);
 541 }
 542
 543 /* Reclaim all kprobes on the free_list */
 544 static void do_free_cleaned_kprobes(void)
 545 {
 546         struct optimized_kprobe *op, *tmp;
 547
 548         list_for_each_entry_safe(op, tmp, &freeing_list, list) {
 549                 BUG_ON(!kprobe_unused(&op->kp));
 550                 list_del_init(&op->list);
 551                 free_aggr_kprobe(&op->kp);
 552         }
 553 }
 554
 555 /* Start optimizer after OPTIMIZE_DELAY passed */
 556 static void kick_kprobe_optimizer(void)
 557 {
 558         schedule_delayed_work(&optimizing_work, OPTIMIZE_DELAY);
 559 }
 560
 561 /* Kprobe jump optimizer */
 562 static void kprobe_optimizer(struct work_struct *work)
 563 {
 564         mutex_lock(&kprobe_mutex);
 565         cpus_read_lock();
 566         /* Lock modules while optimizing kprobes */
 567         mutex_lock(&module_mutex);
 568
 569         /*
 570          * Step 1: Unoptimize kprobes and collect cleaned (unused and disarmed)
 571          * kprobes before waiting for quiesence period.
 572          */
 573         do_unoptimize_kprobes();
 574
 575         /*
 576          * Step 2: Wait for quiesence period to ensure all running interrupts
 577          * are done. Because optprobe may modify multiple instructions
 578          * there is a chance that Nth instruction is interrupted. In that
 579          * case, running interrupt can return to 2nd-Nth byte of jump
 580          * instruction. This wait is for avoiding it.
 581          */
 582         synchronize_sched();
 583
 584         /* Step 3: Optimize kprobes after quiesence period */
 585         do_optimize_kprobes();
 586
 587         /* Step 4: Free cleaned kprobes after quiesence period */
 588         do_free_cleaned_kprobes();
 589
 590         mutex_unlock(&module_mutex);
 591         cpus_read_unlock();
 592         mutex_unlock(&kprobe_mutex);
 593
 594         /* Step 5: Kick optimizer again if needed */
 595         if (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list))
 596                 kick_kprobe_optimizer();
 597 }
 598
 599 /* Wait for completing optimization and unoptimization */
 600 void wait_for_kprobe_optimizer(void)
 601 {
 602         mutex_lock(&kprobe_mutex);
 603
 604         while (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list)) {
 605                 mutex_unlock(&kprobe_mutex);
 606
 607                 /* this will also make optimizing_work execute immmediately */
 608                 flush_delayed_work(&optimizing_work);
 609                 /* @optimizing_work might not have been queued yet, relax */
 610                 cpu_relax();
 611
 612                 mutex_lock(&kprobe_mutex);
 613         }
 614
 615         mutex_unlock(&kprobe_mutex);
 616 }
 617
 618 /* Optimize kprobe if p is ready to be optimized */
 619 static void optimize_kprobe(struct kprobe *p)
 620 {
 621         struct optimized_kprobe *op;
 622
 623         /* Check if the kprobe is disabled or not ready for optimization. */
 624         if (!kprobe_optready(p) || !kprobes_allow_optimization ||
 625             (kprobe_disabled(p) || kprobes_all_disarmed))
 626                 return;
 627
 628         /* Both of break_handler and post_handler are not supported. */
 629         if (p->break_handler || p->post_handler)
 630                 return;
 631
 632         op = container_of(p, struct optimized_kprobe, kp);
 633
 634         /* Check there is no other kprobes at the optimized instructions */
 635         if (arch_check_optimized_kprobe(op) < 0)
 636                 return;
 637
 638         /* Check if it is already optimized. */
 639         if (op->kp.flags & KPROBE_FLAG_OPTIMIZED)
 640                 return;
 641         op->kp.flags |= KPROBE_FLAG_OPTIMIZED;
 642
 643         if (!list_empty(&op->list))
 644                 /* This is under unoptimizing. Just dequeue the probe */
 645                 list_del_init(&op->list);
 646         else {
 647                 list_add(&op->list, &optimizing_list);
 648                 kick_kprobe_optimizer();
 649         }
 650 }
 651
 652 /* Short cut to direct unoptimizing */
 653 static void force_unoptimize_kprobe(struct optimized_kprobe *op)
 654 {
 655         lockdep_assert_cpus_held();
 656         arch_unoptimize_kprobe(op);
 657         if (kprobe_disabled(&op->kp))
 658                 arch_disarm_kprobe(&op->kp);
 659 }
 660
 661 /* Unoptimize a kprobe if p is optimized */
 662 static void unoptimize_kprobe(struct kprobe *p, bool force)
 663 {
 664         struct optimized_kprobe *op;
 665
 666         if (!kprobe_aggrprobe(p) || kprobe_disarmed(p))
 667                 return; /* This is not an optprobe nor optimized */
 668
 669         op = container_of(p, struct optimized_kprobe, kp);
 670         if (!kprobe_optimized(p)) {
 671                 /* Unoptimized or unoptimizing case */
 672                 if (force && !list_empty(&op->list)) {
 673                         /*
 674                          * Only if this is unoptimizing kprobe and forced,
 675                          * forcibly unoptimize it. (No need to unoptimize
 676                          * unoptimized kprobe again :)
 677                          */
 678                         list_del_init(&op->list);
 679                         force_unoptimize_kprobe(op);
 680                 }
 681                 return;
 682         }
 683
 684         op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
 685         if (!list_empty(&op->list)) {
 686                 /* Dequeue from the optimization queue */
 687                 list_del_init(&op->list);
 688                 return;
 689         }
 690         /* Optimized kprobe case */
 691         if (force)
 692                 /* Forcibly update the code: this is a special case */
 693                 force_unoptimize_kprobe(op);
 694         else {
 695                 list_add(&op->list, &unoptimizing_list);
 696                 kick_kprobe_optimizer();
 697         }
 698 }
 699
 700 /* Cancel unoptimizing for reusing */
 701 static void reuse_unused_kprobe(struct kprobe *ap)
 702 {
 703         struct optimized_kprobe *op;
 704
 705         BUG_ON(!kprobe_unused(ap));
 706         /*
 707          * Unused kprobe MUST be on the way of delayed unoptimizing (means
 708          * there is still a relative jump) and disabled.
 709          */
 710         op = container_of(ap, struct optimized_kprobe, kp);
 711         if (unlikely(list_empty(&op->list)))
 712                 printk(KERN_WARNING "Warning: found a stray unused "
 713                         "aggrprobe@%p\n", ap->addr);
 714         /* Enable the probe again */
 715         ap->flags &= ~KPROBE_FLAG_DISABLED;
 716         /* Optimize it again (remove from op->list) */
 717         BUG_ON(!kprobe_optready(ap));
 718         optimize_kprobe(ap);
 719 }
 720
 721 /* Remove optimized instructions */
 722 static void kill_optimized_kprobe(struct kprobe *p)
 723 {
 724         struct optimized_kprobe *op;
 725
 726         op = container_of(p, struct optimized_kprobe, kp);
 727         if (!list_empty(&op->list))
 728                 /* Dequeue from the (un)optimization queue */
 729                 list_del_init(&op->list);
 730         op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
 731
 732         if (kprobe_unused(p)) {
 733                 /* Enqueue if it is unused */
 734                 list_add(&op->list, &freeing_list);
 735                 /*
 736                  * Remove unused probes from the hash list. After waiting
 737                  * for synchronization, this probe is reclaimed.
 738                  * (reclaiming is done by do_free_cleaned_kprobes().)
 739                  */
 740                 hlist_del_rcu(&op->kp.hlist);
 741         }
 742
 743         /* Don't touch the code, because it is already freed. */
 744         arch_remove_optimized_kprobe(op);
 745 }
 746
 747 static inline
 748 void __prepare_optimized_kprobe(struct optimized_kprobe *op, struct kprobe *p)
 749 {
 750         if (!kprobe_ftrace(p))
 751                 arch_prepare_optimized_kprobe(op, p);
 752 }
 753
 754 /* Try to prepare optimized instructions */
 755 static void prepare_optimized_kprobe(struct kprobe *p)
 756 {
 757         struct optimized_kprobe *op;
 758
 759         op = container_of(p, struct optimized_kprobe, kp);
 760         __prepare_optimized_kprobe(op, p);
 761 }
 762
 763 /* Allocate new optimized_kprobe and try to prepare optimized instructions */
 764 static struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
 765 {
 766         struct optimized_kprobe *op;
 767
 768         op = kzalloc(sizeof(struct optimized_kprobe), GFP_KERNEL);
 769         if (!op)
 770                 return NULL;
 771
 772         INIT_LIST_HEAD(&op->list);
 773         op->kp.addr = p->addr;
 774         __prepare_optimized_kprobe(op, p);
 775
 776         return &op->kp;
 777 }
 778
 779 static void init_aggr_kprobe(struct kprobe *ap, struct kprobe *p);
 780
 781 /*
 782  * Prepare an optimized_kprobe and optimize it
 783  * NOTE: p must be a normal registered kprobe
 784  */
 785 static void try_to_optimize_kprobe(struct kprobe *p)
 786 {
 787         struct kprobe *ap;
 788         struct optimized_kprobe *op;
 789
 790         /* Impossible to optimize ftrace-based kprobe */
 791         if (kprobe_ftrace(p))
 792                 return;
 793
 794         /* For preparing optimization, jump_label_text_reserved() is called */
 795         cpus_read_lock();
 796         jump_label_lock();
 797         mutex_lock(&text_mutex);
 798
 799         ap = alloc_aggr_kprobe(p);
 800         if (!ap)
 801                 goto out;
 802
 803         op = container_of(ap, struct optimized_kprobe, kp);
 804         if (!arch_prepared_optinsn(&op->optinsn)) {
 805                 /* If failed to setup optimizing, fallback to kprobe */
 806                 arch_remove_optimized_kprobe(op);
 807                 kfree(op);
 808                 goto out;
 809         }
 810
 811         init_aggr_kprobe(ap, p);
 812         optimize_kprobe(ap);    /* This just kicks optimizer thread */
 813
 814 out:
 815         mutex_unlock(&text_mutex);
 816         jump_label_unlock();
 817         cpus_read_unlock();
 818 }
 819
 820 #ifdef CONFIG_SYSCTL
 821 static void optimize_all_kprobes(void)
 822 {
 823         struct hlist_head *head;
 824         struct kprobe *p;
 825         unsigned int i;
 826
 827         mutex_lock(&kprobe_mutex);
 828         /* If optimization is already allowed, just return */
 829         if (kprobes_allow_optimization)
 830                 goto out;
 831
 832         cpus_read_lock();
 833         kprobes_allow_optimization = true;
 834         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
 835                 head = &kprobe_table[i];
 836                 hlist_for_each_entry_rcu(p, head, hlist)
 837                         if (!kprobe_disabled(p))
 838                                 optimize_kprobe(p);
 839         }
 840         cpus_read_unlock();
 841         printk(KERN_INFO "Kprobes globally optimized\n");
 842 out:
 843         mutex_unlock(&kprobe_mutex);
 844 }
 845
 846 static void unoptimize_all_kprobes(void)
 847 {
 848         struct hlist_head *head;
 849         struct kprobe *p;
 850         unsigned int i;
 851
 852         mutex_lock(&kprobe_mutex);
 853         /* If optimization is already prohibited, just return */
 854         if (!kprobes_allow_optimization) {
 855                 mutex_unlock(&kprobe_mutex);
 856                 return;
 857         }
 858
 859         cpus_read_lock();
 860         kprobes_allow_optimization = false;
 861         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
 862                 head = &kprobe_table[i];
 863                 hlist_for_each_entry_rcu(p, head, hlist) {
 864                         if (!kprobe_disabled(p))
 865                                 unoptimize_kprobe(p, false);
 866                 }
 867         }
 868         cpus_read_unlock();
 869         mutex_unlock(&kprobe_mutex);
 870
 871         /* Wait for unoptimizing completion */
 872         wait_for_kprobe_optimizer();
 873         printk(KERN_INFO "Kprobes globally unoptimized\n");
 874 }
 875
 876 static DEFINE_MUTEX(kprobe_sysctl_mutex);
 877 int sysctl_kprobes_optimization;
 878 int proc_kprobes_optimization_handler(struct ctl_table *table, int write,
 879                                       void __user *buffer, size_t *length,
 880                                       loff_t *ppos)
 881 {
 882         int ret;
 883
 884         mutex_lock(&kprobe_sysctl_mutex);
 885         sysctl_kprobes_optimization = kprobes_allow_optimization ? 1 : 0;
 886         ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
 887
 888         if (sysctl_kprobes_optimization)
 889                 optimize_all_kprobes();
 890         else
 891                 unoptimize_all_kprobes();
 892         mutex_unlock(&kprobe_sysctl_mutex);
 893
 894         return ret;
 895 }
 896 #endif /* CONFIG_SYSCTL */
 897
 898 /* Put a breakpoint for a probe. Must be called with text_mutex locked */
 899 static void __arm_kprobe(struct kprobe *p)
 900 {
 901         struct kprobe *_p;
 902
 903         /* Check collision with other optimized kprobes */
 904         _p = get_optimized_kprobe((unsigned long)p->addr);
 905         if (unlikely(_p))
 906                 /* Fallback to unoptimized kprobe */
 907                 unoptimize_kprobe(_p, true);
 908
 909         arch_arm_kprobe(p);
 910         optimize_kprobe(p);     /* Try to optimize (add kprobe to a list) */
 911 }
 912
 913 /* Remove the breakpoint of a probe. Must be called with text_mutex locked */
 914 static void __disarm_kprobe(struct kprobe *p, bool reopt)
 915 {
 916         struct kprobe *_p;
 917
 918         /* Try to unoptimize */
 919         unoptimize_kprobe(p, kprobes_all_disarmed);
 920
 921         if (!kprobe_queued(p)) {
 922                 arch_disarm_kprobe(p);
 923                 /* If another kprobe was blocked, optimize it. */
 924                 _p = get_optimized_kprobe((unsigned long)p->addr);
 925                 if (unlikely(_p) && reopt)
 926                         optimize_kprobe(_p);
 927         }
 928         /* TODO: reoptimize others after unoptimized this probe */
 929 }
 930
 931 #else /* !CONFIG_OPTPROBES */
 932
 933 #define optimize_kprobe(p)                      do {} while (0)
 934 #define unoptimize_kprobe(p, f)                 do {} while (0)
 935 #define kill_optimized_kprobe(p)                do {} while (0)
 936 #define prepare_optimized_kprobe(p)             do {} while (0)
 937 #define try_to_optimize_kprobe(p)               do {} while (0)
 938 #define __arm_kprobe(p)                         arch_arm_kprobe(p)
 939 #define __disarm_kprobe(p, o)                   arch_disarm_kprobe(p)
 940 #define kprobe_disarmed(p)                      kprobe_disabled(p)
 941 #define wait_for_kprobe_optimizer()             do {} while (0)
 942
 943 /* There should be no unused kprobes can be reused without optimization */
 944 static void reuse_unused_kprobe(struct kprobe *ap)
 945 {
 946         printk(KERN_ERR "Error: There should be no unused kprobe here.\n");
 947         BUG_ON(kprobe_unused(ap));
 948 }
 949
 950 static void free_aggr_kprobe(struct kprobe *p)
 951 {
 952         arch_remove_kprobe(p);
 953         kfree(p);
 954 }
 955
 956 static struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
 957 {
 958         return kzalloc(sizeof(struct kprobe), GFP_KERNEL);
 959 }
 960 #endif /* CONFIG_OPTPROBES */
 961
 962 #ifdef CONFIG_KPROBES_ON_FTRACE
 963 static struct ftrace_ops kprobe_ftrace_ops __read_mostly = {
 964         .func = kprobe_ftrace_handler,
 965         .flags = FTRACE_OPS_FL_SAVE_REGS | FTRACE_OPS_FL_IPMODIFY,
 966 };
 967 static int kprobe_ftrace_enabled;
 968
 969 /* Must ensure p->addr is really on ftrace */
 970 static int prepare_kprobe(struct kprobe *p)
 971 {
 972         if (!kprobe_ftrace(p))
 973                 return arch_prepare_kprobe(p);
 974
 975         return arch_prepare_kprobe_ftrace(p);
 976 }
 977
 978 /* Caller must lock kprobe_mutex */
 979 static void arm_kprobe_ftrace(struct kprobe *p)
 980 {
 981         int ret;
 982
 983         ret = ftrace_set_filter_ip(&kprobe_ftrace_ops,
 984                                    (unsigned long)p->addr, 0, 0);
 985         WARN(ret < 0, "Failed to arm kprobe-ftrace at %p (%d)\n", p->addr, ret);
 986         kprobe_ftrace_enabled++;
 987         if (kprobe_ftrace_enabled == 1) {
 988                 ret = register_ftrace_function(&kprobe_ftrace_ops);
 989                 WARN(ret < 0, "Failed to init kprobe-ftrace (%d)\n", ret);
 990         }
 991 }
 992
 993 /* Caller must lock kprobe_mutex */
 994 static void disarm_kprobe_ftrace(struct kprobe *p)
 995 {
 996         int ret;
 997
 998         kprobe_ftrace_enabled--;
 999         if (kprobe_ftrace_enabled == 0) {
1000                 ret = unregister_ftrace_function(&kprobe_ftrace_ops);
1001                 WARN(ret < 0, "Failed to init kprobe-ftrace (%d)\n", ret);
1002         }
1003         ret = ftrace_set_filter_ip(&kprobe_ftrace_ops,
1004                            (unsigned long)p->addr, 1, 0);
1005         WARN(ret < 0, "Failed to disarm kprobe-ftrace at %p (%d)\n", p->addr, ret);
1006 }
1007 #else   /* !CONFIG_KPROBES_ON_FTRACE */
1008 #define prepare_kprobe(p)       arch_prepare_kprobe(p)
1009 #define arm_kprobe_ftrace(p)    do {} while (0)
1010 #define disarm_kprobe_ftrace(p) do {} while (0)
1011 #endif
1012
1013 /* Arm a kprobe with text_mutex */
1014 static void arm_kprobe(struct kprobe *kp)
1015 {
1016         if (unlikely(kprobe_ftrace(kp))) {
1017                 arm_kprobe_ftrace(kp);
1018                 return;
1019         }
1020         cpus_read_lock();
1021         mutex_lock(&text_mutex);
1022         __arm_kprobe(kp);
1023         mutex_unlock(&text_mutex);
1024         cpus_read_unlock();
1025 }
1026
1027 /* Disarm a kprobe with text_mutex */
1028 static void disarm_kprobe(struct kprobe *kp, bool reopt)
1029 {
1030         if (unlikely(kprobe_ftrace(kp))) {
1031                 disarm_kprobe_ftrace(kp);
1032                 return;
1033         }
1034
1035         cpus_read_lock();
1036         mutex_lock(&text_mutex);
1037         __disarm_kprobe(kp, reopt);
1038         mutex_unlock(&text_mutex);
1039         cpus_read_unlock();
1040 }
1041
1042 /*
1043  * Aggregate handlers for multiple kprobes support - these handlers
1044  * take care of invoking the individual kprobe handlers on p->list
1045  */
1046 static int aggr_pre_handler(struct kprobe *p, struct pt_regs *regs)
1047 {
1048         struct kprobe *kp;
1049
1050         list_for_each_entry_rcu(kp, &p->list, list) {
1051                 if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
1052                         set_kprobe_instance(kp);
1053                         if (kp->pre_handler(kp, regs))
1054                                 return 1;
1055                 }
1056                 reset_kprobe_instance();
1057         }
1058         return 0;
1059 }
1060 NOKPROBE_SYMBOL(aggr_pre_handler);
1061
1062 static void aggr_post_handler(struct kprobe *p, struct pt_regs *regs,
1063                               unsigned long flags)
1064 {
1065         struct kprobe *kp;
1066
1067         list_for_each_entry_rcu(kp, &p->list, list) {
1068                 if (kp->post_handler && likely(!kprobe_disabled(kp))) {
1069                         set_kprobe_instance(kp);
1070                         kp->post_handler(kp, regs, flags);
1071                         reset_kprobe_instance();
1072                 }
1073         }
1074 }
1075 NOKPROBE_SYMBOL(aggr_post_handler);
1076
1077 static int aggr_fault_handler(struct kprobe *p, struct pt_regs *regs,
1078                               int trapnr)
1079 {
1080         struct kprobe *cur = __this_cpu_read(kprobe_instance);
1081
1082         /*
1083          * if we faulted "during" the execution of a user specified
1084          * probe handler, invoke just that probe's fault handler
1085          */
1086         if (cur && cur->fault_handler) {
1087                 if (cur->fault_handler(cur, regs, trapnr))
1088                         return 1;
1089         }
1090         return 0;
1091 }
1092 NOKPROBE_SYMBOL(aggr_fault_handler);
1093
1094 static int aggr_break_handler(struct kprobe *p, struct pt_regs *regs)
1095 {
1096         struct kprobe *cur = __this_cpu_read(kprobe_instance);
1097         int ret = 0;
1098
1099         if (cur && cur->break_handler) {
1100                 if (cur->break_handler(cur, regs))
1101                         ret = 1;
1102         }
1103         reset_kprobe_instance();
1104         return ret;
1105 }
1106 NOKPROBE_SYMBOL(aggr_break_handler);
1107
1108 /* Walks the list and increments nmissed count for multiprobe case */
1109 void kprobes_inc_nmissed_count(struct kprobe *p)
1110 {
1111         struct kprobe *kp;
1112         if (!kprobe_aggrprobe(p)) {
1113                 p->nmissed++;
1114         } else {
1115                 list_for_each_entry_rcu(kp, &p->list, list)
1116                         kp->nmissed++;
1117         }
1118         return;
1119 }
1120 NOKPROBE_SYMBOL(kprobes_inc_nmissed_count);
1121
1122 void recycle_rp_inst(struct kretprobe_instance *ri,
1123                      struct hlist_head *head)
1124 {
1125         struct kretprobe *rp = ri->rp;
1126
1127         /* remove rp inst off the rprobe_inst_table */
1128         hlist_del(&ri->hlist);
1129         INIT_HLIST_NODE(&ri->hlist);
1130         if (likely(rp)) {
1131                 raw_spin_lock(&rp->lock);
1132                 hlist_add_head(&ri->hlist, &rp->free_instances);
1133                 raw_spin_unlock(&rp->lock);
1134         } else
1135                 /* Unregistering */
1136                 hlist_add_head(&ri->hlist, head);
1137 }
1138 NOKPROBE_SYMBOL(recycle_rp_inst);
1139
1140 void kretprobe_hash_lock(struct task_struct *tsk,
1141                          struct hlist_head **head, unsigned long *flags)
1142 __acquires(hlist_lock)
1143 {
1144         unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
1145         raw_spinlock_t *hlist_lock;
1146
1147         *head = &kretprobe_inst_table[hash];
1148         hlist_lock = kretprobe_table_lock_ptr(hash);
1149         raw_spin_lock_irqsave(hlist_lock, *flags);
1150 }
1151 NOKPROBE_SYMBOL(kretprobe_hash_lock);
1152
1153 static void kretprobe_table_lock(unsigned long hash,
1154                                  unsigned long *flags)
1155 __acquires(hlist_lock)
1156 {
1157         raw_spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
1158         raw_spin_lock_irqsave(hlist_lock, *flags);
1159 }
1160 NOKPROBE_SYMBOL(kretprobe_table_lock);
1161
1162 void kretprobe_hash_unlock(struct task_struct *tsk,
1163                            unsigned long *flags)
1164 __releases(hlist_lock)
1165 {
1166         unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
1167         raw_spinlock_t *hlist_lock;
1168
1169         hlist_lock = kretprobe_table_lock_ptr(hash);
1170         raw_spin_unlock_irqrestore(hlist_lock, *flags);
1171 }
1172 NOKPROBE_SYMBOL(kretprobe_hash_unlock);
1173
1174 static void kretprobe_table_unlock(unsigned long hash,
1175                                    unsigned long *flags)
1176 __releases(hlist_lock)
1177 {
1178         raw_spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
1179         raw_spin_unlock_irqrestore(hlist_lock, *flags);
1180 }
1181 NOKPROBE_SYMBOL(kretprobe_table_unlock);
1182
1183 /*
1184  * This function is called from finish_task_switch when task tk becomes dead,
1185  * so that we can recycle any function-return probe instances associated
1186  * with this task. These left over instances represent probed functions
1187  * that have been called but will never return.
1188  */
1189 void kprobe_flush_task(struct task_struct *tk)
1190 {
1191         struct kretprobe_instance *ri;
1192         struct hlist_head *head, empty_rp;
1193         struct hlist_node *tmp;
1194         unsigned long hash, flags = 0;
1195
1196         if (unlikely(!kprobes_initialized))
1197                 /* Early boot.  kretprobe_table_locks not yet initialized. */
1198                 return;
1199
1200         INIT_HLIST_HEAD(&empty_rp);
1201         hash = hash_ptr(tk, KPROBE_HASH_BITS);
1202         head = &kretprobe_inst_table[hash];
1203         kretprobe_table_lock(hash, &flags);
1204         hlist_for_each_entry_safe(ri, tmp, head, hlist) {
1205                 if (ri->task == tk)
1206                         recycle_rp_inst(ri, &empty_rp);
1207         }
1208         kretprobe_table_unlock(hash, &flags);
1209         hlist_for_each_entry_safe(ri, tmp, &empty_rp, hlist) {
1210                 hlist_del(&ri->hlist);
1211                 kfree(ri);
1212         }
1213 }
1214 NOKPROBE_SYMBOL(kprobe_flush_task);
1215
1216 static inline void free_rp_inst(struct kretprobe *rp)
1217 {
1218         struct kretprobe_instance *ri;
1219         struct hlist_node *next;
1220
1221         hlist_for_each_entry_safe(ri, next, &rp->free_instances, hlist) {
1222                 hlist_del(&ri->hlist);
1223                 kfree(ri);
1224         }
1225 }
1226
1227 static void cleanup_rp_inst(struct kretprobe *rp)
1228 {
1229         unsigned long flags, hash;
1230         struct kretprobe_instance *ri;
1231         struct hlist_node *next;
1232         struct hlist_head *head;
1233
1234         /* No race here */
1235         for (hash = 0; hash < KPROBE_TABLE_SIZE; hash++) {
1236                 kretprobe_table_lock(hash, &flags);
1237                 head = &kretprobe_inst_table[hash];
1238                 hlist_for_each_entry_safe(ri, next, head, hlist) {
1239                         if (ri->rp == rp)
1240                                 ri->rp = NULL;
1241                 }
1242                 kretprobe_table_unlock(hash, &flags);
1243         }
1244         free_rp_inst(rp);
1245 }
1246 NOKPROBE_SYMBOL(cleanup_rp_inst);
1247
1248 /*
1249 * Add the new probe to ap->list. Fail if this is the
1250 * second jprobe at the address - two jprobes can't coexist
1251 */
1252 static int add_new_kprobe(struct kprobe *ap, struct kprobe *p)
1253 {
1254         BUG_ON(kprobe_gone(ap) || kprobe_gone(p));
1255
1256         if (p->break_handler || p->post_handler)
1257                 unoptimize_kprobe(ap, true);    /* Fall back to normal kprobe */
1258
1259         if (p->break_handler) {
1260                 if (ap->break_handler)
1261                         return -EEXIST;
1262                 list_add_tail_rcu(&p->list, &ap->list);
1263                 ap->break_handler = aggr_break_handler;
1264         } else
1265                 list_add_rcu(&p->list, &ap->list);
1266         if (p->post_handler && !ap->post_handler)
1267                 ap->post_handler = aggr_post_handler;
1268
1269         return 0;
1270 }
1271
1272 /*
1273  * Fill in the required fields of the "manager kprobe". Replace the
1274  * earlier kprobe in the hlist with the manager kprobe
1275  */
1276 static void init_aggr_kprobe(struct kprobe *ap, struct kprobe *p)
1277 {
1278         /* Copy p's insn slot to ap */
1279         copy_kprobe(p, ap);
1280         flush_insn_slot(ap);
1281         ap->addr = p->addr;
1282         ap->flags = p->flags & ~KPROBE_FLAG_OPTIMIZED;
1283         ap->pre_handler = aggr_pre_handler;
1284         ap->fault_handler = aggr_fault_handler;
1285         /* We don't care the kprobe which has gone. */
1286         if (p->post_handler && !kprobe_gone(p))
1287                 ap->post_handler = aggr_post_handler;
1288         if (p->break_handler && !kprobe_gone(p))
1289                 ap->break_handler = aggr_break_handler;
1290
1291         INIT_LIST_HEAD(&ap->list);
1292         INIT_HLIST_NODE(&ap->hlist);
1293
1294         list_add_rcu(&p->list, &ap->list);
1295         hlist_replace_rcu(&p->hlist, &ap->hlist);
1296 }
1297
1298 /*
1299  * This is the second or subsequent kprobe at the address - handle
1300  * the intricacies
1301  */
1302 static int register_aggr_kprobe(struct kprobe *orig_p, struct kprobe *p)
1303 {
1304         int ret = 0;
1305         struct kprobe *ap = orig_p;
1306
1307         cpus_read_lock();
1308
1309         /* For preparing optimization, jump_label_text_reserved() is called */
1310         jump_label_lock();
1311         mutex_lock(&text_mutex);
1312
1313         if (!kprobe_aggrprobe(orig_p)) {
1314                 /* If orig_p is not an aggr_kprobe, create new aggr_kprobe. */
1315                 ap = alloc_aggr_kprobe(orig_p);
1316                 if (!ap) {
1317                         ret = -ENOMEM;
1318                         goto out;
1319                 }
1320                 init_aggr_kprobe(ap, orig_p);
1321         } else if (kprobe_unused(ap))
1322                 /* This probe is going to die. Rescue it */
1323                 reuse_unused_kprobe(ap);
1324
1325         if (kprobe_gone(ap)) {
1326                 /*
1327                  * Attempting to insert new probe at the same location that
1328                  * had a probe in the module vaddr area which already
1329                  * freed. So, the instruction slot has already been
1330                  * released. We need a new slot for the new probe.
1331                  */
1332                 ret = arch_prepare_kprobe(ap);
1333                 if (ret)
1334                         /*
1335                          * Even if fail to allocate new slot, don't need to
1336                          * free aggr_probe. It will be used next time, or
1337                          * freed by unregister_kprobe.
1338                          */
1339                         goto out;
1340
1341                 /* Prepare optimized instructions if possible. */
1342                 prepare_optimized_kprobe(ap);
1343
1344                 /*
1345                  * Clear gone flag to prevent allocating new slot again, and
1346                  * set disabled flag because it is not armed yet.
1347                  */
1348                 ap->flags = (ap->flags & ~KPROBE_FLAG_GONE)
1349                             | KPROBE_FLAG_DISABLED;
1350         }
1351
1352         /* Copy ap's insn slot to p */
1353         copy_kprobe(ap, p);
1354         ret = add_new_kprobe(ap, p);
1355
1356 out:
1357         mutex_unlock(&text_mutex);
1358         jump_label_unlock();
1359         cpus_read_unlock();
1360
1361         if (ret == 0 && kprobe_disabled(ap) && !kprobe_disabled(p)) {
1362                 ap->flags &= ~KPROBE_FLAG_DISABLED;
1363                 if (!kprobes_all_disarmed)
1364                         /* Arm the breakpoint again. */
1365                         arm_kprobe(ap);
1366         }
1367         return ret;
1368 }
1369
1370 bool __weak arch_within_kprobe_blacklist(unsigned long addr)
1371 {
1372         /* The __kprobes marked functions and entry code must not be probed */
1373         return addr >= (unsigned long)__kprobes_text_start &&
1374                addr < (unsigned long)__kprobes_text_end;
1375 }
1376
1377 bool within_kprobe_blacklist(unsigned long addr)
1378 {
1379         struct kprobe_blacklist_entry *ent;
1380
1381         if (arch_within_kprobe_blacklist(addr))
1382                 return true;
1383         /*
1384          * If there exists a kprobe_blacklist, verify and
1385          * fail any probe registration in the prohibited area
1386          */
1387         list_for_each_entry(ent, &kprobe_blacklist, list) {
1388                 if (addr >= ent->start_addr && addr < ent->end_addr)
1389                         return true;
1390         }
1391
1392         return false;
1393 }
1394
1395 /*
1396  * If we have a symbol_name argument, look it up and add the offset field
1397  * to it. This way, we can specify a relative address to a symbol.
1398  * This returns encoded errors if it fails to look up symbol or invalid
1399  * combination of parameters.
1400  */
1401 static kprobe_opcode_t *_kprobe_addr(kprobe_opcode_t *addr,
1402                         const char *symbol_name, unsigned int offset)
1403 {
1404         if ((symbol_name && addr) || (!symbol_name && !addr))
1405                 goto invalid;
1406
1407         if (symbol_name) {
1408                 addr = kprobe_lookup_name(symbol_name, offset);
1409                 if (!addr)
1410                         return ERR_PTR(-ENOENT);
1411         }
1412
1413         addr = (kprobe_opcode_t *)(((char *)addr) + offset);
1414         if (addr)
1415                 return addr;
1416
1417 invalid:
1418         return ERR_PTR(-EINVAL);
1419 }
1420
1421 static kprobe_opcode_t *kprobe_addr(struct kprobe *p)
1422 {
1423         return _kprobe_addr(p->addr, p->symbol_name, p->offset);
1424 }
1425
1426 /* Check passed kprobe is valid and return kprobe in kprobe_table. */
1427 static struct kprobe *__get_valid_kprobe(struct kprobe *p)
1428 {
1429         struct kprobe *ap, *list_p;
1430
1431         ap = get_kprobe(p->addr);
1432         if (unlikely(!ap))
1433                 return NULL;
1434
1435         if (p != ap) {
1436                 list_for_each_entry_rcu(list_p, &ap->list, list)
1437                         if (list_p == p)
1438                         /* kprobe p is a valid probe */
1439                                 goto valid;
1440                 return NULL;
1441         }
1442 valid:
1443         return ap;
1444 }
1445
1446 /* Return error if the kprobe is being re-registered */
1447 static inline int check_kprobe_rereg(struct kprobe *p)
1448 {
1449         int ret = 0;
1450
1451         mutex_lock(&kprobe_mutex);
1452         if (__get_valid_kprobe(p))
1453                 ret = -EINVAL;
1454         mutex_unlock(&kprobe_mutex);
1455
1456         return ret;
1457 }
1458
1459 int __weak arch_check_ftrace_location(struct kprobe *p)
1460 {
1461         unsigned long ftrace_addr;
1462
1463         ftrace_addr = ftrace_location((unsigned long)p->addr);
1464         if (ftrace_addr) {
1465 #ifdef CONFIG_KPROBES_ON_FTRACE
1466                 /* Given address is not on the instruction boundary */
1467                 if ((unsigned long)p->addr != ftrace_addr)
1468                         return -EILSEQ;
1469                 p->flags |= KPROBE_FLAG_FTRACE;
1470 #else   /* !CONFIG_KPROBES_ON_FTRACE */
1471                 return -EINVAL;
1472 #endif
1473         }
1474         return 0;
1475 }
1476
1477 static int check_kprobe_address_safe(struct kprobe *p,
1478                                      struct module **probed_mod)
1479 {
1480         int ret;
1481
1482         ret = arch_check_ftrace_location(p);
1483         if (ret)
1484                 return ret;
1485         jump_label_lock();
1486         preempt_disable();
1487
1488         /* Ensure it is not in reserved area nor out of text */
1489         if (!kernel_text_address((unsigned long) p->addr) ||
1490             within_kprobe_blacklist((unsigned long) p->addr) ||
1491             jump_label_text_reserved(p->addr, p->addr)) {
1492                 ret = -EINVAL;
1493                 goto out;
1494         }
1495
1496         /* Check if are we probing a module */
1497         *probed_mod = __module_text_address((unsigned long) p->addr);
1498         if (*probed_mod) {
1499                 /*
1500                  * We must hold a refcount of the probed module while updating
1501                  * its code to prohibit unexpected unloading.
1502                  */
1503                 if (unlikely(!try_module_get(*probed_mod))) {
1504                         ret = -ENOENT;
1505                         goto out;
1506                 }
1507
1508                 /*
1509                  * If the module freed .init.text, we couldn't insert
1510                  * kprobes in there.
1511                  */
1512                 if (within_module_init((unsigned long)p->addr, *probed_mod) &&
1513                     (*probed_mod)->state != MODULE_STATE_COMING) {
1514                         module_put(*probed_mod);
1515                         *probed_mod = NULL;
1516                         ret = -ENOENT;
1517                 }
1518         }
1519 out:
1520         preempt_enable();
1521         jump_label_unlock();
1522
1523         return ret;
1524 }
1525
1526 int register_kprobe(struct kprobe *p)
1527 {
1528         int ret;
1529         struct kprobe *old_p;
1530         struct module *probed_mod;
1531         kprobe_opcode_t *addr;
1532
1533         /* Adjust probe address from symbol */
1534         addr = kprobe_addr(p);
1535         if (IS_ERR(addr))
1536                 return PTR_ERR(addr);
1537         p->addr = addr;
1538
1539         ret = check_kprobe_rereg(p);
1540         if (ret)
1541                 return ret;
1542
1543         /* User can pass only KPROBE_FLAG_DISABLED to register_kprobe */
1544         p->flags &= KPROBE_FLAG_DISABLED;
1545         p->nmissed = 0;
1546         INIT_LIST_HEAD(&p->list);
1547
1548         ret = check_kprobe_address_safe(p, &probed_mod);
1549         if (ret)
1550                 return ret;
1551
1552         mutex_lock(&kprobe_mutex);
1553
1554         old_p = get_kprobe(p->addr);
1555         if (old_p) {
1556                 /* Since this may unoptimize old_p, locking text_mutex. */
1557                 ret = register_aggr_kprobe(old_p, p);
1558                 goto out;
1559         }
1560
1561         cpus_read_lock();
1562         /* Prevent text modification */
1563         mutex_lock(&text_mutex);
1564         ret = prepare_kprobe(p);
1565         mutex_unlock(&text_mutex);
1566         cpus_read_unlock();
1567         if (ret)
1568                 goto out;
1569
1570         INIT_HLIST_NODE(&p->hlist);
1571         hlist_add_head_rcu(&p->hlist,
1572                        &kprobe_table[hash_ptr(p->addr, KPROBE_HASH_BITS)]);
1573
1574         if (!kprobes_all_disarmed && !kprobe_disabled(p))
1575                 arm_kprobe(p);
1576
1577         /* Try to optimize kprobe */
1578         try_to_optimize_kprobe(p);
1579 out:
1580         mutex_unlock(&kprobe_mutex);
1581
1582         if (probed_mod)
1583                 module_put(probed_mod);
1584
1585         return ret;
1586 }
1587 EXPORT_SYMBOL_GPL(register_kprobe);
1588
1589 /* Check if all probes on the aggrprobe are disabled */
1590 static int aggr_kprobe_disabled(struct kprobe *ap)
1591 {
1592         struct kprobe *kp;
1593
1594         list_for_each_entry_rcu(kp, &ap->list, list)
1595                 if (!kprobe_disabled(kp))
1596                         /*
1597                          * There is an active probe on the list.
1598                          * We can't disable this ap.
1599                          */
1600                         return 0;
1601
1602         return 1;
1603 }
1604
1605 /* Disable one kprobe: Make sure called under kprobe_mutex is locked */
1606 static struct kprobe *__disable_kprobe(struct kprobe *p)
1607 {
1608         struct kprobe *orig_p;
1609
1610         /* Get an original kprobe for return */
1611         orig_p = __get_valid_kprobe(p);
1612         if (unlikely(orig_p == NULL))
1613                 return NULL;
1614
1615         if (!kprobe_disabled(p)) {
1616                 /* Disable probe if it is a child probe */
1617                 if (p != orig_p)
1618                         p->flags |= KPROBE_FLAG_DISABLED;
1619
1620                 /* Try to disarm and disable this/parent probe */
1621                 if (p == orig_p || aggr_kprobe_disabled(orig_p)) {
1622                         /*
1623                          * If kprobes_all_disarmed is set, orig_p
1624                          * should have already been disarmed, so
1625                          * skip unneed disarming process.
1626                          */
1627                         if (!kprobes_all_disarmed)
1628                                 disarm_kprobe(orig_p, true);
1629                         orig_p->flags |= KPROBE_FLAG_DISABLED;
1630                 }
1631         }
1632
1633         return orig_p;
1634 }
1635
1636 /*
1637  * Unregister a kprobe without a scheduler synchronization.
1638  */
1639 static int __unregister_kprobe_top(struct kprobe *p)
1640 {
1641         struct kprobe *ap, *list_p;
1642
1643         /* Disable kprobe. This will disarm it if needed. */
1644         ap = __disable_kprobe(p);
1645         if (ap == NULL)
1646                 return -EINVAL;
1647
1648         if (ap == p)
1649                 /*
1650                  * This probe is an independent(and non-optimized) kprobe
1651                  * (not an aggrprobe). Remove from the hash list.
1652                  */
1653                 goto disarmed;
1654
1655         /* Following process expects this probe is an aggrprobe */
1656         WARN_ON(!kprobe_aggrprobe(ap));
1657
1658         if (list_is_singular(&ap->list) && kprobe_disarmed(ap))
1659                 /*
1660                  * !disarmed could be happen if the probe is under delayed
1661                  * unoptimizing.
1662                  */
1663                 goto disarmed;
1664         else {
1665                 /* If disabling probe has special handlers, update aggrprobe */
1666                 if (p->break_handler && !kprobe_gone(p))
1667                         ap->break_handler = NULL;
1668                 if (p->post_handler && !kprobe_gone(p)) {
1669                         list_for_each_entry_rcu(list_p, &ap->list, list) {
1670                                 if ((list_p != p) && (list_p->post_handler))
1671                                         goto noclean;
1672                         }
1673                         ap->post_handler = NULL;
1674                 }
1675 noclean:
1676                 /*
1677                  * Remove from the aggrprobe: this path will do nothing in
1678                  * __unregister_kprobe_bottom().
1679                  */
1680                 list_del_rcu(&p->list);
1681                 if (!kprobe_disabled(ap) && !kprobes_all_disarmed)
1682                         /*
1683                          * Try to optimize this probe again, because post
1684                          * handler may have been changed.
1685                          */
1686                         optimize_kprobe(ap);
1687         }
1688         return 0;
1689
1690 disarmed:
1691         BUG_ON(!kprobe_disarmed(ap));
1692         hlist_del_rcu(&ap->hlist);
1693         return 0;
1694 }
1695
1696 static void __unregister_kprobe_bottom(struct kprobe *p)
1697 {
1698         struct kprobe *ap;
1699
1700         if (list_empty(&p->list))
1701                 /* This is an independent kprobe */
1702                 arch_remove_kprobe(p);
1703         else if (list_is_singular(&p->list)) {
1704                 /* This is the last child of an aggrprobe */
1705                 ap = list_entry(p->list.next, struct kprobe, list);
1706                 list_del(&p->list);
1707                 free_aggr_kprobe(ap);
1708         }
1709         /* Otherwise, do nothing. */
1710 }
1711
1712 int register_kprobes(struct kprobe **kps, int num)
1713 {
1714         int i, ret = 0;
1715
1716         if (num <= 0)
1717                 return -EINVAL;
1718         for (i = 0; i < num; i++) {
1719                 ret = register_kprobe(kps[i]);
1720                 if (ret < 0) {
1721                         if (i > 0)
1722                                 unregister_kprobes(kps, i);
1723                         break;
1724                 }
1725         }
1726         return ret;
1727 }
1728 EXPORT_SYMBOL_GPL(register_kprobes);
1729
1730 void unregister_kprobe(struct kprobe *p)
1731 {
1732         unregister_kprobes(&p, 1);
1733 }
1734 EXPORT_SYMBOL_GPL(unregister_kprobe);
1735
1736 void unregister_kprobes(struct kprobe **kps, int num)
1737 {
1738         int i;
1739
1740         if (num <= 0)
1741                 return;
1742         mutex_lock(&kprobe_mutex);
1743         for (i = 0; i < num; i++)
1744                 if (__unregister_kprobe_top(kps[i]) < 0)
1745                         kps[i]->addr = NULL;
1746         mutex_unlock(&kprobe_mutex);
1747
1748         synchronize_sched();
1749         for (i = 0; i < num; i++)
1750                 if (kps[i]->addr)
1751                         __unregister_kprobe_bottom(kps[i]);
1752 }
1753 EXPORT_SYMBOL_GPL(unregister_kprobes);
1754
1755 int __weak kprobe_exceptions_notify(struct notifier_block *self,
1756                                         unsigned long val, void *data)
1757 {
1758         return NOTIFY_DONE;
1759 }
1760 NOKPROBE_SYMBOL(kprobe_exceptions_notify);
1761
1762 static struct notifier_block kprobe_exceptions_nb = {
1763         .notifier_call = kprobe_exceptions_notify,
1764         .priority = 0x7fffffff /* we need to be notified first */
1765 };
1766
1767 unsigned long __weak arch_deref_entry_point(void *entry)
1768 {
1769         return (unsigned long)entry;
1770 }
1771
1772 int register_jprobes(struct jprobe **jps, int num)
1773 {
1774         int ret = 0, i;
1775
1776         if (num <= 0)
1777                 return -EINVAL;
1778
1779         for (i = 0; i < num; i++) {
1780                 ret = register_jprobe(jps[i]);
1781
1782                 if (ret < 0) {
1783                         if (i > 0)
1784                                 unregister_jprobes(jps, i);
1785                         break;
1786                 }
1787         }
1788
1789         return ret;
1790 }
1791 EXPORT_SYMBOL_GPL(register_jprobes);
1792
1793 int register_jprobe(struct jprobe *jp)
1794 {
1795         unsigned long addr, offset;
1796         struct kprobe *kp = &jp->kp;
1797
1798         /*
1799          * Verify probepoint as well as the jprobe handler are
1800          * valid function entry points.
1801          */
1802         addr = arch_deref_entry_point(jp->entry);
1803
1804         if (kallsyms_lookup_size_offset(addr, NULL, &offset) && offset == 0 &&
1805             kprobe_on_func_entry(kp->addr, kp->symbol_name, kp->offset)) {
1806                 kp->pre_handler = setjmp_pre_handler;
1807                 kp->break_handler = longjmp_break_handler;
1808                 return register_kprobe(kp);
1809         }
1810
1811         return -EINVAL;
1812 }
1813 EXPORT_SYMBOL_GPL(register_jprobe);
1814
1815 void unregister_jprobe(struct jprobe *jp)
1816 {
1817         unregister_jprobes(&jp, 1);
1818 }
1819 EXPORT_SYMBOL_GPL(unregister_jprobe);
1820
1821 void unregister_jprobes(struct jprobe **jps, int num)
1822 {
1823         int i;
1824
1825         if (num <= 0)
1826                 return;
1827         mutex_lock(&kprobe_mutex);
1828         for (i = 0; i < num; i++)
1829                 if (__unregister_kprobe_top(&jps[i]->kp) < 0)
1830                         jps[i]->kp.addr = NULL;
1831         mutex_unlock(&kprobe_mutex);
1832
1833         synchronize_sched();
1834         for (i = 0; i < num; i++) {
1835                 if (jps[i]->kp.addr)
1836                         __unregister_kprobe_bottom(&jps[i]->kp);
1837         }
1838 }
1839 EXPORT_SYMBOL_GPL(unregister_jprobes);
1840
1841 #ifdef CONFIG_KRETPROBES
1842 /*
1843  * This kprobe pre_handler is registered with every kretprobe. When probe
1844  * hits it will set up the return probe.
1845  */
1846 static int pre_handler_kretprobe(struct kprobe *p, struct pt_regs *regs)
1847 {
1848         struct kretprobe *rp = container_of(p, struct kretprobe, kp);
1849         unsigned long hash, flags = 0;
1850         struct kretprobe_instance *ri;
1851
1852         /*
1853          * To avoid deadlocks, prohibit return probing in NMI contexts,
1854          * just skip the probe and increase the (inexact) 'nmissed'
1855          * statistical counter, so that the user is informed that
1856          * something happened:
1857          */
1858         if (unlikely(in_nmi())) {
1859                 rp->nmissed++;
1860                 return 0;
1861         }
1862
1863         /* TODO: consider to only swap the RA after the last pre_handler fired */
1864         hash = hash_ptr(current, KPROBE_HASH_BITS);
1865         raw_spin_lock_irqsave(&rp->lock, flags);
1866         if (!hlist_empty(&rp->free_instances)) {
1867                 ri = hlist_entry(rp->free_instances.first,
1868                                 struct kretprobe_instance, hlist);
1869                 hlist_del(&ri->hlist);
1870                 raw_spin_unlock_irqrestore(&rp->lock, flags);
1871
1872                 ri->rp = rp;
1873                 ri->task = current;
1874
1875                 if (rp->entry_handler && rp->entry_handler(ri, regs)) {
1876                         raw_spin_lock_irqsave(&rp->lock, flags);
1877                         hlist_add_head(&ri->hlist, &rp->free_instances);
1878                         raw_spin_unlock_irqrestore(&rp->lock, flags);
1879                         return 0;
1880                 }
1881
1882                 arch_prepare_kretprobe(ri, regs);
1883
1884                 /* XXX(hch): why is there no hlist_move_head? */
1885                 INIT_HLIST_NODE(&ri->hlist);
1886                 kretprobe_table_lock(hash, &flags);
1887                 hlist_add_head(&ri->hlist, &kretprobe_inst_table[hash]);
1888                 kretprobe_table_unlock(hash, &flags);
1889         } else {
1890                 rp->nmissed++;
1891                 raw_spin_unlock_irqrestore(&rp->lock, flags);
1892         }
1893         return 0;
1894 }
1895 NOKPROBE_SYMBOL(pre_handler_kretprobe);
1896
1897 bool __weak arch_kprobe_on_func_entry(unsigned long offset)
1898 {
1899         return !offset;
1900 }
1901
1902 bool kprobe_on_func_entry(kprobe_opcode_t *addr, const char *sym, unsigned long offset)
1903 {
1904         kprobe_opcode_t *kp_addr = _kprobe_addr(addr, sym, offset);
1905
1906         if (IS_ERR(kp_addr))
1907                 return false;
1908
1909         if (!kallsyms_lookup_size_offset((unsigned long)kp_addr, NULL, &offset) ||
1910                                                 !arch_kprobe_on_func_entry(offset))
1911                 return false;
1912
1913         return true;
1914 }
1915
1916 int register_kretprobe(struct kretprobe *rp)
1917 {
1918         int ret = 0;
1919         struct kretprobe_instance *inst;
1920         int i;
1921         void *addr;
1922
1923         if (!kprobe_on_func_entry(rp->kp.addr, rp->kp.symbol_name, rp->kp.offset))
1924                 return -EINVAL;
1925
1926         if (kretprobe_blacklist_size) {
1927                 addr = kprobe_addr(&rp->kp);
1928                 if (IS_ERR(addr))
1929                         return PTR_ERR(addr);
1930
1931                 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
1932                         if (kretprobe_blacklist[i].addr == addr)
1933                                 return -EINVAL;
1934                 }
1935         }
1936
1937         rp->kp.pre_handler = pre_handler_kretprobe;
1938         rp->kp.post_handler = NULL;
1939         rp->kp.fault_handler = NULL;
1940         rp->kp.break_handler = NULL;
1941
1942         /* Pre-allocate memory for max kretprobe instances */
1943         if (rp->maxactive <= 0) {
1944 #ifdef CONFIG_PREEMPT
1945                 rp->maxactive = max_t(unsigned int, 10, 2*num_possible_cpus());
1946 #else
1947                 rp->maxactive = num_possible_cpus();
1948 #endif
1949         }
1950         raw_spin_lock_init(&rp->lock);
1951         INIT_HLIST_HEAD(&rp->free_instances);
1952         for (i = 0; i < rp->maxactive; i++) {
1953                 inst = kmalloc(sizeof(struct kretprobe_instance) +
1954                                rp->data_size, GFP_KERNEL);
1955                 if (inst == NULL) {
1956                         free_rp_inst(rp);
1957                         return -ENOMEM;
1958                 }
1959                 INIT_HLIST_NODE(&inst->hlist);
1960                 hlist_add_head(&inst->hlist, &rp->free_instances);
1961         }
1962
1963         rp->nmissed = 0;
1964         /* Establish function entry probe point */
1965         ret = register_kprobe(&rp->kp);
1966         if (ret != 0)
1967                 free_rp_inst(rp);
1968         return ret;
1969 }
1970 EXPORT_SYMBOL_GPL(register_kretprobe);
1971
1972 int register_kretprobes(struct kretprobe **rps, int num)
1973 {
1974         int ret = 0, i;
1975
1976         if (num <= 0)
1977                 return -EINVAL;
1978         for (i = 0; i < num; i++) {
1979                 ret = register_kretprobe(rps[i]);
1980                 if (ret < 0) {
1981                         if (i > 0)
1982                                 unregister_kretprobes(rps, i);
1983                         break;
1984                 }
1985         }
1986         return ret;
1987 }
1988 EXPORT_SYMBOL_GPL(register_kretprobes);
1989
1990 void unregister_kretprobe(struct kretprobe *rp)
1991 {
1992         unregister_kretprobes(&rp, 1);
1993 }
1994 EXPORT_SYMBOL_GPL(unregister_kretprobe);
1995
1996 void unregister_kretprobes(struct kretprobe **rps, int num)
1997 {
1998         int i;
1999
2000         if (num <= 0)
2001                 return;
2002         mutex_lock(&kprobe_mutex);
2003         for (i = 0; i < num; i++)
2004                 if (__unregister_kprobe_top(&rps[i]->kp) < 0)
2005                         rps[i]->kp.addr = NULL;
2006         mutex_unlock(&kprobe_mutex);
2007
2008         synchronize_sched();
2009         for (i = 0; i < num; i++) {
2010                 if (rps[i]->kp.addr) {
2011                         __unregister_kprobe_bottom(&rps[i]->kp);
2012                         cleanup_rp_inst(rps[i]);
2013                 }
2014         }
2015 }
2016 EXPORT_SYMBOL_GPL(unregister_kretprobes);
2017
2018 #else /* CONFIG_KRETPROBES */
2019 int register_kretprobe(struct kretprobe *rp)
2020 {
2021         return -ENOSYS;
2022 }
2023 EXPORT_SYMBOL_GPL(register_kretprobe);
2024
2025 int register_kretprobes(struct kretprobe **rps, int num)
2026 {
2027         return -ENOSYS;
2028 }
2029 EXPORT_SYMBOL_GPL(register_kretprobes);
2030
2031 void unregister_kretprobe(struct kretprobe *rp)
2032 {
2033 }
2034 EXPORT_SYMBOL_GPL(unregister_kretprobe);
2035
2036 void unregister_kretprobes(struct kretprobe **rps, int num)
2037 {
2038 }
2039 EXPORT_SYMBOL_GPL(unregister_kretprobes);
2040
2041 static int pre_handler_kretprobe(struct kprobe *p, struct pt_regs *regs)
2042 {
2043         return 0;
2044 }
2045 NOKPROBE_SYMBOL(pre_handler_kretprobe);
2046
2047 #endif /* CONFIG_KRETPROBES */
2048
2049 /* Set the kprobe gone and remove its instruction buffer. */
2050 static void kill_kprobe(struct kprobe *p)
2051 {
2052         struct kprobe *kp;
2053
2054         p->flags |= KPROBE_FLAG_GONE;
2055         if (kprobe_aggrprobe(p)) {
2056                 /*
2057                  * If this is an aggr_kprobe, we have to list all the
2058                  * chained probes and mark them GONE.
2059                  */
2060                 list_for_each_entry_rcu(kp, &p->list, list)
2061                         kp->flags |= KPROBE_FLAG_GONE;
2062                 p->post_handler = NULL;
2063                 p->break_handler = NULL;
2064                 kill_optimized_kprobe(p);
2065         }
2066         /*
2067          * Here, we can remove insn_slot safely, because no thread calls
2068          * the original probed function (which will be freed soon) any more.
2069          */
2070         arch_remove_kprobe(p);
2071 }
2072
2073 /* Disable one kprobe */
2074 int disable_kprobe(struct kprobe *kp)
2075 {
2076         int ret = 0;
2077
2078         mutex_lock(&kprobe_mutex);
2079
2080         /* Disable this kprobe */
2081         if (__disable_kprobe(kp) == NULL)
2082                 ret = -EINVAL;
2083
2084         mutex_unlock(&kprobe_mutex);
2085         return ret;
2086 }
2087 EXPORT_SYMBOL_GPL(disable_kprobe);
2088
2089 /* Enable one kprobe */
2090 int enable_kprobe(struct kprobe *kp)
2091 {
2092         int ret = 0;
2093         struct kprobe *p;
2094
2095         mutex_lock(&kprobe_mutex);
2096
2097         /* Check whether specified probe is valid. */
2098         p = __get_valid_kprobe(kp);
2099         if (unlikely(p == NULL)) {
2100                 ret = -EINVAL;
2101                 goto out;
2102         }
2103
2104         if (kprobe_gone(kp)) {
2105                 /* This kprobe has gone, we couldn't enable it. */
2106                 ret = -EINVAL;
2107                 goto out;
2108         }
2109
2110         if (p != kp)
2111                 kp->flags &= ~KPROBE_FLAG_DISABLED;
2112
2113         if (!kprobes_all_disarmed && kprobe_disabled(p)) {
2114                 p->flags &= ~KPROBE_FLAG_DISABLED;
2115                 arm_kprobe(p);
2116         }
2117 out:
2118         mutex_unlock(&kprobe_mutex);
2119         return ret;
2120 }
2121 EXPORT_SYMBOL_GPL(enable_kprobe);
2122
2123 void dump_kprobe(struct kprobe *kp)
2124 {
2125         printk(KERN_WARNING "Dumping kprobe:\n");
2126         printk(KERN_WARNING "Name: %s\nAddress: %p\nOffset: %x\n",
2127                kp->symbol_name, kp->addr, kp->offset);
2128 }
2129 NOKPROBE_SYMBOL(dump_kprobe);
2130
2131 /*
2132  * Lookup and populate the kprobe_blacklist.
2133  *
2134  * Unlike the kretprobe blacklist, we'll need to determine
2135  * the range of addresses that belong to the said functions,
2136  * since a kprobe need not necessarily be at the beginning
2137  * of a function.
2138  */
2139 static int __init populate_kprobe_blacklist(unsigned long *start,
2140                                              unsigned long *end)
2141 {
2142         unsigned long *iter;
2143         struct kprobe_blacklist_entry *ent;
2144         unsigned long entry, offset = 0, size = 0;
2145
2146         for (iter = start; iter < end; iter++) {
2147                 entry = arch_deref_entry_point((void *)*iter);
2148
2149                 if (!kernel_text_address(entry) ||
2150                     !kallsyms_lookup_size_offset(entry, &size, &offset)) {
2151                         pr_err("Failed to find blacklist at %p\n",
2152                                 (void *)entry);
2153                         continue;
2154                 }
2155
2156                 ent = kmalloc(sizeof(*ent), GFP_KERNEL);
2157                 if (!ent)
2158                         return -ENOMEM;
2159                 ent->start_addr = entry;
2160                 ent->end_addr = entry + size;
2161                 INIT_LIST_HEAD(&ent->list);
2162                 list_add_tail(&ent->list, &kprobe_blacklist);
2163         }
2164         return 0;
2165 }
2166
2167 /* Module notifier call back, checking kprobes on the module */
2168 static int kprobes_module_callback(struct notifier_block *nb,
2169                                    unsigned long val, void *data)
2170 {
2171         struct module *mod = data;
2172         struct hlist_head *head;
2173         struct kprobe *p;
2174         unsigned int i;
2175         int checkcore = (val == MODULE_STATE_GOING);
2176
2177         if (val != MODULE_STATE_GOING && val != MODULE_STATE_LIVE)
2178                 return NOTIFY_DONE;
2179
2180         /*
2181          * When MODULE_STATE_GOING was notified, both of module .text and
2182          * .init.text sections would be freed. When MODULE_STATE_LIVE was
2183          * notified, only .init.text section would be freed. We need to
2184          * disable kprobes which have been inserted in the sections.
2185          */
2186         mutex_lock(&kprobe_mutex);
2187         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2188                 head = &kprobe_table[i];
2189                 hlist_for_each_entry_rcu(p, head, hlist)
2190                         if (within_module_init((unsigned long)p->addr, mod) ||
2191                             (checkcore &&
2192                              within_module_core((unsigned long)p->addr, mod))) {
2193                                 /*
2194                                  * The vaddr this probe is installed will soon
2195                                  * be vfreed buy not synced to disk. Hence,
2196                                  * disarming the breakpoint isn't needed.
2197                                  *
2198                                  * Note, this will also move any optimized probes
2199                                  * that are pending to be removed from their
2200                                  * corresponding lists to the freeing_list and
2201                                  * will not be touched by the delayed
2202                                  * kprobe_optimizer work handler.
2203                                  */
2204                                 kill_kprobe(p);
2205                         }
2206         }
2207         mutex_unlock(&kprobe_mutex);
2208         return NOTIFY_DONE;
2209 }
2210
2211 static struct notifier_block kprobe_module_nb = {
2212         .notifier_call = kprobes_module_callback,
2213         .priority = 0
2214 };
2215
2216 /* Markers of _kprobe_blacklist section */
2217 extern unsigned long __start_kprobe_blacklist[];
2218 extern unsigned long __stop_kprobe_blacklist[];
2219
2220 static int __init init_kprobes(void)
2221 {
2222         int i, err = 0;
2223
2224         /* FIXME allocate the probe table, currently defined statically */
2225         /* initialize all list heads */
2226         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2227                 INIT_HLIST_HEAD(&kprobe_table[i]);
2228                 INIT_HLIST_HEAD(&kretprobe_inst_table[i]);
2229                 raw_spin_lock_init(&(kretprobe_table_locks[i].lock));
2230         }
2231
2232         err = populate_kprobe_blacklist(__start_kprobe_blacklist,
2233                                         __stop_kprobe_blacklist);
2234         if (err) {
2235                 pr_err("kprobes: failed to populate blacklist: %d\n", err);
2236                 pr_err("Please take care of using kprobes.\n");
2237         }
2238
2239         if (kretprobe_blacklist_size) {
2240                 /* lookup the function address from its name */
2241                 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
2242                         kretprobe_blacklist[i].addr =
2243                                 kprobe_lookup_name(kretprobe_blacklist[i].name, 0);
2244                         if (!kretprobe_blacklist[i].addr)
2245                                 printk("kretprobe: lookup failed: %s\n",
2246                                        kretprobe_blacklist[i].name);
2247                 }
2248         }
2249
2250 #if defined(CONFIG_OPTPROBES)
2251 #if defined(__ARCH_WANT_KPROBES_INSN_SLOT)
2252         /* Init kprobe_optinsn_slots */
2253         kprobe_optinsn_slots.insn_size = MAX_OPTINSN_SIZE;
2254 #endif
2255         /* By default, kprobes can be optimized */
2256         kprobes_allow_optimization = true;
2257 #endif
2258
2259         /* By default, kprobes are armed */
2260         kprobes_all_disarmed = false;
2261
2262         err = arch_init_kprobes();
2263         if (!err)
2264                 err = register_die_notifier(&kprobe_exceptions_nb);
2265         if (!err)
2266                 err = register_module_notifier(&kprobe_module_nb);
2267
2268         kprobes_initialized = (err == 0);
2269
2270         if (!err)
2271                 init_test_probes();
2272         return err;
2273 }
2274
2275 #ifdef CONFIG_DEBUG_FS
2276 static void report_probe(struct seq_file *pi, struct kprobe *p,
2277                 const char *sym, int offset, char *modname, struct kprobe *pp)
2278 {
2279         char *kprobe_type;
2280
2281         if (p->pre_handler == pre_handler_kretprobe)
2282                 kprobe_type = "r";
2283         else if (p->pre_handler == setjmp_pre_handler)
2284                 kprobe_type = "j";
2285         else
2286                 kprobe_type = "k";
2287
2288         if (sym)
2289                 seq_printf(pi, "%p  %s  %s+0x%x  %s ",
2290                         p->addr, kprobe_type, sym, offset,
2291                         (modname ? modname : " "));
2292         else
2293                 seq_printf(pi, "%p  %s  %p ",
2294                         p->addr, kprobe_type, p->addr);
2295
2296         if (!pp)
2297                 pp = p;
2298         seq_printf(pi, "%s%s%s%s\n",
2299                 (kprobe_gone(p) ? "[GONE]" : ""),
2300                 ((kprobe_disabled(p) && !kprobe_gone(p)) ?  "[DISABLED]" : ""),
2301                 (kprobe_optimized(pp) ? "[OPTIMIZED]" : ""),
2302                 (kprobe_ftrace(pp) ? "[FTRACE]" : ""));
2303 }
2304
2305 static void *kprobe_seq_start(struct seq_file *f, loff_t *pos)
2306 {
2307         return (*pos < KPROBE_TABLE_SIZE) ? pos : NULL;
2308 }
2309
2310 static void *kprobe_seq_next(struct seq_file *f, void *v, loff_t *pos)
2311 {
2312         (*pos)++;
2313         if (*pos >= KPROBE_TABLE_SIZE)
2314                 return NULL;
2315         return pos;
2316 }
2317
2318 static void kprobe_seq_stop(struct seq_file *f, void *v)
2319 {
2320         /* Nothing to do */
2321 }
2322
2323 static int show_kprobe_addr(struct seq_file *pi, void *v)
2324 {
2325         struct hlist_head *head;
2326         struct kprobe *p, *kp;
2327         const char *sym = NULL;
2328         unsigned int i = *(loff_t *) v;
2329         unsigned long offset = 0;
2330         char *modname, namebuf[KSYM_NAME_LEN];
2331
2332         head = &kprobe_table[i];
2333         preempt_disable();
2334         hlist_for_each_entry_rcu(p, head, hlist) {
2335                 sym = kallsyms_lookup((unsigned long)p->addr, NULL,
2336                                         &offset, &modname, namebuf);
2337                 if (kprobe_aggrprobe(p)) {
2338                         list_for_each_entry_rcu(kp, &p->list, list)
2339                                 report_probe(pi, kp, sym, offset, modname, p);
2340                 } else
2341                         report_probe(pi, p, sym, offset, modname, NULL);
2342         }
2343         preempt_enable();
2344         return 0;
2345 }
2346
2347 static const struct seq_operations kprobes_seq_ops = {
2348         .start = kprobe_seq_start,
2349         .next  = kprobe_seq_next,
2350         .stop  = kprobe_seq_stop,
2351         .show  = show_kprobe_addr
2352 };
2353
2354 static int kprobes_open(struct inode *inode, struct file *filp)
2355 {
2356         return seq_open(filp, &kprobes_seq_ops);
2357 }
2358
2359 static const struct file_operations debugfs_kprobes_operations = {
2360         .open           = kprobes_open,
2361         .read           = seq_read,
2362         .llseek         = seq_lseek,
2363         .release        = seq_release,
2364 };
2365
2366 /* kprobes/blacklist -- shows which functions can not be probed */
2367 static void *kprobe_blacklist_seq_start(struct seq_file *m, loff_t *pos)
2368 {
2369         return seq_list_start(&kprobe_blacklist, *pos);
2370 }
2371
2372 static void *kprobe_blacklist_seq_next(struct seq_file *m, void *v, loff_t *pos)
2373 {
2374         return seq_list_next(v, &kprobe_blacklist, pos);
2375 }
2376
2377 static int kprobe_blacklist_seq_show(struct seq_file *m, void *v)
2378 {
2379         struct kprobe_blacklist_entry *ent =
2380                 list_entry(v, struct kprobe_blacklist_entry, list);
2381
2382         seq_printf(m, "0x%p-0x%p\t%ps\n", (void *)ent->start_addr,
2383                    (void *)ent->end_addr, (void *)ent->start_addr);
2384         return 0;
2385 }
2386
2387 static const struct seq_operations kprobe_blacklist_seq_ops = {
2388         .start = kprobe_blacklist_seq_start,
2389         .next  = kprobe_blacklist_seq_next,
2390         .stop  = kprobe_seq_stop,       /* Reuse void function */
2391         .show  = kprobe_blacklist_seq_show,
2392 };
2393
2394 static int kprobe_blacklist_open(struct inode *inode, struct file *filp)
2395 {
2396         return seq_open(filp, &kprobe_blacklist_seq_ops);
2397 }
2398
2399 static const struct file_operations debugfs_kprobe_blacklist_ops = {
2400         .open           = kprobe_blacklist_open,
2401         .read           = seq_read,
2402         .llseek         = seq_lseek,
2403         .release        = seq_release,
2404 };
2405
2406 static void arm_all_kprobes(void)
2407 {
2408         struct hlist_head *head;
2409         struct kprobe *p;
2410         unsigned int i;
2411
2412         mutex_lock(&kprobe_mutex);
2413
2414         /* If kprobes are armed, just return */
2415         if (!kprobes_all_disarmed)
2416                 goto already_enabled;
2417
2418         /*
2419          * optimize_kprobe() called by arm_kprobe() checks
2420          * kprobes_all_disarmed, so set kprobes_all_disarmed before
2421          * arm_kprobe.
2422          */
2423         kprobes_all_disarmed = false;
2424         /* Arming kprobes doesn't optimize kprobe itself */
2425         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2426                 head = &kprobe_table[i];
2427                 hlist_for_each_entry_rcu(p, head, hlist)
2428                         if (!kprobe_disabled(p))
2429                                 arm_kprobe(p);
2430         }
2431
2432         printk(KERN_INFO "Kprobes globally enabled\n");
2433
2434 already_enabled:
2435         mutex_unlock(&kprobe_mutex);
2436         return;
2437 }
2438
2439 static void disarm_all_kprobes(void)
2440 {
2441         struct hlist_head *head;
2442         struct kprobe *p;
2443         unsigned int i;
2444
2445         mutex_lock(&kprobe_mutex);
2446
2447         /* If kprobes are already disarmed, just return */
2448         if (kprobes_all_disarmed) {
2449                 mutex_unlock(&kprobe_mutex);
2450                 return;
2451         }
2452
2453         kprobes_all_disarmed = true;
2454         printk(KERN_INFO "Kprobes globally disabled\n");
2455
2456         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2457                 head = &kprobe_table[i];
2458                 hlist_for_each_entry_rcu(p, head, hlist) {
2459                         if (!arch_trampoline_kprobe(p) && !kprobe_disabled(p))
2460                                 disarm_kprobe(p, false);
2461                 }
2462         }
2463         mutex_unlock(&kprobe_mutex);
2464
2465         /* Wait for disarming all kprobes by optimizer */
2466         wait_for_kprobe_optimizer();
2467 }
2468
2469 /*
2470  * XXX: The debugfs bool file interface doesn't allow for callbacks
2471  * when the bool state is switched. We can reuse that facility when
2472  * available
2473  */
2474 static ssize_t read_enabled_file_bool(struct file *file,
2475                char __user *user_buf, size_t count, loff_t *ppos)
2476 {
2477         char buf[3];
2478
2479         if (!kprobes_all_disarmed)
2480                 buf[0] = '1';
2481         else
2482                 buf[0] = '0';
2483         buf[1] = '\n';
2484         buf[2] = 0x00;
2485         return simple_read_from_buffer(user_buf, count, ppos, buf, 2);
2486 }
2487
2488 static ssize_t write_enabled_file_bool(struct file *file,
2489                const char __user *user_buf, size_t count, loff_t *ppos)
2490 {
2491         char buf[32];
2492         size_t buf_size;
2493
2494         buf_size = min(count, (sizeof(buf)-1));
2495         if (copy_from_user(buf, user_buf, buf_size))
2496                 return -EFAULT;
2497
2498         buf[buf_size] = '\0';
2499         switch (buf[0]) {
2500         case 'y':
2501         case 'Y':
2502         case '1':
2503                 arm_all_kprobes();
2504                 break;
2505         case 'n':
2506         case 'N':
2507         case '0':
2508                 disarm_all_kprobes();
2509                 break;
2510         default:
2511                 return -EINVAL;
2512         }
2513
2514         return count;
2515 }
2516
2517 static const struct file_operations fops_kp = {
2518         .read =         read_enabled_file_bool,
2519         .write =        write_enabled_file_bool,
2520         .llseek =       default_llseek,
2521 };
2522
2523 static int __init debugfs_kprobe_init(void)
2524 {
2525         struct dentry *dir, *file;
2526         unsigned int value = 1;
2527
2528         dir = debugfs_create_dir("kprobes", NULL);
2529         if (!dir)
2530                 return -ENOMEM;
2531
2532         file = debugfs_create_file("list", 0444, dir, NULL,
2533                                 &debugfs_kprobes_operations);
2534         if (!file)
2535                 goto error;
2536
2537         file = debugfs_create_file("enabled", 0600, dir,
2538                                         &value, &fops_kp);
2539         if (!file)
2540                 goto error;
2541
2542         file = debugfs_create_file("blacklist", 0444, dir, NULL,
2543                                 &debugfs_kprobe_blacklist_ops);
2544         if (!file)
2545                 goto error;
2546
2547         return 0;
2548
2549 error:
2550         debugfs_remove(dir);
2551         return -ENOMEM;
2552 }
2553
2554 late_initcall(debugfs_kprobe_init);
2555 #endif /* CONFIG_DEBUG_FS */
2556
2557 module_init(init_kprobes);
2558
2559 /* defined in arch/.../kernel/kprobes.c */
2560 EXPORT_SYMBOL_GPL(jprobe_return);