kernel/kprobes.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  *  Kernel Probes (KProbes)
   4  *  kernel/kprobes.c
   5  *
   6  * Copyright (C) IBM Corporation, 2002, 2004
   7  *
   8  * 2002-Oct     Created by Vamsi Krishna S <vamsi_krishna@in.ibm.com> Kernel
   9  *              Probes initial implementation (includes suggestions from
  10  *              Rusty Russell).
  11  * 2004-Aug     Updated by Prasanna S Panchamukhi <prasanna@in.ibm.com> with
  12  *              hlists and exceptions notifier as suggested by Andi Kleen.
  13  * 2004-July    Suparna Bhattacharya <suparna@in.ibm.com> added jumper probes
  14  *              interface to access function arguments.
  15  * 2004-Sep     Prasanna S Panchamukhi <prasanna@in.ibm.com> Changed Kprobes
  16  *              exceptions notifier to be first on the priority list.
  17  * 2005-May     Hien Nguyen <hien@us.ibm.com>, Jim Keniston
  18  *              <jkenisto@us.ibm.com> and Prasanna S Panchamukhi
  19  *              <prasanna@in.ibm.com> added function-return probes.
  20  */
  21 #include <linux/kprobes.h>
  22 #include <linux/hash.h>
  23 #include <linux/init.h>
  24 #include <linux/slab.h>
  25 #include <linux/stddef.h>
  26 #include <linux/export.h>
  27 #include <linux/moduleloader.h>
  28 #include <linux/kallsyms.h>
  29 #include <linux/freezer.h>
  30 #include <linux/seq_file.h>
  31 #include <linux/debugfs.h>
  32 #include <linux/sysctl.h>
  33 #include <linux/kdebug.h>
  34 #include <linux/memory.h>
  35 #include <linux/ftrace.h>
  36 #include <linux/cpu.h>
  37 #include <linux/jump_label.h>
  38 #include <linux/perf_event.h>
  39
  40 #include <asm/sections.h>
  41 #include <asm/cacheflush.h>
  42 #include <asm/errno.h>
  43 #include <linux/uaccess.h>
  44
  45 #define KPROBE_HASH_BITS 6
  46 #define KPROBE_TABLE_SIZE (1 << KPROBE_HASH_BITS)
  47
  48
  49 static int kprobes_initialized;
  50 /* kprobe_table can be accessed by
  51  * - Normal hlist traversal and RCU add/del under kprobe_mutex is held.
  52  * Or
  53  * - RCU hlist traversal under disabling preempt (breakpoint handlers)
  54  */
  55 static struct hlist_head kprobe_table[KPROBE_TABLE_SIZE];
  56
  57 /* NOTE: change this value only with kprobe_mutex held */
  58 static bool kprobes_all_disarmed;
  59
  60 /* This protects kprobe_table and optimizing_list */
  61 static DEFINE_MUTEX(kprobe_mutex);
  62 static DEFINE_PER_CPU(struct kprobe *, kprobe_instance) = NULL;
  63
  64 kprobe_opcode_t * __weak kprobe_lookup_name(const char *name,
  65                                         unsigned int __unused)
  66 {
  67         return ((kprobe_opcode_t *)(kallsyms_lookup_name(name)));
  68 }
  69
  70 /* Blacklist -- list of struct kprobe_blacklist_entry */
  71 static LIST_HEAD(kprobe_blacklist);
  72
  73 #ifdef __ARCH_WANT_KPROBES_INSN_SLOT
  74 /*
  75  * kprobe->ainsn.insn points to the copy of the instruction to be
  76  * single-stepped. x86_64, POWER4 and above have no-exec support and
  77  * stepping on the instruction on a vmalloced/kmalloced/data page
  78  * is a recipe for disaster
  79  */
  80 struct kprobe_insn_page {
  81         struct list_head list;
  82         kprobe_opcode_t *insns;         /* Page of instruction slots */
  83         struct kprobe_insn_cache *cache;
  84         int nused;
  85         int ngarbage;
  86         char slot_used[];
  87 };
  88
  89 #define KPROBE_INSN_PAGE_SIZE(slots)                    \
  90         (offsetof(struct kprobe_insn_page, slot_used) + \
  91          (sizeof(char) * (slots)))
  92
  93 static int slots_per_page(struct kprobe_insn_cache *c)
  94 {
  95         return PAGE_SIZE/(c->insn_size * sizeof(kprobe_opcode_t));
  96 }
  97
  98 enum kprobe_slot_state {
  99         SLOT_CLEAN = 0,
 100         SLOT_DIRTY = 1,
 101         SLOT_USED = 2,
 102 };
 103
 104 void __weak *alloc_insn_page(void)
 105 {
 106         return module_alloc(PAGE_SIZE);
 107 }
 108
 109 static void free_insn_page(void *page)
 110 {
 111         module_memfree(page);
 112 }
 113
 114 struct kprobe_insn_cache kprobe_insn_slots = {
 115         .mutex = __MUTEX_INITIALIZER(kprobe_insn_slots.mutex),
 116         .alloc = alloc_insn_page,
 117         .free = free_insn_page,
 118         .sym = KPROBE_INSN_PAGE_SYM,
 119         .pages = LIST_HEAD_INIT(kprobe_insn_slots.pages),
 120         .insn_size = MAX_INSN_SIZE,
 121         .nr_garbage = 0,
 122 };
 123 static int collect_garbage_slots(struct kprobe_insn_cache *c);
 124
 125 /**
 126  * __get_insn_slot() - Find a slot on an executable page for an instruction.
 127  * We allocate an executable page if there's no room on existing ones.
 128  */
 129 kprobe_opcode_t *__get_insn_slot(struct kprobe_insn_cache *c)
 130 {
 131         struct kprobe_insn_page *kip;
 132         kprobe_opcode_t *slot = NULL;
 133
 134         /* Since the slot array is not protected by rcu, we need a mutex */
 135         mutex_lock(&c->mutex);
 136  retry:
 137         rcu_read_lock();
 138         list_for_each_entry_rcu(kip, &c->pages, list) {
 139                 if (kip->nused < slots_per_page(c)) {
 140                         int i;
 141                         for (i = 0; i < slots_per_page(c); i++) {
 142                                 if (kip->slot_used[i] == SLOT_CLEAN) {
 143                                         kip->slot_used[i] = SLOT_USED;
 144                                         kip->nused++;
 145                                         slot = kip->insns + (i * c->insn_size);
 146                                         rcu_read_unlock();
 147                                         goto out;
 148                                 }
 149                         }
 150                         /* kip->nused is broken. Fix it. */
 151                         kip->nused = slots_per_page(c);
 152                         WARN_ON(1);
 153                 }
 154         }
 155         rcu_read_unlock();
 156
 157         /* If there are any garbage slots, collect it and try again. */
 158         if (c->nr_garbage && collect_garbage_slots(c) == 0)
 159                 goto retry;
 160
 161         /* All out of space.  Need to allocate a new page. */
 162         kip = kmalloc(KPROBE_INSN_PAGE_SIZE(slots_per_page(c)), GFP_KERNEL);
 163         if (!kip)
 164                 goto out;
 165
 166         /*
 167          * Use module_alloc so this page is within +/- 2GB of where the
 168          * kernel image and loaded module images reside. This is required
 169          * so x86_64 can correctly handle the %rip-relative fixups.
 170          */
 171         kip->insns = c->alloc();
 172         if (!kip->insns) {
 173                 kfree(kip);
 174                 goto out;
 175         }
 176         INIT_LIST_HEAD(&kip->list);
 177         memset(kip->slot_used, SLOT_CLEAN, slots_per_page(c));
 178         kip->slot_used[0] = SLOT_USED;
 179         kip->nused = 1;
 180         kip->ngarbage = 0;
 181         kip->cache = c;
 182         list_add_rcu(&kip->list, &c->pages);
 183         slot = kip->insns;
 184
 185         /* Record the perf ksymbol register event after adding the page */
 186         perf_event_ksymbol(PERF_RECORD_KSYMBOL_TYPE_OOL, (unsigned long)kip->insns,
 187                            PAGE_SIZE, false, c->sym);
 188 out:
 189         mutex_unlock(&c->mutex);
 190         return slot;
 191 }
 192
 193 /* Return 1 if all garbages are collected, otherwise 0. */
 194 static int collect_one_slot(struct kprobe_insn_page *kip, int idx)
 195 {
 196         kip->slot_used[idx] = SLOT_CLEAN;
 197         kip->nused--;
 198         if (kip->nused == 0) {
 199                 /*
 200                  * Page is no longer in use.  Free it unless
 201                  * it's the last one.  We keep the last one
 202                  * so as not to have to set it up again the
 203                  * next time somebody inserts a probe.
 204                  */
 205                 if (!list_is_singular(&kip->list)) {
 206                         /*
 207                          * Record perf ksymbol unregister event before removing
 208                          * the page.
 209                          */
 210                         perf_event_ksymbol(PERF_RECORD_KSYMBOL_TYPE_OOL,
 211                                            (unsigned long)kip->insns, PAGE_SIZE, true,
 212                                            kip->cache->sym);
 213                         list_del_rcu(&kip->list);
 214                         synchronize_rcu();
 215                         kip->cache->free(kip->insns);
 216                         kfree(kip);
 217                 }
 218                 return 1;
 219         }
 220         return 0;
 221 }
 222
 223 static int collect_garbage_slots(struct kprobe_insn_cache *c)
 224 {
 225         struct kprobe_insn_page *kip, *next;
 226
 227         /* Ensure no-one is interrupted on the garbages */
 228         synchronize_rcu();
 229
 230         list_for_each_entry_safe(kip, next, &c->pages, list) {
 231                 int i;
 232                 if (kip->ngarbage == 0)
 233                         continue;
 234                 kip->ngarbage = 0;      /* we will collect all garbages */
 235                 for (i = 0; i < slots_per_page(c); i++) {
 236                         if (kip->slot_used[i] == SLOT_DIRTY && collect_one_slot(kip, i))
 237                                 break;
 238                 }
 239         }
 240         c->nr_garbage = 0;
 241         return 0;
 242 }
 243
 244 void __free_insn_slot(struct kprobe_insn_cache *c,
 245                       kprobe_opcode_t *slot, int dirty)
 246 {
 247         struct kprobe_insn_page *kip;
 248         long idx;
 249
 250         mutex_lock(&c->mutex);
 251         rcu_read_lock();
 252         list_for_each_entry_rcu(kip, &c->pages, list) {
 253                 idx = ((long)slot - (long)kip->insns) /
 254                         (c->insn_size * sizeof(kprobe_opcode_t));
 255                 if (idx >= 0 && idx < slots_per_page(c))
 256                         goto out;
 257         }
 258         /* Could not find this slot. */
 259         WARN_ON(1);
 260         kip = NULL;
 261 out:
 262         rcu_read_unlock();
 263         /* Mark and sweep: this may sleep */
 264         if (kip) {
 265                 /* Check double free */
 266                 WARN_ON(kip->slot_used[idx] != SLOT_USED);
 267                 if (dirty) {
 268                         kip->slot_used[idx] = SLOT_DIRTY;
 269                         kip->ngarbage++;
 270                         if (++c->nr_garbage > slots_per_page(c))
 271                                 collect_garbage_slots(c);
 272                 } else {
 273                         collect_one_slot(kip, idx);
 274                 }
 275         }
 276         mutex_unlock(&c->mutex);
 277 }
 278
 279 /*
 280  * Check given address is on the page of kprobe instruction slots.
 281  * This will be used for checking whether the address on a stack
 282  * is on a text area or not.
 283  */
 284 bool __is_insn_slot_addr(struct kprobe_insn_cache *c, unsigned long addr)
 285 {
 286         struct kprobe_insn_page *kip;
 287         bool ret = false;
 288
 289         rcu_read_lock();
 290         list_for_each_entry_rcu(kip, &c->pages, list) {
 291                 if (addr >= (unsigned long)kip->insns &&
 292                     addr < (unsigned long)kip->insns + PAGE_SIZE) {
 293                         ret = true;
 294                         break;
 295                 }
 296         }
 297         rcu_read_unlock();
 298
 299         return ret;
 300 }
 301
 302 int kprobe_cache_get_kallsym(struct kprobe_insn_cache *c, unsigned int *symnum,
 303                              unsigned long *value, char *type, char *sym)
 304 {
 305         struct kprobe_insn_page *kip;
 306         int ret = -ERANGE;
 307
 308         rcu_read_lock();
 309         list_for_each_entry_rcu(kip, &c->pages, list) {
 310                 if ((*symnum)--)
 311                         continue;
 312                 strlcpy(sym, c->sym, KSYM_NAME_LEN);
 313                 *type = 't';
 314                 *value = (unsigned long)kip->insns;
 315                 ret = 0;
 316                 break;
 317         }
 318         rcu_read_unlock();
 319
 320         return ret;
 321 }
 322
 323 #ifdef CONFIG_OPTPROBES
 324 void __weak *alloc_optinsn_page(void)
 325 {
 326         return alloc_insn_page();
 327 }
 328
 329 void __weak free_optinsn_page(void *page)
 330 {
 331         free_insn_page(page);
 332 }
 333
 334 /* For optimized_kprobe buffer */
 335 struct kprobe_insn_cache kprobe_optinsn_slots = {
 336         .mutex = __MUTEX_INITIALIZER(kprobe_optinsn_slots.mutex),
 337         .alloc = alloc_optinsn_page,
 338         .free = free_optinsn_page,
 339         .sym = KPROBE_OPTINSN_PAGE_SYM,
 340         .pages = LIST_HEAD_INIT(kprobe_optinsn_slots.pages),
 341         /* .insn_size is initialized later */
 342         .nr_garbage = 0,
 343 };
 344 #endif
 345 #endif
 346
 347 /* We have preemption disabled.. so it is safe to use __ versions */
 348 static inline void set_kprobe_instance(struct kprobe *kp)
 349 {
 350         __this_cpu_write(kprobe_instance, kp);
 351 }
 352
 353 static inline void reset_kprobe_instance(void)
 354 {
 355         __this_cpu_write(kprobe_instance, NULL);
 356 }
 357
 358 /*
 359  * This routine is called either:
 360  *      - under the kprobe_mutex - during kprobe_[un]register()
 361  *                              OR
 362  *      - with preemption disabled - from arch/xxx/kernel/kprobes.c
 363  */
 364 struct kprobe *get_kprobe(void *addr)
 365 {
 366         struct hlist_head *head;
 367         struct kprobe *p;
 368
 369         head = &kprobe_table[hash_ptr(addr, KPROBE_HASH_BITS)];
 370         hlist_for_each_entry_rcu(p, head, hlist,
 371                                  lockdep_is_held(&kprobe_mutex)) {
 372                 if (p->addr == addr)
 373                         return p;
 374         }
 375
 376         return NULL;
 377 }
 378 NOKPROBE_SYMBOL(get_kprobe);
 379
 380 static int aggr_pre_handler(struct kprobe *p, struct pt_regs *regs);
 381
 382 /* Return true if the kprobe is an aggregator */
 383 static inline int kprobe_aggrprobe(struct kprobe *p)
 384 {
 385         return p->pre_handler == aggr_pre_handler;
 386 }
 387
 388 /* Return true(!0) if the kprobe is unused */
 389 static inline int kprobe_unused(struct kprobe *p)
 390 {
 391         return kprobe_aggrprobe(p) && kprobe_disabled(p) &&
 392                list_empty(&p->list);
 393 }
 394
 395 /*
 396  * Keep all fields in the kprobe consistent
 397  */
 398 static inline void copy_kprobe(struct kprobe *ap, struct kprobe *p)
 399 {
 400         memcpy(&p->opcode, &ap->opcode, sizeof(kprobe_opcode_t));
 401         memcpy(&p->ainsn, &ap->ainsn, sizeof(struct arch_specific_insn));
 402 }
 403
 404 #ifdef CONFIG_OPTPROBES
 405 /* NOTE: change this value only with kprobe_mutex held */
 406 static bool kprobes_allow_optimization;
 407
 408 /*
 409  * Call all pre_handler on the list, but ignores its return value.
 410  * This must be called from arch-dep optimized caller.
 411  */
 412 void opt_pre_handler(struct kprobe *p, struct pt_regs *regs)
 413 {
 414         struct kprobe *kp;
 415
 416         list_for_each_entry_rcu(kp, &p->list, list) {
 417                 if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
 418                         set_kprobe_instance(kp);
 419                         kp->pre_handler(kp, regs);
 420                 }
 421                 reset_kprobe_instance();
 422         }
 423 }
 424 NOKPROBE_SYMBOL(opt_pre_handler);
 425
 426 /* Free optimized instructions and optimized_kprobe */
 427 static void free_aggr_kprobe(struct kprobe *p)
 428 {
 429         struct optimized_kprobe *op;
 430
 431         op = container_of(p, struct optimized_kprobe, kp);
 432         arch_remove_optimized_kprobe(op);
 433         arch_remove_kprobe(p);
 434         kfree(op);
 435 }
 436
 437 /* Return true(!0) if the kprobe is ready for optimization. */
 438 static inline int kprobe_optready(struct kprobe *p)
 439 {
 440         struct optimized_kprobe *op;
 441
 442         if (kprobe_aggrprobe(p)) {
 443                 op = container_of(p, struct optimized_kprobe, kp);
 444                 return arch_prepared_optinsn(&op->optinsn);
 445         }
 446
 447         return 0;
 448 }
 449
 450 /* Return true(!0) if the kprobe is disarmed. Note: p must be on hash list */
 451 static inline int kprobe_disarmed(struct kprobe *p)
 452 {
 453         struct optimized_kprobe *op;
 454
 455         /* If kprobe is not aggr/opt probe, just return kprobe is disabled */
 456         if (!kprobe_aggrprobe(p))
 457                 return kprobe_disabled(p);
 458
 459         op = container_of(p, struct optimized_kprobe, kp);
 460
 461         return kprobe_disabled(p) && list_empty(&op->list);
 462 }
 463
 464 /* Return true(!0) if the probe is queued on (un)optimizing lists */
 465 static int kprobe_queued(struct kprobe *p)
 466 {
 467         struct optimized_kprobe *op;
 468
 469         if (kprobe_aggrprobe(p)) {
 470                 op = container_of(p, struct optimized_kprobe, kp);
 471                 if (!list_empty(&op->list))
 472                         return 1;
 473         }
 474         return 0;
 475 }
 476
 477 /*
 478  * Return an optimized kprobe whose optimizing code replaces
 479  * instructions including addr (exclude breakpoint).
 480  */
 481 static struct kprobe *get_optimized_kprobe(unsigned long addr)
 482 {
 483         int i;
 484         struct kprobe *p = NULL;
 485         struct optimized_kprobe *op;
 486
 487         /* Don't check i == 0, since that is a breakpoint case. */
 488         for (i = 1; !p && i < MAX_OPTIMIZED_LENGTH; i++)
 489                 p = get_kprobe((void *)(addr - i));
 490
 491         if (p && kprobe_optready(p)) {
 492                 op = container_of(p, struct optimized_kprobe, kp);
 493                 if (arch_within_optimized_kprobe(op, addr))
 494                         return p;
 495         }
 496
 497         return NULL;
 498 }
 499
 500 /* Optimization staging list, protected by kprobe_mutex */
 501 static LIST_HEAD(optimizing_list);
 502 static LIST_HEAD(unoptimizing_list);
 503 static LIST_HEAD(freeing_list);
 504
 505 static void kprobe_optimizer(struct work_struct *work);
 506 static DECLARE_DELAYED_WORK(optimizing_work, kprobe_optimizer);
 507 #define OPTIMIZE_DELAY 5
 508
 509 /*
 510  * Optimize (replace a breakpoint with a jump) kprobes listed on
 511  * optimizing_list.
 512  */
 513 static void do_optimize_kprobes(void)
 514 {
 515         lockdep_assert_held(&text_mutex);
 516         /*
 517          * The optimization/unoptimization refers online_cpus via
 518          * stop_machine() and cpu-hotplug modifies online_cpus.
 519          * And same time, text_mutex will be held in cpu-hotplug and here.
 520          * This combination can cause a deadlock (cpu-hotplug try to lock
 521          * text_mutex but stop_machine can not be done because online_cpus
 522          * has been changed)
 523          * To avoid this deadlock, caller must have locked cpu hotplug
 524          * for preventing cpu-hotplug outside of text_mutex locking.
 525          */
 526         lockdep_assert_cpus_held();
 527
 528         /* Optimization never be done when disarmed */
 529         if (kprobes_all_disarmed || !kprobes_allow_optimization ||
 530             list_empty(&optimizing_list))
 531                 return;
 532
 533         arch_optimize_kprobes(&optimizing_list);
 534 }
 535
 536 /*
 537  * Unoptimize (replace a jump with a breakpoint and remove the breakpoint
 538  * if need) kprobes listed on unoptimizing_list.
 539  */
 540 static void do_unoptimize_kprobes(void)
 541 {
 542         struct optimized_kprobe *op, *tmp;
 543
 544         lockdep_assert_held(&text_mutex);
 545         /* See comment in do_optimize_kprobes() */
 546         lockdep_assert_cpus_held();
 547
 548         /* Unoptimization must be done anytime */
 549         if (list_empty(&unoptimizing_list))
 550                 return;
 551
 552         arch_unoptimize_kprobes(&unoptimizing_list, &freeing_list);
 553         /* Loop free_list for disarming */
 554         list_for_each_entry_safe(op, tmp, &freeing_list, list) {
 555                 /* Switching from detour code to origin */
 556                 op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
 557                 /* Disarm probes if marked disabled */
 558                 if (kprobe_disabled(&op->kp))
 559                         arch_disarm_kprobe(&op->kp);
 560                 if (kprobe_unused(&op->kp)) {
 561                         /*
 562                          * Remove unused probes from hash list. After waiting
 563                          * for synchronization, these probes are reclaimed.
 564                          * (reclaiming is done by do_free_cleaned_kprobes.)
 565                          */
 566                         hlist_del_rcu(&op->kp.hlist);
 567                 } else
 568                         list_del_init(&op->list);
 569         }
 570 }
 571
 572 /* Reclaim all kprobes on the free_list */
 573 static void do_free_cleaned_kprobes(void)
 574 {
 575         struct optimized_kprobe *op, *tmp;
 576
 577         list_for_each_entry_safe(op, tmp, &freeing_list, list) {
 578                 list_del_init(&op->list);
 579                 if (WARN_ON_ONCE(!kprobe_unused(&op->kp))) {
 580                         /*
 581                          * This must not happen, but if there is a kprobe
 582                          * still in use, keep it on kprobes hash list.
 583                          */
 584                         continue;
 585                 }
 586                 free_aggr_kprobe(&op->kp);
 587         }
 588 }
 589
 590 /* Start optimizer after OPTIMIZE_DELAY passed */
 591 static void kick_kprobe_optimizer(void)
 592 {
 593         schedule_delayed_work(&optimizing_work, OPTIMIZE_DELAY);
 594 }
 595
 596 /* Kprobe jump optimizer */
 597 static void kprobe_optimizer(struct work_struct *work)
 598 {
 599         mutex_lock(&kprobe_mutex);
 600         cpus_read_lock();
 601         mutex_lock(&text_mutex);
 602
 603         /*
 604          * Step 1: Unoptimize kprobes and collect cleaned (unused and disarmed)
 605          * kprobes before waiting for quiesence period.
 606          */
 607         do_unoptimize_kprobes();
 608
 609         /*
 610          * Step 2: Wait for quiesence period to ensure all potentially
 611          * preempted tasks to have normally scheduled. Because optprobe
 612          * may modify multiple instructions, there is a chance that Nth
 613          * instruction is preempted. In that case, such tasks can return
 614          * to 2nd-Nth byte of jump instruction. This wait is for avoiding it.
 615          * Note that on non-preemptive kernel, this is transparently converted
 616          * to synchronoze_sched() to wait for all interrupts to have completed.
 617          */
 618         synchronize_rcu_tasks();
 619
 620         /* Step 3: Optimize kprobes after quiesence period */
 621         do_optimize_kprobes();
 622
 623         /* Step 4: Free cleaned kprobes after quiesence period */
 624         do_free_cleaned_kprobes();
 625
 626         mutex_unlock(&text_mutex);
 627         cpus_read_unlock();
 628
 629         /* Step 5: Kick optimizer again if needed */
 630         if (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list))
 631                 kick_kprobe_optimizer();
 632
 633         mutex_unlock(&kprobe_mutex);
 634 }
 635
 636 /* Wait for completing optimization and unoptimization */
 637 void wait_for_kprobe_optimizer(void)
 638 {
 639         mutex_lock(&kprobe_mutex);
 640
 641         while (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list)) {
 642                 mutex_unlock(&kprobe_mutex);
 643
 644                 /* this will also make optimizing_work execute immmediately */
 645                 flush_delayed_work(&optimizing_work);
 646                 /* @optimizing_work might not have been queued yet, relax */
 647                 cpu_relax();
 648
 649                 mutex_lock(&kprobe_mutex);
 650         }
 651
 652         mutex_unlock(&kprobe_mutex);
 653 }
 654
 655 static bool optprobe_queued_unopt(struct optimized_kprobe *op)
 656 {
 657         struct optimized_kprobe *_op;
 658
 659         list_for_each_entry(_op, &unoptimizing_list, list) {
 660                 if (op == _op)
 661                         return true;
 662         }
 663
 664         return false;
 665 }
 666
 667 /* Optimize kprobe if p is ready to be optimized */
 668 static void optimize_kprobe(struct kprobe *p)
 669 {
 670         struct optimized_kprobe *op;
 671
 672         /* Check if the kprobe is disabled or not ready for optimization. */
 673         if (!kprobe_optready(p) || !kprobes_allow_optimization ||
 674             (kprobe_disabled(p) || kprobes_all_disarmed))
 675                 return;
 676
 677         /* kprobes with post_handler can not be optimized */
 678         if (p->post_handler)
 679                 return;
 680
 681         op = container_of(p, struct optimized_kprobe, kp);
 682
 683         /* Check there is no other kprobes at the optimized instructions */
 684         if (arch_check_optimized_kprobe(op) < 0)
 685                 return;
 686
 687         /* Check if it is already optimized. */
 688         if (op->kp.flags & KPROBE_FLAG_OPTIMIZED) {
 689                 if (optprobe_queued_unopt(op)) {
 690                         /* This is under unoptimizing. Just dequeue the probe */
 691                         list_del_init(&op->list);
 692                 }
 693                 return;
 694         }
 695         op->kp.flags |= KPROBE_FLAG_OPTIMIZED;
 696
 697         /* On unoptimizing/optimizing_list, op must have OPTIMIZED flag */
 698         if (WARN_ON_ONCE(!list_empty(&op->list)))
 699                 return;
 700
 701         list_add(&op->list, &optimizing_list);
 702         kick_kprobe_optimizer();
 703 }
 704
 705 /* Short cut to direct unoptimizing */
 706 static void force_unoptimize_kprobe(struct optimized_kprobe *op)
 707 {
 708         lockdep_assert_cpus_held();
 709         arch_unoptimize_kprobe(op);
 710         op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
 711 }
 712
 713 /* Unoptimize a kprobe if p is optimized */
 714 static void unoptimize_kprobe(struct kprobe *p, bool force)
 715 {
 716         struct optimized_kprobe *op;
 717
 718         if (!kprobe_aggrprobe(p) || kprobe_disarmed(p))
 719                 return; /* This is not an optprobe nor optimized */
 720
 721         op = container_of(p, struct optimized_kprobe, kp);
 722         if (!kprobe_optimized(p))
 723                 return;
 724
 725         if (!list_empty(&op->list)) {
 726                 if (optprobe_queued_unopt(op)) {
 727                         /* Queued in unoptimizing queue */
 728                         if (force) {
 729                                 /*
 730                                  * Forcibly unoptimize the kprobe here, and queue it
 731                                  * in the freeing list for release afterwards.
 732                                  */
 733                                 force_unoptimize_kprobe(op);
 734                                 list_move(&op->list, &freeing_list);
 735                         }
 736                 } else {
 737                         /* Dequeue from the optimizing queue */
 738                         list_del_init(&op->list);
 739                         op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
 740                 }
 741                 return;
 742         }
 743
 744         /* Optimized kprobe case */
 745         if (force) {
 746                 /* Forcibly update the code: this is a special case */
 747                 force_unoptimize_kprobe(op);
 748         } else {
 749                 list_add(&op->list, &unoptimizing_list);
 750                 kick_kprobe_optimizer();
 751         }
 752 }
 753
 754 /* Cancel unoptimizing for reusing */
 755 static int reuse_unused_kprobe(struct kprobe *ap)
 756 {
 757         struct optimized_kprobe *op;
 758
 759         /*
 760          * Unused kprobe MUST be on the way of delayed unoptimizing (means
 761          * there is still a relative jump) and disabled.
 762          */
 763         op = container_of(ap, struct optimized_kprobe, kp);
 764         WARN_ON_ONCE(list_empty(&op->list));
 765         /* Enable the probe again */
 766         ap->flags &= ~KPROBE_FLAG_DISABLED;
 767         /* Optimize it again (remove from op->list) */
 768         if (!kprobe_optready(ap))
 769                 return -EINVAL;
 770
 771         optimize_kprobe(ap);
 772         return 0;
 773 }
 774
 775 /* Remove optimized instructions */
 776 static void kill_optimized_kprobe(struct kprobe *p)
 777 {
 778         struct optimized_kprobe *op;
 779
 780         op = container_of(p, struct optimized_kprobe, kp);
 781         if (!list_empty(&op->list))
 782                 /* Dequeue from the (un)optimization queue */
 783                 list_del_init(&op->list);
 784         op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
 785
 786         if (kprobe_unused(p)) {
 787                 /* Enqueue if it is unused */
 788                 list_add(&op->list, &freeing_list);
 789                 /*
 790                  * Remove unused probes from the hash list. After waiting
 791                  * for synchronization, this probe is reclaimed.
 792                  * (reclaiming is done by do_free_cleaned_kprobes().)
 793                  */
 794                 hlist_del_rcu(&op->kp.hlist);
 795         }
 796
 797         /* Don't touch the code, because it is already freed. */
 798         arch_remove_optimized_kprobe(op);
 799 }
 800
 801 static inline
 802 void __prepare_optimized_kprobe(struct optimized_kprobe *op, struct kprobe *p)
 803 {
 804         if (!kprobe_ftrace(p))
 805                 arch_prepare_optimized_kprobe(op, p);
 806 }
 807
 808 /* Try to prepare optimized instructions */
 809 static void prepare_optimized_kprobe(struct kprobe *p)
 810 {
 811         struct optimized_kprobe *op;
 812
 813         op = container_of(p, struct optimized_kprobe, kp);
 814         __prepare_optimized_kprobe(op, p);
 815 }
 816
 817 /* Allocate new optimized_kprobe and try to prepare optimized instructions */
 818 static struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
 819 {
 820         struct optimized_kprobe *op;
 821
 822         op = kzalloc(sizeof(struct optimized_kprobe), GFP_KERNEL);
 823         if (!op)
 824                 return NULL;
 825
 826         INIT_LIST_HEAD(&op->list);
 827         op->kp.addr = p->addr;
 828         __prepare_optimized_kprobe(op, p);
 829
 830         return &op->kp;
 831 }
 832
 833 static void init_aggr_kprobe(struct kprobe *ap, struct kprobe *p);
 834
 835 /*
 836  * Prepare an optimized_kprobe and optimize it
 837  * NOTE: p must be a normal registered kprobe
 838  */
 839 static void try_to_optimize_kprobe(struct kprobe *p)
 840 {
 841         struct kprobe *ap;
 842         struct optimized_kprobe *op;
 843
 844         /* Impossible to optimize ftrace-based kprobe */
 845         if (kprobe_ftrace(p))
 846                 return;
 847
 848         /* For preparing optimization, jump_label_text_reserved() is called */
 849         cpus_read_lock();
 850         jump_label_lock();
 851         mutex_lock(&text_mutex);
 852
 853         ap = alloc_aggr_kprobe(p);
 854         if (!ap)
 855                 goto out;
 856
 857         op = container_of(ap, struct optimized_kprobe, kp);
 858         if (!arch_prepared_optinsn(&op->optinsn)) {
 859                 /* If failed to setup optimizing, fallback to kprobe */
 860                 arch_remove_optimized_kprobe(op);
 861                 kfree(op);
 862                 goto out;
 863         }
 864
 865         init_aggr_kprobe(ap, p);
 866         optimize_kprobe(ap);    /* This just kicks optimizer thread */
 867
 868 out:
 869         mutex_unlock(&text_mutex);
 870         jump_label_unlock();
 871         cpus_read_unlock();
 872 }
 873
 874 static void optimize_all_kprobes(void)
 875 {
 876         struct hlist_head *head;
 877         struct kprobe *p;
 878         unsigned int i;
 879
 880         mutex_lock(&kprobe_mutex);
 881         /* If optimization is already allowed, just return */
 882         if (kprobes_allow_optimization)
 883                 goto out;
 884
 885         cpus_read_lock();
 886         kprobes_allow_optimization = true;
 887         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
 888                 head = &kprobe_table[i];
 889                 hlist_for_each_entry(p, head, hlist)
 890                         if (!kprobe_disabled(p))
 891                                 optimize_kprobe(p);
 892         }
 893         cpus_read_unlock();
 894         printk(KERN_INFO "Kprobes globally optimized\n");
 895 out:
 896         mutex_unlock(&kprobe_mutex);
 897 }
 898
 899 #ifdef CONFIG_SYSCTL
 900 static void unoptimize_all_kprobes(void)
 901 {
 902         struct hlist_head *head;
 903         struct kprobe *p;
 904         unsigned int i;
 905
 906         mutex_lock(&kprobe_mutex);
 907         /* If optimization is already prohibited, just return */
 908         if (!kprobes_allow_optimization) {
 909                 mutex_unlock(&kprobe_mutex);
 910                 return;
 911         }
 912
 913         cpus_read_lock();
 914         kprobes_allow_optimization = false;
 915         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
 916                 head = &kprobe_table[i];
 917                 hlist_for_each_entry(p, head, hlist) {
 918                         if (!kprobe_disabled(p))
 919                                 unoptimize_kprobe(p, false);
 920                 }
 921         }
 922         cpus_read_unlock();
 923         mutex_unlock(&kprobe_mutex);
 924
 925         /* Wait for unoptimizing completion */
 926         wait_for_kprobe_optimizer();
 927         printk(KERN_INFO "Kprobes globally unoptimized\n");
 928 }
 929
 930 static DEFINE_MUTEX(kprobe_sysctl_mutex);
 931 int sysctl_kprobes_optimization;
 932 int proc_kprobes_optimization_handler(struct ctl_table *table, int write,
 933                                       void *buffer, size_t *length,
 934                                       loff_t *ppos)
 935 {
 936         int ret;
 937
 938         mutex_lock(&kprobe_sysctl_mutex);
 939         sysctl_kprobes_optimization = kprobes_allow_optimization ? 1 : 0;
 940         ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
 941
 942         if (sysctl_kprobes_optimization)
 943                 optimize_all_kprobes();
 944         else
 945                 unoptimize_all_kprobes();
 946         mutex_unlock(&kprobe_sysctl_mutex);
 947
 948         return ret;
 949 }
 950 #endif /* CONFIG_SYSCTL */
 951
 952 /* Put a breakpoint for a probe. Must be called with text_mutex locked */
 953 static void __arm_kprobe(struct kprobe *p)
 954 {
 955         struct kprobe *_p;
 956
 957         /* Check collision with other optimized kprobes */
 958         _p = get_optimized_kprobe((unsigned long)p->addr);
 959         if (unlikely(_p))
 960                 /* Fallback to unoptimized kprobe */
 961                 unoptimize_kprobe(_p, true);
 962
 963         arch_arm_kprobe(p);
 964         optimize_kprobe(p);     /* Try to optimize (add kprobe to a list) */
 965 }
 966
 967 /* Remove the breakpoint of a probe. Must be called with text_mutex locked */
 968 static void __disarm_kprobe(struct kprobe *p, bool reopt)
 969 {
 970         struct kprobe *_p;
 971
 972         /* Try to unoptimize */
 973         unoptimize_kprobe(p, kprobes_all_disarmed);
 974
 975         if (!kprobe_queued(p)) {
 976                 arch_disarm_kprobe(p);
 977                 /* If another kprobe was blocked, optimize it. */
 978                 _p = get_optimized_kprobe((unsigned long)p->addr);
 979                 if (unlikely(_p) && reopt)
 980                         optimize_kprobe(_p);
 981         }
 982         /* TODO: reoptimize others after unoptimized this probe */
 983 }
 984
 985 #else /* !CONFIG_OPTPROBES */
 986
 987 #define optimize_kprobe(p)                      do {} while (0)
 988 #define unoptimize_kprobe(p, f)                 do {} while (0)
 989 #define kill_optimized_kprobe(p)                do {} while (0)
 990 #define prepare_optimized_kprobe(p)             do {} while (0)
 991 #define try_to_optimize_kprobe(p)               do {} while (0)
 992 #define __arm_kprobe(p)                         arch_arm_kprobe(p)
 993 #define __disarm_kprobe(p, o)                   arch_disarm_kprobe(p)
 994 #define kprobe_disarmed(p)                      kprobe_disabled(p)
 995 #define wait_for_kprobe_optimizer()             do {} while (0)
 996
 997 static int reuse_unused_kprobe(struct kprobe *ap)
 998 {
 999         /*
1000          * If the optimized kprobe is NOT supported, the aggr kprobe is
1001          * released at the same time that the last aggregated kprobe is
1002          * unregistered.
1003          * Thus there should be no chance to reuse unused kprobe.
1004          */
1005         printk(KERN_ERR "Error: There should be no unused kprobe here.\n");
1006         return -EINVAL;
1007 }
1008
1009 static void free_aggr_kprobe(struct kprobe *p)
1010 {
1011         arch_remove_kprobe(p);
1012         kfree(p);
1013 }
1014
1015 static struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
1016 {
1017         return kzalloc(sizeof(struct kprobe), GFP_KERNEL);
1018 }
1019 #endif /* CONFIG_OPTPROBES */
1020
1021 #ifdef CONFIG_KPROBES_ON_FTRACE
1022 static struct ftrace_ops kprobe_ftrace_ops __read_mostly = {
1023         .func = kprobe_ftrace_handler,
1024         .flags = FTRACE_OPS_FL_SAVE_REGS,
1025 };
1026
1027 static struct ftrace_ops kprobe_ipmodify_ops __read_mostly = {
1028         .func = kprobe_ftrace_handler,
1029         .flags = FTRACE_OPS_FL_SAVE_REGS | FTRACE_OPS_FL_IPMODIFY,
1030 };
1031
1032 static int kprobe_ipmodify_enabled;
1033 static int kprobe_ftrace_enabled;
1034
1035 /* Must ensure p->addr is really on ftrace */
1036 static int prepare_kprobe(struct kprobe *p)
1037 {
1038         if (!kprobe_ftrace(p))
1039                 return arch_prepare_kprobe(p);
1040
1041         return arch_prepare_kprobe_ftrace(p);
1042 }
1043
1044 /* Caller must lock kprobe_mutex */
1045 static int __arm_kprobe_ftrace(struct kprobe *p, struct ftrace_ops *ops,
1046                                int *cnt)
1047 {
1048         int ret = 0;
1049
1050         ret = ftrace_set_filter_ip(ops, (unsigned long)p->addr, 0, 0);
1051         if (ret) {
1052                 pr_debug("Failed to arm kprobe-ftrace at %pS (%d)\n",
1053                          p->addr, ret);
1054                 return ret;
1055         }
1056
1057         if (*cnt == 0) {
1058                 ret = register_ftrace_function(ops);
1059                 if (ret) {
1060                         pr_debug("Failed to init kprobe-ftrace (%d)\n", ret);
1061                         goto err_ftrace;
1062                 }
1063         }
1064
1065         (*cnt)++;
1066         return ret;
1067
1068 err_ftrace:
1069         /*
1070          * At this point, sinec ops is not registered, we should be sefe from
1071          * registering empty filter.
1072          */
1073         ftrace_set_filter_ip(ops, (unsigned long)p->addr, 1, 0);
1074         return ret;
1075 }
1076
1077 static int arm_kprobe_ftrace(struct kprobe *p)
1078 {
1079         bool ipmodify = (p->post_handler != NULL);
1080
1081         return __arm_kprobe_ftrace(p,
1082                 ipmodify ? &kprobe_ipmodify_ops : &kprobe_ftrace_ops,
1083                 ipmodify ? &kprobe_ipmodify_enabled : &kprobe_ftrace_enabled);
1084 }
1085
1086 /* Caller must lock kprobe_mutex */
1087 static int __disarm_kprobe_ftrace(struct kprobe *p, struct ftrace_ops *ops,
1088                                   int *cnt)
1089 {
1090         int ret = 0;
1091
1092         if (*cnt == 1) {
1093                 ret = unregister_ftrace_function(ops);
1094                 if (WARN(ret < 0, "Failed to unregister kprobe-ftrace (%d)\n", ret))
1095                         return ret;
1096         }
1097
1098         (*cnt)--;
1099
1100         ret = ftrace_set_filter_ip(ops, (unsigned long)p->addr, 1, 0);
1101         WARN_ONCE(ret < 0, "Failed to disarm kprobe-ftrace at %pS (%d)\n",
1102                   p->addr, ret);
1103         return ret;
1104 }
1105
1106 static int disarm_kprobe_ftrace(struct kprobe *p)
1107 {
1108         bool ipmodify = (p->post_handler != NULL);
1109
1110         return __disarm_kprobe_ftrace(p,
1111                 ipmodify ? &kprobe_ipmodify_ops : &kprobe_ftrace_ops,
1112                 ipmodify ? &kprobe_ipmodify_enabled : &kprobe_ftrace_enabled);
1113 }
1114 #else   /* !CONFIG_KPROBES_ON_FTRACE */
1115 static inline int prepare_kprobe(struct kprobe *p)
1116 {
1117         return arch_prepare_kprobe(p);
1118 }
1119
1120 static inline int arm_kprobe_ftrace(struct kprobe *p)
1121 {
1122         return -ENODEV;
1123 }
1124
1125 static inline int disarm_kprobe_ftrace(struct kprobe *p)
1126 {
1127         return -ENODEV;
1128 }
1129 #endif
1130
1131 /* Arm a kprobe with text_mutex */
1132 static int arm_kprobe(struct kprobe *kp)
1133 {
1134         if (unlikely(kprobe_ftrace(kp)))
1135                 return arm_kprobe_ftrace(kp);
1136
1137         cpus_read_lock();
1138         mutex_lock(&text_mutex);
1139         __arm_kprobe(kp);
1140         mutex_unlock(&text_mutex);
1141         cpus_read_unlock();
1142
1143         return 0;
1144 }
1145
1146 /* Disarm a kprobe with text_mutex */
1147 static int disarm_kprobe(struct kprobe *kp, bool reopt)
1148 {
1149         if (unlikely(kprobe_ftrace(kp)))
1150                 return disarm_kprobe_ftrace(kp);
1151
1152         cpus_read_lock();
1153         mutex_lock(&text_mutex);
1154         __disarm_kprobe(kp, reopt);
1155         mutex_unlock(&text_mutex);
1156         cpus_read_unlock();
1157
1158         return 0;
1159 }
1160
1161 /*
1162  * Aggregate handlers for multiple kprobes support - these handlers
1163  * take care of invoking the individual kprobe handlers on p->list
1164  */
1165 static int aggr_pre_handler(struct kprobe *p, struct pt_regs *regs)
1166 {
1167         struct kprobe *kp;
1168
1169         list_for_each_entry_rcu(kp, &p->list, list) {
1170                 if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
1171                         set_kprobe_instance(kp);
1172                         if (kp->pre_handler(kp, regs))
1173                                 return 1;
1174                 }
1175                 reset_kprobe_instance();
1176         }
1177         return 0;
1178 }
1179 NOKPROBE_SYMBOL(aggr_pre_handler);
1180
1181 static void aggr_post_handler(struct kprobe *p, struct pt_regs *regs,
1182                               unsigned long flags)
1183 {
1184         struct kprobe *kp;
1185
1186         list_for_each_entry_rcu(kp, &p->list, list) {
1187                 if (kp->post_handler && likely(!kprobe_disabled(kp))) {
1188                         set_kprobe_instance(kp);
1189                         kp->post_handler(kp, regs, flags);
1190                         reset_kprobe_instance();
1191                 }
1192         }
1193 }
1194 NOKPROBE_SYMBOL(aggr_post_handler);
1195
1196 /* Walks the list and increments nmissed count for multiprobe case */
1197 void kprobes_inc_nmissed_count(struct kprobe *p)
1198 {
1199         struct kprobe *kp;
1200         if (!kprobe_aggrprobe(p)) {
1201                 p->nmissed++;
1202         } else {
1203                 list_for_each_entry_rcu(kp, &p->list, list)
1204                         kp->nmissed++;
1205         }
1206         return;
1207 }
1208 NOKPROBE_SYMBOL(kprobes_inc_nmissed_count);
1209
1210 static void free_rp_inst_rcu(struct rcu_head *head)
1211 {
1212         struct kretprobe_instance *ri = container_of(head, struct kretprobe_instance, rcu);
1213
1214         if (refcount_dec_and_test(&ri->rph->ref))
1215                 kfree(ri->rph);
1216         kfree(ri);
1217 }
1218 NOKPROBE_SYMBOL(free_rp_inst_rcu);
1219
1220 static void recycle_rp_inst(struct kretprobe_instance *ri)
1221 {
1222         struct kretprobe *rp = get_kretprobe(ri);
1223
1224         if (likely(rp)) {
1225                 freelist_add(&ri->freelist, &rp->freelist);
1226         } else
1227                 call_rcu(&ri->rcu, free_rp_inst_rcu);
1228 }
1229 NOKPROBE_SYMBOL(recycle_rp_inst);
1230
1231 static struct kprobe kprobe_busy = {
1232         .addr = (void *) get_kprobe,
1233 };
1234
1235 void kprobe_busy_begin(void)
1236 {
1237         struct kprobe_ctlblk *kcb;
1238
1239         preempt_disable();
1240         __this_cpu_write(current_kprobe, &kprobe_busy);
1241         kcb = get_kprobe_ctlblk();
1242         kcb->kprobe_status = KPROBE_HIT_ACTIVE;
1243 }
1244
1245 void kprobe_busy_end(void)
1246 {
1247         __this_cpu_write(current_kprobe, NULL);
1248         preempt_enable();
1249 }
1250
1251 /*
1252  * This function is called from finish_task_switch when task tk becomes dead,
1253  * so that we can recycle any function-return probe instances associated
1254  * with this task. These left over instances represent probed functions
1255  * that have been called but will never return.
1256  */
1257 void kprobe_flush_task(struct task_struct *tk)
1258 {
1259         struct kretprobe_instance *ri;
1260         struct llist_node *node;
1261
1262         /* Early boot, not yet initialized. */
1263         if (unlikely(!kprobes_initialized))
1264                 return;
1265
1266         kprobe_busy_begin();
1267
1268         node = __llist_del_all(&tk->kretprobe_instances);
1269         while (node) {
1270                 ri = container_of(node, struct kretprobe_instance, llist);
1271                 node = node->next;
1272
1273                 recycle_rp_inst(ri);
1274         }
1275
1276         kprobe_busy_end();
1277 }
1278 NOKPROBE_SYMBOL(kprobe_flush_task);
1279
1280 static inline void free_rp_inst(struct kretprobe *rp)
1281 {
1282         struct kretprobe_instance *ri;
1283         struct freelist_node *node;
1284         int count = 0;
1285
1286         node = rp->freelist.head;
1287         while (node) {
1288                 ri = container_of(node, struct kretprobe_instance, freelist);
1289                 node = node->next;
1290
1291                 kfree(ri);
1292                 count++;
1293         }
1294
1295         if (refcount_sub_and_test(count, &rp->rph->ref)) {
1296                 kfree(rp->rph);
1297                 rp->rph = NULL;
1298         }
1299 }
1300
1301 /* Add the new probe to ap->list */
1302 static int add_new_kprobe(struct kprobe *ap, struct kprobe *p)
1303 {
1304         if (p->post_handler)
1305                 unoptimize_kprobe(ap, true);    /* Fall back to normal kprobe */
1306
1307         list_add_rcu(&p->list, &ap->list);
1308         if (p->post_handler && !ap->post_handler)
1309                 ap->post_handler = aggr_post_handler;
1310
1311         return 0;
1312 }
1313
1314 /*
1315  * Fill in the required fields of the "manager kprobe". Replace the
1316  * earlier kprobe in the hlist with the manager kprobe
1317  */
1318 static void init_aggr_kprobe(struct kprobe *ap, struct kprobe *p)
1319 {
1320         /* Copy p's insn slot to ap */
1321         copy_kprobe(p, ap);
1322         flush_insn_slot(ap);
1323         ap->addr = p->addr;
1324         ap->flags = p->flags & ~KPROBE_FLAG_OPTIMIZED;
1325         ap->pre_handler = aggr_pre_handler;
1326         /* We don't care the kprobe which has gone. */
1327         if (p->post_handler && !kprobe_gone(p))
1328                 ap->post_handler = aggr_post_handler;
1329
1330         INIT_LIST_HEAD(&ap->list);
1331         INIT_HLIST_NODE(&ap->hlist);
1332
1333         list_add_rcu(&p->list, &ap->list);
1334         hlist_replace_rcu(&p->hlist, &ap->hlist);
1335 }
1336
1337 /*
1338  * This is the second or subsequent kprobe at the address - handle
1339  * the intricacies
1340  */
1341 static int register_aggr_kprobe(struct kprobe *orig_p, struct kprobe *p)
1342 {
1343         int ret = 0;
1344         struct kprobe *ap = orig_p;
1345
1346         cpus_read_lock();
1347
1348         /* For preparing optimization, jump_label_text_reserved() is called */
1349         jump_label_lock();
1350         mutex_lock(&text_mutex);
1351
1352         if (!kprobe_aggrprobe(orig_p)) {
1353                 /* If orig_p is not an aggr_kprobe, create new aggr_kprobe. */
1354                 ap = alloc_aggr_kprobe(orig_p);
1355                 if (!ap) {
1356                         ret = -ENOMEM;
1357                         goto out;
1358                 }
1359                 init_aggr_kprobe(ap, orig_p);
1360         } else if (kprobe_unused(ap)) {
1361                 /* This probe is going to die. Rescue it */
1362                 ret = reuse_unused_kprobe(ap);
1363                 if (ret)
1364                         goto out;
1365         }
1366
1367         if (kprobe_gone(ap)) {
1368                 /*
1369                  * Attempting to insert new probe at the same location that
1370                  * had a probe in the module vaddr area which already
1371                  * freed. So, the instruction slot has already been
1372                  * released. We need a new slot for the new probe.
1373                  */
1374                 ret = arch_prepare_kprobe(ap);
1375                 if (ret)
1376                         /*
1377                          * Even if fail to allocate new slot, don't need to
1378                          * free aggr_probe. It will be used next time, or
1379                          * freed by unregister_kprobe.
1380                          */
1381                         goto out;
1382
1383                 /* Prepare optimized instructions if possible. */
1384                 prepare_optimized_kprobe(ap);
1385
1386                 /*
1387                  * Clear gone flag to prevent allocating new slot again, and
1388                  * set disabled flag because it is not armed yet.
1389                  */
1390                 ap->flags = (ap->flags & ~KPROBE_FLAG_GONE)
1391                             | KPROBE_FLAG_DISABLED;
1392         }
1393
1394         /* Copy ap's insn slot to p */
1395         copy_kprobe(ap, p);
1396         ret = add_new_kprobe(ap, p);
1397
1398 out:
1399         mutex_unlock(&text_mutex);
1400         jump_label_unlock();
1401         cpus_read_unlock();
1402
1403         if (ret == 0 && kprobe_disabled(ap) && !kprobe_disabled(p)) {
1404                 ap->flags &= ~KPROBE_FLAG_DISABLED;
1405                 if (!kprobes_all_disarmed) {
1406                         /* Arm the breakpoint again. */
1407                         ret = arm_kprobe(ap);
1408                         if (ret) {
1409                                 ap->flags |= KPROBE_FLAG_DISABLED;
1410                                 list_del_rcu(&p->list);
1411                                 synchronize_rcu();
1412                         }
1413                 }
1414         }
1415         return ret;
1416 }
1417
1418 bool __weak arch_within_kprobe_blacklist(unsigned long addr)
1419 {
1420         /* The __kprobes marked functions and entry code must not be probed */
1421         return addr >= (unsigned long)__kprobes_text_start &&
1422                addr < (unsigned long)__kprobes_text_end;
1423 }
1424
1425 static bool __within_kprobe_blacklist(unsigned long addr)
1426 {
1427         struct kprobe_blacklist_entry *ent;
1428
1429         if (arch_within_kprobe_blacklist(addr))
1430                 return true;
1431         /*
1432          * If there exists a kprobe_blacklist, verify and
1433          * fail any probe registration in the prohibited area
1434          */
1435         list_for_each_entry(ent, &kprobe_blacklist, list) {
1436                 if (addr >= ent->start_addr && addr < ent->end_addr)
1437                         return true;
1438         }
1439         return false;
1440 }
1441
1442 bool within_kprobe_blacklist(unsigned long addr)
1443 {
1444         char symname[KSYM_NAME_LEN], *p;
1445
1446         if (__within_kprobe_blacklist(addr))
1447                 return true;
1448
1449         /* Check if the address is on a suffixed-symbol */
1450         if (!lookup_symbol_name(addr, symname)) {
1451                 p = strchr(symname, '.');
1452                 if (!p)
1453                         return false;
1454                 *p = '\0';
1455                 addr = (unsigned long)kprobe_lookup_name(symname, 0);
1456                 if (addr)
1457                         return __within_kprobe_blacklist(addr);
1458         }
1459         return false;
1460 }
1461
1462 /*
1463  * If we have a symbol_name argument, look it up and add the offset field
1464  * to it. This way, we can specify a relative address to a symbol.
1465  * This returns encoded errors if it fails to look up symbol or invalid
1466  * combination of parameters.
1467  */
1468 static kprobe_opcode_t *_kprobe_addr(kprobe_opcode_t *addr,
1469                         const char *symbol_name, unsigned int offset)
1470 {
1471         if ((symbol_name && addr) || (!symbol_name && !addr))
1472                 goto invalid;
1473
1474         if (symbol_name) {
1475                 addr = kprobe_lookup_name(symbol_name, offset);
1476                 if (!addr)
1477                         return ERR_PTR(-ENOENT);
1478         }
1479
1480         addr = (kprobe_opcode_t *)(((char *)addr) + offset);
1481         if (addr)
1482                 return addr;
1483
1484 invalid:
1485         return ERR_PTR(-EINVAL);
1486 }
1487
1488 static kprobe_opcode_t *kprobe_addr(struct kprobe *p)
1489 {
1490         return _kprobe_addr(p->addr, p->symbol_name, p->offset);
1491 }
1492
1493 /* Check passed kprobe is valid and return kprobe in kprobe_table. */
1494 static struct kprobe *__get_valid_kprobe(struct kprobe *p)
1495 {
1496         struct kprobe *ap, *list_p;
1497
1498         lockdep_assert_held(&kprobe_mutex);
1499
1500         ap = get_kprobe(p->addr);
1501         if (unlikely(!ap))
1502                 return NULL;
1503
1504         if (p != ap) {
1505                 list_for_each_entry(list_p, &ap->list, list)
1506                         if (list_p == p)
1507                         /* kprobe p is a valid probe */
1508                                 goto valid;
1509                 return NULL;
1510         }
1511 valid:
1512         return ap;
1513 }
1514
1515 /*
1516  * Warn and return error if the kprobe is being re-registered since
1517  * there must be a software bug.
1518  */
1519 static inline int warn_kprobe_rereg(struct kprobe *p)
1520 {
1521         int ret = 0;
1522
1523         mutex_lock(&kprobe_mutex);
1524         if (WARN_ON_ONCE(__get_valid_kprobe(p)))
1525                 ret = -EINVAL;
1526         mutex_unlock(&kprobe_mutex);
1527
1528         return ret;
1529 }
1530
1531 int __weak arch_check_ftrace_location(struct kprobe *p)
1532 {
1533         unsigned long ftrace_addr;
1534
1535         ftrace_addr = ftrace_location((unsigned long)p->addr);
1536         if (ftrace_addr) {
1537 #ifdef CONFIG_KPROBES_ON_FTRACE
1538                 /* Given address is not on the instruction boundary */
1539                 if ((unsigned long)p->addr != ftrace_addr)
1540                         return -EILSEQ;
1541                 p->flags |= KPROBE_FLAG_FTRACE;
1542 #else   /* !CONFIG_KPROBES_ON_FTRACE */
1543                 return -EINVAL;
1544 #endif
1545         }
1546         return 0;
1547 }
1548
1549 static int check_kprobe_address_safe(struct kprobe *p,
1550                                      struct module **probed_mod)
1551 {
1552         int ret;
1553
1554         ret = arch_check_ftrace_location(p);
1555         if (ret)
1556                 return ret;
1557         jump_label_lock();
1558         preempt_disable();
1559
1560         /* Ensure it is not in reserved area nor out of text */
1561         if (!kernel_text_address((unsigned long) p->addr) ||
1562             within_kprobe_blacklist((unsigned long) p->addr) ||
1563             jump_label_text_reserved(p->addr, p->addr) ||
1564             find_bug((unsigned long)p->addr)) {
1565                 ret = -EINVAL;
1566                 goto out;
1567         }
1568
1569         /* Check if are we probing a module */
1570         *probed_mod = __module_text_address((unsigned long) p->addr);
1571         if (*probed_mod) {
1572                 /*
1573                  * We must hold a refcount of the probed module while updating
1574                  * its code to prohibit unexpected unloading.
1575                  */
1576                 if (unlikely(!try_module_get(*probed_mod))) {
1577                         ret = -ENOENT;
1578                         goto out;
1579                 }
1580
1581                 /*
1582                  * If the module freed .init.text, we couldn't insert
1583                  * kprobes in there.
1584                  */
1585                 if (within_module_init((unsigned long)p->addr, *probed_mod) &&
1586                     (*probed_mod)->state != MODULE_STATE_COMING) {
1587                         module_put(*probed_mod);
1588                         *probed_mod = NULL;
1589                         ret = -ENOENT;
1590                 }
1591         }
1592 out:
1593         preempt_enable();
1594         jump_label_unlock();
1595
1596         return ret;
1597 }
1598
1599 int register_kprobe(struct kprobe *p)
1600 {
1601         int ret;
1602         struct kprobe *old_p;
1603         struct module *probed_mod;
1604         kprobe_opcode_t *addr;
1605
1606         /* Adjust probe address from symbol */
1607         addr = kprobe_addr(p);
1608         if (IS_ERR(addr))
1609                 return PTR_ERR(addr);
1610         p->addr = addr;
1611
1612         ret = warn_kprobe_rereg(p);
1613         if (ret)
1614                 return ret;
1615
1616         /* User can pass only KPROBE_FLAG_DISABLED to register_kprobe */
1617         p->flags &= KPROBE_FLAG_DISABLED;
1618         p->nmissed = 0;
1619         INIT_LIST_HEAD(&p->list);
1620
1621         ret = check_kprobe_address_safe(p, &probed_mod);
1622         if (ret)
1623                 return ret;
1624
1625         mutex_lock(&kprobe_mutex);
1626
1627         old_p = get_kprobe(p->addr);
1628         if (old_p) {
1629                 /* Since this may unoptimize old_p, locking text_mutex. */
1630                 ret = register_aggr_kprobe(old_p, p);
1631                 goto out;
1632         }
1633
1634         cpus_read_lock();
1635         /* Prevent text modification */
1636         mutex_lock(&text_mutex);
1637         ret = prepare_kprobe(p);
1638         mutex_unlock(&text_mutex);
1639         cpus_read_unlock();
1640         if (ret)
1641                 goto out;
1642
1643         INIT_HLIST_NODE(&p->hlist);
1644         hlist_add_head_rcu(&p->hlist,
1645                        &kprobe_table[hash_ptr(p->addr, KPROBE_HASH_BITS)]);
1646
1647         if (!kprobes_all_disarmed && !kprobe_disabled(p)) {
1648                 ret = arm_kprobe(p);
1649                 if (ret) {
1650                         hlist_del_rcu(&p->hlist);
1651                         synchronize_rcu();
1652                         goto out;
1653                 }
1654         }
1655
1656         /* Try to optimize kprobe */
1657         try_to_optimize_kprobe(p);
1658 out:
1659         mutex_unlock(&kprobe_mutex);
1660
1661         if (probed_mod)
1662                 module_put(probed_mod);
1663
1664         return ret;
1665 }
1666 EXPORT_SYMBOL_GPL(register_kprobe);
1667
1668 /* Check if all probes on the aggrprobe are disabled */
1669 static int aggr_kprobe_disabled(struct kprobe *ap)
1670 {
1671         struct kprobe *kp;
1672
1673         lockdep_assert_held(&kprobe_mutex);
1674
1675         list_for_each_entry(kp, &ap->list, list)
1676                 if (!kprobe_disabled(kp))
1677                         /*
1678                          * There is an active probe on the list.
1679                          * We can't disable this ap.
1680                          */
1681                         return 0;
1682
1683         return 1;
1684 }
1685
1686 /* Disable one kprobe: Make sure called under kprobe_mutex is locked */
1687 static struct kprobe *__disable_kprobe(struct kprobe *p)
1688 {
1689         struct kprobe *orig_p;
1690         int ret;
1691
1692         /* Get an original kprobe for return */
1693         orig_p = __get_valid_kprobe(p);
1694         if (unlikely(orig_p == NULL))
1695                 return ERR_PTR(-EINVAL);
1696
1697         if (!kprobe_disabled(p)) {
1698                 /* Disable probe if it is a child probe */
1699                 if (p != orig_p)
1700                         p->flags |= KPROBE_FLAG_DISABLED;
1701
1702                 /* Try to disarm and disable this/parent probe */
1703                 if (p == orig_p || aggr_kprobe_disabled(orig_p)) {
1704                         /*
1705                          * If kprobes_all_disarmed is set, orig_p
1706                          * should have already been disarmed, so
1707                          * skip unneed disarming process.
1708                          */
1709                         if (!kprobes_all_disarmed) {
1710                                 ret = disarm_kprobe(orig_p, true);
1711                                 if (ret) {
1712                                         p->flags &= ~KPROBE_FLAG_DISABLED;
1713                                         return ERR_PTR(ret);
1714                                 }
1715                         }
1716                         orig_p->flags |= KPROBE_FLAG_DISABLED;
1717                 }
1718         }
1719
1720         return orig_p;
1721 }
1722
1723 /*
1724  * Unregister a kprobe without a scheduler synchronization.
1725  */
1726 static int __unregister_kprobe_top(struct kprobe *p)
1727 {
1728         struct kprobe *ap, *list_p;
1729
1730         /* Disable kprobe. This will disarm it if needed. */
1731         ap = __disable_kprobe(p);
1732         if (IS_ERR(ap))
1733                 return PTR_ERR(ap);
1734
1735         if (ap == p)
1736                 /*
1737                  * This probe is an independent(and non-optimized) kprobe
1738                  * (not an aggrprobe). Remove from the hash list.
1739                  */
1740                 goto disarmed;
1741
1742         /* Following process expects this probe is an aggrprobe */
1743         WARN_ON(!kprobe_aggrprobe(ap));
1744
1745         if (list_is_singular(&ap->list) && kprobe_disarmed(ap))
1746                 /*
1747                  * !disarmed could be happen if the probe is under delayed
1748                  * unoptimizing.
1749                  */
1750                 goto disarmed;
1751         else {
1752                 /* If disabling probe has special handlers, update aggrprobe */
1753                 if (p->post_handler && !kprobe_gone(p)) {
1754                         list_for_each_entry(list_p, &ap->list, list) {
1755                                 if ((list_p != p) && (list_p->post_handler))
1756                                         goto noclean;
1757                         }
1758                         ap->post_handler = NULL;
1759                 }
1760 noclean:
1761                 /*
1762                  * Remove from the aggrprobe: this path will do nothing in
1763                  * __unregister_kprobe_bottom().
1764                  */
1765                 list_del_rcu(&p->list);
1766                 if (!kprobe_disabled(ap) && !kprobes_all_disarmed)
1767                         /*
1768                          * Try to optimize this probe again, because post
1769                          * handler may have been changed.
1770                          */
1771                         optimize_kprobe(ap);
1772         }
1773         return 0;
1774
1775 disarmed:
1776         hlist_del_rcu(&ap->hlist);
1777         return 0;
1778 }
1779
1780 static void __unregister_kprobe_bottom(struct kprobe *p)
1781 {
1782         struct kprobe *ap;
1783
1784         if (list_empty(&p->list))
1785                 /* This is an independent kprobe */
1786                 arch_remove_kprobe(p);
1787         else if (list_is_singular(&p->list)) {
1788                 /* This is the last child of an aggrprobe */
1789                 ap = list_entry(p->list.next, struct kprobe, list);
1790                 list_del(&p->list);
1791                 free_aggr_kprobe(ap);
1792         }
1793         /* Otherwise, do nothing. */
1794 }
1795
1796 int register_kprobes(struct kprobe **kps, int num)
1797 {
1798         int i, ret = 0;
1799
1800         if (num <= 0)
1801                 return -EINVAL;
1802         for (i = 0; i < num; i++) {
1803                 ret = register_kprobe(kps[i]);
1804                 if (ret < 0) {
1805                         if (i > 0)
1806                                 unregister_kprobes(kps, i);
1807                         break;
1808                 }
1809         }
1810         return ret;
1811 }
1812 EXPORT_SYMBOL_GPL(register_kprobes);
1813
1814 void unregister_kprobe(struct kprobe *p)
1815 {
1816         unregister_kprobes(&p, 1);
1817 }
1818 EXPORT_SYMBOL_GPL(unregister_kprobe);
1819
1820 void unregister_kprobes(struct kprobe **kps, int num)
1821 {
1822         int i;
1823
1824         if (num <= 0)
1825                 return;
1826         mutex_lock(&kprobe_mutex);
1827         for (i = 0; i < num; i++)
1828                 if (__unregister_kprobe_top(kps[i]) < 0)
1829                         kps[i]->addr = NULL;
1830         mutex_unlock(&kprobe_mutex);
1831
1832         synchronize_rcu();
1833         for (i = 0; i < num; i++)
1834                 if (kps[i]->addr)
1835                         __unregister_kprobe_bottom(kps[i]);
1836 }
1837 EXPORT_SYMBOL_GPL(unregister_kprobes);
1838
1839 int __weak kprobe_exceptions_notify(struct notifier_block *self,
1840                                         unsigned long val, void *data)
1841 {
1842         return NOTIFY_DONE;
1843 }
1844 NOKPROBE_SYMBOL(kprobe_exceptions_notify);
1845
1846 static struct notifier_block kprobe_exceptions_nb = {
1847         .notifier_call = kprobe_exceptions_notify,
1848         .priority = 0x7fffffff /* we need to be notified first */
1849 };
1850
1851 unsigned long __weak arch_deref_entry_point(void *entry)
1852 {
1853         return (unsigned long)entry;
1854 }
1855
1856 #ifdef CONFIG_KRETPROBES
1857
1858 unsigned long __kretprobe_trampoline_handler(struct pt_regs *regs,
1859                                              void *trampoline_address,
1860                                              void *frame_pointer)
1861 {
1862         kprobe_opcode_t *correct_ret_addr = NULL;
1863         struct kretprobe_instance *ri = NULL;
1864         struct llist_node *first, *node;
1865         struct kretprobe *rp;
1866
1867         /* Find all nodes for this frame. */
1868         first = node = current->kretprobe_instances.first;
1869         while (node) {
1870                 ri = container_of(node, struct kretprobe_instance, llist);
1871
1872                 BUG_ON(ri->fp != frame_pointer);
1873
1874                 if (ri->ret_addr != trampoline_address) {
1875                         correct_ret_addr = ri->ret_addr;
1876                         /*
1877                          * This is the real return address. Any other
1878                          * instances associated with this task are for
1879                          * other calls deeper on the call stack
1880                          */
1881                         goto found;
1882                 }
1883
1884                 node = node->next;
1885         }
1886         pr_err("Oops! Kretprobe fails to find correct return address.\n");
1887         BUG_ON(1);
1888
1889 found:
1890         /* Unlink all nodes for this frame. */
1891         current->kretprobe_instances.first = node->next;
1892         node->next = NULL;
1893
1894         /* Run them..  */
1895         while (first) {
1896                 ri = container_of(first, struct kretprobe_instance, llist);
1897                 first = first->next;
1898
1899                 rp = get_kretprobe(ri);
1900                 if (rp && rp->handler) {
1901                         struct kprobe *prev = kprobe_running();
1902
1903                         __this_cpu_write(current_kprobe, &rp->kp);
1904                         ri->ret_addr = correct_ret_addr;
1905                         rp->handler(ri, regs);
1906                         __this_cpu_write(current_kprobe, prev);
1907                 }
1908
1909                 recycle_rp_inst(ri);
1910         }
1911
1912         return (unsigned long)correct_ret_addr;
1913 }
1914 NOKPROBE_SYMBOL(__kretprobe_trampoline_handler)
1915
1916 /*
1917  * This kprobe pre_handler is registered with every kretprobe. When probe
1918  * hits it will set up the return probe.
1919  */
1920 static int pre_handler_kretprobe(struct kprobe *p, struct pt_regs *regs)
1921 {
1922         struct kretprobe *rp = container_of(p, struct kretprobe, kp);
1923         struct kretprobe_instance *ri;
1924         struct freelist_node *fn;
1925
1926         fn = freelist_try_get(&rp->freelist);
1927         if (!fn) {
1928                 rp->nmissed++;
1929                 return 0;
1930         }
1931
1932         ri = container_of(fn, struct kretprobe_instance, freelist);
1933
1934         if (rp->entry_handler && rp->entry_handler(ri, regs)) {
1935                 freelist_add(&ri->freelist, &rp->freelist);
1936                 return 0;
1937         }
1938
1939         arch_prepare_kretprobe(ri, regs);
1940
1941         __llist_add(&ri->llist, &current->kretprobe_instances);
1942
1943         return 0;
1944 }
1945 NOKPROBE_SYMBOL(pre_handler_kretprobe);
1946
1947 bool __weak arch_kprobe_on_func_entry(unsigned long offset)
1948 {
1949         return !offset;
1950 }
1951
1952 /**
1953  * kprobe_on_func_entry() -- check whether given address is function entry
1954  * @addr: Target address
1955  * @sym:  Target symbol name
1956  * @offset: The offset from the symbol or the address
1957  *
1958  * This checks whether the given @addr+@offset or @sym+@offset is on the
1959  * function entry address or not.
1960  * This returns 0 if it is the function entry, or -EINVAL if it is not.
1961  * And also it returns -ENOENT if it fails the symbol or address lookup.
1962  * Caller must pass @addr or @sym (either one must be NULL), or this
1963  * returns -EINVAL.
1964  */
1965 int kprobe_on_func_entry(kprobe_opcode_t *addr, const char *sym, unsigned long offset)
1966 {
1967         kprobe_opcode_t *kp_addr = _kprobe_addr(addr, sym, offset);
1968
1969         if (IS_ERR(kp_addr))
1970                 return PTR_ERR(kp_addr);
1971
1972         if (!kallsyms_lookup_size_offset((unsigned long)kp_addr, NULL, &offset))
1973                 return -ENOENT;
1974
1975         if (!arch_kprobe_on_func_entry(offset))
1976                 return -EINVAL;
1977
1978         return 0;
1979 }
1980
1981 int register_kretprobe(struct kretprobe *rp)
1982 {
1983         int ret;
1984         struct kretprobe_instance *inst;
1985         int i;
1986         void *addr;
1987
1988         ret = kprobe_on_func_entry(rp->kp.addr, rp->kp.symbol_name, rp->kp.offset);
1989         if (ret)
1990                 return ret;
1991
1992         /* If only rp->kp.addr is specified, check reregistering kprobes */
1993         if (rp->kp.addr && warn_kprobe_rereg(&rp->kp))
1994                 return -EINVAL;
1995
1996         if (kretprobe_blacklist_size) {
1997                 addr = kprobe_addr(&rp->kp);
1998                 if (IS_ERR(addr))
1999                         return PTR_ERR(addr);
2000
2001                 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
2002                         if (kretprobe_blacklist[i].addr == addr)
2003                                 return -EINVAL;
2004                 }
2005         }
2006
2007         rp->kp.pre_handler = pre_handler_kretprobe;
2008         rp->kp.post_handler = NULL;
2009
2010         /* Pre-allocate memory for max kretprobe instances */
2011         if (rp->maxactive <= 0) {
2012 #ifdef CONFIG_PREEMPTION
2013                 rp->maxactive = max_t(unsigned int, 10, 2*num_possible_cpus());
2014 #else
2015                 rp->maxactive = num_possible_cpus();
2016 #endif
2017         }
2018         rp->freelist.head = NULL;
2019         rp->rph = kzalloc(sizeof(struct kretprobe_holder), GFP_KERNEL);
2020         if (!rp->rph)
2021                 return -ENOMEM;
2022
2023         rp->rph->rp = rp;
2024         for (i = 0; i < rp->maxactive; i++) {
2025                 inst = kzalloc(sizeof(struct kretprobe_instance) +
2026                                rp->data_size, GFP_KERNEL);
2027                 if (inst == NULL) {
2028                         refcount_set(&rp->rph->ref, i);
2029                         free_rp_inst(rp);
2030                         return -ENOMEM;
2031                 }
2032                 inst->rph = rp->rph;
2033                 freelist_add(&inst->freelist, &rp->freelist);
2034         }
2035         refcount_set(&rp->rph->ref, i);
2036
2037         rp->nmissed = 0;
2038         /* Establish function entry probe point */
2039         ret = register_kprobe(&rp->kp);
2040         if (ret != 0)
2041                 free_rp_inst(rp);
2042         return ret;
2043 }
2044 EXPORT_SYMBOL_GPL(register_kretprobe);
2045
2046 int register_kretprobes(struct kretprobe **rps, int num)
2047 {
2048         int ret = 0, i;
2049
2050         if (num <= 0)
2051                 return -EINVAL;
2052         for (i = 0; i < num; i++) {
2053                 ret = register_kretprobe(rps[i]);
2054                 if (ret < 0) {
2055                         if (i > 0)
2056                                 unregister_kretprobes(rps, i);
2057                         break;
2058                 }
2059         }
2060         return ret;
2061 }
2062 EXPORT_SYMBOL_GPL(register_kretprobes);
2063
2064 void unregister_kretprobe(struct kretprobe *rp)
2065 {
2066         unregister_kretprobes(&rp, 1);
2067 }
2068 EXPORT_SYMBOL_GPL(unregister_kretprobe);
2069
2070 void unregister_kretprobes(struct kretprobe **rps, int num)
2071 {
2072         int i;
2073
2074         if (num <= 0)
2075                 return;
2076         mutex_lock(&kprobe_mutex);
2077         for (i = 0; i < num; i++) {
2078                 if (__unregister_kprobe_top(&rps[i]->kp) < 0)
2079                         rps[i]->kp.addr = NULL;
2080                 rps[i]->rph->rp = NULL;
2081         }
2082         mutex_unlock(&kprobe_mutex);
2083
2084         synchronize_rcu();
2085         for (i = 0; i < num; i++) {
2086                 if (rps[i]->kp.addr) {
2087                         __unregister_kprobe_bottom(&rps[i]->kp);
2088                         free_rp_inst(rps[i]);
2089                 }
2090         }
2091 }
2092 EXPORT_SYMBOL_GPL(unregister_kretprobes);
2093
2094 #else /* CONFIG_KRETPROBES */
2095 int register_kretprobe(struct kretprobe *rp)
2096 {
2097         return -ENOSYS;
2098 }
2099 EXPORT_SYMBOL_GPL(register_kretprobe);
2100
2101 int register_kretprobes(struct kretprobe **rps, int num)
2102 {
2103         return -ENOSYS;
2104 }
2105 EXPORT_SYMBOL_GPL(register_kretprobes);
2106
2107 void unregister_kretprobe(struct kretprobe *rp)
2108 {
2109 }
2110 EXPORT_SYMBOL_GPL(unregister_kretprobe);
2111
2112 void unregister_kretprobes(struct kretprobe **rps, int num)
2113 {
2114 }
2115 EXPORT_SYMBOL_GPL(unregister_kretprobes);
2116
2117 static int pre_handler_kretprobe(struct kprobe *p, struct pt_regs *regs)
2118 {
2119         return 0;
2120 }
2121 NOKPROBE_SYMBOL(pre_handler_kretprobe);
2122
2123 #endif /* CONFIG_KRETPROBES */
2124
2125 /* Set the kprobe gone and remove its instruction buffer. */
2126 static void kill_kprobe(struct kprobe *p)
2127 {
2128         struct kprobe *kp;
2129
2130         lockdep_assert_held(&kprobe_mutex);
2131
2132         p->flags |= KPROBE_FLAG_GONE;
2133         if (kprobe_aggrprobe(p)) {
2134                 /*
2135                  * If this is an aggr_kprobe, we have to list all the
2136                  * chained probes and mark them GONE.
2137                  */
2138                 list_for_each_entry(kp, &p->list, list)
2139                         kp->flags |= KPROBE_FLAG_GONE;
2140                 p->post_handler = NULL;
2141                 kill_optimized_kprobe(p);
2142         }
2143         /*
2144          * Here, we can remove insn_slot safely, because no thread calls
2145          * the original probed function (which will be freed soon) any more.
2146          */
2147         arch_remove_kprobe(p);
2148
2149         /*
2150          * The module is going away. We should disarm the kprobe which
2151          * is using ftrace, because ftrace framework is still available at
2152          * MODULE_STATE_GOING notification.
2153          */
2154         if (kprobe_ftrace(p) && !kprobe_disabled(p) && !kprobes_all_disarmed)
2155                 disarm_kprobe_ftrace(p);
2156 }
2157
2158 /* Disable one kprobe */
2159 int disable_kprobe(struct kprobe *kp)
2160 {
2161         int ret = 0;
2162         struct kprobe *p;
2163
2164         mutex_lock(&kprobe_mutex);
2165
2166         /* Disable this kprobe */
2167         p = __disable_kprobe(kp);
2168         if (IS_ERR(p))
2169                 ret = PTR_ERR(p);
2170
2171         mutex_unlock(&kprobe_mutex);
2172         return ret;
2173 }
2174 EXPORT_SYMBOL_GPL(disable_kprobe);
2175
2176 /* Enable one kprobe */
2177 int enable_kprobe(struct kprobe *kp)
2178 {
2179         int ret = 0;
2180         struct kprobe *p;
2181
2182         mutex_lock(&kprobe_mutex);
2183
2184         /* Check whether specified probe is valid. */
2185         p = __get_valid_kprobe(kp);
2186         if (unlikely(p == NULL)) {
2187                 ret = -EINVAL;
2188                 goto out;
2189         }
2190
2191         if (kprobe_gone(kp)) {
2192                 /* This kprobe has gone, we couldn't enable it. */
2193                 ret = -EINVAL;
2194                 goto out;
2195         }
2196
2197         if (p != kp)
2198                 kp->flags &= ~KPROBE_FLAG_DISABLED;
2199
2200         if (!kprobes_all_disarmed && kprobe_disabled(p)) {
2201                 p->flags &= ~KPROBE_FLAG_DISABLED;
2202                 ret = arm_kprobe(p);
2203                 if (ret)
2204                         p->flags |= KPROBE_FLAG_DISABLED;
2205         }
2206 out:
2207         mutex_unlock(&kprobe_mutex);
2208         return ret;
2209 }
2210 EXPORT_SYMBOL_GPL(enable_kprobe);
2211
2212 /* Caller must NOT call this in usual path. This is only for critical case */
2213 void dump_kprobe(struct kprobe *kp)
2214 {
2215         pr_err("Dumping kprobe:\n");
2216         pr_err("Name: %s\nOffset: %x\nAddress: %pS\n",
2217                kp->symbol_name, kp->offset, kp->addr);
2218 }
2219 NOKPROBE_SYMBOL(dump_kprobe);
2220
2221 int kprobe_add_ksym_blacklist(unsigned long entry)
2222 {
2223         struct kprobe_blacklist_entry *ent;
2224         unsigned long offset = 0, size = 0;
2225
2226         if (!kernel_text_address(entry) ||
2227             !kallsyms_lookup_size_offset(entry, &size, &offset))
2228                 return -EINVAL;
2229
2230         ent = kmalloc(sizeof(*ent), GFP_KERNEL);
2231         if (!ent)
2232                 return -ENOMEM;
2233         ent->start_addr = entry;
2234         ent->end_addr = entry + size;
2235         INIT_LIST_HEAD(&ent->list);
2236         list_add_tail(&ent->list, &kprobe_blacklist);
2237
2238         return (int)size;
2239 }
2240
2241 /* Add all symbols in given area into kprobe blacklist */
2242 int kprobe_add_area_blacklist(unsigned long start, unsigned long end)
2243 {
2244         unsigned long entry;
2245         int ret = 0;
2246
2247         for (entry = start; entry < end; entry += ret) {
2248                 ret = kprobe_add_ksym_blacklist(entry);
2249                 if (ret < 0)
2250                         return ret;
2251                 if (ret == 0)   /* In case of alias symbol */
2252                         ret = 1;
2253         }
2254         return 0;
2255 }
2256
2257 /* Remove all symbols in given area from kprobe blacklist */
2258 static void kprobe_remove_area_blacklist(unsigned long start, unsigned long end)
2259 {
2260         struct kprobe_blacklist_entry *ent, *n;
2261
2262         list_for_each_entry_safe(ent, n, &kprobe_blacklist, list) {
2263                 if (ent->start_addr < start || ent->start_addr >= end)
2264                         continue;
2265                 list_del(&ent->list);
2266                 kfree(ent);
2267         }
2268 }
2269
2270 static void kprobe_remove_ksym_blacklist(unsigned long entry)
2271 {
2272         kprobe_remove_area_blacklist(entry, entry + 1);
2273 }
2274
2275 int __weak arch_kprobe_get_kallsym(unsigned int *symnum, unsigned long *value,
2276                                    char *type, char *sym)
2277 {
2278         return -ERANGE;
2279 }
2280
2281 int kprobe_get_kallsym(unsigned int symnum, unsigned long *value, char *type,
2282                        char *sym)
2283 {
2284 #ifdef __ARCH_WANT_KPROBES_INSN_SLOT
2285         if (!kprobe_cache_get_kallsym(&kprobe_insn_slots, &symnum, value, type, sym))
2286                 return 0;
2287 #ifdef CONFIG_OPTPROBES
2288         if (!kprobe_cache_get_kallsym(&kprobe_optinsn_slots, &symnum, value, type, sym))
2289                 return 0;
2290 #endif
2291 #endif
2292         if (!arch_kprobe_get_kallsym(&symnum, value, type, sym))
2293                 return 0;
2294         return -ERANGE;
2295 }
2296
2297 int __init __weak arch_populate_kprobe_blacklist(void)
2298 {
2299         return 0;
2300 }
2301
2302 /*
2303  * Lookup and populate the kprobe_blacklist.
2304  *
2305  * Unlike the kretprobe blacklist, we'll need to determine
2306  * the range of addresses that belong to the said functions,
2307  * since a kprobe need not necessarily be at the beginning
2308  * of a function.
2309  */
2310 static int __init populate_kprobe_blacklist(unsigned long *start,
2311                                              unsigned long *end)
2312 {
2313         unsigned long entry;
2314         unsigned long *iter;
2315         int ret;
2316
2317         for (iter = start; iter < end; iter++) {
2318                 entry = arch_deref_entry_point((void *)*iter);
2319                 ret = kprobe_add_ksym_blacklist(entry);
2320                 if (ret == -EINVAL)
2321                         continue;
2322                 if (ret < 0)
2323                         return ret;
2324         }
2325
2326         /* Symbols in __kprobes_text are blacklisted */
2327         ret = kprobe_add_area_blacklist((unsigned long)__kprobes_text_start,
2328                                         (unsigned long)__kprobes_text_end);
2329         if (ret)
2330                 return ret;
2331
2332         /* Symbols in noinstr section are blacklisted */
2333         ret = kprobe_add_area_blacklist((unsigned long)__noinstr_text_start,
2334                                         (unsigned long)__noinstr_text_end);
2335
2336         return ret ? : arch_populate_kprobe_blacklist();
2337 }
2338
2339 static void add_module_kprobe_blacklist(struct module *mod)
2340 {
2341         unsigned long start, end;
2342         int i;
2343
2344         if (mod->kprobe_blacklist) {
2345                 for (i = 0; i < mod->num_kprobe_blacklist; i++)
2346                         kprobe_add_ksym_blacklist(mod->kprobe_blacklist[i]);
2347         }
2348
2349         start = (unsigned long)mod->kprobes_text_start;
2350         if (start) {
2351                 end = start + mod->kprobes_text_size;
2352                 kprobe_add_area_blacklist(start, end);
2353         }
2354
2355         start = (unsigned long)mod->noinstr_text_start;
2356         if (start) {
2357                 end = start + mod->noinstr_text_size;
2358                 kprobe_add_area_blacklist(start, end);
2359         }
2360 }
2361
2362 static void remove_module_kprobe_blacklist(struct module *mod)
2363 {
2364         unsigned long start, end;
2365         int i;
2366
2367         if (mod->kprobe_blacklist) {
2368                 for (i = 0; i < mod->num_kprobe_blacklist; i++)
2369                         kprobe_remove_ksym_blacklist(mod->kprobe_blacklist[i]);
2370         }
2371
2372         start = (unsigned long)mod->kprobes_text_start;
2373         if (start) {
2374                 end = start + mod->kprobes_text_size;
2375                 kprobe_remove_area_blacklist(start, end);
2376         }
2377
2378         start = (unsigned long)mod->noinstr_text_start;
2379         if (start) {
2380                 end = start + mod->noinstr_text_size;
2381                 kprobe_remove_area_blacklist(start, end);
2382         }
2383 }
2384
2385 /* Module notifier call back, checking kprobes on the module */
2386 static int kprobes_module_callback(struct notifier_block *nb,
2387                                    unsigned long val, void *data)
2388 {
2389         struct module *mod = data;
2390         struct hlist_head *head;
2391         struct kprobe *p;
2392         unsigned int i;
2393         int checkcore = (val == MODULE_STATE_GOING);
2394
2395         if (val == MODULE_STATE_COMING) {
2396                 mutex_lock(&kprobe_mutex);
2397                 add_module_kprobe_blacklist(mod);
2398                 mutex_unlock(&kprobe_mutex);
2399         }
2400         if (val != MODULE_STATE_GOING && val != MODULE_STATE_LIVE)
2401                 return NOTIFY_DONE;
2402
2403         /*
2404          * When MODULE_STATE_GOING was notified, both of module .text and
2405          * .init.text sections would be freed. When MODULE_STATE_LIVE was
2406          * notified, only .init.text section would be freed. We need to
2407          * disable kprobes which have been inserted in the sections.
2408          */
2409         mutex_lock(&kprobe_mutex);
2410         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2411                 head = &kprobe_table[i];
2412                 hlist_for_each_entry(p, head, hlist)
2413                         if (within_module_init((unsigned long)p->addr, mod) ||
2414                             (checkcore &&
2415                              within_module_core((unsigned long)p->addr, mod))) {
2416                                 /*
2417                                  * The vaddr this probe is installed will soon
2418                                  * be vfreed buy not synced to disk. Hence,
2419                                  * disarming the breakpoint isn't needed.
2420                                  *
2421                                  * Note, this will also move any optimized probes
2422                                  * that are pending to be removed from their
2423                                  * corresponding lists to the freeing_list and
2424                                  * will not be touched by the delayed
2425                                  * kprobe_optimizer work handler.
2426                                  */
2427                                 kill_kprobe(p);
2428                         }
2429         }
2430         if (val == MODULE_STATE_GOING)
2431                 remove_module_kprobe_blacklist(mod);
2432         mutex_unlock(&kprobe_mutex);
2433         return NOTIFY_DONE;
2434 }
2435
2436 static struct notifier_block kprobe_module_nb = {
2437         .notifier_call = kprobes_module_callback,
2438         .priority = 0
2439 };
2440
2441 /* Markers of _kprobe_blacklist section */
2442 extern unsigned long __start_kprobe_blacklist[];
2443 extern unsigned long __stop_kprobe_blacklist[];
2444
2445 void kprobe_free_init_mem(void)
2446 {
2447         void *start = (void *)(&__init_begin);
2448         void *end = (void *)(&__init_end);
2449         struct hlist_head *head;
2450         struct kprobe *p;
2451         int i;
2452
2453         mutex_lock(&kprobe_mutex);
2454
2455         /* Kill all kprobes on initmem */
2456         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2457                 head = &kprobe_table[i];
2458                 hlist_for_each_entry(p, head, hlist) {
2459                         if (start <= (void *)p->addr && (void *)p->addr < end)
2460                                 kill_kprobe(p);
2461                 }
2462         }
2463
2464         mutex_unlock(&kprobe_mutex);
2465 }
2466
2467 static int __init init_kprobes(void)
2468 {
2469         int i, err = 0;
2470
2471         /* FIXME allocate the probe table, currently defined statically */
2472         /* initialize all list heads */
2473         for (i = 0; i < KPROBE_TABLE_SIZE; i++)
2474                 INIT_HLIST_HEAD(&kprobe_table[i]);
2475
2476         err = populate_kprobe_blacklist(__start_kprobe_blacklist,
2477                                         __stop_kprobe_blacklist);
2478         if (err) {
2479                 pr_err("kprobes: failed to populate blacklist: %d\n", err);
2480                 pr_err("Please take care of using kprobes.\n");
2481         }
2482
2483         if (kretprobe_blacklist_size) {
2484                 /* lookup the function address from its name */
2485                 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
2486                         kretprobe_blacklist[i].addr =
2487                                 kprobe_lookup_name(kretprobe_blacklist[i].name, 0);
2488                         if (!kretprobe_blacklist[i].addr)
2489                                 printk("kretprobe: lookup failed: %s\n",
2490                                        kretprobe_blacklist[i].name);
2491                 }
2492         }
2493
2494         /* By default, kprobes are armed */
2495         kprobes_all_disarmed = false;
2496
2497 #if defined(CONFIG_OPTPROBES) && defined(__ARCH_WANT_KPROBES_INSN_SLOT)
2498         /* Init kprobe_optinsn_slots for allocation */
2499         kprobe_optinsn_slots.insn_size = MAX_OPTINSN_SIZE;
2500 #endif
2501
2502         err = arch_init_kprobes();
2503         if (!err)
2504                 err = register_die_notifier(&kprobe_exceptions_nb);
2505         if (!err)
2506                 err = register_module_notifier(&kprobe_module_nb);
2507
2508         kprobes_initialized = (err == 0);
2509
2510         if (!err)
2511                 init_test_probes();
2512         return err;
2513 }
2514 early_initcall(init_kprobes);
2515
2516 #if defined(CONFIG_OPTPROBES)
2517 static int __init init_optprobes(void)
2518 {
2519         /*
2520          * Enable kprobe optimization - this kicks the optimizer which
2521          * depends on synchronize_rcu_tasks() and ksoftirqd, that is
2522          * not spawned in early initcall. So delay the optimization.
2523          */
2524         optimize_all_kprobes();
2525
2526         return 0;
2527 }
2528 subsys_initcall(init_optprobes);
2529 #endif
2530
2531 #ifdef CONFIG_DEBUG_FS
2532 static void report_probe(struct seq_file *pi, struct kprobe *p,
2533                 const char *sym, int offset, char *modname, struct kprobe *pp)
2534 {
2535         char *kprobe_type;
2536         void *addr = p->addr;
2537
2538         if (p->pre_handler == pre_handler_kretprobe)
2539                 kprobe_type = "r";
2540         else
2541                 kprobe_type = "k";
2542
2543         if (!kallsyms_show_value(pi->file->f_cred))
2544                 addr = NULL;
2545
2546         if (sym)
2547                 seq_printf(pi, "%px  %s  %s+0x%x  %s ",
2548                         addr, kprobe_type, sym, offset,
2549                         (modname ? modname : " "));
2550         else    /* try to use %pS */
2551                 seq_printf(pi, "%px  %s  %pS ",
2552                         addr, kprobe_type, p->addr);
2553
2554         if (!pp)
2555                 pp = p;
2556         seq_printf(pi, "%s%s%s%s\n",
2557                 (kprobe_gone(p) ? "[GONE]" : ""),
2558                 ((kprobe_disabled(p) && !kprobe_gone(p)) ?  "[DISABLED]" : ""),
2559                 (kprobe_optimized(pp) ? "[OPTIMIZED]" : ""),
2560                 (kprobe_ftrace(pp) ? "[FTRACE]" : ""));
2561 }
2562
2563 static void *kprobe_seq_start(struct seq_file *f, loff_t *pos)
2564 {
2565         return (*pos < KPROBE_TABLE_SIZE) ? pos : NULL;
2566 }
2567
2568 static void *kprobe_seq_next(struct seq_file *f, void *v, loff_t *pos)
2569 {
2570         (*pos)++;
2571         if (*pos >= KPROBE_TABLE_SIZE)
2572                 return NULL;
2573         return pos;
2574 }
2575
2576 static void kprobe_seq_stop(struct seq_file *f, void *v)
2577 {
2578         /* Nothing to do */
2579 }
2580
2581 static int show_kprobe_addr(struct seq_file *pi, void *v)
2582 {
2583         struct hlist_head *head;
2584         struct kprobe *p, *kp;
2585         const char *sym = NULL;
2586         unsigned int i = *(loff_t *) v;
2587         unsigned long offset = 0;
2588         char *modname, namebuf[KSYM_NAME_LEN];
2589
2590         head = &kprobe_table[i];
2591         preempt_disable();
2592         hlist_for_each_entry_rcu(p, head, hlist) {
2593                 sym = kallsyms_lookup((unsigned long)p->addr, NULL,
2594                                         &offset, &modname, namebuf);
2595                 if (kprobe_aggrprobe(p)) {
2596                         list_for_each_entry_rcu(kp, &p->list, list)
2597                                 report_probe(pi, kp, sym, offset, modname, p);
2598                 } else
2599                         report_probe(pi, p, sym, offset, modname, NULL);
2600         }
2601         preempt_enable();
2602         return 0;
2603 }
2604
2605 static const struct seq_operations kprobes_sops = {
2606         .start = kprobe_seq_start,
2607         .next  = kprobe_seq_next,
2608         .stop  = kprobe_seq_stop,
2609         .show  = show_kprobe_addr
2610 };
2611
2612 DEFINE_SEQ_ATTRIBUTE(kprobes);
2613
2614 /* kprobes/blacklist -- shows which functions can not be probed */
2615 static void *kprobe_blacklist_seq_start(struct seq_file *m, loff_t *pos)
2616 {
2617         mutex_lock(&kprobe_mutex);
2618         return seq_list_start(&kprobe_blacklist, *pos);
2619 }
2620
2621 static void *kprobe_blacklist_seq_next(struct seq_file *m, void *v, loff_t *pos)
2622 {
2623         return seq_list_next(v, &kprobe_blacklist, pos);
2624 }
2625
2626 static int kprobe_blacklist_seq_show(struct seq_file *m, void *v)
2627 {
2628         struct kprobe_blacklist_entry *ent =
2629                 list_entry(v, struct kprobe_blacklist_entry, list);
2630
2631         /*
2632          * If /proc/kallsyms is not showing kernel address, we won't
2633          * show them here either.
2634          */
2635         if (!kallsyms_show_value(m->file->f_cred))
2636                 seq_printf(m, "0x%px-0x%px\t%ps\n", NULL, NULL,
2637                            (void *)ent->start_addr);
2638         else
2639                 seq_printf(m, "0x%px-0x%px\t%ps\n", (void *)ent->start_addr,
2640                            (void *)ent->end_addr, (void *)ent->start_addr);
2641         return 0;
2642 }
2643
2644 static void kprobe_blacklist_seq_stop(struct seq_file *f, void *v)
2645 {
2646         mutex_unlock(&kprobe_mutex);
2647 }
2648
2649 static const struct seq_operations kprobe_blacklist_sops = {
2650         .start = kprobe_blacklist_seq_start,
2651         .next  = kprobe_blacklist_seq_next,
2652         .stop  = kprobe_blacklist_seq_stop,
2653         .show  = kprobe_blacklist_seq_show,
2654 };
2655 DEFINE_SEQ_ATTRIBUTE(kprobe_blacklist);
2656
2657 static int arm_all_kprobes(void)
2658 {
2659         struct hlist_head *head;
2660         struct kprobe *p;
2661         unsigned int i, total = 0, errors = 0;
2662         int err, ret = 0;
2663
2664         mutex_lock(&kprobe_mutex);
2665
2666         /* If kprobes are armed, just return */
2667         if (!kprobes_all_disarmed)
2668                 goto already_enabled;
2669
2670         /*
2671          * optimize_kprobe() called by arm_kprobe() checks
2672          * kprobes_all_disarmed, so set kprobes_all_disarmed before
2673          * arm_kprobe.
2674          */
2675         kprobes_all_disarmed = false;
2676         /* Arming kprobes doesn't optimize kprobe itself */
2677         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2678                 head = &kprobe_table[i];
2679                 /* Arm all kprobes on a best-effort basis */
2680                 hlist_for_each_entry(p, head, hlist) {
2681                         if (!kprobe_disabled(p)) {
2682                                 err = arm_kprobe(p);
2683                                 if (err)  {
2684                                         errors++;
2685                                         ret = err;
2686                                 }
2687                                 total++;
2688                         }
2689                 }
2690         }
2691
2692         if (errors)
2693                 pr_warn("Kprobes globally enabled, but failed to arm %d out of %d probes\n",
2694                         errors, total);
2695         else
2696                 pr_info("Kprobes globally enabled\n");
2697
2698 already_enabled:
2699         mutex_unlock(&kprobe_mutex);
2700         return ret;
2701 }
2702
2703 static int disarm_all_kprobes(void)
2704 {
2705         struct hlist_head *head;
2706         struct kprobe *p;
2707         unsigned int i, total = 0, errors = 0;
2708         int err, ret = 0;
2709
2710         mutex_lock(&kprobe_mutex);
2711
2712         /* If kprobes are already disarmed, just return */
2713         if (kprobes_all_disarmed) {
2714                 mutex_unlock(&kprobe_mutex);
2715                 return 0;
2716         }
2717
2718         kprobes_all_disarmed = true;
2719
2720         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2721                 head = &kprobe_table[i];
2722                 /* Disarm all kprobes on a best-effort basis */
2723                 hlist_for_each_entry(p, head, hlist) {
2724                         if (!arch_trampoline_kprobe(p) && !kprobe_disabled(p)) {
2725                                 err = disarm_kprobe(p, false);
2726                                 if (err) {
2727                                         errors++;
2728                                         ret = err;
2729                                 }
2730                                 total++;
2731                         }
2732                 }
2733         }
2734
2735         if (errors)
2736                 pr_warn("Kprobes globally disabled, but failed to disarm %d out of %d probes\n",
2737                         errors, total);
2738         else
2739                 pr_info("Kprobes globally disabled\n");
2740
2741         mutex_unlock(&kprobe_mutex);
2742
2743         /* Wait for disarming all kprobes by optimizer */
2744         wait_for_kprobe_optimizer();
2745
2746         return ret;
2747 }
2748
2749 /*
2750  * XXX: The debugfs bool file interface doesn't allow for callbacks
2751  * when the bool state is switched. We can reuse that facility when
2752  * available
2753  */
2754 static ssize_t read_enabled_file_bool(struct file *file,
2755                char __user *user_buf, size_t count, loff_t *ppos)
2756 {
2757         char buf[3];
2758
2759         if (!kprobes_all_disarmed)
2760                 buf[0] = '1';
2761         else
2762                 buf[0] = '0';
2763         buf[1] = '\n';
2764         buf[2] = 0x00;
2765         return simple_read_from_buffer(user_buf, count, ppos, buf, 2);
2766 }
2767
2768 static ssize_t write_enabled_file_bool(struct file *file,
2769                const char __user *user_buf, size_t count, loff_t *ppos)
2770 {
2771         char buf[32];
2772         size_t buf_size;
2773         int ret = 0;
2774
2775         buf_size = min(count, (sizeof(buf)-1));
2776         if (copy_from_user(buf, user_buf, buf_size))
2777                 return -EFAULT;
2778
2779         buf[buf_size] = '\0';
2780         switch (buf[0]) {
2781         case 'y':
2782         case 'Y':
2783         case '1':
2784                 ret = arm_all_kprobes();
2785                 break;
2786         case 'n':
2787         case 'N':
2788         case '0':
2789                 ret = disarm_all_kprobes();
2790                 break;
2791         default:
2792                 return -EINVAL;
2793         }
2794
2795         if (ret)
2796                 return ret;
2797
2798         return count;
2799 }
2800
2801 static const struct file_operations fops_kp = {
2802         .read =         read_enabled_file_bool,
2803         .write =        write_enabled_file_bool,
2804         .llseek =       default_llseek,
2805 };
2806
2807 static int __init debugfs_kprobe_init(void)
2808 {
2809         struct dentry *dir;
2810         unsigned int value = 1;
2811
2812         dir = debugfs_create_dir("kprobes", NULL);
2813
2814         debugfs_create_file("list", 0400, dir, NULL, &kprobes_fops);
2815
2816         debugfs_create_file("enabled", 0600, dir, &value, &fops_kp);
2817
2818         debugfs_create_file("blacklist", 0400, dir, NULL,
2819                             &kprobe_blacklist_fops);
2820
2821         return 0;
2822 }
2823
2824 late_initcall(debugfs_kprobe_init);
2825 #endif /* CONFIG_DEBUG_FS */