kernel/kprobes.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  *  Kernel Probes (KProbes)
   4  *
   5  * Copyright (C) IBM Corporation, 2002, 2004
   6  *
   7  * 2002-Oct     Created by Vamsi Krishna S <vamsi_krishna@in.ibm.com> Kernel
   8  *              Probes initial implementation (includes suggestions from
   9  *              Rusty Russell).
  10  * 2004-Aug     Updated by Prasanna S Panchamukhi <prasanna@in.ibm.com> with
  11  *              hlists and exceptions notifier as suggested by Andi Kleen.
  12  * 2004-July    Suparna Bhattacharya <suparna@in.ibm.com> added jumper probes
  13  *              interface to access function arguments.
  14  * 2004-Sep     Prasanna S Panchamukhi <prasanna@in.ibm.com> Changed Kprobes
  15  *              exceptions notifier to be first on the priority list.
  16  * 2005-May     Hien Nguyen <hien@us.ibm.com>, Jim Keniston
  17  *              <jkenisto@us.ibm.com> and Prasanna S Panchamukhi
  18  *              <prasanna@in.ibm.com> added function-return probes.
  19  */
  20
  21 #define pr_fmt(fmt) "kprobes: " fmt
  22
  23 #include <linux/kprobes.h>
  24 #include <linux/hash.h>
  25 #include <linux/init.h>
  26 #include <linux/slab.h>
  27 #include <linux/stddef.h>
  28 #include <linux/export.h>
  29 #include <linux/moduleloader.h>
  30 #include <linux/kallsyms.h>
  31 #include <linux/freezer.h>
  32 #include <linux/seq_file.h>
  33 #include <linux/debugfs.h>
  34 #include <linux/sysctl.h>
  35 #include <linux/kdebug.h>
  36 #include <linux/memory.h>
  37 #include <linux/ftrace.h>
  38 #include <linux/cpu.h>
  39 #include <linux/jump_label.h>
  40 #include <linux/static_call.h>
  41 #include <linux/perf_event.h>
  42
  43 #include <asm/sections.h>
  44 #include <asm/cacheflush.h>
  45 #include <asm/errno.h>
  46 #include <linux/uaccess.h>
  47
  48 #define KPROBE_HASH_BITS 6
  49 #define KPROBE_TABLE_SIZE (1 << KPROBE_HASH_BITS)
  50
  51 #if !defined(CONFIG_OPTPROBES) || !defined(CONFIG_SYSCTL)
  52 #define kprobe_sysctls_init() do { } while (0)
  53 #endif
  54
  55 static int kprobes_initialized;
  56 /* kprobe_table can be accessed by
  57  * - Normal hlist traversal and RCU add/del under 'kprobe_mutex' is held.
  58  * Or
  59  * - RCU hlist traversal under disabling preempt (breakpoint handlers)
  60  */
  61 static struct hlist_head kprobe_table[KPROBE_TABLE_SIZE];
  62
  63 /* NOTE: change this value only with 'kprobe_mutex' held */
  64 static bool kprobes_all_disarmed;
  65
  66 /* This protects 'kprobe_table' and 'optimizing_list' */
  67 static DEFINE_MUTEX(kprobe_mutex);
  68 static DEFINE_PER_CPU(struct kprobe *, kprobe_instance);
  69
  70 kprobe_opcode_t * __weak kprobe_lookup_name(const char *name,
  71                                         unsigned int __unused)
  72 {
  73         return ((kprobe_opcode_t *)(kallsyms_lookup_name(name)));
  74 }
  75
  76 /*
  77  * Blacklist -- list of 'struct kprobe_blacklist_entry' to store info where
  78  * kprobes can not probe.
  79  */
  80 static LIST_HEAD(kprobe_blacklist);
  81
  82 #ifdef __ARCH_WANT_KPROBES_INSN_SLOT
  83 /*
  84  * 'kprobe::ainsn.insn' points to the copy of the instruction to be
  85  * single-stepped. x86_64, POWER4 and above have no-exec support and
  86  * stepping on the instruction on a vmalloced/kmalloced/data page
  87  * is a recipe for disaster
  88  */
  89 struct kprobe_insn_page {
  90         struct list_head list;
  91         kprobe_opcode_t *insns;         /* Page of instruction slots */
  92         struct kprobe_insn_cache *cache;
  93         int nused;
  94         int ngarbage;
  95         char slot_used[];
  96 };
  97
  98 #define KPROBE_INSN_PAGE_SIZE(slots)                    \
  99         (offsetof(struct kprobe_insn_page, slot_used) + \
 100          (sizeof(char) * (slots)))
 101
 102 static int slots_per_page(struct kprobe_insn_cache *c)
 103 {
 104         return PAGE_SIZE/(c->insn_size * sizeof(kprobe_opcode_t));
 105 }
 106
 107 enum kprobe_slot_state {
 108         SLOT_CLEAN = 0,
 109         SLOT_DIRTY = 1,
 110         SLOT_USED = 2,
 111 };
 112
 113 void __weak *alloc_insn_page(void)
 114 {
 115         /*
 116          * Use module_alloc() so this page is within +/- 2GB of where the
 117          * kernel image and loaded module images reside. This is required
 118          * for most of the architectures.
 119          * (e.g. x86-64 needs this to handle the %rip-relative fixups.)
 120          */
 121         return module_alloc(PAGE_SIZE);
 122 }
 123
 124 static void free_insn_page(void *page)
 125 {
 126         module_memfree(page);
 127 }
 128
 129 struct kprobe_insn_cache kprobe_insn_slots = {
 130         .mutex = __MUTEX_INITIALIZER(kprobe_insn_slots.mutex),
 131         .alloc = alloc_insn_page,
 132         .free = free_insn_page,
 133         .sym = KPROBE_INSN_PAGE_SYM,
 134         .pages = LIST_HEAD_INIT(kprobe_insn_slots.pages),
 135         .insn_size = MAX_INSN_SIZE,
 136         .nr_garbage = 0,
 137 };
 138 static int collect_garbage_slots(struct kprobe_insn_cache *c);
 139
 140 /**
 141  * __get_insn_slot() - Find a slot on an executable page for an instruction.
 142  * We allocate an executable page if there's no room on existing ones.
 143  */
 144 kprobe_opcode_t *__get_insn_slot(struct kprobe_insn_cache *c)
 145 {
 146         struct kprobe_insn_page *kip;
 147         kprobe_opcode_t *slot = NULL;
 148
 149         /* Since the slot array is not protected by rcu, we need a mutex */
 150         mutex_lock(&c->mutex);
 151  retry:
 152         rcu_read_lock();
 153         list_for_each_entry_rcu(kip, &c->pages, list) {
 154                 if (kip->nused < slots_per_page(c)) {
 155                         int i;
 156
 157                         for (i = 0; i < slots_per_page(c); i++) {
 158                                 if (kip->slot_used[i] == SLOT_CLEAN) {
 159                                         kip->slot_used[i] = SLOT_USED;
 160                                         kip->nused++;
 161                                         slot = kip->insns + (i * c->insn_size);
 162                                         rcu_read_unlock();
 163                                         goto out;
 164                                 }
 165                         }
 166                         /* kip->nused is broken. Fix it. */
 167                         kip->nused = slots_per_page(c);
 168                         WARN_ON(1);
 169                 }
 170         }
 171         rcu_read_unlock();
 172
 173         /* If there are any garbage slots, collect it and try again. */
 174         if (c->nr_garbage && collect_garbage_slots(c) == 0)
 175                 goto retry;
 176
 177         /* All out of space.  Need to allocate a new page. */
 178         kip = kmalloc(KPROBE_INSN_PAGE_SIZE(slots_per_page(c)), GFP_KERNEL);
 179         if (!kip)
 180                 goto out;
 181
 182         kip->insns = c->alloc();
 183         if (!kip->insns) {
 184                 kfree(kip);
 185                 goto out;
 186         }
 187         INIT_LIST_HEAD(&kip->list);
 188         memset(kip->slot_used, SLOT_CLEAN, slots_per_page(c));
 189         kip->slot_used[0] = SLOT_USED;
 190         kip->nused = 1;
 191         kip->ngarbage = 0;
 192         kip->cache = c;
 193         list_add_rcu(&kip->list, &c->pages);
 194         slot = kip->insns;
 195
 196         /* Record the perf ksymbol register event after adding the page */
 197         perf_event_ksymbol(PERF_RECORD_KSYMBOL_TYPE_OOL, (unsigned long)kip->insns,
 198                            PAGE_SIZE, false, c->sym);
 199 out:
 200         mutex_unlock(&c->mutex);
 201         return slot;
 202 }
 203
 204 /* Return true if all garbages are collected, otherwise false. */
 205 static bool collect_one_slot(struct kprobe_insn_page *kip, int idx)
 206 {
 207         kip->slot_used[idx] = SLOT_CLEAN;
 208         kip->nused--;
 209         if (kip->nused == 0) {
 210                 /*
 211                  * Page is no longer in use.  Free it unless
 212                  * it's the last one.  We keep the last one
 213                  * so as not to have to set it up again the
 214                  * next time somebody inserts a probe.
 215                  */
 216                 if (!list_is_singular(&kip->list)) {
 217                         /*
 218                          * Record perf ksymbol unregister event before removing
 219                          * the page.
 220                          */
 221                         perf_event_ksymbol(PERF_RECORD_KSYMBOL_TYPE_OOL,
 222                                            (unsigned long)kip->insns, PAGE_SIZE, true,
 223                                            kip->cache->sym);
 224                         list_del_rcu(&kip->list);
 225                         synchronize_rcu();
 226                         kip->cache->free(kip->insns);
 227                         kfree(kip);
 228                 }
 229                 return true;
 230         }
 231         return false;
 232 }
 233
 234 static int collect_garbage_slots(struct kprobe_insn_cache *c)
 235 {
 236         struct kprobe_insn_page *kip, *next;
 237
 238         /* Ensure no-one is interrupted on the garbages */
 239         synchronize_rcu();
 240
 241         list_for_each_entry_safe(kip, next, &c->pages, list) {
 242                 int i;
 243
 244                 if (kip->ngarbage == 0)
 245                         continue;
 246                 kip->ngarbage = 0;      /* we will collect all garbages */
 247                 for (i = 0; i < slots_per_page(c); i++) {
 248                         if (kip->slot_used[i] == SLOT_DIRTY && collect_one_slot(kip, i))
 249                                 break;
 250                 }
 251         }
 252         c->nr_garbage = 0;
 253         return 0;
 254 }
 255
 256 void __free_insn_slot(struct kprobe_insn_cache *c,
 257                       kprobe_opcode_t *slot, int dirty)
 258 {
 259         struct kprobe_insn_page *kip;
 260         long idx;
 261
 262         mutex_lock(&c->mutex);
 263         rcu_read_lock();
 264         list_for_each_entry_rcu(kip, &c->pages, list) {
 265                 idx = ((long)slot - (long)kip->insns) /
 266                         (c->insn_size * sizeof(kprobe_opcode_t));
 267                 if (idx >= 0 && idx < slots_per_page(c))
 268                         goto out;
 269         }
 270         /* Could not find this slot. */
 271         WARN_ON(1);
 272         kip = NULL;
 273 out:
 274         rcu_read_unlock();
 275         /* Mark and sweep: this may sleep */
 276         if (kip) {
 277                 /* Check double free */
 278                 WARN_ON(kip->slot_used[idx] != SLOT_USED);
 279                 if (dirty) {
 280                         kip->slot_used[idx] = SLOT_DIRTY;
 281                         kip->ngarbage++;
 282                         if (++c->nr_garbage > slots_per_page(c))
 283                                 collect_garbage_slots(c);
 284                 } else {
 285                         collect_one_slot(kip, idx);
 286                 }
 287         }
 288         mutex_unlock(&c->mutex);
 289 }
 290
 291 /*
 292  * Check given address is on the page of kprobe instruction slots.
 293  * This will be used for checking whether the address on a stack
 294  * is on a text area or not.
 295  */
 296 bool __is_insn_slot_addr(struct kprobe_insn_cache *c, unsigned long addr)
 297 {
 298         struct kprobe_insn_page *kip;
 299         bool ret = false;
 300
 301         rcu_read_lock();
 302         list_for_each_entry_rcu(kip, &c->pages, list) {
 303                 if (addr >= (unsigned long)kip->insns &&
 304                     addr < (unsigned long)kip->insns + PAGE_SIZE) {
 305                         ret = true;
 306                         break;
 307                 }
 308         }
 309         rcu_read_unlock();
 310
 311         return ret;
 312 }
 313
 314 int kprobe_cache_get_kallsym(struct kprobe_insn_cache *c, unsigned int *symnum,
 315                              unsigned long *value, char *type, char *sym)
 316 {
 317         struct kprobe_insn_page *kip;
 318         int ret = -ERANGE;
 319
 320         rcu_read_lock();
 321         list_for_each_entry_rcu(kip, &c->pages, list) {
 322                 if ((*symnum)--)
 323                         continue;
 324                 strscpy(sym, c->sym, KSYM_NAME_LEN);
 325                 *type = 't';
 326                 *value = (unsigned long)kip->insns;
 327                 ret = 0;
 328                 break;
 329         }
 330         rcu_read_unlock();
 331
 332         return ret;
 333 }
 334
 335 #ifdef CONFIG_OPTPROBES
 336 void __weak *alloc_optinsn_page(void)
 337 {
 338         return alloc_insn_page();
 339 }
 340
 341 void __weak free_optinsn_page(void *page)
 342 {
 343         free_insn_page(page);
 344 }
 345
 346 /* For optimized_kprobe buffer */
 347 struct kprobe_insn_cache kprobe_optinsn_slots = {
 348         .mutex = __MUTEX_INITIALIZER(kprobe_optinsn_slots.mutex),
 349         .alloc = alloc_optinsn_page,
 350         .free = free_optinsn_page,
 351         .sym = KPROBE_OPTINSN_PAGE_SYM,
 352         .pages = LIST_HEAD_INIT(kprobe_optinsn_slots.pages),
 353         /* .insn_size is initialized later */
 354         .nr_garbage = 0,
 355 };
 356 #endif
 357 #endif
 358
 359 /* We have preemption disabled.. so it is safe to use __ versions */
 360 static inline void set_kprobe_instance(struct kprobe *kp)
 361 {
 362         __this_cpu_write(kprobe_instance, kp);
 363 }
 364
 365 static inline void reset_kprobe_instance(void)
 366 {
 367         __this_cpu_write(kprobe_instance, NULL);
 368 }
 369
 370 /*
 371  * This routine is called either:
 372  *      - under the 'kprobe_mutex' - during kprobe_[un]register().
 373  *                              OR
 374  *      - with preemption disabled - from architecture specific code.
 375  */
 376 struct kprobe *get_kprobe(void *addr)
 377 {
 378         struct hlist_head *head;
 379         struct kprobe *p;
 380
 381         head = &kprobe_table[hash_ptr(addr, KPROBE_HASH_BITS)];
 382         hlist_for_each_entry_rcu(p, head, hlist,
 383                                  lockdep_is_held(&kprobe_mutex)) {
 384                 if (p->addr == addr)
 385                         return p;
 386         }
 387
 388         return NULL;
 389 }
 390 NOKPROBE_SYMBOL(get_kprobe);
 391
 392 static int aggr_pre_handler(struct kprobe *p, struct pt_regs *regs);
 393
 394 /* Return true if 'p' is an aggregator */
 395 static inline bool kprobe_aggrprobe(struct kprobe *p)
 396 {
 397         return p->pre_handler == aggr_pre_handler;
 398 }
 399
 400 /* Return true if 'p' is unused */
 401 static inline bool kprobe_unused(struct kprobe *p)
 402 {
 403         return kprobe_aggrprobe(p) && kprobe_disabled(p) &&
 404                list_empty(&p->list);
 405 }
 406
 407 /* Keep all fields in the kprobe consistent. */
 408 static inline void copy_kprobe(struct kprobe *ap, struct kprobe *p)
 409 {
 410         memcpy(&p->opcode, &ap->opcode, sizeof(kprobe_opcode_t));
 411         memcpy(&p->ainsn, &ap->ainsn, sizeof(struct arch_specific_insn));
 412 }
 413
 414 #ifdef CONFIG_OPTPROBES
 415 /* NOTE: This is protected by 'kprobe_mutex'. */
 416 static bool kprobes_allow_optimization;
 417
 418 /*
 419  * Call all 'kprobe::pre_handler' on the list, but ignores its return value.
 420  * This must be called from arch-dep optimized caller.
 421  */
 422 void opt_pre_handler(struct kprobe *p, struct pt_regs *regs)
 423 {
 424         struct kprobe *kp;
 425
 426         list_for_each_entry_rcu(kp, &p->list, list) {
 427                 if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
 428                         set_kprobe_instance(kp);
 429                         kp->pre_handler(kp, regs);
 430                 }
 431                 reset_kprobe_instance();
 432         }
 433 }
 434 NOKPROBE_SYMBOL(opt_pre_handler);
 435
 436 /* Free optimized instructions and optimized_kprobe */
 437 static void free_aggr_kprobe(struct kprobe *p)
 438 {
 439         struct optimized_kprobe *op;
 440
 441         op = container_of(p, struct optimized_kprobe, kp);
 442         arch_remove_optimized_kprobe(op);
 443         arch_remove_kprobe(p);
 444         kfree(op);
 445 }
 446
 447 /* Return true if the kprobe is ready for optimization. */
 448 static inline int kprobe_optready(struct kprobe *p)
 449 {
 450         struct optimized_kprobe *op;
 451
 452         if (kprobe_aggrprobe(p)) {
 453                 op = container_of(p, struct optimized_kprobe, kp);
 454                 return arch_prepared_optinsn(&op->optinsn);
 455         }
 456
 457         return 0;
 458 }
 459
 460 /* Return true if the kprobe is disarmed. Note: p must be on hash list */
 461 static inline bool kprobe_disarmed(struct kprobe *p)
 462 {
 463         struct optimized_kprobe *op;
 464
 465         /* If kprobe is not aggr/opt probe, just return kprobe is disabled */
 466         if (!kprobe_aggrprobe(p))
 467                 return kprobe_disabled(p);
 468
 469         op = container_of(p, struct optimized_kprobe, kp);
 470
 471         return kprobe_disabled(p) && list_empty(&op->list);
 472 }
 473
 474 /* Return true if the probe is queued on (un)optimizing lists */
 475 static bool kprobe_queued(struct kprobe *p)
 476 {
 477         struct optimized_kprobe *op;
 478
 479         if (kprobe_aggrprobe(p)) {
 480                 op = container_of(p, struct optimized_kprobe, kp);
 481                 if (!list_empty(&op->list))
 482                         return true;
 483         }
 484         return false;
 485 }
 486
 487 /*
 488  * Return an optimized kprobe whose optimizing code replaces
 489  * instructions including 'addr' (exclude breakpoint).
 490  */
 491 static struct kprobe *get_optimized_kprobe(kprobe_opcode_t *addr)
 492 {
 493         int i;
 494         struct kprobe *p = NULL;
 495         struct optimized_kprobe *op;
 496
 497         /* Don't check i == 0, since that is a breakpoint case. */
 498         for (i = 1; !p && i < MAX_OPTIMIZED_LENGTH / sizeof(kprobe_opcode_t); i++)
 499                 p = get_kprobe(addr - i);
 500
 501         if (p && kprobe_optready(p)) {
 502                 op = container_of(p, struct optimized_kprobe, kp);
 503                 if (arch_within_optimized_kprobe(op, addr))
 504                         return p;
 505         }
 506
 507         return NULL;
 508 }
 509
 510 /* Optimization staging list, protected by 'kprobe_mutex' */
 511 static LIST_HEAD(optimizing_list);
 512 static LIST_HEAD(unoptimizing_list);
 513 static LIST_HEAD(freeing_list);
 514
 515 static void kprobe_optimizer(struct work_struct *work);
 516 static DECLARE_DELAYED_WORK(optimizing_work, kprobe_optimizer);
 517 #define OPTIMIZE_DELAY 5
 518
 519 /*
 520  * Optimize (replace a breakpoint with a jump) kprobes listed on
 521  * 'optimizing_list'.
 522  */
 523 static void do_optimize_kprobes(void)
 524 {
 525         lockdep_assert_held(&text_mutex);
 526         /*
 527          * The optimization/unoptimization refers 'online_cpus' via
 528          * stop_machine() and cpu-hotplug modifies the 'online_cpus'.
 529          * And same time, 'text_mutex' will be held in cpu-hotplug and here.
 530          * This combination can cause a deadlock (cpu-hotplug tries to lock
 531          * 'text_mutex' but stop_machine() can not be done because
 532          * the 'online_cpus' has been changed)
 533          * To avoid this deadlock, caller must have locked cpu-hotplug
 534          * for preventing cpu-hotplug outside of 'text_mutex' locking.
 535          */
 536         lockdep_assert_cpus_held();
 537
 538         /* Optimization never be done when disarmed */
 539         if (kprobes_all_disarmed || !kprobes_allow_optimization ||
 540             list_empty(&optimizing_list))
 541                 return;
 542
 543         arch_optimize_kprobes(&optimizing_list);
 544 }
 545
 546 /*
 547  * Unoptimize (replace a jump with a breakpoint and remove the breakpoint
 548  * if need) kprobes listed on 'unoptimizing_list'.
 549  */
 550 static void do_unoptimize_kprobes(void)
 551 {
 552         struct optimized_kprobe *op, *tmp;
 553
 554         lockdep_assert_held(&text_mutex);
 555         /* See comment in do_optimize_kprobes() */
 556         lockdep_assert_cpus_held();
 557
 558         /* Unoptimization must be done anytime */
 559         if (list_empty(&unoptimizing_list))
 560                 return;
 561
 562         arch_unoptimize_kprobes(&unoptimizing_list, &freeing_list);
 563         /* Loop on 'freeing_list' for disarming */
 564         list_for_each_entry_safe(op, tmp, &freeing_list, list) {
 565                 /* Switching from detour code to origin */
 566                 op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
 567                 /* Disarm probes if marked disabled */
 568                 if (kprobe_disabled(&op->kp))
 569                         arch_disarm_kprobe(&op->kp);
 570                 if (kprobe_unused(&op->kp)) {
 571                         /*
 572                          * Remove unused probes from hash list. After waiting
 573                          * for synchronization, these probes are reclaimed.
 574                          * (reclaiming is done by do_free_cleaned_kprobes().)
 575                          */
 576                         hlist_del_rcu(&op->kp.hlist);
 577                 } else
 578                         list_del_init(&op->list);
 579         }
 580 }
 581
 582 /* Reclaim all kprobes on the 'freeing_list' */
 583 static void do_free_cleaned_kprobes(void)
 584 {
 585         struct optimized_kprobe *op, *tmp;
 586
 587         list_for_each_entry_safe(op, tmp, &freeing_list, list) {
 588                 list_del_init(&op->list);
 589                 if (WARN_ON_ONCE(!kprobe_unused(&op->kp))) {
 590                         /*
 591                          * This must not happen, but if there is a kprobe
 592                          * still in use, keep it on kprobes hash list.
 593                          */
 594                         continue;
 595                 }
 596                 free_aggr_kprobe(&op->kp);
 597         }
 598 }
 599
 600 /* Start optimizer after OPTIMIZE_DELAY passed */
 601 static void kick_kprobe_optimizer(void)
 602 {
 603         schedule_delayed_work(&optimizing_work, OPTIMIZE_DELAY);
 604 }
 605
 606 /* Kprobe jump optimizer */
 607 static void kprobe_optimizer(struct work_struct *work)
 608 {
 609         mutex_lock(&kprobe_mutex);
 610         cpus_read_lock();
 611         mutex_lock(&text_mutex);
 612
 613         /*
 614          * Step 1: Unoptimize kprobes and collect cleaned (unused and disarmed)
 615          * kprobes before waiting for quiesence period.
 616          */
 617         do_unoptimize_kprobes();
 618
 619         /*
 620          * Step 2: Wait for quiesence period to ensure all potentially
 621          * preempted tasks to have normally scheduled. Because optprobe
 622          * may modify multiple instructions, there is a chance that Nth
 623          * instruction is preempted. In that case, such tasks can return
 624          * to 2nd-Nth byte of jump instruction. This wait is for avoiding it.
 625          * Note that on non-preemptive kernel, this is transparently converted
 626          * to synchronoze_sched() to wait for all interrupts to have completed.
 627          */
 628         synchronize_rcu_tasks();
 629
 630         /* Step 3: Optimize kprobes after quiesence period */
 631         do_optimize_kprobes();
 632
 633         /* Step 4: Free cleaned kprobes after quiesence period */
 634         do_free_cleaned_kprobes();
 635
 636         mutex_unlock(&text_mutex);
 637         cpus_read_unlock();
 638
 639         /* Step 5: Kick optimizer again if needed */
 640         if (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list))
 641                 kick_kprobe_optimizer();
 642
 643         mutex_unlock(&kprobe_mutex);
 644 }
 645
 646 /* Wait for completing optimization and unoptimization */
 647 void wait_for_kprobe_optimizer(void)
 648 {
 649         mutex_lock(&kprobe_mutex);
 650
 651         while (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list)) {
 652                 mutex_unlock(&kprobe_mutex);
 653
 654                 /* This will also make 'optimizing_work' execute immmediately */
 655                 flush_delayed_work(&optimizing_work);
 656                 /* 'optimizing_work' might not have been queued yet, relax */
 657                 cpu_relax();
 658
 659                 mutex_lock(&kprobe_mutex);
 660         }
 661
 662         mutex_unlock(&kprobe_mutex);
 663 }
 664
 665 static bool optprobe_queued_unopt(struct optimized_kprobe *op)
 666 {
 667         struct optimized_kprobe *_op;
 668
 669         list_for_each_entry(_op, &unoptimizing_list, list) {
 670                 if (op == _op)
 671                         return true;
 672         }
 673
 674         return false;
 675 }
 676
 677 /* Optimize kprobe if p is ready to be optimized */
 678 static void optimize_kprobe(struct kprobe *p)
 679 {
 680         struct optimized_kprobe *op;
 681
 682         /* Check if the kprobe is disabled or not ready for optimization. */
 683         if (!kprobe_optready(p) || !kprobes_allow_optimization ||
 684             (kprobe_disabled(p) || kprobes_all_disarmed))
 685                 return;
 686
 687         /* kprobes with 'post_handler' can not be optimized */
 688         if (p->post_handler)
 689                 return;
 690
 691         op = container_of(p, struct optimized_kprobe, kp);
 692
 693         /* Check there is no other kprobes at the optimized instructions */
 694         if (arch_check_optimized_kprobe(op) < 0)
 695                 return;
 696
 697         /* Check if it is already optimized. */
 698         if (op->kp.flags & KPROBE_FLAG_OPTIMIZED) {
 699                 if (optprobe_queued_unopt(op)) {
 700                         /* This is under unoptimizing. Just dequeue the probe */
 701                         list_del_init(&op->list);
 702                 }
 703                 return;
 704         }
 705         op->kp.flags |= KPROBE_FLAG_OPTIMIZED;
 706
 707         /*
 708          * On the 'unoptimizing_list' and 'optimizing_list',
 709          * 'op' must have OPTIMIZED flag
 710          */
 711         if (WARN_ON_ONCE(!list_empty(&op->list)))
 712                 return;
 713
 714         list_add(&op->list, &optimizing_list);
 715         kick_kprobe_optimizer();
 716 }
 717
 718 /* Short cut to direct unoptimizing */
 719 static void force_unoptimize_kprobe(struct optimized_kprobe *op)
 720 {
 721         lockdep_assert_cpus_held();
 722         arch_unoptimize_kprobe(op);
 723         op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
 724 }
 725
 726 /* Unoptimize a kprobe if p is optimized */
 727 static void unoptimize_kprobe(struct kprobe *p, bool force)
 728 {
 729         struct optimized_kprobe *op;
 730
 731         if (!kprobe_aggrprobe(p) || kprobe_disarmed(p))
 732                 return; /* This is not an optprobe nor optimized */
 733
 734         op = container_of(p, struct optimized_kprobe, kp);
 735         if (!kprobe_optimized(p))
 736                 return;
 737
 738         if (!list_empty(&op->list)) {
 739                 if (optprobe_queued_unopt(op)) {
 740                         /* Queued in unoptimizing queue */
 741                         if (force) {
 742                                 /*
 743                                  * Forcibly unoptimize the kprobe here, and queue it
 744                                  * in the freeing list for release afterwards.
 745                                  */
 746                                 force_unoptimize_kprobe(op);
 747                                 list_move(&op->list, &freeing_list);
 748                         }
 749                 } else {
 750                         /* Dequeue from the optimizing queue */
 751                         list_del_init(&op->list);
 752                         op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
 753                 }
 754                 return;
 755         }
 756
 757         /* Optimized kprobe case */
 758         if (force) {
 759                 /* Forcibly update the code: this is a special case */
 760                 force_unoptimize_kprobe(op);
 761         } else {
 762                 list_add(&op->list, &unoptimizing_list);
 763                 kick_kprobe_optimizer();
 764         }
 765 }
 766
 767 /* Cancel unoptimizing for reusing */
 768 static int reuse_unused_kprobe(struct kprobe *ap)
 769 {
 770         struct optimized_kprobe *op;
 771
 772         /*
 773          * Unused kprobe MUST be on the way of delayed unoptimizing (means
 774          * there is still a relative jump) and disabled.
 775          */
 776         op = container_of(ap, struct optimized_kprobe, kp);
 777         WARN_ON_ONCE(list_empty(&op->list));
 778         /* Enable the probe again */
 779         ap->flags &= ~KPROBE_FLAG_DISABLED;
 780         /* Optimize it again. (remove from 'op->list') */
 781         if (!kprobe_optready(ap))
 782                 return -EINVAL;
 783
 784         optimize_kprobe(ap);
 785         return 0;
 786 }
 787
 788 /* Remove optimized instructions */
 789 static void kill_optimized_kprobe(struct kprobe *p)
 790 {
 791         struct optimized_kprobe *op;
 792
 793         op = container_of(p, struct optimized_kprobe, kp);
 794         if (!list_empty(&op->list))
 795                 /* Dequeue from the (un)optimization queue */
 796                 list_del_init(&op->list);
 797         op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
 798
 799         if (kprobe_unused(p)) {
 800                 /* Enqueue if it is unused */
 801                 list_add(&op->list, &freeing_list);
 802                 /*
 803                  * Remove unused probes from the hash list. After waiting
 804                  * for synchronization, this probe is reclaimed.
 805                  * (reclaiming is done by do_free_cleaned_kprobes().)
 806                  */
 807                 hlist_del_rcu(&op->kp.hlist);
 808         }
 809
 810         /* Don't touch the code, because it is already freed. */
 811         arch_remove_optimized_kprobe(op);
 812 }
 813
 814 static inline
 815 void __prepare_optimized_kprobe(struct optimized_kprobe *op, struct kprobe *p)
 816 {
 817         if (!kprobe_ftrace(p))
 818                 arch_prepare_optimized_kprobe(op, p);
 819 }
 820
 821 /* Try to prepare optimized instructions */
 822 static void prepare_optimized_kprobe(struct kprobe *p)
 823 {
 824         struct optimized_kprobe *op;
 825
 826         op = container_of(p, struct optimized_kprobe, kp);
 827         __prepare_optimized_kprobe(op, p);
 828 }
 829
 830 /* Allocate new optimized_kprobe and try to prepare optimized instructions. */
 831 static struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
 832 {
 833         struct optimized_kprobe *op;
 834
 835         op = kzalloc(sizeof(struct optimized_kprobe), GFP_KERNEL);
 836         if (!op)
 837                 return NULL;
 838
 839         INIT_LIST_HEAD(&op->list);
 840         op->kp.addr = p->addr;
 841         __prepare_optimized_kprobe(op, p);
 842
 843         return &op->kp;
 844 }
 845
 846 static void init_aggr_kprobe(struct kprobe *ap, struct kprobe *p);
 847
 848 /*
 849  * Prepare an optimized_kprobe and optimize it.
 850  * NOTE: 'p' must be a normal registered kprobe.
 851  */
 852 static void try_to_optimize_kprobe(struct kprobe *p)
 853 {
 854         struct kprobe *ap;
 855         struct optimized_kprobe *op;
 856
 857         /* Impossible to optimize ftrace-based kprobe. */
 858         if (kprobe_ftrace(p))
 859                 return;
 860
 861         /* For preparing optimization, jump_label_text_reserved() is called. */
 862         cpus_read_lock();
 863         jump_label_lock();
 864         mutex_lock(&text_mutex);
 865
 866         ap = alloc_aggr_kprobe(p);
 867         if (!ap)
 868                 goto out;
 869
 870         op = container_of(ap, struct optimized_kprobe, kp);
 871         if (!arch_prepared_optinsn(&op->optinsn)) {
 872                 /* If failed to setup optimizing, fallback to kprobe. */
 873                 arch_remove_optimized_kprobe(op);
 874                 kfree(op);
 875                 goto out;
 876         }
 877
 878         init_aggr_kprobe(ap, p);
 879         optimize_kprobe(ap);    /* This just kicks optimizer thread. */
 880
 881 out:
 882         mutex_unlock(&text_mutex);
 883         jump_label_unlock();
 884         cpus_read_unlock();
 885 }
 886
 887 static void optimize_all_kprobes(void)
 888 {
 889         struct hlist_head *head;
 890         struct kprobe *p;
 891         unsigned int i;
 892
 893         mutex_lock(&kprobe_mutex);
 894         /* If optimization is already allowed, just return. */
 895         if (kprobes_allow_optimization)
 896                 goto out;
 897
 898         cpus_read_lock();
 899         kprobes_allow_optimization = true;
 900         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
 901                 head = &kprobe_table[i];
 902                 hlist_for_each_entry(p, head, hlist)
 903                         if (!kprobe_disabled(p))
 904                                 optimize_kprobe(p);
 905         }
 906         cpus_read_unlock();
 907         pr_info("kprobe jump-optimization is enabled. All kprobes are optimized if possible.\n");
 908 out:
 909         mutex_unlock(&kprobe_mutex);
 910 }
 911
 912 #ifdef CONFIG_SYSCTL
 913 static void unoptimize_all_kprobes(void)
 914 {
 915         struct hlist_head *head;
 916         struct kprobe *p;
 917         unsigned int i;
 918
 919         mutex_lock(&kprobe_mutex);
 920         /* If optimization is already prohibited, just return. */
 921         if (!kprobes_allow_optimization) {
 922                 mutex_unlock(&kprobe_mutex);
 923                 return;
 924         }
 925
 926         cpus_read_lock();
 927         kprobes_allow_optimization = false;
 928         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
 929                 head = &kprobe_table[i];
 930                 hlist_for_each_entry(p, head, hlist) {
 931                         if (!kprobe_disabled(p))
 932                                 unoptimize_kprobe(p, false);
 933                 }
 934         }
 935         cpus_read_unlock();
 936         mutex_unlock(&kprobe_mutex);
 937
 938         /* Wait for unoptimizing completion. */
 939         wait_for_kprobe_optimizer();
 940         pr_info("kprobe jump-optimization is disabled. All kprobes are based on software breakpoint.\n");
 941 }
 942
 943 static DEFINE_MUTEX(kprobe_sysctl_mutex);
 944 static int sysctl_kprobes_optimization;
 945 static int proc_kprobes_optimization_handler(struct ctl_table *table,
 946                                              int write, void *buffer,
 947                                              size_t *length, loff_t *ppos)
 948 {
 949         int ret;
 950
 951         mutex_lock(&kprobe_sysctl_mutex);
 952         sysctl_kprobes_optimization = kprobes_allow_optimization ? 1 : 0;
 953         ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
 954
 955         if (sysctl_kprobes_optimization)
 956                 optimize_all_kprobes();
 957         else
 958                 unoptimize_all_kprobes();
 959         mutex_unlock(&kprobe_sysctl_mutex);
 960
 961         return ret;
 962 }
 963
 964 static struct ctl_table kprobe_sysctls[] = {
 965         {
 966                 .procname       = "kprobes-optimization",
 967                 .data           = &sysctl_kprobes_optimization,
 968                 .maxlen         = sizeof(int),
 969                 .mode           = 0644,
 970                 .proc_handler   = proc_kprobes_optimization_handler,
 971                 .extra1         = SYSCTL_ZERO,
 972                 .extra2         = SYSCTL_ONE,
 973         },
 974         {}
 975 };
 976
 977 static void __init kprobe_sysctls_init(void)
 978 {
 979         register_sysctl_init("debug", kprobe_sysctls);
 980 }
 981 #endif /* CONFIG_SYSCTL */
 982
 983 /* Put a breakpoint for a probe. */
 984 static void __arm_kprobe(struct kprobe *p)
 985 {
 986         struct kprobe *_p;
 987
 988         lockdep_assert_held(&text_mutex);
 989
 990         /* Find the overlapping optimized kprobes. */
 991         _p = get_optimized_kprobe(p->addr);
 992         if (unlikely(_p))
 993                 /* Fallback to unoptimized kprobe */
 994                 unoptimize_kprobe(_p, true);
 995
 996         arch_arm_kprobe(p);
 997         optimize_kprobe(p);     /* Try to optimize (add kprobe to a list) */
 998 }
 999
1000 /* Remove the breakpoint of a probe. */
1001 static void __disarm_kprobe(struct kprobe *p, bool reopt)
1002 {
1003         struct kprobe *_p;
1004
1005         lockdep_assert_held(&text_mutex);
1006
1007         /* Try to unoptimize */
1008         unoptimize_kprobe(p, kprobes_all_disarmed);
1009
1010         if (!kprobe_queued(p)) {
1011                 arch_disarm_kprobe(p);
1012                 /* If another kprobe was blocked, re-optimize it. */
1013                 _p = get_optimized_kprobe(p->addr);
1014                 if (unlikely(_p) && reopt)
1015                         optimize_kprobe(_p);
1016         }
1017         /*
1018          * TODO: Since unoptimization and real disarming will be done by
1019          * the worker thread, we can not check whether another probe are
1020          * unoptimized because of this probe here. It should be re-optimized
1021          * by the worker thread.
1022          */
1023 }
1024
1025 #else /* !CONFIG_OPTPROBES */
1026
1027 #define optimize_kprobe(p)                      do {} while (0)
1028 #define unoptimize_kprobe(p, f)                 do {} while (0)
1029 #define kill_optimized_kprobe(p)                do {} while (0)
1030 #define prepare_optimized_kprobe(p)             do {} while (0)
1031 #define try_to_optimize_kprobe(p)               do {} while (0)
1032 #define __arm_kprobe(p)                         arch_arm_kprobe(p)
1033 #define __disarm_kprobe(p, o)                   arch_disarm_kprobe(p)
1034 #define kprobe_disarmed(p)                      kprobe_disabled(p)
1035 #define wait_for_kprobe_optimizer()             do {} while (0)
1036
1037 static int reuse_unused_kprobe(struct kprobe *ap)
1038 {
1039         /*
1040          * If the optimized kprobe is NOT supported, the aggr kprobe is
1041          * released at the same time that the last aggregated kprobe is
1042          * unregistered.
1043          * Thus there should be no chance to reuse unused kprobe.
1044          */
1045         WARN_ON_ONCE(1);
1046         return -EINVAL;
1047 }
1048
1049 static void free_aggr_kprobe(struct kprobe *p)
1050 {
1051         arch_remove_kprobe(p);
1052         kfree(p);
1053 }
1054
1055 static struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
1056 {
1057         return kzalloc(sizeof(struct kprobe), GFP_KERNEL);
1058 }
1059 #endif /* CONFIG_OPTPROBES */
1060
1061 #ifdef CONFIG_KPROBES_ON_FTRACE
1062 static struct ftrace_ops kprobe_ftrace_ops __read_mostly = {
1063         .func = kprobe_ftrace_handler,
1064         .flags = FTRACE_OPS_FL_SAVE_REGS,
1065 };
1066
1067 static struct ftrace_ops kprobe_ipmodify_ops __read_mostly = {
1068         .func = kprobe_ftrace_handler,
1069         .flags = FTRACE_OPS_FL_SAVE_REGS | FTRACE_OPS_FL_IPMODIFY,
1070 };
1071
1072 static int kprobe_ipmodify_enabled;
1073 static int kprobe_ftrace_enabled;
1074
1075 static int __arm_kprobe_ftrace(struct kprobe *p, struct ftrace_ops *ops,
1076                                int *cnt)
1077 {
1078         int ret = 0;
1079
1080         lockdep_assert_held(&kprobe_mutex);
1081
1082         ret = ftrace_set_filter_ip(ops, (unsigned long)p->addr, 0, 0);
1083         if (WARN_ONCE(ret < 0, "Failed to arm kprobe-ftrace at %pS (error %d)\n", p->addr, ret))
1084                 return ret;
1085
1086         if (*cnt == 0) {
1087                 ret = register_ftrace_function(ops);
1088                 if (WARN(ret < 0, "Failed to register kprobe-ftrace (error %d)\n", ret))
1089                         goto err_ftrace;
1090         }
1091
1092         (*cnt)++;
1093         return ret;
1094
1095 err_ftrace:
1096         /*
1097          * At this point, sinec ops is not registered, we should be sefe from
1098          * registering empty filter.
1099          */
1100         ftrace_set_filter_ip(ops, (unsigned long)p->addr, 1, 0);
1101         return ret;
1102 }
1103
1104 static int arm_kprobe_ftrace(struct kprobe *p)
1105 {
1106         bool ipmodify = (p->post_handler != NULL);
1107
1108         return __arm_kprobe_ftrace(p,
1109                 ipmodify ? &kprobe_ipmodify_ops : &kprobe_ftrace_ops,
1110                 ipmodify ? &kprobe_ipmodify_enabled : &kprobe_ftrace_enabled);
1111 }
1112
1113 static int __disarm_kprobe_ftrace(struct kprobe *p, struct ftrace_ops *ops,
1114                                   int *cnt)
1115 {
1116         int ret = 0;
1117
1118         lockdep_assert_held(&kprobe_mutex);
1119
1120         if (*cnt == 1) {
1121                 ret = unregister_ftrace_function(ops);
1122                 if (WARN(ret < 0, "Failed to unregister kprobe-ftrace (error %d)\n", ret))
1123                         return ret;
1124         }
1125
1126         (*cnt)--;
1127
1128         ret = ftrace_set_filter_ip(ops, (unsigned long)p->addr, 1, 0);
1129         WARN_ONCE(ret < 0, "Failed to disarm kprobe-ftrace at %pS (error %d)\n",
1130                   p->addr, ret);
1131         return ret;
1132 }
1133
1134 static int disarm_kprobe_ftrace(struct kprobe *p)
1135 {
1136         bool ipmodify = (p->post_handler != NULL);
1137
1138         return __disarm_kprobe_ftrace(p,
1139                 ipmodify ? &kprobe_ipmodify_ops : &kprobe_ftrace_ops,
1140                 ipmodify ? &kprobe_ipmodify_enabled : &kprobe_ftrace_enabled);
1141 }
1142 #else   /* !CONFIG_KPROBES_ON_FTRACE */
1143 static inline int arm_kprobe_ftrace(struct kprobe *p)
1144 {
1145         return -ENODEV;
1146 }
1147
1148 static inline int disarm_kprobe_ftrace(struct kprobe *p)
1149 {
1150         return -ENODEV;
1151 }
1152 #endif
1153
1154 static int prepare_kprobe(struct kprobe *p)
1155 {
1156         /* Must ensure p->addr is really on ftrace */
1157         if (kprobe_ftrace(p))
1158                 return arch_prepare_kprobe_ftrace(p);
1159
1160         return arch_prepare_kprobe(p);
1161 }
1162
1163 static int arm_kprobe(struct kprobe *kp)
1164 {
1165         if (unlikely(kprobe_ftrace(kp)))
1166                 return arm_kprobe_ftrace(kp);
1167
1168         cpus_read_lock();
1169         mutex_lock(&text_mutex);
1170         __arm_kprobe(kp);
1171         mutex_unlock(&text_mutex);
1172         cpus_read_unlock();
1173
1174         return 0;
1175 }
1176
1177 static int disarm_kprobe(struct kprobe *kp, bool reopt)
1178 {
1179         if (unlikely(kprobe_ftrace(kp)))
1180                 return disarm_kprobe_ftrace(kp);
1181
1182         cpus_read_lock();
1183         mutex_lock(&text_mutex);
1184         __disarm_kprobe(kp, reopt);
1185         mutex_unlock(&text_mutex);
1186         cpus_read_unlock();
1187
1188         return 0;
1189 }
1190
1191 /*
1192  * Aggregate handlers for multiple kprobes support - these handlers
1193  * take care of invoking the individual kprobe handlers on p->list
1194  */
1195 static int aggr_pre_handler(struct kprobe *p, struct pt_regs *regs)
1196 {
1197         struct kprobe *kp;
1198
1199         list_for_each_entry_rcu(kp, &p->list, list) {
1200                 if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
1201                         set_kprobe_instance(kp);
1202                         if (kp->pre_handler(kp, regs))
1203                                 return 1;
1204                 }
1205                 reset_kprobe_instance();
1206         }
1207         return 0;
1208 }
1209 NOKPROBE_SYMBOL(aggr_pre_handler);
1210
1211 static void aggr_post_handler(struct kprobe *p, struct pt_regs *regs,
1212                               unsigned long flags)
1213 {
1214         struct kprobe *kp;
1215
1216         list_for_each_entry_rcu(kp, &p->list, list) {
1217                 if (kp->post_handler && likely(!kprobe_disabled(kp))) {
1218                         set_kprobe_instance(kp);
1219                         kp->post_handler(kp, regs, flags);
1220                         reset_kprobe_instance();
1221                 }
1222         }
1223 }
1224 NOKPROBE_SYMBOL(aggr_post_handler);
1225
1226 /* Walks the list and increments 'nmissed' if 'p' has child probes. */
1227 void kprobes_inc_nmissed_count(struct kprobe *p)
1228 {
1229         struct kprobe *kp;
1230
1231         if (!kprobe_aggrprobe(p)) {
1232                 p->nmissed++;
1233         } else {
1234                 list_for_each_entry_rcu(kp, &p->list, list)
1235                         kp->nmissed++;
1236         }
1237 }
1238 NOKPROBE_SYMBOL(kprobes_inc_nmissed_count);
1239
1240 static struct kprobe kprobe_busy = {
1241         .addr = (void *) get_kprobe,
1242 };
1243
1244 void kprobe_busy_begin(void)
1245 {
1246         struct kprobe_ctlblk *kcb;
1247
1248         preempt_disable();
1249         __this_cpu_write(current_kprobe, &kprobe_busy);
1250         kcb = get_kprobe_ctlblk();
1251         kcb->kprobe_status = KPROBE_HIT_ACTIVE;
1252 }
1253
1254 void kprobe_busy_end(void)
1255 {
1256         __this_cpu_write(current_kprobe, NULL);
1257         preempt_enable();
1258 }
1259
1260 /* Add the new probe to 'ap->list'. */
1261 static int add_new_kprobe(struct kprobe *ap, struct kprobe *p)
1262 {
1263         if (p->post_handler)
1264                 unoptimize_kprobe(ap, true);    /* Fall back to normal kprobe */
1265
1266         list_add_rcu(&p->list, &ap->list);
1267         if (p->post_handler && !ap->post_handler)
1268                 ap->post_handler = aggr_post_handler;
1269
1270         return 0;
1271 }
1272
1273 /*
1274  * Fill in the required fields of the aggregator kprobe. Replace the
1275  * earlier kprobe in the hlist with the aggregator kprobe.
1276  */
1277 static void init_aggr_kprobe(struct kprobe *ap, struct kprobe *p)
1278 {
1279         /* Copy the insn slot of 'p' to 'ap'. */
1280         copy_kprobe(p, ap);
1281         flush_insn_slot(ap);
1282         ap->addr = p->addr;
1283         ap->flags = p->flags & ~KPROBE_FLAG_OPTIMIZED;
1284         ap->pre_handler = aggr_pre_handler;
1285         /* We don't care the kprobe which has gone. */
1286         if (p->post_handler && !kprobe_gone(p))
1287                 ap->post_handler = aggr_post_handler;
1288
1289         INIT_LIST_HEAD(&ap->list);
1290         INIT_HLIST_NODE(&ap->hlist);
1291
1292         list_add_rcu(&p->list, &ap->list);
1293         hlist_replace_rcu(&p->hlist, &ap->hlist);
1294 }
1295
1296 /*
1297  * This registers the second or subsequent kprobe at the same address.
1298  */
1299 static int register_aggr_kprobe(struct kprobe *orig_p, struct kprobe *p)
1300 {
1301         int ret = 0;
1302         struct kprobe *ap = orig_p;
1303
1304         cpus_read_lock();
1305
1306         /* For preparing optimization, jump_label_text_reserved() is called */
1307         jump_label_lock();
1308         mutex_lock(&text_mutex);
1309
1310         if (!kprobe_aggrprobe(orig_p)) {
1311                 /* If 'orig_p' is not an 'aggr_kprobe', create new one. */
1312                 ap = alloc_aggr_kprobe(orig_p);
1313                 if (!ap) {
1314                         ret = -ENOMEM;
1315                         goto out;
1316                 }
1317                 init_aggr_kprobe(ap, orig_p);
1318         } else if (kprobe_unused(ap)) {
1319                 /* This probe is going to die. Rescue it */
1320                 ret = reuse_unused_kprobe(ap);
1321                 if (ret)
1322                         goto out;
1323         }
1324
1325         if (kprobe_gone(ap)) {
1326                 /*
1327                  * Attempting to insert new probe at the same location that
1328                  * had a probe in the module vaddr area which already
1329                  * freed. So, the instruction slot has already been
1330                  * released. We need a new slot for the new probe.
1331                  */
1332                 ret = arch_prepare_kprobe(ap);
1333                 if (ret)
1334                         /*
1335                          * Even if fail to allocate new slot, don't need to
1336                          * free the 'ap'. It will be used next time, or
1337                          * freed by unregister_kprobe().
1338                          */
1339                         goto out;
1340
1341                 /* Prepare optimized instructions if possible. */
1342                 prepare_optimized_kprobe(ap);
1343
1344                 /*
1345                  * Clear gone flag to prevent allocating new slot again, and
1346                  * set disabled flag because it is not armed yet.
1347                  */
1348                 ap->flags = (ap->flags & ~KPROBE_FLAG_GONE)
1349                             | KPROBE_FLAG_DISABLED;
1350         }
1351
1352         /* Copy the insn slot of 'p' to 'ap'. */
1353         copy_kprobe(ap, p);
1354         ret = add_new_kprobe(ap, p);
1355
1356 out:
1357         mutex_unlock(&text_mutex);
1358         jump_label_unlock();
1359         cpus_read_unlock();
1360
1361         if (ret == 0 && kprobe_disabled(ap) && !kprobe_disabled(p)) {
1362                 ap->flags &= ~KPROBE_FLAG_DISABLED;
1363                 if (!kprobes_all_disarmed) {
1364                         /* Arm the breakpoint again. */
1365                         ret = arm_kprobe(ap);
1366                         if (ret) {
1367                                 ap->flags |= KPROBE_FLAG_DISABLED;
1368                                 list_del_rcu(&p->list);
1369                                 synchronize_rcu();
1370                         }
1371                 }
1372         }
1373         return ret;
1374 }
1375
1376 bool __weak arch_within_kprobe_blacklist(unsigned long addr)
1377 {
1378         /* The '__kprobes' functions and entry code must not be probed. */
1379         return addr >= (unsigned long)__kprobes_text_start &&
1380                addr < (unsigned long)__kprobes_text_end;
1381 }
1382
1383 static bool __within_kprobe_blacklist(unsigned long addr)
1384 {
1385         struct kprobe_blacklist_entry *ent;
1386
1387         if (arch_within_kprobe_blacklist(addr))
1388                 return true;
1389         /*
1390          * If 'kprobe_blacklist' is defined, check the address and
1391          * reject any probe registration in the prohibited area.
1392          */
1393         list_for_each_entry(ent, &kprobe_blacklist, list) {
1394                 if (addr >= ent->start_addr && addr < ent->end_addr)
1395                         return true;
1396         }
1397         return false;
1398 }
1399
1400 bool within_kprobe_blacklist(unsigned long addr)
1401 {
1402         char symname[KSYM_NAME_LEN], *p;
1403
1404         if (__within_kprobe_blacklist(addr))
1405                 return true;
1406
1407         /* Check if the address is on a suffixed-symbol */
1408         if (!lookup_symbol_name(addr, symname)) {
1409                 p = strchr(symname, '.');
1410                 if (!p)
1411                         return false;
1412                 *p = '\0';
1413                 addr = (unsigned long)kprobe_lookup_name(symname, 0);
1414                 if (addr)
1415                         return __within_kprobe_blacklist(addr);
1416         }
1417         return false;
1418 }
1419
1420 /*
1421  * arch_adjust_kprobe_addr - adjust the address
1422  * @addr: symbol base address
1423  * @offset: offset within the symbol
1424  * @on_func_entry: was this @addr+@offset on the function entry
1425  *
1426  * Typically returns @addr + @offset, except for special cases where the
1427  * function might be prefixed by a CFI landing pad, in that case any offset
1428  * inside the landing pad is mapped to the first 'real' instruction of the
1429  * symbol.
1430  *
1431  * Specifically, for things like IBT/BTI, skip the resp. ENDBR/BTI.C
1432  * instruction at +0.
1433  */
1434 kprobe_opcode_t *__weak arch_adjust_kprobe_addr(unsigned long addr,
1435                                                 unsigned long offset,
1436                                                 bool *on_func_entry)
1437 {
1438         *on_func_entry = !offset;
1439         return (kprobe_opcode_t *)(addr + offset);
1440 }
1441
1442 /*
1443  * If 'symbol_name' is specified, look it up and add the 'offset'
1444  * to it. This way, we can specify a relative address to a symbol.
1445  * This returns encoded errors if it fails to look up symbol or invalid
1446  * combination of parameters.
1447  */
1448 static kprobe_opcode_t *
1449 _kprobe_addr(kprobe_opcode_t *addr, const char *symbol_name,
1450              unsigned long offset, bool *on_func_entry)
1451 {
1452         if ((symbol_name && addr) || (!symbol_name && !addr))
1453                 goto invalid;
1454
1455         if (symbol_name) {
1456                 /*
1457                  * Input: @sym + @offset
1458                  * Output: @addr + @offset
1459                  *
1460                  * NOTE: kprobe_lookup_name() does *NOT* fold the offset
1461                  *       argument into it's output!
1462                  */
1463                 addr = kprobe_lookup_name(symbol_name, offset);
1464                 if (!addr)
1465                         return ERR_PTR(-ENOENT);
1466         }
1467
1468         /*
1469          * So here we have @addr + @offset, displace it into a new
1470          * @addr' + @offset' where @addr' is the symbol start address.
1471          */
1472         addr = (void *)addr + offset;
1473         if (!kallsyms_lookup_size_offset((unsigned long)addr, NULL, &offset))
1474                 return ERR_PTR(-ENOENT);
1475         addr = (void *)addr - offset;
1476
1477         /*
1478          * Then ask the architecture to re-combine them, taking care of
1479          * magical function entry details while telling us if this was indeed
1480          * at the start of the function.
1481          */
1482         addr = arch_adjust_kprobe_addr((unsigned long)addr, offset, on_func_entry);
1483         if (addr)
1484                 return addr;
1485
1486 invalid:
1487         return ERR_PTR(-EINVAL);
1488 }
1489
1490 static kprobe_opcode_t *kprobe_addr(struct kprobe *p)
1491 {
1492         bool on_func_entry;
1493         return _kprobe_addr(p->addr, p->symbol_name, p->offset, &on_func_entry);
1494 }
1495
1496 /*
1497  * Check the 'p' is valid and return the aggregator kprobe
1498  * at the same address.
1499  */
1500 static struct kprobe *__get_valid_kprobe(struct kprobe *p)
1501 {
1502         struct kprobe *ap, *list_p;
1503
1504         lockdep_assert_held(&kprobe_mutex);
1505
1506         ap = get_kprobe(p->addr);
1507         if (unlikely(!ap))
1508                 return NULL;
1509
1510         if (p != ap) {
1511                 list_for_each_entry(list_p, &ap->list, list)
1512                         if (list_p == p)
1513                         /* kprobe p is a valid probe */
1514                                 goto valid;
1515                 return NULL;
1516         }
1517 valid:
1518         return ap;
1519 }
1520
1521 /*
1522  * Warn and return error if the kprobe is being re-registered since
1523  * there must be a software bug.
1524  */
1525 static inline int warn_kprobe_rereg(struct kprobe *p)
1526 {
1527         int ret = 0;
1528
1529         mutex_lock(&kprobe_mutex);
1530         if (WARN_ON_ONCE(__get_valid_kprobe(p)))
1531                 ret = -EINVAL;
1532         mutex_unlock(&kprobe_mutex);
1533
1534         return ret;
1535 }
1536
1537 static int check_ftrace_location(struct kprobe *p)
1538 {
1539         unsigned long addr = (unsigned long)p->addr;
1540
1541         if (ftrace_location(addr) == addr) {
1542 #ifdef CONFIG_KPROBES_ON_FTRACE
1543                 p->flags |= KPROBE_FLAG_FTRACE;
1544 #else   /* !CONFIG_KPROBES_ON_FTRACE */
1545                 return -EINVAL;
1546 #endif
1547         }
1548         return 0;
1549 }
1550
1551 static int check_kprobe_address_safe(struct kprobe *p,
1552                                      struct module **probed_mod)
1553 {
1554         int ret;
1555
1556         ret = check_ftrace_location(p);
1557         if (ret)
1558                 return ret;
1559         jump_label_lock();
1560         preempt_disable();
1561
1562         /* Ensure it is not in reserved area nor out of text */
1563         if (!(core_kernel_text((unsigned long) p->addr) ||
1564             is_module_text_address((unsigned long) p->addr)) ||
1565             within_kprobe_blacklist((unsigned long) p->addr) ||
1566             jump_label_text_reserved(p->addr, p->addr) ||
1567             static_call_text_reserved(p->addr, p->addr) ||
1568             find_bug((unsigned long)p->addr)) {
1569                 ret = -EINVAL;
1570                 goto out;
1571         }
1572
1573         /* Check if 'p' is probing a module. */
1574         *probed_mod = __module_text_address((unsigned long) p->addr);
1575         if (*probed_mod) {
1576                 /*
1577                  * We must hold a refcount of the probed module while updating
1578                  * its code to prohibit unexpected unloading.
1579                  */
1580                 if (unlikely(!try_module_get(*probed_mod))) {
1581                         ret = -ENOENT;
1582                         goto out;
1583                 }
1584
1585                 /*
1586                  * If the module freed '.init.text', we couldn't insert
1587                  * kprobes in there.
1588                  */
1589                 if (within_module_init((unsigned long)p->addr, *probed_mod) &&
1590                     (*probed_mod)->state != MODULE_STATE_COMING) {
1591                         module_put(*probed_mod);
1592                         *probed_mod = NULL;
1593                         ret = -ENOENT;
1594                 }
1595         }
1596 out:
1597         preempt_enable();
1598         jump_label_unlock();
1599
1600         return ret;
1601 }
1602
1603 int register_kprobe(struct kprobe *p)
1604 {
1605         int ret;
1606         struct kprobe *old_p;
1607         struct module *probed_mod;
1608         kprobe_opcode_t *addr;
1609
1610         /* Adjust probe address from symbol */
1611         addr = kprobe_addr(p);
1612         if (IS_ERR(addr))
1613                 return PTR_ERR(addr);
1614         p->addr = addr;
1615
1616         ret = warn_kprobe_rereg(p);
1617         if (ret)
1618                 return ret;
1619
1620         /* User can pass only KPROBE_FLAG_DISABLED to register_kprobe */
1621         p->flags &= KPROBE_FLAG_DISABLED;
1622         p->nmissed = 0;
1623         INIT_LIST_HEAD(&p->list);
1624
1625         ret = check_kprobe_address_safe(p, &probed_mod);
1626         if (ret)
1627                 return ret;
1628
1629         mutex_lock(&kprobe_mutex);
1630
1631         old_p = get_kprobe(p->addr);
1632         if (old_p) {
1633                 /* Since this may unoptimize 'old_p', locking 'text_mutex'. */
1634                 ret = register_aggr_kprobe(old_p, p);
1635                 goto out;
1636         }
1637
1638         cpus_read_lock();
1639         /* Prevent text modification */
1640         mutex_lock(&text_mutex);
1641         ret = prepare_kprobe(p);
1642         mutex_unlock(&text_mutex);
1643         cpus_read_unlock();
1644         if (ret)
1645                 goto out;
1646
1647         INIT_HLIST_NODE(&p->hlist);
1648         hlist_add_head_rcu(&p->hlist,
1649                        &kprobe_table[hash_ptr(p->addr, KPROBE_HASH_BITS)]);
1650
1651         if (!kprobes_all_disarmed && !kprobe_disabled(p)) {
1652                 ret = arm_kprobe(p);
1653                 if (ret) {
1654                         hlist_del_rcu(&p->hlist);
1655                         synchronize_rcu();
1656                         goto out;
1657                 }
1658         }
1659
1660         /* Try to optimize kprobe */
1661         try_to_optimize_kprobe(p);
1662 out:
1663         mutex_unlock(&kprobe_mutex);
1664
1665         if (probed_mod)
1666                 module_put(probed_mod);
1667
1668         return ret;
1669 }
1670 EXPORT_SYMBOL_GPL(register_kprobe);
1671
1672 /* Check if all probes on the 'ap' are disabled. */
1673 static bool aggr_kprobe_disabled(struct kprobe *ap)
1674 {
1675         struct kprobe *kp;
1676
1677         lockdep_assert_held(&kprobe_mutex);
1678
1679         list_for_each_entry(kp, &ap->list, list)
1680                 if (!kprobe_disabled(kp))
1681                         /*
1682                          * Since there is an active probe on the list,
1683                          * we can't disable this 'ap'.
1684                          */
1685                         return false;
1686
1687         return true;
1688 }
1689
1690 static struct kprobe *__disable_kprobe(struct kprobe *p)
1691 {
1692         struct kprobe *orig_p;
1693         int ret;
1694
1695         lockdep_assert_held(&kprobe_mutex);
1696
1697         /* Get an original kprobe for return */
1698         orig_p = __get_valid_kprobe(p);
1699         if (unlikely(orig_p == NULL))
1700                 return ERR_PTR(-EINVAL);
1701
1702         if (!kprobe_disabled(p)) {
1703                 /* Disable probe if it is a child probe */
1704                 if (p != orig_p)
1705                         p->flags |= KPROBE_FLAG_DISABLED;
1706
1707                 /* Try to disarm and disable this/parent probe */
1708                 if (p == orig_p || aggr_kprobe_disabled(orig_p)) {
1709                         /*
1710                          * If 'kprobes_all_disarmed' is set, 'orig_p'
1711                          * should have already been disarmed, so
1712                          * skip unneed disarming process.
1713                          */
1714                         if (!kprobes_all_disarmed) {
1715                                 ret = disarm_kprobe(orig_p, true);
1716                                 if (ret) {
1717                                         p->flags &= ~KPROBE_FLAG_DISABLED;
1718                                         return ERR_PTR(ret);
1719                                 }
1720                         }
1721                         orig_p->flags |= KPROBE_FLAG_DISABLED;
1722                 }
1723         }
1724
1725         return orig_p;
1726 }
1727
1728 /*
1729  * Unregister a kprobe without a scheduler synchronization.
1730  */
1731 static int __unregister_kprobe_top(struct kprobe *p)
1732 {
1733         struct kprobe *ap, *list_p;
1734
1735         /* Disable kprobe. This will disarm it if needed. */
1736         ap = __disable_kprobe(p);
1737         if (IS_ERR(ap))
1738                 return PTR_ERR(ap);
1739
1740         if (ap == p)
1741                 /*
1742                  * This probe is an independent(and non-optimized) kprobe
1743                  * (not an aggrprobe). Remove from the hash list.
1744                  */
1745                 goto disarmed;
1746
1747         /* Following process expects this probe is an aggrprobe */
1748         WARN_ON(!kprobe_aggrprobe(ap));
1749
1750         if (list_is_singular(&ap->list) && kprobe_disarmed(ap))
1751                 /*
1752                  * !disarmed could be happen if the probe is under delayed
1753                  * unoptimizing.
1754                  */
1755                 goto disarmed;
1756         else {
1757                 /* If disabling probe has special handlers, update aggrprobe */
1758                 if (p->post_handler && !kprobe_gone(p)) {
1759                         list_for_each_entry(list_p, &ap->list, list) {
1760                                 if ((list_p != p) && (list_p->post_handler))
1761                                         goto noclean;
1762                         }
1763                         ap->post_handler = NULL;
1764                 }
1765 noclean:
1766                 /*
1767                  * Remove from the aggrprobe: this path will do nothing in
1768                  * __unregister_kprobe_bottom().
1769                  */
1770                 list_del_rcu(&p->list);
1771                 if (!kprobe_disabled(ap) && !kprobes_all_disarmed)
1772                         /*
1773                          * Try to optimize this probe again, because post
1774                          * handler may have been changed.
1775                          */
1776                         optimize_kprobe(ap);
1777         }
1778         return 0;
1779
1780 disarmed:
1781         hlist_del_rcu(&ap->hlist);
1782         return 0;
1783 }
1784
1785 static void __unregister_kprobe_bottom(struct kprobe *p)
1786 {
1787         struct kprobe *ap;
1788
1789         if (list_empty(&p->list))
1790                 /* This is an independent kprobe */
1791                 arch_remove_kprobe(p);
1792         else if (list_is_singular(&p->list)) {
1793                 /* This is the last child of an aggrprobe */
1794                 ap = list_entry(p->list.next, struct kprobe, list);
1795                 list_del(&p->list);
1796                 free_aggr_kprobe(ap);
1797         }
1798         /* Otherwise, do nothing. */
1799 }
1800
1801 int register_kprobes(struct kprobe **kps, int num)
1802 {
1803         int i, ret = 0;
1804
1805         if (num <= 0)
1806                 return -EINVAL;
1807         for (i = 0; i < num; i++) {
1808                 ret = register_kprobe(kps[i]);
1809                 if (ret < 0) {
1810                         if (i > 0)
1811                                 unregister_kprobes(kps, i);
1812                         break;
1813                 }
1814         }
1815         return ret;
1816 }
1817 EXPORT_SYMBOL_GPL(register_kprobes);
1818
1819 void unregister_kprobe(struct kprobe *p)
1820 {
1821         unregister_kprobes(&p, 1);
1822 }
1823 EXPORT_SYMBOL_GPL(unregister_kprobe);
1824
1825 void unregister_kprobes(struct kprobe **kps, int num)
1826 {
1827         int i;
1828
1829         if (num <= 0)
1830                 return;
1831         mutex_lock(&kprobe_mutex);
1832         for (i = 0; i < num; i++)
1833                 if (__unregister_kprobe_top(kps[i]) < 0)
1834                         kps[i]->addr = NULL;
1835         mutex_unlock(&kprobe_mutex);
1836
1837         synchronize_rcu();
1838         for (i = 0; i < num; i++)
1839                 if (kps[i]->addr)
1840                         __unregister_kprobe_bottom(kps[i]);
1841 }
1842 EXPORT_SYMBOL_GPL(unregister_kprobes);
1843
1844 int __weak kprobe_exceptions_notify(struct notifier_block *self,
1845                                         unsigned long val, void *data)
1846 {
1847         return NOTIFY_DONE;
1848 }
1849 NOKPROBE_SYMBOL(kprobe_exceptions_notify);
1850
1851 static struct notifier_block kprobe_exceptions_nb = {
1852         .notifier_call = kprobe_exceptions_notify,
1853         .priority = 0x7fffffff /* we need to be notified first */
1854 };
1855
1856 #ifdef CONFIG_KRETPROBES
1857
1858 #if !defined(CONFIG_KRETPROBE_ON_RETHOOK)
1859 static void free_rp_inst_rcu(struct rcu_head *head)
1860 {
1861         struct kretprobe_instance *ri = container_of(head, struct kretprobe_instance, rcu);
1862
1863         if (refcount_dec_and_test(&ri->rph->ref))
1864                 kfree(ri->rph);
1865         kfree(ri);
1866 }
1867 NOKPROBE_SYMBOL(free_rp_inst_rcu);
1868
1869 static void recycle_rp_inst(struct kretprobe_instance *ri)
1870 {
1871         struct kretprobe *rp = get_kretprobe(ri);
1872
1873         if (likely(rp))
1874                 freelist_add(&ri->freelist, &rp->freelist);
1875         else
1876                 call_rcu(&ri->rcu, free_rp_inst_rcu);
1877 }
1878 NOKPROBE_SYMBOL(recycle_rp_inst);
1879
1880 /*
1881  * This function is called from delayed_put_task_struct() when a task is
1882  * dead and cleaned up to recycle any kretprobe instances associated with
1883  * this task. These left over instances represent probed functions that
1884  * have been called but will never return.
1885  */
1886 void kprobe_flush_task(struct task_struct *tk)
1887 {
1888         struct kretprobe_instance *ri;
1889         struct llist_node *node;
1890
1891         /* Early boot, not yet initialized. */
1892         if (unlikely(!kprobes_initialized))
1893                 return;
1894
1895         kprobe_busy_begin();
1896
1897         node = __llist_del_all(&tk->kretprobe_instances);
1898         while (node) {
1899                 ri = container_of(node, struct kretprobe_instance, llist);
1900                 node = node->next;
1901
1902                 recycle_rp_inst(ri);
1903         }
1904
1905         kprobe_busy_end();
1906 }
1907 NOKPROBE_SYMBOL(kprobe_flush_task);
1908
1909 static inline void free_rp_inst(struct kretprobe *rp)
1910 {
1911         struct kretprobe_instance *ri;
1912         struct freelist_node *node;
1913         int count = 0;
1914
1915         node = rp->freelist.head;
1916         while (node) {
1917                 ri = container_of(node, struct kretprobe_instance, freelist);
1918                 node = node->next;
1919
1920                 kfree(ri);
1921                 count++;
1922         }
1923
1924         if (refcount_sub_and_test(count, &rp->rph->ref)) {
1925                 kfree(rp->rph);
1926                 rp->rph = NULL;
1927         }
1928 }
1929
1930 /* This assumes the 'tsk' is the current task or the is not running. */
1931 static kprobe_opcode_t *__kretprobe_find_ret_addr(struct task_struct *tsk,
1932                                                   struct llist_node **cur)
1933 {
1934         struct kretprobe_instance *ri = NULL;
1935         struct llist_node *node = *cur;
1936
1937         if (!node)
1938                 node = tsk->kretprobe_instances.first;
1939         else
1940                 node = node->next;
1941
1942         while (node) {
1943                 ri = container_of(node, struct kretprobe_instance, llist);
1944                 if (ri->ret_addr != kretprobe_trampoline_addr()) {
1945                         *cur = node;
1946                         return ri->ret_addr;
1947                 }
1948                 node = node->next;
1949         }
1950         return NULL;
1951 }
1952 NOKPROBE_SYMBOL(__kretprobe_find_ret_addr);
1953
1954 /**
1955  * kretprobe_find_ret_addr -- Find correct return address modified by kretprobe
1956  * @tsk: Target task
1957  * @fp: A frame pointer
1958  * @cur: a storage of the loop cursor llist_node pointer for next call
1959  *
1960  * Find the correct return address modified by a kretprobe on @tsk in unsigned
1961  * long type. If it finds the return address, this returns that address value,
1962  * or this returns 0.
1963  * The @tsk must be 'current' or a task which is not running. @fp is a hint
1964  * to get the currect return address - which is compared with the
1965  * kretprobe_instance::fp field. The @cur is a loop cursor for searching the
1966  * kretprobe return addresses on the @tsk. The '*@cur' should be NULL at the
1967  * first call, but '@cur' itself must NOT NULL.
1968  */
1969 unsigned long kretprobe_find_ret_addr(struct task_struct *tsk, void *fp,
1970                                       struct llist_node **cur)
1971 {
1972         struct kretprobe_instance *ri = NULL;
1973         kprobe_opcode_t *ret;
1974
1975         if (WARN_ON_ONCE(!cur))
1976                 return 0;
1977
1978         do {
1979                 ret = __kretprobe_find_ret_addr(tsk, cur);
1980                 if (!ret)
1981                         break;
1982                 ri = container_of(*cur, struct kretprobe_instance, llist);
1983         } while (ri->fp != fp);
1984
1985         return (unsigned long)ret;
1986 }
1987 NOKPROBE_SYMBOL(kretprobe_find_ret_addr);
1988
1989 void __weak arch_kretprobe_fixup_return(struct pt_regs *regs,
1990                                         kprobe_opcode_t *correct_ret_addr)
1991 {
1992         /*
1993          * Do nothing by default. Please fill this to update the fake return
1994          * address on the stack with the correct one on each arch if possible.
1995          */
1996 }
1997
1998 unsigned long __kretprobe_trampoline_handler(struct pt_regs *regs,
1999                                              void *frame_pointer)
2000 {
2001         kprobe_opcode_t *correct_ret_addr = NULL;
2002         struct kretprobe_instance *ri = NULL;
2003         struct llist_node *first, *node = NULL;
2004         struct kretprobe *rp;
2005
2006         /* Find correct address and all nodes for this frame. */
2007         correct_ret_addr = __kretprobe_find_ret_addr(current, &node);
2008         if (!correct_ret_addr) {
2009                 pr_err("kretprobe: Return address not found, not execute handler. Maybe there is a bug in the kernel.\n");
2010                 BUG_ON(1);
2011         }
2012
2013         /*
2014          * Set the return address as the instruction pointer, because if the
2015          * user handler calls stack_trace_save_regs() with this 'regs',
2016          * the stack trace will start from the instruction pointer.
2017          */
2018         instruction_pointer_set(regs, (unsigned long)correct_ret_addr);
2019
2020         /* Run the user handler of the nodes. */
2021         first = current->kretprobe_instances.first;
2022         while (first) {
2023                 ri = container_of(first, struct kretprobe_instance, llist);
2024
2025                 if (WARN_ON_ONCE(ri->fp != frame_pointer))
2026                         break;
2027
2028                 rp = get_kretprobe(ri);
2029                 if (rp && rp->handler) {
2030                         struct kprobe *prev = kprobe_running();
2031
2032                         __this_cpu_write(current_kprobe, &rp->kp);
2033                         ri->ret_addr = correct_ret_addr;
2034                         rp->handler(ri, regs);
2035                         __this_cpu_write(current_kprobe, prev);
2036                 }
2037                 if (first == node)
2038                         break;
2039
2040                 first = first->next;
2041         }
2042
2043         arch_kretprobe_fixup_return(regs, correct_ret_addr);
2044
2045         /* Unlink all nodes for this frame. */
2046         first = current->kretprobe_instances.first;
2047         current->kretprobe_instances.first = node->next;
2048         node->next = NULL;
2049
2050         /* Recycle free instances. */
2051         while (first) {
2052                 ri = container_of(first, struct kretprobe_instance, llist);
2053                 first = first->next;
2054
2055                 recycle_rp_inst(ri);
2056         }
2057
2058         return (unsigned long)correct_ret_addr;
2059 }
2060 NOKPROBE_SYMBOL(__kretprobe_trampoline_handler)
2061
2062 /*
2063  * This kprobe pre_handler is registered with every kretprobe. When probe
2064  * hits it will set up the return probe.
2065  */
2066 static int pre_handler_kretprobe(struct kprobe *p, struct pt_regs *regs)
2067 {
2068         struct kretprobe *rp = container_of(p, struct kretprobe, kp);
2069         struct kretprobe_instance *ri;
2070         struct freelist_node *fn;
2071
2072         fn = freelist_try_get(&rp->freelist);
2073         if (!fn) {
2074                 rp->nmissed++;
2075                 return 0;
2076         }
2077
2078         ri = container_of(fn, struct kretprobe_instance, freelist);
2079
2080         if (rp->entry_handler && rp->entry_handler(ri, regs)) {
2081                 freelist_add(&ri->freelist, &rp->freelist);
2082                 return 0;
2083         }
2084
2085         arch_prepare_kretprobe(ri, regs);
2086
2087         __llist_add(&ri->llist, &current->kretprobe_instances);
2088
2089         return 0;
2090 }
2091 NOKPROBE_SYMBOL(pre_handler_kretprobe);
2092 #else /* CONFIG_KRETPROBE_ON_RETHOOK */
2093 /*
2094  * This kprobe pre_handler is registered with every kretprobe. When probe
2095  * hits it will set up the return probe.
2096  */
2097 static int pre_handler_kretprobe(struct kprobe *p, struct pt_regs *regs)
2098 {
2099         struct kretprobe *rp = container_of(p, struct kretprobe, kp);
2100         struct kretprobe_instance *ri;
2101         struct rethook_node *rhn;
2102
2103         rhn = rethook_try_get(rp->rh);
2104         if (!rhn) {
2105                 rp->nmissed++;
2106                 return 0;
2107         }
2108
2109         ri = container_of(rhn, struct kretprobe_instance, node);
2110
2111         if (rp->entry_handler && rp->entry_handler(ri, regs))
2112                 rethook_recycle(rhn);
2113         else
2114                 rethook_hook(rhn, regs, kprobe_ftrace(p));
2115
2116         return 0;
2117 }
2118 NOKPROBE_SYMBOL(pre_handler_kretprobe);
2119
2120 static void kretprobe_rethook_handler(struct rethook_node *rh, void *data,
2121                                       struct pt_regs *regs)
2122 {
2123         struct kretprobe *rp = (struct kretprobe *)data;
2124         struct kretprobe_instance *ri;
2125         struct kprobe_ctlblk *kcb;
2126
2127         /* The data must NOT be null. This means rethook data structure is broken. */
2128         if (WARN_ON_ONCE(!data) || !rp->handler)
2129                 return;
2130
2131         __this_cpu_write(current_kprobe, &rp->kp);
2132         kcb = get_kprobe_ctlblk();
2133         kcb->kprobe_status = KPROBE_HIT_ACTIVE;
2134
2135         ri = container_of(rh, struct kretprobe_instance, node);
2136         rp->handler(ri, regs);
2137
2138         __this_cpu_write(current_kprobe, NULL);
2139 }
2140 NOKPROBE_SYMBOL(kretprobe_rethook_handler);
2141
2142 #endif /* !CONFIG_KRETPROBE_ON_RETHOOK */
2143
2144 /**
2145  * kprobe_on_func_entry() -- check whether given address is function entry
2146  * @addr: Target address
2147  * @sym:  Target symbol name
2148  * @offset: The offset from the symbol or the address
2149  *
2150  * This checks whether the given @addr+@offset or @sym+@offset is on the
2151  * function entry address or not.
2152  * This returns 0 if it is the function entry, or -EINVAL if it is not.
2153  * And also it returns -ENOENT if it fails the symbol or address lookup.
2154  * Caller must pass @addr or @sym (either one must be NULL), or this
2155  * returns -EINVAL.
2156  */
2157 int kprobe_on_func_entry(kprobe_opcode_t *addr, const char *sym, unsigned long offset)
2158 {
2159         bool on_func_entry;
2160         kprobe_opcode_t *kp_addr = _kprobe_addr(addr, sym, offset, &on_func_entry);
2161
2162         if (IS_ERR(kp_addr))
2163                 return PTR_ERR(kp_addr);
2164
2165         if (!on_func_entry)
2166                 return -EINVAL;
2167
2168         return 0;
2169 }
2170
2171 int register_kretprobe(struct kretprobe *rp)
2172 {
2173         int ret;
2174         struct kretprobe_instance *inst;
2175         int i;
2176         void *addr;
2177
2178         ret = kprobe_on_func_entry(rp->kp.addr, rp->kp.symbol_name, rp->kp.offset);
2179         if (ret)
2180                 return ret;
2181
2182         /* If only 'rp->kp.addr' is specified, check reregistering kprobes */
2183         if (rp->kp.addr && warn_kprobe_rereg(&rp->kp))
2184                 return -EINVAL;
2185
2186         if (kretprobe_blacklist_size) {
2187                 addr = kprobe_addr(&rp->kp);
2188                 if (IS_ERR(addr))
2189                         return PTR_ERR(addr);
2190
2191                 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
2192                         if (kretprobe_blacklist[i].addr == addr)
2193                                 return -EINVAL;
2194                 }
2195         }
2196
2197         if (rp->data_size > KRETPROBE_MAX_DATA_SIZE)
2198                 return -E2BIG;
2199
2200         rp->kp.pre_handler = pre_handler_kretprobe;
2201         rp->kp.post_handler = NULL;
2202
2203         /* Pre-allocate memory for max kretprobe instances */
2204         if (rp->maxactive <= 0) {
2205 #ifdef CONFIG_PREEMPTION
2206                 rp->maxactive = max_t(unsigned int, 10, 2*num_possible_cpus());
2207 #else
2208                 rp->maxactive = num_possible_cpus();
2209 #endif
2210         }
2211 #ifdef CONFIG_KRETPROBE_ON_RETHOOK
2212         rp->rh = rethook_alloc((void *)rp, kretprobe_rethook_handler);
2213         if (!rp->rh)
2214                 return -ENOMEM;
2215
2216         for (i = 0; i < rp->maxactive; i++) {
2217                 inst = kzalloc(sizeof(struct kretprobe_instance) +
2218                                rp->data_size, GFP_KERNEL);
2219                 if (inst == NULL) {
2220                         rethook_free(rp->rh);
2221                         rp->rh = NULL;
2222                         return -ENOMEM;
2223                 }
2224                 rethook_add_node(rp->rh, &inst->node);
2225         }
2226         rp->nmissed = 0;
2227         /* Establish function entry probe point */
2228         ret = register_kprobe(&rp->kp);
2229         if (ret != 0) {
2230                 rethook_free(rp->rh);
2231                 rp->rh = NULL;
2232         }
2233 #else   /* !CONFIG_KRETPROBE_ON_RETHOOK */
2234         rp->freelist.head = NULL;
2235         rp->rph = kzalloc(sizeof(struct kretprobe_holder), GFP_KERNEL);
2236         if (!rp->rph)
2237                 return -ENOMEM;
2238
2239         rp->rph->rp = rp;
2240         for (i = 0; i < rp->maxactive; i++) {
2241                 inst = kzalloc(sizeof(struct kretprobe_instance) +
2242                                rp->data_size, GFP_KERNEL);
2243                 if (inst == NULL) {
2244                         refcount_set(&rp->rph->ref, i);
2245                         free_rp_inst(rp);
2246                         return -ENOMEM;
2247                 }
2248                 inst->rph = rp->rph;
2249                 freelist_add(&inst->freelist, &rp->freelist);
2250         }
2251         refcount_set(&rp->rph->ref, i);
2252
2253         rp->nmissed = 0;
2254         /* Establish function entry probe point */
2255         ret = register_kprobe(&rp->kp);
2256         if (ret != 0)
2257                 free_rp_inst(rp);
2258 #endif
2259         return ret;
2260 }
2261 EXPORT_SYMBOL_GPL(register_kretprobe);
2262
2263 int register_kretprobes(struct kretprobe **rps, int num)
2264 {
2265         int ret = 0, i;
2266
2267         if (num <= 0)
2268                 return -EINVAL;
2269         for (i = 0; i < num; i++) {
2270                 ret = register_kretprobe(rps[i]);
2271                 if (ret < 0) {
2272                         if (i > 0)
2273                                 unregister_kretprobes(rps, i);
2274                         break;
2275                 }
2276         }
2277         return ret;
2278 }
2279 EXPORT_SYMBOL_GPL(register_kretprobes);
2280
2281 void unregister_kretprobe(struct kretprobe *rp)
2282 {
2283         unregister_kretprobes(&rp, 1);
2284 }
2285 EXPORT_SYMBOL_GPL(unregister_kretprobe);
2286
2287 void unregister_kretprobes(struct kretprobe **rps, int num)
2288 {
2289         int i;
2290
2291         if (num <= 0)
2292                 return;
2293         mutex_lock(&kprobe_mutex);
2294         for (i = 0; i < num; i++) {
2295                 if (__unregister_kprobe_top(&rps[i]->kp) < 0)
2296                         rps[i]->kp.addr = NULL;
2297 #ifdef CONFIG_KRETPROBE_ON_RETHOOK
2298                 rethook_free(rps[i]->rh);
2299 #else
2300                 rps[i]->rph->rp = NULL;
2301 #endif
2302         }
2303         mutex_unlock(&kprobe_mutex);
2304
2305         synchronize_rcu();
2306         for (i = 0; i < num; i++) {
2307                 if (rps[i]->kp.addr) {
2308                         __unregister_kprobe_bottom(&rps[i]->kp);
2309 #ifndef CONFIG_KRETPROBE_ON_RETHOOK
2310                         free_rp_inst(rps[i]);
2311 #endif
2312                 }
2313         }
2314 }
2315 EXPORT_SYMBOL_GPL(unregister_kretprobes);
2316
2317 #else /* CONFIG_KRETPROBES */
2318 int register_kretprobe(struct kretprobe *rp)
2319 {
2320         return -EOPNOTSUPP;
2321 }
2322 EXPORT_SYMBOL_GPL(register_kretprobe);
2323
2324 int register_kretprobes(struct kretprobe **rps, int num)
2325 {
2326         return -EOPNOTSUPP;
2327 }
2328 EXPORT_SYMBOL_GPL(register_kretprobes);
2329
2330 void unregister_kretprobe(struct kretprobe *rp)
2331 {
2332 }
2333 EXPORT_SYMBOL_GPL(unregister_kretprobe);
2334
2335 void unregister_kretprobes(struct kretprobe **rps, int num)
2336 {
2337 }
2338 EXPORT_SYMBOL_GPL(unregister_kretprobes);
2339
2340 static int pre_handler_kretprobe(struct kprobe *p, struct pt_regs *regs)
2341 {
2342         return 0;
2343 }
2344 NOKPROBE_SYMBOL(pre_handler_kretprobe);
2345
2346 #endif /* CONFIG_KRETPROBES */
2347
2348 /* Set the kprobe gone and remove its instruction buffer. */
2349 static void kill_kprobe(struct kprobe *p)
2350 {
2351         struct kprobe *kp;
2352
2353         lockdep_assert_held(&kprobe_mutex);
2354
2355         p->flags |= KPROBE_FLAG_GONE;
2356         if (kprobe_aggrprobe(p)) {
2357                 /*
2358                  * If this is an aggr_kprobe, we have to list all the
2359                  * chained probes and mark them GONE.
2360                  */
2361                 list_for_each_entry(kp, &p->list, list)
2362                         kp->flags |= KPROBE_FLAG_GONE;
2363                 p->post_handler = NULL;
2364                 kill_optimized_kprobe(p);
2365         }
2366         /*
2367          * Here, we can remove insn_slot safely, because no thread calls
2368          * the original probed function (which will be freed soon) any more.
2369          */
2370         arch_remove_kprobe(p);
2371
2372         /*
2373          * The module is going away. We should disarm the kprobe which
2374          * is using ftrace, because ftrace framework is still available at
2375          * 'MODULE_STATE_GOING' notification.
2376          */
2377         if (kprobe_ftrace(p) && !kprobe_disabled(p) && !kprobes_all_disarmed)
2378                 disarm_kprobe_ftrace(p);
2379 }
2380
2381 /* Disable one kprobe */
2382 int disable_kprobe(struct kprobe *kp)
2383 {
2384         int ret = 0;
2385         struct kprobe *p;
2386
2387         mutex_lock(&kprobe_mutex);
2388
2389         /* Disable this kprobe */
2390         p = __disable_kprobe(kp);
2391         if (IS_ERR(p))
2392                 ret = PTR_ERR(p);
2393
2394         mutex_unlock(&kprobe_mutex);
2395         return ret;
2396 }
2397 EXPORT_SYMBOL_GPL(disable_kprobe);
2398
2399 /* Enable one kprobe */
2400 int enable_kprobe(struct kprobe *kp)
2401 {
2402         int ret = 0;
2403         struct kprobe *p;
2404
2405         mutex_lock(&kprobe_mutex);
2406
2407         /* Check whether specified probe is valid. */
2408         p = __get_valid_kprobe(kp);
2409         if (unlikely(p == NULL)) {
2410                 ret = -EINVAL;
2411                 goto out;
2412         }
2413
2414         if (kprobe_gone(kp)) {
2415                 /* This kprobe has gone, we couldn't enable it. */
2416                 ret = -EINVAL;
2417                 goto out;
2418         }
2419
2420         if (p != kp)
2421                 kp->flags &= ~KPROBE_FLAG_DISABLED;
2422
2423         if (!kprobes_all_disarmed && kprobe_disabled(p)) {
2424                 p->flags &= ~KPROBE_FLAG_DISABLED;
2425                 ret = arm_kprobe(p);
2426                 if (ret)
2427                         p->flags |= KPROBE_FLAG_DISABLED;
2428         }
2429 out:
2430         mutex_unlock(&kprobe_mutex);
2431         return ret;
2432 }
2433 EXPORT_SYMBOL_GPL(enable_kprobe);
2434
2435 /* Caller must NOT call this in usual path. This is only for critical case */
2436 void dump_kprobe(struct kprobe *kp)
2437 {
2438         pr_err("Dump kprobe:\n.symbol_name = %s, .offset = %x, .addr = %pS\n",
2439                kp->symbol_name, kp->offset, kp->addr);
2440 }
2441 NOKPROBE_SYMBOL(dump_kprobe);
2442
2443 int kprobe_add_ksym_blacklist(unsigned long entry)
2444 {
2445         struct kprobe_blacklist_entry *ent;
2446         unsigned long offset = 0, size = 0;
2447
2448         if (!kernel_text_address(entry) ||
2449             !kallsyms_lookup_size_offset(entry, &size, &offset))
2450                 return -EINVAL;
2451
2452         ent = kmalloc(sizeof(*ent), GFP_KERNEL);
2453         if (!ent)
2454                 return -ENOMEM;
2455         ent->start_addr = entry;
2456         ent->end_addr = entry + size;
2457         INIT_LIST_HEAD(&ent->list);
2458         list_add_tail(&ent->list, &kprobe_blacklist);
2459
2460         return (int)size;
2461 }
2462
2463 /* Add all symbols in given area into kprobe blacklist */
2464 int kprobe_add_area_blacklist(unsigned long start, unsigned long end)
2465 {
2466         unsigned long entry;
2467         int ret = 0;
2468
2469         for (entry = start; entry < end; entry += ret) {
2470                 ret = kprobe_add_ksym_blacklist(entry);
2471                 if (ret < 0)
2472                         return ret;
2473                 if (ret == 0)   /* In case of alias symbol */
2474                         ret = 1;
2475         }
2476         return 0;
2477 }
2478
2479 /* Remove all symbols in given area from kprobe blacklist */
2480 static void kprobe_remove_area_blacklist(unsigned long start, unsigned long end)
2481 {
2482         struct kprobe_blacklist_entry *ent, *n;
2483
2484         list_for_each_entry_safe(ent, n, &kprobe_blacklist, list) {
2485                 if (ent->start_addr < start || ent->start_addr >= end)
2486                         continue;
2487                 list_del(&ent->list);
2488                 kfree(ent);
2489         }
2490 }
2491
2492 static void kprobe_remove_ksym_blacklist(unsigned long entry)
2493 {
2494         kprobe_remove_area_blacklist(entry, entry + 1);
2495 }
2496
2497 int __weak arch_kprobe_get_kallsym(unsigned int *symnum, unsigned long *value,
2498                                    char *type, char *sym)
2499 {
2500         return -ERANGE;
2501 }
2502
2503 int kprobe_get_kallsym(unsigned int symnum, unsigned long *value, char *type,
2504                        char *sym)
2505 {
2506 #ifdef __ARCH_WANT_KPROBES_INSN_SLOT
2507         if (!kprobe_cache_get_kallsym(&kprobe_insn_slots, &symnum, value, type, sym))
2508                 return 0;
2509 #ifdef CONFIG_OPTPROBES
2510         if (!kprobe_cache_get_kallsym(&kprobe_optinsn_slots, &symnum, value, type, sym))
2511                 return 0;
2512 #endif
2513 #endif
2514         if (!arch_kprobe_get_kallsym(&symnum, value, type, sym))
2515                 return 0;
2516         return -ERANGE;
2517 }
2518
2519 int __init __weak arch_populate_kprobe_blacklist(void)
2520 {
2521         return 0;
2522 }
2523
2524 /*
2525  * Lookup and populate the kprobe_blacklist.
2526  *
2527  * Unlike the kretprobe blacklist, we'll need to determine
2528  * the range of addresses that belong to the said functions,
2529  * since a kprobe need not necessarily be at the beginning
2530  * of a function.
2531  */
2532 static int __init populate_kprobe_blacklist(unsigned long *start,
2533                                              unsigned long *end)
2534 {
2535         unsigned long entry;
2536         unsigned long *iter;
2537         int ret;
2538
2539         for (iter = start; iter < end; iter++) {
2540                 entry = (unsigned long)dereference_symbol_descriptor((void *)*iter);
2541                 ret = kprobe_add_ksym_blacklist(entry);
2542                 if (ret == -EINVAL)
2543                         continue;
2544                 if (ret < 0)
2545                         return ret;
2546         }
2547
2548         /* Symbols in '__kprobes_text' are blacklisted */
2549         ret = kprobe_add_area_blacklist((unsigned long)__kprobes_text_start,
2550                                         (unsigned long)__kprobes_text_end);
2551         if (ret)
2552                 return ret;
2553
2554         /* Symbols in 'noinstr' section are blacklisted */
2555         ret = kprobe_add_area_blacklist((unsigned long)__noinstr_text_start,
2556                                         (unsigned long)__noinstr_text_end);
2557
2558         return ret ? : arch_populate_kprobe_blacklist();
2559 }
2560
2561 static void add_module_kprobe_blacklist(struct module *mod)
2562 {
2563         unsigned long start, end;
2564         int i;
2565
2566         if (mod->kprobe_blacklist) {
2567                 for (i = 0; i < mod->num_kprobe_blacklist; i++)
2568                         kprobe_add_ksym_blacklist(mod->kprobe_blacklist[i]);
2569         }
2570
2571         start = (unsigned long)mod->kprobes_text_start;
2572         if (start) {
2573                 end = start + mod->kprobes_text_size;
2574                 kprobe_add_area_blacklist(start, end);
2575         }
2576
2577         start = (unsigned long)mod->noinstr_text_start;
2578         if (start) {
2579                 end = start + mod->noinstr_text_size;
2580                 kprobe_add_area_blacklist(start, end);
2581         }
2582 }
2583
2584 static void remove_module_kprobe_blacklist(struct module *mod)
2585 {
2586         unsigned long start, end;
2587         int i;
2588
2589         if (mod->kprobe_blacklist) {
2590                 for (i = 0; i < mod->num_kprobe_blacklist; i++)
2591                         kprobe_remove_ksym_blacklist(mod->kprobe_blacklist[i]);
2592         }
2593
2594         start = (unsigned long)mod->kprobes_text_start;
2595         if (start) {
2596                 end = start + mod->kprobes_text_size;
2597                 kprobe_remove_area_blacklist(start, end);
2598         }
2599
2600         start = (unsigned long)mod->noinstr_text_start;
2601         if (start) {
2602                 end = start + mod->noinstr_text_size;
2603                 kprobe_remove_area_blacklist(start, end);
2604         }
2605 }
2606
2607 /* Module notifier call back, checking kprobes on the module */
2608 static int kprobes_module_callback(struct notifier_block *nb,
2609                                    unsigned long val, void *data)
2610 {
2611         struct module *mod = data;
2612         struct hlist_head *head;
2613         struct kprobe *p;
2614         unsigned int i;
2615         int checkcore = (val == MODULE_STATE_GOING);
2616
2617         if (val == MODULE_STATE_COMING) {
2618                 mutex_lock(&kprobe_mutex);
2619                 add_module_kprobe_blacklist(mod);
2620                 mutex_unlock(&kprobe_mutex);
2621         }
2622         if (val != MODULE_STATE_GOING && val != MODULE_STATE_LIVE)
2623                 return NOTIFY_DONE;
2624
2625         /*
2626          * When 'MODULE_STATE_GOING' was notified, both of module '.text' and
2627          * '.init.text' sections would be freed. When 'MODULE_STATE_LIVE' was
2628          * notified, only '.init.text' section would be freed. We need to
2629          * disable kprobes which have been inserted in the sections.
2630          */
2631         mutex_lock(&kprobe_mutex);
2632         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2633                 head = &kprobe_table[i];
2634                 hlist_for_each_entry(p, head, hlist)
2635                         if (within_module_init((unsigned long)p->addr, mod) ||
2636                             (checkcore &&
2637                              within_module_core((unsigned long)p->addr, mod))) {
2638                                 /*
2639                                  * The vaddr this probe is installed will soon
2640                                  * be vfreed buy not synced to disk. Hence,
2641                                  * disarming the breakpoint isn't needed.
2642                                  *
2643                                  * Note, this will also move any optimized probes
2644                                  * that are pending to be removed from their
2645                                  * corresponding lists to the 'freeing_list' and
2646                                  * will not be touched by the delayed
2647                                  * kprobe_optimizer() work handler.
2648                                  */
2649                                 kill_kprobe(p);
2650                         }
2651         }
2652         if (val == MODULE_STATE_GOING)
2653                 remove_module_kprobe_blacklist(mod);
2654         mutex_unlock(&kprobe_mutex);
2655         return NOTIFY_DONE;
2656 }
2657
2658 static struct notifier_block kprobe_module_nb = {
2659         .notifier_call = kprobes_module_callback,
2660         .priority = 0
2661 };
2662
2663 void kprobe_free_init_mem(void)
2664 {
2665         void *start = (void *)(&__init_begin);
2666         void *end = (void *)(&__init_end);
2667         struct hlist_head *head;
2668         struct kprobe *p;
2669         int i;
2670
2671         mutex_lock(&kprobe_mutex);
2672
2673         /* Kill all kprobes on initmem because the target code has been freed. */
2674         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2675                 head = &kprobe_table[i];
2676                 hlist_for_each_entry(p, head, hlist) {
2677                         if (start <= (void *)p->addr && (void *)p->addr < end)
2678                                 kill_kprobe(p);
2679                 }
2680         }
2681
2682         mutex_unlock(&kprobe_mutex);
2683 }
2684
2685 static int __init init_kprobes(void)
2686 {
2687         int i, err = 0;
2688
2689         /* FIXME allocate the probe table, currently defined statically */
2690         /* initialize all list heads */
2691         for (i = 0; i < KPROBE_TABLE_SIZE; i++)
2692                 INIT_HLIST_HEAD(&kprobe_table[i]);
2693
2694         err = populate_kprobe_blacklist(__start_kprobe_blacklist,
2695                                         __stop_kprobe_blacklist);
2696         if (err)
2697                 pr_err("Failed to populate blacklist (error %d), kprobes not restricted, be careful using them!\n", err);
2698
2699         if (kretprobe_blacklist_size) {
2700                 /* lookup the function address from its name */
2701                 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
2702                         kretprobe_blacklist[i].addr =
2703                                 kprobe_lookup_name(kretprobe_blacklist[i].name, 0);
2704                         if (!kretprobe_blacklist[i].addr)
2705                                 pr_err("Failed to lookup symbol '%s' for kretprobe blacklist. Maybe the target function is removed or renamed.\n",
2706                                        kretprobe_blacklist[i].name);
2707                 }
2708         }
2709
2710         /* By default, kprobes are armed */
2711         kprobes_all_disarmed = false;
2712
2713 #if defined(CONFIG_OPTPROBES) && defined(__ARCH_WANT_KPROBES_INSN_SLOT)
2714         /* Init 'kprobe_optinsn_slots' for allocation */
2715         kprobe_optinsn_slots.insn_size = MAX_OPTINSN_SIZE;
2716 #endif
2717
2718         err = arch_init_kprobes();
2719         if (!err)
2720                 err = register_die_notifier(&kprobe_exceptions_nb);
2721         if (!err)
2722                 err = register_module_notifier(&kprobe_module_nb);
2723
2724         kprobes_initialized = (err == 0);
2725         kprobe_sysctls_init();
2726         return err;
2727 }
2728 early_initcall(init_kprobes);
2729
2730 #if defined(CONFIG_OPTPROBES)
2731 static int __init init_optprobes(void)
2732 {
2733         /*
2734          * Enable kprobe optimization - this kicks the optimizer which
2735          * depends on synchronize_rcu_tasks() and ksoftirqd, that is
2736          * not spawned in early initcall. So delay the optimization.
2737          */
2738         optimize_all_kprobes();
2739
2740         return 0;
2741 }
2742 subsys_initcall(init_optprobes);
2743 #endif
2744
2745 #ifdef CONFIG_DEBUG_FS
2746 static void report_probe(struct seq_file *pi, struct kprobe *p,
2747                 const char *sym, int offset, char *modname, struct kprobe *pp)
2748 {
2749         char *kprobe_type;
2750         void *addr = p->addr;
2751
2752         if (p->pre_handler == pre_handler_kretprobe)
2753                 kprobe_type = "r";
2754         else
2755                 kprobe_type = "k";
2756
2757         if (!kallsyms_show_value(pi->file->f_cred))
2758                 addr = NULL;
2759
2760         if (sym)
2761                 seq_printf(pi, "%px  %s  %s+0x%x  %s ",
2762                         addr, kprobe_type, sym, offset,
2763                         (modname ? modname : " "));
2764         else    /* try to use %pS */
2765                 seq_printf(pi, "%px  %s  %pS ",
2766                         addr, kprobe_type, p->addr);
2767
2768         if (!pp)
2769                 pp = p;
2770         seq_printf(pi, "%s%s%s%s\n",
2771                 (kprobe_gone(p) ? "[GONE]" : ""),
2772                 ((kprobe_disabled(p) && !kprobe_gone(p)) ?  "[DISABLED]" : ""),
2773                 (kprobe_optimized(pp) ? "[OPTIMIZED]" : ""),
2774                 (kprobe_ftrace(pp) ? "[FTRACE]" : ""));
2775 }
2776
2777 static void *kprobe_seq_start(struct seq_file *f, loff_t *pos)
2778 {
2779         return (*pos < KPROBE_TABLE_SIZE) ? pos : NULL;
2780 }
2781
2782 static void *kprobe_seq_next(struct seq_file *f, void *v, loff_t *pos)
2783 {
2784         (*pos)++;
2785         if (*pos >= KPROBE_TABLE_SIZE)
2786                 return NULL;
2787         return pos;
2788 }
2789
2790 static void kprobe_seq_stop(struct seq_file *f, void *v)
2791 {
2792         /* Nothing to do */
2793 }
2794
2795 static int show_kprobe_addr(struct seq_file *pi, void *v)
2796 {
2797         struct hlist_head *head;
2798         struct kprobe *p, *kp;
2799         const char *sym = NULL;
2800         unsigned int i = *(loff_t *) v;
2801         unsigned long offset = 0;
2802         char *modname, namebuf[KSYM_NAME_LEN];
2803
2804         head = &kprobe_table[i];
2805         preempt_disable();
2806         hlist_for_each_entry_rcu(p, head, hlist) {
2807                 sym = kallsyms_lookup((unsigned long)p->addr, NULL,
2808                                         &offset, &modname, namebuf);
2809                 if (kprobe_aggrprobe(p)) {
2810                         list_for_each_entry_rcu(kp, &p->list, list)
2811                                 report_probe(pi, kp, sym, offset, modname, p);
2812                 } else
2813                         report_probe(pi, p, sym, offset, modname, NULL);
2814         }
2815         preempt_enable();
2816         return 0;
2817 }
2818
2819 static const struct seq_operations kprobes_sops = {
2820         .start = kprobe_seq_start,
2821         .next  = kprobe_seq_next,
2822         .stop  = kprobe_seq_stop,
2823         .show  = show_kprobe_addr
2824 };
2825
2826 DEFINE_SEQ_ATTRIBUTE(kprobes);
2827
2828 /* kprobes/blacklist -- shows which functions can not be probed */
2829 static void *kprobe_blacklist_seq_start(struct seq_file *m, loff_t *pos)
2830 {
2831         mutex_lock(&kprobe_mutex);
2832         return seq_list_start(&kprobe_blacklist, *pos);
2833 }
2834
2835 static void *kprobe_blacklist_seq_next(struct seq_file *m, void *v, loff_t *pos)
2836 {
2837         return seq_list_next(v, &kprobe_blacklist, pos);
2838 }
2839
2840 static int kprobe_blacklist_seq_show(struct seq_file *m, void *v)
2841 {
2842         struct kprobe_blacklist_entry *ent =
2843                 list_entry(v, struct kprobe_blacklist_entry, list);
2844
2845         /*
2846          * If '/proc/kallsyms' is not showing kernel address, we won't
2847          * show them here either.
2848          */
2849         if (!kallsyms_show_value(m->file->f_cred))
2850                 seq_printf(m, "0x%px-0x%px\t%ps\n", NULL, NULL,
2851                            (void *)ent->start_addr);
2852         else
2853                 seq_printf(m, "0x%px-0x%px\t%ps\n", (void *)ent->start_addr,
2854                            (void *)ent->end_addr, (void *)ent->start_addr);
2855         return 0;
2856 }
2857
2858 static void kprobe_blacklist_seq_stop(struct seq_file *f, void *v)
2859 {
2860         mutex_unlock(&kprobe_mutex);
2861 }
2862
2863 static const struct seq_operations kprobe_blacklist_sops = {
2864         .start = kprobe_blacklist_seq_start,
2865         .next  = kprobe_blacklist_seq_next,
2866         .stop  = kprobe_blacklist_seq_stop,
2867         .show  = kprobe_blacklist_seq_show,
2868 };
2869 DEFINE_SEQ_ATTRIBUTE(kprobe_blacklist);
2870
2871 static int arm_all_kprobes(void)
2872 {
2873         struct hlist_head *head;
2874         struct kprobe *p;
2875         unsigned int i, total = 0, errors = 0;
2876         int err, ret = 0;
2877
2878         mutex_lock(&kprobe_mutex);
2879
2880         /* If kprobes are armed, just return */
2881         if (!kprobes_all_disarmed)
2882                 goto already_enabled;
2883
2884         /*
2885          * optimize_kprobe() called by arm_kprobe() checks
2886          * kprobes_all_disarmed, so set kprobes_all_disarmed before
2887          * arm_kprobe.
2888          */
2889         kprobes_all_disarmed = false;
2890         /* Arming kprobes doesn't optimize kprobe itself */
2891         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2892                 head = &kprobe_table[i];
2893                 /* Arm all kprobes on a best-effort basis */
2894                 hlist_for_each_entry(p, head, hlist) {
2895                         if (!kprobe_disabled(p)) {
2896                                 err = arm_kprobe(p);
2897                                 if (err)  {
2898                                         errors++;
2899                                         ret = err;
2900                                 }
2901                                 total++;
2902                         }
2903                 }
2904         }
2905
2906         if (errors)
2907                 pr_warn("Kprobes globally enabled, but failed to enable %d out of %d probes. Please check which kprobes are kept disabled via debugfs.\n",
2908                         errors, total);
2909         else
2910                 pr_info("Kprobes globally enabled\n");
2911
2912 already_enabled:
2913         mutex_unlock(&kprobe_mutex);
2914         return ret;
2915 }
2916
2917 static int disarm_all_kprobes(void)
2918 {
2919         struct hlist_head *head;
2920         struct kprobe *p;
2921         unsigned int i, total = 0, errors = 0;
2922         int err, ret = 0;
2923
2924         mutex_lock(&kprobe_mutex);
2925
2926         /* If kprobes are already disarmed, just return */
2927         if (kprobes_all_disarmed) {
2928                 mutex_unlock(&kprobe_mutex);
2929                 return 0;
2930         }
2931
2932         kprobes_all_disarmed = true;
2933
2934         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2935                 head = &kprobe_table[i];
2936                 /* Disarm all kprobes on a best-effort basis */
2937                 hlist_for_each_entry(p, head, hlist) {
2938                         if (!arch_trampoline_kprobe(p) && !kprobe_disabled(p)) {
2939                                 err = disarm_kprobe(p, false);
2940                                 if (err) {
2941                                         errors++;
2942                                         ret = err;
2943                                 }
2944                                 total++;
2945                         }
2946                 }
2947         }
2948
2949         if (errors)
2950                 pr_warn("Kprobes globally disabled, but failed to disable %d out of %d probes. Please check which kprobes are kept enabled via debugfs.\n",
2951                         errors, total);
2952         else
2953                 pr_info("Kprobes globally disabled\n");
2954
2955         mutex_unlock(&kprobe_mutex);
2956
2957         /* Wait for disarming all kprobes by optimizer */
2958         wait_for_kprobe_optimizer();
2959
2960         return ret;
2961 }
2962
2963 /*
2964  * XXX: The debugfs bool file interface doesn't allow for callbacks
2965  * when the bool state is switched. We can reuse that facility when
2966  * available
2967  */
2968 static ssize_t read_enabled_file_bool(struct file *file,
2969                char __user *user_buf, size_t count, loff_t *ppos)
2970 {
2971         char buf[3];
2972
2973         if (!kprobes_all_disarmed)
2974                 buf[0] = '1';
2975         else
2976                 buf[0] = '0';
2977         buf[1] = '\n';
2978         buf[2] = 0x00;
2979         return simple_read_from_buffer(user_buf, count, ppos, buf, 2);
2980 }
2981
2982 static ssize_t write_enabled_file_bool(struct file *file,
2983                const char __user *user_buf, size_t count, loff_t *ppos)
2984 {
2985         bool enable;
2986         int ret;
2987
2988         ret = kstrtobool_from_user(user_buf, count, &enable);
2989         if (ret)
2990                 return ret;
2991
2992         ret = enable ? arm_all_kprobes() : disarm_all_kprobes();
2993         if (ret)
2994                 return ret;
2995
2996         return count;
2997 }
2998
2999 static const struct file_operations fops_kp = {
3000         .read =         read_enabled_file_bool,
3001         .write =        write_enabled_file_bool,
3002         .llseek =       default_llseek,
3003 };
3004
3005 static int __init debugfs_kprobe_init(void)
3006 {
3007         struct dentry *dir;
3008
3009         dir = debugfs_create_dir("kprobes", NULL);
3010
3011         debugfs_create_file("list", 0400, dir, NULL, &kprobes_fops);
3012
3013         debugfs_create_file("enabled", 0600, dir, NULL, &fops_kp);
3014
3015         debugfs_create_file("blacklist", 0400, dir, NULL,
3016                             &kprobe_blacklist_fops);
3017
3018         return 0;
3019 }
3020
3021 late_initcall(debugfs_kprobe_init);
3022 #endif /* CONFIG_DEBUG_FS */